Merge tag 'pinctrl-v7.3-1' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrlgrafted

Pull pin control updates from Linus Walleij:
 "Core changes:

   - Use the non-blocking mux_state_try_select() in the generic MUX pin
     control back-end

   - Free pin maps on pinctrl_generic_to_map() failure in the core
     helpers

  New hardware pin control:

   - Qualcomm Maili TLMM SoC, PMG1110 PMIC, and Eliza LPASS LPI TLMM SoC
     (this is for the low power audio, LPASS = Low Power Audio SubSystem
     portions)

   - Upboard support on top of the Intel pin controller

   - Apple T6030 and T6031 support (just compatible strings)

   - Samsung Exynos 8855 SoC

   - Mediatek MT6858 SoC

   - Rockchip RK3308B and RV1106 SoCs

   - Airoha EN7523 and AN7563 SoCs

  Improvements:

   - Qualcomm TLMM GPIOs are unconditionally marked as wakeup capable

   - Qualcomm improvements on top of the Qualcomm PDC (power domain
     controller) irqchip improvements from tglx:s tree

   - Qualcomm IPQ806x and IPQ9650 non-urgent pin grop fixups

   - The NPCM8xx driver has been put into shape fixing many pin and
     group definitions

   - Some non-urgent Tegra 264 pin assignments are fixed up

   - Some non-urgent fixups to Airoha AN7581 and AN7583 registers and
     pin assignments, missing features etc"

* tag 'pinctrl-v7.3-1' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl: (130 commits)
  dt-bindings: pinctrl: Convert TI DA850 pupd to DT schema
  pinctrl: meson: sync some modify from A4
  pinctrl: meson: a4: Add input enable pin configuration
  pinctrl: realtek: rtd1625: remove unused group name spdif_sel
  pinctrl: realtek: rtd1625: fix base_bit for VE4 GPIO 13
  dt-bindings: pinctrl: microchip,pic32mzda-pinctrl: Convert to DT schema
  Revert "Merge branch 'ib-rsk7204' into devel"
  pinctrl: rockchip: Add RV1106 pinctrl support
  dt-bindings: pinctrl: rockchip: Add RV1106 compatible
  pinctrl: rockchip: Decode drive strength in the get function
  pinctrl: fix PINCTRL_GENERIC_MUX not always being selectable
  pinctrl: airoha: add support of an7563 SoC
  dt-bindings: pinctrl: airoha: add support of an7563 pin controller
  pinctrl: airoha: try to find chip scu node by phandle first
  pinctrl: airoha: add support of en7523 SoC
  dt-bindings: pinctrl: airoha: add support of en7523 pin controller
  pinctrl: airoha: an7583: add support for olt pinmux
  pinctrl: airoha: an7583: add support for pon_alt pinmux
  pinctrl: airoha: an7583: add support for npu_uart pinmux
  dt-bindings: pinctrl: airoha: an7583: add missed features
  ...

57 files changed, 26320 insertions, 0 deletions

diff --git a/arch/x86/crypto/.gitignore b/arch/x86/crypto/.gitignore
new file mode 100644
index 000000000..580c839bb
--- /dev/null
+++ b/arch/x86/crypto/.gitignore
@@ -0,0 +1,2 @@
+# SPDX-License-Identifier: GPL-2.0-only
+poly1305-x86_64-cryptogams.S
diff --git a/arch/x86/crypto/Kconfig b/arch/x86/crypto/Kconfig
new file mode 100644
index 000000000..f65d7b837
--- /dev/null
+++ b/arch/x86/crypto/Kconfig
@@ -0,0 +1,320 @@
+# SPDX-License-Identifier: GPL-2.0
+
+menu "Accelerated Cryptographic Algorithms for CPU (x86)"
+
+config CRYPTO_AES_NI_INTEL
+	tristate "Ciphers: AES, modes: ECB, CBC, CTS, CTR, XCTR, XTS, GCM (AES-NI/VAES)"
+	select CRYPTO_AEAD
+	select CRYPTO_LIB_AES
+	select CRYPTO_LIB_GF128MUL
+	select CRYPTO_SKCIPHER
+	help
+	  AEAD cipher: AES with GCM
+	  Length-preserving ciphers: AES with ECB, CBC, CTS, CTR, XCTR, XTS
+
+	  Architecture: x86 (32-bit and 64-bit) using:
+	  - AES-NI (AES new instructions)
+	  - VAES (Vector AES)
+
+	  Some algorithm implementations are supported only in 64-bit builds,
+	  and some have additional prerequisites such as AVX2 or AVX512.
+
+config CRYPTO_BLOWFISH_X86_64
+	tristate "Ciphers: Blowfish, modes: ECB, CBC"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_BLOWFISH_COMMON
+	imply CRYPTO_CTR
+	help
+	  Block cipher: Blowfish cipher algorithm
+	  Length-preserving ciphers: Blowfish with ECB and CBC modes
+
+	  Architecture: x86_64
+
+config CRYPTO_CAMELLIA_X86_64
+	tristate "Ciphers: Camellia with modes: ECB, CBC"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	imply CRYPTO_CTR
+	help
+	  Block cipher: Camellia cipher algorithms
+	  Length-preserving ciphers: Camellia with ECB and CBC modes
+
+	  Architecture: x86_64
+
+config CRYPTO_CAMELLIA_AESNI_AVX_X86_64
+	tristate "Ciphers: Camellia with modes: ECB, CBC (AES-NI/AVX)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_CAMELLIA_X86_64
+	imply CRYPTO_XTS
+	help
+	  Length-preserving ciphers: Camellia with ECB and CBC modes
+
+	  Architecture: x86_64 using:
+	  - AES-NI (AES New Instructions)
+	  - AVX (Advanced Vector Extensions)
+
+config CRYPTO_CAMELLIA_AESNI_AVX2_X86_64
+	tristate "Ciphers: Camellia with modes: ECB, CBC (AES-NI/AVX2)"
+	depends on 64BIT
+	select CRYPTO_CAMELLIA_AESNI_AVX_X86_64
+	help
+	  Length-preserving ciphers: Camellia with ECB and CBC modes
+
+	  Architecture: x86_64 using:
+	  - AES-NI (AES New Instructions)
+	  - AVX2 (Advanced Vector Extensions 2)
+
+config CRYPTO_CAST5_AVX_X86_64
+	tristate "Ciphers: CAST5 with modes: ECB, CBC (AVX)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_CAST5
+	select CRYPTO_CAST_COMMON
+	imply CRYPTO_CTR
+	help
+	  Length-preserving ciphers: CAST5 (CAST-128) cipher algorithm
+	  (RFC2144) with ECB and CBC modes
+
+	  Architecture: x86_64 using:
+	  - AVX (Advanced Vector Extensions)
+
+	  Processes 16 blocks in parallel.
+
+config CRYPTO_CAST6_AVX_X86_64
+	tristate "Ciphers: CAST6 with modes: ECB, CBC (AVX)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_CAST6
+	select CRYPTO_CAST_COMMON
+	imply CRYPTO_XTS
+	imply CRYPTO_CTR
+	help
+	  Length-preserving ciphers: CAST6 (CAST-256) cipher algorithm
+	  (RFC2612) with ECB and CBC modes
+
+	  Architecture: x86_64 using:
+	  - AVX (Advanced Vector Extensions)
+
+	  Processes eight blocks in parallel.
+
+config CRYPTO_SERPENT_SSE2_X86_64
+	tristate "Ciphers: Serpent with modes: ECB, CBC (SSE2)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_SERPENT
+	imply CRYPTO_CTR
+	help
+	  Length-preserving ciphers: Serpent cipher algorithm
+	  with ECB and CBC modes
+
+	  Architecture: x86_64 using:
+	  - SSE2 (Streaming SIMD Extensions 2)
+
+	  Processes eight blocks in parallel.
+
+config CRYPTO_SERPENT_SSE2_586
+	tristate "Ciphers: Serpent with modes: ECB, CBC (32-bit with SSE2)"
+	depends on !64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_SERPENT
+	imply CRYPTO_CTR
+	help
+	  Length-preserving ciphers: Serpent cipher algorithm
+	  with ECB and CBC modes
+
+	  Architecture: x86 (32-bit) using:
+	  - SSE2 (Streaming SIMD Extensions 2)
+
+	  Processes four blocks in parallel.
+
+config CRYPTO_SERPENT_AVX_X86_64
+	tristate "Ciphers: Serpent with modes: ECB, CBC (AVX)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_SERPENT
+	imply CRYPTO_XTS
+	imply CRYPTO_CTR
+	help
+	  Length-preserving ciphers: Serpent cipher algorithm
+	  with ECB and CBC modes
+
+	  Architecture: x86_64 using:
+	  - AVX (Advanced Vector Extensions)
+
+	  Processes eight blocks in parallel.
+
+config CRYPTO_SERPENT_AVX2_X86_64
+	tristate "Ciphers: Serpent with modes: ECB, CBC (AVX2)"
+	depends on 64BIT
+	select CRYPTO_SERPENT_AVX_X86_64
+	help
+	  Length-preserving ciphers: Serpent cipher algorithm
+	  with ECB and CBC modes
+
+	  Architecture: x86_64 using:
+	  - AVX2 (Advanced Vector Extensions 2)
+
+	  Processes 16 blocks in parallel.
+
+config CRYPTO_SM4_AESNI_AVX_X86_64
+	tristate "Ciphers: SM4 with modes: ECB, CBC, CTR (AES-NI/AVX)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_ALGAPI
+	select CRYPTO_SM4
+	help
+	  Length-preserving ciphers: SM4 cipher algorithms
+	  (OSCCA GB/T 32907-2016) with ECB, CBC, and CTR modes
+
+	  Architecture: x86_64 using:
+	  - AES-NI (AES New Instructions)
+	  - AVX (Advanced Vector Extensions)
+
+	  Through two affine transforms,
+	  we can use the AES S-Box to simulate the SM4 S-Box to achieve the
+	  effect of instruction acceleration.
+
+	  If unsure, say N.
+
+config CRYPTO_SM4_AESNI_AVX2_X86_64
+	tristate "Ciphers: SM4 with modes: ECB, CBC, CTR (AES-NI/AVX2)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_ALGAPI
+	select CRYPTO_SM4
+	select CRYPTO_SM4_AESNI_AVX_X86_64
+	help
+	  Length-preserving ciphers: SM4 cipher algorithms
+	  (OSCCA GB/T 32907-2016) with ECB, CBC, and CTR modes
+
+	  Architecture: x86_64 using:
+	  - AES-NI (AES New Instructions)
+	  - AVX2 (Advanced Vector Extensions 2)
+
+	  Through two affine transforms,
+	  we can use the AES S-Box to simulate the SM4 S-Box to achieve the
+	  effect of instruction acceleration.
+
+	  If unsure, say N.
+
+config CRYPTO_TWOFISH_586
+	tristate "Ciphers: Twofish (32-bit)"
+	depends on !64BIT
+	select CRYPTO_ALGAPI
+	select CRYPTO_TWOFISH_COMMON
+	imply CRYPTO_CTR
+	help
+	  Block cipher: Twofish cipher algorithm
+
+	  Architecture: x86 (32-bit)
+
+config CRYPTO_TWOFISH_X86_64
+	tristate "Ciphers: Twofish"
+	depends on 64BIT
+	select CRYPTO_ALGAPI
+	select CRYPTO_TWOFISH_COMMON
+	imply CRYPTO_CTR
+	help
+	  Block cipher: Twofish cipher algorithm
+
+	  Architecture: x86_64
+
+config CRYPTO_TWOFISH_X86_64_3WAY
+	tristate "Ciphers: Twofish with modes: ECB, CBC (3-way parallel)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_TWOFISH_COMMON
+	select CRYPTO_TWOFISH_X86_64
+	help
+	  Length-preserving cipher: Twofish cipher algorithm
+	  with ECB and CBC modes
+
+	  Architecture: x86_64
+
+	  Processes three blocks in parallel, better utilizing resources of
+	  out-of-order CPUs.
+
+config CRYPTO_TWOFISH_AVX_X86_64
+	tristate "Ciphers: Twofish with modes: ECB, CBC (AVX)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_TWOFISH_COMMON
+	select CRYPTO_TWOFISH_X86_64
+	select CRYPTO_TWOFISH_X86_64_3WAY
+	imply CRYPTO_XTS
+	help
+	  Length-preserving cipher: Twofish cipher algorithm
+	  with ECB and CBC modes
+
+	  Architecture: x86_64 using:
+	  - AVX (Advanced Vector Extensions)
+
+	  Processes eight blocks in parallel.
+
+config CRYPTO_ARIA_AESNI_AVX_X86_64
+	tristate "Ciphers: ARIA with modes: ECB, CTR (AES-NI/AVX/GFNI)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_ALGAPI
+	select CRYPTO_ARIA
+	help
+	  Length-preserving cipher: ARIA cipher algorithms
+	  (RFC 5794) with ECB and CTR modes
+
+	  Architecture: x86_64 using:
+	  - AES-NI (AES New Instructions)
+	  - AVX (Advanced Vector Extensions)
+	  - GFNI (Galois Field New Instructions)
+
+	  Processes 16 blocks in parallel.
+
+config CRYPTO_ARIA_AESNI_AVX2_X86_64
+	tristate "Ciphers: ARIA with modes: ECB, CTR (AES-NI/AVX2/GFNI)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_ALGAPI
+	select CRYPTO_ARIA
+	select CRYPTO_ARIA_AESNI_AVX_X86_64
+	help
+	  Length-preserving cipher: ARIA cipher algorithms
+	  (RFC 5794) with ECB and CTR modes
+
+	  Architecture: x86_64 using:
+	  - AES-NI (AES New Instructions)
+	  - AVX2 (Advanced Vector Extensions)
+	  - GFNI (Galois Field New Instructions)
+
+	  Processes 32 blocks in parallel.
+
+config CRYPTO_ARIA_GFNI_AVX512_X86_64
+	tristate "Ciphers: ARIA with modes: ECB, CTR (AVX512/GFNI)"
+	depends on 64BIT
+	select CRYPTO_SKCIPHER
+	select CRYPTO_ALGAPI
+	select CRYPTO_ARIA
+	select CRYPTO_ARIA_AESNI_AVX_X86_64
+	select CRYPTO_ARIA_AESNI_AVX2_X86_64
+	help
+	  Length-preserving cipher: ARIA cipher algorithms
+	  (RFC 5794) with ECB and CTR modes
+
+	  Architecture: x86_64 using:
+	  - AVX512 (Advanced Vector Extensions)
+	  - GFNI (Galois Field New Instructions)
+
+	  Processes 64 blocks in parallel.
+
+config CRYPTO_AEGIS128_AESNI_SSE2
+	tristate "AEAD ciphers: AEGIS-128 (AES-NI/SSE4.1)"
+	depends on 64BIT
+	select CRYPTO_AEAD
+	help
+	  AEGIS-128 AEAD algorithm
+
+	  Architecture: x86_64 using:
+	  - AES-NI (AES New Instructions)
+	  - SSE4.1 (Streaming SIMD Extensions 4.1)
+
+endmenu
diff --git a/arch/x86/crypto/Makefile b/arch/x86/crypto/Makefile
new file mode 100644
index 000000000..e04ff8718
--- /dev/null
+++ b/arch/x86/crypto/Makefile
@@ -0,0 +1,63 @@
+# SPDX-License-Identifier: GPL-2.0
+#
+# x86 crypto algorithms
+
+obj-$(CONFIG_CRYPTO_TWOFISH_586) += twofish-i586.o
+twofish-i586-y := twofish-i586-asm_32.o twofish_glue.o
+obj-$(CONFIG_CRYPTO_TWOFISH_X86_64) += twofish-x86_64.o
+twofish-x86_64-y := twofish-x86_64-asm_64.o twofish_glue.o
+obj-$(CONFIG_CRYPTO_TWOFISH_X86_64_3WAY) += twofish-x86_64-3way.o
+twofish-x86_64-3way-y := twofish-x86_64-asm_64-3way.o twofish_glue_3way.o
+obj-$(CONFIG_CRYPTO_TWOFISH_AVX_X86_64) += twofish-avx-x86_64.o
+twofish-avx-x86_64-y := twofish-avx-x86_64-asm_64.o twofish_avx_glue.o
+
+obj-$(CONFIG_CRYPTO_SERPENT_SSE2_586) += serpent-sse2-i586.o
+serpent-sse2-i586-y := serpent-sse2-i586-asm_32.o serpent_sse2_glue.o
+obj-$(CONFIG_CRYPTO_SERPENT_SSE2_X86_64) += serpent-sse2-x86_64.o
+serpent-sse2-x86_64-y := serpent-sse2-x86_64-asm_64.o serpent_sse2_glue.o
+obj-$(CONFIG_CRYPTO_SERPENT_AVX_X86_64) += serpent-avx-x86_64.o
+serpent-avx-x86_64-y := serpent-avx-x86_64-asm_64.o serpent_avx_glue.o
+obj-$(CONFIG_CRYPTO_SERPENT_AVX2_X86_64) += serpent-avx2.o
+serpent-avx2-y := serpent-avx2-asm_64.o serpent_avx2_glue.o
+
+obj-$(CONFIG_CRYPTO_CAMELLIA_X86_64) += camellia-x86_64.o
+camellia-x86_64-y := camellia-x86_64-asm_64.o camellia_glue.o
+obj-$(CONFIG_CRYPTO_CAMELLIA_AESNI_AVX_X86_64) += camellia-aesni-avx-x86_64.o
+camellia-aesni-avx-x86_64-y := camellia-aesni-avx-asm_64.o camellia_aesni_avx_glue.o
+obj-$(CONFIG_CRYPTO_CAMELLIA_AESNI_AVX2_X86_64) += camellia-aesni-avx2.o
+camellia-aesni-avx2-y := camellia-aesni-avx2-asm_64.o camellia_aesni_avx2_glue.o
+
+obj-$(CONFIG_CRYPTO_BLOWFISH_X86_64) += blowfish-x86_64.o
+blowfish-x86_64-y := blowfish-x86_64-asm_64.o blowfish_glue.o
+
+obj-$(CONFIG_CRYPTO_CAST5_AVX_X86_64) += cast5-avx-x86_64.o
+cast5-avx-x86_64-y := cast5-avx-x86_64-asm_64.o cast5_avx_glue.o
+
+obj-$(CONFIG_CRYPTO_CAST6_AVX_X86_64) += cast6-avx-x86_64.o
+cast6-avx-x86_64-y := cast6-avx-x86_64-asm_64.o cast6_avx_glue.o
+
+obj-$(CONFIG_CRYPTO_AEGIS128_AESNI_SSE2) += aegis128-aesni.o
+aegis128-aesni-y := aegis128-aesni-asm.o aegis128-aesni-glue.o
+
+obj-$(CONFIG_CRYPTO_AES_NI_INTEL) += aesni-intel.o
+aesni-intel-y := aesni-intel_asm.o aesni-intel_glue.o
+aesni-intel-$(CONFIG_64BIT) += aes-ctr-avx-x86_64.o \
+			       aes-gcm-aesni-x86_64.o \
+			       aes-gcm-vaes-avx2.o \
+			       aes-gcm-vaes-avx512.o \
+			       aes-xts-avx-x86_64.o
+
+obj-$(CONFIG_CRYPTO_SM4_AESNI_AVX_X86_64) += sm4-aesni-avx-x86_64.o
+sm4-aesni-avx-x86_64-y := sm4-aesni-avx-asm_64.o sm4_aesni_avx_glue.o
+
+obj-$(CONFIG_CRYPTO_SM4_AESNI_AVX2_X86_64) += sm4-aesni-avx2-x86_64.o
+sm4-aesni-avx2-x86_64-y := sm4-aesni-avx2-asm_64.o sm4_aesni_avx2_glue.o
+
+obj-$(CONFIG_CRYPTO_ARIA_AESNI_AVX_X86_64) += aria-aesni-avx-x86_64.o
+aria-aesni-avx-x86_64-y := aria-aesni-avx-asm_64.o aria_aesni_avx_glue.o
+
+obj-$(CONFIG_CRYPTO_ARIA_AESNI_AVX2_X86_64) += aria-aesni-avx2-x86_64.o
+aria-aesni-avx2-x86_64-y := aria-aesni-avx2-asm_64.o aria_aesni_avx2_glue.o
+
+obj-$(CONFIG_CRYPTO_ARIA_GFNI_AVX512_X86_64) += aria-gfni-avx512-x86_64.o
+aria-gfni-avx512-x86_64-y := aria-gfni-avx512-asm_64.o aria_gfni_avx512_glue.o
diff --git a/arch/x86/crypto/aegis128-aesni-asm.S b/arch/x86/crypto/aegis128-aesni-asm.S
new file mode 100644
index 000000000..7294dc0ee
--- /dev/null
+++ b/arch/x86/crypto/aegis128-aesni-asm.S
@@ -0,0 +1,602 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * AES-NI + SSE4.1 implementation of AEGIS-128
+ *
+ * Copyright (c) 2017-2018 Ondrej Mosnacek <omosnacek@gmail.com>
+ * Copyright (C) 2017-2018 Red Hat, Inc. All rights reserved.
+ * Copyright 2024 Google LLC
+ */
+
+#include <linux/linkage.h>
+
+#define STATE0	%xmm0
+#define STATE1	%xmm1
+#define STATE2	%xmm2
+#define STATE3	%xmm3
+#define STATE4	%xmm4
+#define KEY	%xmm5
+#define MSG	%xmm5
+#define T0	%xmm6
+#define T1	%xmm7
+
+.section .rodata.cst16.aegis128_const, "aM", @progbits, 32
+.align 16
+.Laegis128_const_0:
+	.byte 0x00, 0x01, 0x01, 0x02, 0x03, 0x05, 0x08, 0x0d
+	.byte 0x15, 0x22, 0x37, 0x59, 0x90, 0xe9, 0x79, 0x62
+.Laegis128_const_1:
+	.byte 0xdb, 0x3d, 0x18, 0x55, 0x6d, 0xc2, 0x2f, 0xf1
+	.byte 0x20, 0x11, 0x31, 0x42, 0x73, 0xb5, 0x28, 0xdd
+
+.section .rodata.cst32.zeropad_mask, "aM", @progbits, 32
+.align 32
+.Lzeropad_mask:
+	.octa 0xffffffffffffffffffffffffffffffff
+	.octa 0
+
+.text
+
+/*
+ * aegis128_update
+ * input:
+ *   STATE[0-4] - input state
+ * output:
+ *   STATE[0-4] - output state (shifted positions)
+ * changed:
+ *   T0
+ */
+.macro aegis128_update
+	movdqa STATE4, T0
+	aesenc STATE0, STATE4
+	aesenc STATE1, STATE0
+	aesenc STATE2, STATE1
+	aesenc STATE3, STATE2
+	aesenc T0,     STATE3
+.endm
+
+/*
+ * Load 1 <= LEN (%ecx) <= 15 bytes from the pointer SRC into the xmm register
+ * MSG and zeroize any remaining bytes.  Clobbers %rax, %rcx, and %r8.
+ */
+.macro load_partial
+	sub $8, %ecx			/* LEN - 8 */
+	jle .Lle8\@
+
+	/* Load 9 <= LEN <= 15 bytes: */
+	movq (SRC), MSG			/* Load first 8 bytes */
+	mov (SRC, %rcx), %rax		/* Load last 8 bytes */
+	neg %ecx
+	shl $3, %ecx
+	shr %cl, %rax			/* Discard overlapping bytes */
+	pinsrq $1, %rax, MSG
+	jmp .Ldone\@
+
+.Lle8\@:
+	add $4, %ecx			/* LEN - 4 */
+	jl .Llt4\@
+
+	/* Load 4 <= LEN <= 8 bytes: */
+	mov (SRC), %eax			/* Load first 4 bytes */
+	mov (SRC, %rcx), %r8d		/* Load last 4 bytes */
+	jmp .Lcombine\@
+
+.Llt4\@:
+	/* Load 1 <= LEN <= 3 bytes: */
+	add $2, %ecx			/* LEN - 2 */
+	movzbl (SRC), %eax		/* Load first byte */
+	jl .Lmovq\@
+	movzwl (SRC, %rcx), %r8d	/* Load last 2 bytes */
+.Lcombine\@:
+	shl $3, %ecx
+	shl %cl, %r8
+	or %r8, %rax			/* Combine the two parts */
+.Lmovq\@:
+	movq %rax, MSG
+.Ldone\@:
+.endm
+
+/*
+ * Store 1 <= LEN (%ecx) <= 15 bytes from the xmm register \msg to the pointer
+ * DST.  Clobbers %rax, %rcx, and %r8.
+ */
+.macro store_partial msg
+	sub $8, %ecx			/* LEN - 8 */
+	jl .Llt8\@
+
+	/* Store 8 <= LEN <= 15 bytes: */
+	pextrq $1, \msg, %rax
+	mov %ecx, %r8d
+	shl $3, %ecx
+	ror %cl, %rax
+	mov %rax, (DST, %r8)		/* Store last LEN - 8 bytes */
+	movq \msg, (DST)		/* Store first 8 bytes */
+	jmp .Ldone\@
+
+.Llt8\@:
+	add $4, %ecx			/* LEN - 4 */
+	jl .Llt4\@
+
+	/* Store 4 <= LEN <= 7 bytes: */
+	pextrd $1, \msg, %eax
+	mov %ecx, %r8d
+	shl $3, %ecx
+	ror %cl, %eax
+	mov %eax, (DST, %r8)		/* Store last LEN - 4 bytes */
+	movd \msg, (DST)		/* Store first 4 bytes */
+	jmp .Ldone\@
+
+.Llt4\@:
+	/* Store 1 <= LEN <= 3 bytes: */
+	pextrb $0, \msg, 0(DST)
+	cmp $-2, %ecx			/* LEN - 4 == -2, i.e. LEN == 2? */
+	jl .Ldone\@
+	pextrb $1, \msg, 1(DST)
+	je .Ldone\@
+	pextrb $2, \msg, 2(DST)
+.Ldone\@:
+.endm
+
+/*
+ * void aegis128_aesni_init(struct aegis_state *state,
+ *			    const struct aegis_block *key,
+ *			    const u8 iv[AEGIS128_NONCE_SIZE]);
+ */
+SYM_FUNC_START(aegis128_aesni_init)
+	.set STATEP, %rdi
+	.set KEYP, %rsi
+	.set IVP, %rdx
+
+	/* load IV: */
+	movdqu (IVP), T1
+
+	/* load key: */
+	movdqa (KEYP), KEY
+	pxor KEY, T1
+	movdqa T1, STATE0
+	movdqa KEY, STATE3
+	movdqa KEY, STATE4
+
+	/* load the constants: */
+	movdqa .Laegis128_const_0(%rip), STATE2
+	movdqa .Laegis128_const_1(%rip), STATE1
+	pxor STATE2, STATE3
+	pxor STATE1, STATE4
+
+	/* update 10 times with KEY / KEY xor IV: */
+	aegis128_update; pxor KEY, STATE4
+	aegis128_update; pxor T1,  STATE3
+	aegis128_update; pxor KEY, STATE2
+	aegis128_update; pxor T1,  STATE1
+	aegis128_update; pxor KEY, STATE0
+	aegis128_update; pxor T1,  STATE4
+	aegis128_update; pxor KEY, STATE3
+	aegis128_update; pxor T1,  STATE2
+	aegis128_update; pxor KEY, STATE1
+	aegis128_update; pxor T1,  STATE0
+
+	/* store the state: */
+	movdqu STATE0, 0x00(STATEP)
+	movdqu STATE1, 0x10(STATEP)
+	movdqu STATE2, 0x20(STATEP)
+	movdqu STATE3, 0x30(STATEP)
+	movdqu STATE4, 0x40(STATEP)
+	RET
+SYM_FUNC_END(aegis128_aesni_init)
+
+/*
+ * void aegis128_aesni_ad(struct aegis_state *state, const u8 *data,
+ *			  unsigned int len);
+ *
+ * len must be a multiple of 16.
+ */
+SYM_FUNC_START(aegis128_aesni_ad)
+	.set STATEP, %rdi
+	.set SRC, %rsi
+	.set LEN, %edx
+
+	test LEN, LEN
+	jz .Lad_out
+
+	/* load the state: */
+	movdqu 0x00(STATEP), STATE0
+	movdqu 0x10(STATEP), STATE1
+	movdqu 0x20(STATEP), STATE2
+	movdqu 0x30(STATEP), STATE3
+	movdqu 0x40(STATEP), STATE4
+
+.align 8
+.Lad_loop:
+	movdqu 0x00(SRC), MSG
+	aegis128_update
+	pxor MSG, STATE4
+	sub $0x10, LEN
+	jz .Lad_out_1
+
+	movdqu 0x10(SRC), MSG
+	aegis128_update
+	pxor MSG, STATE3
+	sub $0x10, LEN
+	jz .Lad_out_2
+
+	movdqu 0x20(SRC), MSG
+	aegis128_update
+	pxor MSG, STATE2
+	sub $0x10, LEN
+	jz .Lad_out_3
+
+	movdqu 0x30(SRC), MSG
+	aegis128_update
+	pxor MSG, STATE1
+	sub $0x10, LEN
+	jz .Lad_out_4
+
+	movdqu 0x40(SRC), MSG
+	aegis128_update
+	pxor MSG, STATE0
+	sub $0x10, LEN
+	jz .Lad_out_0
+
+	add $0x50, SRC
+	jmp .Lad_loop
+
+	/* store the state: */
+.Lad_out_0:
+	movdqu STATE0, 0x00(STATEP)
+	movdqu STATE1, 0x10(STATEP)
+	movdqu STATE2, 0x20(STATEP)
+	movdqu STATE3, 0x30(STATEP)
+	movdqu STATE4, 0x40(STATEP)
+	RET
+
+.Lad_out_1:
+	movdqu STATE4, 0x00(STATEP)
+	movdqu STATE0, 0x10(STATEP)
+	movdqu STATE1, 0x20(STATEP)
+	movdqu STATE2, 0x30(STATEP)
+	movdqu STATE3, 0x40(STATEP)
+	RET
+
+.Lad_out_2:
+	movdqu STATE3, 0x00(STATEP)
+	movdqu STATE4, 0x10(STATEP)
+	movdqu STATE0, 0x20(STATEP)
+	movdqu STATE1, 0x30(STATEP)
+	movdqu STATE2, 0x40(STATEP)
+	RET
+
+.Lad_out_3:
+	movdqu STATE2, 0x00(STATEP)
+	movdqu STATE3, 0x10(STATEP)
+	movdqu STATE4, 0x20(STATEP)
+	movdqu STATE0, 0x30(STATEP)
+	movdqu STATE1, 0x40(STATEP)
+	RET
+
+.Lad_out_4:
+	movdqu STATE1, 0x00(STATEP)
+	movdqu STATE2, 0x10(STATEP)
+	movdqu STATE3, 0x20(STATEP)
+	movdqu STATE4, 0x30(STATEP)
+	movdqu STATE0, 0x40(STATEP)
+.Lad_out:
+	RET
+SYM_FUNC_END(aegis128_aesni_ad)
+
+.macro encrypt_block s0 s1 s2 s3 s4 i
+	movdqu (\i * 0x10)(SRC), MSG
+	movdqa MSG, T0
+	pxor \s1, T0
+	pxor \s4, T0
+	movdqa \s2, T1
+	pand \s3, T1
+	pxor T1, T0
+	movdqu T0, (\i * 0x10)(DST)
+
+	aegis128_update
+	pxor MSG, \s4
+
+	sub $0x10, LEN
+	jz .Lenc_out_\i
+.endm
+
+/*
+ * void aegis128_aesni_enc(struct aegis_state *state, const u8 *src, u8 *dst,
+ *			   unsigned int len);
+ *
+ * len must be nonzero and a multiple of 16.
+ */
+SYM_FUNC_START(aegis128_aesni_enc)
+	.set STATEP, %rdi
+	.set SRC, %rsi
+	.set DST, %rdx
+	.set LEN, %ecx
+
+	/* load the state: */
+	movdqu 0x00(STATEP), STATE0
+	movdqu 0x10(STATEP), STATE1
+	movdqu 0x20(STATEP), STATE2
+	movdqu 0x30(STATEP), STATE3
+	movdqu 0x40(STATEP), STATE4
+
+.align 8
+.Lenc_loop:
+	encrypt_block STATE0 STATE1 STATE2 STATE3 STATE4 0
+	encrypt_block STATE4 STATE0 STATE1 STATE2 STATE3 1
+	encrypt_block STATE3 STATE4 STATE0 STATE1 STATE2 2
+	encrypt_block STATE2 STATE3 STATE4 STATE0 STATE1 3
+	encrypt_block STATE1 STATE2 STATE3 STATE4 STATE0 4
+
+	add $0x50, SRC
+	add $0x50, DST
+	jmp .Lenc_loop
+
+	/* store the state: */
+.Lenc_out_0:
+	movdqu STATE4, 0x00(STATEP)
+	movdqu STATE0, 0x10(STATEP)
+	movdqu STATE1, 0x20(STATEP)
+	movdqu STATE2, 0x30(STATEP)
+	movdqu STATE3, 0x40(STATEP)
+	RET
+
+.Lenc_out_1:
+	movdqu STATE3, 0x00(STATEP)
+	movdqu STATE4, 0x10(STATEP)
+	movdqu STATE0, 0x20(STATEP)
+	movdqu STATE1, 0x30(STATEP)
+	movdqu STATE2, 0x40(STATEP)
+	RET
+
+.Lenc_out_2:
+	movdqu STATE2, 0x00(STATEP)
+	movdqu STATE3, 0x10(STATEP)
+	movdqu STATE4, 0x20(STATEP)
+	movdqu STATE0, 0x30(STATEP)
+	movdqu STATE1, 0x40(STATEP)
+	RET
+
+.Lenc_out_3:
+	movdqu STATE1, 0x00(STATEP)
+	movdqu STATE2, 0x10(STATEP)
+	movdqu STATE3, 0x20(STATEP)
+	movdqu STATE4, 0x30(STATEP)
+	movdqu STATE0, 0x40(STATEP)
+	RET
+
+.Lenc_out_4:
+	movdqu STATE0, 0x00(STATEP)
+	movdqu STATE1, 0x10(STATEP)
+	movdqu STATE2, 0x20(STATEP)
+	movdqu STATE3, 0x30(STATEP)
+	movdqu STATE4, 0x40(STATEP)
+.Lenc_out:
+	RET
+SYM_FUNC_END(aegis128_aesni_enc)
+
+/*
+ * void aegis128_aesni_enc_tail(struct aegis_state *state, const u8 *src,
+ *				u8 *dst, unsigned int len);
+ */
+SYM_FUNC_START(aegis128_aesni_enc_tail)
+	.set STATEP, %rdi
+	.set SRC, %rsi
+	.set DST, %rdx
+	.set LEN, %ecx	/* {load,store}_partial rely on this being %ecx */
+
+	/* load the state: */
+	movdqu 0x00(STATEP), STATE0
+	movdqu 0x10(STATEP), STATE1
+	movdqu 0x20(STATEP), STATE2
+	movdqu 0x30(STATEP), STATE3
+	movdqu 0x40(STATEP), STATE4
+
+	/* encrypt message: */
+	mov LEN, %r9d
+	load_partial
+
+	movdqa MSG, T0
+	pxor STATE1, T0
+	pxor STATE4, T0
+	movdqa STATE2, T1
+	pand STATE3, T1
+	pxor T1, T0
+
+	mov %r9d, LEN
+	store_partial T0
+
+	aegis128_update
+	pxor MSG, STATE4
+
+	/* store the state: */
+	movdqu STATE4, 0x00(STATEP)
+	movdqu STATE0, 0x10(STATEP)
+	movdqu STATE1, 0x20(STATEP)
+	movdqu STATE2, 0x30(STATEP)
+	movdqu STATE3, 0x40(STATEP)
+	RET
+SYM_FUNC_END(aegis128_aesni_enc_tail)
+
+.macro decrypt_block s0 s1 s2 s3 s4 i
+	movdqu (\i * 0x10)(SRC), MSG
+	pxor \s1, MSG
+	pxor \s4, MSG
+	movdqa \s2, T1
+	pand \s3, T1
+	pxor T1, MSG
+	movdqu MSG, (\i * 0x10)(DST)
+
+	aegis128_update
+	pxor MSG, \s4
+
+	sub $0x10, LEN
+	jz .Ldec_out_\i
+.endm
+
+/*
+ * void aegis128_aesni_dec(struct aegis_state *state, const u8 *src, u8 *dst,
+ *			   unsigned int len);
+ *
+ * len must be nonzero and a multiple of 16.
+ */
+SYM_FUNC_START(aegis128_aesni_dec)
+	.set STATEP, %rdi
+	.set SRC, %rsi
+	.set DST, %rdx
+	.set LEN, %ecx
+
+	/* load the state: */
+	movdqu 0x00(STATEP), STATE0
+	movdqu 0x10(STATEP), STATE1
+	movdqu 0x20(STATEP), STATE2
+	movdqu 0x30(STATEP), STATE3
+	movdqu 0x40(STATEP), STATE4
+
+.align 8
+.Ldec_loop:
+	decrypt_block STATE0 STATE1 STATE2 STATE3 STATE4 0
+	decrypt_block STATE4 STATE0 STATE1 STATE2 STATE3 1
+	decrypt_block STATE3 STATE4 STATE0 STATE1 STATE2 2
+	decrypt_block STATE2 STATE3 STATE4 STATE0 STATE1 3
+	decrypt_block STATE1 STATE2 STATE3 STATE4 STATE0 4
+
+	add $0x50, SRC
+	add $0x50, DST
+	jmp .Ldec_loop
+
+	/* store the state: */
+.Ldec_out_0:
+	movdqu STATE4, 0x00(STATEP)
+	movdqu STATE0, 0x10(STATEP)
+	movdqu STATE1, 0x20(STATEP)
+	movdqu STATE2, 0x30(STATEP)
+	movdqu STATE3, 0x40(STATEP)
+	RET
+
+.Ldec_out_1:
+	movdqu STATE3, 0x00(STATEP)
+	movdqu STATE4, 0x10(STATEP)
+	movdqu STATE0, 0x20(STATEP)
+	movdqu STATE1, 0x30(STATEP)
+	movdqu STATE2, 0x40(STATEP)
+	RET
+
+.Ldec_out_2:
+	movdqu STATE2, 0x00(STATEP)
+	movdqu STATE3, 0x10(STATEP)
+	movdqu STATE4, 0x20(STATEP)
+	movdqu STATE0, 0x30(STATEP)
+	movdqu STATE1, 0x40(STATEP)
+	RET
+
+.Ldec_out_3:
+	movdqu STATE1, 0x00(STATEP)
+	movdqu STATE2, 0x10(STATEP)
+	movdqu STATE3, 0x20(STATEP)
+	movdqu STATE4, 0x30(STATEP)
+	movdqu STATE0, 0x40(STATEP)
+	RET
+
+.Ldec_out_4:
+	movdqu STATE0, 0x00(STATEP)
+	movdqu STATE1, 0x10(STATEP)
+	movdqu STATE2, 0x20(STATEP)
+	movdqu STATE3, 0x30(STATEP)
+	movdqu STATE4, 0x40(STATEP)
+.Ldec_out:
+	RET
+SYM_FUNC_END(aegis128_aesni_dec)
+
+/*
+ * void aegis128_aesni_dec_tail(struct aegis_state *state, const u8 *src,
+ *				u8 *dst, unsigned int len);
+ */
+SYM_FUNC_START(aegis128_aesni_dec_tail)
+	.set STATEP, %rdi
+	.set SRC, %rsi
+	.set DST, %rdx
+	.set LEN, %ecx	/* {load,store}_partial rely on this being %ecx */
+
+	/* load the state: */
+	movdqu 0x00(STATEP), STATE0
+	movdqu 0x10(STATEP), STATE1
+	movdqu 0x20(STATEP), STATE2
+	movdqu 0x30(STATEP), STATE3
+	movdqu 0x40(STATEP), STATE4
+
+	/* decrypt message: */
+	mov LEN, %r9d
+	load_partial
+
+	pxor STATE1, MSG
+	pxor STATE4, MSG
+	movdqa STATE2, T1
+	pand STATE3, T1
+	pxor T1, MSG
+
+	mov %r9d, LEN
+	store_partial MSG
+
+	/* mask with byte count: */
+	lea .Lzeropad_mask+16(%rip), %rax
+	sub %r9, %rax
+	movdqu (%rax), T0
+	pand T0, MSG
+
+	aegis128_update
+	pxor MSG, STATE4
+
+	/* store the state: */
+	movdqu STATE4, 0x00(STATEP)
+	movdqu STATE0, 0x10(STATEP)
+	movdqu STATE1, 0x20(STATEP)
+	movdqu STATE2, 0x30(STATEP)
+	movdqu STATE3, 0x40(STATEP)
+	RET
+SYM_FUNC_END(aegis128_aesni_dec_tail)
+
+/*
+ * void aegis128_aesni_final(struct aegis_state *state,
+ *			     struct aegis_block *tag_xor,
+ *			     unsigned int assoclen, unsigned int cryptlen);
+ */
+SYM_FUNC_START(aegis128_aesni_final)
+	.set STATEP, %rdi
+	.set TAG_XOR, %rsi
+	.set ASSOCLEN, %edx
+	.set CRYPTLEN, %ecx
+
+	/* load the state: */
+	movdqu 0x00(STATEP), STATE0
+	movdqu 0x10(STATEP), STATE1
+	movdqu 0x20(STATEP), STATE2
+	movdqu 0x30(STATEP), STATE3
+	movdqu 0x40(STATEP), STATE4
+
+	/* prepare length block: */
+	movd ASSOCLEN, MSG
+	pinsrd $2, CRYPTLEN, MSG
+	psllq $3, MSG /* multiply by 8 (to get bit count) */
+
+	pxor STATE3, MSG
+
+	/* update state: */
+	aegis128_update; pxor MSG, STATE4
+	aegis128_update; pxor MSG, STATE3
+	aegis128_update; pxor MSG, STATE2
+	aegis128_update; pxor MSG, STATE1
+	aegis128_update; pxor MSG, STATE0
+	aegis128_update; pxor MSG, STATE4
+	aegis128_update; pxor MSG, STATE3
+
+	/* xor tag: */
+	movdqu (TAG_XOR), MSG
+
+	pxor STATE0, MSG
+	pxor STATE1, MSG
+	pxor STATE2, MSG
+	pxor STATE3, MSG
+	pxor STATE4, MSG
+
+	movdqu MSG, (TAG_XOR)
+	RET
+SYM_FUNC_END(aegis128_aesni_final)
diff --git a/arch/x86/crypto/aegis128-aesni-glue.c b/arch/x86/crypto/aegis128-aesni-glue.c
new file mode 100644
index 000000000..f1adfba1a
--- /dev/null
+++ b/arch/x86/crypto/aegis128-aesni-glue.c
@@ -0,0 +1,287 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * The AEGIS-128 Authenticated-Encryption Algorithm
+ *   Glue for AES-NI + SSE4.1 implementation
+ *
+ * Copyright (c) 2017-2018 Ondrej Mosnacek <omosnacek@gmail.com>
+ * Copyright (C) 2017-2018 Red Hat, Inc. All rights reserved.
+ */
+
+#include <crypto/internal/aead.h>
+#include <crypto/internal/skcipher.h>
+#include <crypto/scatterwalk.h>
+#include <linux/module.h>
+#include <asm/fpu/api.h>
+#include <asm/cpu_device_id.h>
+
+#define AEGIS128_BLOCK_ALIGN 16
+#define AEGIS128_BLOCK_SIZE 16
+#define AEGIS128_NONCE_SIZE 16
+#define AEGIS128_STATE_BLOCKS 5
+#define AEGIS128_KEY_SIZE 16
+#define AEGIS128_MIN_AUTH_SIZE 8
+#define AEGIS128_MAX_AUTH_SIZE 16
+
+struct aegis_block {
+	u8 bytes[AEGIS128_BLOCK_SIZE] __aligned(AEGIS128_BLOCK_ALIGN);
+};
+
+struct aegis_state {
+	struct aegis_block blocks[AEGIS128_STATE_BLOCKS];
+};
+
+struct aegis_ctx {
+	struct aegis_block key;
+};
+
+asmlinkage void aegis128_aesni_init(struct aegis_state *state,
+				    const struct aegis_block *key,
+				    const u8 iv[AEGIS128_NONCE_SIZE]);
+
+asmlinkage void aegis128_aesni_ad(struct aegis_state *state, const u8 *data,
+				  unsigned int len);
+
+asmlinkage void aegis128_aesni_enc(struct aegis_state *state, const u8 *src,
+				   u8 *dst, unsigned int len);
+
+asmlinkage void aegis128_aesni_dec(struct aegis_state *state, const u8 *src,
+				   u8 *dst, unsigned int len);
+
+asmlinkage void aegis128_aesni_enc_tail(struct aegis_state *state,
+					const u8 *src, u8 *dst,
+					unsigned int len);
+
+asmlinkage void aegis128_aesni_dec_tail(struct aegis_state *state,
+					const u8 *src, u8 *dst,
+					unsigned int len);
+
+asmlinkage void aegis128_aesni_final(struct aegis_state *state,
+				     struct aegis_block *tag_xor,
+				     unsigned int assoclen,
+				     unsigned int cryptlen);
+
+static void crypto_aegis128_aesni_process_ad(
+		struct aegis_state *state, struct scatterlist *sg_src,
+		unsigned int assoclen)
+{
+	struct scatter_walk walk;
+	struct aegis_block buf;
+	unsigned int pos = 0;
+
+	scatterwalk_start(&walk, sg_src);
+	while (assoclen != 0) {
+		unsigned int size = scatterwalk_next(&walk, assoclen);
+		const u8 *src = walk.addr;
+		unsigned int left = size;
+
+		if (pos + size >= AEGIS128_BLOCK_SIZE) {
+			if (pos > 0) {
+				unsigned int fill = AEGIS128_BLOCK_SIZE - pos;
+				memcpy(buf.bytes + pos, src, fill);
+				aegis128_aesni_ad(state, buf.bytes,
+						  AEGIS128_BLOCK_SIZE);
+				pos = 0;
+				left -= fill;
+				src += fill;
+			}
+
+			aegis128_aesni_ad(state, src,
+					  left & ~(AEGIS128_BLOCK_SIZE - 1));
+			src += left & ~(AEGIS128_BLOCK_SIZE - 1);
+			left &= AEGIS128_BLOCK_SIZE - 1;
+		}
+
+		memcpy(buf.bytes + pos, src, left);
+		pos += left;
+		assoclen -= size;
+
+		scatterwalk_done_src(&walk, size);
+	}
+
+	if (pos > 0) {
+		memset(buf.bytes + pos, 0, AEGIS128_BLOCK_SIZE - pos);
+		aegis128_aesni_ad(state, buf.bytes, AEGIS128_BLOCK_SIZE);
+	}
+}
+
+static __always_inline int
+crypto_aegis128_aesni_process_crypt(struct aegis_state *state,
+				    struct skcipher_walk *walk, bool enc)
+{
+	int err = 0;
+
+	while (walk->nbytes >= AEGIS128_BLOCK_SIZE) {
+		if (enc)
+			aegis128_aesni_enc(state, walk->src.virt.addr,
+					   walk->dst.virt.addr,
+					   round_down(walk->nbytes,
+						      AEGIS128_BLOCK_SIZE));
+		else
+			aegis128_aesni_dec(state, walk->src.virt.addr,
+					   walk->dst.virt.addr,
+					   round_down(walk->nbytes,
+						      AEGIS128_BLOCK_SIZE));
+		kernel_fpu_end();
+		err = skcipher_walk_done(walk,
+					 walk->nbytes % AEGIS128_BLOCK_SIZE);
+		kernel_fpu_begin();
+	}
+
+	if (walk->nbytes) {
+		if (enc)
+			aegis128_aesni_enc_tail(state, walk->src.virt.addr,
+						walk->dst.virt.addr,
+						walk->nbytes);
+		else
+			aegis128_aesni_dec_tail(state, walk->src.virt.addr,
+						walk->dst.virt.addr,
+						walk->nbytes);
+		kernel_fpu_end();
+		err = skcipher_walk_done(walk, 0);
+		kernel_fpu_begin();
+	}
+	return err;
+}
+
+static struct aegis_ctx *crypto_aegis128_aesni_ctx(struct crypto_aead *aead)
+{
+	u8 *ctx = crypto_aead_ctx(aead);
+	ctx = PTR_ALIGN(ctx, __alignof__(struct aegis_ctx));
+	return (void *)ctx;
+}
+
+static int crypto_aegis128_aesni_setkey(struct crypto_aead *aead, const u8 *key,
+					unsigned int keylen)
+{
+	struct aegis_ctx *ctx = crypto_aegis128_aesni_ctx(aead);
+
+	if (keylen != AEGIS128_KEY_SIZE)
+		return -EINVAL;
+
+	memcpy(ctx->key.bytes, key, AEGIS128_KEY_SIZE);
+
+	return 0;
+}
+
+static int crypto_aegis128_aesni_setauthsize(struct crypto_aead *tfm,
+						unsigned int authsize)
+{
+	if (authsize > AEGIS128_MAX_AUTH_SIZE)
+		return -EINVAL;
+	if (authsize < AEGIS128_MIN_AUTH_SIZE)
+		return -EINVAL;
+	return 0;
+}
+
+static __always_inline int
+crypto_aegis128_aesni_crypt(struct aead_request *req,
+			    struct aegis_block *tag_xor,
+			    unsigned int cryptlen, bool enc)
+{
+	struct crypto_aead *tfm = crypto_aead_reqtfm(req);
+	struct aegis_ctx *ctx = crypto_aegis128_aesni_ctx(tfm);
+	struct skcipher_walk walk;
+	struct aegis_state state;
+	int err;
+
+	if (enc)
+		err = skcipher_walk_aead_encrypt(&walk, req, false);
+	else
+		err = skcipher_walk_aead_decrypt(&walk, req, false);
+	if (err)
+		return err;
+
+	kernel_fpu_begin();
+
+	aegis128_aesni_init(&state, &ctx->key, req->iv);
+	crypto_aegis128_aesni_process_ad(&state, req->src, req->assoclen);
+	err = crypto_aegis128_aesni_process_crypt(&state, &walk, enc);
+	if (err == 0)
+		aegis128_aesni_final(&state, tag_xor, req->assoclen, cryptlen);
+	kernel_fpu_end();
+	return err;
+}
+
+static int crypto_aegis128_aesni_encrypt(struct aead_request *req)
+{
+	struct crypto_aead *tfm = crypto_aead_reqtfm(req);
+	struct aegis_block tag = {};
+	unsigned int authsize = crypto_aead_authsize(tfm);
+	unsigned int cryptlen = req->cryptlen;
+	int err;
+
+	err = crypto_aegis128_aesni_crypt(req, &tag, cryptlen, true);
+	if (err)
+		return err;
+
+	scatterwalk_map_and_copy(tag.bytes, req->dst,
+				 req->assoclen + cryptlen, authsize, 1);
+	return 0;
+}
+
+static int crypto_aegis128_aesni_decrypt(struct aead_request *req)
+{
+	static const struct aegis_block zeros = {};
+
+	struct crypto_aead *tfm = crypto_aead_reqtfm(req);
+	struct aegis_block tag;
+	unsigned int authsize = crypto_aead_authsize(tfm);
+	unsigned int cryptlen = req->cryptlen - authsize;
+	int err;
+
+	scatterwalk_map_and_copy(tag.bytes, req->src,
+				 req->assoclen + cryptlen, authsize, 0);
+
+	err = crypto_aegis128_aesni_crypt(req, &tag, cryptlen, false);
+	if (err)
+		return err;
+
+	return crypto_memneq(tag.bytes, zeros.bytes, authsize) ? -EBADMSG : 0;
+}
+
+static struct aead_alg crypto_aegis128_aesni_alg = {
+	.setkey = crypto_aegis128_aesni_setkey,
+	.setauthsize = crypto_aegis128_aesni_setauthsize,
+	.encrypt = crypto_aegis128_aesni_encrypt,
+	.decrypt = crypto_aegis128_aesni_decrypt,
+
+	.ivsize = AEGIS128_NONCE_SIZE,
+	.maxauthsize = AEGIS128_MAX_AUTH_SIZE,
+	.chunksize = AEGIS128_BLOCK_SIZE,
+
+	.base = {
+		.cra_blocksize = 1,
+		.cra_ctxsize = sizeof(struct aegis_ctx) +
+			       __alignof__(struct aegis_ctx),
+		.cra_priority = 400,
+
+		.cra_name = "aegis128",
+		.cra_driver_name = "aegis128-aesni",
+
+		.cra_module = THIS_MODULE,
+	}
+};
+
+static int __init crypto_aegis128_aesni_module_init(void)
+{
+	if (!boot_cpu_has(X86_FEATURE_XMM4_1) ||
+	    !boot_cpu_has(X86_FEATURE_AES) ||
+	    !cpu_has_xfeatures(XFEATURE_MASK_SSE, NULL))
+		return -ENODEV;
+
+	return crypto_register_aead(&crypto_aegis128_aesni_alg);
+}
+
+static void __exit crypto_aegis128_aesni_module_exit(void)
+{
+	crypto_unregister_aead(&crypto_aegis128_aesni_alg);
+}
+
+module_init(crypto_aegis128_aesni_module_init);
+module_exit(crypto_aegis128_aesni_module_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Ondrej Mosnacek <omosnacek@gmail.com>");
+MODULE_DESCRIPTION("AEGIS-128 AEAD algorithm -- AESNI+SSE4.1 implementation");
+MODULE_ALIAS_CRYPTO("aegis128");
+MODULE_ALIAS_CRYPTO("aegis128-aesni");
diff --git a/arch/x86/crypto/aes-ctr-avx-x86_64.S b/arch/x86/crypto/aes-ctr-avx-x86_64.S
new file mode 100644
index 000000000..2745918f6
--- /dev/null
+++ b/arch/x86/crypto/aes-ctr-avx-x86_64.S
@@ -0,0 +1,571 @@
+/* SPDX-License-Identifier: Apache-2.0 OR BSD-2-Clause */
+//
+// Copyright 2025 Google LLC
+//
+// Author: Eric Biggers <ebiggers@google.com>
+//
+// This file is dual-licensed, meaning that you can use it under your choice of
+// either of the following two licenses:
+//
+// Licensed under the Apache License 2.0 (the "License").  You may obtain a copy
+// of the License at
+//
+//	http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// or
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are met:
+//
+// 1. Redistributions of source code must retain the above copyright notice,
+//    this list of conditions and the following disclaimer.
+//
+// 2. Redistributions in binary form must reproduce the above copyright
+//    notice, this list of conditions and the following disclaimer in the
+//    documentation and/or other materials provided with the distribution.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+// POSSIBILITY OF SUCH DAMAGE.
+//
+//------------------------------------------------------------------------------
+//
+// This file contains x86_64 assembly implementations of AES-CTR and AES-XCTR
+// using the following sets of CPU features:
+//	- AES-NI && AVX
+//	- VAES && AVX2
+//	- VAES && AVX512BW && AVX512VL && BMI2
+//
+// See the function definitions at the bottom of the file for more information.
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+
+.section .rodata
+.p2align 4
+
+.Lbswap_mask:
+	.octa	0x000102030405060708090a0b0c0d0e0f
+
+.Lctr_pattern:
+	.quad	0, 0
+.Lone:
+	.quad	1, 0
+.Ltwo:
+	.quad	2, 0
+	.quad	3, 0
+
+.Lfour:
+	.quad	4, 0
+
+.text
+
+// Move a vector between memory and a register.
+.macro	_vmovdqu	src, dst
+.if VL < 64
+	vmovdqu		\src, \dst
+.else
+	vmovdqu8	\src, \dst
+.endif
+.endm
+
+// Move a vector between registers.
+.macro	_vmovdqa	src, dst
+.if VL < 64
+	vmovdqa		\src, \dst
+.else
+	vmovdqa64	\src, \dst
+.endif
+.endm
+
+// Broadcast a 128-bit value from memory to all 128-bit lanes of a vector
+// register.
+.macro	_vbroadcast128	src, dst
+.if VL == 16
+	vmovdqu		\src, \dst
+.elseif VL == 32
+	vbroadcasti128	\src, \dst
+.else
+	vbroadcasti32x4	\src, \dst
+.endif
+.endm
+
+// XOR two vectors together.
+.macro	_vpxor	src1, src2, dst
+.if VL < 64
+	vpxor		\src1, \src2, \dst
+.else
+	vpxord		\src1, \src2, \dst
+.endif
+.endm
+
+// Load 1 <= %ecx <= 15 bytes from the pointer \src into the xmm register \dst
+// and zeroize any remaining bytes.  Clobbers %rax, %rcx, and \tmp{64,32}.
+.macro	_load_partial_block	src, dst, tmp64, tmp32
+	sub		$8, %ecx		// LEN - 8
+	jle		.Lle8\@
+
+	// Load 9 <= LEN <= 15 bytes.
+	vmovq		(\src), \dst		// Load first 8 bytes
+	mov		(\src, %rcx), %rax	// Load last 8 bytes
+	neg		%ecx
+	shl		$3, %ecx
+	shr		%cl, %rax		// Discard overlapping bytes
+	vpinsrq		$1, %rax, \dst, \dst
+	jmp		.Ldone\@
+
+.Lle8\@:
+	add		$4, %ecx		// LEN - 4
+	jl		.Llt4\@
+
+	// Load 4 <= LEN <= 8 bytes.
+	mov		(\src), %eax		// Load first 4 bytes
+	mov		(\src, %rcx), \tmp32	// Load last 4 bytes
+	jmp		.Lcombine\@
+
+.Llt4\@:
+	// Load 1 <= LEN <= 3 bytes.
+	add		$2, %ecx		// LEN - 2
+	movzbl		(\src), %eax		// Load first byte
+	jl		.Lmovq\@
+	movzwl		(\src, %rcx), \tmp32	// Load last 2 bytes
+.Lcombine\@:
+	shl		$3, %ecx
+	shl		%cl, \tmp64
+	or		\tmp64, %rax		// Combine the two parts
+.Lmovq\@:
+	vmovq		%rax, \dst
+.Ldone\@:
+.endm
+
+// Store 1 <= %ecx <= 15 bytes from the xmm register \src to the pointer \dst.
+// Clobbers %rax, %rcx, and \tmp{64,32}.
+.macro	_store_partial_block	src, dst, tmp64, tmp32
+	sub		$8, %ecx		// LEN - 8
+	jl		.Llt8\@
+
+	// Store 8 <= LEN <= 15 bytes.
+	vpextrq		$1, \src, %rax
+	mov		%ecx, \tmp32
+	shl		$3, %ecx
+	ror		%cl, %rax
+	mov		%rax, (\dst, \tmp64)	// Store last LEN - 8 bytes
+	vmovq		\src, (\dst)		// Store first 8 bytes
+	jmp		.Ldone\@
+
+.Llt8\@:
+	add		$4, %ecx		// LEN - 4
+	jl		.Llt4\@
+
+	// Store 4 <= LEN <= 7 bytes.
+	vpextrd		$1, \src, %eax
+	mov		%ecx, \tmp32
+	shl		$3, %ecx
+	ror		%cl, %eax
+	mov		%eax, (\dst, \tmp64)	// Store last LEN - 4 bytes
+	vmovd		\src, (\dst)		// Store first 4 bytes
+	jmp		.Ldone\@
+
+.Llt4\@:
+	// Store 1 <= LEN <= 3 bytes.
+	vpextrb		$0, \src, 0(\dst)
+	cmp		$-2, %ecx		// LEN - 4 == -2, i.e. LEN == 2?
+	jl		.Ldone\@
+	vpextrb		$1, \src, 1(\dst)
+	je		.Ldone\@
+	vpextrb		$2, \src, 2(\dst)
+.Ldone\@:
+.endm
+
+// Prepare the next two vectors of AES inputs in AESDATA\i0 and AESDATA\i1, and
+// XOR each with the zero-th round key.  Also update LE_CTR if !\final.
+.macro	_prepare_2_ctr_vecs	is_xctr, i0, i1, final=0
+.if \is_xctr
+  .if USE_AVX512
+	vmovdqa64	LE_CTR, AESDATA\i0
+	vpternlogd	$0x96, XCTR_IV, RNDKEY0, AESDATA\i0
+  .else
+	vpxor		XCTR_IV, LE_CTR, AESDATA\i0
+	vpxor		RNDKEY0, AESDATA\i0, AESDATA\i0
+  .endif
+	vpaddq		LE_CTR_INC1, LE_CTR, AESDATA\i1
+
+  .if USE_AVX512
+	vpternlogd	$0x96, XCTR_IV, RNDKEY0, AESDATA\i1
+  .else
+	vpxor		XCTR_IV, AESDATA\i1, AESDATA\i1
+	vpxor		RNDKEY0, AESDATA\i1, AESDATA\i1
+  .endif
+.else
+	vpshufb		BSWAP_MASK, LE_CTR, AESDATA\i0
+	_vpxor		RNDKEY0, AESDATA\i0, AESDATA\i0
+	vpaddq		LE_CTR_INC1, LE_CTR, AESDATA\i1
+	vpshufb		BSWAP_MASK, AESDATA\i1, AESDATA\i1
+	_vpxor		RNDKEY0, AESDATA\i1, AESDATA\i1
+.endif
+.if !\final
+	vpaddq		LE_CTR_INC2, LE_CTR, LE_CTR
+.endif
+.endm
+
+// Do all AES rounds on the data in the given AESDATA vectors, excluding the
+// zero-th and last rounds.
+.macro	_aesenc_loop	vecs:vararg
+	mov		KEY, %rax
+1:
+	_vbroadcast128	(%rax), RNDKEY
+.irp i, \vecs
+	vaesenc		RNDKEY, AESDATA\i, AESDATA\i
+.endr
+	add		$16, %rax
+	cmp		%rax, RNDKEYLAST_PTR
+	jne		1b
+.endm
+
+// Finalize the keystream blocks in the given AESDATA vectors by doing the last
+// AES round, then XOR those keystream blocks with the corresponding data.
+// Reduce latency by doing the XOR before the vaesenclast, utilizing the
+// property vaesenclast(key, a) ^ b == vaesenclast(key ^ b, a).
+.macro	_aesenclast_and_xor	vecs:vararg
+.irp i, \vecs
+	_vpxor		\i*VL(SRC), RNDKEYLAST, RNDKEY
+	vaesenclast	RNDKEY, AESDATA\i, AESDATA\i
+.endr
+.irp i, \vecs
+	_vmovdqu	AESDATA\i, \i*VL(DST)
+.endr
+.endm
+
+// XOR the keystream blocks in the specified AESDATA vectors with the
+// corresponding data.
+.macro	_xor_data	vecs:vararg
+.irp i, \vecs
+	_vpxor		\i*VL(SRC), AESDATA\i, AESDATA\i
+.endr
+.irp i, \vecs
+	_vmovdqu	AESDATA\i, \i*VL(DST)
+.endr
+.endm
+
+.macro	_aes_ctr_crypt		is_xctr
+
+	// Define register aliases V0-V15 that map to the xmm, ymm, or zmm
+	// registers according to the selected Vector Length (VL).
+.irp i, 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
+  .if VL == 16
+	.set	V\i,		%xmm\i
+  .elseif VL == 32
+	.set	V\i,		%ymm\i
+  .elseif VL == 64
+	.set	V\i,		%zmm\i
+  .else
+	.error "Unsupported Vector Length (VL)"
+  .endif
+.endr
+
+	// Function arguments
+	.set	KEY,		%rdi	// Initially points to the start of the
+					// crypto_aes_ctx, then is advanced to
+					// point to the index 1 round key
+	.set	KEY32,		%edi	// Available as temp register after all
+					// keystream blocks have been generated
+	.set	SRC,		%rsi	// Pointer to next source data
+	.set	DST,		%rdx	// Pointer to next destination data
+	.set	LEN,		%ecx	// Remaining length in bytes.
+					// Note: _load_partial_block relies on
+					// this being in %ecx.
+	.set	LEN64,		%rcx	// Zero-extend LEN before using!
+	.set	LEN8,		%cl
+.if \is_xctr
+	.set	XCTR_IV_PTR,	%r8	// const u8 iv[AES_BLOCK_SIZE];
+	.set	XCTR_CTR,	%r9	// u64 ctr;
+.else
+	.set	LE_CTR_PTR,	%r8	// const u64 le_ctr[2];
+.endif
+
+	// Additional local variables
+	.set	RNDKEYLAST_PTR,	%r10
+	.set	AESDATA0,	V0
+	.set	AESDATA0_XMM,	%xmm0
+	.set	AESDATA1,	V1
+	.set	AESDATA1_XMM,	%xmm1
+	.set	AESDATA2,	V2
+	.set	AESDATA3,	V3
+	.set	AESDATA4,	V4
+	.set	AESDATA5,	V5
+	.set	AESDATA6,	V6
+	.set	AESDATA7,	V7
+.if \is_xctr
+	.set	XCTR_IV,	V8
+.else
+	.set	BSWAP_MASK,	V8
+.endif
+	.set	LE_CTR,		V9
+	.set	LE_CTR_XMM,	%xmm9
+	.set	LE_CTR_INC1,	V10
+	.set	LE_CTR_INC2,	V11
+	.set	RNDKEY0,	V12
+	.set	RNDKEYLAST,	V13
+	.set	RNDKEY,		V14
+
+	// Create the first vector of counters.
+.if \is_xctr
+  .if VL == 16
+	vmovq		XCTR_CTR, LE_CTR
+  .elseif VL == 32
+	vmovq		XCTR_CTR, LE_CTR_XMM
+	inc		XCTR_CTR
+	vmovq		XCTR_CTR, AESDATA0_XMM
+	vinserti128	$1, AESDATA0_XMM, LE_CTR, LE_CTR
+  .else
+	vpbroadcastq	XCTR_CTR, LE_CTR
+	vpsrldq		$8, LE_CTR, LE_CTR
+	vpaddq		.Lctr_pattern(%rip), LE_CTR, LE_CTR
+  .endif
+	_vbroadcast128	(XCTR_IV_PTR), XCTR_IV
+.else
+	_vbroadcast128	(LE_CTR_PTR), LE_CTR
+  .if VL > 16
+	vpaddq		.Lctr_pattern(%rip), LE_CTR, LE_CTR
+  .endif
+	_vbroadcast128	.Lbswap_mask(%rip), BSWAP_MASK
+.endif
+
+.if VL == 16
+	_vbroadcast128	.Lone(%rip), LE_CTR_INC1
+.elseif VL == 32
+	_vbroadcast128	.Ltwo(%rip), LE_CTR_INC1
+.else
+	_vbroadcast128	.Lfour(%rip), LE_CTR_INC1
+.endif
+	vpsllq		$1, LE_CTR_INC1, LE_CTR_INC2
+
+	// Load the AES key length: 16 (AES-128), 24 (AES-192), or 32 (AES-256).
+	movl		480(KEY), %eax
+
+	// Compute the pointer to the last round key.
+	lea		6*16(KEY, %rax, 4), RNDKEYLAST_PTR
+
+	// Load the zero-th and last round keys.
+	_vbroadcast128	(KEY), RNDKEY0
+	_vbroadcast128	(RNDKEYLAST_PTR), RNDKEYLAST
+
+	// Make KEY point to the first round key.
+	add		$16, KEY
+
+	// This is the main loop, which encrypts 8 vectors of data at a time.
+	add		$-8*VL, LEN
+	jl		.Lloop_8x_done\@
+.Lloop_8x\@:
+	_prepare_2_ctr_vecs	\is_xctr, 0, 1
+	_prepare_2_ctr_vecs	\is_xctr, 2, 3
+	_prepare_2_ctr_vecs	\is_xctr, 4, 5
+	_prepare_2_ctr_vecs	\is_xctr, 6, 7
+	_aesenc_loop	0,1,2,3,4,5,6,7
+	_aesenclast_and_xor 0,1,2,3,4,5,6,7
+	sub		$-8*VL, SRC
+	sub		$-8*VL, DST
+	add		$-8*VL, LEN
+	jge		.Lloop_8x\@
+.Lloop_8x_done\@:
+	sub		$-8*VL, LEN
+	jz		.Ldone\@
+
+	// 1 <= LEN < 8*VL.  Generate 2, 4, or 8 more vectors of keystream
+	// blocks, depending on the remaining LEN.
+
+	_prepare_2_ctr_vecs	\is_xctr, 0, 1
+	_prepare_2_ctr_vecs	\is_xctr, 2, 3
+	cmp		$4*VL, LEN
+	jle		.Lenc_tail_atmost4vecs\@
+
+	// 4*VL < LEN < 8*VL.  Generate 8 vectors of keystream blocks.  Use the
+	// first 4 to XOR 4 full vectors of data.  Then XOR the remaining data.
+	_prepare_2_ctr_vecs	\is_xctr, 4, 5
+	_prepare_2_ctr_vecs	\is_xctr, 6, 7, final=1
+	_aesenc_loop	0,1,2,3,4,5,6,7
+	_aesenclast_and_xor 0,1,2,3
+	vaesenclast	RNDKEYLAST, AESDATA4, AESDATA0
+	vaesenclast	RNDKEYLAST, AESDATA5, AESDATA1
+	vaesenclast	RNDKEYLAST, AESDATA6, AESDATA2
+	vaesenclast	RNDKEYLAST, AESDATA7, AESDATA3
+	sub		$-4*VL, SRC
+	sub		$-4*VL, DST
+	add		$-4*VL, LEN
+	cmp		$1*VL-1, LEN
+	jle		.Lxor_tail_partial_vec_0\@
+	_xor_data	0
+	cmp		$2*VL-1, LEN
+	jle		.Lxor_tail_partial_vec_1\@
+	_xor_data	1
+	cmp		$3*VL-1, LEN
+	jle		.Lxor_tail_partial_vec_2\@
+	_xor_data	2
+	cmp		$4*VL-1, LEN
+	jle		.Lxor_tail_partial_vec_3\@
+	_xor_data	3
+	jmp		.Ldone\@
+
+.Lenc_tail_atmost4vecs\@:
+	cmp		$2*VL, LEN
+	jle		.Lenc_tail_atmost2vecs\@
+
+	// 2*VL < LEN <= 4*VL.  Generate 4 vectors of keystream blocks.  Use the
+	// first 2 to XOR 2 full vectors of data.  Then XOR the remaining data.
+	_aesenc_loop	0,1,2,3
+	_aesenclast_and_xor 0,1
+	vaesenclast	RNDKEYLAST, AESDATA2, AESDATA0
+	vaesenclast	RNDKEYLAST, AESDATA3, AESDATA1
+	sub		$-2*VL, SRC
+	sub		$-2*VL, DST
+	add		$-2*VL, LEN
+	jmp		.Lxor_tail_upto2vecs\@
+
+.Lenc_tail_atmost2vecs\@:
+	// 1 <= LEN <= 2*VL.  Generate 2 vectors of keystream blocks.  Then XOR
+	// the remaining data.
+	_aesenc_loop	0,1
+	vaesenclast	RNDKEYLAST, AESDATA0, AESDATA0
+	vaesenclast	RNDKEYLAST, AESDATA1, AESDATA1
+
+.Lxor_tail_upto2vecs\@:
+	cmp		$1*VL-1, LEN
+	jle		.Lxor_tail_partial_vec_0\@
+	_xor_data	0
+	cmp		$2*VL-1, LEN
+	jle		.Lxor_tail_partial_vec_1\@
+	_xor_data	1
+	jmp		.Ldone\@
+
+.Lxor_tail_partial_vec_1\@:
+	add		$-1*VL, LEN
+	jz		.Ldone\@
+	sub		$-1*VL, SRC
+	sub		$-1*VL, DST
+	_vmovdqa	AESDATA1, AESDATA0
+	jmp		.Lxor_tail_partial_vec_0\@
+
+.Lxor_tail_partial_vec_2\@:
+	add		$-2*VL, LEN
+	jz		.Ldone\@
+	sub		$-2*VL, SRC
+	sub		$-2*VL, DST
+	_vmovdqa	AESDATA2, AESDATA0
+	jmp		.Lxor_tail_partial_vec_0\@
+
+.Lxor_tail_partial_vec_3\@:
+	add		$-3*VL, LEN
+	jz		.Ldone\@
+	sub		$-3*VL, SRC
+	sub		$-3*VL, DST
+	_vmovdqa	AESDATA3, AESDATA0
+
+.Lxor_tail_partial_vec_0\@:
+	// XOR the remaining 1 <= LEN < VL bytes.  It's easy if masked
+	// loads/stores are available; otherwise it's a bit harder...
+.if USE_AVX512
+	mov		$-1, %rax
+	bzhi		LEN64, %rax, %rax
+	kmovq		%rax, %k1
+	vmovdqu8	(SRC), AESDATA1{%k1}{z}
+	vpxord		AESDATA1, AESDATA0, AESDATA0
+	vmovdqu8	AESDATA0, (DST){%k1}
+.else
+  .if VL == 32
+	cmp		$16, LEN
+	jl		1f
+	vpxor		(SRC), AESDATA0_XMM, AESDATA1_XMM
+	vmovdqu		AESDATA1_XMM, (DST)
+	add		$16, SRC
+	add		$16, DST
+	sub		$16, LEN
+	jz		.Ldone\@
+	vextracti128	$1, AESDATA0, AESDATA0_XMM
+1:
+  .endif
+	mov		LEN, %r10d
+	_load_partial_block	SRC, AESDATA1_XMM, KEY, KEY32
+	vpxor		AESDATA1_XMM, AESDATA0_XMM, AESDATA0_XMM
+	mov		%r10d, %ecx
+	_store_partial_block	AESDATA0_XMM, DST, KEY, KEY32
+.endif
+
+.Ldone\@:
+.if VL > 16
+	vzeroupper
+.endif
+	RET
+.endm
+
+// Below are the definitions of the functions generated by the above macro.
+// They have the following prototypes:
+//
+//
+// void aes_ctr64_crypt_##suffix(const struct crypto_aes_ctx *key,
+//				 const u8 *src, u8 *dst, int len,
+//				 const u64 le_ctr[2]);
+//
+// void aes_xctr_crypt_##suffix(const struct crypto_aes_ctx *key,
+//				const u8 *src, u8 *dst, int len,
+//				const u8 iv[AES_BLOCK_SIZE], u64 ctr);
+//
+// Both functions generate |len| bytes of keystream, XOR it with the data from
+// |src|, and write the result to |dst|.  On non-final calls, |len| must be a
+// multiple of 16.  On the final call, |len| can be any value.
+//
+// aes_ctr64_crypt_* implement "regular" CTR, where the keystream is generated
+// from a 128-bit big endian counter that increments by 1 for each AES block.
+// HOWEVER, to keep the assembly code simple, some of the counter management is
+// left to the caller.  aes_ctr64_crypt_* take the counter in little endian
+// form, only increment the low 64 bits internally, do the conversion to big
+// endian internally, and don't write the updated counter back to memory.  The
+// caller is responsible for converting the starting IV to the little endian
+// le_ctr, detecting the (very rare) case of a carry out of the low 64 bits
+// being needed and splitting at that point with a carry done in between, and
+// updating le_ctr after each part if the message is multi-part.
+//
+// aes_xctr_crypt_* implement XCTR as specified in "Length-preserving encryption
+// with HCTR2" (https://eprint.iacr.org/2021/1441.pdf).  XCTR is an
+// easier-to-implement variant of CTR that uses little endian byte order and
+// eliminates carries.  |ctr| is the per-message block counter starting at 1.
+
+.set	VL, 16
+.set	USE_AVX512, 0
+SYM_TYPED_FUNC_START(aes_ctr64_crypt_aesni_avx)
+	_aes_ctr_crypt	0
+SYM_FUNC_END(aes_ctr64_crypt_aesni_avx)
+SYM_TYPED_FUNC_START(aes_xctr_crypt_aesni_avx)
+	_aes_ctr_crypt	1
+SYM_FUNC_END(aes_xctr_crypt_aesni_avx)
+
+.set	VL, 32
+.set	USE_AVX512, 0
+SYM_TYPED_FUNC_START(aes_ctr64_crypt_vaes_avx2)
+	_aes_ctr_crypt	0
+SYM_FUNC_END(aes_ctr64_crypt_vaes_avx2)
+SYM_TYPED_FUNC_START(aes_xctr_crypt_vaes_avx2)
+	_aes_ctr_crypt	1
+SYM_FUNC_END(aes_xctr_crypt_vaes_avx2)
+
+.set	VL, 64
+.set	USE_AVX512, 1
+SYM_TYPED_FUNC_START(aes_ctr64_crypt_vaes_avx512)
+	_aes_ctr_crypt	0
+SYM_FUNC_END(aes_ctr64_crypt_vaes_avx512)
+SYM_TYPED_FUNC_START(aes_xctr_crypt_vaes_avx512)
+	_aes_ctr_crypt	1
+SYM_FUNC_END(aes_xctr_crypt_vaes_avx512)
diff --git a/arch/x86/crypto/aes-gcm-aesni-x86_64.S b/arch/x86/crypto/aes-gcm-aesni-x86_64.S
new file mode 100644
index 000000000..6b2abb768
--- /dev/null
+++ b/arch/x86/crypto/aes-gcm-aesni-x86_64.S
@@ -0,0 +1,1129 @@
+/* SPDX-License-Identifier: Apache-2.0 OR BSD-2-Clause */
+//
+// AES-NI optimized AES-GCM for x86_64
+//
+// Copyright 2024 Google LLC
+//
+// Author: Eric Biggers <ebiggers@google.com>
+//
+//------------------------------------------------------------------------------
+//
+// This file is dual-licensed, meaning that you can use it under your choice of
+// either of the following two licenses:
+//
+// Licensed under the Apache License 2.0 (the "License").  You may obtain a copy
+// of the License at
+//
+//	http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// or
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are met:
+//
+// 1. Redistributions of source code must retain the above copyright notice,
+//    this list of conditions and the following disclaimer.
+//
+// 2. Redistributions in binary form must reproduce the above copyright
+//    notice, this list of conditions and the following disclaimer in the
+//    documentation and/or other materials provided with the distribution.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+// POSSIBILITY OF SUCH DAMAGE.
+//
+//------------------------------------------------------------------------------
+//
+// This file implements AES-GCM (Galois/Counter Mode) for x86_64 CPUs that
+// support the original set of AES instructions, i.e. AES-NI.  Two
+// implementations are provided, one that uses AVX and one that doesn't.  They
+// are very similar, being generated by the same macros.  The only difference is
+// that the AVX implementation takes advantage of VEX-coded instructions in some
+// places to avoid some 'movdqu' and 'movdqa' instructions.  The AVX
+// implementation does *not* use 256-bit vectors, as AES is not supported on
+// 256-bit vectors until the VAES feature (which this file doesn't target).
+//
+// The specific CPU feature prerequisites are AES-NI and PCLMULQDQ, plus SSE4.1
+// for the *_aesni functions or AVX for the *_aesni_avx ones.  (But it seems
+// there are no CPUs that support AES-NI without also PCLMULQDQ and SSE4.1.)
+//
+// The design generally follows that of aes-gcm-vaes-avx512.S, and that file is
+// more thoroughly commented.  This file has the following notable changes:
+//
+//    - The vector length is fixed at 128-bit, i.e. xmm registers.  This means
+//      there is only one AES block (and GHASH block) per register.
+//
+//    - Without AVX512, only 16 SIMD registers are available instead of 32.  We
+//      work around this by being much more careful about using registers,
+//      relying heavily on loads to load values as they are needed.
+//
+//    - Masking is not available either.  We work around this by implementing
+//      partial block loads and stores using overlapping scalar loads and stores
+//      combined with shifts and SSE4.1 insertion and extraction instructions.
+//
+//    - The main loop is organized differently due to the different design
+//      constraints.  First, with just one AES block per SIMD register, on some
+//      CPUs 4 registers don't saturate the 'aesenc' throughput.  We therefore
+//      do an 8-register wide loop.  Considering that and the fact that we have
+//      just 16 SIMD registers to work with, it's not feasible to cache AES
+//      round keys and GHASH key powers in registers across loop iterations.
+//      That's not ideal, but also not actually that bad, since loads can run in
+//      parallel with other instructions.  Significantly, this also makes it
+//      possible to roll up the inner loops, relying on hardware loop unrolling
+//      instead of software loop unrolling, greatly reducing code size.
+//
+//    - We implement the GHASH multiplications in the main loop using Karatsuba
+//      multiplication instead of schoolbook multiplication.  This saves one
+//      pclmulqdq instruction per block, at the cost of one 64-bit load, one
+//      pshufd, and 0.25 pxors per block.  (This is without the three-argument
+//      XOR support that would be provided by AVX512, which would be more
+//      beneficial to schoolbook than Karatsuba.)
+//
+//      As a rough approximation, we can assume that Karatsuba multiplication is
+//      faster than schoolbook multiplication in this context if one pshufd and
+//      0.25 pxors are cheaper than a pclmulqdq.  (We assume that the 64-bit
+//      load is "free" due to running in parallel with arithmetic instructions.)
+//      This is true on AMD CPUs, including all that support pclmulqdq up to at
+//      least Zen 3.  It's also true on older Intel CPUs: Westmere through
+//      Haswell on the Core side, and Silvermont through Goldmont Plus on the
+//      low-power side.  On some of these CPUs, pclmulqdq is quite slow, and the
+//      benefit of Karatsuba should be substantial.  On newer Intel CPUs,
+//      schoolbook multiplication should be faster, but only marginally.
+//
+//      Not all these CPUs were available to be tested.  However, benchmarks on
+//      available CPUs suggest that this approximation is plausible.  Switching
+//      to Karatsuba showed negligible change (< 1%) on Intel Broadwell,
+//      Skylake, and Cascade Lake, but it improved AMD Zen 1-3 by 6-7%.
+//      Considering that and the fact that Karatsuba should be even more
+//      beneficial on older Intel CPUs, it seems like the right choice here.
+//
+//      An additional 0.25 pclmulqdq per block (2 per 8 blocks) could be
+//      saved by using a multiplication-less reduction method.  We don't do that
+//      because it would require a large number of shift and xor instructions,
+//      making it less worthwhile and likely harmful on newer CPUs.
+//
+//      It does make sense to sometimes use a different reduction optimization
+//      that saves a pclmulqdq, though: precompute the hash key times x^64, and
+//      multiply the low half of the data block by the hash key with the extra
+//      factor of x^64.  This eliminates one step of the reduction.  However,
+//      this is incompatible with Karatsuba multiplication.  Therefore, for
+//      multi-block processing we use Karatsuba multiplication with a regular
+//      reduction.  For single-block processing, we use the x^64 optimization.
+
+#include <linux/linkage.h>
+
+.section .rodata
+.p2align 4
+.Lbswap_mask:
+	.octa   0x000102030405060708090a0b0c0d0e0f
+.Lgfpoly:
+	.quad	0xc200000000000000
+.Lone:
+	.quad	1
+.Lgfpoly_and_internal_carrybit:
+	.octa	0xc2000000000000010000000000000001
+	// Loading 16 bytes from '.Lzeropad_mask + 16 - len' produces a mask of
+	// 'len' 0xff bytes and the rest zeroes.
+.Lzeropad_mask:
+	.octa	0xffffffffffffffffffffffffffffffff
+	.octa	0
+
+// Offsets in struct aes_gcm_key_aesni
+#define OFFSETOF_AESKEYLEN	0
+#define OFFSETOF_AESROUNDKEYS	16
+#define OFFSETOF_H_POWERS	272
+#define OFFSETOF_H_POWERS_XORED	400
+#define OFFSETOF_H_TIMES_X64	464
+
+.text
+
+// Do a vpclmulqdq, or fall back to a movdqa and a pclmulqdq.  The fallback
+// assumes that all operands are distinct and that any mem operand is aligned.
+.macro	_vpclmulqdq	imm, src1, src2, dst
+.if USE_AVX
+	vpclmulqdq	\imm, \src1, \src2, \dst
+.else
+	movdqa		\src2, \dst
+	pclmulqdq	\imm, \src1, \dst
+.endif
+.endm
+
+// Do a vpshufb, or fall back to a movdqa and a pshufb.  The fallback assumes
+// that all operands are distinct and that any mem operand is aligned.
+.macro	_vpshufb	src1, src2, dst
+.if USE_AVX
+	vpshufb		\src1, \src2, \dst
+.else
+	movdqa		\src2, \dst
+	pshufb		\src1, \dst
+.endif
+.endm
+
+// Do a vpand, or fall back to a movdqu and a pand.  The fallback assumes that
+// all operands are distinct.
+.macro	_vpand		src1, src2, dst
+.if USE_AVX
+	vpand		\src1, \src2, \dst
+.else
+	movdqu		\src1, \dst
+	pand		\src2, \dst
+.endif
+.endm
+
+// XOR the unaligned memory operand \mem into the xmm register \reg.  \tmp must
+// be a temporary xmm register.
+.macro	_xor_mem_to_reg	mem, reg, tmp
+.if USE_AVX
+	vpxor		\mem, \reg, \reg
+.else
+	movdqu		\mem, \tmp
+	pxor		\tmp, \reg
+.endif
+.endm
+
+// Test the unaligned memory operand \mem against the xmm register \reg.  \tmp
+// must be a temporary xmm register.
+.macro	_test_mem	mem, reg, tmp
+.if USE_AVX
+	vptest		\mem, \reg
+.else
+	movdqu		\mem, \tmp
+	ptest		\tmp, \reg
+.endif
+.endm
+
+// Load 1 <= %ecx <= 15 bytes from the pointer \src into the xmm register \dst
+// and zeroize any remaining bytes.  Clobbers %rax, %rcx, and \tmp{64,32}.
+.macro	_load_partial_block	src, dst, tmp64, tmp32
+	sub		$8, %ecx		// LEN - 8
+	jle		.Lle8\@
+
+	// Load 9 <= LEN <= 15 bytes.
+	movq		(\src), \dst		// Load first 8 bytes
+	mov		(\src, %rcx), %rax	// Load last 8 bytes
+	neg		%ecx
+	shl		$3, %ecx
+	shr		%cl, %rax		// Discard overlapping bytes
+	pinsrq		$1, %rax, \dst
+	jmp		.Ldone\@
+
+.Lle8\@:
+	add		$4, %ecx		// LEN - 4
+	jl		.Llt4\@
+
+	// Load 4 <= LEN <= 8 bytes.
+	mov		(\src), %eax		// Load first 4 bytes
+	mov		(\src, %rcx), \tmp32	// Load last 4 bytes
+	jmp		.Lcombine\@
+
+.Llt4\@:
+	// Load 1 <= LEN <= 3 bytes.
+	add		$2, %ecx		// LEN - 2
+	movzbl		(\src), %eax		// Load first byte
+	jl		.Lmovq\@
+	movzwl		(\src, %rcx), \tmp32	// Load last 2 bytes
+.Lcombine\@:
+	shl		$3, %ecx
+	shl		%cl, \tmp64
+	or		\tmp64, %rax		// Combine the two parts
+.Lmovq\@:
+	movq		%rax, \dst
+.Ldone\@:
+.endm
+
+// Store 1 <= %ecx <= 15 bytes from the xmm register \src to the pointer \dst.
+// Clobbers %rax, %rcx, and %rsi.
+.macro	_store_partial_block	src, dst
+	sub		$8, %ecx		// LEN - 8
+	jl		.Llt8\@
+
+	// Store 8 <= LEN <= 15 bytes.
+	pextrq		$1, \src, %rax
+	mov		%ecx, %esi
+	shl		$3, %ecx
+	ror		%cl, %rax
+	mov		%rax, (\dst, %rsi)	// Store last LEN - 8 bytes
+	movq		\src, (\dst)		// Store first 8 bytes
+	jmp		.Ldone\@
+
+.Llt8\@:
+	add		$4, %ecx		// LEN - 4
+	jl		.Llt4\@
+
+	// Store 4 <= LEN <= 7 bytes.
+	pextrd		$1, \src, %eax
+	mov		%ecx, %esi
+	shl		$3, %ecx
+	ror		%cl, %eax
+	mov		%eax, (\dst, %rsi)	// Store last LEN - 4 bytes
+	movd		\src, (\dst)		// Store first 4 bytes
+	jmp		.Ldone\@
+
+.Llt4\@:
+	// Store 1 <= LEN <= 3 bytes.
+	pextrb		$0, \src, 0(\dst)
+	cmp		$-2, %ecx		// LEN - 4 == -2, i.e. LEN == 2?
+	jl		.Ldone\@
+	pextrb		$1, \src, 1(\dst)
+	je		.Ldone\@
+	pextrb		$2, \src, 2(\dst)
+.Ldone\@:
+.endm
+
+// Do one step of GHASH-multiplying \a by \b and storing the reduced product in
+// \b.  To complete all steps, this must be invoked with \i=0 through \i=9.
+// \a_times_x64 must contain \a * x^64 in reduced form, \gfpoly must contain the
+// .Lgfpoly constant, and \t0-\t1 must be temporary registers.
+.macro	_ghash_mul_step	i, a, a_times_x64, b, gfpoly, t0, t1
+
+	// MI = (a_L * b_H) + ((a*x^64)_L * b_L)
+.if \i == 0
+	_vpclmulqdq	$0x01, \a, \b, \t0
+.elseif \i == 1
+	_vpclmulqdq	$0x00, \a_times_x64, \b, \t1
+.elseif \i == 2
+	pxor		\t1, \t0
+
+	// HI = (a_H * b_H) + ((a*x^64)_H * b_L)
+.elseif \i == 3
+	_vpclmulqdq	$0x11, \a, \b, \t1
+.elseif \i == 4
+	pclmulqdq	$0x10, \a_times_x64, \b
+.elseif \i == 5
+	pxor		\t1, \b
+.elseif \i == 6
+
+	// Fold MI into HI.
+	pshufd		$0x4e, \t0, \t1		// Swap halves of MI
+.elseif \i == 7
+	pclmulqdq	$0x00, \gfpoly, \t0	// MI_L*(x^63 + x^62 + x^57)
+.elseif \i == 8
+	pxor		\t1, \b
+.elseif \i == 9
+	pxor		\t0, \b
+.endif
+.endm
+
+// GHASH-multiply \a by \b and store the reduced product in \b.
+// See _ghash_mul_step for details.
+.macro	_ghash_mul	a, a_times_x64, b, gfpoly, t0, t1
+.irp i, 0,1,2,3,4,5,6,7,8,9
+	_ghash_mul_step	\i, \a, \a_times_x64, \b, \gfpoly, \t0, \t1
+.endr
+.endm
+
+// GHASH-multiply \a by \b and add the unreduced product to \lo, \mi, and \hi.
+// This does Karatsuba multiplication and must be paired with _ghash_reduce.  On
+// the first call, \lo, \mi, and \hi must be zero.  \a_xored must contain the
+// two halves of \a XOR'd together, i.e. a_L + a_H.  \b is clobbered.
+.macro	_ghash_mul_noreduce	a, a_xored, b, lo, mi, hi, t0
+
+	// LO += a_L * b_L
+	_vpclmulqdq	$0x00, \a, \b, \t0
+	pxor		\t0, \lo
+
+	// b_L + b_H
+	pshufd		$0x4e, \b, \t0
+	pxor		\b, \t0
+
+	// HI += a_H * b_H
+	pclmulqdq	$0x11, \a, \b
+	pxor		\b, \hi
+
+	// MI += (a_L + a_H) * (b_L + b_H)
+	pclmulqdq	$0x00, \a_xored, \t0
+	pxor		\t0, \mi
+.endm
+
+// Reduce the product from \lo, \mi, and \hi, and store the result in \dst.
+// This assumes that _ghash_mul_noreduce was used.
+.macro	_ghash_reduce	lo, mi, hi, dst, t0
+
+	movq		.Lgfpoly(%rip), \t0
+
+	// MI += LO + HI (needed because we used Karatsuba multiplication)
+	pxor		\lo, \mi
+	pxor		\hi, \mi
+
+	// Fold LO into MI.
+	pshufd		$0x4e, \lo, \dst
+	pclmulqdq	$0x00, \t0, \lo
+	pxor		\dst, \mi
+	pxor		\lo, \mi
+
+	// Fold MI into HI.
+	pshufd		$0x4e, \mi, \dst
+	pclmulqdq	$0x00, \t0, \mi
+	pxor		\hi, \dst
+	pxor		\mi, \dst
+.endm
+
+// Do the first step of the GHASH update of a set of 8 ciphertext blocks.
+//
+// The whole GHASH update does:
+//
+//	GHASH_ACC = (blk0+GHASH_ACC)*H^8 + blk1*H^7 + blk2*H^6 + blk3*H^5 +
+//				blk4*H^4 + blk5*H^3 + blk6*H^2 + blk7*H^1
+//
+// This macro just does the first step: it does the unreduced multiplication
+// (blk0+GHASH_ACC)*H^8 and starts gathering the unreduced product in the xmm
+// registers LO, MI, and GHASH_ACC a.k.a. HI.  It also zero-initializes the
+// inner block counter in %rax, which is a value that counts up by 8 for each
+// block in the set of 8 and is used later to index by 8*blknum and 16*blknum.
+//
+// To reduce the number of pclmulqdq instructions required, both this macro and
+// _ghash_update_continue_8x use Karatsuba multiplication instead of schoolbook
+// multiplication.  See the file comment for more details about this choice.
+//
+// Both macros expect the ciphertext blocks blk[0-7] to be available at DST if
+// encrypting, or SRC if decrypting.  They also expect the precomputed hash key
+// powers H^i and their XOR'd-together halves to be available in the struct
+// pointed to by KEY.  Both macros clobber TMP[0-2].
+.macro	_ghash_update_begin_8x	enc
+
+	// Initialize the inner block counter.
+	xor		%eax, %eax
+
+	// Load the highest hash key power, H^8.
+	movdqa		OFFSETOF_H_POWERS(KEY), TMP0
+
+	// Load the first ciphertext block and byte-reflect it.
+.if \enc
+	movdqu		(DST), TMP1
+.else
+	movdqu		(SRC), TMP1
+.endif
+	pshufb		BSWAP_MASK, TMP1
+
+	// Add the GHASH accumulator to the ciphertext block to get the block
+	// 'b' that needs to be multiplied with the hash key power 'a'.
+	pxor		TMP1, GHASH_ACC
+
+	// b_L + b_H
+	pshufd		$0x4e, GHASH_ACC, MI
+	pxor		GHASH_ACC, MI
+
+	// LO = a_L * b_L
+	_vpclmulqdq	$0x00, TMP0, GHASH_ACC, LO
+
+	// HI = a_H * b_H
+	pclmulqdq	$0x11, TMP0, GHASH_ACC
+
+	// MI = (a_L + a_H) * (b_L + b_H)
+	pclmulqdq	$0x00, OFFSETOF_H_POWERS_XORED(KEY), MI
+.endm
+
+// Continue the GHASH update of 8 ciphertext blocks as described above by doing
+// an unreduced multiplication of the next ciphertext block by the next lowest
+// key power and accumulating the result into LO, MI, and GHASH_ACC a.k.a. HI.
+.macro	_ghash_update_continue_8x enc
+	add		$8, %eax
+
+	// Load the next lowest key power.
+	movdqa		OFFSETOF_H_POWERS(KEY,%rax,2), TMP0
+
+	// Load the next ciphertext block and byte-reflect it.
+.if \enc
+	movdqu		(DST,%rax,2), TMP1
+.else
+	movdqu		(SRC,%rax,2), TMP1
+.endif
+	pshufb		BSWAP_MASK, TMP1
+
+	// LO += a_L * b_L
+	_vpclmulqdq	$0x00, TMP0, TMP1, TMP2
+	pxor		TMP2, LO
+
+	// b_L + b_H
+	pshufd		$0x4e, TMP1, TMP2
+	pxor		TMP1, TMP2
+
+	// HI += a_H * b_H
+	pclmulqdq	$0x11, TMP0, TMP1
+	pxor		TMP1, GHASH_ACC
+
+	// MI += (a_L + a_H) * (b_L + b_H)
+	movq		OFFSETOF_H_POWERS_XORED(KEY,%rax), TMP1
+	pclmulqdq	$0x00, TMP1, TMP2
+	pxor		TMP2, MI
+.endm
+
+// Reduce LO, MI, and GHASH_ACC a.k.a. HI into GHASH_ACC.  This is similar to
+// _ghash_reduce, but it's hardcoded to use the registers of the main loop and
+// it uses the same register for HI and the destination.  It's also divided into
+// two steps.  TMP1 must be preserved across steps.
+//
+// One pshufd could be saved by shuffling MI and XOR'ing LO into it, instead of
+// shuffling LO, XOR'ing LO into MI, and shuffling MI.  However, this would
+// increase the critical path length, and it seems to slightly hurt performance.
+.macro	_ghash_update_end_8x_step	i
+.if \i == 0
+	movq		.Lgfpoly(%rip), TMP1
+	pxor		LO, MI
+	pxor		GHASH_ACC, MI
+	pshufd		$0x4e, LO, TMP2
+	pclmulqdq	$0x00, TMP1, LO
+	pxor		TMP2, MI
+	pxor		LO, MI
+.elseif \i == 1
+	pshufd		$0x4e, MI, TMP2
+	pclmulqdq	$0x00, TMP1, MI
+	pxor		TMP2, GHASH_ACC
+	pxor		MI, GHASH_ACC
+.endif
+.endm
+
+// void aes_gcm_precompute_##suffix(struct aes_gcm_key_aesni *key);
+//
+// Given the expanded AES key, derive the GHASH subkey and initialize the GHASH
+// related fields in the key struct.
+.macro	_aes_gcm_precompute
+
+	// Function arguments
+	.set	KEY,		%rdi
+
+	// Additional local variables.
+	// %xmm0-%xmm1 and %rax are used as temporaries.
+	.set	RNDKEYLAST_PTR,	%rsi
+	.set	H_CUR,		%xmm2
+	.set	H_POW1,		%xmm3	// H^1
+	.set	H_POW1_X64,	%xmm4	// H^1 * x^64
+	.set	GFPOLY,		%xmm5
+
+	// Encrypt an all-zeroes block to get the raw hash subkey.
+	movl		OFFSETOF_AESKEYLEN(KEY), %eax
+	lea		OFFSETOF_AESROUNDKEYS+6*16(KEY,%rax,4), RNDKEYLAST_PTR
+	movdqa		OFFSETOF_AESROUNDKEYS(KEY), H_POW1
+	lea		OFFSETOF_AESROUNDKEYS+16(KEY), %rax
+1:
+	aesenc		(%rax), H_POW1
+	add		$16, %rax
+	cmp		%rax, RNDKEYLAST_PTR
+	jne		1b
+	aesenclast	(RNDKEYLAST_PTR), H_POW1
+
+	// Preprocess the raw hash subkey as needed to operate on GHASH's
+	// bit-reflected values directly: reflect its bytes, then multiply it by
+	// x^-1 (using the backwards interpretation of polynomial coefficients
+	// from the GCM spec) or equivalently x^1 (using the alternative,
+	// natural interpretation of polynomial coefficients).
+	pshufb		.Lbswap_mask(%rip), H_POW1
+	movdqa		H_POW1, %xmm0
+	pshufd		$0xd3, %xmm0, %xmm0
+	psrad		$31, %xmm0
+	paddq		H_POW1, H_POW1
+	pand		.Lgfpoly_and_internal_carrybit(%rip), %xmm0
+	pxor		%xmm0, H_POW1
+
+	// Store H^1.
+	movdqa		H_POW1, OFFSETOF_H_POWERS+7*16(KEY)
+
+	// Compute and store H^1 * x^64.
+	movq		.Lgfpoly(%rip), GFPOLY
+	pshufd		$0x4e, H_POW1, %xmm0
+	_vpclmulqdq	$0x00, H_POW1, GFPOLY, H_POW1_X64
+	pxor		%xmm0, H_POW1_X64
+	movdqa		H_POW1_X64, OFFSETOF_H_TIMES_X64(KEY)
+
+	// Compute and store the halves of H^1 XOR'd together.
+	pxor		H_POW1, %xmm0
+	movq		%xmm0, OFFSETOF_H_POWERS_XORED+7*8(KEY)
+
+	// Compute and store the remaining key powers H^2 through H^8.
+	movdqa		H_POW1, H_CUR
+	mov		$6*8, %eax
+.Lprecompute_next\@:
+	// Compute H^i = H^{i-1} * H^1.
+	_ghash_mul	H_POW1, H_POW1_X64, H_CUR, GFPOLY, %xmm0, %xmm1
+	// Store H^i.
+	movdqa		H_CUR, OFFSETOF_H_POWERS(KEY,%rax,2)
+	// Compute and store the halves of H^i XOR'd together.
+	pshufd		$0x4e, H_CUR, %xmm0
+	pxor		H_CUR, %xmm0
+	movq		%xmm0, OFFSETOF_H_POWERS_XORED(KEY,%rax)
+	sub		$8, %eax
+	jge		.Lprecompute_next\@
+
+	RET
+.endm
+
+// void aes_gcm_aad_update_aesni(const struct aes_gcm_key_aesni *key,
+//				 u8 ghash_acc[16], const u8 *aad, int aadlen);
+//
+// This function processes the AAD (Additional Authenticated Data) in GCM.
+// Using the key |key|, it updates the GHASH accumulator |ghash_acc| with the
+// data given by |aad| and |aadlen|.  On the first call, |ghash_acc| must be all
+// zeroes.  |aadlen| must be a multiple of 16, except on the last call where it
+// can be any length.  The caller must do any buffering needed to ensure this.
+.macro	_aes_gcm_aad_update
+
+	// Function arguments
+	.set	KEY,		%rdi
+	.set	GHASH_ACC_PTR,	%rsi
+	.set	AAD,		%rdx
+	.set	AADLEN,		%ecx
+	// Note: _load_partial_block relies on AADLEN being in %ecx.
+
+	// Additional local variables.
+	// %rax, %r10, and %xmm0-%xmm1 are used as temporary registers.
+	.set	BSWAP_MASK,	%xmm2
+	.set	GHASH_ACC,	%xmm3
+	.set	H_POW1,		%xmm4	// H^1
+	.set	H_POW1_X64,	%xmm5	// H^1 * x^64
+	.set	GFPOLY,		%xmm6
+
+	movdqa		.Lbswap_mask(%rip), BSWAP_MASK
+	movdqu		(GHASH_ACC_PTR), GHASH_ACC
+	movdqa		OFFSETOF_H_POWERS+7*16(KEY), H_POW1
+	movdqa		OFFSETOF_H_TIMES_X64(KEY), H_POW1_X64
+	movq		.Lgfpoly(%rip), GFPOLY
+
+	// Process the AAD one full block at a time.
+	sub		$16, AADLEN
+	jl		.Laad_loop_1x_done\@
+.Laad_loop_1x\@:
+	movdqu		(AAD), %xmm0
+	pshufb		BSWAP_MASK, %xmm0
+	pxor		%xmm0, GHASH_ACC
+	_ghash_mul	H_POW1, H_POW1_X64, GHASH_ACC, GFPOLY, %xmm0, %xmm1
+	add		$16, AAD
+	sub		$16, AADLEN
+	jge		.Laad_loop_1x\@
+.Laad_loop_1x_done\@:
+	// Check whether there is a partial block at the end.
+	add		$16, AADLEN
+	jz		.Laad_done\@
+
+	// Process a partial block of length 1 <= AADLEN <= 15.
+	// _load_partial_block assumes that %ecx contains AADLEN.
+	_load_partial_block	AAD, %xmm0, %r10, %r10d
+	pshufb		BSWAP_MASK, %xmm0
+	pxor		%xmm0, GHASH_ACC
+	_ghash_mul	H_POW1, H_POW1_X64, GHASH_ACC, GFPOLY, %xmm0, %xmm1
+
+.Laad_done\@:
+	movdqu		GHASH_ACC, (GHASH_ACC_PTR)
+	RET
+.endm
+
+// Increment LE_CTR eight times to generate eight little-endian counter blocks,
+// swap each to big-endian, and store them in AESDATA[0-7].  Also XOR them with
+// the zero-th AES round key.  Clobbers TMP0 and TMP1.
+.macro	_ctr_begin_8x
+	movq		.Lone(%rip), TMP0
+	movdqa		OFFSETOF_AESROUNDKEYS(KEY), TMP1 // zero-th round key
+.irp i, 0,1,2,3,4,5,6,7
+	_vpshufb	BSWAP_MASK, LE_CTR, AESDATA\i
+	pxor		TMP1, AESDATA\i
+	paddd		TMP0, LE_CTR
+.endr
+.endm
+
+// Do a non-last round of AES on AESDATA[0-7] using \round_key.
+.macro	_aesenc_8x	round_key
+.irp i, 0,1,2,3,4,5,6,7
+	aesenc		\round_key, AESDATA\i
+.endr
+.endm
+
+// Do the last round of AES on AESDATA[0-7] using \round_key.
+.macro	_aesenclast_8x	round_key
+.irp i, 0,1,2,3,4,5,6,7
+	aesenclast	\round_key, AESDATA\i
+.endr
+.endm
+
+// XOR eight blocks from SRC with the keystream blocks in AESDATA[0-7], and
+// store the result to DST.  Clobbers TMP0.
+.macro	_xor_data_8x
+.irp i, 0,1,2,3,4,5,6,7
+	_xor_mem_to_reg	\i*16(SRC), AESDATA\i, tmp=TMP0
+.endr
+.irp i, 0,1,2,3,4,5,6,7
+	movdqu		AESDATA\i, \i*16(DST)
+.endr
+.endm
+
+// void aes_gcm_{enc,dec}_update_##suffix(const struct aes_gcm_key_aesni *key,
+//					  const u32 le_ctr[4], u8 ghash_acc[16],
+//					  const u8 *src, u8 *dst, int datalen);
+//
+// This macro generates a GCM encryption or decryption update function with the
+// above prototype (with \enc selecting which one).
+//
+// This function computes the next portion of the CTR keystream, XOR's it with
+// |datalen| bytes from |src|, and writes the resulting encrypted or decrypted
+// data to |dst|.  It also updates the GHASH accumulator |ghash_acc| using the
+// next |datalen| ciphertext bytes.
+//
+// |datalen| must be a multiple of 16, except on the last call where it can be
+// any length.  The caller must do any buffering needed to ensure this.  Both
+// in-place and out-of-place en/decryption are supported.
+//
+// |le_ctr| must give the current counter in little-endian format.  For a new
+// message, the low word of the counter must be 2.  This function loads the
+// counter from |le_ctr| and increments the loaded counter as needed, but it
+// does *not* store the updated counter back to |le_ctr|.  The caller must
+// update |le_ctr| if any more data segments follow.  Internally, only the low
+// 32-bit word of the counter is incremented, following the GCM standard.
+.macro	_aes_gcm_update	enc
+
+	// Function arguments
+	.set	KEY,		%rdi
+	.set	LE_CTR_PTR,	%rsi	// Note: overlaps with usage as temp reg
+	.set	GHASH_ACC_PTR,	%rdx
+	.set	SRC,		%rcx
+	.set	DST,		%r8
+	.set	DATALEN,	%r9d
+	.set	DATALEN64,	%r9	// Zero-extend DATALEN before using!
+	// Note: the code setting up for _load_partial_block assumes that SRC is
+	// in %rcx (and that DATALEN is *not* in %rcx).
+
+	// Additional local variables
+
+	// %rax and %rsi are used as temporary registers.  Note: %rsi overlaps
+	// with LE_CTR_PTR, which is used only at the beginning.
+
+	.set	AESKEYLEN,	%r10d	// AES key length in bytes
+	.set	AESKEYLEN64,	%r10
+	.set	RNDKEYLAST_PTR,	%r11	// Pointer to last AES round key
+
+	// Put the most frequently used values in %xmm0-%xmm7 to reduce code
+	// size.  (%xmm0-%xmm7 take fewer bytes to encode than %xmm8-%xmm15.)
+	.set	TMP0,		%xmm0
+	.set	TMP1,		%xmm1
+	.set	TMP2,		%xmm2
+	.set	LO,		%xmm3	// Low part of unreduced product
+	.set	MI,		%xmm4	// Middle part of unreduced product
+	.set	GHASH_ACC,	%xmm5	// GHASH accumulator; in main loop also
+					// the high part of unreduced product
+	.set	BSWAP_MASK,	%xmm6	// Shuffle mask for reflecting bytes
+	.set	LE_CTR,		%xmm7	// Little-endian counter value
+	.set	AESDATA0,	%xmm8
+	.set	AESDATA1,	%xmm9
+	.set	AESDATA2,	%xmm10
+	.set	AESDATA3,	%xmm11
+	.set	AESDATA4,	%xmm12
+	.set	AESDATA5,	%xmm13
+	.set	AESDATA6,	%xmm14
+	.set	AESDATA7,	%xmm15
+
+	movdqa		.Lbswap_mask(%rip), BSWAP_MASK
+	movdqu		(GHASH_ACC_PTR), GHASH_ACC
+	movdqu		(LE_CTR_PTR), LE_CTR
+
+	movl		OFFSETOF_AESKEYLEN(KEY), AESKEYLEN
+	lea		OFFSETOF_AESROUNDKEYS+6*16(KEY,AESKEYLEN64,4), RNDKEYLAST_PTR
+
+	// If there are at least 8*16 bytes of data, then continue into the main
+	// loop, which processes 8*16 bytes of data per iteration.
+	//
+	// The main loop interleaves AES and GHASH to improve performance on
+	// CPUs that can execute these instructions in parallel.  When
+	// decrypting, the GHASH input (the ciphertext) is immediately
+	// available.  When encrypting, we instead encrypt a set of 8 blocks
+	// first and then GHASH those blocks while encrypting the next set of 8,
+	// repeat that as needed, and finally GHASH the last set of 8 blocks.
+	//
+	// Code size optimization: Prefer adding or subtracting -8*16 over 8*16,
+	// as this makes the immediate fit in a signed byte, saving 3 bytes.
+	add		$-8*16, DATALEN
+	jl		.Lcrypt_loop_8x_done\@
+.if \enc
+	// Encrypt the first 8 plaintext blocks.
+	_ctr_begin_8x
+	lea		OFFSETOF_AESROUNDKEYS+16(KEY), %rsi
+	.p2align 4
+1:
+	movdqa		(%rsi), TMP0
+	_aesenc_8x	TMP0
+	add		$16, %rsi
+	cmp		%rsi, RNDKEYLAST_PTR
+	jne		1b
+	movdqa		(%rsi), TMP0
+	_aesenclast_8x	TMP0
+	_xor_data_8x
+	// Don't increment DST until the ciphertext blocks have been hashed.
+	sub		$-8*16, SRC
+	add		$-8*16, DATALEN
+	jl		.Lghash_last_ciphertext_8x\@
+.endif
+
+	.p2align 4
+.Lcrypt_loop_8x\@:
+
+	// Generate the next set of 8 counter blocks and start encrypting them.
+	_ctr_begin_8x
+	lea		OFFSETOF_AESROUNDKEYS+16(KEY), %rsi
+
+	// Do a round of AES, and start the GHASH update of 8 ciphertext blocks
+	// by doing the unreduced multiplication for the first ciphertext block.
+	movdqa		(%rsi), TMP0
+	add		$16, %rsi
+	_aesenc_8x	TMP0
+	_ghash_update_begin_8x \enc
+
+	// Do 7 more rounds of AES, and continue the GHASH update by doing the
+	// unreduced multiplication for the remaining ciphertext blocks.
+	.p2align 4
+1:
+	movdqa		(%rsi), TMP0
+	add		$16, %rsi
+	_aesenc_8x	TMP0
+	_ghash_update_continue_8x \enc
+	cmp		$7*8, %eax
+	jne		1b
+
+	// Do the remaining AES rounds.
+	.p2align 4
+1:
+	movdqa		(%rsi), TMP0
+	add		$16, %rsi
+	_aesenc_8x	TMP0
+	cmp		%rsi, RNDKEYLAST_PTR
+	jne		1b
+
+	// Do the GHASH reduction and the last round of AES.
+	movdqa		(RNDKEYLAST_PTR), TMP0
+	_ghash_update_end_8x_step	0
+	_aesenclast_8x	TMP0
+	_ghash_update_end_8x_step	1
+
+	// XOR the data with the AES-CTR keystream blocks.
+.if \enc
+	sub		$-8*16, DST
+.endif
+	_xor_data_8x
+	sub		$-8*16, SRC
+.if !\enc
+	sub		$-8*16, DST
+.endif
+	add		$-8*16, DATALEN
+	jge		.Lcrypt_loop_8x\@
+
+.if \enc
+.Lghash_last_ciphertext_8x\@:
+	// Update GHASH with the last set of 8 ciphertext blocks.
+	_ghash_update_begin_8x		\enc
+	.p2align 4
+1:
+	_ghash_update_continue_8x	\enc
+	cmp		$7*8, %eax
+	jne		1b
+	_ghash_update_end_8x_step	0
+	_ghash_update_end_8x_step	1
+	sub		$-8*16, DST
+.endif
+
+.Lcrypt_loop_8x_done\@:
+
+	sub		$-8*16, DATALEN
+	jz		.Ldone\@
+
+	// Handle the remainder of length 1 <= DATALEN < 8*16 bytes.  We keep
+	// things simple and keep the code size down by just going one block at
+	// a time, again taking advantage of hardware loop unrolling.  Since
+	// there are enough key powers available for all remaining data, we do
+	// the GHASH multiplications unreduced, and only reduce at the very end.
+
+	.set	HI,		TMP2
+	.set	H_POW,		AESDATA0
+	.set	H_POW_XORED,	AESDATA1
+	.set	ONE,		AESDATA2
+
+	movq		.Lone(%rip), ONE
+
+	// Start collecting the unreduced GHASH intermediate value LO, MI, HI.
+	pxor		LO, LO
+	pxor		MI, MI
+	pxor		HI, HI
+
+	// Set up a block counter %rax to contain 8*(8-n), where n is the number
+	// of blocks that remain, counting any partial block.  This will be used
+	// to access the key powers H^n through H^1.
+	mov		DATALEN, %eax
+	neg		%eax
+	and		$~15, %eax
+	sar		$1, %eax
+	add		$64, %eax
+
+	sub		$16, DATALEN
+	jl		.Lcrypt_loop_1x_done\@
+
+	// Process the data one full block at a time.
+.Lcrypt_loop_1x\@:
+
+	// Encrypt the next counter block.
+	_vpshufb	BSWAP_MASK, LE_CTR, TMP0
+	paddd		ONE, LE_CTR
+	pxor		OFFSETOF_AESROUNDKEYS(KEY), TMP0
+	lea		-6*16(RNDKEYLAST_PTR), %rsi	// Reduce code size
+	cmp		$24, AESKEYLEN
+	jl		128f	// AES-128?
+	je		192f	// AES-192?
+	// AES-256
+	aesenc		-7*16(%rsi), TMP0
+	aesenc		-6*16(%rsi), TMP0
+192:
+	aesenc		-5*16(%rsi), TMP0
+	aesenc		-4*16(%rsi), TMP0
+128:
+.irp i, -3,-2,-1,0,1,2,3,4,5
+	aesenc		\i*16(%rsi), TMP0
+.endr
+	aesenclast	(RNDKEYLAST_PTR), TMP0
+
+	// Load the next key power H^i.
+	movdqa		OFFSETOF_H_POWERS(KEY,%rax,2), H_POW
+	movq		OFFSETOF_H_POWERS_XORED(KEY,%rax), H_POW_XORED
+
+	// XOR the keystream block that was just generated in TMP0 with the next
+	// source data block and store the resulting en/decrypted data to DST.
+.if \enc
+	_xor_mem_to_reg	(SRC), TMP0, tmp=TMP1
+	movdqu		TMP0, (DST)
+.else
+	movdqu		(SRC), TMP1
+	pxor		TMP1, TMP0
+	movdqu		TMP0, (DST)
+.endif
+
+	// Update GHASH with the ciphertext block.
+.if \enc
+	pshufb		BSWAP_MASK, TMP0
+	pxor		TMP0, GHASH_ACC
+.else
+	pshufb		BSWAP_MASK, TMP1
+	pxor		TMP1, GHASH_ACC
+.endif
+	_ghash_mul_noreduce	H_POW, H_POW_XORED, GHASH_ACC, LO, MI, HI, TMP0
+	pxor		GHASH_ACC, GHASH_ACC
+
+	add		$8, %eax
+	add		$16, SRC
+	add		$16, DST
+	sub		$16, DATALEN
+	jge		.Lcrypt_loop_1x\@
+.Lcrypt_loop_1x_done\@:
+	// Check whether there is a partial block at the end.
+	add		$16, DATALEN
+	jz		.Lghash_reduce\@
+
+	// Process a partial block of length 1 <= DATALEN <= 15.
+
+	// Encrypt a counter block for the last time.
+	pshufb		BSWAP_MASK, LE_CTR
+	pxor		OFFSETOF_AESROUNDKEYS(KEY), LE_CTR
+	lea		OFFSETOF_AESROUNDKEYS+16(KEY), %rsi
+1:
+	aesenc		(%rsi), LE_CTR
+	add		$16, %rsi
+	cmp		%rsi, RNDKEYLAST_PTR
+	jne		1b
+	aesenclast	(RNDKEYLAST_PTR), LE_CTR
+
+	// Load the lowest key power, H^1.
+	movdqa		OFFSETOF_H_POWERS(KEY,%rax,2), H_POW
+	movq		OFFSETOF_H_POWERS_XORED(KEY,%rax), H_POW_XORED
+
+	// Load and zero-pad 1 <= DATALEN <= 15 bytes of data from SRC.  SRC is
+	// in %rcx, but _load_partial_block needs DATALEN in %rcx instead.
+	// RNDKEYLAST_PTR is no longer needed, so reuse it for SRC.
+	mov		SRC, RNDKEYLAST_PTR
+	mov		DATALEN, %ecx
+	_load_partial_block	RNDKEYLAST_PTR, TMP0, %rsi, %esi
+
+	// XOR the keystream block that was just generated in LE_CTR with the
+	// source data block and store the resulting en/decrypted data to DST.
+	pxor		TMP0, LE_CTR
+	mov		DATALEN, %ecx
+	_store_partial_block	LE_CTR, DST
+
+	// If encrypting, zero-pad the final ciphertext block for GHASH.  (If
+	// decrypting, this was already done by _load_partial_block.)
+.if \enc
+	lea		.Lzeropad_mask+16(%rip), %rax
+	sub		DATALEN64, %rax
+	_vpand		(%rax), LE_CTR, TMP0
+.endif
+
+	// Update GHASH with the final ciphertext block.
+	pshufb		BSWAP_MASK, TMP0
+	pxor		TMP0, GHASH_ACC
+	_ghash_mul_noreduce	H_POW, H_POW_XORED, GHASH_ACC, LO, MI, HI, TMP0
+
+.Lghash_reduce\@:
+	// Finally, do the GHASH reduction.
+	_ghash_reduce	LO, MI, HI, GHASH_ACC, TMP0
+
+.Ldone\@:
+	// Store the updated GHASH accumulator back to memory.
+	movdqu		GHASH_ACC, (GHASH_ACC_PTR)
+
+	RET
+.endm
+
+// void aes_gcm_enc_final_##suffix(const struct aes_gcm_key_aesni *key,
+//				   const u32 le_ctr[4], u8 ghash_acc[16],
+//				   u64 total_aadlen, u64 total_datalen);
+// bool aes_gcm_dec_final_##suffix(const struct aes_gcm_key_aesni *key,
+//				   const u32 le_ctr[4], const u8 ghash_acc[16],
+//				   u64 total_aadlen, u64 total_datalen,
+//				   const u8 tag[16], int taglen);
+//
+// This macro generates one of the above two functions (with \enc selecting
+// which one).  Both functions finish computing the GCM authentication tag by
+// updating GHASH with the lengths block and encrypting the GHASH accumulator.
+// |total_aadlen| and |total_datalen| must be the total length of the additional
+// authenticated data and the en/decrypted data in bytes, respectively.
+//
+// The encryption function then stores the full-length (16-byte) computed
+// authentication tag to |ghash_acc|.  The decryption function instead loads the
+// expected authentication tag (the one that was transmitted) from the 16-byte
+// buffer |tag|, compares the first 4 <= |taglen| <= 16 bytes of it to the
+// computed tag in constant time, and returns true if and only if they match.
+.macro	_aes_gcm_final	enc
+
+	// Function arguments
+	.set	KEY,		%rdi
+	.set	LE_CTR_PTR,	%rsi
+	.set	GHASH_ACC_PTR,	%rdx
+	.set	TOTAL_AADLEN,	%rcx
+	.set	TOTAL_DATALEN,	%r8
+	.set	TAG,		%r9
+	.set	TAGLEN,		%r10d	// Originally at 8(%rsp)
+	.set	TAGLEN64,	%r10
+
+	// Additional local variables.
+	// %rax and %xmm0-%xmm2 are used as temporary registers.
+	.set	AESKEYLEN,	%r11d
+	.set	AESKEYLEN64,	%r11
+	.set	BSWAP_MASK,	%xmm3
+	.set	GHASH_ACC,	%xmm4
+	.set	H_POW1,		%xmm5	// H^1
+	.set	H_POW1_X64,	%xmm6	// H^1 * x^64
+	.set	GFPOLY,		%xmm7
+
+	movdqa		.Lbswap_mask(%rip), BSWAP_MASK
+	movl		OFFSETOF_AESKEYLEN(KEY), AESKEYLEN
+
+	// Set up a counter block with 1 in the low 32-bit word.  This is the
+	// counter that produces the ciphertext needed to encrypt the auth tag.
+	movdqu		(LE_CTR_PTR), %xmm0
+	mov		$1, %eax
+	pinsrd		$0, %eax, %xmm0
+
+	// Build the lengths block and XOR it into the GHASH accumulator.
+	movq		TOTAL_DATALEN, GHASH_ACC
+	pinsrq		$1, TOTAL_AADLEN, GHASH_ACC
+	psllq		$3, GHASH_ACC	// Bytes to bits
+	_xor_mem_to_reg	(GHASH_ACC_PTR), GHASH_ACC, %xmm1
+
+	movdqa		OFFSETOF_H_POWERS+7*16(KEY), H_POW1
+	movdqa		OFFSETOF_H_TIMES_X64(KEY), H_POW1_X64
+	movq		.Lgfpoly(%rip), GFPOLY
+
+	// Make %rax point to the 6th from last AES round key.  (Using signed
+	// byte offsets -7*16 through 6*16 decreases code size.)
+	lea		OFFSETOF_AESROUNDKEYS(KEY,AESKEYLEN64,4), %rax
+
+	// AES-encrypt the counter block and also multiply GHASH_ACC by H^1.
+	// Interleave the AES and GHASH instructions to improve performance.
+	pshufb		BSWAP_MASK, %xmm0
+	pxor		OFFSETOF_AESROUNDKEYS(KEY), %xmm0
+	cmp		$24, AESKEYLEN
+	jl		128f	// AES-128?
+	je		192f	// AES-192?
+	// AES-256
+	aesenc		-7*16(%rax), %xmm0
+	aesenc		-6*16(%rax), %xmm0
+192:
+	aesenc		-5*16(%rax), %xmm0
+	aesenc		-4*16(%rax), %xmm0
+128:
+.irp i, 0,1,2,3,4,5,6,7,8
+	aesenc		(\i-3)*16(%rax), %xmm0
+	_ghash_mul_step	\i, H_POW1, H_POW1_X64, GHASH_ACC, GFPOLY, %xmm1, %xmm2
+.endr
+	aesenclast	6*16(%rax), %xmm0
+	_ghash_mul_step	9, H_POW1, H_POW1_X64, GHASH_ACC, GFPOLY, %xmm1, %xmm2
+
+	// Undo the byte reflection of the GHASH accumulator.
+	pshufb		BSWAP_MASK, GHASH_ACC
+
+	// Encrypt the GHASH accumulator.
+	pxor		%xmm0, GHASH_ACC
+
+.if \enc
+	// Return the computed auth tag.
+	movdqu		GHASH_ACC, (GHASH_ACC_PTR)
+.else
+	.set		ZEROPAD_MASK_PTR, TOTAL_AADLEN // Reusing TOTAL_AADLEN!
+
+	// Verify the auth tag in constant time by XOR'ing the transmitted and
+	// computed auth tags together and using the ptest instruction to check
+	// whether the first TAGLEN bytes of the result are zero.
+	_xor_mem_to_reg	(TAG), GHASH_ACC, tmp=%xmm0
+	movl		8(%rsp), TAGLEN
+	lea		.Lzeropad_mask+16(%rip), ZEROPAD_MASK_PTR
+	sub		TAGLEN64, ZEROPAD_MASK_PTR
+	xor		%eax, %eax
+	_test_mem	(ZEROPAD_MASK_PTR), GHASH_ACC, tmp=%xmm0
+	sete		%al
+.endif
+	RET
+.endm
+
+.set	USE_AVX, 0
+SYM_FUNC_START(aes_gcm_precompute_aesni)
+	_aes_gcm_precompute
+SYM_FUNC_END(aes_gcm_precompute_aesni)
+SYM_FUNC_START(aes_gcm_aad_update_aesni)
+	_aes_gcm_aad_update
+SYM_FUNC_END(aes_gcm_aad_update_aesni)
+SYM_FUNC_START(aes_gcm_enc_update_aesni)
+	_aes_gcm_update	1
+SYM_FUNC_END(aes_gcm_enc_update_aesni)
+SYM_FUNC_START(aes_gcm_dec_update_aesni)
+	_aes_gcm_update	0
+SYM_FUNC_END(aes_gcm_dec_update_aesni)
+SYM_FUNC_START(aes_gcm_enc_final_aesni)
+	_aes_gcm_final	1
+SYM_FUNC_END(aes_gcm_enc_final_aesni)
+SYM_FUNC_START(aes_gcm_dec_final_aesni)
+	_aes_gcm_final	0
+SYM_FUNC_END(aes_gcm_dec_final_aesni)
+
+.set	USE_AVX, 1
+SYM_FUNC_START(aes_gcm_precompute_aesni_avx)
+	_aes_gcm_precompute
+SYM_FUNC_END(aes_gcm_precompute_aesni_avx)
+SYM_FUNC_START(aes_gcm_aad_update_aesni_avx)
+	_aes_gcm_aad_update
+SYM_FUNC_END(aes_gcm_aad_update_aesni_avx)
+SYM_FUNC_START(aes_gcm_enc_update_aesni_avx)
+	_aes_gcm_update	1
+SYM_FUNC_END(aes_gcm_enc_update_aesni_avx)
+SYM_FUNC_START(aes_gcm_dec_update_aesni_avx)
+	_aes_gcm_update	0
+SYM_FUNC_END(aes_gcm_dec_update_aesni_avx)
+SYM_FUNC_START(aes_gcm_enc_final_aesni_avx)
+	_aes_gcm_final	1
+SYM_FUNC_END(aes_gcm_enc_final_aesni_avx)
+SYM_FUNC_START(aes_gcm_dec_final_aesni_avx)
+	_aes_gcm_final	0
+SYM_FUNC_END(aes_gcm_dec_final_aesni_avx)
diff --git a/arch/x86/crypto/aes-gcm-vaes-avx2.S b/arch/x86/crypto/aes-gcm-vaes-avx2.S
new file mode 100644
index 000000000..9cc387957
--- /dev/null
+++ b/arch/x86/crypto/aes-gcm-vaes-avx2.S
@@ -0,0 +1,1147 @@
+/* SPDX-License-Identifier: Apache-2.0 OR BSD-2-Clause */
+//
+// AES-GCM implementation for x86_64 CPUs that support the following CPU
+// features: VAES && VPCLMULQDQ && AVX2
+//
+// Copyright 2025 Google LLC
+//
+// Author: Eric Biggers <ebiggers@google.com>
+//
+//------------------------------------------------------------------------------
+//
+// This file is dual-licensed, meaning that you can use it under your choice of
+// either of the following two licenses:
+//
+// Licensed under the Apache License 2.0 (the "License").  You may obtain a copy
+// of the License at
+//
+//	http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// or
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are met:
+//
+// 1. Redistributions of source code must retain the above copyright notice,
+//    this list of conditions and the following disclaimer.
+//
+// 2. Redistributions in binary form must reproduce the above copyright
+//    notice, this list of conditions and the following disclaimer in the
+//    documentation and/or other materials provided with the distribution.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+// POSSIBILITY OF SUCH DAMAGE.
+//
+// -----------------------------------------------------------------------------
+//
+// This is similar to aes-gcm-vaes-avx512.S, but it uses AVX2 instead of AVX512.
+// This means it can only use 16 vector registers instead of 32, the maximum
+// vector length is 32 bytes, and some instructions such as vpternlogd and
+// masked loads/stores are unavailable.  However, it is able to run on CPUs that
+// have VAES without AVX512, namely AMD Zen 3 (including "Milan" server CPUs),
+// various Intel client CPUs such as Alder Lake, and Intel Sierra Forest.
+//
+// This implementation also uses Karatsuba multiplication instead of schoolbook
+// multiplication for GHASH in its main loop.  This does not help much on Intel,
+// but it improves performance by ~5% on AMD Zen 3.  Other factors weighing
+// slightly in favor of Karatsuba multiplication in this implementation are the
+// lower maximum vector length (which means there are fewer key powers, so we
+// can cache the halves of each key power XOR'd together and still use less
+// memory than the AVX512 implementation), and the unavailability of the
+// vpternlogd instruction (which helped schoolbook a bit more than Karatsuba).
+
+#include <linux/linkage.h>
+
+.section .rodata
+.p2align 4
+
+	// The below three 16-byte values must be in the order that they are, as
+	// they are really two 32-byte tables and a 16-byte value that overlap:
+	//
+	// - The first 32-byte table begins at .Lselect_high_bytes_table.
+	//   For 0 <= len <= 16, the 16-byte value at
+	//   '.Lselect_high_bytes_table + len' selects the high 'len' bytes of
+	//   another 16-byte value when AND'ed with it.
+	//
+	// - The second 32-byte table begins at .Lrshift_and_bswap_table.
+	//   For 0 <= len <= 16, the 16-byte value at
+	//   '.Lrshift_and_bswap_table + len' is a vpshufb mask that does the
+	//   following operation: right-shift by '16 - len' bytes (shifting in
+	//   zeroes), then reflect all 16 bytes.
+	//
+	// - The 16-byte value at .Lbswap_mask is a vpshufb mask that reflects
+	//   all 16 bytes.
+.Lselect_high_bytes_table:
+	.octa	0
+.Lrshift_and_bswap_table:
+	.octa	0xffffffffffffffffffffffffffffffff
+.Lbswap_mask:
+	.octa	0x000102030405060708090a0b0c0d0e0f
+
+	// Sixteen 0x0f bytes.  By XOR'ing an entry of .Lrshift_and_bswap_table
+	// with this, we get a mask that left-shifts by '16 - len' bytes.
+.Lfifteens:
+	.octa	0x0f0f0f0f0f0f0f0f0f0f0f0f0f0f0f0f
+
+	// This is the GHASH reducing polynomial without its constant term, i.e.
+	// x^128 + x^7 + x^2 + x, represented using the backwards mapping
+	// between bits and polynomial coefficients.
+	//
+	// Alternatively, it can be interpreted as the naturally-ordered
+	// representation of the polynomial x^127 + x^126 + x^121 + 1, i.e. the
+	// "reversed" GHASH reducing polynomial without its x^128 term.
+.Lgfpoly:
+	.octa	0xc2000000000000000000000000000001
+
+	// Same as above, but with the (1 << 64) bit set.
+.Lgfpoly_and_internal_carrybit:
+	.octa	0xc2000000000000010000000000000001
+
+	// Values needed to prepare the initial vector of counter blocks.
+.Lctr_pattern:
+	.octa	0
+	.octa	1
+
+	// The number of AES blocks per vector, as a 128-bit value.
+.Linc_2blocks:
+	.octa	2
+
+// Offsets in struct aes_gcm_key_vaes_avx2
+#define OFFSETOF_AESKEYLEN	0
+#define OFFSETOF_AESROUNDKEYS	16
+#define OFFSETOF_H_POWERS	288
+#define NUM_H_POWERS		8
+#define OFFSETOFEND_H_POWERS    (OFFSETOF_H_POWERS + (NUM_H_POWERS * 16))
+#define OFFSETOF_H_POWERS_XORED	OFFSETOFEND_H_POWERS
+
+.text
+
+// Do one step of GHASH-multiplying the 128-bit lanes of \a by the 128-bit lanes
+// of \b and storing the reduced products in \dst.  Uses schoolbook
+// multiplication.
+.macro	_ghash_mul_step	i, a, b, dst, gfpoly, t0, t1, t2
+.if \i == 0
+	vpclmulqdq	$0x00, \a, \b, \t0	  // LO = a_L * b_L
+	vpclmulqdq	$0x01, \a, \b, \t1	  // MI_0 = a_L * b_H
+.elseif \i == 1
+	vpclmulqdq	$0x10, \a, \b, \t2	  // MI_1 = a_H * b_L
+.elseif \i == 2
+	vpxor		\t2, \t1, \t1		  // MI = MI_0 + MI_1
+.elseif \i == 3
+	vpclmulqdq	$0x01, \t0, \gfpoly, \t2  // LO_L*(x^63 + x^62 + x^57)
+.elseif \i == 4
+	vpshufd		$0x4e, \t0, \t0		  // Swap halves of LO
+.elseif \i == 5
+	vpxor		\t0, \t1, \t1		  // Fold LO into MI (part 1)
+	vpxor		\t2, \t1, \t1		  // Fold LO into MI (part 2)
+.elseif \i == 6
+	vpclmulqdq	$0x11, \a, \b, \dst	  // HI = a_H * b_H
+.elseif \i == 7
+	vpclmulqdq	$0x01, \t1, \gfpoly, \t0  // MI_L*(x^63 + x^62 + x^57)
+.elseif \i == 8
+	vpshufd		$0x4e, \t1, \t1		  // Swap halves of MI
+.elseif \i == 9
+	vpxor		\t1, \dst, \dst		  // Fold MI into HI (part 1)
+	vpxor		\t0, \dst, \dst		  // Fold MI into HI (part 2)
+.endif
+.endm
+
+// GHASH-multiply the 128-bit lanes of \a by the 128-bit lanes of \b and store
+// the reduced products in \dst.  See _ghash_mul_step for full explanation.
+.macro	_ghash_mul	a, b, dst, gfpoly, t0, t1, t2
+.irp i, 0,1,2,3,4,5,6,7,8,9
+	_ghash_mul_step	\i, \a, \b, \dst, \gfpoly, \t0, \t1, \t2
+.endr
+.endm
+
+// GHASH-multiply the 128-bit lanes of \a by the 128-bit lanes of \b and add the
+// *unreduced* products to \lo, \mi, and \hi.
+.macro	_ghash_mul_noreduce	a, b, lo, mi, hi, t0
+	vpclmulqdq	$0x00, \a, \b, \t0	// a_L * b_L
+	vpxor		\t0, \lo, \lo
+	vpclmulqdq	$0x01, \a, \b, \t0	// a_L * b_H
+	vpxor		\t0, \mi, \mi
+	vpclmulqdq	$0x10, \a, \b, \t0	// a_H * b_L
+	vpxor		\t0, \mi, \mi
+	vpclmulqdq	$0x11, \a, \b, \t0	// a_H * b_H
+	vpxor		\t0, \hi, \hi
+.endm
+
+// Reduce the unreduced products from \lo, \mi, and \hi and store the 128-bit
+// reduced products in \hi.  See _ghash_mul_step for explanation of reduction.
+.macro	_ghash_reduce	lo, mi, hi, gfpoly, t0
+	vpclmulqdq	$0x01, \lo, \gfpoly, \t0
+	vpshufd		$0x4e, \lo, \lo
+	vpxor		\lo, \mi, \mi
+	vpxor		\t0, \mi, \mi
+	vpclmulqdq	$0x01, \mi, \gfpoly, \t0
+	vpshufd		$0x4e, \mi, \mi
+	vpxor		\mi, \hi, \hi
+	vpxor		\t0, \hi, \hi
+.endm
+
+// This is a specialized version of _ghash_mul that computes \a * \a, i.e. it
+// squares \a.  It skips computing MI = (a_L * a_H) + (a_H * a_L) = 0.
+.macro	_ghash_square	a, dst, gfpoly, t0, t1
+	vpclmulqdq	$0x00, \a, \a, \t0	  // LO = a_L * a_L
+	vpclmulqdq	$0x11, \a, \a, \dst	  // HI = a_H * a_H
+	vpclmulqdq	$0x01, \t0, \gfpoly, \t1  // LO_L*(x^63 + x^62 + x^57)
+	vpshufd		$0x4e, \t0, \t0		  // Swap halves of LO
+	vpxor		\t0, \t1, \t1		  // Fold LO into MI
+	vpclmulqdq	$0x01, \t1, \gfpoly, \t0  // MI_L*(x^63 + x^62 + x^57)
+	vpshufd		$0x4e, \t1, \t1		  // Swap halves of MI
+	vpxor		\t1, \dst, \dst		  // Fold MI into HI (part 1)
+	vpxor		\t0, \dst, \dst		  // Fold MI into HI (part 2)
+.endm
+
+// void aes_gcm_precompute_vaes_avx2(struct aes_gcm_key_vaes_avx2 *key);
+//
+// Given the expanded AES key |key->base.aes_key|, derive the GHASH subkey and
+// initialize |key->h_powers| and |key->h_powers_xored|.
+//
+// We use h_powers[0..7] to store H^8 through H^1, and h_powers_xored[0..7] to
+// store the 64-bit halves of the key powers XOR'd together (for Karatsuba
+// multiplication) in the order 8,6,7,5,4,2,3,1.
+SYM_FUNC_START(aes_gcm_precompute_vaes_avx2)
+
+	// Function arguments
+	.set	KEY,		%rdi
+
+	// Additional local variables
+	.set	POWERS_PTR,	%rsi
+	.set	RNDKEYLAST_PTR,	%rdx
+	.set	TMP0,		%ymm0
+	.set	TMP0_XMM,	%xmm0
+	.set	TMP1,		%ymm1
+	.set	TMP1_XMM,	%xmm1
+	.set	TMP2,		%ymm2
+	.set	TMP2_XMM,	%xmm2
+	.set	H_CUR,		%ymm3
+	.set	H_CUR_XMM,	%xmm3
+	.set	H_CUR2,		%ymm4
+	.set	H_INC,		%ymm5
+	.set	H_INC_XMM,	%xmm5
+	.set	GFPOLY,		%ymm6
+	.set	GFPOLY_XMM,	%xmm6
+
+	// Encrypt an all-zeroes block to get the raw hash subkey.
+	movl		OFFSETOF_AESKEYLEN(KEY), %eax
+	lea		OFFSETOF_AESROUNDKEYS+6*16(KEY,%rax,4), RNDKEYLAST_PTR
+	vmovdqu		OFFSETOF_AESROUNDKEYS(KEY), H_CUR_XMM
+	lea		OFFSETOF_AESROUNDKEYS+16(KEY), %rax
+1:
+	vaesenc		(%rax), H_CUR_XMM, H_CUR_XMM
+	add		$16, %rax
+	cmp		%rax, RNDKEYLAST_PTR
+	jne		1b
+	vaesenclast	(RNDKEYLAST_PTR), H_CUR_XMM, H_CUR_XMM
+
+	// Reflect the bytes of the raw hash subkey.
+	vpshufb		.Lbswap_mask(%rip), H_CUR_XMM, H_CUR_XMM
+
+	// Finish preprocessing the byte-reflected hash subkey by multiplying it
+	// by x^-1 ("standard" interpretation of polynomial coefficients) or
+	// equivalently x^1 (natural interpretation).  This gets the key into a
+	// format that avoids having to bit-reflect the data blocks later.
+	vpshufd		$0xd3, H_CUR_XMM, TMP0_XMM
+	vpsrad		$31, TMP0_XMM, TMP0_XMM
+	vpaddq		H_CUR_XMM, H_CUR_XMM, H_CUR_XMM
+	vpand		.Lgfpoly_and_internal_carrybit(%rip), TMP0_XMM, TMP0_XMM
+	vpxor		TMP0_XMM, H_CUR_XMM, H_CUR_XMM
+
+	// Load the gfpoly constant.
+	vbroadcasti128	.Lgfpoly(%rip), GFPOLY
+
+	// Square H^1 to get H^2.
+	_ghash_square	H_CUR_XMM, H_INC_XMM, GFPOLY_XMM, TMP0_XMM, TMP1_XMM
+
+	// Create H_CUR = [H^2, H^1] and H_INC = [H^2, H^2].
+	vinserti128	$1, H_CUR_XMM, H_INC, H_CUR
+	vinserti128	$1, H_INC_XMM, H_INC, H_INC
+
+	// Compute H_CUR2 = [H^4, H^3].
+	_ghash_mul	H_INC, H_CUR, H_CUR2, GFPOLY, TMP0, TMP1, TMP2
+
+	// Store [H^2, H^1] and [H^4, H^3].
+	vmovdqu		H_CUR, OFFSETOF_H_POWERS+3*32(KEY)
+	vmovdqu		H_CUR2, OFFSETOF_H_POWERS+2*32(KEY)
+
+	// For Karatsuba multiplication: compute and store the two 64-bit halves
+	// of each key power XOR'd together.  Order is 4,2,3,1.
+	vpunpcklqdq	H_CUR, H_CUR2, TMP0
+	vpunpckhqdq	H_CUR, H_CUR2, TMP1
+	vpxor		TMP1, TMP0, TMP0
+	vmovdqu		TMP0, OFFSETOF_H_POWERS_XORED+32(KEY)
+
+	// Compute and store H_CUR = [H^6, H^5] and H_CUR2 = [H^8, H^7].
+	_ghash_mul	H_INC, H_CUR2, H_CUR, GFPOLY, TMP0, TMP1, TMP2
+	_ghash_mul	H_INC, H_CUR, H_CUR2, GFPOLY, TMP0, TMP1, TMP2
+	vmovdqu		H_CUR, OFFSETOF_H_POWERS+1*32(KEY)
+	vmovdqu		H_CUR2, OFFSETOF_H_POWERS+0*32(KEY)
+
+	// Again, compute and store the two 64-bit halves of each key power
+	// XOR'd together.  Order is 8,6,7,5.
+	vpunpcklqdq	H_CUR, H_CUR2, TMP0
+	vpunpckhqdq	H_CUR, H_CUR2, TMP1
+	vpxor		TMP1, TMP0, TMP0
+	vmovdqu		TMP0, OFFSETOF_H_POWERS_XORED(KEY)
+
+	vzeroupper
+	RET
+SYM_FUNC_END(aes_gcm_precompute_vaes_avx2)
+
+// Do one step of the GHASH update of four vectors of data blocks.
+//   \i: the step to do, 0 through 9
+//   \ghashdata_ptr: pointer to the data blocks (ciphertext or AAD)
+//   KEY: pointer to struct aes_gcm_key_vaes_avx2
+//   BSWAP_MASK: mask for reflecting the bytes of blocks
+//   H_POW[2-1]_XORED: cached values from KEY->h_powers_xored
+//   TMP[0-2]: temporary registers.  TMP[1-2] must be preserved across steps.
+//   LO, MI: working state for this macro that must be preserved across steps
+//   GHASH_ACC: the GHASH accumulator (input/output)
+.macro	_ghash_step_4x	i, ghashdata_ptr
+	.set		HI, GHASH_ACC # alias
+	.set		HI_XMM, GHASH_ACC_XMM
+.if \i == 0
+	// First vector
+	vmovdqu		0*32(\ghashdata_ptr), TMP1
+	vpshufb		BSWAP_MASK, TMP1, TMP1
+	vmovdqu		OFFSETOF_H_POWERS+0*32(KEY), TMP2
+	vpxor		GHASH_ACC, TMP1, TMP1
+	vpclmulqdq	$0x00, TMP2, TMP1, LO
+	vpclmulqdq	$0x11, TMP2, TMP1, HI
+	vpunpckhqdq	TMP1, TMP1, TMP0
+	vpxor		TMP1, TMP0, TMP0
+	vpclmulqdq	$0x00, H_POW2_XORED, TMP0, MI
+.elseif \i == 1
+.elseif \i == 2
+	// Second vector
+	vmovdqu		1*32(\ghashdata_ptr), TMP1
+	vpshufb		BSWAP_MASK, TMP1, TMP1
+	vmovdqu		OFFSETOF_H_POWERS+1*32(KEY), TMP2
+	vpclmulqdq	$0x00, TMP2, TMP1, TMP0
+	vpxor		TMP0, LO, LO
+	vpclmulqdq	$0x11, TMP2, TMP1, TMP0
+	vpxor		TMP0, HI, HI
+	vpunpckhqdq	TMP1, TMP1, TMP0
+	vpxor		TMP1, TMP0, TMP0
+	vpclmulqdq	$0x10, H_POW2_XORED, TMP0, TMP0
+	vpxor		TMP0, MI, MI
+.elseif \i == 3
+	// Third vector
+	vmovdqu		2*32(\ghashdata_ptr), TMP1
+	vpshufb		BSWAP_MASK, TMP1, TMP1
+	vmovdqu		OFFSETOF_H_POWERS+2*32(KEY), TMP2
+.elseif \i == 4
+	vpclmulqdq	$0x00, TMP2, TMP1, TMP0
+	vpxor		TMP0, LO, LO
+	vpclmulqdq	$0x11, TMP2, TMP1, TMP0
+	vpxor		TMP0, HI, HI
+.elseif \i == 5
+	vpunpckhqdq	TMP1, TMP1, TMP0
+	vpxor		TMP1, TMP0, TMP0
+	vpclmulqdq	$0x00, H_POW1_XORED, TMP0, TMP0
+	vpxor		TMP0, MI, MI
+
+	// Fourth vector
+	vmovdqu		3*32(\ghashdata_ptr), TMP1
+	vpshufb		BSWAP_MASK, TMP1, TMP1
+.elseif \i == 6
+	vmovdqu		OFFSETOF_H_POWERS+3*32(KEY), TMP2
+	vpclmulqdq	$0x00, TMP2, TMP1, TMP0
+	vpxor		TMP0, LO, LO
+	vpclmulqdq	$0x11, TMP2, TMP1, TMP0
+	vpxor		TMP0, HI, HI
+	vpunpckhqdq	TMP1, TMP1, TMP0
+	vpxor		TMP1, TMP0, TMP0
+	vpclmulqdq	$0x10, H_POW1_XORED, TMP0, TMP0
+	vpxor		TMP0, MI, MI
+.elseif \i == 7
+	// Finalize 'mi' following Karatsuba multiplication.
+	vpxor		LO, MI, MI
+	vpxor		HI, MI, MI
+
+	// Fold lo into mi.
+	vbroadcasti128	.Lgfpoly(%rip), TMP2
+	vpclmulqdq	$0x01, LO, TMP2, TMP0
+	vpshufd		$0x4e, LO, LO
+	vpxor		LO, MI, MI
+	vpxor		TMP0, MI, MI
+.elseif \i == 8
+	// Fold mi into hi.
+	vpclmulqdq	$0x01, MI, TMP2, TMP0
+	vpshufd		$0x4e, MI, MI
+	vpxor		MI, HI, HI
+	vpxor		TMP0, HI, HI
+.elseif \i == 9
+	vextracti128	$1, HI, TMP0_XMM
+	vpxor		TMP0_XMM, HI_XMM, GHASH_ACC_XMM
+.endif
+.endm
+
+// Update GHASH with four vectors of data blocks.  See _ghash_step_4x for full
+// explanation.
+.macro	_ghash_4x	ghashdata_ptr
+.irp i, 0,1,2,3,4,5,6,7,8,9
+	_ghash_step_4x	\i, \ghashdata_ptr
+.endr
+.endm
+
+// Load 1 <= %ecx <= 16 bytes from the pointer \src into the xmm register \dst
+// and zeroize any remaining bytes.  Clobbers %rax, %rcx, and \tmp{64,32}.
+.macro	_load_partial_block	src, dst, tmp64, tmp32
+	sub		$8, %ecx		// LEN - 8
+	jle		.Lle8\@
+
+	// Load 9 <= LEN <= 16 bytes.
+	vmovq		(\src), \dst		// Load first 8 bytes
+	mov		(\src, %rcx), %rax	// Load last 8 bytes
+	neg		%ecx
+	shl		$3, %ecx
+	shr		%cl, %rax		// Discard overlapping bytes
+	vpinsrq		$1, %rax, \dst, \dst
+	jmp		.Ldone\@
+
+.Lle8\@:
+	add		$4, %ecx		// LEN - 4
+	jl		.Llt4\@
+
+	// Load 4 <= LEN <= 8 bytes.
+	mov		(\src), %eax		// Load first 4 bytes
+	mov		(\src, %rcx), \tmp32	// Load last 4 bytes
+	jmp		.Lcombine\@
+
+.Llt4\@:
+	// Load 1 <= LEN <= 3 bytes.
+	add		$2, %ecx		// LEN - 2
+	movzbl		(\src), %eax		// Load first byte
+	jl		.Lmovq\@
+	movzwl		(\src, %rcx), \tmp32	// Load last 2 bytes
+.Lcombine\@:
+	shl		$3, %ecx
+	shl		%cl, \tmp64
+	or		\tmp64, %rax		// Combine the two parts
+.Lmovq\@:
+	vmovq		%rax, \dst
+.Ldone\@:
+.endm
+
+// Store 1 <= %ecx <= 16 bytes from the xmm register \src to the pointer \dst.
+// Clobbers %rax, %rcx, and \tmp{64,32}.
+.macro	_store_partial_block	src, dst, tmp64, tmp32
+	sub		$8, %ecx		// LEN - 8
+	jl		.Llt8\@
+
+	// Store 8 <= LEN <= 16 bytes.
+	vpextrq		$1, \src, %rax
+	mov		%ecx, \tmp32
+	shl		$3, %ecx
+	ror		%cl, %rax
+	mov		%rax, (\dst, \tmp64)	// Store last LEN - 8 bytes
+	vmovq		\src, (\dst)		// Store first 8 bytes
+	jmp		.Ldone\@
+
+.Llt8\@:
+	add		$4, %ecx		// LEN - 4
+	jl		.Llt4\@
+
+	// Store 4 <= LEN <= 7 bytes.
+	vpextrd		$1, \src, %eax
+	mov		%ecx, \tmp32
+	shl		$3, %ecx
+	ror		%cl, %eax
+	mov		%eax, (\dst, \tmp64)	// Store last LEN - 4 bytes
+	vmovd		\src, (\dst)		// Store first 4 bytes
+	jmp		.Ldone\@
+
+.Llt4\@:
+	// Store 1 <= LEN <= 3 bytes.
+	vpextrb		$0, \src, 0(\dst)
+	cmp		$-2, %ecx		// LEN - 4 == -2, i.e. LEN == 2?
+	jl		.Ldone\@
+	vpextrb		$1, \src, 1(\dst)
+	je		.Ldone\@
+	vpextrb		$2, \src, 2(\dst)
+.Ldone\@:
+.endm
+
+// void aes_gcm_aad_update_vaes_avx2(const struct aes_gcm_key_vaes_avx2 *key,
+//				     u8 ghash_acc[16],
+//				     const u8 *aad, int aadlen);
+//
+// This function processes the AAD (Additional Authenticated Data) in GCM.
+// Using the key |key|, it updates the GHASH accumulator |ghash_acc| with the
+// data given by |aad| and |aadlen|.  On the first call, |ghash_acc| must be all
+// zeroes.  |aadlen| must be a multiple of 16, except on the last call where it
+// can be any length.  The caller must do any buffering needed to ensure this.
+//
+// This handles large amounts of AAD efficiently, while also keeping overhead
+// low for small amounts which is the common case.  TLS and IPsec use less than
+// one block of AAD, but (uncommonly) other use cases may use much more.
+SYM_FUNC_START(aes_gcm_aad_update_vaes_avx2)
+
+	// Function arguments
+	.set	KEY,		%rdi
+	.set	GHASH_ACC_PTR,	%rsi
+	.set	AAD,		%rdx
+	.set	AADLEN,		%ecx	// Must be %ecx for _load_partial_block
+	.set	AADLEN64,	%rcx	// Zero-extend AADLEN before using!
+
+	// Additional local variables.
+	// %rax and %r8 are used as temporary registers.
+	.set	TMP0,		%ymm0
+	.set	TMP0_XMM,	%xmm0
+	.set	TMP1,		%ymm1
+	.set	TMP1_XMM,	%xmm1
+	.set	TMP2,		%ymm2
+	.set	TMP2_XMM,	%xmm2
+	.set	LO,		%ymm3
+	.set	LO_XMM,		%xmm3
+	.set	MI,		%ymm4
+	.set	MI_XMM,		%xmm4
+	.set	GHASH_ACC,	%ymm5
+	.set	GHASH_ACC_XMM,	%xmm5
+	.set	BSWAP_MASK,	%ymm6
+	.set	BSWAP_MASK_XMM,	%xmm6
+	.set	GFPOLY,		%ymm7
+	.set	GFPOLY_XMM,	%xmm7
+	.set	H_POW2_XORED,	%ymm8
+	.set	H_POW1_XORED,	%ymm9
+
+	// Load the bswap_mask and gfpoly constants.  Since AADLEN is usually
+	// small, usually only 128-bit vectors will be used.  So as an
+	// optimization, don't broadcast these constants to both 128-bit lanes
+	// quite yet.
+	vmovdqu		.Lbswap_mask(%rip), BSWAP_MASK_XMM
+	vmovdqu		.Lgfpoly(%rip), GFPOLY_XMM
+
+	// Load the GHASH accumulator.
+	vmovdqu		(GHASH_ACC_PTR), GHASH_ACC_XMM
+
+	// Check for the common case of AADLEN <= 16, as well as AADLEN == 0.
+	test		AADLEN, AADLEN
+	jz		.Laad_done
+	cmp		$16, AADLEN
+	jle		.Laad_lastblock
+
+	// AADLEN > 16, so we'll operate on full vectors.  Broadcast bswap_mask
+	// and gfpoly to both 128-bit lanes.
+	vinserti128	$1, BSWAP_MASK_XMM, BSWAP_MASK, BSWAP_MASK
+	vinserti128	$1, GFPOLY_XMM, GFPOLY, GFPOLY
+
+	// If AADLEN >= 128, update GHASH with 128 bytes of AAD at a time.
+	add		$-128, AADLEN	// 128 is 4 bytes, -128 is 1 byte
+	jl		.Laad_loop_4x_done
+	vmovdqu		OFFSETOF_H_POWERS_XORED(KEY), H_POW2_XORED
+	vmovdqu		OFFSETOF_H_POWERS_XORED+32(KEY), H_POW1_XORED
+.Laad_loop_4x:
+	_ghash_4x	AAD
+	sub		$-128, AAD
+	add		$-128, AADLEN
+	jge		.Laad_loop_4x
+.Laad_loop_4x_done:
+
+	// If AADLEN >= 32, update GHASH with 32 bytes of AAD at a time.
+	add		$96, AADLEN
+	jl		.Laad_loop_1x_done
+.Laad_loop_1x:
+	vmovdqu		(AAD), TMP0
+	vpshufb		BSWAP_MASK, TMP0, TMP0
+	vpxor		TMP0, GHASH_ACC, GHASH_ACC
+	vmovdqu		OFFSETOFEND_H_POWERS-32(KEY), TMP0
+	_ghash_mul	TMP0, GHASH_ACC, GHASH_ACC, GFPOLY, TMP1, TMP2, LO
+	vextracti128	$1, GHASH_ACC, TMP0_XMM
+	vpxor		TMP0_XMM, GHASH_ACC_XMM, GHASH_ACC_XMM
+	add		$32, AAD
+	sub		$32, AADLEN
+	jge		.Laad_loop_1x
+.Laad_loop_1x_done:
+	add		$32, AADLEN
+	// Now 0 <= AADLEN < 32.
+
+	jz		.Laad_done
+	cmp		$16, AADLEN
+	jle		.Laad_lastblock
+
+	// Update GHASH with the remaining 17 <= AADLEN <= 31 bytes of AAD.
+	mov		AADLEN, AADLEN	// Zero-extend AADLEN to AADLEN64.
+	vmovdqu		(AAD), TMP0_XMM
+	vmovdqu		-16(AAD, AADLEN64), TMP1_XMM
+	vpshufb		BSWAP_MASK_XMM, TMP0_XMM, TMP0_XMM
+	vpxor		TMP0_XMM, GHASH_ACC_XMM, GHASH_ACC_XMM
+	lea		.Lrshift_and_bswap_table(%rip), %rax
+	vpshufb		-16(%rax, AADLEN64), TMP1_XMM, TMP1_XMM
+	vinserti128	$1, TMP1_XMM, GHASH_ACC, GHASH_ACC
+	vmovdqu		OFFSETOFEND_H_POWERS-32(KEY), TMP0
+	_ghash_mul	TMP0, GHASH_ACC, GHASH_ACC, GFPOLY, TMP1, TMP2, LO
+	vextracti128	$1, GHASH_ACC, TMP0_XMM
+	vpxor		TMP0_XMM, GHASH_ACC_XMM, GHASH_ACC_XMM
+	jmp		.Laad_done
+
+.Laad_lastblock:
+	// Update GHASH with the remaining 1 <= AADLEN <= 16 bytes of AAD.
+	_load_partial_block	AAD, TMP0_XMM, %r8, %r8d
+	vpshufb		BSWAP_MASK_XMM, TMP0_XMM, TMP0_XMM
+	vpxor		TMP0_XMM, GHASH_ACC_XMM, GHASH_ACC_XMM
+	vmovdqu		OFFSETOFEND_H_POWERS-16(KEY), TMP0_XMM
+	_ghash_mul	TMP0_XMM, GHASH_ACC_XMM, GHASH_ACC_XMM, GFPOLY_XMM, \
+			TMP1_XMM, TMP2_XMM, LO_XMM
+
+.Laad_done:
+	// Store the updated GHASH accumulator back to memory.
+	vmovdqu		GHASH_ACC_XMM, (GHASH_ACC_PTR)
+
+	vzeroupper
+	RET
+SYM_FUNC_END(aes_gcm_aad_update_vaes_avx2)
+
+// Do one non-last round of AES encryption on the blocks in the given AESDATA
+// vectors using the round key that has been broadcast to all 128-bit lanes of
+// \round_key.
+.macro	_vaesenc	round_key, vecs:vararg
+.irp i, \vecs
+	vaesenc		\round_key, AESDATA\i, AESDATA\i
+.endr
+.endm
+
+// Generate counter blocks in the given AESDATA vectors, then do the zero-th AES
+// round on them.  Clobbers TMP0.
+.macro	_ctr_begin	vecs:vararg
+	vbroadcasti128	.Linc_2blocks(%rip), TMP0
+.irp i, \vecs
+	vpshufb		BSWAP_MASK, LE_CTR, AESDATA\i
+	vpaddd		TMP0, LE_CTR, LE_CTR
+.endr
+.irp i, \vecs
+	vpxor		RNDKEY0, AESDATA\i, AESDATA\i
+.endr
+.endm
+
+// Generate and encrypt counter blocks in the given AESDATA vectors, excluding
+// the last AES round.  Clobbers %rax and TMP0.
+.macro	_aesenc_loop	vecs:vararg
+	_ctr_begin	\vecs
+	lea		OFFSETOF_AESROUNDKEYS+16(KEY), %rax
+.Laesenc_loop\@:
+	vbroadcasti128	(%rax), TMP0
+	_vaesenc	TMP0, \vecs
+	add		$16, %rax
+	cmp		%rax, RNDKEYLAST_PTR
+	jne		.Laesenc_loop\@
+.endm
+
+// Finalize the keystream blocks in the given AESDATA vectors by doing the last
+// AES round, then XOR those keystream blocks with the corresponding data.
+// Reduce latency by doing the XOR before the vaesenclast, utilizing the
+// property vaesenclast(key, a) ^ b == vaesenclast(key ^ b, a).  Clobbers TMP0.
+.macro	_aesenclast_and_xor	vecs:vararg
+.irp i, \vecs
+	vpxor		\i*32(SRC), RNDKEYLAST, TMP0
+	vaesenclast	TMP0, AESDATA\i, AESDATA\i
+.endr
+.irp i, \vecs
+	vmovdqu		AESDATA\i, \i*32(DST)
+.endr
+.endm
+
+// void aes_gcm_{enc,dec}_update_vaes_avx2(const struct aes_gcm_key_vaes_avx2 *key,
+//					   const u32 le_ctr[4], u8 ghash_acc[16],
+//					   const u8 *src, u8 *dst, int datalen);
+//
+// This macro generates a GCM encryption or decryption update function with the
+// above prototype (with \enc selecting which one).  The function computes the
+// next portion of the CTR keystream, XOR's it with |datalen| bytes from |src|,
+// and writes the resulting encrypted or decrypted data to |dst|.  It also
+// updates the GHASH accumulator |ghash_acc| using the next |datalen| ciphertext
+// bytes.
+//
+// |datalen| must be a multiple of 16, except on the last call where it can be
+// any length.  The caller must do any buffering needed to ensure this.  Both
+// in-place and out-of-place en/decryption are supported.
+//
+// |le_ctr| must give the current counter in little-endian format.  This
+// function loads the counter from |le_ctr| and increments the loaded counter as
+// needed, but it does *not* store the updated counter back to |le_ctr|.  The
+// caller must update |le_ctr| if any more data segments follow.  Internally,
+// only the low 32-bit word of the counter is incremented, following the GCM
+// standard.
+.macro	_aes_gcm_update	enc
+
+	// Function arguments
+	.set	KEY,		%rdi
+	.set	LE_CTR_PTR,	%rsi
+	.set	LE_CTR_PTR32,	%esi
+	.set	GHASH_ACC_PTR,	%rdx
+	.set	SRC,		%rcx	// Assumed to be %rcx.
+					// See .Ltail_xor_and_ghash_1to16bytes
+	.set	DST,		%r8
+	.set	DATALEN,	%r9d
+	.set	DATALEN64,	%r9	// Zero-extend DATALEN before using!
+
+	// Additional local variables
+
+	// %rax is used as a temporary register.  LE_CTR_PTR is also available
+	// as a temporary register after the counter is loaded.
+
+	// AES key length in bytes
+	.set	AESKEYLEN,	%r10d
+	.set	AESKEYLEN64,	%r10
+
+	// Pointer to the last AES round key for the chosen AES variant
+	.set	RNDKEYLAST_PTR,	%r11
+
+	// BSWAP_MASK is the shuffle mask for byte-reflecting 128-bit values
+	// using vpshufb, copied to all 128-bit lanes.
+	.set	BSWAP_MASK,	%ymm0
+	.set	BSWAP_MASK_XMM,	%xmm0
+
+	// GHASH_ACC is the accumulator variable for GHASH.  When fully reduced,
+	// only the lowest 128-bit lane can be nonzero.  When not fully reduced,
+	// more than one lane may be used, and they need to be XOR'd together.
+	.set	GHASH_ACC,	%ymm1
+	.set	GHASH_ACC_XMM,	%xmm1
+
+	// TMP[0-2] are temporary registers.
+	.set	TMP0,		%ymm2
+	.set	TMP0_XMM,	%xmm2
+	.set	TMP1,		%ymm3
+	.set	TMP1_XMM,	%xmm3
+	.set	TMP2,		%ymm4
+	.set	TMP2_XMM,	%xmm4
+
+	// LO and MI are used to accumulate unreduced GHASH products.
+	.set	LO,		%ymm5
+	.set	LO_XMM,		%xmm5
+	.set	MI,		%ymm6
+	.set	MI_XMM,		%xmm6
+
+	// H_POW[2-1]_XORED contain cached values from KEY->h_powers_xored.  The
+	// descending numbering reflects the order of the key powers.
+	.set	H_POW2_XORED,	%ymm7
+	.set	H_POW2_XORED_XMM, %xmm7
+	.set	H_POW1_XORED,	%ymm8
+
+	// RNDKEY0 caches the zero-th round key, and RNDKEYLAST the last one.
+	.set	RNDKEY0,	%ymm9
+	.set	RNDKEYLAST,	%ymm10
+
+	// LE_CTR contains the next set of little-endian counter blocks.
+	.set	LE_CTR,		%ymm11
+
+	// AESDATA[0-3] hold the counter blocks that are being encrypted by AES.
+	.set	AESDATA0,	%ymm12
+	.set	AESDATA0_XMM,	%xmm12
+	.set	AESDATA1,	%ymm13
+	.set	AESDATA1_XMM,	%xmm13
+	.set	AESDATA2,	%ymm14
+	.set	AESDATA3,	%ymm15
+
+.if \enc
+	.set	GHASHDATA_PTR,	DST
+.else
+	.set	GHASHDATA_PTR,	SRC
+.endif
+
+	vbroadcasti128	.Lbswap_mask(%rip), BSWAP_MASK
+
+	// Load the GHASH accumulator and the starting counter.
+	vmovdqu		(GHASH_ACC_PTR), GHASH_ACC_XMM
+	vbroadcasti128	(LE_CTR_PTR), LE_CTR
+
+	// Load the AES key length in bytes.
+	movl		OFFSETOF_AESKEYLEN(KEY), AESKEYLEN
+
+	// Make RNDKEYLAST_PTR point to the last AES round key.  This is the
+	// round key with index 10, 12, or 14 for AES-128, AES-192, or AES-256
+	// respectively.  Then load the zero-th and last round keys.
+	lea		OFFSETOF_AESROUNDKEYS+6*16(KEY,AESKEYLEN64,4), RNDKEYLAST_PTR
+	vbroadcasti128	OFFSETOF_AESROUNDKEYS(KEY), RNDKEY0
+	vbroadcasti128	(RNDKEYLAST_PTR), RNDKEYLAST
+
+	// Finish initializing LE_CTR by adding 1 to the second block.
+	vpaddd		.Lctr_pattern(%rip), LE_CTR, LE_CTR
+
+	// If there are at least 128 bytes of data, then continue into the loop
+	// that processes 128 bytes of data at a time.  Otherwise skip it.
+	add		$-128, DATALEN	// 128 is 4 bytes, -128 is 1 byte
+	jl		.Lcrypt_loop_4x_done\@
+
+	vmovdqu		OFFSETOF_H_POWERS_XORED(KEY), H_POW2_XORED
+	vmovdqu		OFFSETOF_H_POWERS_XORED+32(KEY), H_POW1_XORED
+
+	// Main loop: en/decrypt and hash 4 vectors (128 bytes) at a time.
+
+.if \enc
+	// Encrypt the first 4 vectors of plaintext blocks.
+	_aesenc_loop	0,1,2,3
+	_aesenclast_and_xor	0,1,2,3
+	sub		$-128, SRC	// 128 is 4 bytes, -128 is 1 byte
+	add		$-128, DATALEN
+	jl		.Lghash_last_ciphertext_4x\@
+.endif
+
+.align 16
+.Lcrypt_loop_4x\@:
+
+	// Start the AES encryption of the counter blocks.
+	_ctr_begin	0,1,2,3
+	cmp		$24, AESKEYLEN
+	jl		128f	// AES-128?
+	je		192f	// AES-192?
+	// AES-256
+	vbroadcasti128	-13*16(RNDKEYLAST_PTR), TMP0
+	_vaesenc	TMP0, 0,1,2,3
+	vbroadcasti128	-12*16(RNDKEYLAST_PTR), TMP0
+	_vaesenc	TMP0, 0,1,2,3
+192:
+	vbroadcasti128	-11*16(RNDKEYLAST_PTR), TMP0
+	_vaesenc	TMP0, 0,1,2,3
+	vbroadcasti128	-10*16(RNDKEYLAST_PTR), TMP0
+	_vaesenc	TMP0, 0,1,2,3
+128:
+
+	// Finish the AES encryption of the counter blocks in AESDATA[0-3],
+	// interleaved with the GHASH update of the ciphertext blocks.
+.irp i, 9,8,7,6,5,4,3,2,1
+	_ghash_step_4x  (9 - \i), GHASHDATA_PTR
+	vbroadcasti128	-\i*16(RNDKEYLAST_PTR), TMP0
+	_vaesenc	TMP0, 0,1,2,3
+.endr
+	_ghash_step_4x	9, GHASHDATA_PTR
+.if \enc
+	sub		$-128, DST	// 128 is 4 bytes, -128 is 1 byte
+.endif
+	_aesenclast_and_xor	0,1,2,3
+	sub		$-128, SRC
+.if !\enc
+	sub		$-128, DST
+.endif
+	add		$-128, DATALEN
+	jge		.Lcrypt_loop_4x\@
+
+.if \enc
+.Lghash_last_ciphertext_4x\@:
+	// Update GHASH with the last set of ciphertext blocks.
+	_ghash_4x	DST
+	sub		$-128, DST
+.endif
+
+.Lcrypt_loop_4x_done\@:
+
+	// Undo the extra subtraction by 128 and check whether data remains.
+	sub		$-128, DATALEN	// 128 is 4 bytes, -128 is 1 byte
+	jz		.Ldone\@
+
+	// The data length isn't a multiple of 128 bytes.  Process the remaining
+	// data of length 1 <= DATALEN < 128.
+	//
+	// Since there are enough key powers available for all remaining data,
+	// there is no need to do a GHASH reduction after each iteration.
+	// Instead, multiply each remaining block by its own key power, and only
+	// do a GHASH reduction at the very end.
+
+	// Make POWERS_PTR point to the key powers [H^N, H^(N-1), ...] where N
+	// is the number of blocks that remain.
+	.set		POWERS_PTR, LE_CTR_PTR	// LE_CTR_PTR is free to be reused.
+	.set		POWERS_PTR32, LE_CTR_PTR32
+	mov		DATALEN, %eax
+	neg		%rax
+	and		$~15, %rax  // -round_up(DATALEN, 16)
+	lea		OFFSETOFEND_H_POWERS(KEY,%rax), POWERS_PTR
+
+	// Start collecting the unreduced GHASH intermediate value LO, MI, HI.
+	.set		HI, H_POW2_XORED	// H_POW2_XORED is free to be reused.
+	.set		HI_XMM, H_POW2_XORED_XMM
+	vpxor		LO_XMM, LO_XMM, LO_XMM
+	vpxor		MI_XMM, MI_XMM, MI_XMM
+	vpxor		HI_XMM, HI_XMM, HI_XMM
+
+	// 1 <= DATALEN < 128.  Generate 2 or 4 more vectors of keystream blocks
+	// excluding the last AES round, depending on the remaining DATALEN.
+	cmp		$64, DATALEN
+	jg		.Ltail_gen_4_keystream_vecs\@
+	_aesenc_loop	0,1
+	cmp		$32, DATALEN
+	jge		.Ltail_xor_and_ghash_full_vec_loop\@
+	jmp		.Ltail_xor_and_ghash_partial_vec\@
+.Ltail_gen_4_keystream_vecs\@:
+	_aesenc_loop	0,1,2,3
+
+	// XOR the remaining data and accumulate the unreduced GHASH products
+	// for DATALEN >= 32, starting with one full 32-byte vector at a time.
+.Ltail_xor_and_ghash_full_vec_loop\@:
+.if \enc
+	_aesenclast_and_xor	0
+	vpshufb		BSWAP_MASK, AESDATA0, AESDATA0
+.else
+	vmovdqu		(SRC), TMP1
+	vpxor		TMP1, RNDKEYLAST, TMP0
+	vaesenclast	TMP0, AESDATA0, AESDATA0
+	vmovdqu		AESDATA0, (DST)
+	vpshufb		BSWAP_MASK, TMP1, AESDATA0
+.endif
+	// The ciphertext blocks (i.e. GHASH input data) are now in AESDATA0.
+	vpxor		GHASH_ACC, AESDATA0, AESDATA0
+	vmovdqu		(POWERS_PTR), TMP2
+	_ghash_mul_noreduce	TMP2, AESDATA0, LO, MI, HI, TMP0
+	vmovdqa		AESDATA1, AESDATA0
+	vmovdqa		AESDATA2, AESDATA1
+	vmovdqa		AESDATA3, AESDATA2
+	vpxor		GHASH_ACC_XMM, GHASH_ACC_XMM, GHASH_ACC_XMM
+	add		$32, SRC
+	add		$32, DST
+	add		$32, POWERS_PTR
+	sub		$32, DATALEN
+	cmp		$32, DATALEN
+	jge		.Ltail_xor_and_ghash_full_vec_loop\@
+	test		DATALEN, DATALEN
+	jz		.Ltail_ghash_reduce\@
+
+.Ltail_xor_and_ghash_partial_vec\@:
+	// XOR the remaining data and accumulate the unreduced GHASH products,
+	// for 1 <= DATALEN < 32.
+	vaesenclast	RNDKEYLAST, AESDATA0, AESDATA0
+	cmp		$16, DATALEN
+	jle		.Ltail_xor_and_ghash_1to16bytes\@
+
+	// Handle 17 <= DATALEN < 32.
+
+	// Load a vpshufb mask that will right-shift by '32 - DATALEN' bytes
+	// (shifting in zeroes), then reflect all 16 bytes.
+	lea		.Lrshift_and_bswap_table(%rip), %rax
+	vmovdqu		-16(%rax, DATALEN64), TMP2_XMM
+
+	// Move the second keystream block to its own register and left-align it
+	vextracti128	$1, AESDATA0, AESDATA1_XMM
+	vpxor		.Lfifteens(%rip), TMP2_XMM, TMP0_XMM
+	vpshufb		TMP0_XMM, AESDATA1_XMM, AESDATA1_XMM
+
+	// Using overlapping loads and stores, XOR the source data with the
+	// keystream and write the destination data.  Then prepare the GHASH
+	// input data: the full ciphertext block and the zero-padded partial
+	// ciphertext block, both byte-reflected, in AESDATA0.
+.if \enc
+	vpxor		-16(SRC, DATALEN64), AESDATA1_XMM, AESDATA1_XMM
+	vpxor		(SRC), AESDATA0_XMM, AESDATA0_XMM
+	vmovdqu		AESDATA1_XMM, -16(DST, DATALEN64)
+	vmovdqu		AESDATA0_XMM, (DST)
+	vpshufb		TMP2_XMM, AESDATA1_XMM, AESDATA1_XMM
+	vpshufb		BSWAP_MASK_XMM, AESDATA0_XMM, AESDATA0_XMM
+.else
+	vmovdqu		-16(SRC, DATALEN64), TMP1_XMM
+	vmovdqu		(SRC), TMP0_XMM
+	vpxor		TMP1_XMM, AESDATA1_XMM, AESDATA1_XMM
+	vpxor		TMP0_XMM, AESDATA0_XMM, AESDATA0_XMM
+	vmovdqu		AESDATA1_XMM, -16(DST, DATALEN64)
+	vmovdqu		AESDATA0_XMM, (DST)
+	vpshufb		TMP2_XMM, TMP1_XMM, AESDATA1_XMM
+	vpshufb		BSWAP_MASK_XMM, TMP0_XMM, AESDATA0_XMM
+.endif
+	vpxor		GHASH_ACC_XMM, AESDATA0_XMM, AESDATA0_XMM
+	vinserti128	$1, AESDATA1_XMM, AESDATA0, AESDATA0
+	vmovdqu		(POWERS_PTR), TMP2
+	jmp		.Ltail_ghash_last_vec\@
+
+.Ltail_xor_and_ghash_1to16bytes\@:
+	// Handle 1 <= DATALEN <= 16.  Carefully load and store the
+	// possibly-partial block, which we mustn't access out of bounds.
+	vmovdqu		(POWERS_PTR), TMP2_XMM
+	mov		SRC, KEY	// Free up %rcx, assuming SRC == %rcx
+	mov		DATALEN, %ecx
+	_load_partial_block	KEY, TMP0_XMM, POWERS_PTR, POWERS_PTR32
+	vpxor		TMP0_XMM, AESDATA0_XMM, AESDATA0_XMM
+	mov		DATALEN, %ecx
+	_store_partial_block	AESDATA0_XMM, DST, POWERS_PTR, POWERS_PTR32
+.if \enc
+	lea		.Lselect_high_bytes_table(%rip), %rax
+	vpshufb		BSWAP_MASK_XMM, AESDATA0_XMM, AESDATA0_XMM
+	vpand		(%rax, DATALEN64), AESDATA0_XMM, AESDATA0_XMM
+.else
+	vpshufb		BSWAP_MASK_XMM, TMP0_XMM, AESDATA0_XMM
+.endif
+	vpxor		GHASH_ACC_XMM, AESDATA0_XMM, AESDATA0_XMM
+
+.Ltail_ghash_last_vec\@:
+	// Accumulate the unreduced GHASH products for the last 1-2 blocks.  The
+	// GHASH input data is in AESDATA0.  If only one block remains, then the
+	// second block in AESDATA0 is zero and does not affect the result.
+	_ghash_mul_noreduce	TMP2, AESDATA0, LO, MI, HI, TMP0
+
+.Ltail_ghash_reduce\@:
+	// Finally, do the GHASH reduction.
+	vbroadcasti128	.Lgfpoly(%rip), TMP0
+	_ghash_reduce	LO, MI, HI, TMP0, TMP1
+	vextracti128	$1, HI, GHASH_ACC_XMM
+	vpxor		HI_XMM, GHASH_ACC_XMM, GHASH_ACC_XMM
+
+.Ldone\@:
+	// Store the updated GHASH accumulator back to memory.
+	vmovdqu		GHASH_ACC_XMM, (GHASH_ACC_PTR)
+
+	vzeroupper
+	RET
+.endm
+
+// void aes_gcm_enc_final_vaes_avx2(const struct aes_gcm_key_vaes_avx2 *key,
+//				    const u32 le_ctr[4], u8 ghash_acc[16],
+//				    u64 total_aadlen, u64 total_datalen);
+// bool aes_gcm_dec_final_vaes_avx2(const struct aes_gcm_key_vaes_avx2 *key,
+//				    const u32 le_ctr[4], const u8 ghash_acc[16],
+//				    u64 total_aadlen, u64 total_datalen,
+//				    const u8 tag[16], int taglen);
+//
+// This macro generates one of the above two functions (with \enc selecting
+// which one).  Both functions finish computing the GCM authentication tag by
+// updating GHASH with the lengths block and encrypting the GHASH accumulator.
+// |total_aadlen| and |total_datalen| must be the total length of the additional
+// authenticated data and the en/decrypted data in bytes, respectively.
+//
+// The encryption function then stores the full-length (16-byte) computed
+// authentication tag to |ghash_acc|.  The decryption function instead loads the
+// expected authentication tag (the one that was transmitted) from the 16-byte
+// buffer |tag|, compares the first 4 <= |taglen| <= 16 bytes of it to the
+// computed tag in constant time, and returns true if and only if they match.
+.macro	_aes_gcm_final	enc
+
+	// Function arguments
+	.set	KEY,		%rdi
+	.set	LE_CTR_PTR,	%rsi
+	.set	GHASH_ACC_PTR,	%rdx
+	.set	TOTAL_AADLEN,	%rcx
+	.set	TOTAL_DATALEN,	%r8
+	.set	TAG,		%r9
+	.set	TAGLEN,		%r10d	// Originally at 8(%rsp)
+	.set	TAGLEN64,	%r10
+
+	// Additional local variables.
+	// %rax and %xmm0-%xmm3 are used as temporary registers.
+	.set	AESKEYLEN,	%r11d
+	.set	AESKEYLEN64,	%r11
+	.set	GFPOLY,		%xmm4
+	.set	BSWAP_MASK,	%xmm5
+	.set	LE_CTR,		%xmm6
+	.set	GHASH_ACC,	%xmm7
+	.set	H_POW1,		%xmm8
+
+	// Load some constants.
+	vmovdqa		.Lgfpoly(%rip), GFPOLY
+	vmovdqa		.Lbswap_mask(%rip), BSWAP_MASK
+
+	// Load the AES key length in bytes.
+	movl		OFFSETOF_AESKEYLEN(KEY), AESKEYLEN
+
+	// Set up a counter block with 1 in the low 32-bit word.  This is the
+	// counter that produces the ciphertext needed to encrypt the auth tag.
+	// GFPOLY has 1 in the low word, so grab the 1 from there using a blend.
+	vpblendd	$0xe, (LE_CTR_PTR), GFPOLY, LE_CTR
+
+	// Build the lengths block and XOR it with the GHASH accumulator.
+	// Although the lengths block is defined as the AAD length followed by
+	// the en/decrypted data length, both in big-endian byte order, a byte
+	// reflection of the full block is needed because of the way we compute
+	// GHASH (see _ghash_mul_step).  By using little-endian values in the
+	// opposite order, we avoid having to reflect any bytes here.
+	vmovq		TOTAL_DATALEN, %xmm0
+	vpinsrq		$1, TOTAL_AADLEN, %xmm0, %xmm0
+	vpsllq		$3, %xmm0, %xmm0	// Bytes to bits
+	vpxor		(GHASH_ACC_PTR), %xmm0, GHASH_ACC
+
+	// Load the first hash key power (H^1), which is stored last.
+	vmovdqu		OFFSETOFEND_H_POWERS-16(KEY), H_POW1
+
+	// Load TAGLEN if decrypting.
+.if !\enc
+	movl		8(%rsp), TAGLEN
+.endif
+
+	// Make %rax point to the last AES round key for the chosen AES variant.
+	lea		OFFSETOF_AESROUNDKEYS+6*16(KEY,AESKEYLEN64,4), %rax
+
+	// Start the AES encryption of the counter block by swapping the counter
+	// block to big-endian and XOR-ing it with the zero-th AES round key.
+	vpshufb		BSWAP_MASK, LE_CTR, %xmm0
+	vpxor		OFFSETOF_AESROUNDKEYS(KEY), %xmm0, %xmm0
+
+	// Complete the AES encryption and multiply GHASH_ACC by H^1.
+	// Interleave the AES and GHASH instructions to improve performance.
+	cmp		$24, AESKEYLEN
+	jl		128f	// AES-128?
+	je		192f	// AES-192?
+	// AES-256
+	vaesenc		-13*16(%rax), %xmm0, %xmm0
+	vaesenc		-12*16(%rax), %xmm0, %xmm0
+192:
+	vaesenc		-11*16(%rax), %xmm0, %xmm0
+	vaesenc		-10*16(%rax), %xmm0, %xmm0
+128:
+.irp i, 0,1,2,3,4,5,6,7,8
+	_ghash_mul_step	\i, H_POW1, GHASH_ACC, GHASH_ACC, GFPOLY, \
+			%xmm1, %xmm2, %xmm3
+	vaesenc		(\i-9)*16(%rax), %xmm0, %xmm0
+.endr
+	_ghash_mul_step	9, H_POW1, GHASH_ACC, GHASH_ACC, GFPOLY, \
+			%xmm1, %xmm2, %xmm3
+
+	// Undo the byte reflection of the GHASH accumulator.
+	vpshufb		BSWAP_MASK, GHASH_ACC, GHASH_ACC
+
+	// Do the last AES round and XOR the resulting keystream block with the
+	// GHASH accumulator to produce the full computed authentication tag.
+	//
+	// Reduce latency by taking advantage of the property vaesenclast(key,
+	// a) ^ b == vaesenclast(key ^ b, a).  I.e., XOR GHASH_ACC into the last
+	// round key, instead of XOR'ing the final AES output with GHASH_ACC.
+	//
+	// enc_final then returns the computed auth tag, while dec_final
+	// compares it with the transmitted one and returns a bool.  To compare
+	// the tags, dec_final XORs them together and uses vptest to check
+	// whether the result is all-zeroes.  This should be constant-time.
+	// dec_final applies the vaesenclast optimization to this additional
+	// value XOR'd too.
+.if \enc
+	vpxor		(%rax), GHASH_ACC, %xmm1
+	vaesenclast	%xmm1, %xmm0, GHASH_ACC
+	vmovdqu		GHASH_ACC, (GHASH_ACC_PTR)
+.else
+	vpxor		(TAG), GHASH_ACC, GHASH_ACC
+	vpxor		(%rax), GHASH_ACC, GHASH_ACC
+	vaesenclast	GHASH_ACC, %xmm0, %xmm0
+	lea		.Lselect_high_bytes_table(%rip), %rax
+	vmovdqu		(%rax, TAGLEN64), %xmm1
+	vpshufb		BSWAP_MASK, %xmm1, %xmm1 // select low bytes, not high
+	xor		%eax, %eax
+	vptest		%xmm1, %xmm0
+	sete		%al
+.endif
+	// No need for vzeroupper here, since only used xmm registers were used.
+	RET
+.endm
+
+SYM_FUNC_START(aes_gcm_enc_update_vaes_avx2)
+	_aes_gcm_update	1
+SYM_FUNC_END(aes_gcm_enc_update_vaes_avx2)
+SYM_FUNC_START(aes_gcm_dec_update_vaes_avx2)
+	_aes_gcm_update	0
+SYM_FUNC_END(aes_gcm_dec_update_vaes_avx2)
+
+SYM_FUNC_START(aes_gcm_enc_final_vaes_avx2)
+	_aes_gcm_final	1
+SYM_FUNC_END(aes_gcm_enc_final_vaes_avx2)
+SYM_FUNC_START(aes_gcm_dec_final_vaes_avx2)
+	_aes_gcm_final	0
+SYM_FUNC_END(aes_gcm_dec_final_vaes_avx2)
diff --git a/arch/x86/crypto/aes-gcm-vaes-avx512.S b/arch/x86/crypto/aes-gcm-vaes-avx512.S
new file mode 100644
index 000000000..516747db4
--- /dev/null
+++ b/arch/x86/crypto/aes-gcm-vaes-avx512.S
@@ -0,0 +1,1166 @@
+/* SPDX-License-Identifier: Apache-2.0 OR BSD-2-Clause */
+//
+// AES-GCM implementation for x86_64 CPUs that support the following CPU
+// features: VAES && VPCLMULQDQ && AVX512BW && AVX512VL && BMI2
+//
+// Copyright 2024 Google LLC
+//
+// Author: Eric Biggers <ebiggers@google.com>
+//
+//------------------------------------------------------------------------------
+//
+// This file is dual-licensed, meaning that you can use it under your choice of
+// either of the following two licenses:
+//
+// Licensed under the Apache License 2.0 (the "License").  You may obtain a copy
+// of the License at
+//
+//	http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// or
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are met:
+//
+// 1. Redistributions of source code must retain the above copyright notice,
+//    this list of conditions and the following disclaimer.
+//
+// 2. Redistributions in binary form must reproduce the above copyright
+//    notice, this list of conditions and the following disclaimer in the
+//    documentation and/or other materials provided with the distribution.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+// POSSIBILITY OF SUCH DAMAGE.
+
+#include <linux/linkage.h>
+
+.section .rodata
+.p2align 6
+
+	// A shuffle mask that reflects the bytes of 16-byte blocks
+.Lbswap_mask:
+	.octa	0x000102030405060708090a0b0c0d0e0f
+
+	// This is the GHASH reducing polynomial without its constant term, i.e.
+	// x^128 + x^7 + x^2 + x, represented using the backwards mapping
+	// between bits and polynomial coefficients.
+	//
+	// Alternatively, it can be interpreted as the naturally-ordered
+	// representation of the polynomial x^127 + x^126 + x^121 + 1, i.e. the
+	// "reversed" GHASH reducing polynomial without its x^128 term.
+.Lgfpoly:
+	.octa	0xc2000000000000000000000000000001
+
+	// Same as above, but with the (1 << 64) bit set.
+.Lgfpoly_and_internal_carrybit:
+	.octa	0xc2000000000000010000000000000001
+
+	// Values needed to prepare the initial vector of counter blocks.
+.Lctr_pattern:
+	.octa	0
+	.octa	1
+	.octa	2
+	.octa	3
+
+	// The number of AES blocks per vector, as a 128-bit value.
+.Linc_4blocks:
+	.octa	4
+
+// Number of powers of the hash key stored in the key struct.  The powers are
+// stored from highest (H^NUM_H_POWERS) to lowest (H^1).
+#define NUM_H_POWERS		16
+
+// Offset to AES key length (in bytes) in the key struct
+#define OFFSETOF_AESKEYLEN	0
+
+// Offset to AES round keys in the key struct
+#define OFFSETOF_AESROUNDKEYS	16
+
+// Offset to start of hash key powers array in the key struct
+#define OFFSETOF_H_POWERS	320
+
+// Offset to end of hash key powers array in the key struct.
+//
+// This is immediately followed by three zeroized padding blocks, which are
+// included so that partial vectors can be handled more easily.  E.g. if two
+// blocks remain, we load the 4 values [H^2, H^1, 0, 0].  The most padding
+// blocks needed is 3, which occurs if [H^1, 0, 0, 0] is loaded.
+#define OFFSETOFEND_H_POWERS	(OFFSETOF_H_POWERS + (NUM_H_POWERS * 16))
+
+.text
+
+// The _ghash_mul_step macro does one step of GHASH multiplication of the
+// 128-bit lanes of \a by the corresponding 128-bit lanes of \b and storing the
+// reduced products in \dst.  \t0, \t1, and \t2 are temporary registers of the
+// same size as \a and \b.  To complete all steps, this must invoked with \i=0
+// through \i=9.  The division into steps allows users of this macro to
+// optionally interleave the computation with other instructions.  Users of this
+// macro must preserve the parameter registers across steps.
+//
+// The multiplications are done in GHASH's representation of the finite field
+// GF(2^128).  Elements of GF(2^128) are represented as binary polynomials
+// (i.e. polynomials whose coefficients are bits) modulo a reducing polynomial
+// G.  The GCM specification uses G = x^128 + x^7 + x^2 + x + 1.  Addition is
+// just XOR, while multiplication is more complex and has two parts: (a) do
+// carryless multiplication of two 128-bit input polynomials to get a 256-bit
+// intermediate product polynomial, and (b) reduce the intermediate product to
+// 128 bits by adding multiples of G that cancel out terms in it.  (Adding
+// multiples of G doesn't change which field element the polynomial represents.)
+//
+// Unfortunately, the GCM specification maps bits to/from polynomial
+// coefficients backwards from the natural order.  In each byte it specifies the
+// highest bit to be the lowest order polynomial coefficient, *not* the highest!
+// This makes it nontrivial to work with the GHASH polynomials.  We could
+// reflect the bits, but x86 doesn't have an instruction that does that.
+//
+// Instead, we operate on the values without bit-reflecting them.  This *mostly*
+// just works, since XOR and carryless multiplication are symmetric with respect
+// to bit order, but it has some consequences.  First, due to GHASH's byte
+// order, by skipping bit reflection, *byte* reflection becomes necessary to
+// give the polynomial terms a consistent order.  E.g., considering an N-bit
+// value interpreted using the G = x^128 + x^7 + x^2 + x + 1 convention, bits 0
+// through N-1 of the byte-reflected value represent the coefficients of x^(N-1)
+// through x^0, whereas bits 0 through N-1 of the non-byte-reflected value
+// represent x^7...x^0, x^15...x^8, ..., x^(N-1)...x^(N-8) which can't be worked
+// with.  Fortunately, x86's vpshufb instruction can do byte reflection.
+//
+// Second, forgoing the bit reflection causes an extra multiple of x (still
+// using the G = x^128 + x^7 + x^2 + x + 1 convention) to be introduced by each
+// multiplication.  This is because an M-bit by N-bit carryless multiplication
+// really produces a (M+N-1)-bit product, but in practice it's zero-extended to
+// M+N bits.  In the G = x^128 + x^7 + x^2 + x + 1 convention, which maps bits
+// to polynomial coefficients backwards, this zero-extension actually changes
+// the product by introducing an extra factor of x.  Therefore, users of this
+// macro must ensure that one of the inputs has an extra factor of x^-1, i.e.
+// the multiplicative inverse of x, to cancel out the extra x.
+//
+// Third, the backwards coefficients convention is just confusing to work with,
+// since it makes "low" and "high" in the polynomial math mean the opposite of
+// their normal meaning in computer programming.  This can be solved by using an
+// alternative interpretation: the polynomial coefficients are understood to be
+// in the natural order, and the multiplication is actually \a * \b * x^-128 mod
+// x^128 + x^127 + x^126 + x^121 + 1.  This doesn't change the inputs, outputs,
+// or the implementation at all; it just changes the mathematical interpretation
+// of what each instruction is doing.  Starting from here, we'll use this
+// alternative interpretation, as it's easier to understand the code that way.
+//
+// Moving onto the implementation, the vpclmulqdq instruction does 64 x 64 =>
+// 128-bit carryless multiplication, so we break the 128 x 128 multiplication
+// into parts as follows (the _L and _H suffixes denote low and high 64 bits):
+//
+//     LO = a_L * b_L
+//     MI = (a_L * b_H) + (a_H * b_L)
+//     HI = a_H * b_H
+//
+// The 256-bit product is x^128*HI + x^64*MI + LO.  LO, MI, and HI are 128-bit.
+// Note that MI "overlaps" with LO and HI.  We don't consolidate MI into LO and
+// HI right away, since the way the reduction works makes that unnecessary.
+//
+// For the reduction, we cancel out the low 128 bits by adding multiples of G =
+// x^128 + x^127 + x^126 + x^121 + 1.  This is done by two iterations, each of
+// which cancels out the next lowest 64 bits.  Consider a value x^64*A + B,
+// where A and B are 128-bit.  Adding B_L*G to that value gives:
+//
+//       x^64*A + B + B_L*G
+//     = x^64*A + x^64*B_H + B_L + B_L*(x^128 + x^127 + x^126 + x^121 + 1)
+//     = x^64*A + x^64*B_H + B_L + x^128*B_L + x^64*B_L*(x^63 + x^62 + x^57) + B_L
+//     = x^64*A + x^64*B_H + x^128*B_L + x^64*B_L*(x^63 + x^62 + x^57) + B_L + B_L
+//     = x^64*(A + B_H + x^64*B_L + B_L*(x^63 + x^62 + x^57))
+//
+// So: if we sum A, B with its halves swapped, and the low half of B times x^63
+// + x^62 + x^57, we get a 128-bit value C where x^64*C is congruent to the
+// original value x^64*A + B.  I.e., the low 64 bits got canceled out.
+//
+// We just need to apply this twice: first to fold LO into MI, and second to
+// fold the updated MI into HI.
+//
+// The needed three-argument XORs are done using the vpternlogd instruction with
+// immediate 0x96, since this is faster than two vpxord instructions.
+//
+// A potential optimization, assuming that b is fixed per-key (if a is fixed
+// per-key it would work the other way around), is to use one iteration of the
+// reduction described above to precompute a value c such that x^64*c = b mod G,
+// and then multiply a_L by c (and implicitly by x^64) instead of by b:
+//
+//     MI = (a_L * c_L) + (a_H * b_L)
+//     HI = (a_L * c_H) + (a_H * b_H)
+//
+// This would eliminate the LO part of the intermediate product, which would
+// eliminate the need to fold LO into MI.  This would save two instructions,
+// including a vpclmulqdq.  However, we currently don't use this optimization
+// because it would require twice as many per-key precomputed values.
+//
+// Using Karatsuba multiplication instead of "schoolbook" multiplication
+// similarly would save a vpclmulqdq but does not seem to be worth it.
+.macro	_ghash_mul_step	i, a, b, dst, gfpoly, t0, t1, t2
+.if \i == 0
+	vpclmulqdq	$0x00, \a, \b, \t0	  // LO = a_L * b_L
+	vpclmulqdq	$0x01, \a, \b, \t1	  // MI_0 = a_L * b_H
+.elseif \i == 1
+	vpclmulqdq	$0x10, \a, \b, \t2	  // MI_1 = a_H * b_L
+.elseif \i == 2
+	vpxord		\t2, \t1, \t1		  // MI = MI_0 + MI_1
+.elseif \i == 3
+	vpclmulqdq	$0x01, \t0, \gfpoly, \t2  // LO_L*(x^63 + x^62 + x^57)
+.elseif \i == 4
+	vpshufd		$0x4e, \t0, \t0		  // Swap halves of LO
+.elseif \i == 5
+	vpternlogd	$0x96, \t2, \t0, \t1	  // Fold LO into MI
+.elseif \i == 6
+	vpclmulqdq	$0x11, \a, \b, \dst	  // HI = a_H * b_H
+.elseif \i == 7
+	vpclmulqdq	$0x01, \t1, \gfpoly, \t0  // MI_L*(x^63 + x^62 + x^57)
+.elseif \i == 8
+	vpshufd		$0x4e, \t1, \t1		  // Swap halves of MI
+.elseif \i == 9
+	vpternlogd	$0x96, \t0, \t1, \dst	  // Fold MI into HI
+.endif
+.endm
+
+// GHASH-multiply the 128-bit lanes of \a by the 128-bit lanes of \b and store
+// the reduced products in \dst.  See _ghash_mul_step for full explanation.
+.macro	_ghash_mul	a, b, dst, gfpoly, t0, t1, t2
+.irp i, 0,1,2,3,4,5,6,7,8,9
+	_ghash_mul_step	\i, \a, \b, \dst, \gfpoly, \t0, \t1, \t2
+.endr
+.endm
+
+// GHASH-multiply the 128-bit lanes of \a by the 128-bit lanes of \b and add the
+// *unreduced* products to \lo, \mi, and \hi.
+.macro	_ghash_mul_noreduce	a, b, lo, mi, hi, t0, t1, t2, t3
+	vpclmulqdq	$0x00, \a, \b, \t0	// a_L * b_L
+	vpclmulqdq	$0x01, \a, \b, \t1	// a_L * b_H
+	vpclmulqdq	$0x10, \a, \b, \t2	// a_H * b_L
+	vpclmulqdq	$0x11, \a, \b, \t3	// a_H * b_H
+	vpxord		\t0, \lo, \lo
+	vpternlogd	$0x96, \t2, \t1, \mi
+	vpxord		\t3, \hi, \hi
+.endm
+
+// Reduce the unreduced products from \lo, \mi, and \hi and store the 128-bit
+// reduced products in \hi.  See _ghash_mul_step for explanation of reduction.
+.macro	_ghash_reduce	lo, mi, hi, gfpoly, t0
+	vpclmulqdq	$0x01, \lo, \gfpoly, \t0
+	vpshufd		$0x4e, \lo, \lo
+	vpternlogd	$0x96, \t0, \lo, \mi
+	vpclmulqdq	$0x01, \mi, \gfpoly, \t0
+	vpshufd		$0x4e, \mi, \mi
+	vpternlogd	$0x96, \t0, \mi, \hi
+.endm
+
+// This is a specialized version of _ghash_mul that computes \a * \a, i.e. it
+// squares \a.  It skips computing MI = (a_L * a_H) + (a_H * a_L) = 0.
+.macro	_ghash_square	a, dst, gfpoly, t0, t1
+	vpclmulqdq	$0x00, \a, \a, \t0	  // LO = a_L * a_L
+	vpclmulqdq	$0x11, \a, \a, \dst	  // HI = a_H * a_H
+	vpclmulqdq	$0x01, \t0, \gfpoly, \t1  // LO_L*(x^63 + x^62 + x^57)
+	vpshufd		$0x4e, \t0, \t0		  // Swap halves of LO
+	vpxord		\t0, \t1, \t1		  // Fold LO into MI
+	vpclmulqdq	$0x01, \t1, \gfpoly, \t0  // MI_L*(x^63 + x^62 + x^57)
+	vpshufd		$0x4e, \t1, \t1		  // Swap halves of MI
+	vpternlogd	$0x96, \t0, \t1, \dst	  // Fold MI into HI
+.endm
+
+// void aes_gcm_precompute_vaes_avx512(struct aes_gcm_key_vaes_avx512 *key);
+//
+// Given the expanded AES key |key->base.aes_key|, derive the GHASH subkey and
+// initialize |key->h_powers| and |key->padding|.
+SYM_FUNC_START(aes_gcm_precompute_vaes_avx512)
+
+	// Function arguments
+	.set	KEY,		%rdi
+
+	// Additional local variables.
+	// %zmm[0-2] and %rax are used as temporaries.
+	.set	POWERS_PTR,	%rsi
+	.set	RNDKEYLAST_PTR,	%rdx
+	.set	H_CUR,		%zmm3
+	.set	H_CUR_YMM,	%ymm3
+	.set	H_CUR_XMM,	%xmm3
+	.set	H_INC,		%zmm4
+	.set	H_INC_YMM,	%ymm4
+	.set	H_INC_XMM,	%xmm4
+	.set	GFPOLY,		%zmm5
+	.set	GFPOLY_YMM,	%ymm5
+	.set	GFPOLY_XMM,	%xmm5
+
+	// Get pointer to lowest set of key powers (located at end of array).
+	lea		OFFSETOFEND_H_POWERS-64(KEY), POWERS_PTR
+
+	// Encrypt an all-zeroes block to get the raw hash subkey.
+	movl		OFFSETOF_AESKEYLEN(KEY), %eax
+	lea		OFFSETOF_AESROUNDKEYS+6*16(KEY,%rax,4), RNDKEYLAST_PTR
+	vmovdqu		OFFSETOF_AESROUNDKEYS(KEY), %xmm0
+	add		$OFFSETOF_AESROUNDKEYS+16, KEY
+1:
+	vaesenc		(KEY), %xmm0, %xmm0
+	add		$16, KEY
+	cmp		KEY, RNDKEYLAST_PTR
+	jne		1b
+	vaesenclast	(RNDKEYLAST_PTR), %xmm0, %xmm0
+
+	// Reflect the bytes of the raw hash subkey.
+	vpshufb		.Lbswap_mask(%rip), %xmm0, H_CUR_XMM
+
+	// Zeroize the padding blocks.
+	vpxor		%xmm0, %xmm0, %xmm0
+	vmovdqu		%ymm0, 64(POWERS_PTR)
+	vmovdqu		%xmm0, 64+2*16(POWERS_PTR)
+
+	// Finish preprocessing the first key power, H^1.  Since this GHASH
+	// implementation operates directly on values with the backwards bit
+	// order specified by the GCM standard, it's necessary to preprocess the
+	// raw key as follows.  First, reflect its bytes.  Second, multiply it
+	// by x^-1 mod x^128 + x^7 + x^2 + x + 1 (if using the backwards
+	// interpretation of polynomial coefficients), which can also be
+	// interpreted as multiplication by x mod x^128 + x^127 + x^126 + x^121
+	// + 1 using the alternative, natural interpretation of polynomial
+	// coefficients.  For details, see the comment above _ghash_mul_step.
+	//
+	// Either way, for the multiplication the concrete operation performed
+	// is a left shift of the 128-bit value by 1 bit, then an XOR with (0xc2
+	// << 120) | 1 if a 1 bit was carried out.  However, there's no 128-bit
+	// wide shift instruction, so instead double each of the two 64-bit
+	// halves and incorporate the internal carry bit into the value XOR'd.
+	vpshufd		$0xd3, H_CUR_XMM, %xmm0
+	vpsrad		$31, %xmm0, %xmm0
+	vpaddq		H_CUR_XMM, H_CUR_XMM, H_CUR_XMM
+	// H_CUR_XMM ^= xmm0 & gfpoly_and_internal_carrybit
+	vpternlogd	$0x78, .Lgfpoly_and_internal_carrybit(%rip), %xmm0, H_CUR_XMM
+
+	// Load the gfpoly constant.
+	vbroadcasti32x4	.Lgfpoly(%rip), GFPOLY
+
+	// Square H^1 to get H^2.
+	//
+	// Note that as with H^1, all higher key powers also need an extra
+	// factor of x^-1 (or x using the natural interpretation).  Nothing
+	// special needs to be done to make this happen, though: H^1 * H^1 would
+	// end up with two factors of x^-1, but the multiplication consumes one.
+	// So the product H^2 ends up with the desired one factor of x^-1.
+	_ghash_square	H_CUR_XMM, H_INC_XMM, GFPOLY_XMM, %xmm0, %xmm1
+
+	// Create H_CUR_YMM = [H^2, H^1] and H_INC_YMM = [H^2, H^2].
+	vinserti128	$1, H_CUR_XMM, H_INC_YMM, H_CUR_YMM
+	vinserti128	$1, H_INC_XMM, H_INC_YMM, H_INC_YMM
+
+	// Create H_CUR = [H^4, H^3, H^2, H^1] and H_INC = [H^4, H^4, H^4, H^4].
+	_ghash_mul	H_INC_YMM, H_CUR_YMM, H_INC_YMM, GFPOLY_YMM, \
+			%ymm0, %ymm1, %ymm2
+	vinserti64x4	$1, H_CUR_YMM, H_INC, H_CUR
+	vshufi64x2	$0, H_INC, H_INC, H_INC
+
+	// Store the lowest set of key powers.
+	vmovdqu8	H_CUR, (POWERS_PTR)
+
+	// Compute and store the remaining key powers.
+	// Repeatedly multiply [H^(i+3), H^(i+2), H^(i+1), H^i] by
+	// [H^4, H^4, H^4, H^4] to get [H^(i+7), H^(i+6), H^(i+5), H^(i+4)].
+	mov		$3, %eax
+.Lprecompute_next:
+	sub		$64, POWERS_PTR
+	_ghash_mul	H_INC, H_CUR, H_CUR, GFPOLY, %zmm0, %zmm1, %zmm2
+	vmovdqu8	H_CUR, (POWERS_PTR)
+	dec		%eax
+	jnz		.Lprecompute_next
+
+	vzeroupper	// This is needed after using ymm or zmm registers.
+	RET
+SYM_FUNC_END(aes_gcm_precompute_vaes_avx512)
+
+// XOR together the 128-bit lanes of \src (whose low lane is \src_xmm) and store
+// the result in \dst_xmm.  This implicitly zeroizes the other lanes of dst.
+.macro	_horizontal_xor	src, src_xmm, dst_xmm, t0_xmm, t1_xmm, t2_xmm
+	vextracti32x4	$1, \src, \t0_xmm
+	vextracti32x4	$2, \src, \t1_xmm
+	vextracti32x4	$3, \src, \t2_xmm
+	vpxord		\t0_xmm, \src_xmm, \dst_xmm
+	vpternlogd	$0x96, \t1_xmm, \t2_xmm, \dst_xmm
+.endm
+
+// Do one step of the GHASH update of the data blocks given in the vector
+// registers GHASHDATA[0-3].  \i specifies the step to do, 0 through 9.  The
+// division into steps allows users of this macro to optionally interleave the
+// computation with other instructions.  This macro uses the vector register
+// GHASH_ACC as input/output; GHASHDATA[0-3] as inputs that are clobbered;
+// H_POW[4-1], GFPOLY, and BSWAP_MASK as inputs that aren't clobbered; and
+// GHASHTMP[0-2] as temporaries.  This macro handles the byte-reflection of the
+// data blocks.  The parameter registers must be preserved across steps.
+//
+// The GHASH update does: GHASH_ACC = H_POW4*(GHASHDATA0 + GHASH_ACC) +
+// H_POW3*GHASHDATA1 + H_POW2*GHASHDATA2 + H_POW1*GHASHDATA3, where the
+// operations are vectorized operations on 512-bit vectors of 128-bit blocks.
+// The vectorized terms correspond to the following non-vectorized terms:
+//
+//       H_POW4*(GHASHDATA0 + GHASH_ACC) => H^16*(blk0 + GHASH_ACC_XMM),
+//              H^15*(blk1 + 0), H^14*(blk2 + 0), and H^13*(blk3 + 0)
+//       H_POW3*GHASHDATA1 => H^12*blk4, H^11*blk5, H^10*blk6, and H^9*blk7
+//       H_POW2*GHASHDATA2 => H^8*blk8,  H^7*blk9,  H^6*blk10, and H^5*blk11
+//       H_POW1*GHASHDATA3 => H^4*blk12, H^3*blk13, H^2*blk14, and H^1*blk15
+//
+// More concretely, this code does:
+//   - Do vectorized "schoolbook" multiplications to compute the intermediate
+//     256-bit product of each block and its corresponding hash key power.
+//   - Sum (XOR) the intermediate 256-bit products across vectors.
+//   - Do a vectorized reduction of these 256-bit intermediate values to
+//     128-bits each.
+//   - Sum (XOR) these values and store the 128-bit result in GHASH_ACC_XMM.
+//
+// See _ghash_mul_step for the full explanation of the operations performed for
+// each individual finite field multiplication and reduction.
+.macro	_ghash_step_4x	i
+.if \i == 0
+	vpshufb		BSWAP_MASK, GHASHDATA0, GHASHDATA0
+	vpxord		GHASH_ACC, GHASHDATA0, GHASHDATA0
+	vpshufb		BSWAP_MASK, GHASHDATA1, GHASHDATA1
+	vpshufb		BSWAP_MASK, GHASHDATA2, GHASHDATA2
+.elseif \i == 1
+	vpshufb		BSWAP_MASK, GHASHDATA3, GHASHDATA3
+	vpclmulqdq	$0x00, H_POW4, GHASHDATA0, GHASH_ACC	// LO_0
+	vpclmulqdq	$0x00, H_POW3, GHASHDATA1, GHASHTMP0	// LO_1
+	vpclmulqdq	$0x00, H_POW2, GHASHDATA2, GHASHTMP1	// LO_2
+.elseif \i == 2
+	vpxord		GHASHTMP0, GHASH_ACC, GHASH_ACC		// sum(LO_{1,0})
+	vpclmulqdq	$0x00, H_POW1, GHASHDATA3, GHASHTMP2	// LO_3
+	vpternlogd	$0x96, GHASHTMP2, GHASHTMP1, GHASH_ACC	// LO = sum(LO_{3,2,1,0})
+	vpclmulqdq	$0x01, H_POW4, GHASHDATA0, GHASHTMP0	// MI_0
+.elseif \i == 3
+	vpclmulqdq	$0x01, H_POW3, GHASHDATA1, GHASHTMP1	// MI_1
+	vpclmulqdq	$0x01, H_POW2, GHASHDATA2, GHASHTMP2	// MI_2
+	vpternlogd	$0x96, GHASHTMP2, GHASHTMP1, GHASHTMP0	// sum(MI_{2,1,0})
+	vpclmulqdq	$0x01, H_POW1, GHASHDATA3, GHASHTMP1	// MI_3
+.elseif \i == 4
+	vpclmulqdq	$0x10, H_POW4, GHASHDATA0, GHASHTMP2	// MI_4
+	vpternlogd	$0x96, GHASHTMP2, GHASHTMP1, GHASHTMP0	// sum(MI_{4,3,2,1,0})
+	vpclmulqdq	$0x10, H_POW3, GHASHDATA1, GHASHTMP1	// MI_5
+	vpclmulqdq	$0x10, H_POW2, GHASHDATA2, GHASHTMP2	// MI_6
+.elseif \i == 5
+	vpternlogd	$0x96, GHASHTMP2, GHASHTMP1, GHASHTMP0	// sum(MI_{6,5,4,3,2,1,0})
+	vpclmulqdq	$0x01, GHASH_ACC, GFPOLY, GHASHTMP2	// LO_L*(x^63 + x^62 + x^57)
+	vpclmulqdq	$0x10, H_POW1, GHASHDATA3, GHASHTMP1	// MI_7
+	vpxord		GHASHTMP1, GHASHTMP0, GHASHTMP0		// MI = sum(MI_{7,6,5,4,3,2,1,0})
+.elseif \i == 6
+	vpshufd		$0x4e, GHASH_ACC, GHASH_ACC		// Swap halves of LO
+	vpclmulqdq	$0x11, H_POW4, GHASHDATA0, GHASHDATA0	// HI_0
+	vpclmulqdq	$0x11, H_POW3, GHASHDATA1, GHASHDATA1	// HI_1
+	vpclmulqdq	$0x11, H_POW2, GHASHDATA2, GHASHDATA2	// HI_2
+.elseif \i == 7
+	vpternlogd	$0x96, GHASHTMP2, GHASH_ACC, GHASHTMP0	// Fold LO into MI
+	vpclmulqdq	$0x11, H_POW1, GHASHDATA3, GHASHDATA3	// HI_3
+	vpternlogd	$0x96, GHASHDATA2, GHASHDATA1, GHASHDATA0 // sum(HI_{2,1,0})
+	vpclmulqdq	$0x01, GHASHTMP0, GFPOLY, GHASHTMP1	// MI_L*(x^63 + x^62 + x^57)
+.elseif \i == 8
+	vpxord		GHASHDATA3, GHASHDATA0, GHASH_ACC	// HI = sum(HI_{3,2,1,0})
+	vpshufd		$0x4e, GHASHTMP0, GHASHTMP0		// Swap halves of MI
+	vpternlogd	$0x96, GHASHTMP1, GHASHTMP0, GHASH_ACC	// Fold MI into HI
+.elseif \i == 9
+	_horizontal_xor	GHASH_ACC, GHASH_ACC_XMM, GHASH_ACC_XMM, \
+			GHASHDATA0_XMM, GHASHDATA1_XMM, GHASHDATA2_XMM
+.endif
+.endm
+
+// Update GHASH with four vectors of data blocks.  See _ghash_step_4x for full
+// explanation.
+.macro	_ghash_4x
+.irp i, 0,1,2,3,4,5,6,7,8,9
+	_ghash_step_4x	\i
+.endr
+.endm
+
+// void aes_gcm_aad_update_vaes_avx512(const struct aes_gcm_key_vaes_avx512 *key,
+//				       u8 ghash_acc[16],
+//				       const u8 *aad, int aadlen);
+//
+// This function processes the AAD (Additional Authenticated Data) in GCM.
+// Using the key |key|, it updates the GHASH accumulator |ghash_acc| with the
+// data given by |aad| and |aadlen|.  On the first call, |ghash_acc| must be all
+// zeroes.  |aadlen| must be a multiple of 16, except on the last call where it
+// can be any length.  The caller must do any buffering needed to ensure this.
+//
+// This handles large amounts of AAD efficiently, while also keeping overhead
+// low for small amounts which is the common case.  TLS and IPsec use less than
+// one block of AAD, but (uncommonly) other use cases may use much more.
+SYM_FUNC_START(aes_gcm_aad_update_vaes_avx512)
+
+	// Function arguments
+	.set	KEY,		%rdi
+	.set	GHASH_ACC_PTR,	%rsi
+	.set	AAD,		%rdx
+	.set	AADLEN,		%ecx
+	.set	AADLEN64,	%rcx	// Zero-extend AADLEN before using!
+
+	// Additional local variables.
+	// %rax and %k1 are used as temporary registers.
+	.set	GHASHDATA0,	%zmm0
+	.set	GHASHDATA0_XMM,	%xmm0
+	.set	GHASHDATA1,	%zmm1
+	.set	GHASHDATA1_XMM,	%xmm1
+	.set	GHASHDATA2,	%zmm2
+	.set	GHASHDATA2_XMM,	%xmm2
+	.set	GHASHDATA3,	%zmm3
+	.set	BSWAP_MASK,	%zmm4
+	.set	BSWAP_MASK_XMM,	%xmm4
+	.set	GHASH_ACC,	%zmm5
+	.set	GHASH_ACC_XMM,	%xmm5
+	.set	H_POW4,		%zmm6
+	.set	H_POW3,		%zmm7
+	.set	H_POW2,		%zmm8
+	.set	H_POW1,		%zmm9
+	.set	H_POW1_XMM,	%xmm9
+	.set	GFPOLY,		%zmm10
+	.set	GFPOLY_XMM,	%xmm10
+	.set	GHASHTMP0,	%zmm11
+	.set	GHASHTMP1,	%zmm12
+	.set	GHASHTMP2,	%zmm13
+
+	// Load the GHASH accumulator.
+	vmovdqu		(GHASH_ACC_PTR), GHASH_ACC_XMM
+
+	// Check for the common case of AADLEN <= 16, as well as AADLEN == 0.
+	cmp		$16, AADLEN
+	jg		.Laad_more_than_16bytes
+	test		AADLEN, AADLEN
+	jz		.Laad_done
+
+	// Fast path: update GHASH with 1 <= AADLEN <= 16 bytes of AAD.
+	vmovdqu		.Lbswap_mask(%rip), BSWAP_MASK_XMM
+	vmovdqu		.Lgfpoly(%rip), GFPOLY_XMM
+	mov		$-1, %eax
+	bzhi		AADLEN, %eax, %eax
+	kmovd		%eax, %k1
+	vmovdqu8	(AAD), GHASHDATA0_XMM{%k1}{z}
+	vmovdqu		OFFSETOFEND_H_POWERS-16(KEY), H_POW1_XMM
+	vpshufb		BSWAP_MASK_XMM, GHASHDATA0_XMM, GHASHDATA0_XMM
+	vpxor		GHASHDATA0_XMM, GHASH_ACC_XMM, GHASH_ACC_XMM
+	_ghash_mul	H_POW1_XMM, GHASH_ACC_XMM, GHASH_ACC_XMM, GFPOLY_XMM, \
+			GHASHDATA0_XMM, GHASHDATA1_XMM, GHASHDATA2_XMM
+	jmp		.Laad_done
+
+.Laad_more_than_16bytes:
+	vbroadcasti32x4	.Lbswap_mask(%rip), BSWAP_MASK
+	vbroadcasti32x4	.Lgfpoly(%rip), GFPOLY
+
+	// If AADLEN >= 256, update GHASH with 256 bytes of AAD at a time.
+	sub		$256, AADLEN
+	jl		.Laad_loop_4x_done
+	vmovdqu8	OFFSETOFEND_H_POWERS-4*64(KEY), H_POW4
+	vmovdqu8	OFFSETOFEND_H_POWERS-3*64(KEY), H_POW3
+	vmovdqu8	OFFSETOFEND_H_POWERS-2*64(KEY), H_POW2
+	vmovdqu8	OFFSETOFEND_H_POWERS-1*64(KEY), H_POW1
+.Laad_loop_4x:
+	vmovdqu8	0*64(AAD), GHASHDATA0
+	vmovdqu8	1*64(AAD), GHASHDATA1
+	vmovdqu8	2*64(AAD), GHASHDATA2
+	vmovdqu8	3*64(AAD), GHASHDATA3
+	_ghash_4x
+	add		$256, AAD
+	sub		$256, AADLEN
+	jge		.Laad_loop_4x
+.Laad_loop_4x_done:
+
+	// If AADLEN >= 64, update GHASH with 64 bytes of AAD at a time.
+	add		$192, AADLEN
+	jl		.Laad_loop_1x_done
+	vmovdqu8	OFFSETOFEND_H_POWERS-1*64(KEY), H_POW1
+.Laad_loop_1x:
+	vmovdqu8	(AAD), GHASHDATA0
+	vpshufb		BSWAP_MASK, GHASHDATA0, GHASHDATA0
+	vpxord		GHASHDATA0, GHASH_ACC, GHASH_ACC
+	_ghash_mul	H_POW1, GHASH_ACC, GHASH_ACC, GFPOLY, \
+			GHASHDATA0, GHASHDATA1, GHASHDATA2
+	_horizontal_xor GHASH_ACC, GHASH_ACC_XMM, GHASH_ACC_XMM, \
+			GHASHDATA0_XMM, GHASHDATA1_XMM, GHASHDATA2_XMM
+	add		$64, AAD
+	sub		$64, AADLEN
+	jge		.Laad_loop_1x
+.Laad_loop_1x_done:
+
+	// Update GHASH with the remaining 0 <= AADLEN < 64 bytes of AAD.
+	add		$64, AADLEN
+	jz		.Laad_done
+	mov		$-1, %rax
+	bzhi		AADLEN64, %rax, %rax
+	kmovq		%rax, %k1
+	vmovdqu8	(AAD), GHASHDATA0{%k1}{z}
+	neg		AADLEN64
+	and		$~15, AADLEN64  // -round_up(AADLEN, 16)
+	vmovdqu8	OFFSETOFEND_H_POWERS(KEY,AADLEN64), H_POW1
+	vpshufb		BSWAP_MASK, GHASHDATA0, GHASHDATA0
+	vpxord		GHASHDATA0, GHASH_ACC, GHASH_ACC
+	_ghash_mul	H_POW1, GHASH_ACC, GHASH_ACC, GFPOLY, \
+			GHASHDATA0, GHASHDATA1, GHASHDATA2
+	_horizontal_xor GHASH_ACC, GHASH_ACC_XMM, GHASH_ACC_XMM, \
+			GHASHDATA0_XMM, GHASHDATA1_XMM, GHASHDATA2_XMM
+
+.Laad_done:
+	// Store the updated GHASH accumulator back to memory.
+	vmovdqu		GHASH_ACC_XMM, (GHASH_ACC_PTR)
+
+	vzeroupper	// This is needed after using ymm or zmm registers.
+	RET
+SYM_FUNC_END(aes_gcm_aad_update_vaes_avx512)
+
+// Do one non-last round of AES encryption on the blocks in %zmm[0-3] using the
+// round key that has been broadcast to all 128-bit lanes of \round_key.
+.macro	_vaesenc_4x	round_key
+	vaesenc		\round_key, %zmm0, %zmm0
+	vaesenc		\round_key, %zmm1, %zmm1
+	vaesenc		\round_key, %zmm2, %zmm2
+	vaesenc		\round_key, %zmm3, %zmm3
+.endm
+
+// Start the AES encryption of four vectors of counter blocks.
+.macro	_ctr_begin_4x
+
+	// Increment LE_CTR four times to generate four vectors of little-endian
+	// counter blocks, swap each to big-endian, and store them in %zmm[0-3].
+	vpshufb		BSWAP_MASK, LE_CTR, %zmm0
+	vpaddd		LE_CTR_INC, LE_CTR, LE_CTR
+	vpshufb		BSWAP_MASK, LE_CTR, %zmm1
+	vpaddd		LE_CTR_INC, LE_CTR, LE_CTR
+	vpshufb		BSWAP_MASK, LE_CTR, %zmm2
+	vpaddd		LE_CTR_INC, LE_CTR, LE_CTR
+	vpshufb		BSWAP_MASK, LE_CTR, %zmm3
+	vpaddd		LE_CTR_INC, LE_CTR, LE_CTR
+
+	// AES "round zero": XOR in the zero-th round key.
+	vpxord		RNDKEY0, %zmm0, %zmm0
+	vpxord		RNDKEY0, %zmm1, %zmm1
+	vpxord		RNDKEY0, %zmm2, %zmm2
+	vpxord		RNDKEY0, %zmm3, %zmm3
+.endm
+
+// Do the last AES round for four vectors of counter blocks %zmm[0-3], XOR
+// source data with the resulting keystream, and write the result to DST and
+// GHASHDATA[0-3].  (Implementation differs slightly, but has the same effect.)
+.macro	_aesenclast_and_xor_4x
+	// XOR the source data with the last round key, saving the result in
+	// GHASHDATA[0-3].  This reduces latency by taking advantage of the
+	// property vaesenclast(key, a) ^ b == vaesenclast(key ^ b, a).
+	vpxord		0*64(SRC), RNDKEYLAST, GHASHDATA0
+	vpxord		1*64(SRC), RNDKEYLAST, GHASHDATA1
+	vpxord		2*64(SRC), RNDKEYLAST, GHASHDATA2
+	vpxord		3*64(SRC), RNDKEYLAST, GHASHDATA3
+
+	// Do the last AES round.  This handles the XOR with the source data
+	// too, as per the optimization described above.
+	vaesenclast	GHASHDATA0, %zmm0, GHASHDATA0
+	vaesenclast	GHASHDATA1, %zmm1, GHASHDATA1
+	vaesenclast	GHASHDATA2, %zmm2, GHASHDATA2
+	vaesenclast	GHASHDATA3, %zmm3, GHASHDATA3
+
+	// Store the en/decrypted data to DST.
+	vmovdqu8	GHASHDATA0, 0*64(DST)
+	vmovdqu8	GHASHDATA1, 1*64(DST)
+	vmovdqu8	GHASHDATA2, 2*64(DST)
+	vmovdqu8	GHASHDATA3, 3*64(DST)
+.endm
+
+// void aes_gcm_{enc,dec}_update_vaes_avx512(const struct aes_gcm_key_vaes_avx512 *key,
+//					     const u32 le_ctr[4], u8 ghash_acc[16],
+//					     const u8 *src, u8 *dst, int datalen);
+//
+// This macro generates a GCM encryption or decryption update function with the
+// above prototype (with \enc selecting which one).  The function computes the
+// next portion of the CTR keystream, XOR's it with |datalen| bytes from |src|,
+// and writes the resulting encrypted or decrypted data to |dst|.  It also
+// updates the GHASH accumulator |ghash_acc| using the next |datalen| ciphertext
+// bytes.
+//
+// |datalen| must be a multiple of 16, except on the last call where it can be
+// any length.  The caller must do any buffering needed to ensure this.  Both
+// in-place and out-of-place en/decryption are supported.
+//
+// |le_ctr| must give the current counter in little-endian format.  This
+// function loads the counter from |le_ctr| and increments the loaded counter as
+// needed, but it does *not* store the updated counter back to |le_ctr|.  The
+// caller must update |le_ctr| if any more data segments follow.  Internally,
+// only the low 32-bit word of the counter is incremented, following the GCM
+// standard.
+.macro	_aes_gcm_update	enc
+
+	// Function arguments
+	.set	KEY,		%rdi
+	.set	LE_CTR_PTR,	%rsi
+	.set	GHASH_ACC_PTR,	%rdx
+	.set	SRC,		%rcx
+	.set	DST,		%r8
+	.set	DATALEN,	%r9d
+	.set	DATALEN64,	%r9	// Zero-extend DATALEN before using!
+
+	// Additional local variables
+
+	// %rax and %k1 are used as temporary registers.  LE_CTR_PTR is also
+	// available as a temporary register after the counter is loaded.
+
+	// AES key length in bytes
+	.set	AESKEYLEN,	%r10d
+	.set	AESKEYLEN64,	%r10
+
+	// Pointer to the last AES round key for the chosen AES variant
+	.set	RNDKEYLAST_PTR,	%r11
+
+	// In the main loop, %zmm[0-3] are used as AES input and output.
+	// Elsewhere they are used as temporary registers.
+
+	// GHASHDATA[0-3] hold the ciphertext blocks and GHASH input data.
+	.set	GHASHDATA0,	%zmm4
+	.set	GHASHDATA0_XMM,	%xmm4
+	.set	GHASHDATA1,	%zmm5
+	.set	GHASHDATA1_XMM,	%xmm5
+	.set	GHASHDATA2,	%zmm6
+	.set	GHASHDATA2_XMM,	%xmm6
+	.set	GHASHDATA3,	%zmm7
+
+	// BSWAP_MASK is the shuffle mask for byte-reflecting 128-bit values
+	// using vpshufb, copied to all 128-bit lanes.
+	.set	BSWAP_MASK,	%zmm8
+
+	// RNDKEY temporarily holds the next AES round key.
+	.set	RNDKEY,		%zmm9
+
+	// GHASH_ACC is the accumulator variable for GHASH.  When fully reduced,
+	// only the lowest 128-bit lane can be nonzero.  When not fully reduced,
+	// more than one lane may be used, and they need to be XOR'd together.
+	.set	GHASH_ACC,	%zmm10
+	.set	GHASH_ACC_XMM,	%xmm10
+
+	// LE_CTR_INC is the vector of 32-bit words that need to be added to a
+	// vector of little-endian counter blocks to advance it forwards.
+	.set	LE_CTR_INC,	%zmm11
+
+	// LE_CTR contains the next set of little-endian counter blocks.
+	.set	LE_CTR,		%zmm12
+
+	// RNDKEY0, RNDKEYLAST, and RNDKEY_M[9-1] contain cached AES round keys,
+	// copied to all 128-bit lanes.  RNDKEY0 is the zero-th round key,
+	// RNDKEYLAST the last, and RNDKEY_M\i the one \i-th from the last.
+	.set	RNDKEY0,	%zmm13
+	.set	RNDKEYLAST,	%zmm14
+	.set	RNDKEY_M9,	%zmm15
+	.set	RNDKEY_M8,	%zmm16
+	.set	RNDKEY_M7,	%zmm17
+	.set	RNDKEY_M6,	%zmm18
+	.set	RNDKEY_M5,	%zmm19
+	.set	RNDKEY_M4,	%zmm20
+	.set	RNDKEY_M3,	%zmm21
+	.set	RNDKEY_M2,	%zmm22
+	.set	RNDKEY_M1,	%zmm23
+
+	// GHASHTMP[0-2] are temporary variables used by _ghash_step_4x.  These
+	// cannot coincide with anything used for AES encryption, since for
+	// performance reasons GHASH and AES encryption are interleaved.
+	.set	GHASHTMP0,	%zmm24
+	.set	GHASHTMP1,	%zmm25
+	.set	GHASHTMP2,	%zmm26
+
+	// H_POW[4-1] contain the powers of the hash key H^16...H^1.  The
+	// descending numbering reflects the order of the key powers.
+	.set	H_POW4,		%zmm27
+	.set	H_POW3,		%zmm28
+	.set	H_POW2,		%zmm29
+	.set	H_POW1,		%zmm30
+
+	// GFPOLY contains the .Lgfpoly constant, copied to all 128-bit lanes.
+	.set	GFPOLY,		%zmm31
+
+	// Load some constants.
+	vbroadcasti32x4	.Lbswap_mask(%rip), BSWAP_MASK
+	vbroadcasti32x4	.Lgfpoly(%rip), GFPOLY
+
+	// Load the GHASH accumulator and the starting counter.
+	vmovdqu		(GHASH_ACC_PTR), GHASH_ACC_XMM
+	vbroadcasti32x4	(LE_CTR_PTR), LE_CTR
+
+	// Load the AES key length in bytes.
+	movl		OFFSETOF_AESKEYLEN(KEY), AESKEYLEN
+
+	// Make RNDKEYLAST_PTR point to the last AES round key.  This is the
+	// round key with index 10, 12, or 14 for AES-128, AES-192, or AES-256
+	// respectively.  Then load the zero-th and last round keys.
+	lea		OFFSETOF_AESROUNDKEYS+6*16(KEY,AESKEYLEN64,4), RNDKEYLAST_PTR
+	vbroadcasti32x4	OFFSETOF_AESROUNDKEYS(KEY), RNDKEY0
+	vbroadcasti32x4	(RNDKEYLAST_PTR), RNDKEYLAST
+
+	// Finish initializing LE_CTR by adding [0, 1, ...] to its low words.
+	vpaddd		.Lctr_pattern(%rip), LE_CTR, LE_CTR
+
+	// Load 4 into all 128-bit lanes of LE_CTR_INC.
+	vbroadcasti32x4	.Linc_4blocks(%rip), LE_CTR_INC
+
+	// If there are at least 256 bytes of data, then continue into the loop
+	// that processes 256 bytes of data at a time.  Otherwise skip it.
+	//
+	// Pre-subtracting 256 from DATALEN saves an instruction from the main
+	// loop and also ensures that at least one write always occurs to
+	// DATALEN, zero-extending it and allowing DATALEN64 to be used later.
+	sub		$256, DATALEN
+	jl		.Lcrypt_loop_4x_done\@
+
+	// Load powers of the hash key.
+	vmovdqu8	OFFSETOFEND_H_POWERS-4*64(KEY), H_POW4
+	vmovdqu8	OFFSETOFEND_H_POWERS-3*64(KEY), H_POW3
+	vmovdqu8	OFFSETOFEND_H_POWERS-2*64(KEY), H_POW2
+	vmovdqu8	OFFSETOFEND_H_POWERS-1*64(KEY), H_POW1
+
+	// Main loop: en/decrypt and hash 4 vectors at a time.
+	//
+	// When possible, interleave the AES encryption of the counter blocks
+	// with the GHASH update of the ciphertext blocks.  This improves
+	// performance on many CPUs because the execution ports used by the VAES
+	// instructions often differ from those used by vpclmulqdq and other
+	// instructions used in GHASH.  For example, many Intel CPUs dispatch
+	// vaesenc to ports 0 and 1 and vpclmulqdq to port 5.
+	//
+	// The interleaving is easiest to do during decryption, since during
+	// decryption the ciphertext blocks are immediately available.  For
+	// encryption, instead encrypt the first set of blocks, then hash those
+	// blocks while encrypting the next set of blocks, repeat that as
+	// needed, and finally hash the last set of blocks.
+
+.if \enc
+	// Encrypt the first 4 vectors of plaintext blocks.  Leave the resulting
+	// ciphertext in GHASHDATA[0-3] for GHASH.
+	_ctr_begin_4x
+	lea		OFFSETOF_AESROUNDKEYS+16(KEY), %rax
+1:
+	vbroadcasti32x4	(%rax), RNDKEY
+	_vaesenc_4x	RNDKEY
+	add		$16, %rax
+	cmp		%rax, RNDKEYLAST_PTR
+	jne		1b
+	_aesenclast_and_xor_4x
+	add		$256, SRC
+	add		$256, DST
+	sub		$256, DATALEN
+	jl		.Lghash_last_ciphertext_4x\@
+.endif
+
+	// Cache as many additional AES round keys as possible.
+.irp i, 9,8,7,6,5,4,3,2,1
+	vbroadcasti32x4	-\i*16(RNDKEYLAST_PTR), RNDKEY_M\i
+.endr
+
+.Lcrypt_loop_4x\@:
+
+	// If decrypting, load more ciphertext blocks into GHASHDATA[0-3].  If
+	// encrypting, GHASHDATA[0-3] already contain the previous ciphertext.
+.if !\enc
+	vmovdqu8	0*64(SRC), GHASHDATA0
+	vmovdqu8	1*64(SRC), GHASHDATA1
+	vmovdqu8	2*64(SRC), GHASHDATA2
+	vmovdqu8	3*64(SRC), GHASHDATA3
+.endif
+
+	// Start the AES encryption of the counter blocks.
+	_ctr_begin_4x
+	cmp		$24, AESKEYLEN
+	jl		128f	// AES-128?
+	je		192f	// AES-192?
+	// AES-256
+	vbroadcasti32x4	-13*16(RNDKEYLAST_PTR), RNDKEY
+	_vaesenc_4x	RNDKEY
+	vbroadcasti32x4	-12*16(RNDKEYLAST_PTR), RNDKEY
+	_vaesenc_4x	RNDKEY
+192:
+	vbroadcasti32x4	-11*16(RNDKEYLAST_PTR), RNDKEY
+	_vaesenc_4x	RNDKEY
+	vbroadcasti32x4	-10*16(RNDKEYLAST_PTR), RNDKEY
+	_vaesenc_4x	RNDKEY
+128:
+
+	// Finish the AES encryption of the counter blocks in %zmm[0-3],
+	// interleaved with the GHASH update of the ciphertext blocks in
+	// GHASHDATA[0-3].
+.irp i, 9,8,7,6,5,4,3,2,1
+	_ghash_step_4x  (9 - \i)
+	_vaesenc_4x	RNDKEY_M\i
+.endr
+	_ghash_step_4x	9
+	_aesenclast_and_xor_4x
+	add		$256, SRC
+	add		$256, DST
+	sub		$256, DATALEN
+	jge		.Lcrypt_loop_4x\@
+
+.if \enc
+.Lghash_last_ciphertext_4x\@:
+	// Update GHASH with the last set of ciphertext blocks.
+	_ghash_4x
+.endif
+
+.Lcrypt_loop_4x_done\@:
+
+	// Undo the extra subtraction by 256 and check whether data remains.
+	add		$256, DATALEN
+	jz		.Ldone\@
+
+	// The data length isn't a multiple of 256 bytes.  Process the remaining
+	// data of length 1 <= DATALEN < 256, up to one 64-byte vector at a
+	// time.  Going one vector at a time may seem inefficient compared to
+	// having separate code paths for each possible number of vectors
+	// remaining.  However, using a loop keeps the code size down, and it
+	// performs surprising well; modern CPUs will start executing the next
+	// iteration before the previous one finishes and also predict the
+	// number of loop iterations.  For a similar reason, we roll up the AES
+	// rounds.
+	//
+	// On the last iteration, the remaining length may be less than 64
+	// bytes.  Handle this using masking.
+	//
+	// Since there are enough key powers available for all remaining data,
+	// there is no need to do a GHASH reduction after each iteration.
+	// Instead, multiply each remaining block by its own key power, and only
+	// do a GHASH reduction at the very end.
+
+	// Make POWERS_PTR point to the key powers [H^N, H^(N-1), ...] where N
+	// is the number of blocks that remain.
+	.set		POWERS_PTR, LE_CTR_PTR	// LE_CTR_PTR is free to be reused.
+	mov		DATALEN, %eax
+	neg		%rax
+	and		$~15, %rax  // -round_up(DATALEN, 16)
+	lea		OFFSETOFEND_H_POWERS(KEY,%rax), POWERS_PTR
+
+	// Start collecting the unreduced GHASH intermediate value LO, MI, HI.
+	.set		LO, GHASHDATA0
+	.set		LO_XMM, GHASHDATA0_XMM
+	.set		MI, GHASHDATA1
+	.set		MI_XMM, GHASHDATA1_XMM
+	.set		HI, GHASHDATA2
+	.set		HI_XMM, GHASHDATA2_XMM
+	vpxor		LO_XMM, LO_XMM, LO_XMM
+	vpxor		MI_XMM, MI_XMM, MI_XMM
+	vpxor		HI_XMM, HI_XMM, HI_XMM
+
+.Lcrypt_loop_1x\@:
+
+	// Select the appropriate mask for this iteration: all 1's if
+	// DATALEN >= 64, otherwise DATALEN 1's.  Do this branchlessly using the
+	// bzhi instruction from BMI2.  (This relies on DATALEN <= 255.)
+	mov		$-1, %rax
+	bzhi		DATALEN64, %rax, %rax
+	kmovq		%rax, %k1
+
+	// Encrypt a vector of counter blocks.  This does not need to be masked.
+	vpshufb		BSWAP_MASK, LE_CTR, %zmm0
+	vpaddd		LE_CTR_INC, LE_CTR, LE_CTR
+	vpxord		RNDKEY0, %zmm0, %zmm0
+	lea		OFFSETOF_AESROUNDKEYS+16(KEY), %rax
+1:
+	vbroadcasti32x4	(%rax), RNDKEY
+	vaesenc		RNDKEY, %zmm0, %zmm0
+	add		$16, %rax
+	cmp		%rax, RNDKEYLAST_PTR
+	jne		1b
+	vaesenclast	RNDKEYLAST, %zmm0, %zmm0
+
+	// XOR the data with the appropriate number of keystream bytes.
+	vmovdqu8	(SRC), %zmm1{%k1}{z}
+	vpxord		%zmm1, %zmm0, %zmm0
+	vmovdqu8	%zmm0, (DST){%k1}
+
+	// Update GHASH with the ciphertext block(s), without reducing.
+	//
+	// In the case of DATALEN < 64, the ciphertext is zero-padded to 64
+	// bytes.  (If decrypting, it's done by the above masked load.  If
+	// encrypting, it's done by the below masked register-to-register move.)
+	// Note that if DATALEN <= 48, there will be additional padding beyond
+	// the padding of the last block specified by GHASH itself; i.e., there
+	// may be whole block(s) that get processed by the GHASH multiplication
+	// and reduction instructions but should not actually be included in the
+	// GHASH.  However, any such blocks are all-zeroes, and the values that
+	// they're multiplied with are also all-zeroes.  Therefore they just add
+	// 0 * 0 = 0 to the final GHASH result, which makes no difference.
+	vmovdqu8	(POWERS_PTR), H_POW1
+.if \enc
+	vmovdqu8	%zmm0, %zmm1{%k1}{z}
+.endif
+	vpshufb		BSWAP_MASK, %zmm1, %zmm0
+	vpxord		GHASH_ACC, %zmm0, %zmm0
+	_ghash_mul_noreduce	H_POW1, %zmm0, LO, MI, HI, \
+				GHASHDATA3, %zmm1, %zmm2, %zmm3
+	vpxor		GHASH_ACC_XMM, GHASH_ACC_XMM, GHASH_ACC_XMM
+
+	add		$64, POWERS_PTR
+	add		$64, SRC
+	add		$64, DST
+	sub		$64, DATALEN
+	jg		.Lcrypt_loop_1x\@
+
+	// Finally, do the GHASH reduction.
+	_ghash_reduce	LO, MI, HI, GFPOLY, %zmm0
+	_horizontal_xor	HI, HI_XMM, GHASH_ACC_XMM, %xmm0, %xmm1, %xmm2
+
+.Ldone\@:
+	// Store the updated GHASH accumulator back to memory.
+	vmovdqu		GHASH_ACC_XMM, (GHASH_ACC_PTR)
+
+	vzeroupper	// This is needed after using ymm or zmm registers.
+	RET
+.endm
+
+// void aes_gcm_enc_final_vaes_avx512(const struct aes_gcm_key_vaes_avx512 *key,
+//				      const u32 le_ctr[4], u8 ghash_acc[16],
+//				      u64 total_aadlen, u64 total_datalen);
+// bool aes_gcm_dec_final_vaes_avx512(const struct aes_gcm_key_vaes_avx512 *key,
+//				      const u32 le_ctr[4],
+//				      const u8 ghash_acc[16],
+//				      u64 total_aadlen, u64 total_datalen,
+//				      const u8 tag[16], int taglen);
+//
+// This macro generates one of the above two functions (with \enc selecting
+// which one).  Both functions finish computing the GCM authentication tag by
+// updating GHASH with the lengths block and encrypting the GHASH accumulator.
+// |total_aadlen| and |total_datalen| must be the total length of the additional
+// authenticated data and the en/decrypted data in bytes, respectively.
+//
+// The encryption function then stores the full-length (16-byte) computed
+// authentication tag to |ghash_acc|.  The decryption function instead loads the
+// expected authentication tag (the one that was transmitted) from the 16-byte
+// buffer |tag|, compares the first 4 <= |taglen| <= 16 bytes of it to the
+// computed tag in constant time, and returns true if and only if they match.
+.macro	_aes_gcm_final	enc
+
+	// Function arguments
+	.set	KEY,		%rdi
+	.set	LE_CTR_PTR,	%rsi
+	.set	GHASH_ACC_PTR,	%rdx
+	.set	TOTAL_AADLEN,	%rcx
+	.set	TOTAL_DATALEN,	%r8
+	.set	TAG,		%r9
+	.set	TAGLEN,		%r10d	// Originally at 8(%rsp)
+
+	// Additional local variables.
+	// %rax, %xmm0-%xmm3, and %k1 are used as temporary registers.
+	.set	AESKEYLEN,	%r11d
+	.set	AESKEYLEN64,	%r11
+	.set	GFPOLY,		%xmm4
+	.set	BSWAP_MASK,	%xmm5
+	.set	LE_CTR,		%xmm6
+	.set	GHASH_ACC,	%xmm7
+	.set	H_POW1,		%xmm8
+
+	// Load some constants.
+	vmovdqa		.Lgfpoly(%rip), GFPOLY
+	vmovdqa		.Lbswap_mask(%rip), BSWAP_MASK
+
+	// Load the AES key length in bytes.
+	movl		OFFSETOF_AESKEYLEN(KEY), AESKEYLEN
+
+	// Set up a counter block with 1 in the low 32-bit word.  This is the
+	// counter that produces the ciphertext needed to encrypt the auth tag.
+	// GFPOLY has 1 in the low word, so grab the 1 from there using a blend.
+	vpblendd	$0xe, (LE_CTR_PTR), GFPOLY, LE_CTR
+
+	// Build the lengths block and XOR it with the GHASH accumulator.
+	// Although the lengths block is defined as the AAD length followed by
+	// the en/decrypted data length, both in big-endian byte order, a byte
+	// reflection of the full block is needed because of the way we compute
+	// GHASH (see _ghash_mul_step).  By using little-endian values in the
+	// opposite order, we avoid having to reflect any bytes here.
+	vmovq		TOTAL_DATALEN, %xmm0
+	vpinsrq		$1, TOTAL_AADLEN, %xmm0, %xmm0
+	vpsllq		$3, %xmm0, %xmm0	// Bytes to bits
+	vpxor		(GHASH_ACC_PTR), %xmm0, GHASH_ACC
+
+	// Load the first hash key power (H^1), which is stored last.
+	vmovdqu8	OFFSETOFEND_H_POWERS-16(KEY), H_POW1
+
+.if !\enc
+	// Prepare a mask of TAGLEN one bits.
+	movl		8(%rsp), TAGLEN
+	mov		$-1, %eax
+	bzhi		TAGLEN, %eax, %eax
+	kmovd		%eax, %k1
+.endif
+
+	// Make %rax point to the last AES round key for the chosen AES variant.
+	lea		OFFSETOF_AESROUNDKEYS+6*16(KEY,AESKEYLEN64,4), %rax
+
+	// Start the AES encryption of the counter block by swapping the counter
+	// block to big-endian and XOR-ing it with the zero-th AES round key.
+	vpshufb		BSWAP_MASK, LE_CTR, %xmm0
+	vpxor		OFFSETOF_AESROUNDKEYS(KEY), %xmm0, %xmm0
+
+	// Complete the AES encryption and multiply GHASH_ACC by H^1.
+	// Interleave the AES and GHASH instructions to improve performance.
+	cmp		$24, AESKEYLEN
+	jl		128f	// AES-128?
+	je		192f	// AES-192?
+	// AES-256
+	vaesenc		-13*16(%rax), %xmm0, %xmm0
+	vaesenc		-12*16(%rax), %xmm0, %xmm0
+192:
+	vaesenc		-11*16(%rax), %xmm0, %xmm0
+	vaesenc		-10*16(%rax), %xmm0, %xmm0
+128:
+.irp i, 0,1,2,3,4,5,6,7,8
+	_ghash_mul_step	\i, H_POW1, GHASH_ACC, GHASH_ACC, GFPOLY, \
+			%xmm1, %xmm2, %xmm3
+	vaesenc		(\i-9)*16(%rax), %xmm0, %xmm0
+.endr
+	_ghash_mul_step	9, H_POW1, GHASH_ACC, GHASH_ACC, GFPOLY, \
+			%xmm1, %xmm2, %xmm3
+
+	// Undo the byte reflection of the GHASH accumulator.
+	vpshufb		BSWAP_MASK, GHASH_ACC, GHASH_ACC
+
+	// Do the last AES round and XOR the resulting keystream block with the
+	// GHASH accumulator to produce the full computed authentication tag.
+	//
+	// Reduce latency by taking advantage of the property vaesenclast(key,
+	// a) ^ b == vaesenclast(key ^ b, a).  I.e., XOR GHASH_ACC into the last
+	// round key, instead of XOR'ing the final AES output with GHASH_ACC.
+	//
+	// enc_final then returns the computed auth tag, while dec_final
+	// compares it with the transmitted one and returns a bool.  To compare
+	// the tags, dec_final XORs them together and uses vptest to check
+	// whether the result is all-zeroes.  This should be constant-time.
+	// dec_final applies the vaesenclast optimization to this additional
+	// value XOR'd too, using vpternlogd to XOR the last round key, GHASH
+	// accumulator, and transmitted auth tag together in one instruction.
+.if \enc
+	vpxor		(%rax), GHASH_ACC, %xmm1
+	vaesenclast	%xmm1, %xmm0, GHASH_ACC
+	vmovdqu		GHASH_ACC, (GHASH_ACC_PTR)
+.else
+	vmovdqu		(TAG), %xmm1
+	vpternlogd	$0x96, (%rax), GHASH_ACC, %xmm1
+	vaesenclast	%xmm1, %xmm0, %xmm0
+	xor		%eax, %eax
+	vmovdqu8	%xmm0, %xmm0{%k1}{z}	// Truncate to TAGLEN bytes
+	vptest		%xmm0, %xmm0
+	sete		%al
+.endif
+	// No need for vzeroupper here, since only used xmm registers were used.
+	RET
+.endm
+
+SYM_FUNC_START(aes_gcm_enc_update_vaes_avx512)
+	_aes_gcm_update	1
+SYM_FUNC_END(aes_gcm_enc_update_vaes_avx512)
+SYM_FUNC_START(aes_gcm_dec_update_vaes_avx512)
+	_aes_gcm_update	0
+SYM_FUNC_END(aes_gcm_dec_update_vaes_avx512)
+
+SYM_FUNC_START(aes_gcm_enc_final_vaes_avx512)
+	_aes_gcm_final	1
+SYM_FUNC_END(aes_gcm_enc_final_vaes_avx512)
+SYM_FUNC_START(aes_gcm_dec_final_vaes_avx512)
+	_aes_gcm_final	0
+SYM_FUNC_END(aes_gcm_dec_final_vaes_avx512)
diff --git a/arch/x86/crypto/aes-xts-avx-x86_64.S b/arch/x86/crypto/aes-xts-avx-x86_64.S
new file mode 100644
index 000000000..a30753a3e
--- /dev/null
+++ b/arch/x86/crypto/aes-xts-avx-x86_64.S
@@ -0,0 +1,905 @@
+/* SPDX-License-Identifier: Apache-2.0 OR BSD-2-Clause */
+//
+// AES-XTS for modern x86_64 CPUs
+//
+// Copyright 2024 Google LLC
+//
+// Author: Eric Biggers <ebiggers@google.com>
+//
+//------------------------------------------------------------------------------
+//
+// This file is dual-licensed, meaning that you can use it under your choice of
+// either of the following two licenses:
+//
+// Licensed under the Apache License 2.0 (the "License").  You may obtain a copy
+// of the License at
+//
+//	http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+// or
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are met:
+//
+// 1. Redistributions of source code must retain the above copyright notice,
+//    this list of conditions and the following disclaimer.
+//
+// 2. Redistributions in binary form must reproduce the above copyright
+//    notice, this list of conditions and the following disclaimer in the
+//    documentation and/or other materials provided with the distribution.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+// POSSIBILITY OF SUCH DAMAGE.
+
+/*
+ * This file implements AES-XTS for modern x86_64 CPUs.  To handle the
+ * complexities of coding for x86 SIMD, e.g. where every vector length needs
+ * different code, it uses a macro to generate several implementations that
+ * share similar source code but are targeted at different CPUs, listed below:
+ *
+ * AES-NI && AVX
+ *    - 128-bit vectors (1 AES block per vector)
+ *    - VEX-coded instructions
+ *    - xmm0-xmm15
+ *    - This is for older CPUs that lack VAES but do have AVX.
+ *
+ * VAES && VPCLMULQDQ && AVX2
+ *    - 256-bit vectors (2 AES blocks per vector)
+ *    - VEX-coded instructions
+ *    - ymm0-ymm15
+ *    - This is for CPUs that have VAES but either lack AVX512 (e.g. Intel's
+ *      Alder Lake and AMD's Zen 3) or downclock too eagerly when using zmm
+ *      registers (e.g. Intel's Ice Lake).
+ *
+ * VAES && VPCLMULQDQ && AVX512BW && AVX512VL && BMI2
+ *    - 512-bit vectors (4 AES blocks per vector)
+ *    - EVEX-coded instructions
+ *    - zmm0-zmm31
+ *    - This is for CPUs that have good AVX512 support.
+ *
+ * This file doesn't have an implementation for AES-NI alone (without AVX), as
+ * the lack of VEX would make all the assembly code different.
+ *
+ * When we use VAES, we also use VPCLMULQDQ to parallelize the computation of
+ * the XTS tweaks.  This avoids a bottleneck.  Currently there don't seem to be
+ * any CPUs that support VAES but not VPCLMULQDQ.  If that changes, we might
+ * need to start also providing an implementation using VAES alone.
+ *
+ * The AES-XTS implementations in this file support everything required by the
+ * crypto API, including support for arbitrary input lengths and multi-part
+ * processing.  However, they are most heavily optimized for the common case of
+ * power-of-2 length inputs that are processed in a single part (disk sectors).
+ */
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+
+.section .rodata
+.p2align 4
+.Lgf_poly:
+	// The low 64 bits of this value represent the polynomial x^7 + x^2 + x
+	// + 1.  It is the value that must be XOR'd into the low 64 bits of the
+	// tweak each time a 1 is carried out of the high 64 bits.
+	//
+	// The high 64 bits of this value is just the internal carry bit that
+	// exists when there's a carry out of the low 64 bits of the tweak.
+	.quad	0x87, 1
+
+	// These are the shift amounts that are needed when multiplying by [x^0,
+	// x^1, x^2, x^3] to compute the first vector of tweaks when VL=64.
+	//
+	// The right shifts by 64 are expected to zeroize the destination.
+	// 'vpsrlvq' is indeed defined to do that; i.e. it doesn't truncate the
+	// amount to 64 & 63 = 0 like the 'shr' scalar shift instruction would.
+.Lrshift_amounts:
+	.byte	64, 64, 63, 63, 62, 62, 61, 61
+.Llshift_amounts:
+	.byte	0, 0, 1, 1, 2, 2, 3, 3
+
+	// This table contains constants for vpshufb and vpblendvb, used to
+	// handle variable byte shifts and blending during ciphertext stealing
+	// on CPUs that don't support AVX512-style masking.
+.Lcts_permute_table:
+	.byte	0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+	.byte	0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+	.byte	0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07
+	.byte	0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f
+	.byte	0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+	.byte	0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+.text
+
+.macro	_define_Vi	i
+.if VL == 16
+	.set	V\i,		%xmm\i
+.elseif VL == 32
+	.set	V\i,		%ymm\i
+.elseif VL == 64
+	.set	V\i,		%zmm\i
+.else
+	.error "Unsupported Vector Length (VL)"
+.endif
+.endm
+
+.macro _define_aliases
+	// Define register aliases V0-V15, or V0-V31 if all 32 SIMD registers
+	// are available, that map to the xmm, ymm, or zmm registers according
+	// to the selected Vector Length (VL).
+.irp i, 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
+	_define_Vi	\i
+.endr
+.if USE_AVX512
+.irp i, 16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31
+	_define_Vi	\i
+.endr
+.endif
+
+	// Function parameters
+	.set	KEY,		%rdi	// Initially points to crypto_aes_ctx, then is
+					// advanced to point to 7th-from-last round key
+	.set	SRC,		%rsi	// Pointer to next source data
+	.set	DST,		%rdx	// Pointer to next destination data
+	.set	LEN,		%ecx	// Remaining length in bytes
+	.set	LEN8,		%cl
+	.set	LEN64,		%rcx
+	.set	TWEAK,		%r8	// Pointer to next tweak
+
+	// %rax holds the AES key length in bytes.
+	.set	KEYLEN,		%eax
+	.set	KEYLEN64,	%rax
+
+	// %r9-r11 are available as temporaries.
+
+	// V0-V3 hold the data blocks during the main loop, or temporary values
+	// otherwise.  V4-V5 hold temporary values.
+
+	// V6-V9 hold XTS tweaks.  Each 128-bit lane holds one tweak.
+	.set	TWEAK0_XMM,	%xmm6
+	.set	TWEAK0,		V6
+	.set	TWEAK1_XMM,	%xmm7
+	.set	TWEAK1,		V7
+	.set	TWEAK2,		V8
+	.set	TWEAK3,		V9
+
+	// V10-V13 are used for computing the next values of TWEAK[0-3].
+	.set	NEXT_TWEAK0,	V10
+	.set	NEXT_TWEAK1,	V11
+	.set	NEXT_TWEAK2,	V12
+	.set	NEXT_TWEAK3,	V13
+
+	// V14 holds the constant from .Lgf_poly, copied to all 128-bit lanes.
+	.set	GF_POLY_XMM,	%xmm14
+	.set	GF_POLY,	V14
+
+	// V15 holds the key for AES "round 0", copied to all 128-bit lanes.
+	.set	KEY0_XMM,	%xmm15
+	.set	KEY0,		V15
+
+	// If 32 SIMD registers are available, then V16-V29 hold the remaining
+	// AES round keys, copied to all 128-bit lanes.
+	//
+	// AES-128, AES-192, and AES-256 use different numbers of round keys.
+	// To allow handling all three variants efficiently, we align the round
+	// keys to the *end* of this register range.  I.e., AES-128 uses
+	// KEY5-KEY14, AES-192 uses KEY3-KEY14, and AES-256 uses KEY1-KEY14.
+	// (All also use KEY0 for the XOR-only "round" at the beginning.)
+.if USE_AVX512
+	.set	KEY1_XMM,	%xmm16
+	.set	KEY1,		V16
+	.set	KEY2_XMM,	%xmm17
+	.set	KEY2,		V17
+	.set	KEY3_XMM,	%xmm18
+	.set	KEY3,		V18
+	.set	KEY4_XMM,	%xmm19
+	.set	KEY4,		V19
+	.set	KEY5_XMM,	%xmm20
+	.set	KEY5,		V20
+	.set	KEY6_XMM,	%xmm21
+	.set	KEY6,		V21
+	.set	KEY7_XMM,	%xmm22
+	.set	KEY7,		V22
+	.set	KEY8_XMM,	%xmm23
+	.set	KEY8,		V23
+	.set	KEY9_XMM,	%xmm24
+	.set	KEY9,		V24
+	.set	KEY10_XMM,	%xmm25
+	.set	KEY10,		V25
+	.set	KEY11_XMM,	%xmm26
+	.set	KEY11,		V26
+	.set	KEY12_XMM,	%xmm27
+	.set	KEY12,		V27
+	.set	KEY13_XMM,	%xmm28
+	.set	KEY13,		V28
+	.set	KEY14_XMM,	%xmm29
+	.set	KEY14,		V29
+.endif
+	// V30-V31 are currently unused.
+.endm
+
+// Move a vector between memory and a register.
+.macro	_vmovdqu	src, dst
+.if VL < 64
+	vmovdqu		\src, \dst
+.else
+	vmovdqu8	\src, \dst
+.endif
+.endm
+
+// Broadcast a 128-bit value into a vector.
+.macro	_vbroadcast128	src, dst
+.if VL == 16
+	vmovdqu		\src, \dst
+.elseif VL == 32
+	vbroadcasti128	\src, \dst
+.else
+	vbroadcasti32x4	\src, \dst
+.endif
+.endm
+
+// XOR two vectors together.
+.macro	_vpxor	src1, src2, dst
+.if VL < 64
+	vpxor		\src1, \src2, \dst
+.else
+	vpxord		\src1, \src2, \dst
+.endif
+.endm
+
+// XOR three vectors together.
+.macro	_xor3	src1, src2, src3_and_dst
+.if USE_AVX512
+	// vpternlogd with immediate 0x96 is a three-argument XOR.
+	vpternlogd	$0x96, \src1, \src2, \src3_and_dst
+.else
+	vpxor		\src1, \src3_and_dst, \src3_and_dst
+	vpxor		\src2, \src3_and_dst, \src3_and_dst
+.endif
+.endm
+
+// Given a 128-bit XTS tweak in the xmm register \src, compute the next tweak
+// (by multiplying by the polynomial 'x') and write it to \dst.
+.macro	_next_tweak	src, tmp, dst
+	vpshufd		$0x13, \src, \tmp
+	vpaddq		\src, \src, \dst
+	vpsrad		$31, \tmp, \tmp
+.if USE_AVX512
+	vpternlogd	$0x78, GF_POLY_XMM, \tmp, \dst
+.else
+	vpand		GF_POLY_XMM, \tmp, \tmp
+	vpxor		\tmp, \dst, \dst
+.endif
+.endm
+
+// Given the XTS tweak(s) in the vector \src, compute the next vector of
+// tweak(s) (by multiplying by the polynomial 'x^(VL/16)') and write it to \dst.
+//
+// If VL > 16, then there are multiple tweaks, and we use vpclmulqdq to compute
+// all tweaks in the vector in parallel.  If VL=16, we just do the regular
+// computation without vpclmulqdq, as it's the faster method for a single tweak.
+.macro	_next_tweakvec	src, tmp1, tmp2, dst
+.if VL == 16
+	_next_tweak	\src, \tmp1, \dst
+.else
+	vpsrlq		$64 - VL/16, \src, \tmp1
+	vpclmulqdq	$0x01, GF_POLY, \tmp1, \tmp2
+	vpslldq		$8, \tmp1, \tmp1
+	vpsllq		$VL/16, \src, \dst
+	_xor3		\tmp1, \tmp2, \dst
+.endif
+.endm
+
+// Given the first XTS tweak at (TWEAK), compute the first set of tweaks and
+// store them in the vector registers TWEAK0-TWEAK3.  Clobbers V0-V5.
+.macro	_compute_first_set_of_tweaks
+.if VL == 16
+	vmovdqu		(TWEAK), TWEAK0_XMM
+	vmovdqu		.Lgf_poly(%rip), GF_POLY
+	_next_tweak	TWEAK0, %xmm0, TWEAK1
+	_next_tweak	TWEAK1, %xmm0, TWEAK2
+	_next_tweak	TWEAK2, %xmm0, TWEAK3
+.elseif VL == 32
+	vmovdqu		(TWEAK), TWEAK0_XMM
+	vbroadcasti128	.Lgf_poly(%rip), GF_POLY
+
+	// Compute the first vector of tweaks.
+	_next_tweak	TWEAK0_XMM, %xmm0, %xmm1
+	vinserti128	$1, %xmm1, TWEAK0, TWEAK0
+
+	// Compute the next three vectors of tweaks:
+	//	TWEAK1 = TWEAK0 * [x^2, x^2]
+	//	TWEAK2 = TWEAK0 * [x^4, x^4]
+	//	TWEAK3 = TWEAK0 * [x^6, x^6]
+	vpsrlq		$64 - 2, TWEAK0, V0
+	vpsrlq		$64 - 4, TWEAK0, V2
+	vpsrlq		$64 - 6, TWEAK0, V4
+	vpclmulqdq	$0x01, GF_POLY, V0, V1
+	vpclmulqdq	$0x01, GF_POLY, V2, V3
+	vpclmulqdq	$0x01, GF_POLY, V4, V5
+	vpslldq		$8, V0, V0
+	vpslldq		$8, V2, V2
+	vpslldq		$8, V4, V4
+	vpsllq		$2, TWEAK0, TWEAK1
+	vpsllq		$4, TWEAK0, TWEAK2
+	vpsllq		$6, TWEAK0, TWEAK3
+	vpxor		V0, TWEAK1, TWEAK1
+	vpxor		V2, TWEAK2, TWEAK2
+	vpxor		V4, TWEAK3, TWEAK3
+	vpxor		V1, TWEAK1, TWEAK1
+	vpxor		V3, TWEAK2, TWEAK2
+	vpxor		V5, TWEAK3, TWEAK3
+.else
+	vbroadcasti32x4	(TWEAK), TWEAK0
+	vbroadcasti32x4	.Lgf_poly(%rip), GF_POLY
+
+	// Compute the first vector of tweaks:
+	//	TWEAK0 = broadcast128(TWEAK) * [x^0, x^1, x^2, x^3]
+	vpmovzxbq	.Lrshift_amounts(%rip), V4
+	vpsrlvq		V4, TWEAK0, V0
+	vpclmulqdq	$0x01, GF_POLY, V0, V1
+	vpmovzxbq	.Llshift_amounts(%rip), V4
+	vpslldq		$8, V0, V0
+	vpsllvq		V4, TWEAK0, TWEAK0
+	vpternlogd	$0x96, V0, V1, TWEAK0
+
+	// Compute the next three vectors of tweaks:
+	//	TWEAK1 = TWEAK0 * [x^4, x^4, x^4, x^4]
+	//	TWEAK2 = TWEAK0 * [x^8, x^8, x^8, x^8]
+	//	TWEAK3 = TWEAK0 * [x^12, x^12, x^12, x^12]
+	// x^8 only needs byte-aligned shifts, so optimize accordingly.
+	vpsrlq		$64 - 4, TWEAK0, V0
+	vpsrldq		$(64 - 8) / 8, TWEAK0, V2
+	vpsrlq		$64 - 12, TWEAK0, V4
+	vpclmulqdq	$0x01, GF_POLY, V0, V1
+	vpclmulqdq	$0x01, GF_POLY, V2, V3
+	vpclmulqdq	$0x01, GF_POLY, V4, V5
+	vpslldq		$8, V0, V0
+	vpslldq		$8, V4, V4
+	vpsllq		$4, TWEAK0, TWEAK1
+	vpslldq		$8 / 8, TWEAK0, TWEAK2
+	vpsllq		$12, TWEAK0, TWEAK3
+	vpternlogd	$0x96, V0, V1, TWEAK1
+	vpxord		V3, TWEAK2, TWEAK2
+	vpternlogd	$0x96, V4, V5, TWEAK3
+.endif
+.endm
+
+// Do one step in computing the next set of tweaks using the method of just
+// multiplying by x repeatedly (the same method _next_tweak uses).
+.macro	_tweak_step_mulx	i
+.if \i == 0
+	.set PREV_TWEAK, TWEAK3
+	.set NEXT_TWEAK, NEXT_TWEAK0
+.elseif \i == 5
+	.set PREV_TWEAK, NEXT_TWEAK0
+	.set NEXT_TWEAK, NEXT_TWEAK1
+.elseif \i == 10
+	.set PREV_TWEAK, NEXT_TWEAK1
+	.set NEXT_TWEAK, NEXT_TWEAK2
+.elseif \i == 15
+	.set PREV_TWEAK, NEXT_TWEAK2
+	.set NEXT_TWEAK, NEXT_TWEAK3
+.endif
+.if \i >= 0 && \i < 20 && \i % 5 == 0
+	vpshufd		$0x13, PREV_TWEAK, V5
+.elseif \i >= 0 && \i < 20 && \i % 5 == 1
+	vpaddq		PREV_TWEAK, PREV_TWEAK, NEXT_TWEAK
+.elseif \i >= 0 && \i < 20 && \i % 5 == 2
+	vpsrad		$31, V5, V5
+.elseif \i >= 0 && \i < 20 && \i % 5 == 3
+	vpand		GF_POLY, V5, V5
+.elseif \i >= 0 && \i < 20 && \i % 5 == 4
+	vpxor		V5, NEXT_TWEAK, NEXT_TWEAK
+.elseif \i == 1000
+	vmovdqa		NEXT_TWEAK0, TWEAK0
+	vmovdqa		NEXT_TWEAK1, TWEAK1
+	vmovdqa		NEXT_TWEAK2, TWEAK2
+	vmovdqa		NEXT_TWEAK3, TWEAK3
+.endif
+.endm
+
+// Do one step in computing the next set of tweaks using the VPCLMULQDQ method
+// (the same method _next_tweakvec uses for VL > 16).  This means multiplying
+// each tweak by x^(4*VL/16) independently.
+//
+// Since 4*VL/16 is a multiple of 8 when VL > 16 (which it is here), the needed
+// shift amounts are byte-aligned, which allows the use of vpsrldq and vpslldq
+// to do 128-bit wide shifts.  The 128-bit left shift (vpslldq) saves
+// instructions directly.  The 128-bit right shift (vpsrldq) performs better
+// than a 64-bit right shift on Intel CPUs in the context where it is used here,
+// because it runs on a different execution port from the AES instructions.
+.macro	_tweak_step_pclmul	i
+.if \i == 0
+	vpsrldq		$(128 - 4*VL/16) / 8, TWEAK0, NEXT_TWEAK0
+.elseif \i == 2
+	vpsrldq		$(128 - 4*VL/16) / 8, TWEAK1, NEXT_TWEAK1
+.elseif \i == 4
+	vpsrldq		$(128 - 4*VL/16) / 8, TWEAK2, NEXT_TWEAK2
+.elseif \i == 6
+	vpsrldq		$(128 - 4*VL/16) / 8, TWEAK3, NEXT_TWEAK3
+.elseif \i == 8
+	vpclmulqdq	$0x00, GF_POLY, NEXT_TWEAK0, NEXT_TWEAK0
+.elseif \i == 10
+	vpclmulqdq	$0x00, GF_POLY, NEXT_TWEAK1, NEXT_TWEAK1
+.elseif \i == 12
+	vpclmulqdq	$0x00, GF_POLY, NEXT_TWEAK2, NEXT_TWEAK2
+.elseif \i == 14
+	vpclmulqdq	$0x00, GF_POLY, NEXT_TWEAK3, NEXT_TWEAK3
+.elseif \i == 1000
+	vpslldq		$(4*VL/16) / 8, TWEAK0, TWEAK0
+	vpslldq		$(4*VL/16) / 8, TWEAK1, TWEAK1
+	vpslldq		$(4*VL/16) / 8, TWEAK2, TWEAK2
+	vpslldq		$(4*VL/16) / 8, TWEAK3, TWEAK3
+	_vpxor		NEXT_TWEAK0, TWEAK0, TWEAK0
+	_vpxor		NEXT_TWEAK1, TWEAK1, TWEAK1
+	_vpxor		NEXT_TWEAK2, TWEAK2, TWEAK2
+	_vpxor		NEXT_TWEAK3, TWEAK3, TWEAK3
+.endif
+.endm
+
+// _tweak_step does one step of the computation of the next set of tweaks from
+// TWEAK[0-3].  To complete all steps, this is invoked with increasing values of
+// \i that include at least 0 through 19, then 1000 which signals the last step.
+//
+// This is used to interleave the computation of the next set of tweaks with the
+// AES en/decryptions, which increases performance in some cases.  Clobbers V5.
+.macro	_tweak_step	i
+.if VL == 16
+	_tweak_step_mulx	\i
+.else
+	_tweak_step_pclmul	\i
+.endif
+.endm
+
+.macro	_setup_round_keys	enc
+
+	// Select either the encryption round keys or the decryption round keys.
+.if \enc
+	.set	OFFS, 0
+.else
+	.set	OFFS, 240
+.endif
+
+	// Load the round key for "round 0".
+	_vbroadcast128	OFFS(KEY), KEY0
+
+	// Increment KEY to make it so that 7*16(KEY) is the last round key.
+	// For AES-128, increment by 3*16, resulting in the 10 round keys (not
+	// counting the zero-th round key which was just loaded into KEY0) being
+	// -2*16(KEY) through 7*16(KEY).  For AES-192, increment by 5*16 and use
+	// 12 round keys -4*16(KEY) through 7*16(KEY).  For AES-256, increment
+	// by 7*16 and use 14 round keys -6*16(KEY) through 7*16(KEY).
+	//
+	// This rebasing provides two benefits.  First, it makes the offset to
+	// any round key be in the range [-96, 112], fitting in a signed byte.
+	// This shortens VEX-encoded instructions that access the later round
+	// keys which otherwise would need 4-byte offsets.  Second, it makes it
+	// easy to do AES-128 and AES-192 by skipping irrelevant rounds at the
+	// beginning.  Skipping rounds at the end doesn't work as well because
+	// the last round needs different instructions.
+	//
+	// An alternative approach would be to roll up all the round loops.  We
+	// don't do that because (a) it isn't compatible with caching the round
+	// keys in registers which we do when possible (see below), (b) we
+	// interleave the AES rounds with the XTS tweak computation, and (c) it
+	// seems unwise to rely *too* heavily on the CPU's branch predictor.
+	lea		OFFS-16(KEY, KEYLEN64, 4), KEY
+
+	// If all 32 SIMD registers are available, cache all the round keys.
+.if USE_AVX512
+	cmp		$24, KEYLEN
+	jl		.Laes128\@
+	je		.Laes192\@
+	vbroadcasti32x4	-6*16(KEY), KEY1
+	vbroadcasti32x4	-5*16(KEY), KEY2
+.Laes192\@:
+	vbroadcasti32x4	-4*16(KEY), KEY3
+	vbroadcasti32x4	-3*16(KEY), KEY4
+.Laes128\@:
+	vbroadcasti32x4	-2*16(KEY), KEY5
+	vbroadcasti32x4	-1*16(KEY), KEY6
+	vbroadcasti32x4	0*16(KEY), KEY7
+	vbroadcasti32x4	1*16(KEY), KEY8
+	vbroadcasti32x4	2*16(KEY), KEY9
+	vbroadcasti32x4	3*16(KEY), KEY10
+	vbroadcasti32x4	4*16(KEY), KEY11
+	vbroadcasti32x4	5*16(KEY), KEY12
+	vbroadcasti32x4	6*16(KEY), KEY13
+	vbroadcasti32x4	7*16(KEY), KEY14
+.endif
+.endm
+
+// Do a single non-last round of AES encryption (if \enc==1) or decryption (if
+// \enc==0) on the block(s) in \data using the round key(s) in \key.  The
+// register length determines the number of AES blocks en/decrypted.
+.macro	_vaes	enc, key, data
+.if \enc
+	vaesenc		\key, \data, \data
+.else
+	vaesdec		\key, \data, \data
+.endif
+.endm
+
+// Same as _vaes, but does the last round.
+.macro	_vaeslast	enc, key, data
+.if \enc
+	vaesenclast	\key, \data, \data
+.else
+	vaesdeclast	\key, \data, \data
+.endif
+.endm
+
+// Do a single non-last round of AES en/decryption on the block(s) in \data,
+// using the same key for all block(s).  The round key is loaded from the
+// appropriate register or memory location for round \i.  May clobber \tmp.
+.macro _vaes_1x		enc, i, xmm_suffix, data, tmp
+.if USE_AVX512
+	_vaes		\enc, KEY\i\xmm_suffix, \data
+.else
+.ifnb \xmm_suffix
+	_vaes		\enc, (\i-7)*16(KEY), \data
+.else
+	_vbroadcast128	(\i-7)*16(KEY), \tmp
+	_vaes		\enc, \tmp, \data
+.endif
+.endif
+.endm
+
+// Do a single non-last round of AES en/decryption on the blocks in registers
+// V0-V3, using the same key for all blocks.  The round key is loaded from the
+// appropriate register or memory location for round \i.  In addition, does two
+// steps of the computation of the next set of tweaks.  May clobber V4 and V5.
+.macro	_vaes_4x	enc, i
+.if USE_AVX512
+	_tweak_step	(2*(\i-5))
+	_vaes		\enc, KEY\i, V0
+	_vaes		\enc, KEY\i, V1
+	_tweak_step	(2*(\i-5) + 1)
+	_vaes		\enc, KEY\i, V2
+	_vaes		\enc, KEY\i, V3
+.else
+	_vbroadcast128	(\i-7)*16(KEY), V4
+	_tweak_step	(2*(\i-5))
+	_vaes		\enc, V4, V0
+	_vaes		\enc, V4, V1
+	_tweak_step	(2*(\i-5) + 1)
+	_vaes		\enc, V4, V2
+	_vaes		\enc, V4, V3
+.endif
+.endm
+
+// Do tweaked AES en/decryption (i.e., XOR with \tweak, then AES en/decrypt,
+// then XOR with \tweak again) of the block(s) in \data.  To process a single
+// block, use xmm registers and set \xmm_suffix=_XMM.  To process a vector of
+// length VL, use V* registers and leave \xmm_suffix empty.  Clobbers \tmp.
+.macro	_aes_crypt	enc, xmm_suffix, tweak, data, tmp
+	_xor3		KEY0\xmm_suffix, \tweak, \data
+	cmp		$24, KEYLEN
+	jl		.Laes128\@
+	je		.Laes192\@
+	_vaes_1x	\enc, 1, \xmm_suffix, \data, tmp=\tmp
+	_vaes_1x	\enc, 2, \xmm_suffix, \data, tmp=\tmp
+.Laes192\@:
+	_vaes_1x	\enc, 3, \xmm_suffix, \data, tmp=\tmp
+	_vaes_1x	\enc, 4, \xmm_suffix, \data, tmp=\tmp
+.Laes128\@:
+.irp i, 5,6,7,8,9,10,11,12,13
+	_vaes_1x	\enc, \i, \xmm_suffix, \data, tmp=\tmp
+.endr
+.if USE_AVX512
+	vpxord		KEY14\xmm_suffix, \tweak, \tmp
+.else
+.ifnb \xmm_suffix
+	vpxor		7*16(KEY), \tweak, \tmp
+.else
+	_vbroadcast128	7*16(KEY), \tmp
+	vpxor		\tweak, \tmp, \tmp
+.endif
+.endif
+	_vaeslast	\enc, \tmp, \data
+.endm
+
+.macro	_aes_xts_crypt	enc
+	_define_aliases
+
+.if !\enc
+	// When decrypting a message whose length isn't a multiple of the AES
+	// block length, exclude the last full block from the main loop by
+	// subtracting 16 from LEN.  This is needed because ciphertext stealing
+	// decryption uses the last two tweaks in reverse order.  We'll handle
+	// the last full block and the partial block specially at the end.
+	lea		-16(LEN), %eax
+	test		$15, LEN8
+	cmovnz		%eax, LEN
+.endif
+
+	// Load the AES key length: 16 (AES-128), 24 (AES-192), or 32 (AES-256).
+	movl		480(KEY), KEYLEN
+
+	// Setup the pointer to the round keys and cache as many as possible.
+	_setup_round_keys	\enc
+
+	// Compute the first set of tweaks TWEAK[0-3].
+	_compute_first_set_of_tweaks
+
+	add		$-4*VL, LEN  // shorter than 'sub 4*VL' when VL=32
+	jl		.Lhandle_remainder\@
+
+.Lmain_loop\@:
+	// This is the main loop, en/decrypting 4*VL bytes per iteration.
+
+	// XOR each source block with its tweak and the zero-th round key.
+.if USE_AVX512
+	vmovdqu8	0*VL(SRC), V0
+	vmovdqu8	1*VL(SRC), V1
+	vmovdqu8	2*VL(SRC), V2
+	vmovdqu8	3*VL(SRC), V3
+	vpternlogd	$0x96, TWEAK0, KEY0, V0
+	vpternlogd	$0x96, TWEAK1, KEY0, V1
+	vpternlogd	$0x96, TWEAK2, KEY0, V2
+	vpternlogd	$0x96, TWEAK3, KEY0, V3
+.else
+	vpxor		0*VL(SRC), KEY0, V0
+	vpxor		1*VL(SRC), KEY0, V1
+	vpxor		2*VL(SRC), KEY0, V2
+	vpxor		3*VL(SRC), KEY0, V3
+	vpxor		TWEAK0, V0, V0
+	vpxor		TWEAK1, V1, V1
+	vpxor		TWEAK2, V2, V2
+	vpxor		TWEAK3, V3, V3
+.endif
+	cmp		$24, KEYLEN
+	jl		.Laes128\@
+	je		.Laes192\@
+	// Do all the AES rounds on the data blocks, interleaved with
+	// the computation of the next set of tweaks.
+	_vaes_4x	\enc, 1
+	_vaes_4x	\enc, 2
+.Laes192\@:
+	_vaes_4x	\enc, 3
+	_vaes_4x	\enc, 4
+.Laes128\@:
+.irp i, 5,6,7,8,9,10,11,12,13
+	_vaes_4x	\enc, \i
+.endr
+	// Do the last AES round, then XOR the results with the tweaks again.
+	// Reduce latency by doing the XOR before the vaesenclast, utilizing the
+	// property vaesenclast(key, a) ^ b == vaesenclast(key ^ b, a)
+	// (and likewise for vaesdeclast).
+.if USE_AVX512
+	_tweak_step	18
+	_tweak_step	19
+	vpxord		TWEAK0, KEY14, V4
+	vpxord		TWEAK1, KEY14, V5
+	_vaeslast	\enc, V4, V0
+	_vaeslast	\enc, V5, V1
+	vpxord		TWEAK2, KEY14, V4
+	vpxord		TWEAK3, KEY14, V5
+	_vaeslast	\enc, V4, V2
+	_vaeslast	\enc, V5, V3
+.else
+	_vbroadcast128	7*16(KEY), V4
+	_tweak_step	18 // uses V5
+	_tweak_step	19 // uses V5
+	vpxor		TWEAK0, V4, V5
+	_vaeslast	\enc, V5, V0
+	vpxor		TWEAK1, V4, V5
+	_vaeslast	\enc, V5, V1
+	vpxor		TWEAK2, V4, V5
+	vpxor		TWEAK3, V4, V4
+	_vaeslast	\enc, V5, V2
+	_vaeslast	\enc, V4, V3
+.endif
+
+	// Store the destination blocks.
+	_vmovdqu	V0, 0*VL(DST)
+	_vmovdqu	V1, 1*VL(DST)
+	_vmovdqu	V2, 2*VL(DST)
+	_vmovdqu	V3, 3*VL(DST)
+
+	// Finish computing the next set of tweaks.
+	_tweak_step	1000
+
+	sub		$-4*VL, SRC  // shorter than 'add 4*VL' when VL=32
+	sub		$-4*VL, DST
+	add		$-4*VL, LEN
+	jge		.Lmain_loop\@
+
+	// Check for the uncommon case where the data length isn't a multiple of
+	// 4*VL.  Handle it out-of-line in order to optimize for the common
+	// case.  In the common case, just fall through to the ret.
+	test		$4*VL-1, LEN8
+	jnz		.Lhandle_remainder\@
+.Ldone\@:
+	// Store the next tweak back to *TWEAK to support continuation calls.
+	vmovdqu		TWEAK0_XMM, (TWEAK)
+.if VL > 16
+	vzeroupper
+.endif
+	RET
+
+.Lhandle_remainder\@:
+
+	// En/decrypt any remaining full blocks, one vector at a time.
+.if VL > 16
+	add		$3*VL, LEN	// Undo extra sub of 4*VL, then sub VL.
+	jl		.Lvec_at_a_time_done\@
+.Lvec_at_a_time\@:
+	_vmovdqu	(SRC), V0
+	_aes_crypt	\enc, , TWEAK0, V0, tmp=V1
+	_vmovdqu	V0, (DST)
+	_next_tweakvec	TWEAK0, V0, V1, TWEAK0
+	add		$VL, SRC
+	add		$VL, DST
+	sub		$VL, LEN
+	jge		.Lvec_at_a_time\@
+.Lvec_at_a_time_done\@:
+	add		$VL-16, LEN	// Undo extra sub of VL, then sub 16.
+.else
+	add		$4*VL-16, LEN	// Undo extra sub of 4*VL, then sub 16.
+.endif
+
+	// En/decrypt any remaining full blocks, one at a time.
+	jl		.Lblock_at_a_time_done\@
+.Lblock_at_a_time\@:
+	vmovdqu		(SRC), %xmm0
+	_aes_crypt	\enc, _XMM, TWEAK0_XMM, %xmm0, tmp=%xmm1
+	vmovdqu		%xmm0, (DST)
+	_next_tweak	TWEAK0_XMM, %xmm0, TWEAK0_XMM
+	add		$16, SRC
+	add		$16, DST
+	sub		$16, LEN
+	jge		.Lblock_at_a_time\@
+.Lblock_at_a_time_done\@:
+	add		$16, LEN	// Undo the extra sub of 16.
+	// Now 0 <= LEN <= 15.  If LEN is zero, we're done.
+	jz		.Ldone\@
+
+	// Otherwise 1 <= LEN <= 15, but the real remaining length is 16 + LEN.
+	// Do ciphertext stealing to process the last 16 + LEN bytes.
+
+.if \enc
+	// If encrypting, the main loop already encrypted the last full block to
+	// create the CTS intermediate ciphertext.  Prepare for the rest of CTS
+	// by rewinding the pointers and loading the intermediate ciphertext.
+	sub		$16, SRC
+	sub		$16, DST
+	vmovdqu		(DST), %xmm0
+.else
+	// If decrypting, the main loop didn't decrypt the last full block
+	// because CTS decryption uses the last two tweaks in reverse order.
+	// Do it now by advancing the tweak and decrypting the last full block.
+	_next_tweak	TWEAK0_XMM, %xmm0, TWEAK1_XMM
+	vmovdqu		(SRC), %xmm0
+	_aes_crypt	\enc, _XMM, TWEAK1_XMM, %xmm0, tmp=%xmm1
+.endif
+
+.if USE_AVX512
+	// Create a mask that has the first LEN bits set.
+	mov		$-1, %r9d
+	bzhi		LEN, %r9d, %r9d
+	kmovd		%r9d, %k1
+
+	// Swap the first LEN bytes of the en/decryption of the last full block
+	// with the partial block.  Note that to support in-place en/decryption,
+	// the load from the src partial block must happen before the store to
+	// the dst partial block.
+	vmovdqa		%xmm0, %xmm1
+	vmovdqu8	16(SRC), %xmm0{%k1}
+	vmovdqu8	%xmm1, 16(DST){%k1}
+.else
+	lea		.Lcts_permute_table(%rip), %r9
+
+	// Load the src partial block, left-aligned.  Note that to support
+	// in-place en/decryption, this must happen before the store to the dst
+	// partial block.
+	vmovdqu		(SRC, LEN64, 1), %xmm1
+
+	// Shift the first LEN bytes of the en/decryption of the last full block
+	// to the end of a register, then store it to DST+LEN.  This stores the
+	// dst partial block.  It also writes to the second part of the dst last
+	// full block, but that part is overwritten later.
+	vpshufb		(%r9, LEN64, 1), %xmm0, %xmm2
+	vmovdqu		%xmm2, (DST, LEN64, 1)
+
+	// Make xmm3 contain [16-LEN,16-LEN+1,...,14,15,0x80,0x80,...].
+	sub		LEN64, %r9
+	vmovdqu		32(%r9), %xmm3
+
+	// Shift the src partial block to the beginning of its register.
+	vpshufb		%xmm3, %xmm1, %xmm1
+
+	// Do a blend to generate the src partial block followed by the second
+	// part of the en/decryption of the last full block.
+	vpblendvb	%xmm3, %xmm0, %xmm1, %xmm0
+.endif
+	// En/decrypt again and store the last full block.
+	_aes_crypt	\enc, _XMM, TWEAK0_XMM, %xmm0, tmp=%xmm1
+	vmovdqu		%xmm0, (DST)
+	jmp		.Ldone\@
+.endm
+
+// void aes_xts_encrypt_iv(const struct crypto_aes_ctx *tweak_key,
+//			   u8 iv[AES_BLOCK_SIZE]);
+//
+// Encrypt |iv| using the AES key |tweak_key| to get the first tweak.  Assumes
+// that the CPU supports AES-NI and AVX, but not necessarily VAES or AVX512.
+SYM_TYPED_FUNC_START(aes_xts_encrypt_iv)
+	.set	TWEAK_KEY,	%rdi
+	.set	IV,		%rsi
+	.set	KEYLEN,		%eax
+	.set	KEYLEN64,	%rax
+
+	vmovdqu		(IV), %xmm0
+	vpxor		(TWEAK_KEY), %xmm0, %xmm0
+	movl		480(TWEAK_KEY), KEYLEN
+	lea		-16(TWEAK_KEY, KEYLEN64, 4), TWEAK_KEY
+	cmp		$24, KEYLEN
+	jl		.Lencrypt_iv_aes128
+	je		.Lencrypt_iv_aes192
+	vaesenc		-6*16(TWEAK_KEY), %xmm0, %xmm0
+	vaesenc		-5*16(TWEAK_KEY), %xmm0, %xmm0
+.Lencrypt_iv_aes192:
+	vaesenc		-4*16(TWEAK_KEY), %xmm0, %xmm0
+	vaesenc		-3*16(TWEAK_KEY), %xmm0, %xmm0
+.Lencrypt_iv_aes128:
+.irp i, -2,-1,0,1,2,3,4,5,6
+	vaesenc		\i*16(TWEAK_KEY), %xmm0, %xmm0
+.endr
+	vaesenclast	7*16(TWEAK_KEY), %xmm0, %xmm0
+	vmovdqu		%xmm0, (IV)
+	RET
+SYM_FUNC_END(aes_xts_encrypt_iv)
+
+// Below are the actual AES-XTS encryption and decryption functions,
+// instantiated from the above macro.  They all have the following prototype:
+//
+// void (*xts_crypt_func)(const struct crypto_aes_ctx *key,
+//			  const u8 *src, u8 *dst, int len,
+//			  u8 tweak[AES_BLOCK_SIZE]);
+//
+// |key| is the data key.  |tweak| contains the next tweak; the encryption of
+// the original IV with the tweak key was already done.  This function supports
+// incremental computation, but |len| must always be >= 16 (AES_BLOCK_SIZE), and
+// |len| must be a multiple of 16 except on the last call.  If |len| is a
+// multiple of 16, then this function updates |tweak| to contain the next tweak.
+
+.set	VL, 16
+.set	USE_AVX512, 0
+SYM_TYPED_FUNC_START(aes_xts_encrypt_aesni_avx)
+	_aes_xts_crypt	1
+SYM_FUNC_END(aes_xts_encrypt_aesni_avx)
+SYM_TYPED_FUNC_START(aes_xts_decrypt_aesni_avx)
+	_aes_xts_crypt	0
+SYM_FUNC_END(aes_xts_decrypt_aesni_avx)
+
+.set	VL, 32
+.set	USE_AVX512, 0
+SYM_TYPED_FUNC_START(aes_xts_encrypt_vaes_avx2)
+	_aes_xts_crypt	1
+SYM_FUNC_END(aes_xts_encrypt_vaes_avx2)
+SYM_TYPED_FUNC_START(aes_xts_decrypt_vaes_avx2)
+	_aes_xts_crypt	0
+SYM_FUNC_END(aes_xts_decrypt_vaes_avx2)
+
+.set	VL, 64
+.set	USE_AVX512, 1
+SYM_TYPED_FUNC_START(aes_xts_encrypt_vaes_avx512)
+	_aes_xts_crypt	1
+SYM_FUNC_END(aes_xts_encrypt_vaes_avx512)
+SYM_TYPED_FUNC_START(aes_xts_decrypt_vaes_avx512)
+	_aes_xts_crypt	0
+SYM_FUNC_END(aes_xts_decrypt_vaes_avx512)
diff --git a/arch/x86/crypto/aesni-intel_asm.S b/arch/x86/crypto/aesni-intel_asm.S
new file mode 100644
index 000000000..6abe5e38a
--- /dev/null
+++ b/arch/x86/crypto/aesni-intel_asm.S
@@ -0,0 +1,1338 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Implement AES algorithm in Intel AES-NI instructions.
+ *
+ * The white paper of AES-NI instructions can be downloaded from:
+ *   http://softwarecommunity.intel.com/isn/downloads/intelavx/AES-Instructions-Set_WP.pdf
+ *
+ * Copyright (C) 2008, Intel Corp.
+ *    Author: Huang Ying <ying.huang@intel.com>
+ *            Vinodh Gopal <vinodh.gopal@intel.com>
+ *            Kahraman Akdemir
+ *
+ * Copyright (c) 2010, Intel Corporation.
+ *
+ * Ported x86_64 version to x86:
+ *    Author: Mathias Krause <minipli@googlemail.com>
+ */
+
+#include <linux/linkage.h>
+#include <linux/objtool.h>
+#include <asm/frame.h>
+
+#define STATE1	%xmm0
+#define STATE2	%xmm4
+#define STATE3	%xmm5
+#define STATE4	%xmm6
+#define STATE	STATE1
+#define IN1	%xmm1
+#define IN2	%xmm7
+#define IN3	%xmm8
+#define IN4	%xmm9
+#define IN	IN1
+#define KEY	%xmm2
+#define IV	%xmm3
+
+#define BSWAP_MASK %xmm10
+#define CTR	%xmm11
+#define INC	%xmm12
+
+#define GF128MUL_MASK %xmm7
+
+#ifdef __x86_64__
+#define AREG	%rax
+#define KEYP	%rdi
+#define OUTP	%rsi
+#define UKEYP	OUTP
+#define INP	%rdx
+#define LEN	%rcx
+#define IVP	%r8
+#define KLEN	%r9d
+#define T1	%r10
+#define TKEYP	T1
+#define T2	%r11
+#define TCTR_LOW T2
+#else
+#define AREG	%eax
+#define KEYP	%edi
+#define OUTP	AREG
+#define UKEYP	OUTP
+#define INP	%edx
+#define LEN	%esi
+#define IVP	%ebp
+#define KLEN	%ebx
+#define T1	%ecx
+#define TKEYP	T1
+#endif
+
+SYM_FUNC_START_LOCAL(_key_expansion_256a)
+	pshufd $0b11111111, %xmm1, %xmm1
+	shufps $0b00010000, %xmm0, %xmm4
+	pxor %xmm4, %xmm0
+	shufps $0b10001100, %xmm0, %xmm4
+	pxor %xmm4, %xmm0
+	pxor %xmm1, %xmm0
+	movaps %xmm0, (TKEYP)
+	add $0x10, TKEYP
+	RET
+SYM_FUNC_END(_key_expansion_256a)
+SYM_FUNC_ALIAS_LOCAL(_key_expansion_128, _key_expansion_256a)
+
+SYM_FUNC_START_LOCAL(_key_expansion_192a)
+	pshufd $0b01010101, %xmm1, %xmm1
+	shufps $0b00010000, %xmm0, %xmm4
+	pxor %xmm4, %xmm0
+	shufps $0b10001100, %xmm0, %xmm4
+	pxor %xmm4, %xmm0
+	pxor %xmm1, %xmm0
+
+	movaps %xmm2, %xmm5
+	movaps %xmm2, %xmm6
+	pslldq $4, %xmm5
+	pshufd $0b11111111, %xmm0, %xmm3
+	pxor %xmm3, %xmm2
+	pxor %xmm5, %xmm2
+
+	movaps %xmm0, %xmm1
+	shufps $0b01000100, %xmm0, %xmm6
+	movaps %xmm6, (TKEYP)
+	shufps $0b01001110, %xmm2, %xmm1
+	movaps %xmm1, 0x10(TKEYP)
+	add $0x20, TKEYP
+	RET
+SYM_FUNC_END(_key_expansion_192a)
+
+SYM_FUNC_START_LOCAL(_key_expansion_192b)
+	pshufd $0b01010101, %xmm1, %xmm1
+	shufps $0b00010000, %xmm0, %xmm4
+	pxor %xmm4, %xmm0
+	shufps $0b10001100, %xmm0, %xmm4
+	pxor %xmm4, %xmm0
+	pxor %xmm1, %xmm0
+
+	movaps %xmm2, %xmm5
+	pslldq $4, %xmm5
+	pshufd $0b11111111, %xmm0, %xmm3
+	pxor %xmm3, %xmm2
+	pxor %xmm5, %xmm2
+
+	movaps %xmm0, (TKEYP)
+	add $0x10, TKEYP
+	RET
+SYM_FUNC_END(_key_expansion_192b)
+
+SYM_FUNC_START_LOCAL(_key_expansion_256b)
+	pshufd $0b10101010, %xmm1, %xmm1
+	shufps $0b00010000, %xmm2, %xmm4
+	pxor %xmm4, %xmm2
+	shufps $0b10001100, %xmm2, %xmm4
+	pxor %xmm4, %xmm2
+	pxor %xmm1, %xmm2
+	movaps %xmm2, (TKEYP)
+	add $0x10, TKEYP
+	RET
+SYM_FUNC_END(_key_expansion_256b)
+
+/*
+ * void aesni_set_key(struct crypto_aes_ctx *ctx, const u8 *in_key,
+ *                    unsigned int key_len)
+ */
+SYM_FUNC_START(aesni_set_key)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl KEYP
+	movl (FRAME_OFFSET+8)(%esp), KEYP	# ctx
+	movl (FRAME_OFFSET+12)(%esp), UKEYP	# in_key
+	movl (FRAME_OFFSET+16)(%esp), %edx	# key_len
+#endif
+	movups (UKEYP), %xmm0		# user key (first 16 bytes)
+	movaps %xmm0, (KEYP)
+	lea 0x10(KEYP), TKEYP		# key addr
+	movl %edx, 480(KEYP)
+	pxor %xmm4, %xmm4		# xmm4 is assumed 0 in _key_expansion_x
+	cmp $24, %dl
+	jb .Lenc_key128
+	je .Lenc_key192
+	movups 0x10(UKEYP), %xmm2	# other user key
+	movaps %xmm2, (TKEYP)
+	add $0x10, TKEYP
+	aeskeygenassist $0x1, %xmm2, %xmm1	# round 1
+	call _key_expansion_256a
+	aeskeygenassist $0x1, %xmm0, %xmm1
+	call _key_expansion_256b
+	aeskeygenassist $0x2, %xmm2, %xmm1	# round 2
+	call _key_expansion_256a
+	aeskeygenassist $0x2, %xmm0, %xmm1
+	call _key_expansion_256b
+	aeskeygenassist $0x4, %xmm2, %xmm1	# round 3
+	call _key_expansion_256a
+	aeskeygenassist $0x4, %xmm0, %xmm1
+	call _key_expansion_256b
+	aeskeygenassist $0x8, %xmm2, %xmm1	# round 4
+	call _key_expansion_256a
+	aeskeygenassist $0x8, %xmm0, %xmm1
+	call _key_expansion_256b
+	aeskeygenassist $0x10, %xmm2, %xmm1	# round 5
+	call _key_expansion_256a
+	aeskeygenassist $0x10, %xmm0, %xmm1
+	call _key_expansion_256b
+	aeskeygenassist $0x20, %xmm2, %xmm1	# round 6
+	call _key_expansion_256a
+	aeskeygenassist $0x20, %xmm0, %xmm1
+	call _key_expansion_256b
+	aeskeygenassist $0x40, %xmm2, %xmm1	# round 7
+	call _key_expansion_256a
+	jmp .Ldec_key
+.Lenc_key192:
+	movq 0x10(UKEYP), %xmm2		# other user key
+	aeskeygenassist $0x1, %xmm2, %xmm1	# round 1
+	call _key_expansion_192a
+	aeskeygenassist $0x2, %xmm2, %xmm1	# round 2
+	call _key_expansion_192b
+	aeskeygenassist $0x4, %xmm2, %xmm1	# round 3
+	call _key_expansion_192a
+	aeskeygenassist $0x8, %xmm2, %xmm1	# round 4
+	call _key_expansion_192b
+	aeskeygenassist $0x10, %xmm2, %xmm1	# round 5
+	call _key_expansion_192a
+	aeskeygenassist $0x20, %xmm2, %xmm1	# round 6
+	call _key_expansion_192b
+	aeskeygenassist $0x40, %xmm2, %xmm1	# round 7
+	call _key_expansion_192a
+	aeskeygenassist $0x80, %xmm2, %xmm1	# round 8
+	call _key_expansion_192b
+	jmp .Ldec_key
+.Lenc_key128:
+	aeskeygenassist $0x1, %xmm0, %xmm1	# round 1
+	call _key_expansion_128
+	aeskeygenassist $0x2, %xmm0, %xmm1	# round 2
+	call _key_expansion_128
+	aeskeygenassist $0x4, %xmm0, %xmm1	# round 3
+	call _key_expansion_128
+	aeskeygenassist $0x8, %xmm0, %xmm1	# round 4
+	call _key_expansion_128
+	aeskeygenassist $0x10, %xmm0, %xmm1	# round 5
+	call _key_expansion_128
+	aeskeygenassist $0x20, %xmm0, %xmm1	# round 6
+	call _key_expansion_128
+	aeskeygenassist $0x40, %xmm0, %xmm1	# round 7
+	call _key_expansion_128
+	aeskeygenassist $0x80, %xmm0, %xmm1	# round 8
+	call _key_expansion_128
+	aeskeygenassist $0x1b, %xmm0, %xmm1	# round 9
+	call _key_expansion_128
+	aeskeygenassist $0x36, %xmm0, %xmm1	# round 10
+	call _key_expansion_128
+.Ldec_key:
+	sub $0x10, TKEYP
+	movaps (KEYP), %xmm0
+	movaps (TKEYP), %xmm1
+	movaps %xmm0, 240(TKEYP)
+	movaps %xmm1, 240(KEYP)
+	add $0x10, KEYP
+	lea 240-16(TKEYP), UKEYP
+.align 4
+.Ldec_key_loop:
+	movaps (KEYP), %xmm0
+	aesimc %xmm0, %xmm1
+	movaps %xmm1, (UKEYP)
+	add $0x10, KEYP
+	sub $0x10, UKEYP
+	cmp TKEYP, KEYP
+	jb .Ldec_key_loop
+#ifndef __x86_64__
+	popl KEYP
+#endif
+	FRAME_END
+	RET
+SYM_FUNC_END(aesni_set_key)
+
+/*
+ * void aesni_enc(const void *ctx, u8 *dst, const u8 *src)
+ */
+SYM_FUNC_START(aesni_enc)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl KEYP
+	pushl KLEN
+	movl (FRAME_OFFSET+12)(%esp), KEYP	# ctx
+	movl (FRAME_OFFSET+16)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+20)(%esp), INP	# src
+#endif
+	movl 480(KEYP), KLEN		# key length
+	movups (INP), STATE		# input
+	call _aesni_enc1
+	movups STATE, (OUTP)		# output
+#ifndef __x86_64__
+	popl KLEN
+	popl KEYP
+#endif
+	FRAME_END
+	RET
+SYM_FUNC_END(aesni_enc)
+
+/*
+ * _aesni_enc1:		internal ABI
+ * input:
+ *	KEYP:		key struct pointer
+ *	KLEN:		round count
+ *	STATE:		initial state (input)
+ * output:
+ *	STATE:		finial state (output)
+ * changed:
+ *	KEY
+ *	TKEYP (T1)
+ */
+SYM_FUNC_START_LOCAL(_aesni_enc1)
+	movaps (KEYP), KEY		# key
+	mov KEYP, TKEYP
+	pxor KEY, STATE		# round 0
+	add $0x30, TKEYP
+	cmp $24, KLEN
+	jb .Lenc128
+	lea 0x20(TKEYP), TKEYP
+	je .Lenc192
+	add $0x20, TKEYP
+	movaps -0x60(TKEYP), KEY
+	aesenc KEY, STATE
+	movaps -0x50(TKEYP), KEY
+	aesenc KEY, STATE
+.align 4
+.Lenc192:
+	movaps -0x40(TKEYP), KEY
+	aesenc KEY, STATE
+	movaps -0x30(TKEYP), KEY
+	aesenc KEY, STATE
+.align 4
+.Lenc128:
+	movaps -0x20(TKEYP), KEY
+	aesenc KEY, STATE
+	movaps -0x10(TKEYP), KEY
+	aesenc KEY, STATE
+	movaps (TKEYP), KEY
+	aesenc KEY, STATE
+	movaps 0x10(TKEYP), KEY
+	aesenc KEY, STATE
+	movaps 0x20(TKEYP), KEY
+	aesenc KEY, STATE
+	movaps 0x30(TKEYP), KEY
+	aesenc KEY, STATE
+	movaps 0x40(TKEYP), KEY
+	aesenc KEY, STATE
+	movaps 0x50(TKEYP), KEY
+	aesenc KEY, STATE
+	movaps 0x60(TKEYP), KEY
+	aesenc KEY, STATE
+	movaps 0x70(TKEYP), KEY
+	aesenclast KEY, STATE
+	RET
+SYM_FUNC_END(_aesni_enc1)
+
+/*
+ * _aesni_enc4:	internal ABI
+ * input:
+ *	KEYP:		key struct pointer
+ *	KLEN:		round count
+ *	STATE1:		initial state (input)
+ *	STATE2
+ *	STATE3
+ *	STATE4
+ * output:
+ *	STATE1:		finial state (output)
+ *	STATE2
+ *	STATE3
+ *	STATE4
+ * changed:
+ *	KEY
+ *	TKEYP (T1)
+ */
+SYM_FUNC_START_LOCAL(_aesni_enc4)
+	movaps (KEYP), KEY		# key
+	mov KEYP, TKEYP
+	pxor KEY, STATE1		# round 0
+	pxor KEY, STATE2
+	pxor KEY, STATE3
+	pxor KEY, STATE4
+	add $0x30, TKEYP
+	cmp $24, KLEN
+	jb .L4enc128
+	lea 0x20(TKEYP), TKEYP
+	je .L4enc192
+	add $0x20, TKEYP
+	movaps -0x60(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+	movaps -0x50(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+#.align 4
+.L4enc192:
+	movaps -0x40(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+	movaps -0x30(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+#.align 4
+.L4enc128:
+	movaps -0x20(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+	movaps -0x10(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+	movaps (TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+	movaps 0x10(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+	movaps 0x20(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+	movaps 0x30(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+	movaps 0x40(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+	movaps 0x50(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+	movaps 0x60(TKEYP), KEY
+	aesenc KEY, STATE1
+	aesenc KEY, STATE2
+	aesenc KEY, STATE3
+	aesenc KEY, STATE4
+	movaps 0x70(TKEYP), KEY
+	aesenclast KEY, STATE1		# last round
+	aesenclast KEY, STATE2
+	aesenclast KEY, STATE3
+	aesenclast KEY, STATE4
+	RET
+SYM_FUNC_END(_aesni_enc4)
+
+/*
+ * _aesni_dec1:		internal ABI
+ * input:
+ *	KEYP:		key struct pointer
+ *	KLEN:		key length
+ *	STATE:		initial state (input)
+ * output:
+ *	STATE:		finial state (output)
+ * changed:
+ *	KEY
+ *	TKEYP (T1)
+ */
+SYM_FUNC_START_LOCAL(_aesni_dec1)
+	movaps (KEYP), KEY		# key
+	mov KEYP, TKEYP
+	pxor KEY, STATE		# round 0
+	add $0x30, TKEYP
+	cmp $24, KLEN
+	jb .Ldec128
+	lea 0x20(TKEYP), TKEYP
+	je .Ldec192
+	add $0x20, TKEYP
+	movaps -0x60(TKEYP), KEY
+	aesdec KEY, STATE
+	movaps -0x50(TKEYP), KEY
+	aesdec KEY, STATE
+.align 4
+.Ldec192:
+	movaps -0x40(TKEYP), KEY
+	aesdec KEY, STATE
+	movaps -0x30(TKEYP), KEY
+	aesdec KEY, STATE
+.align 4
+.Ldec128:
+	movaps -0x20(TKEYP), KEY
+	aesdec KEY, STATE
+	movaps -0x10(TKEYP), KEY
+	aesdec KEY, STATE
+	movaps (TKEYP), KEY
+	aesdec KEY, STATE
+	movaps 0x10(TKEYP), KEY
+	aesdec KEY, STATE
+	movaps 0x20(TKEYP), KEY
+	aesdec KEY, STATE
+	movaps 0x30(TKEYP), KEY
+	aesdec KEY, STATE
+	movaps 0x40(TKEYP), KEY
+	aesdec KEY, STATE
+	movaps 0x50(TKEYP), KEY
+	aesdec KEY, STATE
+	movaps 0x60(TKEYP), KEY
+	aesdec KEY, STATE
+	movaps 0x70(TKEYP), KEY
+	aesdeclast KEY, STATE
+	RET
+SYM_FUNC_END(_aesni_dec1)
+
+/*
+ * _aesni_dec4:	internal ABI
+ * input:
+ *	KEYP:		key struct pointer
+ *	KLEN:		key length
+ *	STATE1:		initial state (input)
+ *	STATE2
+ *	STATE3
+ *	STATE4
+ * output:
+ *	STATE1:		finial state (output)
+ *	STATE2
+ *	STATE3
+ *	STATE4
+ * changed:
+ *	KEY
+ *	TKEYP (T1)
+ */
+SYM_FUNC_START_LOCAL(_aesni_dec4)
+	movaps (KEYP), KEY		# key
+	mov KEYP, TKEYP
+	pxor KEY, STATE1		# round 0
+	pxor KEY, STATE2
+	pxor KEY, STATE3
+	pxor KEY, STATE4
+	add $0x30, TKEYP
+	cmp $24, KLEN
+	jb .L4dec128
+	lea 0x20(TKEYP), TKEYP
+	je .L4dec192
+	add $0x20, TKEYP
+	movaps -0x60(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+	movaps -0x50(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+.align 4
+.L4dec192:
+	movaps -0x40(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+	movaps -0x30(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+.align 4
+.L4dec128:
+	movaps -0x20(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+	movaps -0x10(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+	movaps (TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+	movaps 0x10(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+	movaps 0x20(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+	movaps 0x30(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+	movaps 0x40(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+	movaps 0x50(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+	movaps 0x60(TKEYP), KEY
+	aesdec KEY, STATE1
+	aesdec KEY, STATE2
+	aesdec KEY, STATE3
+	aesdec KEY, STATE4
+	movaps 0x70(TKEYP), KEY
+	aesdeclast KEY, STATE1		# last round
+	aesdeclast KEY, STATE2
+	aesdeclast KEY, STATE3
+	aesdeclast KEY, STATE4
+	RET
+SYM_FUNC_END(_aesni_dec4)
+
+/*
+ * void aesni_ecb_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
+ *		      size_t len)
+ */
+SYM_FUNC_START(aesni_ecb_enc)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl LEN
+	pushl KEYP
+	pushl KLEN
+	movl (FRAME_OFFSET+16)(%esp), KEYP	# ctx
+	movl (FRAME_OFFSET+20)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+24)(%esp), INP	# src
+	movl (FRAME_OFFSET+28)(%esp), LEN	# len
+#endif
+	test LEN, LEN		# check length
+	jz .Lecb_enc_ret
+	mov 480(KEYP), KLEN
+	cmp $16, LEN
+	jb .Lecb_enc_ret
+	cmp $64, LEN
+	jb .Lecb_enc_loop1
+.align 4
+.Lecb_enc_loop4:
+	movups (INP), STATE1
+	movups 0x10(INP), STATE2
+	movups 0x20(INP), STATE3
+	movups 0x30(INP), STATE4
+	call _aesni_enc4
+	movups STATE1, (OUTP)
+	movups STATE2, 0x10(OUTP)
+	movups STATE3, 0x20(OUTP)
+	movups STATE4, 0x30(OUTP)
+	sub $64, LEN
+	add $64, INP
+	add $64, OUTP
+	cmp $64, LEN
+	jge .Lecb_enc_loop4
+	cmp $16, LEN
+	jb .Lecb_enc_ret
+.align 4
+.Lecb_enc_loop1:
+	movups (INP), STATE1
+	call _aesni_enc1
+	movups STATE1, (OUTP)
+	sub $16, LEN
+	add $16, INP
+	add $16, OUTP
+	cmp $16, LEN
+	jge .Lecb_enc_loop1
+.Lecb_enc_ret:
+#ifndef __x86_64__
+	popl KLEN
+	popl KEYP
+	popl LEN
+#endif
+	FRAME_END
+	RET
+SYM_FUNC_END(aesni_ecb_enc)
+
+/*
+ * void aesni_ecb_dec(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
+ *		      size_t len);
+ */
+SYM_FUNC_START(aesni_ecb_dec)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl LEN
+	pushl KEYP
+	pushl KLEN
+	movl (FRAME_OFFSET+16)(%esp), KEYP	# ctx
+	movl (FRAME_OFFSET+20)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+24)(%esp), INP	# src
+	movl (FRAME_OFFSET+28)(%esp), LEN	# len
+#endif
+	test LEN, LEN
+	jz .Lecb_dec_ret
+	mov 480(KEYP), KLEN
+	add $240, KEYP
+	cmp $16, LEN
+	jb .Lecb_dec_ret
+	cmp $64, LEN
+	jb .Lecb_dec_loop1
+.align 4
+.Lecb_dec_loop4:
+	movups (INP), STATE1
+	movups 0x10(INP), STATE2
+	movups 0x20(INP), STATE3
+	movups 0x30(INP), STATE4
+	call _aesni_dec4
+	movups STATE1, (OUTP)
+	movups STATE2, 0x10(OUTP)
+	movups STATE3, 0x20(OUTP)
+	movups STATE4, 0x30(OUTP)
+	sub $64, LEN
+	add $64, INP
+	add $64, OUTP
+	cmp $64, LEN
+	jge .Lecb_dec_loop4
+	cmp $16, LEN
+	jb .Lecb_dec_ret
+.align 4
+.Lecb_dec_loop1:
+	movups (INP), STATE1
+	call _aesni_dec1
+	movups STATE1, (OUTP)
+	sub $16, LEN
+	add $16, INP
+	add $16, OUTP
+	cmp $16, LEN
+	jge .Lecb_dec_loop1
+.Lecb_dec_ret:
+#ifndef __x86_64__
+	popl KLEN
+	popl KEYP
+	popl LEN
+#endif
+	FRAME_END
+	RET
+SYM_FUNC_END(aesni_ecb_dec)
+
+/*
+ * void aesni_cbc_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
+ *		      size_t len, u8 *iv)
+ */
+SYM_FUNC_START(aesni_cbc_enc)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl IVP
+	pushl LEN
+	pushl KEYP
+	pushl KLEN
+	movl (FRAME_OFFSET+20)(%esp), KEYP	# ctx
+	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+28)(%esp), INP	# src
+	movl (FRAME_OFFSET+32)(%esp), LEN	# len
+	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
+#endif
+	cmp $16, LEN
+	jb .Lcbc_enc_ret
+	mov 480(KEYP), KLEN
+	movups (IVP), STATE	# load iv as initial state
+.align 4
+.Lcbc_enc_loop:
+	movups (INP), IN	# load input
+	pxor IN, STATE
+	call _aesni_enc1
+	movups STATE, (OUTP)	# store output
+	sub $16, LEN
+	add $16, INP
+	add $16, OUTP
+	cmp $16, LEN
+	jge .Lcbc_enc_loop
+	movups STATE, (IVP)
+.Lcbc_enc_ret:
+#ifndef __x86_64__
+	popl KLEN
+	popl KEYP
+	popl LEN
+	popl IVP
+#endif
+	FRAME_END
+	RET
+SYM_FUNC_END(aesni_cbc_enc)
+
+/*
+ * void aesni_cbc_dec(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
+ *		      size_t len, u8 *iv)
+ */
+SYM_FUNC_START(aesni_cbc_dec)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl IVP
+	pushl LEN
+	pushl KEYP
+	pushl KLEN
+	movl (FRAME_OFFSET+20)(%esp), KEYP	# ctx
+	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+28)(%esp), INP	# src
+	movl (FRAME_OFFSET+32)(%esp), LEN	# len
+	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
+#endif
+	cmp $16, LEN
+	jb .Lcbc_dec_just_ret
+	mov 480(KEYP), KLEN
+	add $240, KEYP
+	movups (IVP), IV
+	cmp $64, LEN
+	jb .Lcbc_dec_loop1
+.align 4
+.Lcbc_dec_loop4:
+	movups (INP), IN1
+	movaps IN1, STATE1
+	movups 0x10(INP), IN2
+	movaps IN2, STATE2
+#ifdef __x86_64__
+	movups 0x20(INP), IN3
+	movaps IN3, STATE3
+	movups 0x30(INP), IN4
+	movaps IN4, STATE4
+#else
+	movups 0x20(INP), IN1
+	movaps IN1, STATE3
+	movups 0x30(INP), IN2
+	movaps IN2, STATE4
+#endif
+	call _aesni_dec4
+	pxor IV, STATE1
+#ifdef __x86_64__
+	pxor IN1, STATE2
+	pxor IN2, STATE3
+	pxor IN3, STATE4
+	movaps IN4, IV
+#else
+	pxor IN1, STATE4
+	movaps IN2, IV
+	movups (INP), IN1
+	pxor IN1, STATE2
+	movups 0x10(INP), IN2
+	pxor IN2, STATE3
+#endif
+	movups STATE1, (OUTP)
+	movups STATE2, 0x10(OUTP)
+	movups STATE3, 0x20(OUTP)
+	movups STATE4, 0x30(OUTP)
+	sub $64, LEN
+	add $64, INP
+	add $64, OUTP
+	cmp $64, LEN
+	jge .Lcbc_dec_loop4
+	cmp $16, LEN
+	jb .Lcbc_dec_ret
+.align 4
+.Lcbc_dec_loop1:
+	movups (INP), IN
+	movaps IN, STATE
+	call _aesni_dec1
+	pxor IV, STATE
+	movups STATE, (OUTP)
+	movaps IN, IV
+	sub $16, LEN
+	add $16, INP
+	add $16, OUTP
+	cmp $16, LEN
+	jge .Lcbc_dec_loop1
+.Lcbc_dec_ret:
+	movups IV, (IVP)
+.Lcbc_dec_just_ret:
+#ifndef __x86_64__
+	popl KLEN
+	popl KEYP
+	popl LEN
+	popl IVP
+#endif
+	FRAME_END
+	RET
+SYM_FUNC_END(aesni_cbc_dec)
+
+/*
+ * void aesni_cts_cbc_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
+ *			  size_t len, u8 *iv)
+ */
+SYM_FUNC_START(aesni_cts_cbc_enc)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl IVP
+	pushl LEN
+	pushl KEYP
+	pushl KLEN
+	movl (FRAME_OFFSET+20)(%esp), KEYP	# ctx
+	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+28)(%esp), INP	# src
+	movl (FRAME_OFFSET+32)(%esp), LEN	# len
+	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
+	lea .Lcts_permute_table, T1
+#else
+	lea .Lcts_permute_table(%rip), T1
+#endif
+	mov 480(KEYP), KLEN
+	movups (IVP), STATE
+	sub $16, LEN
+	mov T1, IVP
+	add $32, IVP
+	add LEN, T1
+	sub LEN, IVP
+	movups (T1), %xmm4
+	movups (IVP), %xmm5
+
+	movups (INP), IN1
+	add LEN, INP
+	movups (INP), IN2
+
+	pxor IN1, STATE
+	call _aesni_enc1
+
+	pshufb %xmm5, IN2
+	pxor STATE, IN2
+	pshufb %xmm4, STATE
+	add OUTP, LEN
+	movups STATE, (LEN)
+
+	movaps IN2, STATE
+	call _aesni_enc1
+	movups STATE, (OUTP)
+
+#ifndef __x86_64__
+	popl KLEN
+	popl KEYP
+	popl LEN
+	popl IVP
+#endif
+	FRAME_END
+	RET
+SYM_FUNC_END(aesni_cts_cbc_enc)
+
+/*
+ * void aesni_cts_cbc_dec(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
+ *			  size_t len, u8 *iv)
+ */
+SYM_FUNC_START(aesni_cts_cbc_dec)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl IVP
+	pushl LEN
+	pushl KEYP
+	pushl KLEN
+	movl (FRAME_OFFSET+20)(%esp), KEYP	# ctx
+	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+28)(%esp), INP	# src
+	movl (FRAME_OFFSET+32)(%esp), LEN	# len
+	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
+	lea .Lcts_permute_table, T1
+#else
+	lea .Lcts_permute_table(%rip), T1
+#endif
+	mov 480(KEYP), KLEN
+	add $240, KEYP
+	movups (IVP), IV
+	sub $16, LEN
+	mov T1, IVP
+	add $32, IVP
+	add LEN, T1
+	sub LEN, IVP
+	movups (T1), %xmm4
+
+	movups (INP), STATE
+	add LEN, INP
+	movups (INP), IN1
+
+	call _aesni_dec1
+	movaps STATE, IN2
+	pshufb %xmm4, STATE
+	pxor IN1, STATE
+
+	add OUTP, LEN
+	movups STATE, (LEN)
+
+	movups (IVP), %xmm0
+	pshufb %xmm0, IN1
+	pblendvb IN2, IN1
+	movaps IN1, STATE
+	call _aesni_dec1
+
+	pxor IV, STATE
+	movups STATE, (OUTP)
+
+#ifndef __x86_64__
+	popl KLEN
+	popl KEYP
+	popl LEN
+	popl IVP
+#endif
+	FRAME_END
+	RET
+SYM_FUNC_END(aesni_cts_cbc_dec)
+
+.pushsection .rodata
+.align 16
+.Lcts_permute_table:
+	.byte		0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+	.byte		0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+	.byte		0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07
+	.byte		0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f
+	.byte		0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+	.byte		0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80
+#ifdef __x86_64__
+.Lbswap_mask:
+	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+#endif
+.popsection
+
+#ifdef __x86_64__
+/*
+ * _aesni_inc_init:	internal ABI
+ *	setup registers used by _aesni_inc
+ * input:
+ *	IV
+ * output:
+ *	CTR:	== IV, in little endian
+ *	TCTR_LOW: == lower qword of CTR
+ *	INC:	== 1, in little endian
+ *	BSWAP_MASK == endian swapping mask
+ */
+SYM_FUNC_START_LOCAL(_aesni_inc_init)
+	movaps .Lbswap_mask(%rip), BSWAP_MASK
+	movaps IV, CTR
+	pshufb BSWAP_MASK, CTR
+	mov $1, TCTR_LOW
+	movq TCTR_LOW, INC
+	movq CTR, TCTR_LOW
+	RET
+SYM_FUNC_END(_aesni_inc_init)
+
+/*
+ * _aesni_inc:		internal ABI
+ *	Increase IV by 1, IV is in big endian
+ * input:
+ *	IV
+ *	CTR:	== IV, in little endian
+ *	TCTR_LOW: == lower qword of CTR
+ *	INC:	== 1, in little endian
+ *	BSWAP_MASK == endian swapping mask
+ * output:
+ *	IV:	Increase by 1
+ * changed:
+ *	CTR:	== output IV, in little endian
+ *	TCTR_LOW: == lower qword of CTR
+ */
+SYM_FUNC_START_LOCAL(_aesni_inc)
+	paddq INC, CTR
+	add $1, TCTR_LOW
+	jnc .Linc_low
+	pslldq $8, INC
+	paddq INC, CTR
+	psrldq $8, INC
+.Linc_low:
+	movaps CTR, IV
+	pshufb BSWAP_MASK, IV
+	RET
+SYM_FUNC_END(_aesni_inc)
+
+/*
+ * void aesni_ctr_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
+ *		      size_t len, u8 *iv)
+ */
+SYM_FUNC_START(aesni_ctr_enc)
+	ANNOTATE_NOENDBR
+	FRAME_BEGIN
+	cmp $16, LEN
+	jb .Lctr_enc_just_ret
+	mov 480(KEYP), KLEN
+	movups (IVP), IV
+	call _aesni_inc_init
+	cmp $64, LEN
+	jb .Lctr_enc_loop1
+.align 4
+.Lctr_enc_loop4:
+	movaps IV, STATE1
+	call _aesni_inc
+	movups (INP), IN1
+	movaps IV, STATE2
+	call _aesni_inc
+	movups 0x10(INP), IN2
+	movaps IV, STATE3
+	call _aesni_inc
+	movups 0x20(INP), IN3
+	movaps IV, STATE4
+	call _aesni_inc
+	movups 0x30(INP), IN4
+	call _aesni_enc4
+	pxor IN1, STATE1
+	movups STATE1, (OUTP)
+	pxor IN2, STATE2
+	movups STATE2, 0x10(OUTP)
+	pxor IN3, STATE3
+	movups STATE3, 0x20(OUTP)
+	pxor IN4, STATE4
+	movups STATE4, 0x30(OUTP)
+	sub $64, LEN
+	add $64, INP
+	add $64, OUTP
+	cmp $64, LEN
+	jge .Lctr_enc_loop4
+	cmp $16, LEN
+	jb .Lctr_enc_ret
+.align 4
+.Lctr_enc_loop1:
+	movaps IV, STATE
+	call _aesni_inc
+	movups (INP), IN
+	call _aesni_enc1
+	pxor IN, STATE
+	movups STATE, (OUTP)
+	sub $16, LEN
+	add $16, INP
+	add $16, OUTP
+	cmp $16, LEN
+	jge .Lctr_enc_loop1
+.Lctr_enc_ret:
+	movups IV, (IVP)
+.Lctr_enc_just_ret:
+	FRAME_END
+	RET
+SYM_FUNC_END(aesni_ctr_enc)
+
+#endif
+
+.section	.rodata.cst16.gf128mul_x_ble_mask, "aM", @progbits, 16
+.align 16
+.Lgf128mul_x_ble_mask:
+	.octa 0x00000000000000010000000000000087
+.previous
+
+/*
+ * _aesni_gf128mul_x_ble: Multiply in GF(2^128) for XTS IVs
+ * input:
+ *	IV:	current IV
+ *	GF128MUL_MASK == mask with 0x87 and 0x01
+ * output:
+ *	IV:	next IV
+ * changed:
+ *	KEY:	== temporary value
+ */
+.macro _aesni_gf128mul_x_ble
+	pshufd $0x13, IV, KEY
+	paddq IV, IV
+	psrad $31, KEY
+	pand GF128MUL_MASK, KEY
+	pxor KEY, IV
+.endm
+
+.macro	_aesni_xts_crypt	enc
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl IVP
+	pushl LEN
+	pushl KEYP
+	pushl KLEN
+	movl (FRAME_OFFSET+20)(%esp), KEYP	# ctx
+	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+28)(%esp), INP	# src
+	movl (FRAME_OFFSET+32)(%esp), LEN	# len
+	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
+	movdqa .Lgf128mul_x_ble_mask, GF128MUL_MASK
+#else
+	movdqa .Lgf128mul_x_ble_mask(%rip), GF128MUL_MASK
+#endif
+	movups (IVP), IV
+
+	mov 480(KEYP), KLEN
+.if !\enc
+	add $240, KEYP
+
+	test $15, LEN
+	jz .Lxts_loop4\@
+	sub $16, LEN
+.endif
+
+.Lxts_loop4\@:
+	sub $64, LEN
+	jl .Lxts_1x\@
+
+	movdqa IV, STATE1
+	movdqu 0x00(INP), IN
+	pxor IN, STATE1
+	movdqu IV, 0x00(OUTP)
+
+	_aesni_gf128mul_x_ble
+	movdqa IV, STATE2
+	movdqu 0x10(INP), IN
+	pxor IN, STATE2
+	movdqu IV, 0x10(OUTP)
+
+	_aesni_gf128mul_x_ble
+	movdqa IV, STATE3
+	movdqu 0x20(INP), IN
+	pxor IN, STATE3
+	movdqu IV, 0x20(OUTP)
+
+	_aesni_gf128mul_x_ble
+	movdqa IV, STATE4
+	movdqu 0x30(INP), IN
+	pxor IN, STATE4
+	movdqu IV, 0x30(OUTP)
+
+.if \enc
+	call _aesni_enc4
+.else
+	call _aesni_dec4
+.endif
+
+	movdqu 0x00(OUTP), IN
+	pxor IN, STATE1
+	movdqu STATE1, 0x00(OUTP)
+
+	movdqu 0x10(OUTP), IN
+	pxor IN, STATE2
+	movdqu STATE2, 0x10(OUTP)
+
+	movdqu 0x20(OUTP), IN
+	pxor IN, STATE3
+	movdqu STATE3, 0x20(OUTP)
+
+	movdqu 0x30(OUTP), IN
+	pxor IN, STATE4
+	movdqu STATE4, 0x30(OUTP)
+
+	_aesni_gf128mul_x_ble
+
+	add $64, INP
+	add $64, OUTP
+	test LEN, LEN
+	jnz .Lxts_loop4\@
+
+.Lxts_ret_iv\@:
+	movups IV, (IVP)
+
+.Lxts_ret\@:
+#ifndef __x86_64__
+	popl KLEN
+	popl KEYP
+	popl LEN
+	popl IVP
+#endif
+	FRAME_END
+	RET
+
+.Lxts_1x\@:
+	add $64, LEN
+	jz .Lxts_ret_iv\@
+.if \enc
+	sub $16, LEN
+	jl .Lxts_cts4\@
+.endif
+
+.Lxts_loop1\@:
+	movdqu (INP), STATE
+.if \enc
+	pxor IV, STATE
+	call _aesni_enc1
+.else
+	add $16, INP
+	sub $16, LEN
+	jl .Lxts_cts1\@
+	pxor IV, STATE
+	call _aesni_dec1
+.endif
+	pxor IV, STATE
+	_aesni_gf128mul_x_ble
+
+	test LEN, LEN
+	jz .Lxts_out\@
+
+.if \enc
+	add $16, INP
+	sub $16, LEN
+	jl .Lxts_cts1\@
+.endif
+
+	movdqu STATE, (OUTP)
+	add $16, OUTP
+	jmp .Lxts_loop1\@
+
+.Lxts_out\@:
+	movdqu STATE, (OUTP)
+	jmp .Lxts_ret_iv\@
+
+.if \enc
+.Lxts_cts4\@:
+	movdqa STATE4, STATE
+	sub $16, OUTP
+.Lxts_cts1\@:
+.else
+.Lxts_cts1\@:
+	movdqa IV, STATE4
+	_aesni_gf128mul_x_ble
+
+	pxor IV, STATE
+	call _aesni_dec1
+	pxor IV, STATE
+.endif
+#ifndef __x86_64__
+	lea .Lcts_permute_table, T1
+#else
+	lea .Lcts_permute_table(%rip), T1
+#endif
+	add LEN, INP		/* rewind input pointer */
+	add $16, LEN		/* # bytes in final block */
+	movups (INP), IN1
+
+	mov T1, IVP
+	add $32, IVP
+	add LEN, T1
+	sub LEN, IVP
+	add OUTP, LEN
+
+	movups (T1), %xmm4
+	movaps STATE, IN2
+	pshufb %xmm4, STATE
+	movups STATE, (LEN)
+
+	movups (IVP), %xmm0
+	pshufb %xmm0, IN1
+	pblendvb IN2, IN1
+	movaps IN1, STATE
+
+.if \enc
+	pxor IV, STATE
+	call _aesni_enc1
+	pxor IV, STATE
+.else
+	pxor STATE4, STATE
+	call _aesni_dec1
+	pxor STATE4, STATE
+.endif
+
+	movups STATE, (OUTP)
+	jmp .Lxts_ret\@
+.endm
+
+/*
+ * void aesni_xts_enc(const struct crypto_aes_ctx *ctx, u8 *dst,
+ *		      const u8 *src, unsigned int len, le128 *iv)
+ */
+SYM_FUNC_START(aesni_xts_enc)
+	_aesni_xts_crypt	1
+SYM_FUNC_END(aesni_xts_enc)
+
+/*
+ * void aesni_xts_dec(const struct crypto_aes_ctx *ctx, u8 *dst,
+ *		      const u8 *src, unsigned int len, le128 *iv)
+ */
+SYM_FUNC_START(aesni_xts_dec)
+	_aesni_xts_crypt	0
+SYM_FUNC_END(aesni_xts_dec)
diff --git a/arch/x86/crypto/aesni-intel_glue.c b/arch/x86/crypto/aesni-intel_glue.c
new file mode 100644
index 000000000..f522fff92
--- /dev/null
+++ b/arch/x86/crypto/aesni-intel_glue.c
@@ -0,0 +1,1673 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Support for AES-NI and VAES instructions.  This file contains glue code.
+ * The real AES implementations are in aesni-intel_asm.S and other .S files.
+ *
+ * Copyright (C) 2008, Intel Corp.
+ *    Author: Huang Ying <ying.huang@intel.com>
+ *
+ * Added RFC4106 AES-GCM support for 128-bit keys under the AEAD
+ * interface for 64-bit kernels.
+ *    Authors: Adrian Hoban <adrian.hoban@intel.com>
+ *             Gabriele Paoloni <gabriele.paoloni@intel.com>
+ *             Tadeusz Struk (tadeusz.struk@intel.com)
+ *             Aidan O'Mahony (aidan.o.mahony@intel.com)
+ *    Copyright (c) 2010, Intel Corporation.
+ *
+ * Copyright 2024 Google LLC
+ */
+
+#include <linux/hardirq.h>
+#include <linux/types.h>
+#include <linux/module.h>
+#include <linux/err.h>
+#include <crypto/algapi.h>
+#include <crypto/aes.h>
+#include <crypto/b128ops.h>
+#include <crypto/gcm.h>
+#include <crypto/gf128mul.h>
+#include <crypto/xts.h>
+#include <asm/cpu_device_id.h>
+#include <asm/simd.h>
+#include <crypto/scatterwalk.h>
+#include <crypto/internal/aead.h>
+#include <crypto/internal/simd.h>
+#include <crypto/internal/skcipher.h>
+#include <linux/jump_label.h>
+#include <linux/workqueue.h>
+#include <linux/spinlock.h>
+#include <linux/static_call.h>
+
+
+#define AESNI_ALIGN	16
+#define AESNI_ALIGN_ATTR __attribute__ ((__aligned__(AESNI_ALIGN)))
+#define AES_BLOCK_MASK	(~(AES_BLOCK_SIZE - 1))
+#define AESNI_ALIGN_EXTRA ((AESNI_ALIGN - 1) & ~(CRYPTO_MINALIGN - 1))
+#define CRYPTO_AES_CTX_SIZE (sizeof(struct crypto_aes_ctx) + AESNI_ALIGN_EXTRA)
+#define XTS_AES_CTX_SIZE (sizeof(struct aesni_xts_ctx) + AESNI_ALIGN_EXTRA)
+
+struct aesni_xts_ctx {
+	struct crypto_aes_ctx tweak_ctx AESNI_ALIGN_ATTR;
+	struct crypto_aes_ctx crypt_ctx AESNI_ALIGN_ATTR;
+};
+
+static inline void *aes_align_addr(void *addr)
+{
+	if (crypto_tfm_ctx_alignment() >= AESNI_ALIGN)
+		return addr;
+	return PTR_ALIGN(addr, AESNI_ALIGN);
+}
+
+asmlinkage void aesni_set_key(struct crypto_aes_ctx *ctx, const u8 *in_key,
+			      unsigned int key_len);
+asmlinkage void aesni_enc(const void *ctx, u8 *out, const u8 *in);
+asmlinkage void aesni_ecb_enc(struct crypto_aes_ctx *ctx, u8 *out,
+			      const u8 *in, unsigned int len);
+asmlinkage void aesni_ecb_dec(struct crypto_aes_ctx *ctx, u8 *out,
+			      const u8 *in, unsigned int len);
+asmlinkage void aesni_cbc_enc(struct crypto_aes_ctx *ctx, u8 *out,
+			      const u8 *in, unsigned int len, u8 *iv);
+asmlinkage void aesni_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out,
+			      const u8 *in, unsigned int len, u8 *iv);
+asmlinkage void aesni_cts_cbc_enc(struct crypto_aes_ctx *ctx, u8 *out,
+				  const u8 *in, unsigned int len, u8 *iv);
+asmlinkage void aesni_cts_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out,
+				  const u8 *in, unsigned int len, u8 *iv);
+
+asmlinkage void aesni_xts_enc(const struct crypto_aes_ctx *ctx, u8 *out,
+			      const u8 *in, unsigned int len, u8 *iv);
+
+asmlinkage void aesni_xts_dec(const struct crypto_aes_ctx *ctx, u8 *out,
+			      const u8 *in, unsigned int len, u8 *iv);
+
+#ifdef CONFIG_X86_64
+asmlinkage void aesni_ctr_enc(struct crypto_aes_ctx *ctx, u8 *out,
+			      const u8 *in, unsigned int len, u8 *iv);
+#endif
+
+static inline struct crypto_aes_ctx *aes_ctx(void *raw_ctx)
+{
+	return aes_align_addr(raw_ctx);
+}
+
+static inline struct aesni_xts_ctx *aes_xts_ctx(struct crypto_skcipher *tfm)
+{
+	return aes_align_addr(crypto_skcipher_ctx(tfm));
+}
+
+static int aes_set_key_common(struct crypto_aes_ctx *ctx,
+			      const u8 *in_key, unsigned int key_len)
+{
+	int err;
+
+	if (!crypto_simd_usable())
+		return aes_expandkey(ctx, in_key, key_len);
+
+	err = aes_check_keylen(key_len);
+	if (err)
+		return err;
+
+	kernel_fpu_begin();
+	aesni_set_key(ctx, in_key, key_len);
+	kernel_fpu_end();
+	return 0;
+}
+
+static int aesni_skcipher_setkey(struct crypto_skcipher *tfm, const u8 *key,
+			         unsigned int len)
+{
+	return aes_set_key_common(aes_ctx(crypto_skcipher_ctx(tfm)), key, len);
+}
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes)) {
+		kernel_fpu_begin();
+		aesni_ecb_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
+			      nbytes & AES_BLOCK_MASK);
+		kernel_fpu_end();
+		nbytes &= AES_BLOCK_SIZE - 1;
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	return err;
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes)) {
+		kernel_fpu_begin();
+		aesni_ecb_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
+			      nbytes & AES_BLOCK_MASK);
+		kernel_fpu_end();
+		nbytes &= AES_BLOCK_SIZE - 1;
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	return err;
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes)) {
+		kernel_fpu_begin();
+		aesni_cbc_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
+			      nbytes & AES_BLOCK_MASK, walk.iv);
+		kernel_fpu_end();
+		nbytes &= AES_BLOCK_SIZE - 1;
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	return err;
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes)) {
+		kernel_fpu_begin();
+		aesni_cbc_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
+			      nbytes & AES_BLOCK_MASK, walk.iv);
+		kernel_fpu_end();
+		nbytes &= AES_BLOCK_SIZE - 1;
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	return err;
+}
+
+static int cts_cbc_encrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
+	int cbc_blocks = DIV_ROUND_UP(req->cryptlen, AES_BLOCK_SIZE) - 2;
+	struct scatterlist *src = req->src, *dst = req->dst;
+	struct scatterlist sg_src[2], sg_dst[2];
+	struct skcipher_request subreq;
+	struct skcipher_walk walk;
+	int err;
+
+	skcipher_request_set_tfm(&subreq, tfm);
+	skcipher_request_set_callback(&subreq, skcipher_request_flags(req),
+				      NULL, NULL);
+
+	if (req->cryptlen <= AES_BLOCK_SIZE) {
+		if (req->cryptlen < AES_BLOCK_SIZE)
+			return -EINVAL;
+		cbc_blocks = 1;
+	}
+
+	if (cbc_blocks > 0) {
+		skcipher_request_set_crypt(&subreq, req->src, req->dst,
+					   cbc_blocks * AES_BLOCK_SIZE,
+					   req->iv);
+
+		err = cbc_encrypt(&subreq);
+		if (err)
+			return err;
+
+		if (req->cryptlen == AES_BLOCK_SIZE)
+			return 0;
+
+		dst = src = scatterwalk_ffwd(sg_src, req->src, subreq.cryptlen);
+		if (req->dst != req->src)
+			dst = scatterwalk_ffwd(sg_dst, req->dst,
+					       subreq.cryptlen);
+	}
+
+	/* handle ciphertext stealing */
+	skcipher_request_set_crypt(&subreq, src, dst,
+				   req->cryptlen - cbc_blocks * AES_BLOCK_SIZE,
+				   req->iv);
+
+	err = skcipher_walk_virt(&walk, &subreq, false);
+	if (err)
+		return err;
+
+	kernel_fpu_begin();
+	aesni_cts_cbc_enc(ctx, walk.dst.virt.addr, walk.src.virt.addr,
+			  walk.nbytes, walk.iv);
+	kernel_fpu_end();
+
+	return skcipher_walk_done(&walk, 0);
+}
+
+static int cts_cbc_decrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
+	int cbc_blocks = DIV_ROUND_UP(req->cryptlen, AES_BLOCK_SIZE) - 2;
+	struct scatterlist *src = req->src, *dst = req->dst;
+	struct scatterlist sg_src[2], sg_dst[2];
+	struct skcipher_request subreq;
+	struct skcipher_walk walk;
+	int err;
+
+	skcipher_request_set_tfm(&subreq, tfm);
+	skcipher_request_set_callback(&subreq, skcipher_request_flags(req),
+				      NULL, NULL);
+
+	if (req->cryptlen <= AES_BLOCK_SIZE) {
+		if (req->cryptlen < AES_BLOCK_SIZE)
+			return -EINVAL;
+		cbc_blocks = 1;
+	}
+
+	if (cbc_blocks > 0) {
+		skcipher_request_set_crypt(&subreq, req->src, req->dst,
+					   cbc_blocks * AES_BLOCK_SIZE,
+					   req->iv);
+
+		err = cbc_decrypt(&subreq);
+		if (err)
+			return err;
+
+		if (req->cryptlen == AES_BLOCK_SIZE)
+			return 0;
+
+		dst = src = scatterwalk_ffwd(sg_src, req->src, subreq.cryptlen);
+		if (req->dst != req->src)
+			dst = scatterwalk_ffwd(sg_dst, req->dst,
+					       subreq.cryptlen);
+	}
+
+	/* handle ciphertext stealing */
+	skcipher_request_set_crypt(&subreq, src, dst,
+				   req->cryptlen - cbc_blocks * AES_BLOCK_SIZE,
+				   req->iv);
+
+	err = skcipher_walk_virt(&walk, &subreq, false);
+	if (err)
+		return err;
+
+	kernel_fpu_begin();
+	aesni_cts_cbc_dec(ctx, walk.dst.virt.addr, walk.src.virt.addr,
+			  walk.nbytes, walk.iv);
+	kernel_fpu_end();
+
+	return skcipher_walk_done(&walk, 0);
+}
+
+#ifdef CONFIG_X86_64
+/* This is the non-AVX version. */
+static int ctr_crypt_aesni(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct crypto_aes_ctx *ctx = aes_ctx(crypto_skcipher_ctx(tfm));
+	u8 keystream[AES_BLOCK_SIZE];
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes) > 0) {
+		kernel_fpu_begin();
+		if (nbytes & AES_BLOCK_MASK)
+			aesni_ctr_enc(ctx, walk.dst.virt.addr,
+				      walk.src.virt.addr,
+				      nbytes & AES_BLOCK_MASK, walk.iv);
+		nbytes &= ~AES_BLOCK_MASK;
+
+		if (walk.nbytes == walk.total && nbytes > 0) {
+			aesni_enc(ctx, keystream, walk.iv);
+			crypto_xor_cpy(walk.dst.virt.addr + walk.nbytes - nbytes,
+				       walk.src.virt.addr + walk.nbytes - nbytes,
+				       keystream, nbytes);
+			crypto_inc(walk.iv, AES_BLOCK_SIZE);
+			nbytes = 0;
+		}
+		kernel_fpu_end();
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+	return err;
+}
+#endif
+
+static int xts_setkey_aesni(struct crypto_skcipher *tfm, const u8 *key,
+			    unsigned int keylen)
+{
+	struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
+	int err;
+
+	err = xts_verify_key(tfm, key, keylen);
+	if (err)
+		return err;
+
+	keylen /= 2;
+
+	/* first half of xts-key is for crypt */
+	err = aes_set_key_common(&ctx->crypt_ctx, key, keylen);
+	if (err)
+		return err;
+
+	/* second half of xts-key is for tweak */
+	return aes_set_key_common(&ctx->tweak_ctx, key + keylen, keylen);
+}
+
+typedef void (*xts_encrypt_iv_func)(const struct crypto_aes_ctx *tweak_key,
+				    u8 iv[AES_BLOCK_SIZE]);
+typedef void (*xts_crypt_func)(const struct crypto_aes_ctx *key,
+			       const u8 *src, u8 *dst, int len,
+			       u8 tweak[AES_BLOCK_SIZE]);
+
+/* This handles cases where the source and/or destination span pages. */
+static noinline int
+xts_crypt_slowpath(struct skcipher_request *req, xts_crypt_func crypt_func)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	const struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
+	int tail = req->cryptlen % AES_BLOCK_SIZE;
+	struct scatterlist sg_src[2], sg_dst[2];
+	struct skcipher_request subreq;
+	struct skcipher_walk walk;
+	struct scatterlist *src, *dst;
+	int err;
+
+	/*
+	 * If the message length isn't divisible by the AES block size, then
+	 * separate off the last full block and the partial block.  This ensures
+	 * that they are processed in the same call to the assembly function,
+	 * which is required for ciphertext stealing.
+	 */
+	if (tail) {
+		skcipher_request_set_tfm(&subreq, tfm);
+		skcipher_request_set_callback(&subreq,
+					      skcipher_request_flags(req),
+					      NULL, NULL);
+		skcipher_request_set_crypt(&subreq, req->src, req->dst,
+					   req->cryptlen - tail - AES_BLOCK_SIZE,
+					   req->iv);
+		req = &subreq;
+	}
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while (walk.nbytes) {
+		kernel_fpu_begin();
+		(*crypt_func)(&ctx->crypt_ctx,
+			      walk.src.virt.addr, walk.dst.virt.addr,
+			      walk.nbytes & ~(AES_BLOCK_SIZE - 1), req->iv);
+		kernel_fpu_end();
+		err = skcipher_walk_done(&walk,
+					 walk.nbytes & (AES_BLOCK_SIZE - 1));
+	}
+
+	if (err || !tail)
+		return err;
+
+	/* Do ciphertext stealing with the last full block and partial block. */
+
+	dst = src = scatterwalk_ffwd(sg_src, req->src, req->cryptlen);
+	if (req->dst != req->src)
+		dst = scatterwalk_ffwd(sg_dst, req->dst, req->cryptlen);
+
+	skcipher_request_set_crypt(req, src, dst, AES_BLOCK_SIZE + tail,
+				   req->iv);
+
+	err = skcipher_walk_virt(&walk, req, false);
+	if (err)
+		return err;
+
+	kernel_fpu_begin();
+	(*crypt_func)(&ctx->crypt_ctx, walk.src.virt.addr, walk.dst.virt.addr,
+		      walk.nbytes, req->iv);
+	kernel_fpu_end();
+
+	return skcipher_walk_done(&walk, 0);
+}
+
+/* __always_inline to avoid indirect call in fastpath */
+static __always_inline int
+xts_crypt(struct skcipher_request *req, xts_encrypt_iv_func encrypt_iv,
+	  xts_crypt_func crypt_func)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	const struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
+
+	if (unlikely(req->cryptlen < AES_BLOCK_SIZE))
+		return -EINVAL;
+
+	kernel_fpu_begin();
+	(*encrypt_iv)(&ctx->tweak_ctx, req->iv);
+
+	/*
+	 * In practice, virtually all XTS plaintexts and ciphertexts are either
+	 * 512 or 4096 bytes and do not use multiple scatterlist elements.  To
+	 * optimize the performance of these cases, the below fast-path handles
+	 * single-scatterlist-element messages as efficiently as possible.  The
+	 * code is 64-bit specific, as it assumes no page mapping is needed.
+	 */
+	if (IS_ENABLED(CONFIG_X86_64) &&
+	    likely(req->src->length >= req->cryptlen &&
+		   req->dst->length >= req->cryptlen)) {
+		(*crypt_func)(&ctx->crypt_ctx, sg_virt(req->src),
+			      sg_virt(req->dst), req->cryptlen, req->iv);
+		kernel_fpu_end();
+		return 0;
+	}
+	kernel_fpu_end();
+	return xts_crypt_slowpath(req, crypt_func);
+}
+
+static void aesni_xts_encrypt_iv(const struct crypto_aes_ctx *tweak_key,
+				 u8 iv[AES_BLOCK_SIZE])
+{
+	aesni_enc(tweak_key, iv, iv);
+}
+
+static void aesni_xts_encrypt(const struct crypto_aes_ctx *key,
+			      const u8 *src, u8 *dst, int len,
+			      u8 tweak[AES_BLOCK_SIZE])
+{
+	aesni_xts_enc(key, dst, src, len, tweak);
+}
+
+static void aesni_xts_decrypt(const struct crypto_aes_ctx *key,
+			      const u8 *src, u8 *dst, int len,
+			      u8 tweak[AES_BLOCK_SIZE])
+{
+	aesni_xts_dec(key, dst, src, len, tweak);
+}
+
+static int xts_encrypt_aesni(struct skcipher_request *req)
+{
+	return xts_crypt(req, aesni_xts_encrypt_iv, aesni_xts_encrypt);
+}
+
+static int xts_decrypt_aesni(struct skcipher_request *req)
+{
+	return xts_crypt(req, aesni_xts_encrypt_iv, aesni_xts_decrypt);
+}
+
+static struct skcipher_alg aesni_skciphers[] = {
+	{
+		.base = {
+			.cra_name		= "ecb(aes)",
+			.cra_driver_name	= "ecb-aes-aesni",
+			.cra_priority		= 400,
+			.cra_blocksize		= AES_BLOCK_SIZE,
+			.cra_ctxsize		= CRYPTO_AES_CTX_SIZE,
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= AES_MIN_KEY_SIZE,
+		.max_keysize	= AES_MAX_KEY_SIZE,
+		.setkey		= aesni_skcipher_setkey,
+		.encrypt	= ecb_encrypt,
+		.decrypt	= ecb_decrypt,
+	}, {
+		.base = {
+			.cra_name		= "cbc(aes)",
+			.cra_driver_name	= "cbc-aes-aesni",
+			.cra_priority		= 400,
+			.cra_blocksize		= AES_BLOCK_SIZE,
+			.cra_ctxsize		= CRYPTO_AES_CTX_SIZE,
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= AES_MIN_KEY_SIZE,
+		.max_keysize	= AES_MAX_KEY_SIZE,
+		.ivsize		= AES_BLOCK_SIZE,
+		.setkey		= aesni_skcipher_setkey,
+		.encrypt	= cbc_encrypt,
+		.decrypt	= cbc_decrypt,
+	}, {
+		.base = {
+			.cra_name		= "cts(cbc(aes))",
+			.cra_driver_name	= "cts-cbc-aes-aesni",
+			.cra_priority		= 400,
+			.cra_blocksize		= AES_BLOCK_SIZE,
+			.cra_ctxsize		= CRYPTO_AES_CTX_SIZE,
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= AES_MIN_KEY_SIZE,
+		.max_keysize	= AES_MAX_KEY_SIZE,
+		.ivsize		= AES_BLOCK_SIZE,
+		.walksize	= 2 * AES_BLOCK_SIZE,
+		.setkey		= aesni_skcipher_setkey,
+		.encrypt	= cts_cbc_encrypt,
+		.decrypt	= cts_cbc_decrypt,
+#ifdef CONFIG_X86_64
+	}, {
+		.base = {
+			.cra_name		= "ctr(aes)",
+			.cra_driver_name	= "ctr-aes-aesni",
+			.cra_priority		= 400,
+			.cra_blocksize		= 1,
+			.cra_ctxsize		= CRYPTO_AES_CTX_SIZE,
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= AES_MIN_KEY_SIZE,
+		.max_keysize	= AES_MAX_KEY_SIZE,
+		.ivsize		= AES_BLOCK_SIZE,
+		.chunksize	= AES_BLOCK_SIZE,
+		.setkey		= aesni_skcipher_setkey,
+		.encrypt	= ctr_crypt_aesni,
+		.decrypt	= ctr_crypt_aesni,
+#endif
+	}, {
+		.base = {
+			.cra_name		= "xts(aes)",
+			.cra_driver_name	= "xts-aes-aesni",
+			.cra_priority		= 401,
+			.cra_blocksize		= AES_BLOCK_SIZE,
+			.cra_ctxsize		= XTS_AES_CTX_SIZE,
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= 2 * AES_MIN_KEY_SIZE,
+		.max_keysize	= 2 * AES_MAX_KEY_SIZE,
+		.ivsize		= AES_BLOCK_SIZE,
+		.walksize	= 2 * AES_BLOCK_SIZE,
+		.setkey		= xts_setkey_aesni,
+		.encrypt	= xts_encrypt_aesni,
+		.decrypt	= xts_decrypt_aesni,
+	}
+};
+
+#ifdef CONFIG_X86_64
+asmlinkage void aes_xts_encrypt_iv(const struct crypto_aes_ctx *tweak_key,
+				   u8 iv[AES_BLOCK_SIZE]);
+
+/* __always_inline to avoid indirect call */
+static __always_inline int
+ctr_crypt(struct skcipher_request *req,
+	  void (*ctr64_func)(const struct crypto_aes_ctx *key,
+			     const u8 *src, u8 *dst, int len,
+			     const u64 le_ctr[2]))
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	const struct crypto_aes_ctx *key = aes_ctx(crypto_skcipher_ctx(tfm));
+	unsigned int nbytes, p1_nbytes, nblocks;
+	struct skcipher_walk walk;
+	u64 le_ctr[2];
+	u64 ctr64;
+	int err;
+
+	ctr64 = le_ctr[0] = get_unaligned_be64(&req->iv[8]);
+	le_ctr[1] = get_unaligned_be64(&req->iv[0]);
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes) != 0) {
+		if (nbytes < walk.total) {
+			/* Not the end yet, so keep the length block-aligned. */
+			nbytes = round_down(nbytes, AES_BLOCK_SIZE);
+			nblocks = nbytes / AES_BLOCK_SIZE;
+		} else {
+			/* It's the end, so include any final partial block. */
+			nblocks = DIV_ROUND_UP(nbytes, AES_BLOCK_SIZE);
+		}
+		ctr64 += nblocks;
+
+		kernel_fpu_begin();
+		if (likely(ctr64 >= nblocks)) {
+			/* The low 64 bits of the counter won't overflow. */
+			(*ctr64_func)(key, walk.src.virt.addr,
+				      walk.dst.virt.addr, nbytes, le_ctr);
+		} else {
+			/*
+			 * The low 64 bits of the counter will overflow.  The
+			 * assembly doesn't handle this case, so split the
+			 * operation into two at the point where the overflow
+			 * will occur.  After the first part, add the carry bit.
+			 */
+			p1_nbytes = min(nbytes, (nblocks - ctr64) * AES_BLOCK_SIZE);
+			(*ctr64_func)(key, walk.src.virt.addr,
+				      walk.dst.virt.addr, p1_nbytes, le_ctr);
+			le_ctr[0] = 0;
+			le_ctr[1]++;
+			(*ctr64_func)(key, walk.src.virt.addr + p1_nbytes,
+				      walk.dst.virt.addr + p1_nbytes,
+				      nbytes - p1_nbytes, le_ctr);
+		}
+		kernel_fpu_end();
+		le_ctr[0] = ctr64;
+
+		err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
+	}
+
+	put_unaligned_be64(ctr64, &req->iv[8]);
+	put_unaligned_be64(le_ctr[1], &req->iv[0]);
+
+	return err;
+}
+
+/* __always_inline to avoid indirect call */
+static __always_inline int
+xctr_crypt(struct skcipher_request *req,
+	   void (*xctr_func)(const struct crypto_aes_ctx *key,
+			     const u8 *src, u8 *dst, int len,
+			     const u8 iv[AES_BLOCK_SIZE], u64 ctr))
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	const struct crypto_aes_ctx *key = aes_ctx(crypto_skcipher_ctx(tfm));
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	u64 ctr = 1;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+	while ((nbytes = walk.nbytes) != 0) {
+		if (nbytes < walk.total)
+			nbytes = round_down(nbytes, AES_BLOCK_SIZE);
+
+		kernel_fpu_begin();
+		(*xctr_func)(key, walk.src.virt.addr, walk.dst.virt.addr,
+			     nbytes, req->iv, ctr);
+		kernel_fpu_end();
+
+		ctr += DIV_ROUND_UP(nbytes, AES_BLOCK_SIZE);
+		err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
+	}
+	return err;
+}
+
+#define DEFINE_AVX_SKCIPHER_ALGS(suffix, driver_name_suffix, priority)	       \
+									       \
+asmlinkage void								       \
+aes_xts_encrypt_##suffix(const struct crypto_aes_ctx *key, const u8 *src,      \
+			 u8 *dst, int len, u8 tweak[AES_BLOCK_SIZE]);	       \
+asmlinkage void								       \
+aes_xts_decrypt_##suffix(const struct crypto_aes_ctx *key, const u8 *src,      \
+			 u8 *dst, int len, u8 tweak[AES_BLOCK_SIZE]);	       \
+									       \
+static int xts_encrypt_##suffix(struct skcipher_request *req)		       \
+{									       \
+	return xts_crypt(req, aes_xts_encrypt_iv, aes_xts_encrypt_##suffix);   \
+}									       \
+									       \
+static int xts_decrypt_##suffix(struct skcipher_request *req)		       \
+{									       \
+	return xts_crypt(req, aes_xts_encrypt_iv, aes_xts_decrypt_##suffix);   \
+}									       \
+									       \
+asmlinkage void								       \
+aes_ctr64_crypt_##suffix(const struct crypto_aes_ctx *key,		       \
+			 const u8 *src, u8 *dst, int len, const u64 le_ctr[2]);\
+									       \
+static int ctr_crypt_##suffix(struct skcipher_request *req)		       \
+{									       \
+	return ctr_crypt(req, aes_ctr64_crypt_##suffix);		       \
+}									       \
+									       \
+asmlinkage void								       \
+aes_xctr_crypt_##suffix(const struct crypto_aes_ctx *key,		       \
+			const u8 *src, u8 *dst, int len,		       \
+			const u8 iv[AES_BLOCK_SIZE], u64 ctr);		       \
+									       \
+static int xctr_crypt_##suffix(struct skcipher_request *req)		       \
+{									       \
+	return xctr_crypt(req, aes_xctr_crypt_##suffix);		       \
+}									       \
+									       \
+static struct skcipher_alg skcipher_algs_##suffix[] = {{		       \
+	.base.cra_name		= "xts(aes)",				       \
+	.base.cra_driver_name	= "xts-aes-" driver_name_suffix,	       \
+	.base.cra_priority	= priority,				       \
+	.base.cra_blocksize	= AES_BLOCK_SIZE,			       \
+	.base.cra_ctxsize	= XTS_AES_CTX_SIZE,			       \
+	.base.cra_module	= THIS_MODULE,				       \
+	.min_keysize		= 2 * AES_MIN_KEY_SIZE,			       \
+	.max_keysize		= 2 * AES_MAX_KEY_SIZE,			       \
+	.ivsize			= AES_BLOCK_SIZE,			       \
+	.walksize		= 2 * AES_BLOCK_SIZE,			       \
+	.setkey			= xts_setkey_aesni,			       \
+	.encrypt		= xts_encrypt_##suffix,			       \
+	.decrypt		= xts_decrypt_##suffix,			       \
+}, {									       \
+	.base.cra_name		= "ctr(aes)",				       \
+	.base.cra_driver_name	= "ctr-aes-" driver_name_suffix,	       \
+	.base.cra_priority	= priority,				       \
+	.base.cra_blocksize	= 1,					       \
+	.base.cra_ctxsize	= CRYPTO_AES_CTX_SIZE,			       \
+	.base.cra_module	= THIS_MODULE,				       \
+	.min_keysize		= AES_MIN_KEY_SIZE,			       \
+	.max_keysize		= AES_MAX_KEY_SIZE,			       \
+	.ivsize			= AES_BLOCK_SIZE,			       \
+	.chunksize		= AES_BLOCK_SIZE,			       \
+	.setkey			= aesni_skcipher_setkey,		       \
+	.encrypt		= ctr_crypt_##suffix,			       \
+	.decrypt		= ctr_crypt_##suffix,			       \
+}, {									       \
+	.base.cra_name		= "xctr(aes)",				       \
+	.base.cra_driver_name	= "xctr-aes-" driver_name_suffix,	       \
+	.base.cra_priority	= priority,				       \
+	.base.cra_blocksize	= 1,					       \
+	.base.cra_ctxsize	= CRYPTO_AES_CTX_SIZE,			       \
+	.base.cra_module	= THIS_MODULE,				       \
+	.min_keysize		= AES_MIN_KEY_SIZE,			       \
+	.max_keysize		= AES_MAX_KEY_SIZE,			       \
+	.ivsize			= AES_BLOCK_SIZE,			       \
+	.chunksize		= AES_BLOCK_SIZE,			       \
+	.setkey			= aesni_skcipher_setkey,		       \
+	.encrypt		= xctr_crypt_##suffix,			       \
+	.decrypt		= xctr_crypt_##suffix,			       \
+}}
+
+DEFINE_AVX_SKCIPHER_ALGS(aesni_avx, "aesni-avx", 500);
+DEFINE_AVX_SKCIPHER_ALGS(vaes_avx2, "vaes-avx2", 600);
+DEFINE_AVX_SKCIPHER_ALGS(vaes_avx512, "vaes-avx512", 800);
+
+/* The common part of the x86_64 AES-GCM key struct */
+struct aes_gcm_key {
+	/* Expanded AES key and the AES key length in bytes */
+	struct aes_enckey aes_key;
+
+	/* RFC4106 nonce (used only by the rfc4106 algorithms) */
+	u32 rfc4106_nonce;
+};
+
+/* Key struct used by the AES-NI implementations of AES-GCM */
+struct aes_gcm_key_aesni {
+	/*
+	 * Common part of the key.  16-byte alignment is required by the
+	 * assembly code.
+	 */
+	struct aes_gcm_key base __aligned(16);
+
+	/*
+	 * Powers of the hash key H^8 through H^1.  These are 128-bit values.
+	 * They all have an extra factor of x^-1 and are byte-reversed.  16-byte
+	 * alignment is required by the assembly code.
+	 */
+	u64 h_powers[8][2] __aligned(16);
+
+	/*
+	 * h_powers_xored[i] contains the two 64-bit halves of h_powers[i] XOR'd
+	 * together.  It's used for Karatsuba multiplication.  16-byte alignment
+	 * is required by the assembly code.
+	 */
+	u64 h_powers_xored[8] __aligned(16);
+
+	/*
+	 * H^1 times x^64 (and also the usual extra factor of x^-1).  16-byte
+	 * alignment is required by the assembly code.
+	 */
+	u64 h_times_x64[2] __aligned(16);
+};
+#define AES_GCM_KEY_AESNI(key)	\
+	container_of((key), struct aes_gcm_key_aesni, base)
+#define AES_GCM_KEY_AESNI_SIZE	\
+	(sizeof(struct aes_gcm_key_aesni) + (15 & ~(CRYPTO_MINALIGN - 1)))
+
+/* Key struct used by the VAES + AVX2 implementation of AES-GCM */
+struct aes_gcm_key_vaes_avx2 {
+	/*
+	 * Common part of the key.  The assembly code prefers 16-byte alignment
+	 * for this.
+	 */
+	struct aes_gcm_key base __aligned(16);
+
+	/*
+	 * Powers of the hash key H^8 through H^1.  These are 128-bit values.
+	 * They all have an extra factor of x^-1 and are byte-reversed.
+	 * The assembly code prefers 32-byte alignment for this.
+	 */
+	u64 h_powers[8][2] __aligned(32);
+
+	/*
+	 * Each entry in this array contains the two halves of an entry of
+	 * h_powers XOR'd together, in the following order:
+	 * H^8,H^6,H^7,H^5,H^4,H^2,H^3,H^1 i.e. indices 0,2,1,3,4,6,5,7.
+	 * This is used for Karatsuba multiplication.
+	 */
+	u64 h_powers_xored[8];
+};
+
+#define AES_GCM_KEY_VAES_AVX2(key) \
+	container_of((key), struct aes_gcm_key_vaes_avx2, base)
+#define AES_GCM_KEY_VAES_AVX2_SIZE \
+	(sizeof(struct aes_gcm_key_vaes_avx2) + (31 & ~(CRYPTO_MINALIGN - 1)))
+
+/* Key struct used by the VAES + AVX512 implementation of AES-GCM */
+struct aes_gcm_key_vaes_avx512 {
+	/*
+	 * Common part of the key.  The assembly code prefers 16-byte alignment
+	 * for this.
+	 */
+	struct aes_gcm_key base __aligned(16);
+
+	/*
+	 * Powers of the hash key H^16 through H^1.  These are 128-bit values.
+	 * They all have an extra factor of x^-1 and are byte-reversed.  This
+	 * array is aligned to a 64-byte boundary to make it naturally aligned
+	 * for 512-bit loads, which can improve performance.  (The assembly code
+	 * doesn't *need* the alignment; this is just an optimization.)
+	 */
+	u64 h_powers[16][2] __aligned(64);
+
+	/* Three padding blocks required by the assembly code */
+	u64 padding[3][2];
+};
+#define AES_GCM_KEY_VAES_AVX512(key) \
+	container_of((key), struct aes_gcm_key_vaes_avx512, base)
+#define AES_GCM_KEY_VAES_AVX512_SIZE \
+	(sizeof(struct aes_gcm_key_vaes_avx512) + (63 & ~(CRYPTO_MINALIGN - 1)))
+
+/*
+ * These flags are passed to the AES-GCM helper functions to specify the
+ * specific version of AES-GCM (RFC4106 or not), whether it's encryption or
+ * decryption, and which assembly functions should be called.  Assembly
+ * functions are selected using flags instead of function pointers to avoid
+ * indirect calls (which are very expensive on x86) regardless of inlining.
+ */
+#define FLAG_RFC4106	BIT(0)
+#define FLAG_ENC	BIT(1)
+#define FLAG_AVX	BIT(2)
+#define FLAG_VAES_AVX2	BIT(3)
+#define FLAG_VAES_AVX512 BIT(4)
+
+static inline struct aes_gcm_key *
+aes_gcm_key_get(struct crypto_aead *tfm, int flags)
+{
+	if (flags & FLAG_VAES_AVX512)
+		return PTR_ALIGN(crypto_aead_ctx(tfm), 64);
+	else if (flags & FLAG_VAES_AVX2)
+		return PTR_ALIGN(crypto_aead_ctx(tfm), 32);
+	else
+		return PTR_ALIGN(crypto_aead_ctx(tfm), 16);
+}
+
+asmlinkage void
+aes_gcm_precompute_aesni(struct aes_gcm_key_aesni *key);
+asmlinkage void
+aes_gcm_precompute_aesni_avx(struct aes_gcm_key_aesni *key);
+asmlinkage void
+aes_gcm_precompute_vaes_avx2(struct aes_gcm_key_vaes_avx2 *key);
+asmlinkage void
+aes_gcm_precompute_vaes_avx512(struct aes_gcm_key_vaes_avx512 *key);
+
+static void aes_gcm_precompute(struct aes_gcm_key *key, int flags)
+{
+	if (flags & FLAG_VAES_AVX512)
+		aes_gcm_precompute_vaes_avx512(AES_GCM_KEY_VAES_AVX512(key));
+	else if (flags & FLAG_VAES_AVX2)
+		aes_gcm_precompute_vaes_avx2(AES_GCM_KEY_VAES_AVX2(key));
+	else if (flags & FLAG_AVX)
+		aes_gcm_precompute_aesni_avx(AES_GCM_KEY_AESNI(key));
+	else
+		aes_gcm_precompute_aesni(AES_GCM_KEY_AESNI(key));
+}
+
+asmlinkage void
+aes_gcm_aad_update_aesni(const struct aes_gcm_key_aesni *key,
+			 u8 ghash_acc[16], const u8 *aad, int aadlen);
+asmlinkage void
+aes_gcm_aad_update_aesni_avx(const struct aes_gcm_key_aesni *key,
+			     u8 ghash_acc[16], const u8 *aad, int aadlen);
+asmlinkage void
+aes_gcm_aad_update_vaes_avx2(const struct aes_gcm_key_vaes_avx2 *key,
+			     u8 ghash_acc[16], const u8 *aad, int aadlen);
+asmlinkage void
+aes_gcm_aad_update_vaes_avx512(const struct aes_gcm_key_vaes_avx512 *key,
+			       u8 ghash_acc[16], const u8 *aad, int aadlen);
+
+static void aes_gcm_aad_update(const struct aes_gcm_key *key, u8 ghash_acc[16],
+			       const u8 *aad, int aadlen, int flags)
+{
+	if (flags & FLAG_VAES_AVX512)
+		aes_gcm_aad_update_vaes_avx512(AES_GCM_KEY_VAES_AVX512(key),
+					       ghash_acc, aad, aadlen);
+	else if (flags & FLAG_VAES_AVX2)
+		aes_gcm_aad_update_vaes_avx2(AES_GCM_KEY_VAES_AVX2(key),
+					     ghash_acc, aad, aadlen);
+	else if (flags & FLAG_AVX)
+		aes_gcm_aad_update_aesni_avx(AES_GCM_KEY_AESNI(key), ghash_acc,
+					     aad, aadlen);
+	else
+		aes_gcm_aad_update_aesni(AES_GCM_KEY_AESNI(key), ghash_acc,
+					 aad, aadlen);
+}
+
+asmlinkage void
+aes_gcm_enc_update_aesni(const struct aes_gcm_key_aesni *key,
+			 const u32 le_ctr[4], u8 ghash_acc[16],
+			 const u8 *src, u8 *dst, int datalen);
+asmlinkage void
+aes_gcm_enc_update_aesni_avx(const struct aes_gcm_key_aesni *key,
+			     const u32 le_ctr[4], u8 ghash_acc[16],
+			     const u8 *src, u8 *dst, int datalen);
+asmlinkage void
+aes_gcm_enc_update_vaes_avx2(const struct aes_gcm_key_vaes_avx2 *key,
+			     const u32 le_ctr[4], u8 ghash_acc[16],
+			     const u8 *src, u8 *dst, int datalen);
+asmlinkage void
+aes_gcm_enc_update_vaes_avx512(const struct aes_gcm_key_vaes_avx512 *key,
+			       const u32 le_ctr[4], u8 ghash_acc[16],
+			       const u8 *src, u8 *dst, int datalen);
+
+asmlinkage void
+aes_gcm_dec_update_aesni(const struct aes_gcm_key_aesni *key,
+			 const u32 le_ctr[4], u8 ghash_acc[16],
+			 const u8 *src, u8 *dst, int datalen);
+asmlinkage void
+aes_gcm_dec_update_aesni_avx(const struct aes_gcm_key_aesni *key,
+			     const u32 le_ctr[4], u8 ghash_acc[16],
+			     const u8 *src, u8 *dst, int datalen);
+asmlinkage void
+aes_gcm_dec_update_vaes_avx2(const struct aes_gcm_key_vaes_avx2 *key,
+			     const u32 le_ctr[4], u8 ghash_acc[16],
+			     const u8 *src, u8 *dst, int datalen);
+asmlinkage void
+aes_gcm_dec_update_vaes_avx512(const struct aes_gcm_key_vaes_avx512 *key,
+			       const u32 le_ctr[4], u8 ghash_acc[16],
+			       const u8 *src, u8 *dst, int datalen);
+
+/* __always_inline to optimize out the branches based on @flags */
+static __always_inline void
+aes_gcm_update(const struct aes_gcm_key *key,
+	       const u32 le_ctr[4], u8 ghash_acc[16],
+	       const u8 *src, u8 *dst, int datalen, int flags)
+{
+	if (flags & FLAG_ENC) {
+		if (flags & FLAG_VAES_AVX512)
+			aes_gcm_enc_update_vaes_avx512(AES_GCM_KEY_VAES_AVX512(key),
+						       le_ctr, ghash_acc,
+						       src, dst, datalen);
+		else if (flags & FLAG_VAES_AVX2)
+			aes_gcm_enc_update_vaes_avx2(AES_GCM_KEY_VAES_AVX2(key),
+						     le_ctr, ghash_acc,
+						     src, dst, datalen);
+		else if (flags & FLAG_AVX)
+			aes_gcm_enc_update_aesni_avx(AES_GCM_KEY_AESNI(key),
+						     le_ctr, ghash_acc,
+						     src, dst, datalen);
+		else
+			aes_gcm_enc_update_aesni(AES_GCM_KEY_AESNI(key), le_ctr,
+						 ghash_acc, src, dst, datalen);
+	} else {
+		if (flags & FLAG_VAES_AVX512)
+			aes_gcm_dec_update_vaes_avx512(AES_GCM_KEY_VAES_AVX512(key),
+						       le_ctr, ghash_acc,
+						       src, dst, datalen);
+		else if (flags & FLAG_VAES_AVX2)
+			aes_gcm_dec_update_vaes_avx2(AES_GCM_KEY_VAES_AVX2(key),
+						     le_ctr, ghash_acc,
+						     src, dst, datalen);
+		else if (flags & FLAG_AVX)
+			aes_gcm_dec_update_aesni_avx(AES_GCM_KEY_AESNI(key),
+						     le_ctr, ghash_acc,
+						     src, dst, datalen);
+		else
+			aes_gcm_dec_update_aesni(AES_GCM_KEY_AESNI(key),
+						 le_ctr, ghash_acc,
+						 src, dst, datalen);
+	}
+}
+
+asmlinkage void
+aes_gcm_enc_final_aesni(const struct aes_gcm_key_aesni *key,
+			const u32 le_ctr[4], u8 ghash_acc[16],
+			u64 total_aadlen, u64 total_datalen);
+asmlinkage void
+aes_gcm_enc_final_aesni_avx(const struct aes_gcm_key_aesni *key,
+			    const u32 le_ctr[4], u8 ghash_acc[16],
+			    u64 total_aadlen, u64 total_datalen);
+asmlinkage void
+aes_gcm_enc_final_vaes_avx2(const struct aes_gcm_key_vaes_avx2 *key,
+			    const u32 le_ctr[4], u8 ghash_acc[16],
+			    u64 total_aadlen, u64 total_datalen);
+asmlinkage void
+aes_gcm_enc_final_vaes_avx512(const struct aes_gcm_key_vaes_avx512 *key,
+			      const u32 le_ctr[4], u8 ghash_acc[16],
+			      u64 total_aadlen, u64 total_datalen);
+
+/* __always_inline to optimize out the branches based on @flags */
+static __always_inline void
+aes_gcm_enc_final(const struct aes_gcm_key *key,
+		  const u32 le_ctr[4], u8 ghash_acc[16],
+		  u64 total_aadlen, u64 total_datalen, int flags)
+{
+	if (flags & FLAG_VAES_AVX512)
+		aes_gcm_enc_final_vaes_avx512(AES_GCM_KEY_VAES_AVX512(key),
+					      le_ctr, ghash_acc,
+					      total_aadlen, total_datalen);
+	else if (flags & FLAG_VAES_AVX2)
+		aes_gcm_enc_final_vaes_avx2(AES_GCM_KEY_VAES_AVX2(key),
+					    le_ctr, ghash_acc,
+					    total_aadlen, total_datalen);
+	else if (flags & FLAG_AVX)
+		aes_gcm_enc_final_aesni_avx(AES_GCM_KEY_AESNI(key),
+					    le_ctr, ghash_acc,
+					    total_aadlen, total_datalen);
+	else
+		aes_gcm_enc_final_aesni(AES_GCM_KEY_AESNI(key),
+					le_ctr, ghash_acc,
+					total_aadlen, total_datalen);
+}
+
+asmlinkage bool __must_check
+aes_gcm_dec_final_aesni(const struct aes_gcm_key_aesni *key,
+			const u32 le_ctr[4], const u8 ghash_acc[16],
+			u64 total_aadlen, u64 total_datalen,
+			const u8 tag[16], int taglen);
+asmlinkage bool __must_check
+aes_gcm_dec_final_aesni_avx(const struct aes_gcm_key_aesni *key,
+			    const u32 le_ctr[4], const u8 ghash_acc[16],
+			    u64 total_aadlen, u64 total_datalen,
+			    const u8 tag[16], int taglen);
+asmlinkage bool __must_check
+aes_gcm_dec_final_vaes_avx2(const struct aes_gcm_key_vaes_avx2 *key,
+			    const u32 le_ctr[4], const u8 ghash_acc[16],
+			    u64 total_aadlen, u64 total_datalen,
+			    const u8 tag[16], int taglen);
+asmlinkage bool __must_check
+aes_gcm_dec_final_vaes_avx512(const struct aes_gcm_key_vaes_avx512 *key,
+			      const u32 le_ctr[4], const u8 ghash_acc[16],
+			      u64 total_aadlen, u64 total_datalen,
+			      const u8 tag[16], int taglen);
+
+/* __always_inline to optimize out the branches based on @flags */
+static __always_inline bool __must_check
+aes_gcm_dec_final(const struct aes_gcm_key *key, const u32 le_ctr[4],
+		  u8 ghash_acc[16], u64 total_aadlen, u64 total_datalen,
+		  u8 tag[16], int taglen, int flags)
+{
+	if (flags & FLAG_VAES_AVX512)
+		return aes_gcm_dec_final_vaes_avx512(AES_GCM_KEY_VAES_AVX512(key),
+						     le_ctr, ghash_acc,
+						     total_aadlen, total_datalen,
+						     tag, taglen);
+	else if (flags & FLAG_VAES_AVX2)
+		return aes_gcm_dec_final_vaes_avx2(AES_GCM_KEY_VAES_AVX2(key),
+						   le_ctr, ghash_acc,
+						   total_aadlen, total_datalen,
+						   tag, taglen);
+	else if (flags & FLAG_AVX)
+		return aes_gcm_dec_final_aesni_avx(AES_GCM_KEY_AESNI(key),
+						   le_ctr, ghash_acc,
+						   total_aadlen, total_datalen,
+						   tag, taglen);
+	else
+		return aes_gcm_dec_final_aesni(AES_GCM_KEY_AESNI(key),
+					       le_ctr, ghash_acc,
+					       total_aadlen, total_datalen,
+					       tag, taglen);
+}
+
+/*
+ * This is the Integrity Check Value (aka the authentication tag) length and can
+ * be 8, 12 or 16 bytes long.
+ */
+static int common_rfc4106_set_authsize(struct crypto_aead *aead,
+				       unsigned int authsize)
+{
+	switch (authsize) {
+	case 8:
+	case 12:
+	case 16:
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int generic_gcmaes_set_authsize(struct crypto_aead *tfm,
+				       unsigned int authsize)
+{
+	switch (authsize) {
+	case 4:
+	case 8:
+	case 12:
+	case 13:
+	case 14:
+	case 15:
+	case 16:
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+/*
+ * This is the setkey function for the x86_64 implementations of AES-GCM.  It
+ * saves the RFC4106 nonce if applicable, expands the AES key, and precomputes
+ * powers of the hash key.
+ *
+ * To comply with the crypto_aead API, this has to be usable in no-SIMD context.
+ * For that reason, this function includes a portable C implementation of the
+ * needed logic.  However, the portable C implementation is very slow, taking
+ * about the same time as encrypting 37 KB of data.  To be ready for users that
+ * may set a key even somewhat frequently, we therefore also include a SIMD
+ * assembly implementation, expanding the AES key using AES-NI and precomputing
+ * the hash key powers using PCLMULQDQ or VPCLMULQDQ.
+ */
+static int gcm_setkey(struct crypto_aead *tfm, const u8 *raw_key,
+		      unsigned int keylen, int flags)
+{
+	struct aes_gcm_key *key = aes_gcm_key_get(tfm, flags);
+	int err;
+
+	if (flags & FLAG_RFC4106) {
+		if (keylen < 4)
+			return -EINVAL;
+		keylen -= 4;
+		key->rfc4106_nonce = get_unaligned_be32(raw_key + keylen);
+	}
+
+	/* The assembly code assumes the following offsets. */
+	static_assert(offsetof(struct aes_gcm_key_aesni, base.aes_key.len) == 0);
+	static_assert(offsetof(struct aes_gcm_key_aesni, base.aes_key.k.rndkeys) == 16);
+	static_assert(offsetof(struct aes_gcm_key_aesni, h_powers) == 272);
+	static_assert(offsetof(struct aes_gcm_key_aesni, h_powers_xored) == 400);
+	static_assert(offsetof(struct aes_gcm_key_aesni, h_times_x64) == 464);
+	static_assert(offsetof(struct aes_gcm_key_vaes_avx2, base.aes_key.len) == 0);
+	static_assert(offsetof(struct aes_gcm_key_vaes_avx2, base.aes_key.k.rndkeys) == 16);
+	static_assert(offsetof(struct aes_gcm_key_vaes_avx2, h_powers) == 288);
+	static_assert(offsetof(struct aes_gcm_key_vaes_avx2, h_powers_xored) == 416);
+	static_assert(offsetof(struct aes_gcm_key_vaes_avx512, base.aes_key.len) == 0);
+	static_assert(offsetof(struct aes_gcm_key_vaes_avx512, base.aes_key.k.rndkeys) == 16);
+	static_assert(offsetof(struct aes_gcm_key_vaes_avx512, h_powers) == 320);
+	static_assert(offsetof(struct aes_gcm_key_vaes_avx512, padding) == 576);
+
+	err = aes_prepareenckey(&key->aes_key, raw_key, keylen);
+	if (err)
+		return err;
+
+	if (likely(crypto_simd_usable())) {
+		kernel_fpu_begin();
+		aes_gcm_precompute(key, flags);
+		kernel_fpu_end();
+	} else {
+		static const u8 x_to_the_minus1[16] __aligned(__alignof__(be128)) = {
+			[0] = 0xc2, [15] = 1
+		};
+		static const u8 x_to_the_63[16] __aligned(__alignof__(be128)) = {
+			[7] = 1,
+		};
+		be128 h1 = {};
+		be128 h;
+		int i;
+
+		/* Encrypt the all-zeroes block to get the hash key H^1 */
+		aes_encrypt(&key->aes_key, (u8 *)&h1, (u8 *)&h1);
+
+		/* Compute H^1 * x^-1 */
+		h = h1;
+		gf128mul_lle(&h, (const be128 *)x_to_the_minus1);
+
+		/* Compute the needed key powers */
+		if (flags & FLAG_VAES_AVX512) {
+			struct aes_gcm_key_vaes_avx512 *k =
+				AES_GCM_KEY_VAES_AVX512(key);
+
+			for (i = ARRAY_SIZE(k->h_powers) - 1; i >= 0; i--) {
+				k->h_powers[i][0] = be64_to_cpu(h.b);
+				k->h_powers[i][1] = be64_to_cpu(h.a);
+				gf128mul_lle(&h, &h1);
+			}
+			memset(k->padding, 0, sizeof(k->padding));
+		} else if (flags & FLAG_VAES_AVX2) {
+			struct aes_gcm_key_vaes_avx2 *k =
+				AES_GCM_KEY_VAES_AVX2(key);
+			static const u8 indices[8] = { 0, 2, 1, 3, 4, 6, 5, 7 };
+
+			for (i = ARRAY_SIZE(k->h_powers) - 1; i >= 0; i--) {
+				k->h_powers[i][0] = be64_to_cpu(h.b);
+				k->h_powers[i][1] = be64_to_cpu(h.a);
+				gf128mul_lle(&h, &h1);
+			}
+			for (i = 0; i < ARRAY_SIZE(k->h_powers_xored); i++) {
+				int j = indices[i];
+
+				k->h_powers_xored[i] = k->h_powers[j][0] ^
+						       k->h_powers[j][1];
+			}
+		} else {
+			struct aes_gcm_key_aesni *k = AES_GCM_KEY_AESNI(key);
+
+			for (i = ARRAY_SIZE(k->h_powers) - 1; i >= 0; i--) {
+				k->h_powers[i][0] = be64_to_cpu(h.b);
+				k->h_powers[i][1] = be64_to_cpu(h.a);
+				k->h_powers_xored[i] = k->h_powers[i][0] ^
+						       k->h_powers[i][1];
+				gf128mul_lle(&h, &h1);
+			}
+			gf128mul_lle(&h1, (const be128 *)x_to_the_63);
+			k->h_times_x64[0] = be64_to_cpu(h1.b);
+			k->h_times_x64[1] = be64_to_cpu(h1.a);
+		}
+	}
+	return 0;
+}
+
+/*
+ * Initialize @ghash_acc, then pass all @assoclen bytes of associated data
+ * (a.k.a. additional authenticated data) from @sg_src through the GHASH update
+ * assembly function.  kernel_fpu_begin() must have already been called.
+ */
+static void gcm_process_assoc(const struct aes_gcm_key *key, u8 ghash_acc[16],
+			      struct scatterlist *sg_src, unsigned int assoclen,
+			      int flags)
+{
+	struct scatter_walk walk;
+	/*
+	 * The assembly function requires that the length of any non-last
+	 * segment of associated data be a multiple of 16 bytes, so this
+	 * function does the buffering needed to achieve that.
+	 */
+	unsigned int pos = 0;
+	u8 buf[16];
+
+	memset(ghash_acc, 0, 16);
+	scatterwalk_start(&walk, sg_src);
+
+	while (assoclen) {
+		unsigned int orig_len_this_step = scatterwalk_next(
+			&walk, assoclen);
+		unsigned int len_this_step = orig_len_this_step;
+		unsigned int len;
+		const u8 *src = walk.addr;
+
+		if (unlikely(pos)) {
+			len = min(len_this_step, 16 - pos);
+			memcpy(&buf[pos], src, len);
+			pos += len;
+			src += len;
+			len_this_step -= len;
+			if (pos < 16)
+				goto next;
+			aes_gcm_aad_update(key, ghash_acc, buf, 16, flags);
+			pos = 0;
+		}
+		len = len_this_step;
+		if (unlikely(assoclen)) /* Not the last segment yet? */
+			len = round_down(len, 16);
+		aes_gcm_aad_update(key, ghash_acc, src, len, flags);
+		src += len;
+		len_this_step -= len;
+		if (unlikely(len_this_step)) {
+			memcpy(buf, src, len_this_step);
+			pos = len_this_step;
+		}
+next:
+		scatterwalk_done_src(&walk, orig_len_this_step);
+		if (need_resched()) {
+			kernel_fpu_end();
+			kernel_fpu_begin();
+		}
+		assoclen -= orig_len_this_step;
+	}
+	if (unlikely(pos))
+		aes_gcm_aad_update(key, ghash_acc, buf, pos, flags);
+}
+
+
+/* __always_inline to optimize out the branches based on @flags */
+static __always_inline int
+gcm_crypt(struct aead_request *req, int flags)
+{
+	struct crypto_aead *tfm = crypto_aead_reqtfm(req);
+	const struct aes_gcm_key *key = aes_gcm_key_get(tfm, flags);
+	unsigned int assoclen = req->assoclen;
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	u8 ghash_acc[16]; /* GHASH accumulator */
+	u32 le_ctr[4]; /* Counter in little-endian format */
+	int taglen;
+	int err;
+
+	/* Initialize the counter and determine the associated data length. */
+	le_ctr[0] = 2;
+	if (flags & FLAG_RFC4106) {
+		if (unlikely(assoclen != 16 && assoclen != 20))
+			return -EINVAL;
+		assoclen -= 8;
+		le_ctr[1] = get_unaligned_be32(req->iv + 4);
+		le_ctr[2] = get_unaligned_be32(req->iv + 0);
+		le_ctr[3] = key->rfc4106_nonce; /* already byte-swapped */
+	} else {
+		le_ctr[1] = get_unaligned_be32(req->iv + 8);
+		le_ctr[2] = get_unaligned_be32(req->iv + 4);
+		le_ctr[3] = get_unaligned_be32(req->iv + 0);
+	}
+
+	/* Begin walking through the plaintext or ciphertext. */
+	if (flags & FLAG_ENC)
+		err = skcipher_walk_aead_encrypt(&walk, req, false);
+	else
+		err = skcipher_walk_aead_decrypt(&walk, req, false);
+	if (err)
+		return err;
+
+	/*
+	 * Since the AES-GCM assembly code requires that at least three assembly
+	 * functions be called to process any message (this is needed to support
+	 * incremental updates cleanly), to reduce overhead we try to do all
+	 * three calls in the same kernel FPU section if possible.  We close the
+	 * section and start a new one if there are multiple data segments or if
+	 * rescheduling is needed while processing the associated data.
+	 */
+	kernel_fpu_begin();
+
+	/* Pass the associated data through GHASH. */
+	gcm_process_assoc(key, ghash_acc, req->src, assoclen, flags);
+
+	/* En/decrypt the data and pass the ciphertext through GHASH. */
+	while (unlikely((nbytes = walk.nbytes) < walk.total)) {
+		/*
+		 * Non-last segment.  In this case, the assembly function
+		 * requires that the length be a multiple of 16 (AES_BLOCK_SIZE)
+		 * bytes.  The needed buffering of up to 16 bytes is handled by
+		 * the skcipher_walk.  Here we just need to round down to a
+		 * multiple of 16.
+		 */
+		nbytes = round_down(nbytes, AES_BLOCK_SIZE);
+		aes_gcm_update(key, le_ctr, ghash_acc, walk.src.virt.addr,
+			       walk.dst.virt.addr, nbytes, flags);
+		le_ctr[0] += nbytes / AES_BLOCK_SIZE;
+		kernel_fpu_end();
+		err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
+		if (err)
+			return err;
+		kernel_fpu_begin();
+	}
+	/* Last segment: process all remaining data. */
+	aes_gcm_update(key, le_ctr, ghash_acc, walk.src.virt.addr,
+		       walk.dst.virt.addr, nbytes, flags);
+	/*
+	 * The low word of the counter isn't used by the finalize, so there's no
+	 * need to increment it here.
+	 */
+
+	/* Finalize */
+	taglen = crypto_aead_authsize(tfm);
+	if (flags & FLAG_ENC) {
+		/* Finish computing the auth tag. */
+		aes_gcm_enc_final(key, le_ctr, ghash_acc, assoclen,
+				  req->cryptlen, flags);
+
+		/* Store the computed auth tag in the dst scatterlist. */
+		scatterwalk_map_and_copy(ghash_acc, req->dst, req->assoclen +
+					 req->cryptlen, taglen, 1);
+	} else {
+		unsigned int datalen = req->cryptlen - taglen;
+		u8 tag[16];
+
+		/* Get the transmitted auth tag from the src scatterlist. */
+		scatterwalk_map_and_copy(tag, req->src, req->assoclen + datalen,
+					 taglen, 0);
+		/*
+		 * Finish computing the auth tag and compare it to the
+		 * transmitted one.  The assembly function does the actual tag
+		 * comparison.  Here, just check the boolean result.
+		 */
+		if (!aes_gcm_dec_final(key, le_ctr, ghash_acc, assoclen,
+				       datalen, tag, taglen, flags))
+			err = -EBADMSG;
+	}
+	kernel_fpu_end();
+	if (nbytes)
+		skcipher_walk_done(&walk, 0);
+	return err;
+}
+
+#define DEFINE_GCM_ALGS(suffix, flags, generic_driver_name, rfc_driver_name,   \
+			ctxsize, priority)				       \
+									       \
+static int gcm_setkey_##suffix(struct crypto_aead *tfm, const u8 *raw_key,     \
+			       unsigned int keylen)			       \
+{									       \
+	return gcm_setkey(tfm, raw_key, keylen, (flags));		       \
+}									       \
+									       \
+static int gcm_encrypt_##suffix(struct aead_request *req)		       \
+{									       \
+	return gcm_crypt(req, (flags) | FLAG_ENC);			       \
+}									       \
+									       \
+static int gcm_decrypt_##suffix(struct aead_request *req)		       \
+{									       \
+	return gcm_crypt(req, (flags));					       \
+}									       \
+									       \
+static int rfc4106_setkey_##suffix(struct crypto_aead *tfm, const u8 *raw_key, \
+				   unsigned int keylen)			       \
+{									       \
+	return gcm_setkey(tfm, raw_key, keylen, (flags) | FLAG_RFC4106);       \
+}									       \
+									       \
+static int rfc4106_encrypt_##suffix(struct aead_request *req)		       \
+{									       \
+	return gcm_crypt(req, (flags) | FLAG_RFC4106 | FLAG_ENC);	       \
+}									       \
+									       \
+static int rfc4106_decrypt_##suffix(struct aead_request *req)		       \
+{									       \
+	return gcm_crypt(req, (flags) | FLAG_RFC4106);			       \
+}									       \
+									       \
+static struct aead_alg aes_gcm_algs_##suffix[] = { {			       \
+	.setkey			= gcm_setkey_##suffix,			       \
+	.setauthsize		= generic_gcmaes_set_authsize,		       \
+	.encrypt		= gcm_encrypt_##suffix,			       \
+	.decrypt		= gcm_decrypt_##suffix,			       \
+	.ivsize			= GCM_AES_IV_SIZE,			       \
+	.chunksize		= AES_BLOCK_SIZE,			       \
+	.maxauthsize		= 16,					       \
+	.base = {							       \
+		.cra_name		= "gcm(aes)",			       \
+		.cra_driver_name	= generic_driver_name,		       \
+		.cra_priority		= (priority),			       \
+		.cra_blocksize		= 1,				       \
+		.cra_ctxsize		= (ctxsize),			       \
+		.cra_module		= THIS_MODULE,			       \
+	},								       \
+}, {									       \
+	.setkey			= rfc4106_setkey_##suffix,		       \
+	.setauthsize		= common_rfc4106_set_authsize,		       \
+	.encrypt		= rfc4106_encrypt_##suffix,		       \
+	.decrypt		= rfc4106_decrypt_##suffix,		       \
+	.ivsize			= GCM_RFC4106_IV_SIZE,			       \
+	.chunksize		= AES_BLOCK_SIZE,			       \
+	.maxauthsize		= 16,					       \
+	.base = {							       \
+		.cra_name		= "rfc4106(gcm(aes))",		       \
+		.cra_driver_name	= rfc_driver_name,		       \
+		.cra_priority		= (priority),			       \
+		.cra_blocksize		= 1,				       \
+		.cra_ctxsize		= (ctxsize),			       \
+		.cra_module		= THIS_MODULE,			       \
+	},								       \
+} }
+
+/* aes_gcm_algs_aesni */
+DEFINE_GCM_ALGS(aesni, /* no flags */ 0,
+		"generic-gcm-aesni", "rfc4106-gcm-aesni",
+		AES_GCM_KEY_AESNI_SIZE, 400);
+
+/* aes_gcm_algs_aesni_avx */
+DEFINE_GCM_ALGS(aesni_avx, FLAG_AVX,
+		"generic-gcm-aesni-avx", "rfc4106-gcm-aesni-avx",
+		AES_GCM_KEY_AESNI_SIZE, 500);
+
+/* aes_gcm_algs_vaes_avx2 */
+DEFINE_GCM_ALGS(vaes_avx2, FLAG_VAES_AVX2,
+		"generic-gcm-vaes-avx2", "rfc4106-gcm-vaes-avx2",
+		AES_GCM_KEY_VAES_AVX2_SIZE, 600);
+
+/* aes_gcm_algs_vaes_avx512 */
+DEFINE_GCM_ALGS(vaes_avx512, FLAG_VAES_AVX512,
+		"generic-gcm-vaes-avx512", "rfc4106-gcm-vaes-avx512",
+		AES_GCM_KEY_VAES_AVX512_SIZE, 800);
+
+static int __init register_avx_algs(void)
+{
+	int err;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX))
+		return 0;
+	err = crypto_register_skciphers(skcipher_algs_aesni_avx,
+					ARRAY_SIZE(skcipher_algs_aesni_avx));
+	if (err)
+		return err;
+	err = crypto_register_aeads(aes_gcm_algs_aesni_avx,
+				    ARRAY_SIZE(aes_gcm_algs_aesni_avx));
+	if (err)
+		return err;
+	/*
+	 * Note: not all the algorithms registered below actually require
+	 * VPCLMULQDQ.  But in practice every CPU with VAES also has VPCLMULQDQ.
+	 * Similarly, the assembler support was added at about the same time.
+	 * For simplicity, just always check for VAES and VPCLMULQDQ together.
+	 */
+	if (!boot_cpu_has(X86_FEATURE_AVX2) ||
+	    !boot_cpu_has(X86_FEATURE_VAES) ||
+	    !boot_cpu_has(X86_FEATURE_VPCLMULQDQ) ||
+	    !boot_cpu_has(X86_FEATURE_PCLMULQDQ) ||
+	    !cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM, NULL))
+		return 0;
+	err = crypto_register_skciphers(skcipher_algs_vaes_avx2,
+					ARRAY_SIZE(skcipher_algs_vaes_avx2));
+	if (err)
+		return err;
+	err = crypto_register_aeads(aes_gcm_algs_vaes_avx2,
+				    ARRAY_SIZE(aes_gcm_algs_vaes_avx2));
+	if (err)
+		return err;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX512BW) ||
+	    !boot_cpu_has(X86_FEATURE_AVX512VL) ||
+	    !boot_cpu_has(X86_FEATURE_BMI2) ||
+	    !cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM |
+			       XFEATURE_MASK_AVX512, NULL))
+		return 0;
+
+	if (boot_cpu_has(X86_FEATURE_PREFER_YMM)) {
+		int i;
+
+		for (i = 0; i < ARRAY_SIZE(skcipher_algs_vaes_avx512); i++)
+			skcipher_algs_vaes_avx512[i].base.cra_priority = 1;
+		for (i = 0; i < ARRAY_SIZE(aes_gcm_algs_vaes_avx512); i++)
+			aes_gcm_algs_vaes_avx512[i].base.cra_priority = 1;
+	}
+
+	err = crypto_register_skciphers(skcipher_algs_vaes_avx512,
+					ARRAY_SIZE(skcipher_algs_vaes_avx512));
+	if (err)
+		return err;
+	err = crypto_register_aeads(aes_gcm_algs_vaes_avx512,
+				    ARRAY_SIZE(aes_gcm_algs_vaes_avx512));
+	if (err)
+		return err;
+
+	return 0;
+}
+
+#define unregister_skciphers(A) \
+	if (refcount_read(&(A)[0].base.cra_refcnt) != 0) \
+		crypto_unregister_skciphers((A), ARRAY_SIZE(A))
+#define unregister_aeads(A) \
+	if (refcount_read(&(A)[0].base.cra_refcnt) != 0) \
+		crypto_unregister_aeads((A), ARRAY_SIZE(A))
+
+static void unregister_avx_algs(void)
+{
+	unregister_skciphers(skcipher_algs_aesni_avx);
+	unregister_aeads(aes_gcm_algs_aesni_avx);
+	unregister_skciphers(skcipher_algs_vaes_avx2);
+	unregister_skciphers(skcipher_algs_vaes_avx512);
+	unregister_aeads(aes_gcm_algs_vaes_avx2);
+	unregister_aeads(aes_gcm_algs_vaes_avx512);
+}
+#else /* CONFIG_X86_64 */
+static struct aead_alg aes_gcm_algs_aesni[0];
+
+static int __init register_avx_algs(void)
+{
+	return 0;
+}
+
+static void unregister_avx_algs(void)
+{
+}
+#endif /* !CONFIG_X86_64 */
+
+static const struct x86_cpu_id aesni_cpu_id[] = {
+	X86_MATCH_FEATURE(X86_FEATURE_AES, NULL),
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, aesni_cpu_id);
+
+static int __init aesni_init(void)
+{
+	int err;
+
+	if (!x86_match_cpu(aesni_cpu_id))
+		return -ENODEV;
+
+	err = crypto_register_skciphers(aesni_skciphers,
+					ARRAY_SIZE(aesni_skciphers));
+	if (err)
+		return err;
+
+	err = crypto_register_aeads(aes_gcm_algs_aesni,
+				    ARRAY_SIZE(aes_gcm_algs_aesni));
+	if (err)
+		goto unregister_skciphers;
+
+	err = register_avx_algs();
+	if (err)
+		goto unregister_avx;
+
+	return 0;
+
+unregister_avx:
+	unregister_avx_algs();
+	crypto_unregister_aeads(aes_gcm_algs_aesni,
+				ARRAY_SIZE(aes_gcm_algs_aesni));
+unregister_skciphers:
+	crypto_unregister_skciphers(aesni_skciphers,
+				    ARRAY_SIZE(aesni_skciphers));
+	return err;
+}
+
+static void __exit aesni_exit(void)
+{
+	crypto_unregister_aeads(aes_gcm_algs_aesni,
+				ARRAY_SIZE(aes_gcm_algs_aesni));
+	crypto_unregister_skciphers(aesni_skciphers,
+				    ARRAY_SIZE(aesni_skciphers));
+	unregister_avx_algs();
+}
+
+module_init(aesni_init);
+module_exit(aesni_exit);
+
+MODULE_DESCRIPTION("AES cipher and modes, optimized with AES-NI or VAES instructions");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_CRYPTO("aes");
diff --git a/arch/x86/crypto/aria-aesni-avx-asm_64.S b/arch/x86/crypto/aria-aesni-avx-asm_64.S
new file mode 100644
index 000000000..932fb1730
--- /dev/null
+++ b/arch/x86/crypto/aria-aesni-avx-asm_64.S
@@ -0,0 +1,1352 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * ARIA Cipher 16-way parallel algorithm (AVX)
+ *
+ * Copyright (c) 2022 Taehee Yoo <ap420073@gmail.com>
+ *
+ */
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+#include <asm/asm-offsets.h>
+#include <asm/frame.h>
+
+/* register macros */
+#define CTX %rdi
+
+
+#define BV8(a0, a1, a2, a3, a4, a5, a6, a7)		\
+	( (((a0) & 1) << 0) |				\
+	  (((a1) & 1) << 1) |				\
+	  (((a2) & 1) << 2) |				\
+	  (((a3) & 1) << 3) |				\
+	  (((a4) & 1) << 4) |				\
+	  (((a5) & 1) << 5) |				\
+	  (((a6) & 1) << 6) |				\
+	  (((a7) & 1) << 7) )
+
+#define BM8X8(l0, l1, l2, l3, l4, l5, l6, l7)		\
+	( ((l7) << (0 * 8)) |				\
+	  ((l6) << (1 * 8)) |				\
+	  ((l5) << (2 * 8)) |				\
+	  ((l4) << (3 * 8)) |				\
+	  ((l3) << (4 * 8)) |				\
+	  ((l2) << (5 * 8)) |				\
+	  ((l1) << (6 * 8)) |				\
+	  ((l0) << (7 * 8)) )
+
+#define inc_le128(x, minus_one, tmp)			\
+	vpcmpeqq minus_one, x, tmp;			\
+	vpsubq minus_one, x, x;				\
+	vpslldq $8, tmp, tmp;				\
+	vpsubq tmp, x, x;
+
+#define filter_8bit(x, lo_t, hi_t, mask4bit, tmp0)	\
+	vpand x, mask4bit, tmp0;			\
+	vpandn x, mask4bit, x;				\
+	vpsrld $4, x, x;				\
+							\
+	vpshufb tmp0, lo_t, tmp0;			\
+	vpshufb x, hi_t, x;				\
+	vpxor tmp0, x, x;
+
+#define transpose_4x4(x0, x1, x2, x3, t1, t2)		\
+	vpunpckhdq x1, x0, t2;				\
+	vpunpckldq x1, x0, x0;				\
+							\
+	vpunpckldq x3, x2, t1;				\
+	vpunpckhdq x3, x2, x2;				\
+							\
+	vpunpckhqdq t1, x0, x1;				\
+	vpunpcklqdq t1, x0, x0;				\
+							\
+	vpunpckhqdq x2, t2, x3;				\
+	vpunpcklqdq x2, t2, x2;
+
+#define byteslice_16x16b(a0, b0, c0, d0,		\
+			 a1, b1, c1, d1,		\
+			 a2, b2, c2, d2,		\
+			 a3, b3, c3, d3,		\
+			 st0, st1)			\
+	vmovdqu d2, st0;				\
+	vmovdqu d3, st1;				\
+	transpose_4x4(a0, a1, a2, a3, d2, d3);		\
+	transpose_4x4(b0, b1, b2, b3, d2, d3);		\
+	vmovdqu st0, d2;				\
+	vmovdqu st1, d3;				\
+							\
+	vmovdqu a0, st0;				\
+	vmovdqu a1, st1;				\
+	transpose_4x4(c0, c1, c2, c3, a0, a1);		\
+	transpose_4x4(d0, d1, d2, d3, a0, a1);		\
+							\
+	vmovdqu .Lshufb_16x16b(%rip), a0;		\
+	vmovdqu st1, a1;				\
+	vpshufb a0, a2, a2;				\
+	vpshufb a0, a3, a3;				\
+	vpshufb a0, b0, b0;				\
+	vpshufb a0, b1, b1;				\
+	vpshufb a0, b2, b2;				\
+	vpshufb a0, b3, b3;				\
+	vpshufb a0, a1, a1;				\
+	vpshufb a0, c0, c0;				\
+	vpshufb a0, c1, c1;				\
+	vpshufb a0, c2, c2;				\
+	vpshufb a0, c3, c3;				\
+	vpshufb a0, d0, d0;				\
+	vpshufb a0, d1, d1;				\
+	vpshufb a0, d2, d2;				\
+	vpshufb a0, d3, d3;				\
+	vmovdqu d3, st1;				\
+	vmovdqu st0, d3;				\
+	vpshufb a0, d3, a0;				\
+	vmovdqu d2, st0;				\
+							\
+	transpose_4x4(a0, b0, c0, d0, d2, d3);		\
+	transpose_4x4(a1, b1, c1, d1, d2, d3);		\
+	vmovdqu st0, d2;				\
+	vmovdqu st1, d3;				\
+							\
+	vmovdqu b0, st0;				\
+	vmovdqu b1, st1;				\
+	transpose_4x4(a2, b2, c2, d2, b0, b1);		\
+	transpose_4x4(a3, b3, c3, d3, b0, b1);		\
+	vmovdqu st0, b0;				\
+	vmovdqu st1, b1;				\
+	/* does not adjust output bytes inside vectors */
+
+#define debyteslice_16x16b(a0, b0, c0, d0,		\
+			   a1, b1, c1, d1,		\
+			   a2, b2, c2, d2,		\
+			   a3, b3, c3, d3,		\
+			   st0, st1)			\
+	vmovdqu d2, st0;				\
+	vmovdqu d3, st1;				\
+	transpose_4x4(a0, a1, a2, a3, d2, d3);		\
+	transpose_4x4(b0, b1, b2, b3, d2, d3);		\
+	vmovdqu st0, d2;				\
+	vmovdqu st1, d3;				\
+							\
+	vmovdqu a0, st0;				\
+	vmovdqu a1, st1;				\
+	transpose_4x4(c0, c1, c2, c3, a0, a1);		\
+	transpose_4x4(d0, d1, d2, d3, a0, a1);		\
+							\
+	vmovdqu .Lshufb_16x16b(%rip), a0;		\
+	vmovdqu st1, a1;				\
+	vpshufb a0, a2, a2;				\
+	vpshufb a0, a3, a3;				\
+	vpshufb a0, b0, b0;				\
+	vpshufb a0, b1, b1;				\
+	vpshufb a0, b2, b2;				\
+	vpshufb a0, b3, b3;				\
+	vpshufb a0, a1, a1;				\
+	vpshufb a0, c0, c0;				\
+	vpshufb a0, c1, c1;				\
+	vpshufb a0, c2, c2;				\
+	vpshufb a0, c3, c3;				\
+	vpshufb a0, d0, d0;				\
+	vpshufb a0, d1, d1;				\
+	vpshufb a0, d2, d2;				\
+	vpshufb a0, d3, d3;				\
+	vmovdqu d3, st1;				\
+	vmovdqu st0, d3;				\
+	vpshufb a0, d3, a0;				\
+	vmovdqu d2, st0;				\
+							\
+	transpose_4x4(c0, d0, a0, b0, d2, d3);		\
+	transpose_4x4(c1, d1, a1, b1, d2, d3);		\
+	vmovdqu st0, d2;				\
+	vmovdqu st1, d3;				\
+							\
+	vmovdqu b0, st0;				\
+	vmovdqu b1, st1;				\
+	transpose_4x4(c2, d2, a2, b2, b0, b1);		\
+	transpose_4x4(c3, d3, a3, b3, b0, b1);		\
+	vmovdqu st0, b0;				\
+	vmovdqu st1, b1;				\
+	/* does not adjust output bytes inside vectors */
+
+/* load blocks to registers and apply pre-whitening */
+#define inpack16_pre(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     rio)				\
+	vmovdqu (0 * 16)(rio), x0;			\
+	vmovdqu (1 * 16)(rio), x1;			\
+	vmovdqu (2 * 16)(rio), x2;			\
+	vmovdqu (3 * 16)(rio), x3;			\
+	vmovdqu (4 * 16)(rio), x4;			\
+	vmovdqu (5 * 16)(rio), x5;			\
+	vmovdqu (6 * 16)(rio), x6;			\
+	vmovdqu (7 * 16)(rio), x7;			\
+	vmovdqu (8 * 16)(rio), y0;			\
+	vmovdqu (9 * 16)(rio), y1;			\
+	vmovdqu (10 * 16)(rio), y2;			\
+	vmovdqu (11 * 16)(rio), y3;			\
+	vmovdqu (12 * 16)(rio), y4;			\
+	vmovdqu (13 * 16)(rio), y5;			\
+	vmovdqu (14 * 16)(rio), y6;			\
+	vmovdqu (15 * 16)(rio), y7;
+
+/* byteslice pre-whitened blocks and store to temporary memory */
+#define inpack16_post(x0, x1, x2, x3,			\
+		      x4, x5, x6, x7,			\
+		      y0, y1, y2, y3,			\
+		      y4, y5, y6, y7,			\
+		      mem_ab, mem_cd)			\
+	byteslice_16x16b(x0, x1, x2, x3,		\
+			 x4, x5, x6, x7,		\
+			 y0, y1, y2, y3,		\
+			 y4, y5, y6, y7,		\
+			 (mem_ab), (mem_cd));		\
+							\
+	vmovdqu x0, 0 * 16(mem_ab);			\
+	vmovdqu x1, 1 * 16(mem_ab);			\
+	vmovdqu x2, 2 * 16(mem_ab);			\
+	vmovdqu x3, 3 * 16(mem_ab);			\
+	vmovdqu x4, 4 * 16(mem_ab);			\
+	vmovdqu x5, 5 * 16(mem_ab);			\
+	vmovdqu x6, 6 * 16(mem_ab);			\
+	vmovdqu x7, 7 * 16(mem_ab);			\
+	vmovdqu y0, 0 * 16(mem_cd);			\
+	vmovdqu y1, 1 * 16(mem_cd);			\
+	vmovdqu y2, 2 * 16(mem_cd);			\
+	vmovdqu y3, 3 * 16(mem_cd);			\
+	vmovdqu y4, 4 * 16(mem_cd);			\
+	vmovdqu y5, 5 * 16(mem_cd);			\
+	vmovdqu y6, 6 * 16(mem_cd);			\
+	vmovdqu y7, 7 * 16(mem_cd);
+
+#define write_output(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     mem)				\
+	vmovdqu x0, 0 * 16(mem);			\
+	vmovdqu x1, 1 * 16(mem);			\
+	vmovdqu x2, 2 * 16(mem);			\
+	vmovdqu x3, 3 * 16(mem);			\
+	vmovdqu x4, 4 * 16(mem);			\
+	vmovdqu x5, 5 * 16(mem);			\
+	vmovdqu x6, 6 * 16(mem);			\
+	vmovdqu x7, 7 * 16(mem);			\
+	vmovdqu y0, 8 * 16(mem);			\
+	vmovdqu y1, 9 * 16(mem);			\
+	vmovdqu y2, 10 * 16(mem);			\
+	vmovdqu y3, 11 * 16(mem);			\
+	vmovdqu y4, 12 * 16(mem);			\
+	vmovdqu y5, 13 * 16(mem);			\
+	vmovdqu y6, 14 * 16(mem);			\
+	vmovdqu y7, 15 * 16(mem);			\
+
+#define aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, idx)		\
+	vmovdqu x0, ((idx + 0) * 16)(mem_tmp);		\
+	vmovdqu x1, ((idx + 1) * 16)(mem_tmp);		\
+	vmovdqu x2, ((idx + 2) * 16)(mem_tmp);		\
+	vmovdqu x3, ((idx + 3) * 16)(mem_tmp);		\
+	vmovdqu x4, ((idx + 4) * 16)(mem_tmp);		\
+	vmovdqu x5, ((idx + 5) * 16)(mem_tmp);		\
+	vmovdqu x6, ((idx + 6) * 16)(mem_tmp);		\
+	vmovdqu x7, ((idx + 7) * 16)(mem_tmp);
+
+#define aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, idx)		\
+	vmovdqu ((idx + 0) * 16)(mem_tmp), x0;		\
+	vmovdqu ((idx + 1) * 16)(mem_tmp), x1;		\
+	vmovdqu ((idx + 2) * 16)(mem_tmp), x2;		\
+	vmovdqu ((idx + 3) * 16)(mem_tmp), x3;		\
+	vmovdqu ((idx + 4) * 16)(mem_tmp), x4;		\
+	vmovdqu ((idx + 5) * 16)(mem_tmp), x5;		\
+	vmovdqu ((idx + 6) * 16)(mem_tmp), x6;		\
+	vmovdqu ((idx + 7) * 16)(mem_tmp), x7;
+
+#define aria_ark_8way(x0, x1, x2, x3,			\
+		      x4, x5, x6, x7,			\
+		      t0, t1, t2, rk,			\
+		      idx, round)			\
+	/* AddRoundKey */                               \
+	vbroadcastss ((round * 16) + idx + 0)(rk), t0;	\
+	vpsrld $24, t0, t2;				\
+	vpshufb t1, t2, t2;				\
+	vpxor t2, x0, x0;				\
+	vpsrld $16, t0, t2;				\
+	vpshufb t1, t2, t2;				\
+	vpxor t2, x1, x1;				\
+	vpsrld $8, t0, t2;				\
+	vpshufb t1, t2, t2;				\
+	vpxor t2, x2, x2;				\
+	vpshufb t1, t0, t2;				\
+	vpxor t2, x3, x3;				\
+	vbroadcastss ((round * 16) + idx + 4)(rk), t0;	\
+	vpsrld $24, t0, t2;				\
+	vpshufb t1, t2, t2;				\
+	vpxor t2, x4, x4;				\
+	vpsrld $16, t0, t2;				\
+	vpshufb t1, t2, t2;				\
+	vpxor t2, x5, x5;				\
+	vpsrld $8, t0, t2;				\
+	vpshufb t1, t2, t2;				\
+	vpxor t2, x6, x6;				\
+	vpshufb t1, t0, t2;				\
+	vpxor t2, x7, x7;
+
+#define aria_sbox_8way_gfni(x0, x1, x2, x3,		\
+			    x4, x5, x6, x7,		\
+			    t0, t1, t2, t3,		\
+			    t4, t5, t6, t7)		\
+	vmovdqa .Ltf_s2_bitmatrix(%rip), t0;		\
+	vmovdqa .Ltf_inv_bitmatrix(%rip), t1;		\
+	vmovdqa .Ltf_id_bitmatrix(%rip), t2;		\
+	vmovdqa .Ltf_aff_bitmatrix(%rip), t3;		\
+	vmovdqa .Ltf_x2_bitmatrix(%rip), t4;		\
+	vgf2p8affineinvqb $(tf_s2_const), t0, x1, x1;	\
+	vgf2p8affineinvqb $(tf_s2_const), t0, x5, x5;	\
+	vgf2p8affineqb $(tf_inv_const), t1, x2, x2;	\
+	vgf2p8affineqb $(tf_inv_const), t1, x6, x6;	\
+	vgf2p8affineinvqb $0, t2, x2, x2;		\
+	vgf2p8affineinvqb $0, t2, x6, x6;		\
+	vgf2p8affineinvqb $(tf_aff_const), t3, x0, x0;	\
+	vgf2p8affineinvqb $(tf_aff_const), t3, x4, x4;	\
+	vgf2p8affineqb $(tf_x2_const), t4, x3, x3;	\
+	vgf2p8affineqb $(tf_x2_const), t4, x7, x7;	\
+	vgf2p8affineinvqb $0, t2, x3, x3;		\
+	vgf2p8affineinvqb $0, t2, x7, x7
+
+#define aria_sbox_8way(x0, x1, x2, x3,            	\
+		       x4, x5, x6, x7,			\
+		       t0, t1, t2, t3,			\
+		       t4, t5, t6, t7)			\
+	vmovdqa .Linv_shift_row(%rip), t0;		\
+	vmovdqa .Lshift_row(%rip), t1;			\
+	vbroadcastss .L0f0f0f0f(%rip), t6;		\
+	vmovdqa .Ltf_lo__inv_aff__and__s2(%rip), t2;	\
+	vmovdqa .Ltf_hi__inv_aff__and__s2(%rip), t3;	\
+	vmovdqa .Ltf_lo__x2__and__fwd_aff(%rip), t4;	\
+	vmovdqa .Ltf_hi__x2__and__fwd_aff(%rip), t5;	\
+							\
+	vaesenclast t7, x0, x0;				\
+	vaesenclast t7, x4, x4;				\
+	vaesenclast t7, x1, x1;				\
+	vaesenclast t7, x5, x5;				\
+	vaesdeclast t7, x2, x2;				\
+	vaesdeclast t7, x6, x6;				\
+							\
+	/* AES inverse shift rows */			\
+	vpshufb t0, x0, x0;				\
+	vpshufb t0, x4, x4;				\
+	vpshufb t0, x1, x1;				\
+	vpshufb t0, x5, x5;				\
+	vpshufb t1, x3, x3;				\
+	vpshufb t1, x7, x7;				\
+	vpshufb t1, x2, x2;				\
+	vpshufb t1, x6, x6;				\
+							\
+	/* affine transformation for S2 */		\
+	filter_8bit(x1, t2, t3, t6, t0);		\
+	/* affine transformation for S2 */		\
+	filter_8bit(x5, t2, t3, t6, t0);		\
+							\
+	/* affine transformation for X2 */		\
+	filter_8bit(x3, t4, t5, t6, t0);		\
+	/* affine transformation for X2 */		\
+	filter_8bit(x7, t4, t5, t6, t0);		\
+	vaesdeclast t7, x3, x3;				\
+	vaesdeclast t7, x7, x7;
+
+#define aria_diff_m(x0, x1, x2, x3,			\
+		    t0, t1, t2, t3)			\
+	/* T = rotr32(X, 8); */				\
+	/* X ^= T */					\
+	vpxor x0, x3, t0;				\
+	vpxor x1, x0, t1;				\
+	vpxor x2, x1, t2;				\
+	vpxor x3, x2, t3;				\
+	/* X = T ^ rotr(X, 16); */			\
+	vpxor t2, x0, x0;				\
+	vpxor x1, t3, t3;				\
+	vpxor t0, x2, x2;				\
+	vpxor t1, x3, x1;				\
+	vmovdqu t3, x3;
+
+#define aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7)			\
+	/* t1 ^= t2; */					\
+	vpxor y0, x4, x4;				\
+	vpxor y1, x5, x5;				\
+	vpxor y2, x6, x6;				\
+	vpxor y3, x7, x7;				\
+							\
+	/* t2 ^= t3; */					\
+	vpxor y4, y0, y0;				\
+	vpxor y5, y1, y1;				\
+	vpxor y6, y2, y2;				\
+	vpxor y7, y3, y3;				\
+							\
+	/* t0 ^= t1; */					\
+	vpxor x4, x0, x0;				\
+	vpxor x5, x1, x1;				\
+	vpxor x6, x2, x2;				\
+	vpxor x7, x3, x3;				\
+							\
+	/* t3 ^= t1; */					\
+	vpxor x4, y4, y4;				\
+	vpxor x5, y5, y5;				\
+	vpxor x6, y6, y6;				\
+	vpxor x7, y7, y7;				\
+							\
+	/* t2 ^= t0; */					\
+	vpxor x0, y0, y0;				\
+	vpxor x1, y1, y1;				\
+	vpxor x2, y2, y2;				\
+	vpxor x3, y3, y3;				\
+							\
+	/* t1 ^= t2; */					\
+	vpxor y0, x4, x4;				\
+	vpxor y1, x5, x5;				\
+	vpxor y2, x6, x6;				\
+	vpxor y3, x7, x7;
+
+#define aria_fe(x0, x1, x2, x3,				\
+		x4, x5, x6, x7,				\
+		y0, y1, y2, y3,				\
+		y4, y5, y6, y7,				\
+		mem_tmp, rk, round)			\
+	vpxor y7, y7, y7;				\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 8, round);	\
+							\
+	aria_sbox_8way(x2, x3, x0, x1, x6, x7, x4, x5,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 0, round);	\
+							\
+	aria_sbox_8way(x2, x3, x0, x1, x6, x7, x4, x5,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 0);		\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);		\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7);			\
+	/* aria_diff_byte() 				\
+	 * T3 = ABCD -> BADC 				\
+	 * T3 = y4, y5, y6, y7 -> y5, y4, y7, y6 	\
+	 * T0 = ABCD -> CDAB 				\
+	 * T0 = x0, x1, x2, x3 -> x2, x3, x0, x1 	\
+	 * T1 = ABCD -> DCBA 				\
+	 * T1 = x4, x5, x6, x7 -> x7, x6, x5, x4	\
+	 */						\
+	aria_diff_word(x2, x3, x0, x1,			\
+		       x7, x6, x5, x4,			\
+		       y0, y1, y2, y3,			\
+		       y5, y4, y7, y6);			\
+	aria_store_state_8way(x3, x2, x1, x0,		\
+			      x6, x7, x4, x5,		\
+			      mem_tmp, 0);
+
+#define aria_fo(x0, x1, x2, x3,				\
+		x4, x5, x6, x7,				\
+		y0, y1, y2, y3,				\
+		y4, y5, y6, y7,				\
+		mem_tmp, rk, round)			\
+	vpxor y7, y7, y7;				\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 8, round);	\
+							\
+	aria_sbox_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 0, round);	\
+							\
+	aria_sbox_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 0);		\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);		\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7);			\
+	/* aria_diff_byte() 				\
+	 * T1 = ABCD -> BADC 				\
+	 * T1 = x4, x5, x6, x7 -> x5, x4, x7, x6	\
+	 * T2 = ABCD -> CDAB 				\
+	 * T2 = y0, y1, y2, y3, -> y2, y3, y0, y1 	\
+	 * T3 = ABCD -> DCBA 				\
+	 * T3 = y4, y5, y6, y7 -> y7, y6, y5, y4 	\
+	 */						\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x5, x4, x7, x6,			\
+		       y2, y3, y0, y1,			\
+		       y7, y6, y5, y4);			\
+	aria_store_state_8way(x3, x2, x1, x0,		\
+			      x6, x7, x4, x5,		\
+			      mem_tmp, 0);
+
+#define aria_ff(x0, x1, x2, x3,				\
+		x4, x5, x6, x7,				\
+		y0, y1, y2, y3,				\
+		y4, y5, y6, y7,				\
+		mem_tmp, rk, round, last_round)		\
+	vpxor y7, y7, y7;				\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 8, round);	\
+							\
+	aria_sbox_8way(x2, x3, x0, x1, x6, x7, x4, x5,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 8, last_round);	\
+							\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 0, round);	\
+							\
+	aria_sbox_8way(x2, x3, x0, x1, x6, x7, x4, x5,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 0, last_round);	\
+							\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);
+
+#define aria_fe_gfni(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     mem_tmp, rk, round)		\
+	vpxor y7, y7, y7;				\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 8, round);	\
+							\
+	aria_sbox_8way_gfni(x2, x3, x0, x1, 		\
+			    x6, x7, x4, x5,		\
+			    y0, y1, y2, y3, 		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 0, round);	\
+							\
+	aria_sbox_8way_gfni(x2, x3, x0, x1, 		\
+			    x6, x7, x4, x5,		\
+			    y0, y1, y2, y3, 		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 0);		\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);		\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7);			\
+	/* aria_diff_byte() 				\
+	 * T3 = ABCD -> BADC 				\
+	 * T3 = y4, y5, y6, y7 -> y5, y4, y7, y6 	\
+	 * T0 = ABCD -> CDAB 				\
+	 * T0 = x0, x1, x2, x3 -> x2, x3, x0, x1 	\
+	 * T1 = ABCD -> DCBA 				\
+	 * T1 = x4, x5, x6, x7 -> x7, x6, x5, x4	\
+	 */						\
+	aria_diff_word(x2, x3, x0, x1,			\
+		       x7, x6, x5, x4,			\
+		       y0, y1, y2, y3,			\
+		       y5, y4, y7, y6);			\
+	aria_store_state_8way(x3, x2, x1, x0,		\
+			      x6, x7, x4, x5,		\
+			      mem_tmp, 0);
+
+#define aria_fo_gfni(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     mem_tmp, rk, round)		\
+	vpxor y7, y7, y7;				\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 8, round);	\
+							\
+	aria_sbox_8way_gfni(x0, x1, x2, x3, 		\
+			    x4, x5, x6, x7,		\
+			    y0, y1, y2, y3, 		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 0, round);	\
+							\
+	aria_sbox_8way_gfni(x0, x1, x2, x3, 		\
+			    x4, x5, x6, x7,		\
+			    y0, y1, y2, y3, 		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 0);		\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);		\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7);			\
+	/* aria_diff_byte() 				\
+	 * T1 = ABCD -> BADC 				\
+	 * T1 = x4, x5, x6, x7 -> x5, x4, x7, x6	\
+	 * T2 = ABCD -> CDAB 				\
+	 * T2 = y0, y1, y2, y3, -> y2, y3, y0, y1 	\
+	 * T3 = ABCD -> DCBA 				\
+	 * T3 = y4, y5, y6, y7 -> y7, y6, y5, y4 	\
+	 */						\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x5, x4, x7, x6,			\
+		       y2, y3, y0, y1,			\
+		       y7, y6, y5, y4);			\
+	aria_store_state_8way(x3, x2, x1, x0,		\
+			      x6, x7, x4, x5,		\
+			      mem_tmp, 0);
+
+#define aria_ff_gfni(x0, x1, x2, x3,			\
+		x4, x5, x6, x7,				\
+		y0, y1, y2, y3,				\
+		y4, y5, y6, y7,				\
+		mem_tmp, rk, round, last_round)		\
+	vpxor y7, y7, y7;				\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 8, round);	\
+							\
+	aria_sbox_8way_gfni(x2, x3, x0, x1, 		\
+			    x6, x7, x4, x5,		\
+			    y0, y1, y2, y3, 		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 8, last_round);	\
+							\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 0, round);	\
+							\
+	aria_sbox_8way_gfni(x2, x3, x0, x1, 		\
+			    x6, x7, x4, x5,		\
+			    y0, y1, y2, y3, 		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, y7, y2, rk, 0, last_round);	\
+							\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);
+
+/* NB: section is mergeable, all elements must be aligned 16-byte blocks */
+.section	.rodata.cst16, "aM", @progbits, 16
+.align 16
+
+#define SHUFB_BYTES(idx) \
+	0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
+
+.Lshufb_16x16b:
+	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3);
+/* For isolating SubBytes from AESENCLAST, inverse shift row */
+.Linv_shift_row:
+	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
+	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
+.Lshift_row:
+	.byte 0x00, 0x05, 0x0a, 0x0f, 0x04, 0x09, 0x0e, 0x03
+	.byte 0x08, 0x0d, 0x02, 0x07, 0x0c, 0x01, 0x06, 0x0b
+/* For CTR-mode IV byteswap */
+.Lbswap128_mask:
+	.byte 0x0f, 0x0e, 0x0d, 0x0c, 0x0b, 0x0a, 0x09, 0x08
+	.byte 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00
+
+/* AES inverse affine and S2 combined:
+ *      1 1 0 0 0 0 0 1     x0     0
+ *      0 1 0 0 1 0 0 0     x1     0
+ *      1 1 0 0 1 1 1 1     x2     0
+ *      0 1 1 0 1 0 0 1     x3     1
+ *      0 1 0 0 1 1 0 0  *  x4  +  0
+ *      0 1 0 1 1 0 0 0     x5     0
+ *      0 0 0 0 0 1 0 1     x6     0
+ *      1 1 1 0 0 1 1 1     x7     1
+ */
+.Ltf_lo__inv_aff__and__s2:
+	.octa 0x92172DA81A9FA520B2370D883ABF8500
+.Ltf_hi__inv_aff__and__s2:
+	.octa 0x2B15FFC1AF917B45E6D8320C625CB688
+
+/* X2 and AES forward affine combined:
+ *      1 0 1 1 0 0 0 1     x0     0
+ *      0 1 1 1 1 0 1 1     x1     0
+ *      0 0 0 1 1 0 1 0     x2     1
+ *      0 1 0 0 0 1 0 0     x3     0
+ *      0 0 1 1 1 0 1 1  *  x4  +  0
+ *      0 1 0 0 1 0 0 0     x5     0
+ *      1 1 0 1 0 0 1 1     x6     0
+ *      0 1 0 0 1 0 1 0     x7     0
+ */
+.Ltf_lo__x2__and__fwd_aff:
+	.octa 0xEFAE0544FCBD1657B8F95213ABEA4100
+.Ltf_hi__x2__and__fwd_aff:
+	.octa 0x3F893781E95FE1576CDA64D2BA0CB204
+
+/* AES affine: */
+#define tf_aff_const BV8(1, 1, 0, 0, 0, 1, 1, 0)
+.Ltf_aff_bitmatrix:
+	.quad BM8X8(BV8(1, 0, 0, 0, 1, 1, 1, 1),
+		    BV8(1, 1, 0, 0, 0, 1, 1, 1),
+		    BV8(1, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 1, 0, 0, 0, 1),
+		    BV8(1, 1, 1, 1, 1, 0, 0, 0),
+		    BV8(0, 1, 1, 1, 1, 1, 0, 0),
+		    BV8(0, 0, 1, 1, 1, 1, 1, 0),
+		    BV8(0, 0, 0, 1, 1, 1, 1, 1))
+	.quad BM8X8(BV8(1, 0, 0, 0, 1, 1, 1, 1),
+		    BV8(1, 1, 0, 0, 0, 1, 1, 1),
+		    BV8(1, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 1, 0, 0, 0, 1),
+		    BV8(1, 1, 1, 1, 1, 0, 0, 0),
+		    BV8(0, 1, 1, 1, 1, 1, 0, 0),
+		    BV8(0, 0, 1, 1, 1, 1, 1, 0),
+		    BV8(0, 0, 0, 1, 1, 1, 1, 1))
+
+/* AES inverse affine: */
+#define tf_inv_const BV8(1, 0, 1, 0, 0, 0, 0, 0)
+.Ltf_inv_bitmatrix:
+	.quad BM8X8(BV8(0, 0, 1, 0, 0, 1, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 0, 1, 0),
+		    BV8(0, 1, 0, 0, 1, 0, 0, 1),
+		    BV8(1, 0, 1, 0, 0, 1, 0, 0),
+		    BV8(0, 1, 0, 1, 0, 0, 1, 0),
+		    BV8(0, 0, 1, 0, 1, 0, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 1, 0, 0),
+		    BV8(0, 1, 0, 0, 1, 0, 1, 0))
+	.quad BM8X8(BV8(0, 0, 1, 0, 0, 1, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 0, 1, 0),
+		    BV8(0, 1, 0, 0, 1, 0, 0, 1),
+		    BV8(1, 0, 1, 0, 0, 1, 0, 0),
+		    BV8(0, 1, 0, 1, 0, 0, 1, 0),
+		    BV8(0, 0, 1, 0, 1, 0, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 1, 0, 0),
+		    BV8(0, 1, 0, 0, 1, 0, 1, 0))
+
+/* S2: */
+#define tf_s2_const BV8(0, 1, 0, 0, 0, 1, 1, 1)
+.Ltf_s2_bitmatrix:
+	.quad BM8X8(BV8(0, 1, 0, 1, 0, 1, 1, 1),
+		    BV8(0, 0, 1, 1, 1, 1, 1, 1),
+		    BV8(1, 1, 1, 0, 1, 1, 0, 1),
+		    BV8(1, 1, 0, 0, 0, 0, 1, 1),
+		    BV8(0, 1, 0, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 0, 0, 1, 1, 1, 0),
+		    BV8(0, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 1, 0, 1, 1, 0))
+	.quad BM8X8(BV8(0, 1, 0, 1, 0, 1, 1, 1),
+		    BV8(0, 0, 1, 1, 1, 1, 1, 1),
+		    BV8(1, 1, 1, 0, 1, 1, 0, 1),
+		    BV8(1, 1, 0, 0, 0, 0, 1, 1),
+		    BV8(0, 1, 0, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 0, 0, 1, 1, 1, 0),
+		    BV8(0, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 1, 0, 1, 1, 0))
+
+/* X2: */
+#define tf_x2_const BV8(0, 0, 1, 1, 0, 1, 0, 0)
+.Ltf_x2_bitmatrix:
+	.quad BM8X8(BV8(0, 0, 0, 1, 1, 0, 0, 0),
+		    BV8(0, 0, 1, 0, 0, 1, 1, 0),
+		    BV8(0, 0, 0, 0, 1, 0, 1, 0),
+		    BV8(1, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 0, 1, 1, 0, 0),
+		    BV8(0, 1, 1, 0, 1, 0, 1, 1),
+		    BV8(1, 0, 1, 1, 1, 1, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 0, 1, 1))
+	.quad BM8X8(BV8(0, 0, 0, 1, 1, 0, 0, 0),
+		    BV8(0, 0, 1, 0, 0, 1, 1, 0),
+		    BV8(0, 0, 0, 0, 1, 0, 1, 0),
+		    BV8(1, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 0, 1, 1, 0, 0),
+		    BV8(0, 1, 1, 0, 1, 0, 1, 1),
+		    BV8(1, 0, 1, 1, 1, 1, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 0, 1, 1))
+
+/* Identity matrix: */
+.Ltf_id_bitmatrix:
+	.quad BM8X8(BV8(1, 0, 0, 0, 0, 0, 0, 0),
+		    BV8(0, 1, 0, 0, 0, 0, 0, 0),
+		    BV8(0, 0, 1, 0, 0, 0, 0, 0),
+		    BV8(0, 0, 0, 1, 0, 0, 0, 0),
+		    BV8(0, 0, 0, 0, 1, 0, 0, 0),
+		    BV8(0, 0, 0, 0, 0, 1, 0, 0),
+		    BV8(0, 0, 0, 0, 0, 0, 1, 0),
+		    BV8(0, 0, 0, 0, 0, 0, 0, 1))
+	.quad BM8X8(BV8(1, 0, 0, 0, 0, 0, 0, 0),
+		    BV8(0, 1, 0, 0, 0, 0, 0, 0),
+		    BV8(0, 0, 1, 0, 0, 0, 0, 0),
+		    BV8(0, 0, 0, 1, 0, 0, 0, 0),
+		    BV8(0, 0, 0, 0, 1, 0, 0, 0),
+		    BV8(0, 0, 0, 0, 0, 1, 0, 0),
+		    BV8(0, 0, 0, 0, 0, 0, 1, 0),
+		    BV8(0, 0, 0, 0, 0, 0, 0, 1))
+
+/* 4-bit mask */
+.section	.rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
+.align 4
+.L0f0f0f0f:
+	.long 0x0f0f0f0f
+
+.text
+
+SYM_FUNC_START_LOCAL(__aria_aesni_avx_crypt_16way)
+	/* input:
+	*      %r9: rk
+	*      %rsi: dst
+	*      %rdx: src
+	*      %xmm0..%xmm15: 16 byte-sliced blocks
+	*/
+
+	FRAME_BEGIN
+
+	movq %rsi, %rax;
+	leaq 8 * 16(%rax), %r8;
+
+	inpack16_post(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		      %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		      %xmm15, %rax, %r8);
+	aria_fo(%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14, %xmm15,
+		%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		%rax, %r9, 0);
+	aria_fe(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		%xmm15, %rax, %r9, 1);
+	aria_fo(%xmm9, %xmm8, %xmm11, %xmm10, %xmm12, %xmm13, %xmm14, %xmm15,
+		%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		%rax, %r9, 2);
+	aria_fe(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		%xmm15, %rax, %r9, 3);
+	aria_fo(%xmm9, %xmm8, %xmm11, %xmm10, %xmm12, %xmm13, %xmm14, %xmm15,
+		%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		%rax, %r9, 4);
+	aria_fe(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		%xmm15, %rax, %r9, 5);
+	aria_fo(%xmm9, %xmm8, %xmm11, %xmm10, %xmm12, %xmm13, %xmm14, %xmm15,
+		%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		%rax, %r9, 6);
+	aria_fe(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		%xmm15, %rax, %r9, 7);
+	aria_fo(%xmm9, %xmm8, %xmm11, %xmm10, %xmm12, %xmm13, %xmm14, %xmm15,
+		%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		%rax, %r9, 8);
+	aria_fe(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		%xmm15, %rax, %r9, 9);
+	aria_fo(%xmm9, %xmm8, %xmm11, %xmm10, %xmm12, %xmm13, %xmm14, %xmm15,
+		%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		%rax, %r9, 10);
+	cmpl $12, ARIA_CTX_rounds(CTX);
+	jne .Laria_192;
+	aria_ff(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		%xmm15, %rax, %r9, 11, 12);
+	jmp .Laria_end;
+.Laria_192:
+	aria_fe(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		%xmm15, %rax, %r9, 11);
+	aria_fo(%xmm9, %xmm8, %xmm11, %xmm10, %xmm12, %xmm13, %xmm14, %xmm15,
+		%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		%rax, %r9, 12);
+	cmpl $14, ARIA_CTX_rounds(CTX);
+	jne .Laria_256;
+	aria_ff(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		%xmm15, %rax, %r9, 13, 14);
+	jmp .Laria_end;
+.Laria_256:
+	aria_fe(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		%xmm15, %rax, %r9, 13);
+	aria_fo(%xmm9, %xmm8, %xmm11, %xmm10, %xmm12, %xmm13, %xmm14, %xmm15,
+		%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		%rax, %r9, 14);
+	aria_ff(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		%xmm15, %rax, %r9, 15, 16);
+.Laria_end:
+	debyteslice_16x16b(%xmm8, %xmm12, %xmm1, %xmm4,
+			   %xmm9, %xmm13, %xmm0, %xmm5,
+			   %xmm10, %xmm14, %xmm3, %xmm6,
+			   %xmm11, %xmm15, %xmm2, %xmm7,
+			   (%rax), (%r8));
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(__aria_aesni_avx_crypt_16way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx_encrypt_16way)
+	/* input:
+	*      %rdi: ctx, CTX
+	*      %rsi: dst
+	*      %rdx: src
+	*/
+
+	FRAME_BEGIN
+
+	leaq ARIA_CTX_enc_key(CTX), %r9;
+
+	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rdx);
+
+	call __aria_aesni_avx_crypt_16way;
+
+	write_output(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx_encrypt_16way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx_decrypt_16way)
+	/* input:
+	*      %rdi: ctx, CTX
+	*      %rsi: dst
+	*      %rdx: src
+	*/
+
+	FRAME_BEGIN
+
+	leaq ARIA_CTX_dec_key(CTX), %r9;
+
+	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rdx);
+
+	call __aria_aesni_avx_crypt_16way;
+
+	write_output(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx_decrypt_16way)
+
+SYM_FUNC_START_LOCAL(__aria_aesni_avx_ctr_gen_keystream_16way)
+	/* input:
+	*      %rdi: ctx
+	*      %rsi: dst
+	*      %rdx: src
+	*      %rcx: keystream
+	*      %r8: iv (big endian, 128bit)
+	*/
+
+	FRAME_BEGIN
+	/* load IV and byteswap */
+	vmovdqu (%r8), %xmm8;
+
+	vmovdqa .Lbswap128_mask (%rip), %xmm1;
+	vpshufb %xmm1, %xmm8, %xmm3; /* be => le */
+
+	vpcmpeqd %xmm0, %xmm0, %xmm0;
+	vpsrldq $8, %xmm0, %xmm0; /* low: -1, high: 0 */
+
+	/* construct IVs */
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm9;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm10;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm11;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm12;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm13;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm14;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm15;
+	vmovdqu %xmm8, (0 * 16)(%rcx);
+	vmovdqu %xmm9, (1 * 16)(%rcx);
+	vmovdqu %xmm10, (2 * 16)(%rcx);
+	vmovdqu %xmm11, (3 * 16)(%rcx);
+	vmovdqu %xmm12, (4 * 16)(%rcx);
+	vmovdqu %xmm13, (5 * 16)(%rcx);
+	vmovdqu %xmm14, (6 * 16)(%rcx);
+	vmovdqu %xmm15, (7 * 16)(%rcx);
+
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm8;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm9;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm10;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm11;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm12;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm13;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm14;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm15;
+	inc_le128(%xmm3, %xmm0, %xmm5); /* +1 */
+	vpshufb %xmm1, %xmm3, %xmm4;
+	vmovdqu %xmm4, (%r8);
+
+	vmovdqu (0 * 16)(%rcx), %xmm0;
+	vmovdqu (1 * 16)(%rcx), %xmm1;
+	vmovdqu (2 * 16)(%rcx), %xmm2;
+	vmovdqu (3 * 16)(%rcx), %xmm3;
+	vmovdqu (4 * 16)(%rcx), %xmm4;
+	vmovdqu (5 * 16)(%rcx), %xmm5;
+	vmovdqu (6 * 16)(%rcx), %xmm6;
+	vmovdqu (7 * 16)(%rcx), %xmm7;
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(__aria_aesni_avx_ctr_gen_keystream_16way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx_ctr_crypt_16way)
+	/* input:
+	*      %rdi: ctx
+	*      %rsi: dst
+	*      %rdx: src
+	*      %rcx: keystream
+	*      %r8: iv (big endian, 128bit)
+	*/
+	FRAME_BEGIN
+
+	call __aria_aesni_avx_ctr_gen_keystream_16way;
+
+	leaq (%rsi), %r10;
+	leaq (%rdx), %r11;
+	leaq (%rcx), %rsi;
+	leaq (%rcx), %rdx;
+	leaq ARIA_CTX_enc_key(CTX), %r9;
+
+	call __aria_aesni_avx_crypt_16way;
+
+	vpxor (0 * 16)(%r11), %xmm1, %xmm1;
+	vpxor (1 * 16)(%r11), %xmm0, %xmm0;
+	vpxor (2 * 16)(%r11), %xmm3, %xmm3;
+	vpxor (3 * 16)(%r11), %xmm2, %xmm2;
+	vpxor (4 * 16)(%r11), %xmm4, %xmm4;
+	vpxor (5 * 16)(%r11), %xmm5, %xmm5;
+	vpxor (6 * 16)(%r11), %xmm6, %xmm6;
+	vpxor (7 * 16)(%r11), %xmm7, %xmm7;
+	vpxor (8 * 16)(%r11), %xmm8, %xmm8;
+	vpxor (9 * 16)(%r11), %xmm9, %xmm9;
+	vpxor (10 * 16)(%r11), %xmm10, %xmm10;
+	vpxor (11 * 16)(%r11), %xmm11, %xmm11;
+	vpxor (12 * 16)(%r11), %xmm12, %xmm12;
+	vpxor (13 * 16)(%r11), %xmm13, %xmm13;
+	vpxor (14 * 16)(%r11), %xmm14, %xmm14;
+	vpxor (15 * 16)(%r11), %xmm15, %xmm15;
+	write_output(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %r10);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx_ctr_crypt_16way)
+
+SYM_FUNC_START_LOCAL(__aria_aesni_avx_gfni_crypt_16way)
+	/* input:
+	*      %r9: rk
+	*      %rsi: dst
+	*      %rdx: src
+	*      %xmm0..%xmm15: 16 byte-sliced blocks
+	*/
+
+	FRAME_BEGIN
+
+	movq %rsi, %rax;
+	leaq 8 * 16(%rax), %r8;
+
+	inpack16_post(%xmm0, %xmm1, %xmm2, %xmm3,
+		      %xmm4, %xmm5, %xmm6, %xmm7,
+		      %xmm8, %xmm9, %xmm10, %xmm11,
+		      %xmm12, %xmm13, %xmm14,
+		      %xmm15, %rax, %r8);
+	aria_fo_gfni(%xmm8, %xmm9, %xmm10, %xmm11,
+		     %xmm12, %xmm13, %xmm14, %xmm15,
+		     %xmm0, %xmm1, %xmm2, %xmm3,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %rax, %r9, 0);
+	aria_fe_gfni(%xmm1, %xmm0, %xmm3, %xmm2,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11,
+		     %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %r9, 1);
+	aria_fo_gfni(%xmm9, %xmm8, %xmm11, %xmm10,
+		     %xmm12, %xmm13, %xmm14, %xmm15,
+		     %xmm0, %xmm1, %xmm2, %xmm3,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %rax, %r9, 2);
+	aria_fe_gfni(%xmm1, %xmm0, %xmm3, %xmm2,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11,
+		     %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %r9, 3);
+	aria_fo_gfni(%xmm9, %xmm8, %xmm11, %xmm10,
+		     %xmm12, %xmm13, %xmm14, %xmm15,
+		     %xmm0, %xmm1, %xmm2, %xmm3,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %rax, %r9, 4);
+	aria_fe_gfni(%xmm1, %xmm0, %xmm3, %xmm2,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11,
+		     %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %r9, 5);
+	aria_fo_gfni(%xmm9, %xmm8, %xmm11, %xmm10,
+		     %xmm12, %xmm13, %xmm14, %xmm15,
+		     %xmm0, %xmm1, %xmm2, %xmm3,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %rax, %r9, 6);
+	aria_fe_gfni(%xmm1, %xmm0, %xmm3, %xmm2,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11,
+		     %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %r9, 7);
+	aria_fo_gfni(%xmm9, %xmm8, %xmm11, %xmm10,
+		     %xmm12, %xmm13, %xmm14, %xmm15,
+		     %xmm0, %xmm1, %xmm2, %xmm3,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %rax, %r9, 8);
+	aria_fe_gfni(%xmm1, %xmm0, %xmm3, %xmm2,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11,
+		     %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %r9, 9);
+	aria_fo_gfni(%xmm9, %xmm8, %xmm11, %xmm10,
+		     %xmm12, %xmm13, %xmm14, %xmm15,
+		     %xmm0, %xmm1, %xmm2, %xmm3,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %rax, %r9, 10);
+	cmpl $12, ARIA_CTX_rounds(CTX);
+	jne .Laria_gfni_192;
+	aria_ff_gfni(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		%xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		%xmm15, %rax, %r9, 11, 12);
+	jmp .Laria_gfni_end;
+.Laria_gfni_192:
+	aria_fe_gfni(%xmm1, %xmm0, %xmm3, %xmm2,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11,
+		     %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %r9, 11);
+	aria_fo_gfni(%xmm9, %xmm8, %xmm11, %xmm10,
+		     %xmm12, %xmm13, %xmm14, %xmm15,
+		     %xmm0, %xmm1, %xmm2, %xmm3,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %rax, %r9, 12);
+	cmpl $14, ARIA_CTX_rounds(CTX);
+	jne .Laria_gfni_256;
+	aria_ff_gfni(%xmm1, %xmm0, %xmm3, %xmm2,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11,
+		     %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %r9, 13, 14);
+	jmp .Laria_gfni_end;
+.Laria_gfni_256:
+	aria_fe_gfni(%xmm1, %xmm0, %xmm3, %xmm2,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11,
+		     %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %r9, 13);
+	aria_fo_gfni(%xmm9, %xmm8, %xmm11, %xmm10,
+		     %xmm12, %xmm13, %xmm14, %xmm15,
+		     %xmm0, %xmm1, %xmm2, %xmm3,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %rax, %r9, 14);
+	aria_ff_gfni(%xmm1, %xmm0, %xmm3, %xmm2,
+		     %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11,
+		     %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %r9, 15, 16);
+.Laria_gfni_end:
+	debyteslice_16x16b(%xmm8, %xmm12, %xmm1, %xmm4,
+			   %xmm9, %xmm13, %xmm0, %xmm5,
+			   %xmm10, %xmm14, %xmm3, %xmm6,
+			   %xmm11, %xmm15, %xmm2, %xmm7,
+			   (%rax), (%r8));
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(__aria_aesni_avx_gfni_crypt_16way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx_gfni_encrypt_16way)
+	/* input:
+	*      %rdi: ctx, CTX
+	*      %rsi: dst
+	*      %rdx: src
+	*/
+
+	FRAME_BEGIN
+
+	leaq ARIA_CTX_enc_key(CTX), %r9;
+
+	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rdx);
+
+	call __aria_aesni_avx_gfni_crypt_16way;
+
+	write_output(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx_gfni_encrypt_16way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx_gfni_decrypt_16way)
+	/* input:
+	*      %rdi: ctx, CTX
+	*      %rsi: dst
+	*      %rdx: src
+	*/
+
+	FRAME_BEGIN
+
+	leaq ARIA_CTX_dec_key(CTX), %r9;
+
+	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rdx);
+
+	call __aria_aesni_avx_gfni_crypt_16way;
+
+	write_output(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx_gfni_decrypt_16way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx_gfni_ctr_crypt_16way)
+	/* input:
+	*      %rdi: ctx
+	*      %rsi: dst
+	*      %rdx: src
+	*      %rcx: keystream
+	*      %r8: iv (big endian, 128bit)
+	*/
+	FRAME_BEGIN
+
+	call __aria_aesni_avx_ctr_gen_keystream_16way
+
+	leaq (%rsi), %r10;
+	leaq (%rdx), %r11;
+	leaq (%rcx), %rsi;
+	leaq (%rcx), %rdx;
+	leaq ARIA_CTX_enc_key(CTX), %r9;
+
+	call __aria_aesni_avx_gfni_crypt_16way;
+
+	vpxor (0 * 16)(%r11), %xmm1, %xmm1;
+	vpxor (1 * 16)(%r11), %xmm0, %xmm0;
+	vpxor (2 * 16)(%r11), %xmm3, %xmm3;
+	vpxor (3 * 16)(%r11), %xmm2, %xmm2;
+	vpxor (4 * 16)(%r11), %xmm4, %xmm4;
+	vpxor (5 * 16)(%r11), %xmm5, %xmm5;
+	vpxor (6 * 16)(%r11), %xmm6, %xmm6;
+	vpxor (7 * 16)(%r11), %xmm7, %xmm7;
+	vpxor (8 * 16)(%r11), %xmm8, %xmm8;
+	vpxor (9 * 16)(%r11), %xmm9, %xmm9;
+	vpxor (10 * 16)(%r11), %xmm10, %xmm10;
+	vpxor (11 * 16)(%r11), %xmm11, %xmm11;
+	vpxor (12 * 16)(%r11), %xmm12, %xmm12;
+	vpxor (13 * 16)(%r11), %xmm13, %xmm13;
+	vpxor (14 * 16)(%r11), %xmm14, %xmm14;
+	vpxor (15 * 16)(%r11), %xmm15, %xmm15;
+	write_output(%xmm1, %xmm0, %xmm3, %xmm2, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %r10);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx_gfni_ctr_crypt_16way)
diff --git a/arch/x86/crypto/aria-aesni-avx2-asm_64.S b/arch/x86/crypto/aria-aesni-avx2-asm_64.S
new file mode 100644
index 000000000..ed53d4f46
--- /dev/null
+++ b/arch/x86/crypto/aria-aesni-avx2-asm_64.S
@@ -0,0 +1,1433 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * ARIA Cipher 32-way parallel algorithm (AVX2)
+ *
+ * Copyright (c) 2022 Taehee Yoo <ap420073@gmail.com>
+ *
+ */
+
+#include <linux/linkage.h>
+#include <asm/frame.h>
+#include <asm/asm-offsets.h>
+#include <linux/cfi_types.h>
+
+/* register macros */
+#define CTX %rdi
+
+#define ymm0_x xmm0
+#define ymm1_x xmm1
+#define ymm2_x xmm2
+#define ymm3_x xmm3
+#define ymm4_x xmm4
+#define ymm5_x xmm5
+#define ymm6_x xmm6
+#define ymm7_x xmm7
+#define ymm8_x xmm8
+#define ymm9_x xmm9
+#define ymm10_x xmm10
+#define ymm11_x xmm11
+#define ymm12_x xmm12
+#define ymm13_x xmm13
+#define ymm14_x xmm14
+#define ymm15_x xmm15
+
+#define BV8(a0, a1, a2, a3, a4, a5, a6, a7)		\
+	( (((a0) & 1) << 0) |				\
+	  (((a1) & 1) << 1) |				\
+	  (((a2) & 1) << 2) |				\
+	  (((a3) & 1) << 3) |				\
+	  (((a4) & 1) << 4) |				\
+	  (((a5) & 1) << 5) |				\
+	  (((a6) & 1) << 6) |				\
+	  (((a7) & 1) << 7) )
+
+#define BM8X8(l0, l1, l2, l3, l4, l5, l6, l7)		\
+	( ((l7) << (0 * 8)) |				\
+	  ((l6) << (1 * 8)) |				\
+	  ((l5) << (2 * 8)) |				\
+	  ((l4) << (3 * 8)) |				\
+	  ((l3) << (4 * 8)) |				\
+	  ((l2) << (5 * 8)) |				\
+	  ((l1) << (6 * 8)) |				\
+	  ((l0) << (7 * 8)) )
+
+#define inc_le128(x, minus_one, tmp)			\
+	vpcmpeqq minus_one, x, tmp;			\
+	vpsubq minus_one, x, x;				\
+	vpslldq $8, tmp, tmp;				\
+	vpsubq tmp, x, x;
+
+#define filter_8bit(x, lo_t, hi_t, mask4bit, tmp0)	\
+	vpand x, mask4bit, tmp0;			\
+	vpandn x, mask4bit, x;				\
+	vpsrld $4, x, x;				\
+							\
+	vpshufb tmp0, lo_t, tmp0;			\
+	vpshufb x, hi_t, x;				\
+	vpxor tmp0, x, x;
+
+#define transpose_4x4(x0, x1, x2, x3, t1, t2)		\
+	vpunpckhdq x1, x0, t2;				\
+	vpunpckldq x1, x0, x0;				\
+							\
+	vpunpckldq x3, x2, t1;				\
+	vpunpckhdq x3, x2, x2;				\
+							\
+	vpunpckhqdq t1, x0, x1;				\
+	vpunpcklqdq t1, x0, x0;				\
+							\
+	vpunpckhqdq x2, t2, x3;				\
+	vpunpcklqdq x2, t2, x2;
+
+#define byteslice_16x16b(a0, b0, c0, d0,		\
+			 a1, b1, c1, d1,		\
+			 a2, b2, c2, d2,		\
+			 a3, b3, c3, d3,		\
+			 st0, st1)			\
+	vmovdqu d2, st0;				\
+	vmovdqu d3, st1;				\
+	transpose_4x4(a0, a1, a2, a3, d2, d3);		\
+	transpose_4x4(b0, b1, b2, b3, d2, d3);		\
+	vmovdqu st0, d2;				\
+	vmovdqu st1, d3;				\
+							\
+	vmovdqu a0, st0;				\
+	vmovdqu a1, st1;				\
+	transpose_4x4(c0, c1, c2, c3, a0, a1);		\
+	transpose_4x4(d0, d1, d2, d3, a0, a1);		\
+							\
+	vbroadcasti128 .Lshufb_16x16b(%rip), a0;	\
+	vmovdqu st1, a1;				\
+	vpshufb a0, a2, a2;				\
+	vpshufb a0, a3, a3;				\
+	vpshufb a0, b0, b0;				\
+	vpshufb a0, b1, b1;				\
+	vpshufb a0, b2, b2;				\
+	vpshufb a0, b3, b3;				\
+	vpshufb a0, a1, a1;				\
+	vpshufb a0, c0, c0;				\
+	vpshufb a0, c1, c1;				\
+	vpshufb a0, c2, c2;				\
+	vpshufb a0, c3, c3;				\
+	vpshufb a0, d0, d0;				\
+	vpshufb a0, d1, d1;				\
+	vpshufb a0, d2, d2;				\
+	vpshufb a0, d3, d3;				\
+	vmovdqu d3, st1;				\
+	vmovdqu st0, d3;				\
+	vpshufb a0, d3, a0;				\
+	vmovdqu d2, st0;				\
+							\
+	transpose_4x4(a0, b0, c0, d0, d2, d3);		\
+	transpose_4x4(a1, b1, c1, d1, d2, d3);		\
+	vmovdqu st0, d2;				\
+	vmovdqu st1, d3;				\
+							\
+	vmovdqu b0, st0;				\
+	vmovdqu b1, st1;				\
+	transpose_4x4(a2, b2, c2, d2, b0, b1);		\
+	transpose_4x4(a3, b3, c3, d3, b0, b1);		\
+	vmovdqu st0, b0;				\
+	vmovdqu st1, b1;				\
+	/* does not adjust output bytes inside vectors */
+
+#define debyteslice_16x16b(a0, b0, c0, d0,		\
+			   a1, b1, c1, d1,		\
+			   a2, b2, c2, d2,		\
+			   a3, b3, c3, d3,		\
+			   st0, st1)			\
+	vmovdqu d2, st0;				\
+	vmovdqu d3, st1;				\
+	transpose_4x4(a0, a1, a2, a3, d2, d3);		\
+	transpose_4x4(b0, b1, b2, b3, d2, d3);		\
+	vmovdqu st0, d2;				\
+	vmovdqu st1, d3;				\
+							\
+	vmovdqu a0, st0;				\
+	vmovdqu a1, st1;				\
+	transpose_4x4(c0, c1, c2, c3, a0, a1);		\
+	transpose_4x4(d0, d1, d2, d3, a0, a1);		\
+							\
+	vbroadcasti128 .Lshufb_16x16b(%rip), a0;	\
+	vmovdqu st1, a1;				\
+	vpshufb a0, a2, a2;				\
+	vpshufb a0, a3, a3;				\
+	vpshufb a0, b0, b0;				\
+	vpshufb a0, b1, b1;				\
+	vpshufb a0, b2, b2;				\
+	vpshufb a0, b3, b3;				\
+	vpshufb a0, a1, a1;				\
+	vpshufb a0, c0, c0;				\
+	vpshufb a0, c1, c1;				\
+	vpshufb a0, c2, c2;				\
+	vpshufb a0, c3, c3;				\
+	vpshufb a0, d0, d0;				\
+	vpshufb a0, d1, d1;				\
+	vpshufb a0, d2, d2;				\
+	vpshufb a0, d3, d3;				\
+	vmovdqu d3, st1;				\
+	vmovdqu st0, d3;				\
+	vpshufb a0, d3, a0;				\
+	vmovdqu d2, st0;				\
+							\
+	transpose_4x4(c0, d0, a0, b0, d2, d3);		\
+	transpose_4x4(c1, d1, a1, b1, d2, d3);		\
+	vmovdqu st0, d2;				\
+	vmovdqu st1, d3;				\
+							\
+	vmovdqu b0, st0;				\
+	vmovdqu b1, st1;				\
+	transpose_4x4(c2, d2, a2, b2, b0, b1);		\
+	transpose_4x4(c3, d3, a3, b3, b0, b1);		\
+	vmovdqu st0, b0;				\
+	vmovdqu st1, b1;				\
+	/* does not adjust output bytes inside vectors */
+
+/* load blocks to registers and apply pre-whitening */
+#define inpack16_pre(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     rio)				\
+	vmovdqu (0 * 32)(rio), x0;			\
+	vmovdqu (1 * 32)(rio), x1;			\
+	vmovdqu (2 * 32)(rio), x2;			\
+	vmovdqu (3 * 32)(rio), x3;			\
+	vmovdqu (4 * 32)(rio), x4;			\
+	vmovdqu (5 * 32)(rio), x5;			\
+	vmovdqu (6 * 32)(rio), x6;			\
+	vmovdqu (7 * 32)(rio), x7;			\
+	vmovdqu (8 * 32)(rio), y0;			\
+	vmovdqu (9 * 32)(rio), y1;			\
+	vmovdqu (10 * 32)(rio), y2;			\
+	vmovdqu (11 * 32)(rio), y3;			\
+	vmovdqu (12 * 32)(rio), y4;			\
+	vmovdqu (13 * 32)(rio), y5;			\
+	vmovdqu (14 * 32)(rio), y6;			\
+	vmovdqu (15 * 32)(rio), y7;
+
+/* byteslice pre-whitened blocks and store to temporary memory */
+#define inpack16_post(x0, x1, x2, x3,			\
+		      x4, x5, x6, x7,			\
+		      y0, y1, y2, y3,			\
+		      y4, y5, y6, y7,			\
+		      mem_ab, mem_cd)			\
+	byteslice_16x16b(x0, x1, x2, x3,		\
+			 x4, x5, x6, x7,		\
+			 y0, y1, y2, y3,		\
+			 y4, y5, y6, y7,		\
+			 (mem_ab), (mem_cd));		\
+							\
+	vmovdqu x0, 0 * 32(mem_ab);			\
+	vmovdqu x1, 1 * 32(mem_ab);			\
+	vmovdqu x2, 2 * 32(mem_ab);			\
+	vmovdqu x3, 3 * 32(mem_ab);			\
+	vmovdqu x4, 4 * 32(mem_ab);			\
+	vmovdqu x5, 5 * 32(mem_ab);			\
+	vmovdqu x6, 6 * 32(mem_ab);			\
+	vmovdqu x7, 7 * 32(mem_ab);			\
+	vmovdqu y0, 0 * 32(mem_cd);			\
+	vmovdqu y1, 1 * 32(mem_cd);			\
+	vmovdqu y2, 2 * 32(mem_cd);			\
+	vmovdqu y3, 3 * 32(mem_cd);			\
+	vmovdqu y4, 4 * 32(mem_cd);			\
+	vmovdqu y5, 5 * 32(mem_cd);			\
+	vmovdqu y6, 6 * 32(mem_cd);			\
+	vmovdqu y7, 7 * 32(mem_cd);
+
+#define write_output(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     mem)				\
+	vmovdqu x0, 0 * 32(mem);			\
+	vmovdqu x1, 1 * 32(mem);			\
+	vmovdqu x2, 2 * 32(mem);			\
+	vmovdqu x3, 3 * 32(mem);			\
+	vmovdqu x4, 4 * 32(mem);			\
+	vmovdqu x5, 5 * 32(mem);			\
+	vmovdqu x6, 6 * 32(mem);			\
+	vmovdqu x7, 7 * 32(mem);			\
+	vmovdqu y0, 8 * 32(mem);			\
+	vmovdqu y1, 9 * 32(mem);			\
+	vmovdqu y2, 10 * 32(mem);			\
+	vmovdqu y3, 11 * 32(mem);			\
+	vmovdqu y4, 12 * 32(mem);			\
+	vmovdqu y5, 13 * 32(mem);			\
+	vmovdqu y6, 14 * 32(mem);			\
+	vmovdqu y7, 15 * 32(mem);			\
+
+#define aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, idx)		\
+	vmovdqu x0, ((idx + 0) * 32)(mem_tmp);		\
+	vmovdqu x1, ((idx + 1) * 32)(mem_tmp);		\
+	vmovdqu x2, ((idx + 2) * 32)(mem_tmp);		\
+	vmovdqu x3, ((idx + 3) * 32)(mem_tmp);		\
+	vmovdqu x4, ((idx + 4) * 32)(mem_tmp);		\
+	vmovdqu x5, ((idx + 5) * 32)(mem_tmp);		\
+	vmovdqu x6, ((idx + 6) * 32)(mem_tmp);		\
+	vmovdqu x7, ((idx + 7) * 32)(mem_tmp);
+
+#define aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, idx)		\
+	vmovdqu ((idx + 0) * 32)(mem_tmp), x0;		\
+	vmovdqu ((idx + 1) * 32)(mem_tmp), x1;		\
+	vmovdqu ((idx + 2) * 32)(mem_tmp), x2;		\
+	vmovdqu ((idx + 3) * 32)(mem_tmp), x3;		\
+	vmovdqu ((idx + 4) * 32)(mem_tmp), x4;		\
+	vmovdqu ((idx + 5) * 32)(mem_tmp), x5;		\
+	vmovdqu ((idx + 6) * 32)(mem_tmp), x6;		\
+	vmovdqu ((idx + 7) * 32)(mem_tmp), x7;
+
+#define aria_ark_8way(x0, x1, x2, x3,			\
+		      x4, x5, x6, x7,			\
+		      t0, rk, idx, round)		\
+	/* AddRoundKey */                               \
+	vpbroadcastb ((round * 16) + idx + 3)(rk), t0;	\
+	vpxor t0, x0, x0;				\
+	vpbroadcastb ((round * 16) + idx + 2)(rk), t0;	\
+	vpxor t0, x1, x1;				\
+	vpbroadcastb ((round * 16) + idx + 1)(rk), t0;	\
+	vpxor t0, x2, x2;				\
+	vpbroadcastb ((round * 16) + idx + 0)(rk), t0;	\
+	vpxor t0, x3, x3;				\
+	vpbroadcastb ((round * 16) + idx + 7)(rk), t0;	\
+	vpxor t0, x4, x4;				\
+	vpbroadcastb ((round * 16) + idx + 6)(rk), t0;	\
+	vpxor t0, x5, x5;				\
+	vpbroadcastb ((round * 16) + idx + 5)(rk), t0;	\
+	vpxor t0, x6, x6;				\
+	vpbroadcastb ((round * 16) + idx + 4)(rk), t0;	\
+	vpxor t0, x7, x7;
+
+#define aria_sbox_8way_gfni(x0, x1, x2, x3,		\
+			    x4, x5, x6, x7,		\
+			    t0, t1, t2, t3,		\
+			    t4, t5, t6, t7)		\
+	vpbroadcastq .Ltf_s2_bitmatrix(%rip), t0;	\
+	vpbroadcastq .Ltf_inv_bitmatrix(%rip), t1;	\
+	vpbroadcastq .Ltf_id_bitmatrix(%rip), t2;	\
+	vpbroadcastq .Ltf_aff_bitmatrix(%rip), t3;	\
+	vpbroadcastq .Ltf_x2_bitmatrix(%rip), t4;	\
+	vgf2p8affineinvqb $(tf_s2_const), t0, x1, x1;	\
+	vgf2p8affineinvqb $(tf_s2_const), t0, x5, x5;	\
+	vgf2p8affineqb $(tf_inv_const), t1, x2, x2;	\
+	vgf2p8affineqb $(tf_inv_const), t1, x6, x6;	\
+	vgf2p8affineinvqb $0, t2, x2, x2;		\
+	vgf2p8affineinvqb $0, t2, x6, x6;		\
+	vgf2p8affineinvqb $(tf_aff_const), t3, x0, x0;	\
+	vgf2p8affineinvqb $(tf_aff_const), t3, x4, x4;	\
+	vgf2p8affineqb $(tf_x2_const), t4, x3, x3;	\
+	vgf2p8affineqb $(tf_x2_const), t4, x7, x7;	\
+	vgf2p8affineinvqb $0, t2, x3, x3;		\
+	vgf2p8affineinvqb $0, t2, x7, x7
+
+#define aria_sbox_8way(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       t0, t1, t2, t3,			\
+		       t4, t5, t6, t7)			\
+	vpxor t7, t7, t7;				\
+	vpxor t6, t6, t6;				\
+	vbroadcasti128 .Linv_shift_row(%rip), t0;	\
+	vbroadcasti128 .Lshift_row(%rip), t1;		\
+	vbroadcasti128 .Ltf_lo__inv_aff__and__s2(%rip), t2; \
+	vbroadcasti128 .Ltf_hi__inv_aff__and__s2(%rip), t3; \
+	vbroadcasti128 .Ltf_lo__x2__and__fwd_aff(%rip), t4; \
+	vbroadcasti128 .Ltf_hi__x2__and__fwd_aff(%rip), t5; \
+							\
+	vextracti128 $1, x0, t6##_x;			\
+	vaesenclast t7##_x, x0##_x, x0##_x;		\
+	vaesenclast t7##_x, t6##_x, t6##_x;		\
+	vinserti128 $1, t6##_x, x0, x0;			\
+							\
+	vextracti128 $1, x4, t6##_x;			\
+	vaesenclast t7##_x, x4##_x, x4##_x;		\
+	vaesenclast t7##_x, t6##_x, t6##_x;		\
+	vinserti128 $1, t6##_x, x4, x4;			\
+							\
+	vextracti128 $1, x1, t6##_x;			\
+	vaesenclast t7##_x, x1##_x, x1##_x;		\
+	vaesenclast t7##_x, t6##_x, t6##_x;		\
+	vinserti128 $1, t6##_x, x1, x1;			\
+							\
+	vextracti128 $1, x5, t6##_x;			\
+	vaesenclast t7##_x, x5##_x, x5##_x;		\
+	vaesenclast t7##_x, t6##_x, t6##_x;		\
+	vinserti128 $1, t6##_x, x5, x5;			\
+							\
+	vextracti128 $1, x2, t6##_x;			\
+	vaesdeclast t7##_x, x2##_x, x2##_x;		\
+	vaesdeclast t7##_x, t6##_x, t6##_x;		\
+	vinserti128 $1, t6##_x, x2, x2;			\
+							\
+	vextracti128 $1, x6, t6##_x;			\
+	vaesdeclast t7##_x, x6##_x, x6##_x;		\
+	vaesdeclast t7##_x, t6##_x, t6##_x;		\
+	vinserti128 $1, t6##_x, x6, x6;			\
+							\
+	vpbroadcastd .L0f0f0f0f(%rip), t6;		\
+							\
+	/* AES inverse shift rows */			\
+	vpshufb t0, x0, x0;				\
+	vpshufb t0, x4, x4;				\
+	vpshufb t0, x1, x1;				\
+	vpshufb t0, x5, x5;				\
+	vpshufb t1, x3, x3;				\
+	vpshufb t1, x7, x7;				\
+	vpshufb t1, x2, x2;				\
+	vpshufb t1, x6, x6;				\
+							\
+	/* affine transformation for S2 */		\
+	filter_8bit(x1, t2, t3, t6, t0);		\
+	/* affine transformation for S2 */		\
+	filter_8bit(x5, t2, t3, t6, t0);		\
+							\
+	/* affine transformation for X2 */		\
+	filter_8bit(x3, t4, t5, t6, t0);		\
+	/* affine transformation for X2 */		\
+	filter_8bit(x7, t4, t5, t6, t0);		\
+							\
+	vpxor t6, t6, t6;				\
+	vextracti128 $1, x3, t6##_x;			\
+	vaesdeclast t7##_x, x3##_x, x3##_x;		\
+	vaesdeclast t7##_x, t6##_x, t6##_x;		\
+	vinserti128 $1, t6##_x, x3, x3;			\
+							\
+	vextracti128 $1, x7, t6##_x;			\
+	vaesdeclast t7##_x, x7##_x, x7##_x;		\
+	vaesdeclast t7##_x, t6##_x, t6##_x;		\
+	vinserti128 $1, t6##_x, x7, x7;			\
+
+#define aria_diff_m(x0, x1, x2, x3,			\
+		    t0, t1, t2, t3)			\
+	/* T = rotr32(X, 8); */				\
+	/* X ^= T */					\
+	vpxor x0, x3, t0;				\
+	vpxor x1, x0, t1;				\
+	vpxor x2, x1, t2;				\
+	vpxor x3, x2, t3;				\
+	/* X = T ^ rotr(X, 16); */			\
+	vpxor t2, x0, x0;				\
+	vpxor x1, t3, t3;				\
+	vpxor t0, x2, x2;				\
+	vpxor t1, x3, x1;				\
+	vmovdqu t3, x3;
+
+#define aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7)			\
+	/* t1 ^= t2; */					\
+	vpxor y0, x4, x4;				\
+	vpxor y1, x5, x5;				\
+	vpxor y2, x6, x6;				\
+	vpxor y3, x7, x7;				\
+							\
+	/* t2 ^= t3; */					\
+	vpxor y4, y0, y0;				\
+	vpxor y5, y1, y1;				\
+	vpxor y6, y2, y2;				\
+	vpxor y7, y3, y3;				\
+							\
+	/* t0 ^= t1; */					\
+	vpxor x4, x0, x0;				\
+	vpxor x5, x1, x1;				\
+	vpxor x6, x2, x2;				\
+	vpxor x7, x3, x3;				\
+							\
+	/* t3 ^= t1; */					\
+	vpxor x4, y4, y4;				\
+	vpxor x5, y5, y5;				\
+	vpxor x6, y6, y6;				\
+	vpxor x7, y7, y7;				\
+							\
+	/* t2 ^= t0; */					\
+	vpxor x0, y0, y0;				\
+	vpxor x1, y1, y1;				\
+	vpxor x2, y2, y2;				\
+	vpxor x3, y3, y3;				\
+							\
+	/* t1 ^= t2; */					\
+	vpxor y0, x4, x4;				\
+	vpxor y1, x5, x5;				\
+	vpxor y2, x6, x6;				\
+	vpxor y3, x7, x7;
+
+#define aria_fe(x0, x1, x2, x3,				\
+		x4, x5, x6, x7,				\
+		y0, y1, y2, y3,				\
+		y4, y5, y6, y7,				\
+		mem_tmp, rk, round)			\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 8, round);		\
+							\
+	aria_sbox_8way(x2, x3, x0, x1, x6, x7, x4, x5,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 0, round);		\
+							\
+	aria_sbox_8way(x2, x3, x0, x1, x6, x7, x4, x5,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 0);		\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);		\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7);			\
+	/* aria_diff_byte()				\
+	 * T3 = ABCD -> BADC				\
+	 * T3 = y4, y5, y6, y7 -> y5, y4, y7, y6	\
+	 * T0 = ABCD -> CDAB				\
+	 * T0 = x0, x1, x2, x3 -> x2, x3, x0, x1	\
+	 * T1 = ABCD -> DCBA				\
+	 * T1 = x4, x5, x6, x7 -> x7, x6, x5, x4	\
+	 */						\
+	aria_diff_word(x2, x3, x0, x1,			\
+		       x7, x6, x5, x4,			\
+		       y0, y1, y2, y3,			\
+		       y5, y4, y7, y6);			\
+	aria_store_state_8way(x3, x2, x1, x0,		\
+			      x6, x7, x4, x5,		\
+			      mem_tmp, 0);
+
+#define aria_fo(x0, x1, x2, x3,				\
+		x4, x5, x6, x7,				\
+		y0, y1, y2, y3,				\
+		y4, y5, y6, y7,				\
+		mem_tmp, rk, round)			\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 8, round);		\
+							\
+	aria_sbox_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 0, round);		\
+							\
+	aria_sbox_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 0);		\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);		\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7);			\
+	/* aria_diff_byte()				\
+	 * T1 = ABCD -> BADC				\
+	 * T1 = x4, x5, x6, x7 -> x5, x4, x7, x6	\
+	 * T2 = ABCD -> CDAB				\
+	 * T2 = y0, y1, y2, y3, -> y2, y3, y0, y1	\
+	 * T3 = ABCD -> DCBA				\
+	 * T3 = y4, y5, y6, y7 -> y7, y6, y5, y4	\
+	 */						\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x5, x4, x7, x6,			\
+		       y2, y3, y0, y1,			\
+		       y7, y6, y5, y4);			\
+	aria_store_state_8way(x3, x2, x1, x0,		\
+			      x6, x7, x4, x5,		\
+			      mem_tmp, 0);
+
+#define aria_ff(x0, x1, x2, x3,				\
+		x4, x5, x6, x7,				\
+		y0, y1, y2, y3,				\
+		y4, y5, y6, y7,				\
+		mem_tmp, rk, round, last_round)		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 8, round);		\
+							\
+	aria_sbox_8way(x2, x3, x0, x1, x6, x7, x4, x5,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 8, last_round);		\
+							\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 0, round);		\
+							\
+	aria_sbox_8way(x2, x3, x0, x1, x6, x7, x4, x5,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7);	\
+							\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 0, last_round);		\
+							\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);
+
+#define aria_fe_gfni(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     mem_tmp, rk, round)		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 8, round);		\
+							\
+	aria_sbox_8way_gfni(x2, x3, x0, x1,		\
+			    x6, x7, x4, x5,		\
+			    y0, y1, y2, y3,		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 0, round);		\
+							\
+	aria_sbox_8way_gfni(x2, x3, x0, x1,		\
+			    x6, x7, x4, x5,		\
+			    y0, y1, y2, y3,		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 0);		\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);		\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7);			\
+	/* aria_diff_byte()				\
+	 * T3 = ABCD -> BADC				\
+	 * T3 = y4, y5, y6, y7 -> y5, y4, y7, y6	\
+	 * T0 = ABCD -> CDAB				\
+	 * T0 = x0, x1, x2, x3 -> x2, x3, x0, x1	\
+	 * T1 = ABCD -> DCBA				\
+	 * T1 = x4, x5, x6, x7 -> x7, x6, x5, x4	\
+	 */						\
+	aria_diff_word(x2, x3, x0, x1,			\
+		       x7, x6, x5, x4,			\
+		       y0, y1, y2, y3,			\
+		       y5, y4, y7, y6);			\
+	aria_store_state_8way(x3, x2, x1, x0,		\
+			      x6, x7, x4, x5,		\
+			      mem_tmp, 0);
+
+#define aria_fo_gfni(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     mem_tmp, rk, round)		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 8, round);		\
+							\
+	aria_sbox_8way_gfni(x0, x1, x2, x3,		\
+			    x4, x5, x6, x7,		\
+			    y0, y1, y2, y3,		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 0, round);		\
+							\
+	aria_sbox_8way_gfni(x0, x1, x2, x3,		\
+			    x4, x5, x6, x7,		\
+			    y0, y1, y2, y3,		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_diff_m(x0, x1, x2, x3, y0, y1, y2, y3);	\
+	aria_diff_m(x4, x5, x6, x7, y0, y1, y2, y3);	\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 0);		\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);		\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7);			\
+	/* aria_diff_byte()				\
+	 * T1 = ABCD -> BADC				\
+	 * T1 = x4, x5, x6, x7 -> x5, x4, x7, x6	\
+	 * T2 = ABCD -> CDAB				\
+	 * T2 = y0, y1, y2, y3, -> y2, y3, y0, y1	\
+	 * T3 = ABCD -> DCBA				\
+	 * T3 = y4, y5, y6, y7 -> y7, y6, y5, y4	\
+	 */						\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x5, x4, x7, x6,			\
+		       y2, y3, y0, y1,			\
+		       y7, y6, y5, y4);			\
+	aria_store_state_8way(x3, x2, x1, x0,		\
+			      x6, x7, x4, x5,		\
+			      mem_tmp, 0);
+
+#define aria_ff_gfni(x0, x1, x2, x3,			\
+		x4, x5, x6, x7,				\
+		y0, y1, y2, y3,				\
+		y4, y5, y6, y7,				\
+		mem_tmp, rk, round, last_round)		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 8, round);		\
+							\
+	aria_sbox_8way_gfni(x2, x3, x0, x1,		\
+			    x6, x7, x4, x5,		\
+			    y0, y1, y2, y3,		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 8, last_round);		\
+							\
+	aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, 8);		\
+							\
+	aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, 0);		\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 0, round);		\
+							\
+	aria_sbox_8way_gfni(x2, x3, x0, x1,		\
+			    x6, x7, x4, x5,		\
+			    y0, y1, y2, y3,		\
+			    y4, y5, y6, y7);		\
+							\
+	aria_ark_8way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		      y0, rk, 0, last_round);		\
+							\
+	aria_load_state_8way(y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     mem_tmp, 8);
+
+.section        .rodata.cst32.shufb_16x16b, "aM", @progbits, 32
+.align 32
+#define SHUFB_BYTES(idx) \
+	0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
+.Lshufb_16x16b:
+	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
+	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
+
+.section	.rodata.cst16, "aM", @progbits, 16
+.align 16
+/* For isolating SubBytes from AESENCLAST, inverse shift row */
+.Linv_shift_row:
+	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
+	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
+.Lshift_row:
+	.byte 0x00, 0x05, 0x0a, 0x0f, 0x04, 0x09, 0x0e, 0x03
+	.byte 0x08, 0x0d, 0x02, 0x07, 0x0c, 0x01, 0x06, 0x0b
+/* For CTR-mode IV byteswap */
+.Lbswap128_mask:
+	.byte 0x0f, 0x0e, 0x0d, 0x0c, 0x0b, 0x0a, 0x09, 0x08
+	.byte 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00
+
+/* AES inverse affine and S2 combined:
+ *      1 1 0 0 0 0 0 1     x0     0
+ *      0 1 0 0 1 0 0 0     x1     0
+ *      1 1 0 0 1 1 1 1     x2     0
+ *      0 1 1 0 1 0 0 1     x3     1
+ *      0 1 0 0 1 1 0 0  *  x4  +  0
+ *      0 1 0 1 1 0 0 0     x5     0
+ *      0 0 0 0 0 1 0 1     x6     0
+ *      1 1 1 0 0 1 1 1     x7     1
+ */
+.Ltf_lo__inv_aff__and__s2:
+	.octa 0x92172DA81A9FA520B2370D883ABF8500
+.Ltf_hi__inv_aff__and__s2:
+	.octa 0x2B15FFC1AF917B45E6D8320C625CB688
+
+/* X2 and AES forward affine combined:
+ *      1 0 1 1 0 0 0 1     x0     0
+ *      0 1 1 1 1 0 1 1     x1     0
+ *      0 0 0 1 1 0 1 0     x2     1
+ *      0 1 0 0 0 1 0 0     x3     0
+ *      0 0 1 1 1 0 1 1  *  x4  +  0
+ *      0 1 0 0 1 0 0 0     x5     0
+ *      1 1 0 1 0 0 1 1     x6     0
+ *      0 1 0 0 1 0 1 0     x7     0
+ */
+.Ltf_lo__x2__and__fwd_aff:
+	.octa 0xEFAE0544FCBD1657B8F95213ABEA4100
+.Ltf_hi__x2__and__fwd_aff:
+	.octa 0x3F893781E95FE1576CDA64D2BA0CB204
+
+.section	.rodata.cst8, "aM", @progbits, 8
+.align 8
+/* AES affine: */
+#define tf_aff_const BV8(1, 1, 0, 0, 0, 1, 1, 0)
+.Ltf_aff_bitmatrix:
+	.quad BM8X8(BV8(1, 0, 0, 0, 1, 1, 1, 1),
+		    BV8(1, 1, 0, 0, 0, 1, 1, 1),
+		    BV8(1, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 1, 0, 0, 0, 1),
+		    BV8(1, 1, 1, 1, 1, 0, 0, 0),
+		    BV8(0, 1, 1, 1, 1, 1, 0, 0),
+		    BV8(0, 0, 1, 1, 1, 1, 1, 0),
+		    BV8(0, 0, 0, 1, 1, 1, 1, 1))
+
+/* AES inverse affine: */
+#define tf_inv_const BV8(1, 0, 1, 0, 0, 0, 0, 0)
+.Ltf_inv_bitmatrix:
+	.quad BM8X8(BV8(0, 0, 1, 0, 0, 1, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 0, 1, 0),
+		    BV8(0, 1, 0, 0, 1, 0, 0, 1),
+		    BV8(1, 0, 1, 0, 0, 1, 0, 0),
+		    BV8(0, 1, 0, 1, 0, 0, 1, 0),
+		    BV8(0, 0, 1, 0, 1, 0, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 1, 0, 0),
+		    BV8(0, 1, 0, 0, 1, 0, 1, 0))
+
+/* S2: */
+#define tf_s2_const BV8(0, 1, 0, 0, 0, 1, 1, 1)
+.Ltf_s2_bitmatrix:
+	.quad BM8X8(BV8(0, 1, 0, 1, 0, 1, 1, 1),
+		    BV8(0, 0, 1, 1, 1, 1, 1, 1),
+		    BV8(1, 1, 1, 0, 1, 1, 0, 1),
+		    BV8(1, 1, 0, 0, 0, 0, 1, 1),
+		    BV8(0, 1, 0, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 0, 0, 1, 1, 1, 0),
+		    BV8(0, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 1, 0, 1, 1, 0))
+
+/* X2: */
+#define tf_x2_const BV8(0, 0, 1, 1, 0, 1, 0, 0)
+.Ltf_x2_bitmatrix:
+	.quad BM8X8(BV8(0, 0, 0, 1, 1, 0, 0, 0),
+		    BV8(0, 0, 1, 0, 0, 1, 1, 0),
+		    BV8(0, 0, 0, 0, 1, 0, 1, 0),
+		    BV8(1, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 0, 1, 1, 0, 0),
+		    BV8(0, 1, 1, 0, 1, 0, 1, 1),
+		    BV8(1, 0, 1, 1, 1, 1, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 0, 1, 1))
+
+/* Identity matrix: */
+.Ltf_id_bitmatrix:
+	.quad BM8X8(BV8(1, 0, 0, 0, 0, 0, 0, 0),
+		    BV8(0, 1, 0, 0, 0, 0, 0, 0),
+		    BV8(0, 0, 1, 0, 0, 0, 0, 0),
+		    BV8(0, 0, 0, 1, 0, 0, 0, 0),
+		    BV8(0, 0, 0, 0, 1, 0, 0, 0),
+		    BV8(0, 0, 0, 0, 0, 1, 0, 0),
+		    BV8(0, 0, 0, 0, 0, 0, 1, 0),
+		    BV8(0, 0, 0, 0, 0, 0, 0, 1))
+
+/* 4-bit mask */
+.section	.rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
+.align 4
+.L0f0f0f0f:
+	.long 0x0f0f0f0f
+
+.text
+
+SYM_FUNC_START_LOCAL(__aria_aesni_avx2_crypt_32way)
+	/* input:
+	 *      %r9: rk
+	 *      %rsi: dst
+	 *      %rdx: src
+	 *      %ymm0..%ymm15: byte-sliced blocks
+	 */
+
+	FRAME_BEGIN
+
+	movq %rsi, %rax;
+	leaq 8 * 32(%rax), %r8;
+
+	inpack16_post(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		      %ymm15, %rax, %r8);
+	aria_fo(%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14, %ymm15,
+		%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		%rax, %r9, 0);
+	aria_fe(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		%ymm15, %rax, %r9, 1);
+	aria_fo(%ymm9, %ymm8, %ymm11, %ymm10, %ymm12, %ymm13, %ymm14, %ymm15,
+		%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		%rax, %r9, 2);
+	aria_fe(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		%ymm15, %rax, %r9, 3);
+	aria_fo(%ymm9, %ymm8, %ymm11, %ymm10, %ymm12, %ymm13, %ymm14, %ymm15,
+		%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		%rax, %r9, 4);
+	aria_fe(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		%ymm15, %rax, %r9, 5);
+	aria_fo(%ymm9, %ymm8, %ymm11, %ymm10, %ymm12, %ymm13, %ymm14, %ymm15,
+		%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		%rax, %r9, 6);
+	aria_fe(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		%ymm15, %rax, %r9, 7);
+	aria_fo(%ymm9, %ymm8, %ymm11, %ymm10, %ymm12, %ymm13, %ymm14, %ymm15,
+		%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		%rax, %r9, 8);
+	aria_fe(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		%ymm15, %rax, %r9, 9);
+	aria_fo(%ymm9, %ymm8, %ymm11, %ymm10, %ymm12, %ymm13, %ymm14, %ymm15,
+		%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		%rax, %r9, 10);
+	cmpl $12, ARIA_CTX_rounds(CTX);
+	jne .Laria_192;
+	aria_ff(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		%ymm15, %rax, %r9, 11, 12);
+	jmp .Laria_end;
+.Laria_192:
+	aria_fe(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		%ymm15, %rax, %r9, 11);
+	aria_fo(%ymm9, %ymm8, %ymm11, %ymm10, %ymm12, %ymm13, %ymm14, %ymm15,
+		%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		%rax, %r9, 12);
+	cmpl $14, ARIA_CTX_rounds(CTX);
+	jne .Laria_256;
+	aria_ff(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		%ymm15, %rax, %r9, 13, 14);
+	jmp .Laria_end;
+.Laria_256:
+	aria_fe(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		%ymm15, %rax, %r9, 13);
+	aria_fo(%ymm9, %ymm8, %ymm11, %ymm10, %ymm12, %ymm13, %ymm14, %ymm15,
+		%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		%rax, %r9, 14);
+	aria_ff(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		%ymm15, %rax, %r9, 15, 16);
+.Laria_end:
+	debyteslice_16x16b(%ymm8, %ymm12, %ymm1, %ymm4,
+			   %ymm9, %ymm13, %ymm0, %ymm5,
+			   %ymm10, %ymm14, %ymm3, %ymm6,
+			   %ymm11, %ymm15, %ymm2, %ymm7,
+			   (%rax), (%r8));
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(__aria_aesni_avx2_crypt_32way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx2_encrypt_32way)
+	/* input:
+	 *      %rdi: ctx, CTX
+	 *      %rsi: dst
+	 *      %rdx: src
+	 */
+
+	FRAME_BEGIN
+
+	leaq ARIA_CTX_enc_key(CTX), %r9;
+
+	inpack16_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rdx);
+
+	call __aria_aesni_avx2_crypt_32way;
+
+	write_output(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx2_encrypt_32way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx2_decrypt_32way)
+	/* input:
+	 *      %rdi: ctx, CTX
+	 *      %rsi: dst
+	 *      %rdx: src
+	 */
+
+	FRAME_BEGIN
+
+	leaq ARIA_CTX_dec_key(CTX), %r9;
+
+	inpack16_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rdx);
+
+	call __aria_aesni_avx2_crypt_32way;
+
+	write_output(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx2_decrypt_32way)
+
+SYM_FUNC_START_LOCAL(__aria_aesni_avx2_ctr_gen_keystream_32way)
+	/* input:
+	 *      %rdi: ctx
+	 *      %rsi: dst
+	 *      %rdx: src
+	 *      %rcx: keystream
+	 *      %r8: iv (big endian, 128bit)
+	 */
+
+	FRAME_BEGIN
+	movq 8(%r8), %r11;
+	bswapq %r11;
+
+	vbroadcasti128 .Lbswap128_mask (%rip), %ymm6;
+	vpcmpeqd %ymm0, %ymm0, %ymm0;
+	vpsrldq $8, %ymm0, %ymm0;   /* ab: -1:0 ; cd: -1:0 */
+	vpaddq %ymm0, %ymm0, %ymm5; /* ab: -2:0 ; cd: -2:0 */
+
+	/* load IV and byteswap */
+	vmovdqu (%r8), %xmm7;
+	vpshufb %xmm6, %xmm7, %xmm7;
+	vmovdqa %xmm7, %xmm3;
+	inc_le128(%xmm7, %xmm0, %xmm4);
+	vinserti128 $1, %xmm7, %ymm3, %ymm3;
+	vpshufb %ymm6, %ymm3, %ymm8; /* +1 ; +0 */
+
+	/* check need for handling 64-bit overflow and carry */
+	cmpq $(0xffffffffffffffff - 32), %r11;
+	ja .Lhandle_ctr_carry;
+
+	/* construct IVs */
+	vpsubq %ymm5, %ymm3, %ymm3; /* +3 ; +2 */
+	vpshufb %ymm6, %ymm3, %ymm9;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +5 ; +4 */
+	vpshufb %ymm6, %ymm3, %ymm10;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +7 ; +6 */
+	vpshufb %ymm6, %ymm3, %ymm11;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +9 ; +8 */
+	vpshufb %ymm6, %ymm3, %ymm12;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +11 ; +10 */
+	vpshufb %ymm6, %ymm3, %ymm13;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +13 ; +12 */
+	vpshufb %ymm6, %ymm3, %ymm14;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +15 ; +14 */
+	vpshufb %ymm6, %ymm3, %ymm15;
+	vmovdqu %ymm8, (0 * 32)(%rcx);
+	vmovdqu %ymm9, (1 * 32)(%rcx);
+	vmovdqu %ymm10, (2 * 32)(%rcx);
+	vmovdqu %ymm11, (3 * 32)(%rcx);
+	vmovdqu %ymm12, (4 * 32)(%rcx);
+	vmovdqu %ymm13, (5 * 32)(%rcx);
+	vmovdqu %ymm14, (6 * 32)(%rcx);
+	vmovdqu %ymm15, (7 * 32)(%rcx);
+
+	vpsubq %ymm5, %ymm3, %ymm3; /* +17 ; +16 */
+	vpshufb %ymm6, %ymm3, %ymm8;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +19 ; +18 */
+	vpshufb %ymm6, %ymm3, %ymm9;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +21 ; +20 */
+	vpshufb %ymm6, %ymm3, %ymm10;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +23 ; +22 */
+	vpshufb %ymm6, %ymm3, %ymm11;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +25 ; +24 */
+	vpshufb %ymm6, %ymm3, %ymm12;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +27 ; +26 */
+	vpshufb %ymm6, %ymm3, %ymm13;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +29 ; +28 */
+	vpshufb %ymm6, %ymm3, %ymm14;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +31 ; +30 */
+	vpshufb %ymm6, %ymm3, %ymm15;
+	vpsubq %ymm5, %ymm3, %ymm3; /* +32 */
+	vpshufb %xmm6, %xmm3, %xmm3;
+	vmovdqu %xmm3, (%r8);
+	vmovdqu (0 * 32)(%rcx), %ymm0;
+	vmovdqu (1 * 32)(%rcx), %ymm1;
+	vmovdqu (2 * 32)(%rcx), %ymm2;
+	vmovdqu (3 * 32)(%rcx), %ymm3;
+	vmovdqu (4 * 32)(%rcx), %ymm4;
+	vmovdqu (5 * 32)(%rcx), %ymm5;
+	vmovdqu (6 * 32)(%rcx), %ymm6;
+	vmovdqu (7 * 32)(%rcx), %ymm7;
+	jmp .Lctr_carry_done;
+
+	.Lhandle_ctr_carry:
+	/* construct IVs */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm9; /* +3 ; +2 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm10; /* +5 ; +4 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm11; /* +7 ; +6 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm12; /* +9 ; +8 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm13; /* +11 ; +10 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm14; /* +13 ; +12 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm15; /* +15 ; +14 */
+	vmovdqu %ymm8, (0 * 32)(%rcx);
+	vmovdqu %ymm9, (1 * 32)(%rcx);
+	vmovdqu %ymm10, (2 * 32)(%rcx);
+	vmovdqu %ymm11, (3 * 32)(%rcx);
+	vmovdqu %ymm12, (4 * 32)(%rcx);
+	vmovdqu %ymm13, (5 * 32)(%rcx);
+	vmovdqu %ymm14, (6 * 32)(%rcx);
+	vmovdqu %ymm15, (7 * 32)(%rcx);
+
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm8; /* +17 ; +16 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm9; /* +19 ; +18 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm10; /* +21 ; +20 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm11; /* +23 ; +22 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm12; /* +25 ; +24 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm13; /* +27 ; +26 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm14; /* +29 ; +28 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vpshufb %ymm6, %ymm3, %ymm15; /* +31 ; +30 */
+	inc_le128(%ymm3, %ymm0, %ymm4);
+	vextracti128 $1, %ymm3, %xmm3;
+	vpshufb %xmm6, %xmm3, %xmm3; /* +32 */
+	vmovdqu %xmm3, (%r8);
+	vmovdqu (0 * 32)(%rcx), %ymm0;
+	vmovdqu (1 * 32)(%rcx), %ymm1;
+	vmovdqu (2 * 32)(%rcx), %ymm2;
+	vmovdqu (3 * 32)(%rcx), %ymm3;
+	vmovdqu (4 * 32)(%rcx), %ymm4;
+	vmovdqu (5 * 32)(%rcx), %ymm5;
+	vmovdqu (6 * 32)(%rcx), %ymm6;
+	vmovdqu (7 * 32)(%rcx), %ymm7;
+
+	.Lctr_carry_done:
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(__aria_aesni_avx2_ctr_gen_keystream_32way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx2_ctr_crypt_32way)
+	/* input:
+	 *      %rdi: ctx
+	 *      %rsi: dst
+	 *      %rdx: src
+	 *      %rcx: keystream
+	 *      %r8: iv (big endian, 128bit)
+	 */
+	FRAME_BEGIN
+
+	call __aria_aesni_avx2_ctr_gen_keystream_32way;
+
+	leaq (%rsi), %r10;
+	leaq (%rdx), %r11;
+	leaq (%rcx), %rsi;
+	leaq (%rcx), %rdx;
+	leaq ARIA_CTX_enc_key(CTX), %r9;
+
+	call __aria_aesni_avx2_crypt_32way;
+
+	vpxor (0 * 32)(%r11), %ymm1, %ymm1;
+	vpxor (1 * 32)(%r11), %ymm0, %ymm0;
+	vpxor (2 * 32)(%r11), %ymm3, %ymm3;
+	vpxor (3 * 32)(%r11), %ymm2, %ymm2;
+	vpxor (4 * 32)(%r11), %ymm4, %ymm4;
+	vpxor (5 * 32)(%r11), %ymm5, %ymm5;
+	vpxor (6 * 32)(%r11), %ymm6, %ymm6;
+	vpxor (7 * 32)(%r11), %ymm7, %ymm7;
+	vpxor (8 * 32)(%r11), %ymm8, %ymm8;
+	vpxor (9 * 32)(%r11), %ymm9, %ymm9;
+	vpxor (10 * 32)(%r11), %ymm10, %ymm10;
+	vpxor (11 * 32)(%r11), %ymm11, %ymm11;
+	vpxor (12 * 32)(%r11), %ymm12, %ymm12;
+	vpxor (13 * 32)(%r11), %ymm13, %ymm13;
+	vpxor (14 * 32)(%r11), %ymm14, %ymm14;
+	vpxor (15 * 32)(%r11), %ymm15, %ymm15;
+	write_output(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %r10);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx2_ctr_crypt_32way)
+
+SYM_FUNC_START_LOCAL(__aria_aesni_avx2_gfni_crypt_32way)
+	/* input:
+	 *      %r9: rk
+	 *      %rsi: dst
+	 *      %rdx: src
+	 *      %ymm0..%ymm15: 16 byte-sliced blocks
+	 */
+
+	FRAME_BEGIN
+
+	movq %rsi, %rax;
+	leaq 8 * 32(%rax), %r8;
+
+	inpack16_post(%ymm0, %ymm1, %ymm2, %ymm3,
+		      %ymm4, %ymm5, %ymm6, %ymm7,
+		      %ymm8, %ymm9, %ymm10, %ymm11,
+		      %ymm12, %ymm13, %ymm14,
+		      %ymm15, %rax, %r8);
+	aria_fo_gfni(%ymm8, %ymm9, %ymm10, %ymm11,
+		     %ymm12, %ymm13, %ymm14, %ymm15,
+		     %ymm0, %ymm1, %ymm2, %ymm3,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %rax, %r9, 0);
+	aria_fe_gfni(%ymm1, %ymm0, %ymm3, %ymm2,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11,
+		     %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %r9, 1);
+	aria_fo_gfni(%ymm9, %ymm8, %ymm11, %ymm10,
+		     %ymm12, %ymm13, %ymm14, %ymm15,
+		     %ymm0, %ymm1, %ymm2, %ymm3,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %rax, %r9, 2);
+	aria_fe_gfni(%ymm1, %ymm0, %ymm3, %ymm2,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11,
+		     %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %r9, 3);
+	aria_fo_gfni(%ymm9, %ymm8, %ymm11, %ymm10,
+		     %ymm12, %ymm13, %ymm14, %ymm15,
+		     %ymm0, %ymm1, %ymm2, %ymm3,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %rax, %r9, 4);
+	aria_fe_gfni(%ymm1, %ymm0, %ymm3, %ymm2,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11,
+		     %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %r9, 5);
+	aria_fo_gfni(%ymm9, %ymm8, %ymm11, %ymm10,
+		     %ymm12, %ymm13, %ymm14, %ymm15,
+		     %ymm0, %ymm1, %ymm2, %ymm3,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %rax, %r9, 6);
+	aria_fe_gfni(%ymm1, %ymm0, %ymm3, %ymm2,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11,
+		     %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %r9, 7);
+	aria_fo_gfni(%ymm9, %ymm8, %ymm11, %ymm10,
+		     %ymm12, %ymm13, %ymm14, %ymm15,
+		     %ymm0, %ymm1, %ymm2, %ymm3,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %rax, %r9, 8);
+	aria_fe_gfni(%ymm1, %ymm0, %ymm3, %ymm2,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11,
+		     %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %r9, 9);
+	aria_fo_gfni(%ymm9, %ymm8, %ymm11, %ymm10,
+		     %ymm12, %ymm13, %ymm14, %ymm15,
+		     %ymm0, %ymm1, %ymm2, %ymm3,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %rax, %r9, 10);
+	cmpl $12, ARIA_CTX_rounds(CTX);
+	jne .Laria_gfni_192;
+	aria_ff_gfni(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		%ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		%ymm15, %rax, %r9, 11, 12);
+	jmp .Laria_gfni_end;
+.Laria_gfni_192:
+	aria_fe_gfni(%ymm1, %ymm0, %ymm3, %ymm2,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11,
+		     %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %r9, 11);
+	aria_fo_gfni(%ymm9, %ymm8, %ymm11, %ymm10,
+		     %ymm12, %ymm13, %ymm14, %ymm15,
+		     %ymm0, %ymm1, %ymm2, %ymm3,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %rax, %r9, 12);
+	cmpl $14, ARIA_CTX_rounds(CTX);
+	jne .Laria_gfni_256;
+	aria_ff_gfni(%ymm1, %ymm0, %ymm3, %ymm2,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11,
+		     %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %r9, 13, 14);
+	jmp .Laria_gfni_end;
+.Laria_gfni_256:
+	aria_fe_gfni(%ymm1, %ymm0, %ymm3, %ymm2,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11,
+		     %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %r9, 13);
+	aria_fo_gfni(%ymm9, %ymm8, %ymm11, %ymm10,
+		     %ymm12, %ymm13, %ymm14, %ymm15,
+		     %ymm0, %ymm1, %ymm2, %ymm3,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %rax, %r9, 14);
+	aria_ff_gfni(%ymm1, %ymm0, %ymm3, %ymm2,
+		     %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11,
+		     %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %r9, 15, 16);
+.Laria_gfni_end:
+	debyteslice_16x16b(%ymm8, %ymm12, %ymm1, %ymm4,
+			   %ymm9, %ymm13, %ymm0, %ymm5,
+			   %ymm10, %ymm14, %ymm3, %ymm6,
+			   %ymm11, %ymm15, %ymm2, %ymm7,
+			   (%rax), (%r8));
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(__aria_aesni_avx2_gfni_crypt_32way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx2_gfni_encrypt_32way)
+	/* input:
+	 *      %rdi: ctx, CTX
+	 *      %rsi: dst
+	 *      %rdx: src
+	 */
+
+	FRAME_BEGIN
+
+	leaq ARIA_CTX_enc_key(CTX), %r9;
+
+	inpack16_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rdx);
+
+	call __aria_aesni_avx2_gfni_crypt_32way;
+
+	write_output(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx2_gfni_encrypt_32way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx2_gfni_decrypt_32way)
+	/* input:
+	 *      %rdi: ctx, CTX
+	 *      %rsi: dst
+	 *      %rdx: src
+	 */
+
+	FRAME_BEGIN
+
+	leaq ARIA_CTX_dec_key(CTX), %r9;
+
+	inpack16_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rdx);
+
+	call __aria_aesni_avx2_gfni_crypt_32way;
+
+	write_output(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx2_gfni_decrypt_32way)
+
+SYM_TYPED_FUNC_START(aria_aesni_avx2_gfni_ctr_crypt_32way)
+	/* input:
+	 *      %rdi: ctx
+	 *      %rsi: dst
+	 *      %rdx: src
+	 *      %rcx: keystream
+	 *      %r8: iv (big endian, 128bit)
+	 */
+	FRAME_BEGIN
+
+	call __aria_aesni_avx2_ctr_gen_keystream_32way
+
+	leaq (%rsi), %r10;
+	leaq (%rdx), %r11;
+	leaq (%rcx), %rsi;
+	leaq (%rcx), %rdx;
+	leaq ARIA_CTX_enc_key(CTX), %r9;
+
+	call __aria_aesni_avx2_gfni_crypt_32way;
+
+	vpxor (0 * 32)(%r11), %ymm1, %ymm1;
+	vpxor (1 * 32)(%r11), %ymm0, %ymm0;
+	vpxor (2 * 32)(%r11), %ymm3, %ymm3;
+	vpxor (3 * 32)(%r11), %ymm2, %ymm2;
+	vpxor (4 * 32)(%r11), %ymm4, %ymm4;
+	vpxor (5 * 32)(%r11), %ymm5, %ymm5;
+	vpxor (6 * 32)(%r11), %ymm6, %ymm6;
+	vpxor (7 * 32)(%r11), %ymm7, %ymm7;
+	vpxor (8 * 32)(%r11), %ymm8, %ymm8;
+	vpxor (9 * 32)(%r11), %ymm9, %ymm9;
+	vpxor (10 * 32)(%r11), %ymm10, %ymm10;
+	vpxor (11 * 32)(%r11), %ymm11, %ymm11;
+	vpxor (12 * 32)(%r11), %ymm12, %ymm12;
+	vpxor (13 * 32)(%r11), %ymm13, %ymm13;
+	vpxor (14 * 32)(%r11), %ymm14, %ymm14;
+	vpxor (15 * 32)(%r11), %ymm15, %ymm15;
+	write_output(%ymm1, %ymm0, %ymm3, %ymm2, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %r10);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_aesni_avx2_gfni_ctr_crypt_32way)
diff --git a/arch/x86/crypto/aria-avx.h b/arch/x86/crypto/aria-avx.h
new file mode 100644
index 000000000..6e1b2d8a3
--- /dev/null
+++ b/arch/x86/crypto/aria-avx.h
@@ -0,0 +1,62 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#ifndef ASM_X86_ARIA_AVX_H
+#define ASM_X86_ARIA_AVX_H
+
+#include <linux/types.h>
+
+#define ARIA_AESNI_PARALLEL_BLOCKS 16
+#define ARIA_AESNI_PARALLEL_BLOCK_SIZE  (ARIA_BLOCK_SIZE * ARIA_AESNI_PARALLEL_BLOCKS)
+
+#define ARIA_AESNI_AVX2_PARALLEL_BLOCKS 32
+#define ARIA_AESNI_AVX2_PARALLEL_BLOCK_SIZE  (ARIA_BLOCK_SIZE * ARIA_AESNI_AVX2_PARALLEL_BLOCKS)
+
+#define ARIA_GFNI_AVX512_PARALLEL_BLOCKS 64
+#define ARIA_GFNI_AVX512_PARALLEL_BLOCK_SIZE  (ARIA_BLOCK_SIZE * ARIA_GFNI_AVX512_PARALLEL_BLOCKS)
+
+asmlinkage void aria_aesni_avx_encrypt_16way(const void *ctx, u8 *dst,
+					     const u8 *src);
+asmlinkage void aria_aesni_avx_decrypt_16way(const void *ctx, u8 *dst,
+					     const u8 *src);
+asmlinkage void aria_aesni_avx_ctr_crypt_16way(const void *ctx, u8 *dst,
+					       const u8 *src,
+					       u8 *keystream, u8 *iv);
+asmlinkage void aria_aesni_avx_gfni_encrypt_16way(const void *ctx, u8 *dst,
+						  const u8 *src);
+asmlinkage void aria_aesni_avx_gfni_decrypt_16way(const void *ctx, u8 *dst,
+						  const u8 *src);
+asmlinkage void aria_aesni_avx_gfni_ctr_crypt_16way(const void *ctx, u8 *dst,
+						    const u8 *src,
+						    u8 *keystream, u8 *iv);
+
+asmlinkage void aria_aesni_avx2_encrypt_32way(const void *ctx, u8 *dst,
+					      const u8 *src);
+asmlinkage void aria_aesni_avx2_decrypt_32way(const void *ctx, u8 *dst,
+					      const u8 *src);
+asmlinkage void aria_aesni_avx2_ctr_crypt_32way(const void *ctx, u8 *dst,
+						const u8 *src,
+						u8 *keystream, u8 *iv);
+asmlinkage void aria_aesni_avx2_gfni_encrypt_32way(const void *ctx, u8 *dst,
+						   const u8 *src);
+asmlinkage void aria_aesni_avx2_gfni_decrypt_32way(const void *ctx, u8 *dst,
+						   const u8 *src);
+asmlinkage void aria_aesni_avx2_gfni_ctr_crypt_32way(const void *ctx, u8 *dst,
+						     const u8 *src,
+						     u8 *keystream, u8 *iv);
+
+struct aria_avx_ops {
+	void (*aria_encrypt_16way)(const void *ctx, u8 *dst, const u8 *src);
+	void (*aria_decrypt_16way)(const void *ctx, u8 *dst, const u8 *src);
+	void (*aria_ctr_crypt_16way)(const void *ctx, u8 *dst, const u8 *src,
+				     u8 *keystream, u8 *iv);
+	void (*aria_encrypt_32way)(const void *ctx, u8 *dst, const u8 *src);
+	void (*aria_decrypt_32way)(const void *ctx, u8 *dst, const u8 *src);
+	void (*aria_ctr_crypt_32way)(const void *ctx, u8 *dst, const u8 *src,
+				     u8 *keystream, u8 *iv);
+	void (*aria_encrypt_64way)(const void *ctx, u8 *dst, const u8 *src);
+	void (*aria_decrypt_64way)(const void *ctx, u8 *dst, const u8 *src);
+	void (*aria_ctr_crypt_64way)(const void *ctx, u8 *dst, const u8 *src,
+				     u8 *keystream, u8 *iv);
+
+
+};
+#endif
diff --git a/arch/x86/crypto/aria-gfni-avx512-asm_64.S b/arch/x86/crypto/aria-gfni-avx512-asm_64.S
new file mode 100644
index 000000000..860887e5d
--- /dev/null
+++ b/arch/x86/crypto/aria-gfni-avx512-asm_64.S
@@ -0,0 +1,971 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * ARIA Cipher 64-way parallel algorithm (AVX512)
+ *
+ * Copyright (c) 2022 Taehee Yoo <ap420073@gmail.com>
+ *
+ */
+
+#include <linux/linkage.h>
+#include <asm/frame.h>
+#include <asm/asm-offsets.h>
+#include <linux/cfi_types.h>
+
+/* register macros */
+#define CTX %rdi
+
+
+#define BV8(a0, a1, a2, a3, a4, a5, a6, a7)		\
+	( (((a0) & 1) << 0) |				\
+	  (((a1) & 1) << 1) |				\
+	  (((a2) & 1) << 2) |				\
+	  (((a3) & 1) << 3) |				\
+	  (((a4) & 1) << 4) |				\
+	  (((a5) & 1) << 5) |				\
+	  (((a6) & 1) << 6) |				\
+	  (((a7) & 1) << 7) )
+
+#define BM8X8(l0, l1, l2, l3, l4, l5, l6, l7)		\
+	( ((l7) << (0 * 8)) |				\
+	  ((l6) << (1 * 8)) |				\
+	  ((l5) << (2 * 8)) |				\
+	  ((l4) << (3 * 8)) |				\
+	  ((l3) << (4 * 8)) |				\
+	  ((l2) << (5 * 8)) |				\
+	  ((l1) << (6 * 8)) |				\
+	  ((l0) << (7 * 8)) )
+
+#define add_le128(out, in, lo_counter, hi_counter1)	\
+	vpaddq lo_counter, in, out;			\
+	vpcmpuq $1, lo_counter, out, %k1;		\
+	kaddb %k1, %k1, %k1;				\
+	vpaddq hi_counter1, out, out{%k1};
+
+#define filter_8bit(x, lo_t, hi_t, mask4bit, tmp0)	\
+	vpandq x, mask4bit, tmp0;			\
+	vpandqn x, mask4bit, x;				\
+	vpsrld $4, x, x;				\
+							\
+	vpshufb tmp0, lo_t, tmp0;			\
+	vpshufb x, hi_t, x;				\
+	vpxorq tmp0, x, x;
+
+#define transpose_4x4(x0, x1, x2, x3, t1, t2)		\
+	vpunpckhdq x1, x0, t2;				\
+	vpunpckldq x1, x0, x0;				\
+							\
+	vpunpckldq x3, x2, t1;				\
+	vpunpckhdq x3, x2, x2;				\
+							\
+	vpunpckhqdq t1, x0, x1;				\
+	vpunpcklqdq t1, x0, x0;				\
+							\
+	vpunpckhqdq x2, t2, x3;				\
+	vpunpcklqdq x2, t2, x2;
+
+#define byteslice_16x16b(a0, b0, c0, d0,		\
+			 a1, b1, c1, d1,		\
+			 a2, b2, c2, d2,		\
+			 a3, b3, c3, d3,		\
+			 st0, st1)			\
+	vmovdqu64 d2, st0;				\
+	vmovdqu64 d3, st1;				\
+	transpose_4x4(a0, a1, a2, a3, d2, d3);		\
+	transpose_4x4(b0, b1, b2, b3, d2, d3);		\
+	vmovdqu64 st0, d2;				\
+	vmovdqu64 st1, d3;				\
+							\
+	vmovdqu64 a0, st0;				\
+	vmovdqu64 a1, st1;				\
+	transpose_4x4(c0, c1, c2, c3, a0, a1);		\
+	transpose_4x4(d0, d1, d2, d3, a0, a1);		\
+							\
+	vbroadcasti64x2 .Lshufb_16x16b(%rip), a0;	\
+	vmovdqu64 st1, a1;				\
+	vpshufb a0, a2, a2;				\
+	vpshufb a0, a3, a3;				\
+	vpshufb a0, b0, b0;				\
+	vpshufb a0, b1, b1;				\
+	vpshufb a0, b2, b2;				\
+	vpshufb a0, b3, b3;				\
+	vpshufb a0, a1, a1;				\
+	vpshufb a0, c0, c0;				\
+	vpshufb a0, c1, c1;				\
+	vpshufb a0, c2, c2;				\
+	vpshufb a0, c3, c3;				\
+	vpshufb a0, d0, d0;				\
+	vpshufb a0, d1, d1;				\
+	vpshufb a0, d2, d2;				\
+	vpshufb a0, d3, d3;				\
+	vmovdqu64 d3, st1;				\
+	vmovdqu64 st0, d3;				\
+	vpshufb a0, d3, a0;				\
+	vmovdqu64 d2, st0;				\
+							\
+	transpose_4x4(a0, b0, c0, d0, d2, d3);		\
+	transpose_4x4(a1, b1, c1, d1, d2, d3);		\
+	vmovdqu64 st0, d2;				\
+	vmovdqu64 st1, d3;				\
+							\
+	vmovdqu64 b0, st0;				\
+	vmovdqu64 b1, st1;				\
+	transpose_4x4(a2, b2, c2, d2, b0, b1);		\
+	transpose_4x4(a3, b3, c3, d3, b0, b1);		\
+	vmovdqu64 st0, b0;				\
+	vmovdqu64 st1, b1;				\
+	/* does not adjust output bytes inside vectors */
+
+#define debyteslice_16x16b(a0, b0, c0, d0,		\
+			   a1, b1, c1, d1,		\
+			   a2, b2, c2, d2,		\
+			   a3, b3, c3, d3,		\
+			   st0, st1)			\
+	vmovdqu64 d2, st0;				\
+	vmovdqu64 d3, st1;				\
+	transpose_4x4(a0, a1, a2, a3, d2, d3);		\
+	transpose_4x4(b0, b1, b2, b3, d2, d3);		\
+	vmovdqu64 st0, d2;				\
+	vmovdqu64 st1, d3;				\
+							\
+	vmovdqu64 a0, st0;				\
+	vmovdqu64 a1, st1;				\
+	transpose_4x4(c0, c1, c2, c3, a0, a1);		\
+	transpose_4x4(d0, d1, d2, d3, a0, a1);		\
+							\
+	vbroadcasti64x2 .Lshufb_16x16b(%rip), a0;	\
+	vmovdqu64 st1, a1;				\
+	vpshufb a0, a2, a2;				\
+	vpshufb a0, a3, a3;				\
+	vpshufb a0, b0, b0;				\
+	vpshufb a0, b1, b1;				\
+	vpshufb a0, b2, b2;				\
+	vpshufb a0, b3, b3;				\
+	vpshufb a0, a1, a1;				\
+	vpshufb a0, c0, c0;				\
+	vpshufb a0, c1, c1;				\
+	vpshufb a0, c2, c2;				\
+	vpshufb a0, c3, c3;				\
+	vpshufb a0, d0, d0;				\
+	vpshufb a0, d1, d1;				\
+	vpshufb a0, d2, d2;				\
+	vpshufb a0, d3, d3;				\
+	vmovdqu64 d3, st1;				\
+	vmovdqu64 st0, d3;				\
+	vpshufb a0, d3, a0;				\
+	vmovdqu64 d2, st0;				\
+							\
+	transpose_4x4(c0, d0, a0, b0, d2, d3);		\
+	transpose_4x4(c1, d1, a1, b1, d2, d3);		\
+	vmovdqu64 st0, d2;				\
+	vmovdqu64 st1, d3;				\
+							\
+	vmovdqu64 b0, st0;				\
+	vmovdqu64 b1, st1;				\
+	transpose_4x4(c2, d2, a2, b2, b0, b1);		\
+	transpose_4x4(c3, d3, a3, b3, b0, b1);		\
+	vmovdqu64 st0, b0;				\
+	vmovdqu64 st1, b1;				\
+	/* does not adjust output bytes inside vectors */
+
+/* load blocks to registers and apply pre-whitening */
+#define inpack16_pre(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     rio)				\
+	vmovdqu64 (0 * 64)(rio), x0;			\
+	vmovdqu64 (1 * 64)(rio), x1;			\
+	vmovdqu64 (2 * 64)(rio), x2;			\
+	vmovdqu64 (3 * 64)(rio), x3;			\
+	vmovdqu64 (4 * 64)(rio), x4;			\
+	vmovdqu64 (5 * 64)(rio), x5;			\
+	vmovdqu64 (6 * 64)(rio), x6;			\
+	vmovdqu64 (7 * 64)(rio), x7;			\
+	vmovdqu64 (8 * 64)(rio), y0;			\
+	vmovdqu64 (9 * 64)(rio), y1;			\
+	vmovdqu64 (10 * 64)(rio), y2;			\
+	vmovdqu64 (11 * 64)(rio), y3;			\
+	vmovdqu64 (12 * 64)(rio), y4;			\
+	vmovdqu64 (13 * 64)(rio), y5;			\
+	vmovdqu64 (14 * 64)(rio), y6;			\
+	vmovdqu64 (15 * 64)(rio), y7;
+
+/* byteslice pre-whitened blocks and store to temporary memory */
+#define inpack16_post(x0, x1, x2, x3,			\
+		      x4, x5, x6, x7,			\
+		      y0, y1, y2, y3,			\
+		      y4, y5, y6, y7,			\
+		      mem_ab, mem_cd)			\
+	byteslice_16x16b(x0, x1, x2, x3,		\
+			 x4, x5, x6, x7,		\
+			 y0, y1, y2, y3,		\
+			 y4, y5, y6, y7,		\
+			 (mem_ab), (mem_cd));		\
+							\
+	vmovdqu64 x0, 0 * 64(mem_ab);			\
+	vmovdqu64 x1, 1 * 64(mem_ab);			\
+	vmovdqu64 x2, 2 * 64(mem_ab);			\
+	vmovdqu64 x3, 3 * 64(mem_ab);			\
+	vmovdqu64 x4, 4 * 64(mem_ab);			\
+	vmovdqu64 x5, 5 * 64(mem_ab);			\
+	vmovdqu64 x6, 6 * 64(mem_ab);			\
+	vmovdqu64 x7, 7 * 64(mem_ab);			\
+	vmovdqu64 y0, 0 * 64(mem_cd);			\
+	vmovdqu64 y1, 1 * 64(mem_cd);			\
+	vmovdqu64 y2, 2 * 64(mem_cd);			\
+	vmovdqu64 y3, 3 * 64(mem_cd);			\
+	vmovdqu64 y4, 4 * 64(mem_cd);			\
+	vmovdqu64 y5, 5 * 64(mem_cd);			\
+	vmovdqu64 y6, 6 * 64(mem_cd);			\
+	vmovdqu64 y7, 7 * 64(mem_cd);
+
+#define write_output(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     mem)				\
+	vmovdqu64 x0, 0 * 64(mem);			\
+	vmovdqu64 x1, 1 * 64(mem);			\
+	vmovdqu64 x2, 2 * 64(mem);			\
+	vmovdqu64 x3, 3 * 64(mem);			\
+	vmovdqu64 x4, 4 * 64(mem);			\
+	vmovdqu64 x5, 5 * 64(mem);			\
+	vmovdqu64 x6, 6 * 64(mem);			\
+	vmovdqu64 x7, 7 * 64(mem);			\
+	vmovdqu64 y0, 8 * 64(mem);			\
+	vmovdqu64 y1, 9 * 64(mem);			\
+	vmovdqu64 y2, 10 * 64(mem);			\
+	vmovdqu64 y3, 11 * 64(mem);			\
+	vmovdqu64 y4, 12 * 64(mem);			\
+	vmovdqu64 y5, 13 * 64(mem);			\
+	vmovdqu64 y6, 14 * 64(mem);			\
+	vmovdqu64 y7, 15 * 64(mem);			\
+
+#define aria_store_state_8way(x0, x1, x2, x3,		\
+			      x4, x5, x6, x7,		\
+			      mem_tmp, idx)		\
+	vmovdqu64 x0, ((idx + 0) * 64)(mem_tmp);	\
+	vmovdqu64 x1, ((idx + 1) * 64)(mem_tmp);	\
+	vmovdqu64 x2, ((idx + 2) * 64)(mem_tmp);	\
+	vmovdqu64 x3, ((idx + 3) * 64)(mem_tmp);	\
+	vmovdqu64 x4, ((idx + 4) * 64)(mem_tmp);	\
+	vmovdqu64 x5, ((idx + 5) * 64)(mem_tmp);	\
+	vmovdqu64 x6, ((idx + 6) * 64)(mem_tmp);	\
+	vmovdqu64 x7, ((idx + 7) * 64)(mem_tmp);
+
+#define aria_load_state_8way(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     mem_tmp, idx)		\
+	vmovdqu64 ((idx + 0) * 64)(mem_tmp), x0;	\
+	vmovdqu64 ((idx + 1) * 64)(mem_tmp), x1;	\
+	vmovdqu64 ((idx + 2) * 64)(mem_tmp), x2;	\
+	vmovdqu64 ((idx + 3) * 64)(mem_tmp), x3;	\
+	vmovdqu64 ((idx + 4) * 64)(mem_tmp), x4;	\
+	vmovdqu64 ((idx + 5) * 64)(mem_tmp), x5;	\
+	vmovdqu64 ((idx + 6) * 64)(mem_tmp), x6;	\
+	vmovdqu64 ((idx + 7) * 64)(mem_tmp), x7;
+
+#define aria_ark_16way(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7,			\
+		       t0, rk, round)			\
+	/* AddRoundKey */                               \
+	vpbroadcastb ((round * 16) + 3)(rk), t0;	\
+	vpxorq t0, x0, x0;				\
+	vpbroadcastb ((round * 16) + 2)(rk), t0;	\
+	vpxorq t0, x1, x1;				\
+	vpbroadcastb ((round * 16) + 1)(rk), t0;	\
+	vpxorq t0, x2, x2;				\
+	vpbroadcastb ((round * 16) + 0)(rk), t0;	\
+	vpxorq t0, x3, x3;				\
+	vpbroadcastb ((round * 16) + 7)(rk), t0;	\
+	vpxorq t0, x4, x4;				\
+	vpbroadcastb ((round * 16) + 6)(rk), t0;	\
+	vpxorq t0, x5, x5;				\
+	vpbroadcastb ((round * 16) + 5)(rk), t0;	\
+	vpxorq t0, x6, x6;				\
+	vpbroadcastb ((round * 16) + 4)(rk), t0;	\
+	vpxorq t0, x7, x7;				\
+	vpbroadcastb ((round * 16) + 11)(rk), t0;	\
+	vpxorq t0, y0, y0;				\
+	vpbroadcastb ((round * 16) + 10)(rk), t0;	\
+	vpxorq t0, y1, y1;				\
+	vpbroadcastb ((round * 16) + 9)(rk), t0;	\
+	vpxorq t0, y2, y2;				\
+	vpbroadcastb ((round * 16) + 8)(rk), t0;	\
+	vpxorq t0, y3, y3;				\
+	vpbroadcastb ((round * 16) + 15)(rk), t0;	\
+	vpxorq t0, y4, y4;				\
+	vpbroadcastb ((round * 16) + 14)(rk), t0;	\
+	vpxorq t0, y5, y5;				\
+	vpbroadcastb ((round * 16) + 13)(rk), t0;	\
+	vpxorq t0, y6, y6;				\
+	vpbroadcastb ((round * 16) + 12)(rk), t0;	\
+	vpxorq t0, y7, y7;
+
+#define aria_sbox_8way_gfni(x0, x1, x2, x3,		\
+			    x4, x5, x6, x7,		\
+			    t0, t1, t2, t3,		\
+			    t4, t5, t6, t7)		\
+	vpbroadcastq .Ltf_s2_bitmatrix(%rip), t0;	\
+	vpbroadcastq .Ltf_inv_bitmatrix(%rip), t1;	\
+	vpbroadcastq .Ltf_id_bitmatrix(%rip), t2;	\
+	vpbroadcastq .Ltf_aff_bitmatrix(%rip), t3;	\
+	vpbroadcastq .Ltf_x2_bitmatrix(%rip), t4;	\
+	vgf2p8affineinvqb $(tf_s2_const), t0, x1, x1;	\
+	vgf2p8affineinvqb $(tf_s2_const), t0, x5, x5;	\
+	vgf2p8affineqb $(tf_inv_const), t1, x2, x2;	\
+	vgf2p8affineqb $(tf_inv_const), t1, x6, x6;	\
+	vgf2p8affineinvqb $0, t2, x2, x2;		\
+	vgf2p8affineinvqb $0, t2, x6, x6;		\
+	vgf2p8affineinvqb $(tf_aff_const), t3, x0, x0;	\
+	vgf2p8affineinvqb $(tf_aff_const), t3, x4, x4;	\
+	vgf2p8affineqb $(tf_x2_const), t4, x3, x3;	\
+	vgf2p8affineqb $(tf_x2_const), t4, x7, x7;	\
+	vgf2p8affineinvqb $0, t2, x3, x3;		\
+	vgf2p8affineinvqb $0, t2, x7, x7;
+
+#define aria_sbox_16way_gfni(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     t0, t1, t2, t3,		\
+			     t4, t5, t6, t7)		\
+	vpbroadcastq .Ltf_s2_bitmatrix(%rip), t0;	\
+	vpbroadcastq .Ltf_inv_bitmatrix(%rip), t1;	\
+	vpbroadcastq .Ltf_id_bitmatrix(%rip), t2;	\
+	vpbroadcastq .Ltf_aff_bitmatrix(%rip), t3;	\
+	vpbroadcastq .Ltf_x2_bitmatrix(%rip), t4;	\
+	vgf2p8affineinvqb $(tf_s2_const), t0, x1, x1;	\
+	vgf2p8affineinvqb $(tf_s2_const), t0, x5, x5;	\
+	vgf2p8affineqb $(tf_inv_const), t1, x2, x2;	\
+	vgf2p8affineqb $(tf_inv_const), t1, x6, x6;	\
+	vgf2p8affineinvqb $0, t2, x2, x2;		\
+	vgf2p8affineinvqb $0, t2, x6, x6;		\
+	vgf2p8affineinvqb $(tf_aff_const), t3, x0, x0;	\
+	vgf2p8affineinvqb $(tf_aff_const), t3, x4, x4;	\
+	vgf2p8affineqb $(tf_x2_const), t4, x3, x3;	\
+	vgf2p8affineqb $(tf_x2_const), t4, x7, x7;	\
+	vgf2p8affineinvqb $0, t2, x3, x3;		\
+	vgf2p8affineinvqb $0, t2, x7, x7;		\
+	vgf2p8affineinvqb $(tf_s2_const), t0, y1, y1;	\
+	vgf2p8affineinvqb $(tf_s2_const), t0, y5, y5;	\
+	vgf2p8affineqb $(tf_inv_const), t1, y2, y2;	\
+	vgf2p8affineqb $(tf_inv_const), t1, y6, y6;	\
+	vgf2p8affineinvqb $0, t2, y2, y2;		\
+	vgf2p8affineinvqb $0, t2, y6, y6;		\
+	vgf2p8affineinvqb $(tf_aff_const), t3, y0, y0;	\
+	vgf2p8affineinvqb $(tf_aff_const), t3, y4, y4;	\
+	vgf2p8affineqb $(tf_x2_const), t4, y3, y3;	\
+	vgf2p8affineqb $(tf_x2_const), t4, y7, y7;	\
+	vgf2p8affineinvqb $0, t2, y3, y3;		\
+	vgf2p8affineinvqb $0, t2, y7, y7;
+
+
+#define aria_diff_m(x0, x1, x2, x3,			\
+		    t0, t1, t2, t3)			\
+	/* T = rotr32(X, 8); */				\
+	/* X ^= T */					\
+	vpxorq x0, x3, t0;				\
+	vpxorq x1, x0, t1;				\
+	vpxorq x2, x1, t2;				\
+	vpxorq x3, x2, t3;				\
+	/* X = T ^ rotr(X, 16); */			\
+	vpxorq t2, x0, x0;				\
+	vpxorq x1, t3, t3;				\
+	vpxorq t0, x2, x2;				\
+	vpxorq t1, x3, x1;				\
+	vmovdqu64 t3, x3;
+
+#define aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7)			\
+	/* t1 ^= t2; */					\
+	vpxorq y0, x4, x4;				\
+	vpxorq y1, x5, x5;				\
+	vpxorq y2, x6, x6;				\
+	vpxorq y3, x7, x7;				\
+							\
+	/* t2 ^= t3; */					\
+	vpxorq y4, y0, y0;				\
+	vpxorq y5, y1, y1;				\
+	vpxorq y6, y2, y2;				\
+	vpxorq y7, y3, y3;				\
+							\
+	/* t0 ^= t1; */					\
+	vpxorq x4, x0, x0;				\
+	vpxorq x5, x1, x1;				\
+	vpxorq x6, x2, x2;				\
+	vpxorq x7, x3, x3;				\
+							\
+	/* t3 ^= t1; */					\
+	vpxorq x4, y4, y4;				\
+	vpxorq x5, y5, y5;				\
+	vpxorq x6, y6, y6;				\
+	vpxorq x7, y7, y7;				\
+							\
+	/* t2 ^= t0; */					\
+	vpxorq x0, y0, y0;				\
+	vpxorq x1, y1, y1;				\
+	vpxorq x2, y2, y2;				\
+	vpxorq x3, y3, y3;				\
+							\
+	/* t1 ^= t2; */					\
+	vpxorq y0, x4, x4;				\
+	vpxorq y1, x5, x5;				\
+	vpxorq y2, x6, x6;				\
+	vpxorq y3, x7, x7;
+
+#define aria_fe_gfni(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     z0, z1, z2, z3,			\
+		     z4, z5, z6, z7,			\
+		     mem_tmp, rk, round)		\
+	aria_ark_16way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7,	\
+		       z0, rk, round);			\
+							\
+	aria_sbox_16way_gfni(x2, x3, x0, x1,		\
+			     x6, x7, x4, x5,		\
+			     y2, y3, y0, y1,		\
+			     y6, y7, y4, y5,		\
+			     z0, z1, z2, z3,		\
+			     z4, z5, z6, z7);		\
+							\
+	aria_diff_m(x0, x1, x2, x3, z0, z1, z2, z3);	\
+	aria_diff_m(x4, x5, x6, x7, z0, z1, z2, z3);	\
+	aria_diff_m(y0, y1, y2, y3, z0, z1, z2, z3);	\
+	aria_diff_m(y4, y5, y6, y7, z0, z1, z2, z3);	\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7);			\
+	/* aria_diff_byte()				\
+	 * T3 = ABCD -> BADC				\
+	 * T3 = y4, y5, y6, y7 -> y5, y4, y7, y6	\
+	 * T0 = ABCD -> CDAB				\
+	 * T0 = x0, x1, x2, x3 -> x2, x3, x0, x1	\
+	 * T1 = ABCD -> DCBA				\
+	 * T1 = x4, x5, x6, x7 -> x7, x6, x5, x4	\
+	 */						\
+	aria_diff_word(x2, x3, x0, x1,			\
+		       x7, x6, x5, x4,			\
+		       y0, y1, y2, y3,			\
+		       y5, y4, y7, y6);			\
+
+
+#define aria_fo_gfni(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     z0, z1, z2, z3,			\
+		     z4, z5, z6, z7,			\
+		     mem_tmp, rk, round)		\
+	aria_ark_16way(x0, x1, x2, x3, x4, x5, x6, x7,	\
+		       y0, y1, y2, y3, y4, y5, y6, y7,	\
+		       z0, rk, round);			\
+							\
+	aria_sbox_16way_gfni(x0, x1, x2, x3,		\
+			     x4, x5, x6, x7,		\
+			     y0, y1, y2, y3,		\
+			     y4, y5, y6, y7,		\
+			     z0, z1, z2, z3,		\
+			     z4, z5, z6, z7);		\
+							\
+	aria_diff_m(x0, x1, x2, x3, z0, z1, z2, z3);	\
+	aria_diff_m(x4, x5, x6, x7, z0, z1, z2, z3);	\
+	aria_diff_m(y0, y1, y2, y3, z0, z1, z2, z3);	\
+	aria_diff_m(y4, y5, y6, y7, z0, z1, z2, z3);	\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7);			\
+	/* aria_diff_byte()				\
+	 * T1 = ABCD -> BADC				\
+	 * T1 = x4, x5, x6, x7 -> x5, x4, x7, x6	\
+	 * T2 = ABCD -> CDAB				\
+	 * T2 = y0, y1, y2, y3, -> y2, y3, y0, y1	\
+	 * T3 = ABCD -> DCBA				\
+	 * T3 = y4, y5, y6, y7 -> y7, y6, y5, y4	\
+	 */						\
+	aria_diff_word(x0, x1, x2, x3,			\
+		       x5, x4, x7, x6,			\
+		       y2, y3, y0, y1,			\
+		       y7, y6, y5, y4);
+
+#define aria_ff_gfni(x0, x1, x2, x3,			\
+		     x4, x5, x6, x7,			\
+		     y0, y1, y2, y3,			\
+		     y4, y5, y6, y7,			\
+		     z0, z1, z2, z3,			\
+		     z4, z5, z6, z7,			\
+		     mem_tmp, rk, round, last_round)	\
+	aria_ark_16way(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7,			\
+		       z0, rk, round);			\
+	aria_sbox_16way_gfni(x2, x3, x0, x1,		\
+			     x6, x7, x4, x5,		\
+			     y2, y3, y0, y1,		\
+			     y6, y7, y4, y5,		\
+			     z0, z1, z2, z3,		\
+			     z4, z5, z6, z7);		\
+	aria_ark_16way(x0, x1, x2, x3,			\
+		       x4, x5, x6, x7,			\
+		       y0, y1, y2, y3,			\
+		       y4, y5, y6, y7,			\
+		       z0, rk, last_round);
+
+
+.section        .rodata.cst64, "aM", @progbits, 64
+.align 64
+.Lcounter0123_lo:
+	.quad 0, 0
+	.quad 1, 0
+	.quad 2, 0
+	.quad 3, 0
+
+.section        .rodata.cst32.shufb_16x16b, "aM", @progbits, 32
+.align 32
+#define SHUFB_BYTES(idx) \
+	0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
+.Lshufb_16x16b:
+	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
+	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
+
+.section	.rodata.cst16, "aM", @progbits, 16
+.align 16
+
+.Lcounter4444_lo:
+	.quad 4, 0
+.Lcounter8888_lo:
+	.quad 8, 0
+.Lcounter16161616_lo:
+	.quad 16, 0
+.Lcounter1111_hi:
+	.quad 0, 1
+
+/* For CTR-mode IV byteswap */
+.Lbswap128_mask:
+	.byte 0x0f, 0x0e, 0x0d, 0x0c, 0x0b, 0x0a, 0x09, 0x08
+	.byte 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00
+
+.section	.rodata.cst8, "aM", @progbits, 8
+.align 8
+/* AES affine: */
+#define tf_aff_const BV8(1, 1, 0, 0, 0, 1, 1, 0)
+.Ltf_aff_bitmatrix:
+	.quad BM8X8(BV8(1, 0, 0, 0, 1, 1, 1, 1),
+		    BV8(1, 1, 0, 0, 0, 1, 1, 1),
+		    BV8(1, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 1, 0, 0, 0, 1),
+		    BV8(1, 1, 1, 1, 1, 0, 0, 0),
+		    BV8(0, 1, 1, 1, 1, 1, 0, 0),
+		    BV8(0, 0, 1, 1, 1, 1, 1, 0),
+		    BV8(0, 0, 0, 1, 1, 1, 1, 1))
+
+/* AES inverse affine: */
+#define tf_inv_const BV8(1, 0, 1, 0, 0, 0, 0, 0)
+.Ltf_inv_bitmatrix:
+	.quad BM8X8(BV8(0, 0, 1, 0, 0, 1, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 0, 1, 0),
+		    BV8(0, 1, 0, 0, 1, 0, 0, 1),
+		    BV8(1, 0, 1, 0, 0, 1, 0, 0),
+		    BV8(0, 1, 0, 1, 0, 0, 1, 0),
+		    BV8(0, 0, 1, 0, 1, 0, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 1, 0, 0),
+		    BV8(0, 1, 0, 0, 1, 0, 1, 0))
+
+/* S2: */
+#define tf_s2_const BV8(0, 1, 0, 0, 0, 1, 1, 1)
+.Ltf_s2_bitmatrix:
+	.quad BM8X8(BV8(0, 1, 0, 1, 0, 1, 1, 1),
+		    BV8(0, 0, 1, 1, 1, 1, 1, 1),
+		    BV8(1, 1, 1, 0, 1, 1, 0, 1),
+		    BV8(1, 1, 0, 0, 0, 0, 1, 1),
+		    BV8(0, 1, 0, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 0, 0, 1, 1, 1, 0),
+		    BV8(0, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 1, 0, 1, 1, 0))
+
+/* X2: */
+#define tf_x2_const BV8(0, 0, 1, 1, 0, 1, 0, 0)
+.Ltf_x2_bitmatrix:
+	.quad BM8X8(BV8(0, 0, 0, 1, 1, 0, 0, 0),
+		    BV8(0, 0, 1, 0, 0, 1, 1, 0),
+		    BV8(0, 0, 0, 0, 1, 0, 1, 0),
+		    BV8(1, 1, 1, 0, 0, 0, 1, 1),
+		    BV8(1, 1, 1, 0, 1, 1, 0, 0),
+		    BV8(0, 1, 1, 0, 1, 0, 1, 1),
+		    BV8(1, 0, 1, 1, 1, 1, 0, 1),
+		    BV8(1, 0, 0, 1, 0, 0, 1, 1))
+
+/* Identity matrix: */
+.Ltf_id_bitmatrix:
+	.quad BM8X8(BV8(1, 0, 0, 0, 0, 0, 0, 0),
+		    BV8(0, 1, 0, 0, 0, 0, 0, 0),
+		    BV8(0, 0, 1, 0, 0, 0, 0, 0),
+		    BV8(0, 0, 0, 1, 0, 0, 0, 0),
+		    BV8(0, 0, 0, 0, 1, 0, 0, 0),
+		    BV8(0, 0, 0, 0, 0, 1, 0, 0),
+		    BV8(0, 0, 0, 0, 0, 0, 1, 0),
+		    BV8(0, 0, 0, 0, 0, 0, 0, 1))
+
+.text
+SYM_FUNC_START_LOCAL(__aria_gfni_avx512_crypt_64way)
+	/* input:
+	 *      %r9: rk
+	 *      %rsi: dst
+	 *      %rdx: src
+	 *      %zmm0..%zmm15: byte-sliced blocks
+	 */
+
+	FRAME_BEGIN
+
+	movq %rsi, %rax;
+	leaq 8 * 64(%rax), %r8;
+
+	inpack16_post(%zmm0, %zmm1, %zmm2, %zmm3,
+		      %zmm4, %zmm5, %zmm6, %zmm7,
+		      %zmm8, %zmm9, %zmm10, %zmm11,
+		      %zmm12, %zmm13, %zmm14,
+		      %zmm15, %rax, %r8);
+	aria_fo_gfni(%zmm0, %zmm1, %zmm2, %zmm3,
+		     %zmm4, %zmm5, %zmm6, %zmm7,
+		     %zmm8, %zmm9, %zmm10, %zmm11,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 0);
+	aria_fe_gfni(%zmm3, %zmm2, %zmm1, %zmm0,
+		     %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 1);
+	aria_fo_gfni(%zmm0, %zmm1, %zmm2, %zmm3,
+		     %zmm4, %zmm5, %zmm6, %zmm7,
+		     %zmm8, %zmm9, %zmm10, %zmm11,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 2);
+	aria_fe_gfni(%zmm3, %zmm2, %zmm1, %zmm0,
+		     %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 3);
+	aria_fo_gfni(%zmm0, %zmm1, %zmm2, %zmm3,
+		     %zmm4, %zmm5, %zmm6, %zmm7,
+		     %zmm8, %zmm9, %zmm10, %zmm11,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 4);
+	aria_fe_gfni(%zmm3, %zmm2, %zmm1, %zmm0,
+		     %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 5);
+	aria_fo_gfni(%zmm0, %zmm1, %zmm2, %zmm3,
+		     %zmm4, %zmm5, %zmm6, %zmm7,
+		     %zmm8, %zmm9, %zmm10, %zmm11,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 6);
+	aria_fe_gfni(%zmm3, %zmm2, %zmm1, %zmm0,
+		     %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 7);
+	aria_fo_gfni(%zmm0, %zmm1, %zmm2, %zmm3,
+		     %zmm4, %zmm5, %zmm6, %zmm7,
+		     %zmm8, %zmm9, %zmm10, %zmm11,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 8);
+	aria_fe_gfni(%zmm3, %zmm2, %zmm1, %zmm0,
+		     %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 9);
+	aria_fo_gfni(%zmm0, %zmm1, %zmm2, %zmm3,
+		     %zmm4, %zmm5, %zmm6, %zmm7,
+		     %zmm8, %zmm9, %zmm10, %zmm11,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 10);
+	cmpl $12, ARIA_CTX_rounds(CTX);
+	jne .Laria_gfni_192;
+	aria_ff_gfni(%zmm3, %zmm2, %zmm1, %zmm0,
+		     %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 11, 12);
+	jmp .Laria_gfni_end;
+.Laria_gfni_192:
+	aria_fe_gfni(%zmm3, %zmm2, %zmm1, %zmm0,
+		     %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 11);
+	aria_fo_gfni(%zmm0, %zmm1, %zmm2, %zmm3,
+		     %zmm4, %zmm5, %zmm6, %zmm7,
+		     %zmm8, %zmm9, %zmm10, %zmm11,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 12);
+	cmpl $14, ARIA_CTX_rounds(CTX);
+	jne .Laria_gfni_256;
+	aria_ff_gfni(%zmm3, %zmm2, %zmm1, %zmm0,
+		     %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 13, 14);
+	jmp .Laria_gfni_end;
+.Laria_gfni_256:
+	aria_fe_gfni(%zmm3, %zmm2, %zmm1, %zmm0,
+		     %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 13);
+	aria_fo_gfni(%zmm0, %zmm1, %zmm2, %zmm3,
+		     %zmm4, %zmm5, %zmm6, %zmm7,
+		     %zmm8, %zmm9, %zmm10, %zmm11,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 14);
+	aria_ff_gfni(%zmm3, %zmm2, %zmm1, %zmm0,
+		     %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10,
+		     %zmm12, %zmm13, %zmm14, %zmm15,
+		     %zmm24, %zmm25, %zmm26, %zmm27,
+		     %zmm28, %zmm29, %zmm30, %zmm31,
+		     %rax, %r9, 15, 16);
+.Laria_gfni_end:
+	debyteslice_16x16b(%zmm9, %zmm12, %zmm3, %zmm6,
+			   %zmm8, %zmm13, %zmm2, %zmm7,
+			   %zmm11, %zmm14, %zmm1, %zmm4,
+			   %zmm10, %zmm15, %zmm0, %zmm5,
+			   (%rax), (%r8));
+	FRAME_END
+	RET;
+SYM_FUNC_END(__aria_gfni_avx512_crypt_64way)
+
+SYM_TYPED_FUNC_START(aria_gfni_avx512_encrypt_64way)
+	/* input:
+	 *      %rdi: ctx, CTX
+	 *      %rsi: dst
+	 *      %rdx: src
+	 */
+
+	FRAME_BEGIN
+
+	leaq ARIA_CTX_enc_key(CTX), %r9;
+
+	inpack16_pre(%zmm0, %zmm1, %zmm2, %zmm3, %zmm4, %zmm5, %zmm6, %zmm7,
+		     %zmm8, %zmm9, %zmm10, %zmm11, %zmm12, %zmm13, %zmm14,
+		     %zmm15, %rdx);
+
+	call __aria_gfni_avx512_crypt_64way;
+
+	write_output(%zmm3, %zmm2, %zmm1, %zmm0, %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10, %zmm12, %zmm13, %zmm14,
+		     %zmm15, %rax);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_gfni_avx512_encrypt_64way)
+
+SYM_TYPED_FUNC_START(aria_gfni_avx512_decrypt_64way)
+	/* input:
+	 *      %rdi: ctx, CTX
+	 *      %rsi: dst
+	 *      %rdx: src
+	 */
+
+	FRAME_BEGIN
+
+	leaq ARIA_CTX_dec_key(CTX), %r9;
+
+	inpack16_pre(%zmm0, %zmm1, %zmm2, %zmm3, %zmm4, %zmm5, %zmm6, %zmm7,
+		     %zmm8, %zmm9, %zmm10, %zmm11, %zmm12, %zmm13, %zmm14,
+		     %zmm15, %rdx);
+
+	call __aria_gfni_avx512_crypt_64way;
+
+	write_output(%zmm3, %zmm2, %zmm1, %zmm0, %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10, %zmm12, %zmm13, %zmm14,
+		     %zmm15, %rax);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_gfni_avx512_decrypt_64way)
+
+SYM_FUNC_START_LOCAL(__aria_gfni_avx512_ctr_gen_keystream_64way)
+	/* input:
+	 *      %rdi: ctx
+	 *      %rsi: dst
+	 *      %rdx: src
+	 *      %rcx: keystream
+	 *      %r8: iv (big endian, 128bit)
+	 */
+
+	FRAME_BEGIN
+
+	vbroadcasti64x2 .Lbswap128_mask (%rip), %zmm19;
+	vmovdqa64 .Lcounter0123_lo (%rip), %zmm21;
+	vbroadcasti64x2 .Lcounter4444_lo (%rip), %zmm22;
+	vbroadcasti64x2 .Lcounter8888_lo (%rip), %zmm23;
+	vbroadcasti64x2 .Lcounter16161616_lo (%rip), %zmm24;
+	vbroadcasti64x2 .Lcounter1111_hi (%rip), %zmm25;
+
+	/* load IV and byteswap */
+	movq 8(%r8), %r11;
+	movq (%r8), %r10;
+	bswapq %r11;
+	bswapq %r10;
+	vbroadcasti64x2 (%r8), %zmm20;
+	vpshufb %zmm19, %zmm20, %zmm20;
+
+	/* check need for handling 64-bit overflow and carry */
+	cmpq $(0xffffffffffffffff - 64), %r11;
+	ja .Lload_ctr_carry;
+
+	/* construct IVs */
+	vpaddq %zmm21, %zmm20, %zmm0;  /* +0:+1:+2:+3 */
+	vpaddq %zmm22, %zmm0, %zmm1; /* +4:+5:+6:+7 */
+	vpaddq %zmm23, %zmm0, %zmm2; /* +8:+9:+10:+11 */
+	vpaddq %zmm23, %zmm1, %zmm3; /* +12:+13:+14:+15 */
+	vpaddq %zmm24, %zmm0, %zmm4; /* +16... */
+	vpaddq %zmm24, %zmm1, %zmm5; /* +20... */
+	vpaddq %zmm24, %zmm2, %zmm6; /* +24... */
+	vpaddq %zmm24, %zmm3, %zmm7; /* +28... */
+	vpaddq %zmm24, %zmm4, %zmm8; /* +32... */
+	vpaddq %zmm24, %zmm5, %zmm9; /* +36... */
+	vpaddq %zmm24, %zmm6, %zmm10; /* +40... */
+	vpaddq %zmm24, %zmm7, %zmm11; /* +44... */
+	vpaddq %zmm24, %zmm8, %zmm12; /* +48... */
+	vpaddq %zmm24, %zmm9, %zmm13; /* +52... */
+	vpaddq %zmm24, %zmm10, %zmm14; /* +56... */
+	vpaddq %zmm24, %zmm11, %zmm15; /* +60... */
+	jmp .Lload_ctr_done;
+
+.Lload_ctr_carry:
+	/* construct IVs */
+	add_le128(%zmm0, %zmm20, %zmm21, %zmm25);  /* +0:+1:+2:+3 */
+	add_le128(%zmm1, %zmm0, %zmm22, %zmm25); /* +4:+5:+6:+7 */
+	add_le128(%zmm2, %zmm0, %zmm23, %zmm25); /* +8:+9:+10:+11 */
+	add_le128(%zmm3, %zmm1, %zmm23, %zmm25); /* +12:+13:+14:+15 */
+	add_le128(%zmm4, %zmm0, %zmm24, %zmm25); /* +16... */
+	add_le128(%zmm5, %zmm1, %zmm24, %zmm25); /* +20... */
+	add_le128(%zmm6, %zmm2, %zmm24, %zmm25); /* +24... */
+	add_le128(%zmm7, %zmm3, %zmm24, %zmm25); /* +28... */
+	add_le128(%zmm8, %zmm4, %zmm24, %zmm25); /* +32... */
+	add_le128(%zmm9, %zmm5, %zmm24, %zmm25); /* +36... */
+	add_le128(%zmm10, %zmm6, %zmm24, %zmm25); /* +40... */
+	add_le128(%zmm11, %zmm7, %zmm24, %zmm25); /* +44... */
+	add_le128(%zmm12, %zmm8, %zmm24, %zmm25); /* +48... */
+	add_le128(%zmm13, %zmm9, %zmm24, %zmm25); /* +52... */
+	add_le128(%zmm14, %zmm10, %zmm24, %zmm25); /* +56... */
+	add_le128(%zmm15, %zmm11, %zmm24, %zmm25); /* +60... */
+
+.Lload_ctr_done:
+	/* Byte-swap IVs and update counter. */
+	addq $64, %r11;
+	adcq $0, %r10;
+	vpshufb %zmm19, %zmm15, %zmm15;
+	vpshufb %zmm19, %zmm14, %zmm14;
+	vpshufb %zmm19, %zmm13, %zmm13;
+	vpshufb %zmm19, %zmm12, %zmm12;
+	vpshufb %zmm19, %zmm11, %zmm11;
+	vpshufb %zmm19, %zmm10, %zmm10;
+	vpshufb %zmm19, %zmm9, %zmm9;
+	vpshufb %zmm19, %zmm8, %zmm8;
+	bswapq %r11;
+	bswapq %r10;
+	vpshufb %zmm19, %zmm7, %zmm7;
+	vpshufb %zmm19, %zmm6, %zmm6;
+	vpshufb %zmm19, %zmm5, %zmm5;
+	vpshufb %zmm19, %zmm4, %zmm4;
+	vpshufb %zmm19, %zmm3, %zmm3;
+	vpshufb %zmm19, %zmm2, %zmm2;
+	vpshufb %zmm19, %zmm1, %zmm1;
+	vpshufb %zmm19, %zmm0, %zmm0;
+	movq %r11, 8(%r8);
+	movq %r10, (%r8);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(__aria_gfni_avx512_ctr_gen_keystream_64way)
+
+SYM_TYPED_FUNC_START(aria_gfni_avx512_ctr_crypt_64way)
+	/* input:
+	 *      %rdi: ctx
+	 *      %rsi: dst
+	 *      %rdx: src
+	 *      %rcx: keystream
+	 *      %r8: iv (big endian, 128bit)
+	 */
+	FRAME_BEGIN
+
+	call __aria_gfni_avx512_ctr_gen_keystream_64way
+
+	leaq (%rsi), %r10;
+	leaq (%rdx), %r11;
+	leaq (%rcx), %rsi;
+	leaq (%rcx), %rdx;
+	leaq ARIA_CTX_enc_key(CTX), %r9;
+
+	call __aria_gfni_avx512_crypt_64way;
+
+	vpxorq (0 * 64)(%r11), %zmm3, %zmm3;
+	vpxorq (1 * 64)(%r11), %zmm2, %zmm2;
+	vpxorq (2 * 64)(%r11), %zmm1, %zmm1;
+	vpxorq (3 * 64)(%r11), %zmm0, %zmm0;
+	vpxorq (4 * 64)(%r11), %zmm6, %zmm6;
+	vpxorq (5 * 64)(%r11), %zmm7, %zmm7;
+	vpxorq (6 * 64)(%r11), %zmm4, %zmm4;
+	vpxorq (7 * 64)(%r11), %zmm5, %zmm5;
+	vpxorq (8 * 64)(%r11), %zmm9, %zmm9;
+	vpxorq (9 * 64)(%r11), %zmm8, %zmm8;
+	vpxorq (10 * 64)(%r11), %zmm11, %zmm11;
+	vpxorq (11 * 64)(%r11), %zmm10, %zmm10;
+	vpxorq (12 * 64)(%r11), %zmm12, %zmm12;
+	vpxorq (13 * 64)(%r11), %zmm13, %zmm13;
+	vpxorq (14 * 64)(%r11), %zmm14, %zmm14;
+	vpxorq (15 * 64)(%r11), %zmm15, %zmm15;
+	write_output(%zmm3, %zmm2, %zmm1, %zmm0, %zmm6, %zmm7, %zmm4, %zmm5,
+		     %zmm9, %zmm8, %zmm11, %zmm10, %zmm12, %zmm13, %zmm14,
+		     %zmm15, %r10);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(aria_gfni_avx512_ctr_crypt_64way)
diff --git a/arch/x86/crypto/aria_aesni_avx2_glue.c b/arch/x86/crypto/aria_aesni_avx2_glue.c
new file mode 100644
index 000000000..1487a49bf
--- /dev/null
+++ b/arch/x86/crypto/aria_aesni_avx2_glue.c
@@ -0,0 +1,245 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Glue Code for the AVX2/AES-NI/GFNI assembler implementation of the ARIA Cipher
+ *
+ * Copyright (c) 2022 Taehee Yoo <ap420073@gmail.com>
+ */
+
+#include <crypto/algapi.h>
+#include <crypto/aria.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <linux/export.h>
+#include <linux/module.h>
+#include <linux/types.h>
+
+#include "ecb_cbc_helpers.h"
+#include "aria-avx.h"
+
+asmlinkage void aria_aesni_avx2_encrypt_32way(const void *ctx, u8 *dst,
+					      const u8 *src);
+EXPORT_SYMBOL_GPL(aria_aesni_avx2_encrypt_32way);
+asmlinkage void aria_aesni_avx2_decrypt_32way(const void *ctx, u8 *dst,
+					      const u8 *src);
+EXPORT_SYMBOL_GPL(aria_aesni_avx2_decrypt_32way);
+asmlinkage void aria_aesni_avx2_ctr_crypt_32way(const void *ctx, u8 *dst,
+						const u8 *src,
+						u8 *keystream, u8 *iv);
+EXPORT_SYMBOL_GPL(aria_aesni_avx2_ctr_crypt_32way);
+asmlinkage void aria_aesni_avx2_gfni_encrypt_32way(const void *ctx, u8 *dst,
+						   const u8 *src);
+EXPORT_SYMBOL_GPL(aria_aesni_avx2_gfni_encrypt_32way);
+asmlinkage void aria_aesni_avx2_gfni_decrypt_32way(const void *ctx, u8 *dst,
+						   const u8 *src);
+EXPORT_SYMBOL_GPL(aria_aesni_avx2_gfni_decrypt_32way);
+asmlinkage void aria_aesni_avx2_gfni_ctr_crypt_32way(const void *ctx, u8 *dst,
+						     const u8 *src,
+						     u8 *keystream, u8 *iv);
+EXPORT_SYMBOL_GPL(aria_aesni_avx2_gfni_ctr_crypt_32way);
+
+static struct aria_avx_ops aria_ops;
+
+struct aria_avx2_request_ctx {
+	u8 keystream[ARIA_AESNI_AVX2_PARALLEL_BLOCK_SIZE];
+};
+
+static int ecb_do_encrypt(struct skcipher_request *req, const u32 *rkey)
+{
+	ECB_WALK_START(req, ARIA_BLOCK_SIZE, ARIA_AESNI_PARALLEL_BLOCKS);
+	ECB_BLOCK(ARIA_AESNI_AVX2_PARALLEL_BLOCKS, aria_ops.aria_encrypt_32way);
+	ECB_BLOCK(ARIA_AESNI_PARALLEL_BLOCKS, aria_ops.aria_encrypt_16way);
+	ECB_BLOCK(1, aria_encrypt);
+	ECB_WALK_END();
+}
+
+static int ecb_do_decrypt(struct skcipher_request *req, const u32 *rkey)
+{
+	ECB_WALK_START(req, ARIA_BLOCK_SIZE, ARIA_AESNI_PARALLEL_BLOCKS);
+	ECB_BLOCK(ARIA_AESNI_AVX2_PARALLEL_BLOCKS, aria_ops.aria_decrypt_32way);
+	ECB_BLOCK(ARIA_AESNI_PARALLEL_BLOCKS, aria_ops.aria_decrypt_16way);
+	ECB_BLOCK(1, aria_decrypt);
+	ECB_WALK_END();
+}
+
+static int aria_avx2_ecb_encrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct aria_ctx *ctx = crypto_skcipher_ctx(tfm);
+
+	return ecb_do_encrypt(req, ctx->enc_key[0]);
+}
+
+static int aria_avx2_ecb_decrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct aria_ctx *ctx = crypto_skcipher_ctx(tfm);
+
+	return ecb_do_decrypt(req, ctx->dec_key[0]);
+}
+
+static int aria_avx2_set_key(struct crypto_skcipher *tfm, const u8 *key,
+			    unsigned int keylen)
+{
+	return aria_set_key(&tfm->base, key, keylen);
+}
+
+static int aria_avx2_ctr_encrypt(struct skcipher_request *req)
+{
+	struct aria_avx2_request_ctx *req_ctx = skcipher_request_ctx(req);
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct aria_ctx *ctx = crypto_skcipher_ctx(tfm);
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes) > 0) {
+		const u8 *src = walk.src.virt.addr;
+		u8 *dst = walk.dst.virt.addr;
+
+		while (nbytes >= ARIA_AESNI_AVX2_PARALLEL_BLOCK_SIZE) {
+			kernel_fpu_begin();
+			aria_ops.aria_ctr_crypt_32way(ctx, dst, src,
+						      &req_ctx->keystream[0],
+						      walk.iv);
+			kernel_fpu_end();
+			dst += ARIA_AESNI_AVX2_PARALLEL_BLOCK_SIZE;
+			src += ARIA_AESNI_AVX2_PARALLEL_BLOCK_SIZE;
+			nbytes -= ARIA_AESNI_AVX2_PARALLEL_BLOCK_SIZE;
+		}
+
+		while (nbytes >= ARIA_AESNI_PARALLEL_BLOCK_SIZE) {
+			kernel_fpu_begin();
+			aria_ops.aria_ctr_crypt_16way(ctx, dst, src,
+						      &req_ctx->keystream[0],
+						      walk.iv);
+			kernel_fpu_end();
+			dst += ARIA_AESNI_PARALLEL_BLOCK_SIZE;
+			src += ARIA_AESNI_PARALLEL_BLOCK_SIZE;
+			nbytes -= ARIA_AESNI_PARALLEL_BLOCK_SIZE;
+		}
+
+		while (nbytes >= ARIA_BLOCK_SIZE) {
+			memcpy(&req_ctx->keystream[0], walk.iv, ARIA_BLOCK_SIZE);
+			crypto_inc(walk.iv, ARIA_BLOCK_SIZE);
+
+			aria_encrypt(ctx, &req_ctx->keystream[0],
+				     &req_ctx->keystream[0]);
+
+			crypto_xor_cpy(dst, src, &req_ctx->keystream[0],
+				       ARIA_BLOCK_SIZE);
+			dst += ARIA_BLOCK_SIZE;
+			src += ARIA_BLOCK_SIZE;
+			nbytes -= ARIA_BLOCK_SIZE;
+		}
+
+		if (walk.nbytes == walk.total && nbytes > 0) {
+			memcpy(&req_ctx->keystream[0], walk.iv,
+			       ARIA_BLOCK_SIZE);
+			crypto_inc(walk.iv, ARIA_BLOCK_SIZE);
+
+			aria_encrypt(ctx, &req_ctx->keystream[0],
+				     &req_ctx->keystream[0]);
+
+			crypto_xor_cpy(dst, src, &req_ctx->keystream[0],
+				       nbytes);
+			dst += nbytes;
+			src += nbytes;
+			nbytes = 0;
+		}
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	return err;
+}
+
+static int aria_avx2_init_tfm(struct crypto_skcipher *tfm)
+{
+	crypto_skcipher_set_reqsize(tfm, sizeof(struct aria_avx2_request_ctx));
+
+	return 0;
+}
+
+static struct skcipher_alg aria_algs[] = {
+	{
+		.base.cra_name		= "ecb(aria)",
+		.base.cra_driver_name	= "ecb-aria-avx2",
+		.base.cra_priority	= 500,
+		.base.cra_blocksize	= ARIA_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct aria_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= ARIA_MIN_KEY_SIZE,
+		.max_keysize		= ARIA_MAX_KEY_SIZE,
+		.setkey			= aria_avx2_set_key,
+		.encrypt		= aria_avx2_ecb_encrypt,
+		.decrypt		= aria_avx2_ecb_decrypt,
+	}, {
+		.base.cra_name		= "ctr(aria)",
+		.base.cra_driver_name	= "ctr-aria-avx2",
+		.base.cra_priority	= 500,
+		.base.cra_flags		= CRYPTO_ALG_SKCIPHER_REQSIZE_LARGE,
+		.base.cra_blocksize	= 1,
+		.base.cra_ctxsize	= sizeof(struct aria_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= ARIA_MIN_KEY_SIZE,
+		.max_keysize		= ARIA_MAX_KEY_SIZE,
+		.ivsize			= ARIA_BLOCK_SIZE,
+		.chunksize		= ARIA_BLOCK_SIZE,
+		.setkey			= aria_avx2_set_key,
+		.encrypt		= aria_avx2_ctr_encrypt,
+		.decrypt		= aria_avx2_ctr_encrypt,
+		.init                   = aria_avx2_init_tfm,
+	}
+};
+
+static int __init aria_avx2_init(void)
+{
+	const char *feature_name;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX) ||
+	    !boot_cpu_has(X86_FEATURE_AVX2) ||
+	    !boot_cpu_has(X86_FEATURE_AES) ||
+	    !boot_cpu_has(X86_FEATURE_OSXSAVE)) {
+		pr_info("AVX2 or AES-NI instructions are not detected.\n");
+		return -ENODEV;
+	}
+
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM,
+				&feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	if (boot_cpu_has(X86_FEATURE_GFNI)) {
+		aria_ops.aria_encrypt_16way = aria_aesni_avx_gfni_encrypt_16way;
+		aria_ops.aria_decrypt_16way = aria_aesni_avx_gfni_decrypt_16way;
+		aria_ops.aria_ctr_crypt_16way = aria_aesni_avx_gfni_ctr_crypt_16way;
+		aria_ops.aria_encrypt_32way = aria_aesni_avx2_gfni_encrypt_32way;
+		aria_ops.aria_decrypt_32way = aria_aesni_avx2_gfni_decrypt_32way;
+		aria_ops.aria_ctr_crypt_32way = aria_aesni_avx2_gfni_ctr_crypt_32way;
+	} else {
+		aria_ops.aria_encrypt_16way = aria_aesni_avx_encrypt_16way;
+		aria_ops.aria_decrypt_16way = aria_aesni_avx_decrypt_16way;
+		aria_ops.aria_ctr_crypt_16way = aria_aesni_avx_ctr_crypt_16way;
+		aria_ops.aria_encrypt_32way = aria_aesni_avx2_encrypt_32way;
+		aria_ops.aria_decrypt_32way = aria_aesni_avx2_decrypt_32way;
+		aria_ops.aria_ctr_crypt_32way = aria_aesni_avx2_ctr_crypt_32way;
+	}
+
+	return crypto_register_skciphers(aria_algs, ARRAY_SIZE(aria_algs));
+}
+
+static void __exit aria_avx2_exit(void)
+{
+	crypto_unregister_skciphers(aria_algs, ARRAY_SIZE(aria_algs));
+}
+
+module_init(aria_avx2_init);
+module_exit(aria_avx2_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Taehee Yoo <ap420073@gmail.com>");
+MODULE_DESCRIPTION("ARIA Cipher Algorithm, AVX2/AES-NI/GFNI optimized");
+MODULE_ALIAS_CRYPTO("aria");
+MODULE_ALIAS_CRYPTO("aria-aesni-avx2");
diff --git a/arch/x86/crypto/aria_aesni_avx_glue.c b/arch/x86/crypto/aria_aesni_avx_glue.c
new file mode 100644
index 000000000..e4e3d7891
--- /dev/null
+++ b/arch/x86/crypto/aria_aesni_avx_glue.c
@@ -0,0 +1,225 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Glue Code for the AVX/AES-NI/GFNI assembler implementation of the ARIA Cipher
+ *
+ * Copyright (c) 2022 Taehee Yoo <ap420073@gmail.com>
+ */
+
+#include <crypto/algapi.h>
+#include <crypto/aria.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <linux/export.h>
+#include <linux/module.h>
+#include <linux/types.h>
+
+#include "ecb_cbc_helpers.h"
+#include "aria-avx.h"
+
+asmlinkage void aria_aesni_avx_encrypt_16way(const void *ctx, u8 *dst,
+					     const u8 *src);
+EXPORT_SYMBOL_GPL(aria_aesni_avx_encrypt_16way);
+asmlinkage void aria_aesni_avx_decrypt_16way(const void *ctx, u8 *dst,
+					     const u8 *src);
+EXPORT_SYMBOL_GPL(aria_aesni_avx_decrypt_16way);
+asmlinkage void aria_aesni_avx_ctr_crypt_16way(const void *ctx, u8 *dst,
+					       const u8 *src,
+					       u8 *keystream, u8 *iv);
+EXPORT_SYMBOL_GPL(aria_aesni_avx_ctr_crypt_16way);
+asmlinkage void aria_aesni_avx_gfni_encrypt_16way(const void *ctx, u8 *dst,
+						  const u8 *src);
+EXPORT_SYMBOL_GPL(aria_aesni_avx_gfni_encrypt_16way);
+asmlinkage void aria_aesni_avx_gfni_decrypt_16way(const void *ctx, u8 *dst,
+						  const u8 *src);
+EXPORT_SYMBOL_GPL(aria_aesni_avx_gfni_decrypt_16way);
+asmlinkage void aria_aesni_avx_gfni_ctr_crypt_16way(const void *ctx, u8 *dst,
+						    const u8 *src,
+						    u8 *keystream, u8 *iv);
+EXPORT_SYMBOL_GPL(aria_aesni_avx_gfni_ctr_crypt_16way);
+
+static struct aria_avx_ops aria_ops;
+
+struct aria_avx_request_ctx {
+	u8 keystream[ARIA_AESNI_PARALLEL_BLOCK_SIZE];
+};
+
+static int ecb_do_encrypt(struct skcipher_request *req, const u32 *rkey)
+{
+	ECB_WALK_START(req, ARIA_BLOCK_SIZE, ARIA_AESNI_PARALLEL_BLOCKS);
+	ECB_BLOCK(ARIA_AESNI_PARALLEL_BLOCKS, aria_ops.aria_encrypt_16way);
+	ECB_BLOCK(1, aria_encrypt);
+	ECB_WALK_END();
+}
+
+static int ecb_do_decrypt(struct skcipher_request *req, const u32 *rkey)
+{
+	ECB_WALK_START(req, ARIA_BLOCK_SIZE, ARIA_AESNI_PARALLEL_BLOCKS);
+	ECB_BLOCK(ARIA_AESNI_PARALLEL_BLOCKS, aria_ops.aria_decrypt_16way);
+	ECB_BLOCK(1, aria_decrypt);
+	ECB_WALK_END();
+}
+
+static int aria_avx_ecb_encrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct aria_ctx *ctx = crypto_skcipher_ctx(tfm);
+
+	return ecb_do_encrypt(req, ctx->enc_key[0]);
+}
+
+static int aria_avx_ecb_decrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct aria_ctx *ctx = crypto_skcipher_ctx(tfm);
+
+	return ecb_do_decrypt(req, ctx->dec_key[0]);
+}
+
+static int aria_avx_set_key(struct crypto_skcipher *tfm, const u8 *key,
+			    unsigned int keylen)
+{
+	return aria_set_key(&tfm->base, key, keylen);
+}
+
+static int aria_avx_ctr_encrypt(struct skcipher_request *req)
+{
+	struct aria_avx_request_ctx *req_ctx = skcipher_request_ctx(req);
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct aria_ctx *ctx = crypto_skcipher_ctx(tfm);
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes) > 0) {
+		const u8 *src = walk.src.virt.addr;
+		u8 *dst = walk.dst.virt.addr;
+
+		while (nbytes >= ARIA_AESNI_PARALLEL_BLOCK_SIZE) {
+			kernel_fpu_begin();
+			aria_ops.aria_ctr_crypt_16way(ctx, dst, src,
+						      &req_ctx->keystream[0],
+						      walk.iv);
+			kernel_fpu_end();
+			dst += ARIA_AESNI_PARALLEL_BLOCK_SIZE;
+			src += ARIA_AESNI_PARALLEL_BLOCK_SIZE;
+			nbytes -= ARIA_AESNI_PARALLEL_BLOCK_SIZE;
+		}
+
+		while (nbytes >= ARIA_BLOCK_SIZE) {
+			memcpy(&req_ctx->keystream[0], walk.iv, ARIA_BLOCK_SIZE);
+			crypto_inc(walk.iv, ARIA_BLOCK_SIZE);
+
+			aria_encrypt(ctx, &req_ctx->keystream[0],
+				     &req_ctx->keystream[0]);
+
+			crypto_xor_cpy(dst, src, &req_ctx->keystream[0],
+				       ARIA_BLOCK_SIZE);
+			dst += ARIA_BLOCK_SIZE;
+			src += ARIA_BLOCK_SIZE;
+			nbytes -= ARIA_BLOCK_SIZE;
+		}
+
+		if (walk.nbytes == walk.total && nbytes > 0) {
+			memcpy(&req_ctx->keystream[0], walk.iv,
+			       ARIA_BLOCK_SIZE);
+			crypto_inc(walk.iv, ARIA_BLOCK_SIZE);
+
+			aria_encrypt(ctx, &req_ctx->keystream[0],
+				     &req_ctx->keystream[0]);
+
+			crypto_xor_cpy(dst, src, &req_ctx->keystream[0],
+				       nbytes);
+			dst += nbytes;
+			src += nbytes;
+			nbytes = 0;
+		}
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	return err;
+}
+
+static int aria_avx_init_tfm(struct crypto_skcipher *tfm)
+{
+	crypto_skcipher_set_reqsize(tfm, sizeof(struct aria_avx_request_ctx));
+
+	return 0;
+}
+
+static struct skcipher_alg aria_algs[] = {
+	{
+		.base.cra_name		= "ecb(aria)",
+		.base.cra_driver_name	= "ecb-aria-avx",
+		.base.cra_priority	= 400,
+		.base.cra_blocksize	= ARIA_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct aria_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= ARIA_MIN_KEY_SIZE,
+		.max_keysize		= ARIA_MAX_KEY_SIZE,
+		.setkey			= aria_avx_set_key,
+		.encrypt		= aria_avx_ecb_encrypt,
+		.decrypt		= aria_avx_ecb_decrypt,
+	}, {
+		.base.cra_name		= "ctr(aria)",
+		.base.cra_driver_name	= "ctr-aria-avx",
+		.base.cra_priority	= 400,
+		.base.cra_blocksize	= 1,
+		.base.cra_ctxsize	= sizeof(struct aria_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= ARIA_MIN_KEY_SIZE,
+		.max_keysize		= ARIA_MAX_KEY_SIZE,
+		.ivsize			= ARIA_BLOCK_SIZE,
+		.chunksize		= ARIA_BLOCK_SIZE,
+		.walksize		= 16 * ARIA_BLOCK_SIZE,
+		.setkey			= aria_avx_set_key,
+		.encrypt		= aria_avx_ctr_encrypt,
+		.decrypt		= aria_avx_ctr_encrypt,
+		.init			= aria_avx_init_tfm,
+	}
+};
+
+static int __init aria_avx_init(void)
+{
+	const char *feature_name;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX) ||
+	    !boot_cpu_has(X86_FEATURE_AES) ||
+	    !boot_cpu_has(X86_FEATURE_OSXSAVE)) {
+		pr_info("AVX or AES-NI instructions are not detected.\n");
+		return -ENODEV;
+	}
+
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM,
+				&feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	if (boot_cpu_has(X86_FEATURE_GFNI)) {
+		aria_ops.aria_encrypt_16way = aria_aesni_avx_gfni_encrypt_16way;
+		aria_ops.aria_decrypt_16way = aria_aesni_avx_gfni_decrypt_16way;
+		aria_ops.aria_ctr_crypt_16way = aria_aesni_avx_gfni_ctr_crypt_16way;
+	} else {
+		aria_ops.aria_encrypt_16way = aria_aesni_avx_encrypt_16way;
+		aria_ops.aria_decrypt_16way = aria_aesni_avx_decrypt_16way;
+		aria_ops.aria_ctr_crypt_16way = aria_aesni_avx_ctr_crypt_16way;
+	}
+
+	return crypto_register_skciphers(aria_algs, ARRAY_SIZE(aria_algs));
+}
+
+static void __exit aria_avx_exit(void)
+{
+	crypto_unregister_skciphers(aria_algs, ARRAY_SIZE(aria_algs));
+}
+
+module_init(aria_avx_init);
+module_exit(aria_avx_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Taehee Yoo <ap420073@gmail.com>");
+MODULE_DESCRIPTION("ARIA Cipher Algorithm, AVX/AES-NI/GFNI optimized");
+MODULE_ALIAS_CRYPTO("aria");
+MODULE_ALIAS_CRYPTO("aria-aesni-avx");
diff --git a/arch/x86/crypto/aria_gfni_avx512_glue.c b/arch/x86/crypto/aria_gfni_avx512_glue.c
new file mode 100644
index 000000000..363cbf439
--- /dev/null
+++ b/arch/x86/crypto/aria_gfni_avx512_glue.c
@@ -0,0 +1,242 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Glue Code for the AVX512/GFNI assembler implementation of the ARIA Cipher
+ *
+ * Copyright (c) 2022 Taehee Yoo <ap420073@gmail.com>
+ */
+
+#include <crypto/algapi.h>
+#include <crypto/aria.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <linux/module.h>
+#include <linux/types.h>
+
+#include "ecb_cbc_helpers.h"
+#include "aria-avx.h"
+
+asmlinkage void aria_gfni_avx512_encrypt_64way(const void *ctx, u8 *dst,
+					       const u8 *src);
+asmlinkage void aria_gfni_avx512_decrypt_64way(const void *ctx, u8 *dst,
+					       const u8 *src);
+asmlinkage void aria_gfni_avx512_ctr_crypt_64way(const void *ctx, u8 *dst,
+						 const u8 *src,
+						 u8 *keystream, u8 *iv);
+
+static struct aria_avx_ops aria_ops;
+
+struct aria_avx512_request_ctx {
+	u8 keystream[ARIA_GFNI_AVX512_PARALLEL_BLOCK_SIZE];
+};
+
+static int ecb_do_encrypt(struct skcipher_request *req, const u32 *rkey)
+{
+	ECB_WALK_START(req, ARIA_BLOCK_SIZE, ARIA_AESNI_PARALLEL_BLOCKS);
+	ECB_BLOCK(ARIA_GFNI_AVX512_PARALLEL_BLOCKS, aria_ops.aria_encrypt_64way);
+	ECB_BLOCK(ARIA_AESNI_AVX2_PARALLEL_BLOCKS, aria_ops.aria_encrypt_32way);
+	ECB_BLOCK(ARIA_AESNI_PARALLEL_BLOCKS, aria_ops.aria_encrypt_16way);
+	ECB_BLOCK(1, aria_encrypt);
+	ECB_WALK_END();
+}
+
+static int ecb_do_decrypt(struct skcipher_request *req, const u32 *rkey)
+{
+	ECB_WALK_START(req, ARIA_BLOCK_SIZE, ARIA_AESNI_PARALLEL_BLOCKS);
+	ECB_BLOCK(ARIA_GFNI_AVX512_PARALLEL_BLOCKS, aria_ops.aria_decrypt_64way);
+	ECB_BLOCK(ARIA_AESNI_AVX2_PARALLEL_BLOCKS, aria_ops.aria_decrypt_32way);
+	ECB_BLOCK(ARIA_AESNI_PARALLEL_BLOCKS, aria_ops.aria_decrypt_16way);
+	ECB_BLOCK(1, aria_decrypt);
+	ECB_WALK_END();
+}
+
+static int aria_avx512_ecb_encrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct aria_ctx *ctx = crypto_skcipher_ctx(tfm);
+
+	return ecb_do_encrypt(req, ctx->enc_key[0]);
+}
+
+static int aria_avx512_ecb_decrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct aria_ctx *ctx = crypto_skcipher_ctx(tfm);
+
+	return ecb_do_decrypt(req, ctx->dec_key[0]);
+}
+
+static int aria_avx512_set_key(struct crypto_skcipher *tfm, const u8 *key,
+			    unsigned int keylen)
+{
+	return aria_set_key(&tfm->base, key, keylen);
+}
+
+static int aria_avx512_ctr_encrypt(struct skcipher_request *req)
+{
+	struct aria_avx512_request_ctx *req_ctx = skcipher_request_ctx(req);
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct aria_ctx *ctx = crypto_skcipher_ctx(tfm);
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes) > 0) {
+		const u8 *src = walk.src.virt.addr;
+		u8 *dst = walk.dst.virt.addr;
+
+		while (nbytes >= ARIA_GFNI_AVX512_PARALLEL_BLOCK_SIZE) {
+			kernel_fpu_begin();
+			aria_ops.aria_ctr_crypt_64way(ctx, dst, src,
+						      &req_ctx->keystream[0],
+						      walk.iv);
+			kernel_fpu_end();
+			dst += ARIA_GFNI_AVX512_PARALLEL_BLOCK_SIZE;
+			src += ARIA_GFNI_AVX512_PARALLEL_BLOCK_SIZE;
+			nbytes -= ARIA_GFNI_AVX512_PARALLEL_BLOCK_SIZE;
+		}
+
+		while (nbytes >= ARIA_AESNI_AVX2_PARALLEL_BLOCK_SIZE) {
+			kernel_fpu_begin();
+			aria_ops.aria_ctr_crypt_32way(ctx, dst, src,
+						      &req_ctx->keystream[0],
+						      walk.iv);
+			kernel_fpu_end();
+			dst += ARIA_AESNI_AVX2_PARALLEL_BLOCK_SIZE;
+			src += ARIA_AESNI_AVX2_PARALLEL_BLOCK_SIZE;
+			nbytes -= ARIA_AESNI_AVX2_PARALLEL_BLOCK_SIZE;
+		}
+
+		while (nbytes >= ARIA_AESNI_PARALLEL_BLOCK_SIZE) {
+			kernel_fpu_begin();
+			aria_ops.aria_ctr_crypt_16way(ctx, dst, src,
+						      &req_ctx->keystream[0],
+						      walk.iv);
+			kernel_fpu_end();
+			dst += ARIA_AESNI_PARALLEL_BLOCK_SIZE;
+			src += ARIA_AESNI_PARALLEL_BLOCK_SIZE;
+			nbytes -= ARIA_AESNI_PARALLEL_BLOCK_SIZE;
+		}
+
+		while (nbytes >= ARIA_BLOCK_SIZE) {
+			memcpy(&req_ctx->keystream[0], walk.iv,
+			       ARIA_BLOCK_SIZE);
+			crypto_inc(walk.iv, ARIA_BLOCK_SIZE);
+
+			aria_encrypt(ctx, &req_ctx->keystream[0],
+				     &req_ctx->keystream[0]);
+
+			crypto_xor_cpy(dst, src, &req_ctx->keystream[0],
+				       ARIA_BLOCK_SIZE);
+			dst += ARIA_BLOCK_SIZE;
+			src += ARIA_BLOCK_SIZE;
+			nbytes -= ARIA_BLOCK_SIZE;
+		}
+
+		if (walk.nbytes == walk.total && nbytes > 0) {
+			memcpy(&req_ctx->keystream[0], walk.iv,
+			       ARIA_BLOCK_SIZE);
+			crypto_inc(walk.iv, ARIA_BLOCK_SIZE);
+
+			aria_encrypt(ctx, &req_ctx->keystream[0],
+				     &req_ctx->keystream[0]);
+
+			crypto_xor_cpy(dst, src, &req_ctx->keystream[0],
+				       nbytes);
+			dst += nbytes;
+			src += nbytes;
+			nbytes = 0;
+		}
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	return err;
+}
+
+static int aria_avx512_init_tfm(struct crypto_skcipher *tfm)
+{
+	crypto_skcipher_set_reqsize(tfm,
+				    sizeof(struct aria_avx512_request_ctx));
+
+	return 0;
+}
+
+static struct skcipher_alg aria_algs[] = {
+	{
+		.base.cra_name		= "ecb(aria)",
+		.base.cra_driver_name	= "ecb-aria-avx512",
+		.base.cra_priority	= 600,
+		.base.cra_blocksize	= ARIA_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct aria_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= ARIA_MIN_KEY_SIZE,
+		.max_keysize		= ARIA_MAX_KEY_SIZE,
+		.setkey			= aria_avx512_set_key,
+		.encrypt		= aria_avx512_ecb_encrypt,
+		.decrypt		= aria_avx512_ecb_decrypt,
+	}, {
+		.base.cra_name		= "ctr(aria)",
+		.base.cra_driver_name	= "ctr-aria-avx512",
+		.base.cra_priority	= 600,
+		.base.cra_flags		= CRYPTO_ALG_SKCIPHER_REQSIZE_LARGE,
+		.base.cra_blocksize	= 1,
+		.base.cra_ctxsize	= sizeof(struct aria_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= ARIA_MIN_KEY_SIZE,
+		.max_keysize		= ARIA_MAX_KEY_SIZE,
+		.ivsize			= ARIA_BLOCK_SIZE,
+		.chunksize		= ARIA_BLOCK_SIZE,
+		.setkey			= aria_avx512_set_key,
+		.encrypt		= aria_avx512_ctr_encrypt,
+		.decrypt		= aria_avx512_ctr_encrypt,
+		.init                   = aria_avx512_init_tfm,
+	}
+};
+
+static int __init aria_avx512_init(void)
+{
+	const char *feature_name;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX) ||
+	    !boot_cpu_has(X86_FEATURE_AVX2) ||
+	    !boot_cpu_has(X86_FEATURE_AVX512F) ||
+	    !boot_cpu_has(X86_FEATURE_AVX512VL) ||
+	    !boot_cpu_has(X86_FEATURE_GFNI) ||
+	    !boot_cpu_has(X86_FEATURE_OSXSAVE)) {
+		pr_info("AVX512/GFNI instructions are not detected.\n");
+		return -ENODEV;
+	}
+
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM |
+			       XFEATURE_MASK_AVX512, &feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	aria_ops.aria_encrypt_16way = aria_aesni_avx_gfni_encrypt_16way;
+	aria_ops.aria_decrypt_16way = aria_aesni_avx_gfni_decrypt_16way;
+	aria_ops.aria_ctr_crypt_16way = aria_aesni_avx_gfni_ctr_crypt_16way;
+	aria_ops.aria_encrypt_32way = aria_aesni_avx2_gfni_encrypt_32way;
+	aria_ops.aria_decrypt_32way = aria_aesni_avx2_gfni_decrypt_32way;
+	aria_ops.aria_ctr_crypt_32way = aria_aesni_avx2_gfni_ctr_crypt_32way;
+	aria_ops.aria_encrypt_64way = aria_gfni_avx512_encrypt_64way;
+	aria_ops.aria_decrypt_64way = aria_gfni_avx512_decrypt_64way;
+	aria_ops.aria_ctr_crypt_64way = aria_gfni_avx512_ctr_crypt_64way;
+
+	return crypto_register_skciphers(aria_algs, ARRAY_SIZE(aria_algs));
+}
+
+static void __exit aria_avx512_exit(void)
+{
+	crypto_unregister_skciphers(aria_algs, ARRAY_SIZE(aria_algs));
+}
+
+module_init(aria_avx512_init);
+module_exit(aria_avx512_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Taehee Yoo <ap420073@gmail.com>");
+MODULE_DESCRIPTION("ARIA Cipher Algorithm, AVX512/GFNI optimized");
+MODULE_ALIAS_CRYPTO("aria");
+MODULE_ALIAS_CRYPTO("aria-gfni-avx512");
diff --git a/arch/x86/crypto/blowfish-x86_64-asm_64.S b/arch/x86/crypto/blowfish-x86_64-asm_64.S
new file mode 100644
index 000000000..e88c8e4f0
--- /dev/null
+++ b/arch/x86/crypto/blowfish-x86_64-asm_64.S
@@ -0,0 +1,354 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Blowfish Cipher Algorithm (x86_64)
+ *
+ * Copyright (C) 2011 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ */
+
+#include <linux/linkage.h>
+
+.file "blowfish-x86_64-asm.S"
+.text
+
+/* structure of crypto context */
+#define p	0
+#define s0	((16 + 2) * 4)
+#define s1	((16 + 2 + (1 * 256)) * 4)
+#define s2	((16 + 2 + (2 * 256)) * 4)
+#define s3	((16 + 2 + (3 * 256)) * 4)
+
+/* register macros */
+#define CTX %r12
+#define RIO %rsi
+
+#define RX0 %rax
+#define RX1 %rbx
+#define RX2 %rcx
+#define RX3 %rdx
+
+#define RX0d %eax
+#define RX1d %ebx
+#define RX2d %ecx
+#define RX3d %edx
+
+#define RX0bl %al
+#define RX1bl %bl
+#define RX2bl %cl
+#define RX3bl %dl
+
+#define RX0bh %ah
+#define RX1bh %bh
+#define RX2bh %ch
+#define RX3bh %dh
+
+#define RT0 %rdi
+#define RT1 %rsi
+#define RT2 %r8
+#define RT3 %r9
+
+#define RT0d %edi
+#define RT1d %esi
+#define RT2d %r8d
+#define RT3d %r9d
+
+#define RKEY %r10
+
+/***********************************************************************
+ * 1-way blowfish
+ ***********************************************************************/
+#define F() \
+	rorq $16,		RX0; \
+	movzbl RX0bh,		RT0d; \
+	movzbl RX0bl,		RT1d; \
+	rolq $16,		RX0; \
+	movl s0(CTX,RT0,4),	RT0d; \
+	addl s1(CTX,RT1,4),	RT0d; \
+	movzbl RX0bh,		RT1d; \
+	movzbl RX0bl,		RT2d; \
+	rolq $32,		RX0; \
+	xorl s2(CTX,RT1,4),	RT0d; \
+	addl s3(CTX,RT2,4),	RT0d; \
+	xorq RT0,		RX0;
+
+#define add_roundkey_enc(n) \
+	xorq p+4*(n)(CTX), 	RX0;
+
+#define round_enc(n) \
+	add_roundkey_enc(n); \
+	\
+	F(); \
+	F();
+
+#define add_roundkey_dec(n) \
+	movq p+4*(n-1)(CTX),	RT0; \
+	rorq $32,		RT0; \
+	xorq RT0,		RX0;
+
+#define round_dec(n) \
+	add_roundkey_dec(n); \
+	\
+	F(); \
+	F(); \
+
+#define read_block() \
+	movq (RIO), 		RX0; \
+	rorq $32, 		RX0; \
+	bswapq 			RX0;
+
+#define write_block() \
+	bswapq 			RX0; \
+	movq RX0, 		(RIO);
+
+SYM_FUNC_START(blowfish_enc_blk)
+	/* input:
+	 *	%rdi: ctx
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	movq %r12, %r11;
+
+	movq %rdi, CTX;
+	movq %rsi, %r10;
+	movq %rdx, RIO;
+
+	read_block();
+
+	round_enc(0);
+	round_enc(2);
+	round_enc(4);
+	round_enc(6);
+	round_enc(8);
+	round_enc(10);
+	round_enc(12);
+	round_enc(14);
+	add_roundkey_enc(16);
+
+	movq %r11, %r12;
+	movq %r10, RIO;
+
+	write_block();
+	RET;
+SYM_FUNC_END(blowfish_enc_blk)
+
+SYM_FUNC_START(blowfish_dec_blk)
+	/* input:
+	 *	%rdi: ctx
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	movq %r12, %r11;
+
+	movq %rdi, CTX;
+	movq %rsi, %r10;
+	movq %rdx, RIO;
+
+	read_block();
+
+	round_dec(17);
+	round_dec(15);
+	round_dec(13);
+	round_dec(11);
+	round_dec(9);
+	round_dec(7);
+	round_dec(5);
+	round_dec(3);
+	add_roundkey_dec(1);
+
+	movq %r10, RIO;
+	write_block();
+
+	movq %r11, %r12;
+
+	RET;
+SYM_FUNC_END(blowfish_dec_blk)
+
+/**********************************************************************
+  4-way blowfish, four blocks parallel
+ **********************************************************************/
+
+/* F() for 4-way. Slower when used alone/1-way, but faster when used
+ * parallel/4-way (tested on AMD Phenom II & Intel Xeon E7330).
+ */
+#define F4(x) \
+	movzbl x ## bh,		RT1d; \
+	movzbl x ## bl,		RT3d; \
+	rorq $16,		x; \
+	movzbl x ## bh,		RT0d; \
+	movzbl x ## bl,		RT2d; \
+	rorq $16,		x; \
+	movl s0(CTX,RT0,4),	RT0d; \
+	addl s1(CTX,RT2,4),	RT0d; \
+	xorl s2(CTX,RT1,4),	RT0d; \
+	addl s3(CTX,RT3,4),	RT0d; \
+	xorq RT0,		x;
+
+#define add_preloaded_roundkey4() \
+	xorq RKEY,		RX0; \
+	xorq RKEY,		RX1; \
+	xorq RKEY,		RX2; \
+	xorq RKEY,		RX3;
+
+#define preload_roundkey_enc(n) \
+	movq p+4*(n)(CTX),	RKEY;
+
+#define add_roundkey_enc4(n) \
+	add_preloaded_roundkey4(); \
+	preload_roundkey_enc(n + 2);
+
+#define round_enc4(n) \
+	add_roundkey_enc4(n); \
+	\
+	F4(RX0); \
+	F4(RX1); \
+	F4(RX2); \
+	F4(RX3); \
+	\
+	F4(RX0); \
+	F4(RX1); \
+	F4(RX2); \
+	F4(RX3);
+
+#define preload_roundkey_dec(n) \
+	movq p+4*((n)-1)(CTX),	RKEY; \
+	rorq $32,		RKEY;
+
+#define add_roundkey_dec4(n) \
+	add_preloaded_roundkey4(); \
+	preload_roundkey_dec(n - 2);
+
+#define round_dec4(n) \
+	add_roundkey_dec4(n); \
+	\
+	F4(RX0); \
+	F4(RX1); \
+	F4(RX2); \
+	F4(RX3); \
+	\
+	F4(RX0); \
+	F4(RX1); \
+	F4(RX2); \
+	F4(RX3);
+
+#define read_block4() \
+	movq (RIO),		RX0; \
+	rorq $32,		RX0; \
+	bswapq 			RX0; \
+	\
+	movq 8(RIO),		RX1; \
+	rorq $32,		RX1; \
+	bswapq 			RX1; \
+	\
+	movq 16(RIO),		RX2; \
+	rorq $32,		RX2; \
+	bswapq 			RX2; \
+	\
+	movq 24(RIO),		RX3; \
+	rorq $32,		RX3; \
+	bswapq 			RX3;
+
+#define write_block4() \
+	bswapq 			RX0; \
+	movq RX0,		(RIO); \
+	\
+	bswapq 			RX1; \
+	movq RX1,		8(RIO); \
+	\
+	bswapq 			RX2; \
+	movq RX2,		16(RIO); \
+	\
+	bswapq 			RX3; \
+	movq RX3,		24(RIO);
+
+#define xor_block4() \
+	movq (RIO),		RT0; \
+	bswapq			RT0; \
+	xorq RT0,		RX1; \
+	\
+	movq 8(RIO),		RT2; \
+	bswapq			RT2; \
+	xorq RT2,		RX2; \
+	\
+	movq 16(RIO),		RT3; \
+	bswapq			RT3; \
+	xorq RT3,		RX3;
+
+SYM_FUNC_START(blowfish_enc_blk_4way)
+	/* input:
+	 *	%rdi: ctx
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	pushq %r12;
+	pushq %rbx;
+
+	movq %rdi, CTX
+	movq %rsi, %r11;
+	movq %rdx, RIO;
+
+	preload_roundkey_enc(0);
+
+	read_block4();
+
+	round_enc4(0);
+	round_enc4(2);
+	round_enc4(4);
+	round_enc4(6);
+	round_enc4(8);
+	round_enc4(10);
+	round_enc4(12);
+	round_enc4(14);
+	add_preloaded_roundkey4();
+
+	movq %r11, RIO;
+	write_block4();
+
+	popq %rbx;
+	popq %r12;
+	RET;
+SYM_FUNC_END(blowfish_enc_blk_4way)
+
+SYM_FUNC_START(__blowfish_dec_blk_4way)
+	/* input:
+	 *	%rdi: ctx
+	 *	%rsi: dst
+	 *	%rdx: src
+	 *	%rcx: cbc (bool)
+	 */
+	pushq %r12;
+	pushq %rbx;
+	pushq %rcx;
+	pushq %rdx;
+
+	movq %rdi, CTX;
+	movq %rsi, %r11;
+	movq %rdx, RIO;
+
+	preload_roundkey_dec(17);
+	read_block4();
+
+	round_dec4(17);
+	round_dec4(15);
+	round_dec4(13);
+	round_dec4(11);
+	round_dec4(9);
+	round_dec4(7);
+	round_dec4(5);
+	round_dec4(3);
+	add_preloaded_roundkey4();
+
+	popq RIO;
+	popq %r12;
+	testq %r12, %r12;
+	jz .L_no_cbc_xor;
+
+	xor_block4();
+
+.L_no_cbc_xor:
+	movq %r11, RIO;
+	write_block4();
+
+	popq %rbx;
+	popq %r12;
+
+	RET;
+SYM_FUNC_END(__blowfish_dec_blk_4way)
diff --git a/arch/x86/crypto/blowfish_glue.c b/arch/x86/crypto/blowfish_glue.c
new file mode 100644
index 000000000..26c5f2ee5
--- /dev/null
+++ b/arch/x86/crypto/blowfish_glue.c
@@ -0,0 +1,196 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Glue Code for assembler optimized version of Blowfish
+ *
+ * Copyright (c) 2011 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ *
+ * CBC & ECB parts based on code (crypto/cbc.c,ecb.c) by:
+ *   Copyright (c) 2006 Herbert Xu <herbert@gondor.apana.org.au>
+ */
+
+#include <crypto/algapi.h>
+#include <crypto/blowfish.h>
+#include <crypto/internal/skcipher.h>
+#include <linux/crypto.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/types.h>
+
+#include "ecb_cbc_helpers.h"
+
+/* regular block cipher functions */
+asmlinkage void blowfish_enc_blk(struct bf_ctx *ctx, u8 *dst, const u8 *src);
+asmlinkage void blowfish_dec_blk(struct bf_ctx *ctx, u8 *dst, const u8 *src);
+
+/* 4-way parallel cipher functions */
+asmlinkage void blowfish_enc_blk_4way(struct bf_ctx *ctx, u8 *dst,
+				      const u8 *src);
+asmlinkage void __blowfish_dec_blk_4way(struct bf_ctx *ctx, u8 *dst,
+					const u8 *src, bool cbc);
+
+static inline void blowfish_dec_ecb_4way(struct bf_ctx *ctx, u8 *dst,
+					     const u8 *src)
+{
+	return __blowfish_dec_blk_4way(ctx, dst, src, false);
+}
+
+static inline void blowfish_dec_cbc_4way(struct bf_ctx *ctx, u8 *dst,
+					     const u8 *src)
+{
+	return __blowfish_dec_blk_4way(ctx, dst, src, true);
+}
+
+static void blowfish_encrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
+{
+	blowfish_enc_blk(crypto_tfm_ctx(tfm), dst, src);
+}
+
+static void blowfish_decrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
+{
+	blowfish_dec_blk(crypto_tfm_ctx(tfm), dst, src);
+}
+
+static int blowfish_setkey_skcipher(struct crypto_skcipher *tfm,
+				    const u8 *key, unsigned int keylen)
+{
+	return blowfish_setkey(&tfm->base, key, keylen);
+}
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, BF_BLOCK_SIZE, -1);
+	ECB_BLOCK(4, blowfish_enc_blk_4way);
+	ECB_BLOCK(1, blowfish_enc_blk);
+	ECB_WALK_END();
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, BF_BLOCK_SIZE, -1);
+	ECB_BLOCK(4, blowfish_dec_ecb_4way);
+	ECB_BLOCK(1, blowfish_dec_blk);
+	ECB_WALK_END();
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, BF_BLOCK_SIZE, -1);
+	CBC_ENC_BLOCK(blowfish_enc_blk);
+	CBC_WALK_END();
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, BF_BLOCK_SIZE, -1);
+	CBC_DEC_BLOCK(4, blowfish_dec_cbc_4way);
+	CBC_DEC_BLOCK(1, blowfish_dec_blk);
+	CBC_WALK_END();
+}
+
+static struct crypto_alg bf_cipher_alg = {
+	.cra_name		= "blowfish",
+	.cra_driver_name	= "blowfish-asm",
+	.cra_priority		= 200,
+	.cra_flags		= CRYPTO_ALG_TYPE_CIPHER,
+	.cra_blocksize		= BF_BLOCK_SIZE,
+	.cra_ctxsize		= sizeof(struct bf_ctx),
+	.cra_module		= THIS_MODULE,
+	.cra_u = {
+		.cipher = {
+			.cia_min_keysize	= BF_MIN_KEY_SIZE,
+			.cia_max_keysize	= BF_MAX_KEY_SIZE,
+			.cia_setkey		= blowfish_setkey,
+			.cia_encrypt		= blowfish_encrypt,
+			.cia_decrypt		= blowfish_decrypt,
+		}
+	}
+};
+
+static struct skcipher_alg bf_skcipher_algs[] = {
+	{
+		.base.cra_name		= "ecb(blowfish)",
+		.base.cra_driver_name	= "ecb-blowfish-asm",
+		.base.cra_priority	= 300,
+		.base.cra_blocksize	= BF_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct bf_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= BF_MIN_KEY_SIZE,
+		.max_keysize		= BF_MAX_KEY_SIZE,
+		.setkey			= blowfish_setkey_skcipher,
+		.encrypt		= ecb_encrypt,
+		.decrypt		= ecb_decrypt,
+	}, {
+		.base.cra_name		= "cbc(blowfish)",
+		.base.cra_driver_name	= "cbc-blowfish-asm",
+		.base.cra_priority	= 300,
+		.base.cra_blocksize	= BF_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct bf_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= BF_MIN_KEY_SIZE,
+		.max_keysize		= BF_MAX_KEY_SIZE,
+		.ivsize			= BF_BLOCK_SIZE,
+		.setkey			= blowfish_setkey_skcipher,
+		.encrypt		= cbc_encrypt,
+		.decrypt		= cbc_decrypt,
+	},
+};
+
+static bool is_blacklisted_cpu(void)
+{
+	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
+		return false;
+
+	if (boot_cpu_data.x86 == 0x0f) {
+		/*
+		 * On Pentium 4, blowfish-x86_64 is slower than generic C
+		 * implementation because use of 64bit rotates (which are really
+		 * slow on P4). Therefore blacklist P4s.
+		 */
+		return true;
+	}
+
+	return false;
+}
+
+static int force;
+module_param(force, int, 0);
+MODULE_PARM_DESC(force, "Force module load, ignore CPU blacklist");
+
+static int __init blowfish_init(void)
+{
+	int err;
+
+	if (!force && is_blacklisted_cpu()) {
+		printk(KERN_INFO
+			"blowfish-x86_64: performance on this CPU "
+			"would be suboptimal: disabling "
+			"blowfish-x86_64.\n");
+		return -ENODEV;
+	}
+
+	err = crypto_register_alg(&bf_cipher_alg);
+	if (err)
+		return err;
+
+	err = crypto_register_skciphers(bf_skcipher_algs,
+					ARRAY_SIZE(bf_skcipher_algs));
+	if (err)
+		crypto_unregister_alg(&bf_cipher_alg);
+
+	return err;
+}
+
+static void __exit blowfish_fini(void)
+{
+	crypto_unregister_alg(&bf_cipher_alg);
+	crypto_unregister_skciphers(bf_skcipher_algs,
+				    ARRAY_SIZE(bf_skcipher_algs));
+}
+
+module_init(blowfish_init);
+module_exit(blowfish_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Blowfish Cipher Algorithm, asm optimized");
+MODULE_ALIAS_CRYPTO("blowfish");
+MODULE_ALIAS_CRYPTO("blowfish-asm");
diff --git a/arch/x86/crypto/camellia-aesni-avx-asm_64.S b/arch/x86/crypto/camellia-aesni-avx-asm_64.S
new file mode 100644
index 000000000..1dfef28c1
--- /dev/null
+++ b/arch/x86/crypto/camellia-aesni-avx-asm_64.S
@@ -0,0 +1,990 @@
+/*
+ * x86_64/AVX/AES-NI assembler implementation of Camellia
+ *
+ * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ */
+
+/*
+ * Version licensed under 2-clause BSD License is available at:
+ *	http://koti.mbnet.fi/axh/crypto/camellia-BSD-1.2.0-aesni1.tar.xz
+ */
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+#include <asm/frame.h>
+
+#define CAMELLIA_TABLE_BYTE_LEN 272
+
+/* struct camellia_ctx: */
+#define key_table 0
+#define key_length CAMELLIA_TABLE_BYTE_LEN
+
+/* register macros */
+#define CTX %rdi
+
+/**********************************************************************
+  16-way camellia
+ **********************************************************************/
+#define filter_8bit(x, lo_t, hi_t, mask4bit, tmp0) \
+	vpand x, mask4bit, tmp0; \
+	vpandn x, mask4bit, x; \
+	vpsrld $4, x, x; \
+	\
+	vpshufb tmp0, lo_t, tmp0; \
+	vpshufb x, hi_t, x; \
+	vpxor tmp0, x, x;
+
+/*
+ * IN:
+ *   x0..x7: byte-sliced AB state
+ *   mem_cd: register pointer storing CD state
+ *   key: index for key material
+ * OUT:
+ *   x0..x7: new byte-sliced CD state
+ */
+#define roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, t0, t1, t2, t3, t4, t5, t6, \
+		  t7, mem_cd, key) \
+	/* \
+	 * S-function with AES subbytes \
+	 */ \
+	vmovdqa .Linv_shift_row(%rip), t4; \
+	vbroadcastss .L0f0f0f0f(%rip), t7; \
+	vmovdqa .Lpre_tf_lo_s1(%rip), t0; \
+	vmovdqa .Lpre_tf_hi_s1(%rip), t1; \
+	\
+	/* AES inverse shift rows */ \
+	vpshufb t4, x0, x0; \
+	vpshufb t4, x7, x7; \
+	vpshufb t4, x1, x1; \
+	vpshufb t4, x4, x4; \
+	vpshufb t4, x2, x2; \
+	vpshufb t4, x5, x5; \
+	vpshufb t4, x3, x3; \
+	vpshufb t4, x6, x6; \
+	\
+	/* prefilter sboxes 1, 2 and 3 */ \
+	vmovdqa .Lpre_tf_lo_s4(%rip), t2; \
+	vmovdqa .Lpre_tf_hi_s4(%rip), t3; \
+	filter_8bit(x0, t0, t1, t7, t6); \
+	filter_8bit(x7, t0, t1, t7, t6); \
+	filter_8bit(x1, t0, t1, t7, t6); \
+	filter_8bit(x4, t0, t1, t7, t6); \
+	filter_8bit(x2, t0, t1, t7, t6); \
+	filter_8bit(x5, t0, t1, t7, t6); \
+	\
+	/* prefilter sbox 4 */ \
+	vpxor t4, t4, t4; \
+	filter_8bit(x3, t2, t3, t7, t6); \
+	filter_8bit(x6, t2, t3, t7, t6); \
+	\
+	/* AES subbytes + AES shift rows */ \
+	vmovdqa .Lpost_tf_lo_s1(%rip), t0; \
+	vmovdqa .Lpost_tf_hi_s1(%rip), t1; \
+	vaesenclast t4, x0, x0; \
+	vaesenclast t4, x7, x7; \
+	vaesenclast t4, x1, x1; \
+	vaesenclast t4, x4, x4; \
+	vaesenclast t4, x2, x2; \
+	vaesenclast t4, x5, x5; \
+	vaesenclast t4, x3, x3; \
+	vaesenclast t4, x6, x6; \
+	\
+	/* postfilter sboxes 1 and 4 */ \
+	vmovdqa .Lpost_tf_lo_s3(%rip), t2; \
+	vmovdqa .Lpost_tf_hi_s3(%rip), t3; \
+	filter_8bit(x0, t0, t1, t7, t6); \
+	filter_8bit(x7, t0, t1, t7, t6); \
+	filter_8bit(x3, t0, t1, t7, t6); \
+	filter_8bit(x6, t0, t1, t7, t6); \
+	\
+	/* postfilter sbox 3 */ \
+	vmovdqa .Lpost_tf_lo_s2(%rip), t4; \
+	vmovdqa .Lpost_tf_hi_s2(%rip), t5; \
+	filter_8bit(x2, t2, t3, t7, t6); \
+	filter_8bit(x5, t2, t3, t7, t6); \
+	\
+	vpxor t6, t6, t6; \
+	vmovq key, t0; \
+	\
+	/* postfilter sbox 2 */ \
+	filter_8bit(x1, t4, t5, t7, t2); \
+	filter_8bit(x4, t4, t5, t7, t2); \
+	\
+	vpsrldq $5, t0, t5; \
+	vpsrldq $1, t0, t1; \
+	vpsrldq $2, t0, t2; \
+	vpsrldq $3, t0, t3; \
+	vpsrldq $4, t0, t4; \
+	vpshufb t6, t0, t0; \
+	vpshufb t6, t1, t1; \
+	vpshufb t6, t2, t2; \
+	vpshufb t6, t3, t3; \
+	vpshufb t6, t4, t4; \
+	vpsrldq $2, t5, t7; \
+	vpshufb t6, t7, t7; \
+	\
+	/* \
+	 * P-function \
+	 */ \
+	vpxor x5, x0, x0; \
+	vpxor x6, x1, x1; \
+	vpxor x7, x2, x2; \
+	vpxor x4, x3, x3; \
+	\
+	vpxor x2, x4, x4; \
+	vpxor x3, x5, x5; \
+	vpxor x0, x6, x6; \
+	vpxor x1, x7, x7; \
+	\
+	vpxor x7, x0, x0; \
+	vpxor x4, x1, x1; \
+	vpxor x5, x2, x2; \
+	vpxor x6, x3, x3; \
+	\
+	vpxor x3, x4, x4; \
+	vpxor x0, x5, x5; \
+	vpxor x1, x6, x6; \
+	vpxor x2, x7, x7; /* note: high and low parts swapped */ \
+	\
+	/* \
+	 * Add key material and result to CD (x becomes new CD) \
+	 */ \
+	\
+	vpxor t3, x4, x4; \
+	vpxor 0 * 16(mem_cd), x4, x4; \
+	\
+	vpxor t2, x5, x5; \
+	vpxor 1 * 16(mem_cd), x5, x5; \
+	\
+	vpsrldq $1, t5, t3; \
+	vpshufb t6, t5, t5; \
+	vpshufb t6, t3, t6; \
+	\
+	vpxor t1, x6, x6; \
+	vpxor 2 * 16(mem_cd), x6, x6; \
+	\
+	vpxor t0, x7, x7; \
+	vpxor 3 * 16(mem_cd), x7, x7; \
+	\
+	vpxor t7, x0, x0; \
+	vpxor 4 * 16(mem_cd), x0, x0; \
+	\
+	vpxor t6, x1, x1; \
+	vpxor 5 * 16(mem_cd), x1, x1; \
+	\
+	vpxor t5, x2, x2; \
+	vpxor 6 * 16(mem_cd), x2, x2; \
+	\
+	vpxor t4, x3, x3; \
+	vpxor 7 * 16(mem_cd), x3, x3;
+
+/*
+ * Size optimization... with inlined roundsm16, binary would be over 5 times
+ * larger and would only be 0.5% faster (on sandy-bridge).
+ */
+.align 8
+SYM_FUNC_START_LOCAL(roundsm16_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
+	roundsm16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		  %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14, %xmm15,
+		  %rcx, (%r9));
+	RET;
+SYM_FUNC_END(roundsm16_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
+
+.align 8
+SYM_FUNC_START_LOCAL(roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
+	roundsm16(%xmm4, %xmm5, %xmm6, %xmm7, %xmm0, %xmm1, %xmm2, %xmm3,
+		  %xmm12, %xmm13, %xmm14, %xmm15, %xmm8, %xmm9, %xmm10, %xmm11,
+		  %rax, (%r9));
+	RET;
+SYM_FUNC_END(roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
+
+/*
+ * IN/OUT:
+ *  x0..x7: byte-sliced AB state preloaded
+ *  mem_ab: byte-sliced AB state in memory
+ *  mem_cb: byte-sliced CD state in memory
+ */
+#define two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, i, dir, store_ab) \
+	leaq (key_table + (i) * 8)(CTX), %r9; \
+	call roundsm16_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd; \
+	\
+	vmovdqu x4, 0 * 16(mem_cd); \
+	vmovdqu x5, 1 * 16(mem_cd); \
+	vmovdqu x6, 2 * 16(mem_cd); \
+	vmovdqu x7, 3 * 16(mem_cd); \
+	vmovdqu x0, 4 * 16(mem_cd); \
+	vmovdqu x1, 5 * 16(mem_cd); \
+	vmovdqu x2, 6 * 16(mem_cd); \
+	vmovdqu x3, 7 * 16(mem_cd); \
+	\
+	leaq (key_table + ((i) + (dir)) * 8)(CTX), %r9; \
+	call roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab; \
+	\
+	store_ab(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab);
+
+#define dummy_store(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) /* do nothing */
+
+#define store_ab_state(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) \
+	/* Store new AB state */ \
+	vmovdqu x0, 0 * 16(mem_ab); \
+	vmovdqu x1, 1 * 16(mem_ab); \
+	vmovdqu x2, 2 * 16(mem_ab); \
+	vmovdqu x3, 3 * 16(mem_ab); \
+	vmovdqu x4, 4 * 16(mem_ab); \
+	vmovdqu x5, 5 * 16(mem_ab); \
+	vmovdqu x6, 6 * 16(mem_ab); \
+	vmovdqu x7, 7 * 16(mem_ab);
+
+#define enc_rounds16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, i) \
+	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 2, 1, store_ab_state); \
+	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 4, 1, store_ab_state); \
+	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 6, 1, dummy_store);
+
+#define dec_rounds16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, i) \
+	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 7, -1, store_ab_state); \
+	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 5, -1, store_ab_state); \
+	two_roundsm16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 3, -1, dummy_store);
+
+/*
+ * IN:
+ *  v0..3: byte-sliced 32-bit integers
+ * OUT:
+ *  v0..3: (IN <<< 1)
+ */
+#define rol32_1_16(v0, v1, v2, v3, t0, t1, t2, zero) \
+	vpcmpgtb v0, zero, t0; \
+	vpaddb v0, v0, v0; \
+	vpabsb t0, t0; \
+	\
+	vpcmpgtb v1, zero, t1; \
+	vpaddb v1, v1, v1; \
+	vpabsb t1, t1; \
+	\
+	vpcmpgtb v2, zero, t2; \
+	vpaddb v2, v2, v2; \
+	vpabsb t2, t2; \
+	\
+	vpor t0, v1, v1; \
+	\
+	vpcmpgtb v3, zero, t0; \
+	vpaddb v3, v3, v3; \
+	vpabsb t0, t0; \
+	\
+	vpor t1, v2, v2; \
+	vpor t2, v3, v3; \
+	vpor t0, v0, v0;
+
+/*
+ * IN:
+ *   r: byte-sliced AB state in memory
+ *   l: byte-sliced CD state in memory
+ * OUT:
+ *   x0..x7: new byte-sliced CD state
+ */
+#define fls16(l, l0, l1, l2, l3, l4, l5, l6, l7, r, t0, t1, t2, t3, tt0, \
+	      tt1, tt2, tt3, kll, klr, krl, krr) \
+	/* \
+	 * t0 = kll; \
+	 * t0 &= ll; \
+	 * lr ^= rol32(t0, 1); \
+	 */ \
+	vpxor tt0, tt0, tt0; \
+	vmovd kll, t0; \
+	vpshufb tt0, t0, t3; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t2; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t1; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t0; \
+	\
+	vpand l0, t0, t0; \
+	vpand l1, t1, t1; \
+	vpand l2, t2, t2; \
+	vpand l3, t3, t3; \
+	\
+	rol32_1_16(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
+	\
+	vpxor l4, t0, l4; \
+	vmovdqu l4, 4 * 16(l); \
+	vpxor l5, t1, l5; \
+	vmovdqu l5, 5 * 16(l); \
+	vpxor l6, t2, l6; \
+	vmovdqu l6, 6 * 16(l); \
+	vpxor l7, t3, l7; \
+	vmovdqu l7, 7 * 16(l); \
+	\
+	/* \
+	 * t2 = krr; \
+	 * t2 |= rr; \
+	 * rl ^= t2; \
+	 */ \
+	\
+	vmovd krr, t0; \
+	vpshufb tt0, t0, t3; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t2; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t1; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t0; \
+	\
+	vpor 4 * 16(r), t0, t0; \
+	vpor 5 * 16(r), t1, t1; \
+	vpor 6 * 16(r), t2, t2; \
+	vpor 7 * 16(r), t3, t3; \
+	\
+	vpxor 0 * 16(r), t0, t0; \
+	vpxor 1 * 16(r), t1, t1; \
+	vpxor 2 * 16(r), t2, t2; \
+	vpxor 3 * 16(r), t3, t3; \
+	vmovdqu t0, 0 * 16(r); \
+	vmovdqu t1, 1 * 16(r); \
+	vmovdqu t2, 2 * 16(r); \
+	vmovdqu t3, 3 * 16(r); \
+	\
+	/* \
+	 * t2 = krl; \
+	 * t2 &= rl; \
+	 * rr ^= rol32(t2, 1); \
+	 */ \
+	vmovd krl, t0; \
+	vpshufb tt0, t0, t3; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t2; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t1; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t0; \
+	\
+	vpand 0 * 16(r), t0, t0; \
+	vpand 1 * 16(r), t1, t1; \
+	vpand 2 * 16(r), t2, t2; \
+	vpand 3 * 16(r), t3, t3; \
+	\
+	rol32_1_16(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
+	\
+	vpxor 4 * 16(r), t0, t0; \
+	vpxor 5 * 16(r), t1, t1; \
+	vpxor 6 * 16(r), t2, t2; \
+	vpxor 7 * 16(r), t3, t3; \
+	vmovdqu t0, 4 * 16(r); \
+	vmovdqu t1, 5 * 16(r); \
+	vmovdqu t2, 6 * 16(r); \
+	vmovdqu t3, 7 * 16(r); \
+	\
+	/* \
+	 * t0 = klr; \
+	 * t0 |= lr; \
+	 * ll ^= t0; \
+	 */ \
+	\
+	vmovd klr, t0; \
+	vpshufb tt0, t0, t3; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t2; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t1; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t0; \
+	\
+	vpor l4, t0, t0; \
+	vpor l5, t1, t1; \
+	vpor l6, t2, t2; \
+	vpor l7, t3, t3; \
+	\
+	vpxor l0, t0, l0; \
+	vmovdqu l0, 0 * 16(l); \
+	vpxor l1, t1, l1; \
+	vmovdqu l1, 1 * 16(l); \
+	vpxor l2, t2, l2; \
+	vmovdqu l2, 2 * 16(l); \
+	vpxor l3, t3, l3; \
+	vmovdqu l3, 3 * 16(l);
+
+#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
+	vpunpckhdq x1, x0, t2; \
+	vpunpckldq x1, x0, x0; \
+	\
+	vpunpckldq x3, x2, t1; \
+	vpunpckhdq x3, x2, x2; \
+	\
+	vpunpckhqdq t1, x0, x1; \
+	vpunpcklqdq t1, x0, x0; \
+	\
+	vpunpckhqdq x2, t2, x3; \
+	vpunpcklqdq x2, t2, x2;
+
+#define byteslice_16x16b(a0, b0, c0, d0, a1, b1, c1, d1, a2, b2, c2, d2, a3, \
+			 b3, c3, d3, st0, st1) \
+	vmovdqu d2, st0; \
+	vmovdqu d3, st1; \
+	transpose_4x4(a0, a1, a2, a3, d2, d3); \
+	transpose_4x4(b0, b1, b2, b3, d2, d3); \
+	vmovdqu st0, d2; \
+	vmovdqu st1, d3; \
+	\
+	vmovdqu a0, st0; \
+	vmovdqu a1, st1; \
+	transpose_4x4(c0, c1, c2, c3, a0, a1); \
+	transpose_4x4(d0, d1, d2, d3, a0, a1); \
+	\
+	vmovdqu .Lshufb_16x16b(%rip), a0; \
+	vmovdqu st1, a1; \
+	vpshufb a0, a2, a2; \
+	vpshufb a0, a3, a3; \
+	vpshufb a0, b0, b0; \
+	vpshufb a0, b1, b1; \
+	vpshufb a0, b2, b2; \
+	vpshufb a0, b3, b3; \
+	vpshufb a0, a1, a1; \
+	vpshufb a0, c0, c0; \
+	vpshufb a0, c1, c1; \
+	vpshufb a0, c2, c2; \
+	vpshufb a0, c3, c3; \
+	vpshufb a0, d0, d0; \
+	vpshufb a0, d1, d1; \
+	vpshufb a0, d2, d2; \
+	vpshufb a0, d3, d3; \
+	vmovdqu d3, st1; \
+	vmovdqu st0, d3; \
+	vpshufb a0, d3, a0; \
+	vmovdqu d2, st0; \
+	\
+	transpose_4x4(a0, b0, c0, d0, d2, d3); \
+	transpose_4x4(a1, b1, c1, d1, d2, d3); \
+	vmovdqu st0, d2; \
+	vmovdqu st1, d3; \
+	\
+	vmovdqu b0, st0; \
+	vmovdqu b1, st1; \
+	transpose_4x4(a2, b2, c2, d2, b0, b1); \
+	transpose_4x4(a3, b3, c3, d3, b0, b1); \
+	vmovdqu st0, b0; \
+	vmovdqu st1, b1; \
+	/* does not adjust output bytes inside vectors */
+
+/* load blocks to registers and apply pre-whitening */
+#define inpack16_pre(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		     y6, y7, rio, key) \
+	vmovq key, x0; \
+	vpshufb .Lpack_bswap(%rip), x0, x0; \
+	\
+	vpxor 0 * 16(rio), x0, y7; \
+	vpxor 1 * 16(rio), x0, y6; \
+	vpxor 2 * 16(rio), x0, y5; \
+	vpxor 3 * 16(rio), x0, y4; \
+	vpxor 4 * 16(rio), x0, y3; \
+	vpxor 5 * 16(rio), x0, y2; \
+	vpxor 6 * 16(rio), x0, y1; \
+	vpxor 7 * 16(rio), x0, y0; \
+	vpxor 8 * 16(rio), x0, x7; \
+	vpxor 9 * 16(rio), x0, x6; \
+	vpxor 10 * 16(rio), x0, x5; \
+	vpxor 11 * 16(rio), x0, x4; \
+	vpxor 12 * 16(rio), x0, x3; \
+	vpxor 13 * 16(rio), x0, x2; \
+	vpxor 14 * 16(rio), x0, x1; \
+	vpxor 15 * 16(rio), x0, x0;
+
+/* byteslice pre-whitened blocks and store to temporary memory */
+#define inpack16_post(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd) \
+	byteslice_16x16b(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, \
+			 y5, y6, y7, (mem_ab), (mem_cd)); \
+	\
+	vmovdqu x0, 0 * 16(mem_ab); \
+	vmovdqu x1, 1 * 16(mem_ab); \
+	vmovdqu x2, 2 * 16(mem_ab); \
+	vmovdqu x3, 3 * 16(mem_ab); \
+	vmovdqu x4, 4 * 16(mem_ab); \
+	vmovdqu x5, 5 * 16(mem_ab); \
+	vmovdqu x6, 6 * 16(mem_ab); \
+	vmovdqu x7, 7 * 16(mem_ab); \
+	vmovdqu y0, 0 * 16(mem_cd); \
+	vmovdqu y1, 1 * 16(mem_cd); \
+	vmovdqu y2, 2 * 16(mem_cd); \
+	vmovdqu y3, 3 * 16(mem_cd); \
+	vmovdqu y4, 4 * 16(mem_cd); \
+	vmovdqu y5, 5 * 16(mem_cd); \
+	vmovdqu y6, 6 * 16(mem_cd); \
+	vmovdqu y7, 7 * 16(mem_cd);
+
+/* de-byteslice, apply post-whitening and store blocks */
+#define outunpack16(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, \
+		    y5, y6, y7, key, stack_tmp0, stack_tmp1) \
+	byteslice_16x16b(y0, y4, x0, x4, y1, y5, x1, x5, y2, y6, x2, x6, y3, \
+			 y7, x3, x7, stack_tmp0, stack_tmp1); \
+	\
+	vmovdqu x0, stack_tmp0; \
+	\
+	vmovq key, x0; \
+	vpshufb .Lpack_bswap(%rip), x0, x0; \
+	\
+	vpxor x0, y7, y7; \
+	vpxor x0, y6, y6; \
+	vpxor x0, y5, y5; \
+	vpxor x0, y4, y4; \
+	vpxor x0, y3, y3; \
+	vpxor x0, y2, y2; \
+	vpxor x0, y1, y1; \
+	vpxor x0, y0, y0; \
+	vpxor x0, x7, x7; \
+	vpxor x0, x6, x6; \
+	vpxor x0, x5, x5; \
+	vpxor x0, x4, x4; \
+	vpxor x0, x3, x3; \
+	vpxor x0, x2, x2; \
+	vpxor x0, x1, x1; \
+	vpxor stack_tmp0, x0, x0;
+
+#define write_output(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		     y6, y7, rio) \
+	vmovdqu x0, 0 * 16(rio); \
+	vmovdqu x1, 1 * 16(rio); \
+	vmovdqu x2, 2 * 16(rio); \
+	vmovdqu x3, 3 * 16(rio); \
+	vmovdqu x4, 4 * 16(rio); \
+	vmovdqu x5, 5 * 16(rio); \
+	vmovdqu x6, 6 * 16(rio); \
+	vmovdqu x7, 7 * 16(rio); \
+	vmovdqu y0, 8 * 16(rio); \
+	vmovdqu y1, 9 * 16(rio); \
+	vmovdqu y2, 10 * 16(rio); \
+	vmovdqu y3, 11 * 16(rio); \
+	vmovdqu y4, 12 * 16(rio); \
+	vmovdqu y5, 13 * 16(rio); \
+	vmovdqu y6, 14 * 16(rio); \
+	vmovdqu y7, 15 * 16(rio);
+
+
+/* NB: section is mergeable, all elements must be aligned 16-byte blocks */
+.section	.rodata.cst16, "aM", @progbits, 16
+.align 16
+
+#define SHUFB_BYTES(idx) \
+	0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
+
+.Lshufb_16x16b:
+	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3);
+
+.Lpack_bswap:
+	.long 0x00010203
+	.long 0x04050607
+	.long 0x80808080
+	.long 0x80808080
+
+/*
+ * pre-SubByte transform
+ *
+ * pre-lookup for sbox1, sbox2, sbox3:
+ *   swap_bitendianness(
+ *       isom_map_camellia_to_aes(
+ *           camellia_f(
+ *               swap_bitendianess(in)
+ *           )
+ *       )
+ *   )
+ *
+ * (note: '⊕ 0xc5' inside camellia_f())
+ */
+.Lpre_tf_lo_s1:
+	.byte 0x45, 0xe8, 0x40, 0xed, 0x2e, 0x83, 0x2b, 0x86
+	.byte 0x4b, 0xe6, 0x4e, 0xe3, 0x20, 0x8d, 0x25, 0x88
+.Lpre_tf_hi_s1:
+	.byte 0x00, 0x51, 0xf1, 0xa0, 0x8a, 0xdb, 0x7b, 0x2a
+	.byte 0x09, 0x58, 0xf8, 0xa9, 0x83, 0xd2, 0x72, 0x23
+
+/*
+ * pre-SubByte transform
+ *
+ * pre-lookup for sbox4:
+ *   swap_bitendianness(
+ *       isom_map_camellia_to_aes(
+ *           camellia_f(
+ *               swap_bitendianess(in <<< 1)
+ *           )
+ *       )
+ *   )
+ *
+ * (note: '⊕ 0xc5' inside camellia_f())
+ */
+.Lpre_tf_lo_s4:
+	.byte 0x45, 0x40, 0x2e, 0x2b, 0x4b, 0x4e, 0x20, 0x25
+	.byte 0x14, 0x11, 0x7f, 0x7a, 0x1a, 0x1f, 0x71, 0x74
+.Lpre_tf_hi_s4:
+	.byte 0x00, 0xf1, 0x8a, 0x7b, 0x09, 0xf8, 0x83, 0x72
+	.byte 0xad, 0x5c, 0x27, 0xd6, 0xa4, 0x55, 0x2e, 0xdf
+
+/*
+ * post-SubByte transform
+ *
+ * post-lookup for sbox1, sbox4:
+ *  swap_bitendianness(
+ *      camellia_h(
+ *          isom_map_aes_to_camellia(
+ *              swap_bitendianness(
+ *                  aes_inverse_affine_transform(in)
+ *              )
+ *          )
+ *      )
+ *  )
+ *
+ * (note: '⊕ 0x6e' inside camellia_h())
+ */
+.Lpost_tf_lo_s1:
+	.byte 0x3c, 0xcc, 0xcf, 0x3f, 0x32, 0xc2, 0xc1, 0x31
+	.byte 0xdc, 0x2c, 0x2f, 0xdf, 0xd2, 0x22, 0x21, 0xd1
+.Lpost_tf_hi_s1:
+	.byte 0x00, 0xf9, 0x86, 0x7f, 0xd7, 0x2e, 0x51, 0xa8
+	.byte 0xa4, 0x5d, 0x22, 0xdb, 0x73, 0x8a, 0xf5, 0x0c
+
+/*
+ * post-SubByte transform
+ *
+ * post-lookup for sbox2:
+ *  swap_bitendianness(
+ *      camellia_h(
+ *          isom_map_aes_to_camellia(
+ *              swap_bitendianness(
+ *                  aes_inverse_affine_transform(in)
+ *              )
+ *          )
+ *      )
+ *  ) <<< 1
+ *
+ * (note: '⊕ 0x6e' inside camellia_h())
+ */
+.Lpost_tf_lo_s2:
+	.byte 0x78, 0x99, 0x9f, 0x7e, 0x64, 0x85, 0x83, 0x62
+	.byte 0xb9, 0x58, 0x5e, 0xbf, 0xa5, 0x44, 0x42, 0xa3
+.Lpost_tf_hi_s2:
+	.byte 0x00, 0xf3, 0x0d, 0xfe, 0xaf, 0x5c, 0xa2, 0x51
+	.byte 0x49, 0xba, 0x44, 0xb7, 0xe6, 0x15, 0xeb, 0x18
+
+/*
+ * post-SubByte transform
+ *
+ * post-lookup for sbox3:
+ *  swap_bitendianness(
+ *      camellia_h(
+ *          isom_map_aes_to_camellia(
+ *              swap_bitendianness(
+ *                  aes_inverse_affine_transform(in)
+ *              )
+ *          )
+ *      )
+ *  ) >>> 1
+ *
+ * (note: '⊕ 0x6e' inside camellia_h())
+ */
+.Lpost_tf_lo_s3:
+	.byte 0x1e, 0x66, 0xe7, 0x9f, 0x19, 0x61, 0xe0, 0x98
+	.byte 0x6e, 0x16, 0x97, 0xef, 0x69, 0x11, 0x90, 0xe8
+.Lpost_tf_hi_s3:
+	.byte 0x00, 0xfc, 0x43, 0xbf, 0xeb, 0x17, 0xa8, 0x54
+	.byte 0x52, 0xae, 0x11, 0xed, 0xb9, 0x45, 0xfa, 0x06
+
+/* For isolating SubBytes from AESENCLAST, inverse shift row */
+.Linv_shift_row:
+	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
+	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
+
+/* 4-bit mask */
+.section	.rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
+.align 4
+.L0f0f0f0f:
+	.long 0x0f0f0f0f
+
+.text
+
+SYM_FUNC_START_LOCAL(__camellia_enc_blk16)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rax: temporary storage, 256 bytes
+	 *	%xmm0..%xmm15: 16 plaintext blocks
+	 * output:
+	 *	%xmm0..%xmm15: 16 encrypted blocks, order swapped:
+	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
+	 */
+	FRAME_BEGIN
+
+	leaq 8 * 16(%rax), %rcx;
+
+	inpack16_post(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		      %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		      %xmm15, %rax, %rcx);
+
+	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %rcx, 0);
+
+	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+	      %xmm15,
+	      ((key_table + (8) * 8) + 0)(CTX),
+	      ((key_table + (8) * 8) + 4)(CTX),
+	      ((key_table + (8) * 8) + 8)(CTX),
+	      ((key_table + (8) * 8) + 12)(CTX));
+
+	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %rcx, 8);
+
+	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+	      %xmm15,
+	      ((key_table + (16) * 8) + 0)(CTX),
+	      ((key_table + (16) * 8) + 4)(CTX),
+	      ((key_table + (16) * 8) + 8)(CTX),
+	      ((key_table + (16) * 8) + 12)(CTX));
+
+	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %rcx, 16);
+
+	movl $24, %r8d;
+	cmpl $16, key_length(CTX);
+	jne .Lenc_max32;
+
+.Lenc_done:
+	/* load CD for output */
+	vmovdqu 0 * 16(%rcx), %xmm8;
+	vmovdqu 1 * 16(%rcx), %xmm9;
+	vmovdqu 2 * 16(%rcx), %xmm10;
+	vmovdqu 3 * 16(%rcx), %xmm11;
+	vmovdqu 4 * 16(%rcx), %xmm12;
+	vmovdqu 5 * 16(%rcx), %xmm13;
+	vmovdqu 6 * 16(%rcx), %xmm14;
+	vmovdqu 7 * 16(%rcx), %xmm15;
+
+	outunpack16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		    %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		    %xmm15, (key_table)(CTX, %r8, 8), (%rax), 1 * 16(%rax));
+
+	FRAME_END
+	RET;
+
+.align 8
+.Lenc_max32:
+	movl $32, %r8d;
+
+	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+	      %xmm15,
+	      ((key_table + (24) * 8) + 0)(CTX),
+	      ((key_table + (24) * 8) + 4)(CTX),
+	      ((key_table + (24) * 8) + 8)(CTX),
+	      ((key_table + (24) * 8) + 12)(CTX));
+
+	enc_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %rcx, 24);
+
+	jmp .Lenc_done;
+SYM_FUNC_END(__camellia_enc_blk16)
+
+SYM_FUNC_START_LOCAL(__camellia_dec_blk16)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rax: temporary storage, 256 bytes
+	 *	%r8d: 24 for 16 byte key, 32 for larger
+	 *	%xmm0..%xmm15: 16 encrypted blocks
+	 * output:
+	 *	%xmm0..%xmm15: 16 plaintext blocks, order swapped:
+	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
+	 */
+	FRAME_BEGIN
+
+	leaq 8 * 16(%rax), %rcx;
+
+	inpack16_post(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		      %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		      %xmm15, %rax, %rcx);
+
+	cmpl $32, %r8d;
+	je .Ldec_max32;
+
+.Ldec_max24:
+	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %rcx, 16);
+
+	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+	      %xmm15,
+	      ((key_table + (16) * 8) + 8)(CTX),
+	      ((key_table + (16) * 8) + 12)(CTX),
+	      ((key_table + (16) * 8) + 0)(CTX),
+	      ((key_table + (16) * 8) + 4)(CTX));
+
+	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %rcx, 8);
+
+	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+	      %xmm15,
+	      ((key_table + (8) * 8) + 8)(CTX),
+	      ((key_table + (8) * 8) + 12)(CTX),
+	      ((key_table + (8) * 8) + 0)(CTX),
+	      ((key_table + (8) * 8) + 4)(CTX));
+
+	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %rcx, 0);
+
+	/* load CD for output */
+	vmovdqu 0 * 16(%rcx), %xmm8;
+	vmovdqu 1 * 16(%rcx), %xmm9;
+	vmovdqu 2 * 16(%rcx), %xmm10;
+	vmovdqu 3 * 16(%rcx), %xmm11;
+	vmovdqu 4 * 16(%rcx), %xmm12;
+	vmovdqu 5 * 16(%rcx), %xmm13;
+	vmovdqu 6 * 16(%rcx), %xmm14;
+	vmovdqu 7 * 16(%rcx), %xmm15;
+
+	outunpack16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		    %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		    %xmm15, (key_table)(CTX), (%rax), 1 * 16(%rax));
+
+	FRAME_END
+	RET;
+
+.align 8
+.Ldec_max32:
+	dec_rounds16(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rax, %rcx, 24);
+
+	fls16(%rax, %xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+	      %rcx, %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+	      %xmm15,
+	      ((key_table + (24) * 8) + 8)(CTX),
+	      ((key_table + (24) * 8) + 12)(CTX),
+	      ((key_table + (24) * 8) + 0)(CTX),
+	      ((key_table + (24) * 8) + 4)(CTX));
+
+	jmp .Ldec_max24;
+SYM_FUNC_END(__camellia_dec_blk16)
+
+SYM_TYPED_FUNC_START(camellia_ecb_enc_16way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst (16 blocks)
+	 *	%rdx: src (16 blocks)
+	 */
+	 FRAME_BEGIN
+
+	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rdx, (key_table)(CTX));
+
+	/* now dst can be used as temporary buffer (even in src == dst case) */
+	movq	%rsi, %rax;
+
+	call __camellia_enc_blk16;
+
+	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
+		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
+		     %xmm8, %rsi);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(camellia_ecb_enc_16way)
+
+SYM_TYPED_FUNC_START(camellia_ecb_dec_16way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst (16 blocks)
+	 *	%rdx: src (16 blocks)
+	 */
+	 FRAME_BEGIN
+
+	cmpl $16, key_length(CTX);
+	movl $32, %r8d;
+	movl $24, %eax;
+	cmovel %eax, %r8d; /* max */
+
+	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rdx, (key_table)(CTX, %r8, 8));
+
+	/* now dst can be used as temporary buffer (even in src == dst case) */
+	movq	%rsi, %rax;
+
+	call __camellia_dec_blk16;
+
+	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
+		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
+		     %xmm8, %rsi);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(camellia_ecb_dec_16way)
+
+SYM_TYPED_FUNC_START(camellia_cbc_dec_16way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst (16 blocks)
+	 *	%rdx: src (16 blocks)
+	 */
+	FRAME_BEGIN
+
+	cmpl $16, key_length(CTX);
+	movl $32, %r8d;
+	movl $24, %eax;
+	cmovel %eax, %r8d; /* max */
+
+	inpack16_pre(%xmm0, %xmm1, %xmm2, %xmm3, %xmm4, %xmm5, %xmm6, %xmm7,
+		     %xmm8, %xmm9, %xmm10, %xmm11, %xmm12, %xmm13, %xmm14,
+		     %xmm15, %rdx, (key_table)(CTX, %r8, 8));
+
+	/*
+	 * dst might still be in-use (in case dst == src), so use stack for
+	 * temporary storage.
+	 */
+	subq $(16 * 16), %rsp;
+	movq %rsp, %rax;
+
+	call __camellia_dec_blk16;
+
+	addq $(16 * 16), %rsp;
+
+	vpxor (0 * 16)(%rdx), %xmm6, %xmm6;
+	vpxor (1 * 16)(%rdx), %xmm5, %xmm5;
+	vpxor (2 * 16)(%rdx), %xmm4, %xmm4;
+	vpxor (3 * 16)(%rdx), %xmm3, %xmm3;
+	vpxor (4 * 16)(%rdx), %xmm2, %xmm2;
+	vpxor (5 * 16)(%rdx), %xmm1, %xmm1;
+	vpxor (6 * 16)(%rdx), %xmm0, %xmm0;
+	vpxor (7 * 16)(%rdx), %xmm15, %xmm15;
+	vpxor (8 * 16)(%rdx), %xmm14, %xmm14;
+	vpxor (9 * 16)(%rdx), %xmm13, %xmm13;
+	vpxor (10 * 16)(%rdx), %xmm12, %xmm12;
+	vpxor (11 * 16)(%rdx), %xmm11, %xmm11;
+	vpxor (12 * 16)(%rdx), %xmm10, %xmm10;
+	vpxor (13 * 16)(%rdx), %xmm9, %xmm9;
+	vpxor (14 * 16)(%rdx), %xmm8, %xmm8;
+	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
+		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
+		     %xmm8, %rsi);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(camellia_cbc_dec_16way)
diff --git a/arch/x86/crypto/camellia-aesni-avx2-asm_64.S b/arch/x86/crypto/camellia-aesni-avx2-asm_64.S
new file mode 100644
index 000000000..b1c9b9450
--- /dev/null
+++ b/arch/x86/crypto/camellia-aesni-avx2-asm_64.S
@@ -0,0 +1,1048 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * x86_64/AVX2/AES-NI assembler implementation of Camellia
+ *
+ * Copyright © 2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ */
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+#include <asm/frame.h>
+
+#define CAMELLIA_TABLE_BYTE_LEN 272
+
+/* struct camellia_ctx: */
+#define key_table 0
+#define key_length CAMELLIA_TABLE_BYTE_LEN
+
+/* register macros */
+#define CTX %rdi
+#define RIO %r8
+
+/**********************************************************************
+  helper macros
+ **********************************************************************/
+#define filter_8bit(x, lo_t, hi_t, mask4bit, tmp0) \
+	vpand x, mask4bit, tmp0; \
+	vpandn x, mask4bit, x; \
+	vpsrld $4, x, x; \
+	\
+	vpshufb tmp0, lo_t, tmp0; \
+	vpshufb x, hi_t, x; \
+	vpxor tmp0, x, x;
+
+#define ymm0_x xmm0
+#define ymm1_x xmm1
+#define ymm2_x xmm2
+#define ymm3_x xmm3
+#define ymm4_x xmm4
+#define ymm5_x xmm5
+#define ymm6_x xmm6
+#define ymm7_x xmm7
+#define ymm8_x xmm8
+#define ymm9_x xmm9
+#define ymm10_x xmm10
+#define ymm11_x xmm11
+#define ymm12_x xmm12
+#define ymm13_x xmm13
+#define ymm14_x xmm14
+#define ymm15_x xmm15
+
+/**********************************************************************
+  32-way camellia
+ **********************************************************************/
+
+/*
+ * IN:
+ *   x0..x7: byte-sliced AB state
+ *   mem_cd: register pointer storing CD state
+ *   key: index for key material
+ * OUT:
+ *   x0..x7: new byte-sliced CD state
+ */
+#define roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, t0, t1, t2, t3, t4, t5, t6, \
+		  t7, mem_cd, key) \
+	/* \
+	 * S-function with AES subbytes \
+	 */ \
+	vbroadcasti128 .Linv_shift_row(%rip), t4; \
+	vpbroadcastd .L0f0f0f0f(%rip), t7; \
+	vbroadcasti128 .Lpre_tf_lo_s1(%rip), t5; \
+	vbroadcasti128 .Lpre_tf_hi_s1(%rip), t6; \
+	vbroadcasti128 .Lpre_tf_lo_s4(%rip), t2; \
+	vbroadcasti128 .Lpre_tf_hi_s4(%rip), t3; \
+	\
+	/* AES inverse shift rows */ \
+	vpshufb t4, x0, x0; \
+	vpshufb t4, x7, x7; \
+	vpshufb t4, x3, x3; \
+	vpshufb t4, x6, x6; \
+	vpshufb t4, x2, x2; \
+	vpshufb t4, x5, x5; \
+	vpshufb t4, x1, x1; \
+	vpshufb t4, x4, x4; \
+	\
+	/* prefilter sboxes 1, 2 and 3 */ \
+	/* prefilter sbox 4 */ \
+	filter_8bit(x0, t5, t6, t7, t4); \
+	filter_8bit(x7, t5, t6, t7, t4); \
+	vextracti128 $1, x0, t0##_x; \
+	vextracti128 $1, x7, t1##_x; \
+	filter_8bit(x3, t2, t3, t7, t4); \
+	filter_8bit(x6, t2, t3, t7, t4); \
+	vextracti128 $1, x3, t3##_x; \
+	vextracti128 $1, x6, t2##_x; \
+	filter_8bit(x2, t5, t6, t7, t4); \
+	filter_8bit(x5, t5, t6, t7, t4); \
+	filter_8bit(x1, t5, t6, t7, t4); \
+	filter_8bit(x4, t5, t6, t7, t4); \
+	\
+	vpxor t4##_x, t4##_x, t4##_x; \
+	\
+	/* AES subbytes + AES shift rows */ \
+	vextracti128 $1, x2, t6##_x; \
+	vextracti128 $1, x5, t5##_x; \
+	vaesenclast t4##_x, x0##_x, x0##_x; \
+	vaesenclast t4##_x, t0##_x, t0##_x; \
+	vinserti128 $1, t0##_x, x0, x0; \
+	vaesenclast t4##_x, x7##_x, x7##_x; \
+	vaesenclast t4##_x, t1##_x, t1##_x; \
+	vinserti128 $1, t1##_x, x7, x7; \
+	vaesenclast t4##_x, x3##_x, x3##_x; \
+	vaesenclast t4##_x, t3##_x, t3##_x; \
+	vinserti128 $1, t3##_x, x3, x3; \
+	vaesenclast t4##_x, x6##_x, x6##_x; \
+	vaesenclast t4##_x, t2##_x, t2##_x; \
+	vinserti128 $1, t2##_x, x6, x6; \
+	vextracti128 $1, x1, t3##_x; \
+	vextracti128 $1, x4, t2##_x; \
+	vbroadcasti128 .Lpost_tf_lo_s1(%rip), t0; \
+	vbroadcasti128 .Lpost_tf_hi_s1(%rip), t1; \
+	vaesenclast t4##_x, x2##_x, x2##_x; \
+	vaesenclast t4##_x, t6##_x, t6##_x; \
+	vinserti128 $1, t6##_x, x2, x2; \
+	vaesenclast t4##_x, x5##_x, x5##_x; \
+	vaesenclast t4##_x, t5##_x, t5##_x; \
+	vinserti128 $1, t5##_x, x5, x5; \
+	vaesenclast t4##_x, x1##_x, x1##_x; \
+	vaesenclast t4##_x, t3##_x, t3##_x; \
+	vinserti128 $1, t3##_x, x1, x1; \
+	vaesenclast t4##_x, x4##_x, x4##_x; \
+	vaesenclast t4##_x, t2##_x, t2##_x; \
+	vinserti128 $1, t2##_x, x4, x4; \
+	\
+	/* postfilter sboxes 1 and 4 */ \
+	vbroadcasti128 .Lpost_tf_lo_s3(%rip), t2; \
+	vbroadcasti128 .Lpost_tf_hi_s3(%rip), t3; \
+	filter_8bit(x0, t0, t1, t7, t6); \
+	filter_8bit(x7, t0, t1, t7, t6); \
+	filter_8bit(x3, t0, t1, t7, t6); \
+	filter_8bit(x6, t0, t1, t7, t6); \
+	\
+	/* postfilter sbox 3 */ \
+	vbroadcasti128 .Lpost_tf_lo_s2(%rip), t4; \
+	vbroadcasti128 .Lpost_tf_hi_s2(%rip), t5; \
+	filter_8bit(x2, t2, t3, t7, t6); \
+	filter_8bit(x5, t2, t3, t7, t6); \
+	\
+	vpbroadcastq key, t0; /* higher 64-bit duplicate ignored */ \
+	\
+	/* postfilter sbox 2 */ \
+	filter_8bit(x1, t4, t5, t7, t2); \
+	filter_8bit(x4, t4, t5, t7, t2); \
+	vpxor t7, t7, t7; \
+	\
+	vpsrldq $1, t0, t1; \
+	vpsrldq $2, t0, t2; \
+	vpshufb t7, t1, t1; \
+	vpsrldq $3, t0, t3; \
+	\
+	/* P-function */ \
+	vpxor x5, x0, x0; \
+	vpxor x6, x1, x1; \
+	vpxor x7, x2, x2; \
+	vpxor x4, x3, x3; \
+	\
+	vpshufb t7, t2, t2; \
+	vpsrldq $4, t0, t4; \
+	vpshufb t7, t3, t3; \
+	vpsrldq $5, t0, t5; \
+	vpshufb t7, t4, t4; \
+	\
+	vpxor x2, x4, x4; \
+	vpxor x3, x5, x5; \
+	vpxor x0, x6, x6; \
+	vpxor x1, x7, x7; \
+	\
+	vpsrldq $6, t0, t6; \
+	vpshufb t7, t5, t5; \
+	vpshufb t7, t6, t6; \
+	\
+	vpxor x7, x0, x0; \
+	vpxor x4, x1, x1; \
+	vpxor x5, x2, x2; \
+	vpxor x6, x3, x3; \
+	\
+	vpxor x3, x4, x4; \
+	vpxor x0, x5, x5; \
+	vpxor x1, x6, x6; \
+	vpxor x2, x7, x7; /* note: high and low parts swapped */ \
+	\
+	/* Add key material and result to CD (x becomes new CD) */ \
+	\
+	vpxor t6, x1, x1; \
+	vpxor 5 * 32(mem_cd), x1, x1; \
+	\
+	vpsrldq $7, t0, t6; \
+	vpshufb t7, t0, t0; \
+	vpshufb t7, t6, t7; \
+	\
+	vpxor t7, x0, x0; \
+	vpxor 4 * 32(mem_cd), x0, x0; \
+	\
+	vpxor t5, x2, x2; \
+	vpxor 6 * 32(mem_cd), x2, x2; \
+	\
+	vpxor t4, x3, x3; \
+	vpxor 7 * 32(mem_cd), x3, x3; \
+	\
+	vpxor t3, x4, x4; \
+	vpxor 0 * 32(mem_cd), x4, x4; \
+	\
+	vpxor t2, x5, x5; \
+	vpxor 1 * 32(mem_cd), x5, x5; \
+	\
+	vpxor t1, x6, x6; \
+	vpxor 2 * 32(mem_cd), x6, x6; \
+	\
+	vpxor t0, x7, x7; \
+	vpxor 3 * 32(mem_cd), x7, x7;
+
+/*
+ * Size optimization... with inlined roundsm32 binary would be over 5 times
+ * larger and would only marginally faster.
+ */
+SYM_FUNC_START_LOCAL(roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
+	roundsm32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		  %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14, %ymm15,
+		  %rcx, (%r9));
+	RET;
+SYM_FUNC_END(roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd)
+
+SYM_FUNC_START_LOCAL(roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
+	roundsm32(%ymm4, %ymm5, %ymm6, %ymm7, %ymm0, %ymm1, %ymm2, %ymm3,
+		  %ymm12, %ymm13, %ymm14, %ymm15, %ymm8, %ymm9, %ymm10, %ymm11,
+		  %rax, (%r9));
+	RET;
+SYM_FUNC_END(roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
+
+/*
+ * IN/OUT:
+ *  x0..x7: byte-sliced AB state preloaded
+ *  mem_ab: byte-sliced AB state in memory
+ *  mem_cb: byte-sliced CD state in memory
+ */
+#define two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, i, dir, store_ab) \
+	leaq (key_table + (i) * 8)(CTX), %r9; \
+	call roundsm32_x0_x1_x2_x3_x4_x5_x6_x7_y0_y1_y2_y3_y4_y5_y6_y7_cd; \
+	\
+	vmovdqu x0, 4 * 32(mem_cd); \
+	vmovdqu x1, 5 * 32(mem_cd); \
+	vmovdqu x2, 6 * 32(mem_cd); \
+	vmovdqu x3, 7 * 32(mem_cd); \
+	vmovdqu x4, 0 * 32(mem_cd); \
+	vmovdqu x5, 1 * 32(mem_cd); \
+	vmovdqu x6, 2 * 32(mem_cd); \
+	vmovdqu x7, 3 * 32(mem_cd); \
+	\
+	leaq (key_table + ((i) + (dir)) * 8)(CTX), %r9; \
+	call roundsm32_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab; \
+	\
+	store_ab(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab);
+
+#define dummy_store(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) /* do nothing */
+
+#define store_ab_state(x0, x1, x2, x3, x4, x5, x6, x7, mem_ab) \
+	/* Store new AB state */ \
+	vmovdqu x4, 4 * 32(mem_ab); \
+	vmovdqu x5, 5 * 32(mem_ab); \
+	vmovdqu x6, 6 * 32(mem_ab); \
+	vmovdqu x7, 7 * 32(mem_ab); \
+	vmovdqu x0, 0 * 32(mem_ab); \
+	vmovdqu x1, 1 * 32(mem_ab); \
+	vmovdqu x2, 2 * 32(mem_ab); \
+	vmovdqu x3, 3 * 32(mem_ab);
+
+#define enc_rounds32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, i) \
+	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 2, 1, store_ab_state); \
+	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 4, 1, store_ab_state); \
+	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 6, 1, dummy_store);
+
+#define dec_rounds32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, i) \
+	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 7, -1, store_ab_state); \
+	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 5, -1, store_ab_state); \
+	two_roundsm32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd, (i) + 3, -1, dummy_store);
+
+/*
+ * IN:
+ *  v0..3: byte-sliced 32-bit integers
+ * OUT:
+ *  v0..3: (IN <<< 1)
+ */
+#define rol32_1_32(v0, v1, v2, v3, t0, t1, t2, zero) \
+	vpcmpgtb v0, zero, t0; \
+	vpaddb v0, v0, v0; \
+	vpabsb t0, t0; \
+	\
+	vpcmpgtb v1, zero, t1; \
+	vpaddb v1, v1, v1; \
+	vpabsb t1, t1; \
+	\
+	vpcmpgtb v2, zero, t2; \
+	vpaddb v2, v2, v2; \
+	vpabsb t2, t2; \
+	\
+	vpor t0, v1, v1; \
+	\
+	vpcmpgtb v3, zero, t0; \
+	vpaddb v3, v3, v3; \
+	vpabsb t0, t0; \
+	\
+	vpor t1, v2, v2; \
+	vpor t2, v3, v3; \
+	vpor t0, v0, v0;
+
+/*
+ * IN:
+ *   r: byte-sliced AB state in memory
+ *   l: byte-sliced CD state in memory
+ * OUT:
+ *   x0..x7: new byte-sliced CD state
+ */
+#define fls32(l, l0, l1, l2, l3, l4, l5, l6, l7, r, t0, t1, t2, t3, tt0, \
+	      tt1, tt2, tt3, kll, klr, krl, krr) \
+	/* \
+	 * t0 = kll; \
+	 * t0 &= ll; \
+	 * lr ^= rol32(t0, 1); \
+	 */ \
+	vpbroadcastd kll, t0; /* only lowest 32-bit used */ \
+	vpxor tt0, tt0, tt0; \
+	vpshufb tt0, t0, t3; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t2; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t1; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t0; \
+	\
+	vpand l0, t0, t0; \
+	vpand l1, t1, t1; \
+	vpand l2, t2, t2; \
+	vpand l3, t3, t3; \
+	\
+	rol32_1_32(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
+	\
+	vpxor l4, t0, l4; \
+	vpbroadcastd krr, t0; /* only lowest 32-bit used */ \
+	vmovdqu l4, 4 * 32(l); \
+	vpxor l5, t1, l5; \
+	vmovdqu l5, 5 * 32(l); \
+	vpxor l6, t2, l6; \
+	vmovdqu l6, 6 * 32(l); \
+	vpxor l7, t3, l7; \
+	vmovdqu l7, 7 * 32(l); \
+	\
+	/* \
+	 * t2 = krr; \
+	 * t2 |= rr; \
+	 * rl ^= t2; \
+	 */ \
+	\
+	vpshufb tt0, t0, t3; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t2; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t1; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t0; \
+	\
+	vpor 4 * 32(r), t0, t0; \
+	vpor 5 * 32(r), t1, t1; \
+	vpor 6 * 32(r), t2, t2; \
+	vpor 7 * 32(r), t3, t3; \
+	\
+	vpxor 0 * 32(r), t0, t0; \
+	vpxor 1 * 32(r), t1, t1; \
+	vpxor 2 * 32(r), t2, t2; \
+	vpxor 3 * 32(r), t3, t3; \
+	vmovdqu t0, 0 * 32(r); \
+	vpbroadcastd krl, t0; /* only lowest 32-bit used */ \
+	vmovdqu t1, 1 * 32(r); \
+	vmovdqu t2, 2 * 32(r); \
+	vmovdqu t3, 3 * 32(r); \
+	\
+	/* \
+	 * t2 = krl; \
+	 * t2 &= rl; \
+	 * rr ^= rol32(t2, 1); \
+	 */ \
+	vpshufb tt0, t0, t3; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t2; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t1; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t0; \
+	\
+	vpand 0 * 32(r), t0, t0; \
+	vpand 1 * 32(r), t1, t1; \
+	vpand 2 * 32(r), t2, t2; \
+	vpand 3 * 32(r), t3, t3; \
+	\
+	rol32_1_32(t3, t2, t1, t0, tt1, tt2, tt3, tt0); \
+	\
+	vpxor 4 * 32(r), t0, t0; \
+	vpxor 5 * 32(r), t1, t1; \
+	vpxor 6 * 32(r), t2, t2; \
+	vpxor 7 * 32(r), t3, t3; \
+	vmovdqu t0, 4 * 32(r); \
+	vpbroadcastd klr, t0; /* only lowest 32-bit used */ \
+	vmovdqu t1, 5 * 32(r); \
+	vmovdqu t2, 6 * 32(r); \
+	vmovdqu t3, 7 * 32(r); \
+	\
+	/* \
+	 * t0 = klr; \
+	 * t0 |= lr; \
+	 * ll ^= t0; \
+	 */ \
+	\
+	vpshufb tt0, t0, t3; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t2; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t1; \
+	vpsrldq $1, t0, t0; \
+	vpshufb tt0, t0, t0; \
+	\
+	vpor l4, t0, t0; \
+	vpor l5, t1, t1; \
+	vpor l6, t2, t2; \
+	vpor l7, t3, t3; \
+	\
+	vpxor l0, t0, l0; \
+	vmovdqu l0, 0 * 32(l); \
+	vpxor l1, t1, l1; \
+	vmovdqu l1, 1 * 32(l); \
+	vpxor l2, t2, l2; \
+	vmovdqu l2, 2 * 32(l); \
+	vpxor l3, t3, l3; \
+	vmovdqu l3, 3 * 32(l);
+
+#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
+	vpunpckhdq x1, x0, t2; \
+	vpunpckldq x1, x0, x0; \
+	\
+	vpunpckldq x3, x2, t1; \
+	vpunpckhdq x3, x2, x2; \
+	\
+	vpunpckhqdq t1, x0, x1; \
+	vpunpcklqdq t1, x0, x0; \
+	\
+	vpunpckhqdq x2, t2, x3; \
+	vpunpcklqdq x2, t2, x2;
+
+#define byteslice_16x16b_fast(a0, b0, c0, d0, a1, b1, c1, d1, a2, b2, c2, d2, \
+			      a3, b3, c3, d3, st0, st1) \
+	vmovdqu d2, st0; \
+	vmovdqu d3, st1; \
+	transpose_4x4(a0, a1, a2, a3, d2, d3); \
+	transpose_4x4(b0, b1, b2, b3, d2, d3); \
+	vmovdqu st0, d2; \
+	vmovdqu st1, d3; \
+	\
+	vmovdqu a0, st0; \
+	vmovdqu a1, st1; \
+	transpose_4x4(c0, c1, c2, c3, a0, a1); \
+	transpose_4x4(d0, d1, d2, d3, a0, a1); \
+	\
+	vbroadcasti128 .Lshufb_16x16b(%rip), a0; \
+	vmovdqu st1, a1; \
+	vpshufb a0, a2, a2; \
+	vpshufb a0, a3, a3; \
+	vpshufb a0, b0, b0; \
+	vpshufb a0, b1, b1; \
+	vpshufb a0, b2, b2; \
+	vpshufb a0, b3, b3; \
+	vpshufb a0, a1, a1; \
+	vpshufb a0, c0, c0; \
+	vpshufb a0, c1, c1; \
+	vpshufb a0, c2, c2; \
+	vpshufb a0, c3, c3; \
+	vpshufb a0, d0, d0; \
+	vpshufb a0, d1, d1; \
+	vpshufb a0, d2, d2; \
+	vpshufb a0, d3, d3; \
+	vmovdqu d3, st1; \
+	vmovdqu st0, d3; \
+	vpshufb a0, d3, a0; \
+	vmovdqu d2, st0; \
+	\
+	transpose_4x4(a0, b0, c0, d0, d2, d3); \
+	transpose_4x4(a1, b1, c1, d1, d2, d3); \
+	vmovdqu st0, d2; \
+	vmovdqu st1, d3; \
+	\
+	vmovdqu b0, st0; \
+	vmovdqu b1, st1; \
+	transpose_4x4(a2, b2, c2, d2, b0, b1); \
+	transpose_4x4(a3, b3, c3, d3, b0, b1); \
+	vmovdqu st0, b0; \
+	vmovdqu st1, b1; \
+	/* does not adjust output bytes inside vectors */
+
+/* load blocks to registers and apply pre-whitening */
+#define inpack32_pre(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		     y6, y7, rio, key) \
+	vpbroadcastq key, x0; \
+	vpshufb .Lpack_bswap(%rip), x0, x0; \
+	\
+	vpxor 0 * 32(rio), x0, y7; \
+	vpxor 1 * 32(rio), x0, y6; \
+	vpxor 2 * 32(rio), x0, y5; \
+	vpxor 3 * 32(rio), x0, y4; \
+	vpxor 4 * 32(rio), x0, y3; \
+	vpxor 5 * 32(rio), x0, y2; \
+	vpxor 6 * 32(rio), x0, y1; \
+	vpxor 7 * 32(rio), x0, y0; \
+	vpxor 8 * 32(rio), x0, x7; \
+	vpxor 9 * 32(rio), x0, x6; \
+	vpxor 10 * 32(rio), x0, x5; \
+	vpxor 11 * 32(rio), x0, x4; \
+	vpxor 12 * 32(rio), x0, x3; \
+	vpxor 13 * 32(rio), x0, x2; \
+	vpxor 14 * 32(rio), x0, x1; \
+	vpxor 15 * 32(rio), x0, x0;
+
+/* byteslice pre-whitened blocks and store to temporary memory */
+#define inpack32_post(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		      y6, y7, mem_ab, mem_cd) \
+	byteslice_16x16b_fast(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, \
+			      y4, y5, y6, y7, (mem_ab), (mem_cd)); \
+	\
+	vmovdqu x0, 0 * 32(mem_ab); \
+	vmovdqu x1, 1 * 32(mem_ab); \
+	vmovdqu x2, 2 * 32(mem_ab); \
+	vmovdqu x3, 3 * 32(mem_ab); \
+	vmovdqu x4, 4 * 32(mem_ab); \
+	vmovdqu x5, 5 * 32(mem_ab); \
+	vmovdqu x6, 6 * 32(mem_ab); \
+	vmovdqu x7, 7 * 32(mem_ab); \
+	vmovdqu y0, 0 * 32(mem_cd); \
+	vmovdqu y1, 1 * 32(mem_cd); \
+	vmovdqu y2, 2 * 32(mem_cd); \
+	vmovdqu y3, 3 * 32(mem_cd); \
+	vmovdqu y4, 4 * 32(mem_cd); \
+	vmovdqu y5, 5 * 32(mem_cd); \
+	vmovdqu y6, 6 * 32(mem_cd); \
+	vmovdqu y7, 7 * 32(mem_cd);
+
+/* de-byteslice, apply post-whitening and store blocks */
+#define outunpack32(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, \
+		    y5, y6, y7, key, stack_tmp0, stack_tmp1) \
+	byteslice_16x16b_fast(y0, y4, x0, x4, y1, y5, x1, x5, y2, y6, x2, x6, \
+			      y3, y7, x3, x7, stack_tmp0, stack_tmp1); \
+	\
+	vmovdqu x0, stack_tmp0; \
+	\
+	vpbroadcastq key, x0; \
+	vpshufb .Lpack_bswap(%rip), x0, x0; \
+	\
+	vpxor x0, y7, y7; \
+	vpxor x0, y6, y6; \
+	vpxor x0, y5, y5; \
+	vpxor x0, y4, y4; \
+	vpxor x0, y3, y3; \
+	vpxor x0, y2, y2; \
+	vpxor x0, y1, y1; \
+	vpxor x0, y0, y0; \
+	vpxor x0, x7, x7; \
+	vpxor x0, x6, x6; \
+	vpxor x0, x5, x5; \
+	vpxor x0, x4, x4; \
+	vpxor x0, x3, x3; \
+	vpxor x0, x2, x2; \
+	vpxor x0, x1, x1; \
+	vpxor stack_tmp0, x0, x0;
+
+#define write_output(x0, x1, x2, x3, x4, x5, x6, x7, y0, y1, y2, y3, y4, y5, \
+		     y6, y7, rio) \
+	vmovdqu x0, 0 * 32(rio); \
+	vmovdqu x1, 1 * 32(rio); \
+	vmovdqu x2, 2 * 32(rio); \
+	vmovdqu x3, 3 * 32(rio); \
+	vmovdqu x4, 4 * 32(rio); \
+	vmovdqu x5, 5 * 32(rio); \
+	vmovdqu x6, 6 * 32(rio); \
+	vmovdqu x7, 7 * 32(rio); \
+	vmovdqu y0, 8 * 32(rio); \
+	vmovdqu y1, 9 * 32(rio); \
+	vmovdqu y2, 10 * 32(rio); \
+	vmovdqu y3, 11 * 32(rio); \
+	vmovdqu y4, 12 * 32(rio); \
+	vmovdqu y5, 13 * 32(rio); \
+	vmovdqu y6, 14 * 32(rio); \
+	vmovdqu y7, 15 * 32(rio);
+
+
+.section	.rodata.cst32.shufb_16x16b, "aM", @progbits, 32
+.align 32
+#define SHUFB_BYTES(idx) \
+	0 + (idx), 4 + (idx), 8 + (idx), 12 + (idx)
+.Lshufb_16x16b:
+	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
+	.byte SHUFB_BYTES(0), SHUFB_BYTES(1), SHUFB_BYTES(2), SHUFB_BYTES(3)
+
+.section	.rodata.cst32.pack_bswap, "aM", @progbits, 32
+.align 32
+.Lpack_bswap:
+	.long 0x00010203, 0x04050607, 0x80808080, 0x80808080
+	.long 0x00010203, 0x04050607, 0x80808080, 0x80808080
+
+/* NB: section is mergeable, all elements must be aligned 16-byte blocks */
+.section	.rodata.cst16, "aM", @progbits, 16
+.align 16
+
+/*
+ * pre-SubByte transform
+ *
+ * pre-lookup for sbox1, sbox2, sbox3:
+ *   swap_bitendianness(
+ *       isom_map_camellia_to_aes(
+ *           camellia_f(
+ *               swap_bitendianess(in)
+ *           )
+ *       )
+ *   )
+ *
+ * (note: '⊕ 0xc5' inside camellia_f())
+ */
+.Lpre_tf_lo_s1:
+	.byte 0x45, 0xe8, 0x40, 0xed, 0x2e, 0x83, 0x2b, 0x86
+	.byte 0x4b, 0xe6, 0x4e, 0xe3, 0x20, 0x8d, 0x25, 0x88
+.Lpre_tf_hi_s1:
+	.byte 0x00, 0x51, 0xf1, 0xa0, 0x8a, 0xdb, 0x7b, 0x2a
+	.byte 0x09, 0x58, 0xf8, 0xa9, 0x83, 0xd2, 0x72, 0x23
+
+/*
+ * pre-SubByte transform
+ *
+ * pre-lookup for sbox4:
+ *   swap_bitendianness(
+ *       isom_map_camellia_to_aes(
+ *           camellia_f(
+ *               swap_bitendianess(in <<< 1)
+ *           )
+ *       )
+ *   )
+ *
+ * (note: '⊕ 0xc5' inside camellia_f())
+ */
+.Lpre_tf_lo_s4:
+	.byte 0x45, 0x40, 0x2e, 0x2b, 0x4b, 0x4e, 0x20, 0x25
+	.byte 0x14, 0x11, 0x7f, 0x7a, 0x1a, 0x1f, 0x71, 0x74
+.Lpre_tf_hi_s4:
+	.byte 0x00, 0xf1, 0x8a, 0x7b, 0x09, 0xf8, 0x83, 0x72
+	.byte 0xad, 0x5c, 0x27, 0xd6, 0xa4, 0x55, 0x2e, 0xdf
+
+/*
+ * post-SubByte transform
+ *
+ * post-lookup for sbox1, sbox4:
+ *  swap_bitendianness(
+ *      camellia_h(
+ *          isom_map_aes_to_camellia(
+ *              swap_bitendianness(
+ *                  aes_inverse_affine_transform(in)
+ *              )
+ *          )
+ *      )
+ *  )
+ *
+ * (note: '⊕ 0x6e' inside camellia_h())
+ */
+.Lpost_tf_lo_s1:
+	.byte 0x3c, 0xcc, 0xcf, 0x3f, 0x32, 0xc2, 0xc1, 0x31
+	.byte 0xdc, 0x2c, 0x2f, 0xdf, 0xd2, 0x22, 0x21, 0xd1
+.Lpost_tf_hi_s1:
+	.byte 0x00, 0xf9, 0x86, 0x7f, 0xd7, 0x2e, 0x51, 0xa8
+	.byte 0xa4, 0x5d, 0x22, 0xdb, 0x73, 0x8a, 0xf5, 0x0c
+
+/*
+ * post-SubByte transform
+ *
+ * post-lookup for sbox2:
+ *  swap_bitendianness(
+ *      camellia_h(
+ *          isom_map_aes_to_camellia(
+ *              swap_bitendianness(
+ *                  aes_inverse_affine_transform(in)
+ *              )
+ *          )
+ *      )
+ *  ) <<< 1
+ *
+ * (note: '⊕ 0x6e' inside camellia_h())
+ */
+.Lpost_tf_lo_s2:
+	.byte 0x78, 0x99, 0x9f, 0x7e, 0x64, 0x85, 0x83, 0x62
+	.byte 0xb9, 0x58, 0x5e, 0xbf, 0xa5, 0x44, 0x42, 0xa3
+.Lpost_tf_hi_s2:
+	.byte 0x00, 0xf3, 0x0d, 0xfe, 0xaf, 0x5c, 0xa2, 0x51
+	.byte 0x49, 0xba, 0x44, 0xb7, 0xe6, 0x15, 0xeb, 0x18
+
+/*
+ * post-SubByte transform
+ *
+ * post-lookup for sbox3:
+ *  swap_bitendianness(
+ *      camellia_h(
+ *          isom_map_aes_to_camellia(
+ *              swap_bitendianness(
+ *                  aes_inverse_affine_transform(in)
+ *              )
+ *          )
+ *      )
+ *  ) >>> 1
+ *
+ * (note: '⊕ 0x6e' inside camellia_h())
+ */
+.Lpost_tf_lo_s3:
+	.byte 0x1e, 0x66, 0xe7, 0x9f, 0x19, 0x61, 0xe0, 0x98
+	.byte 0x6e, 0x16, 0x97, 0xef, 0x69, 0x11, 0x90, 0xe8
+.Lpost_tf_hi_s3:
+	.byte 0x00, 0xfc, 0x43, 0xbf, 0xeb, 0x17, 0xa8, 0x54
+	.byte 0x52, 0xae, 0x11, 0xed, 0xb9, 0x45, 0xfa, 0x06
+
+/* For isolating SubBytes from AESENCLAST, inverse shift row */
+.Linv_shift_row:
+	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
+	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
+
+.section	.rodata.cst4.L0f0f0f0f, "aM", @progbits, 4
+.align 4
+/* 4-bit mask */
+.L0f0f0f0f:
+	.long 0x0f0f0f0f
+
+.text
+
+SYM_FUNC_START_LOCAL(__camellia_enc_blk32)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rax: temporary storage, 512 bytes
+	 *	%ymm0..%ymm15: 32 plaintext blocks
+	 * output:
+	 *	%ymm0..%ymm15: 32 encrypted blocks, order swapped:
+	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
+	 */
+	FRAME_BEGIN
+
+	leaq 8 * 32(%rax), %rcx;
+
+	inpack32_post(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		      %ymm15, %rax, %rcx);
+
+	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %rcx, 0);
+
+	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+	      %ymm15,
+	      ((key_table + (8) * 8) + 0)(CTX),
+	      ((key_table + (8) * 8) + 4)(CTX),
+	      ((key_table + (8) * 8) + 8)(CTX),
+	      ((key_table + (8) * 8) + 12)(CTX));
+
+	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %rcx, 8);
+
+	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+	      %ymm15,
+	      ((key_table + (16) * 8) + 0)(CTX),
+	      ((key_table + (16) * 8) + 4)(CTX),
+	      ((key_table + (16) * 8) + 8)(CTX),
+	      ((key_table + (16) * 8) + 12)(CTX));
+
+	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %rcx, 16);
+
+	movl $24, %r8d;
+	cmpl $16, key_length(CTX);
+	jne .Lenc_max32;
+
+.Lenc_done:
+	/* load CD for output */
+	vmovdqu 0 * 32(%rcx), %ymm8;
+	vmovdqu 1 * 32(%rcx), %ymm9;
+	vmovdqu 2 * 32(%rcx), %ymm10;
+	vmovdqu 3 * 32(%rcx), %ymm11;
+	vmovdqu 4 * 32(%rcx), %ymm12;
+	vmovdqu 5 * 32(%rcx), %ymm13;
+	vmovdqu 6 * 32(%rcx), %ymm14;
+	vmovdqu 7 * 32(%rcx), %ymm15;
+
+	outunpack32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		    %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		    %ymm15, (key_table)(CTX, %r8, 8), (%rax), 1 * 32(%rax));
+
+	FRAME_END
+	RET;
+
+.align 8
+.Lenc_max32:
+	movl $32, %r8d;
+
+	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+	      %ymm15,
+	      ((key_table + (24) * 8) + 0)(CTX),
+	      ((key_table + (24) * 8) + 4)(CTX),
+	      ((key_table + (24) * 8) + 8)(CTX),
+	      ((key_table + (24) * 8) + 12)(CTX));
+
+	enc_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %rcx, 24);
+
+	jmp .Lenc_done;
+SYM_FUNC_END(__camellia_enc_blk32)
+
+SYM_FUNC_START_LOCAL(__camellia_dec_blk32)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rax: temporary storage, 512 bytes
+	 *	%r8d: 24 for 16 byte key, 32 for larger
+	 *	%ymm0..%ymm15: 16 encrypted blocks
+	 * output:
+	 *	%ymm0..%ymm15: 16 plaintext blocks, order swapped:
+	 *       7, 8, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8
+	 */
+	FRAME_BEGIN
+
+	leaq 8 * 32(%rax), %rcx;
+
+	inpack32_post(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		      %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		      %ymm15, %rax, %rcx);
+
+	cmpl $32, %r8d;
+	je .Ldec_max32;
+
+.Ldec_max24:
+	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %rcx, 16);
+
+	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+	      %ymm15,
+	      ((key_table + (16) * 8) + 8)(CTX),
+	      ((key_table + (16) * 8) + 12)(CTX),
+	      ((key_table + (16) * 8) + 0)(CTX),
+	      ((key_table + (16) * 8) + 4)(CTX));
+
+	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %rcx, 8);
+
+	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+	      %ymm15,
+	      ((key_table + (8) * 8) + 8)(CTX),
+	      ((key_table + (8) * 8) + 12)(CTX),
+	      ((key_table + (8) * 8) + 0)(CTX),
+	      ((key_table + (8) * 8) + 4)(CTX));
+
+	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %rcx, 0);
+
+	/* load CD for output */
+	vmovdqu 0 * 32(%rcx), %ymm8;
+	vmovdqu 1 * 32(%rcx), %ymm9;
+	vmovdqu 2 * 32(%rcx), %ymm10;
+	vmovdqu 3 * 32(%rcx), %ymm11;
+	vmovdqu 4 * 32(%rcx), %ymm12;
+	vmovdqu 5 * 32(%rcx), %ymm13;
+	vmovdqu 6 * 32(%rcx), %ymm14;
+	vmovdqu 7 * 32(%rcx), %ymm15;
+
+	outunpack32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		    %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		    %ymm15, (key_table)(CTX), (%rax), 1 * 32(%rax));
+
+	FRAME_END
+	RET;
+
+.align 8
+.Ldec_max32:
+	dec_rounds32(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rax, %rcx, 24);
+
+	fls32(%rax, %ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+	      %rcx, %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+	      %ymm15,
+	      ((key_table + (24) * 8) + 8)(CTX),
+	      ((key_table + (24) * 8) + 12)(CTX),
+	      ((key_table + (24) * 8) + 0)(CTX),
+	      ((key_table + (24) * 8) + 4)(CTX));
+
+	jmp .Ldec_max24;
+SYM_FUNC_END(__camellia_dec_blk32)
+
+SYM_FUNC_START(camellia_ecb_enc_32way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst (32 blocks)
+	 *	%rdx: src (32 blocks)
+	 */
+	FRAME_BEGIN
+
+	vzeroupper;
+
+	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rdx, (key_table)(CTX));
+
+	/* now dst can be used as temporary buffer (even in src == dst case) */
+	movq	%rsi, %rax;
+
+	call __camellia_enc_blk32;
+
+	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
+		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
+		     %ymm8, %rsi);
+
+	vzeroupper;
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(camellia_ecb_enc_32way)
+
+SYM_FUNC_START(camellia_ecb_dec_32way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst (32 blocks)
+	 *	%rdx: src (32 blocks)
+	 */
+	FRAME_BEGIN
+
+	vzeroupper;
+
+	cmpl $16, key_length(CTX);
+	movl $32, %r8d;
+	movl $24, %eax;
+	cmovel %eax, %r8d; /* max */
+
+	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rdx, (key_table)(CTX, %r8, 8));
+
+	/* now dst can be used as temporary buffer (even in src == dst case) */
+	movq	%rsi, %rax;
+
+	call __camellia_dec_blk32;
+
+	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
+		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
+		     %ymm8, %rsi);
+
+	vzeroupper;
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(camellia_ecb_dec_32way)
+
+SYM_FUNC_START(camellia_cbc_dec_32way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst (32 blocks)
+	 *	%rdx: src (32 blocks)
+	 */
+	FRAME_BEGIN
+	subq $(16 * 32), %rsp;
+
+	vzeroupper;
+
+	cmpl $16, key_length(CTX);
+	movl $32, %r8d;
+	movl $24, %eax;
+	cmovel %eax, %r8d; /* max */
+
+	inpack32_pre(%ymm0, %ymm1, %ymm2, %ymm3, %ymm4, %ymm5, %ymm6, %ymm7,
+		     %ymm8, %ymm9, %ymm10, %ymm11, %ymm12, %ymm13, %ymm14,
+		     %ymm15, %rdx, (key_table)(CTX, %r8, 8));
+
+	cmpq %rsi, %rdx;
+	je .Lcbc_dec_use_stack;
+
+	/* dst can be used as temporary storage, src is not overwritten. */
+	movq %rsi, %rax;
+	jmp .Lcbc_dec_continue;
+
+.Lcbc_dec_use_stack:
+	/*
+	 * dst still in-use (because dst == src), so use stack for temporary
+	 * storage.
+	 */
+	movq %rsp, %rax;
+
+.Lcbc_dec_continue:
+	call __camellia_dec_blk32;
+
+	vmovdqu %ymm7, (%rax);
+	vpxor %ymm7, %ymm7, %ymm7;
+	vinserti128 $1, (%rdx), %ymm7, %ymm7;
+	vpxor (%rax), %ymm7, %ymm7;
+	vpxor (0 * 32 + 16)(%rdx), %ymm6, %ymm6;
+	vpxor (1 * 32 + 16)(%rdx), %ymm5, %ymm5;
+	vpxor (2 * 32 + 16)(%rdx), %ymm4, %ymm4;
+	vpxor (3 * 32 + 16)(%rdx), %ymm3, %ymm3;
+	vpxor (4 * 32 + 16)(%rdx), %ymm2, %ymm2;
+	vpxor (5 * 32 + 16)(%rdx), %ymm1, %ymm1;
+	vpxor (6 * 32 + 16)(%rdx), %ymm0, %ymm0;
+	vpxor (7 * 32 + 16)(%rdx), %ymm15, %ymm15;
+	vpxor (8 * 32 + 16)(%rdx), %ymm14, %ymm14;
+	vpxor (9 * 32 + 16)(%rdx), %ymm13, %ymm13;
+	vpxor (10 * 32 + 16)(%rdx), %ymm12, %ymm12;
+	vpxor (11 * 32 + 16)(%rdx), %ymm11, %ymm11;
+	vpxor (12 * 32 + 16)(%rdx), %ymm10, %ymm10;
+	vpxor (13 * 32 + 16)(%rdx), %ymm9, %ymm9;
+	vpxor (14 * 32 + 16)(%rdx), %ymm8, %ymm8;
+	write_output(%ymm7, %ymm6, %ymm5, %ymm4, %ymm3, %ymm2, %ymm1, %ymm0,
+		     %ymm15, %ymm14, %ymm13, %ymm12, %ymm11, %ymm10, %ymm9,
+		     %ymm8, %rsi);
+
+	vzeroupper;
+
+	addq $(16 * 32), %rsp;
+	FRAME_END
+	RET;
+SYM_FUNC_END(camellia_cbc_dec_32way)
diff --git a/arch/x86/crypto/camellia-x86_64-asm_64.S b/arch/x86/crypto/camellia-x86_64-asm_64.S
new file mode 100644
index 000000000..824cb94de
--- /dev/null
+++ b/arch/x86/crypto/camellia-x86_64-asm_64.S
@@ -0,0 +1,502 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Camellia Cipher Algorithm (x86_64)
+ *
+ * Copyright (C) 2012 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ */
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+
+.file "camellia-x86_64-asm_64.S"
+.text
+
+.extern camellia_sp10011110;
+.extern camellia_sp22000222;
+.extern camellia_sp03303033;
+.extern camellia_sp00444404;
+.extern camellia_sp02220222;
+.extern camellia_sp30333033;
+.extern camellia_sp44044404;
+.extern camellia_sp11101110;
+
+#define sp10011110 camellia_sp10011110
+#define sp22000222 camellia_sp22000222
+#define sp03303033 camellia_sp03303033
+#define sp00444404 camellia_sp00444404
+#define sp02220222 camellia_sp02220222
+#define sp30333033 camellia_sp30333033
+#define sp44044404 camellia_sp44044404
+#define sp11101110 camellia_sp11101110
+
+#define CAMELLIA_TABLE_BYTE_LEN 272
+
+/* struct camellia_ctx: */
+#define key_table 0
+#define key_length CAMELLIA_TABLE_BYTE_LEN
+
+/* register macros */
+#define CTX %rdi
+#define RIO %rsi
+#define RIOd %esi
+
+#define RAB0 %rax
+#define RCD0 %rcx
+#define RAB1 %rbx
+#define RCD1 %rdx
+
+#define RAB0d %eax
+#define RCD0d %ecx
+#define RAB1d %ebx
+#define RCD1d %edx
+
+#define RAB0bl %al
+#define RCD0bl %cl
+#define RAB1bl %bl
+#define RCD1bl %dl
+
+#define RAB0bh %ah
+#define RCD0bh %ch
+#define RAB1bh %bh
+#define RCD1bh %dh
+
+#define RT0 %rsi
+#define RT1 %r12
+#define RT2 %r8
+
+#define RT0d %esi
+#define RT1d %r12d
+#define RT2d %r8d
+
+#define RT2bl %r8b
+
+#define RXOR %r9
+#define RR12 %r10
+#define RDST %r11
+
+#define RXORd %r9d
+#define RXORbl %r9b
+
+#define xor2ror16(T0, T1, tmp1, tmp2, ab, dst) \
+	leaq T0(%rip), 			tmp1; \
+	movzbl ab ## bl,		tmp2 ## d; \
+	xorq (tmp1, tmp2, 8),		dst; \
+	leaq T1(%rip), 			tmp2; \
+	movzbl ab ## bh,		tmp1 ## d; \
+	rorq $16,			ab; \
+	xorq (tmp2, tmp1, 8),		dst;
+
+/**********************************************************************
+  1-way camellia
+ **********************************************************************/
+#define roundsm(ab, subkey, cd) \
+	movq (key_table + ((subkey) * 2) * 4)(CTX),	RT2; \
+	\
+	xor2ror16(sp00444404, sp03303033, RT0, RT1, ab ## 0, cd ## 0); \
+	xor2ror16(sp22000222, sp10011110, RT0, RT1, ab ## 0, RT2); \
+	xor2ror16(sp11101110, sp44044404, RT0, RT1, ab ## 0, cd ## 0); \
+	xor2ror16(sp30333033, sp02220222, RT0, RT1, ab ## 0, RT2); \
+	\
+	xorq RT2,					cd ## 0;
+
+#define fls(l, r, kl, kr) \
+	movl (key_table + ((kl) * 2) * 4)(CTX),		RT0d; \
+	andl l ## 0d,					RT0d; \
+	roll $1,					RT0d; \
+	shlq $32,					RT0; \
+	xorq RT0,					l ## 0; \
+	movq (key_table + ((kr) * 2) * 4)(CTX),		RT1; \
+	orq r ## 0,					RT1; \
+	shrq $32,					RT1; \
+	xorq RT1,					r ## 0; \
+	\
+	movq (key_table + ((kl) * 2) * 4)(CTX),		RT2; \
+	orq l ## 0,					RT2; \
+	shrq $32,					RT2; \
+	xorq RT2,					l ## 0; \
+	movl (key_table + ((kr) * 2) * 4)(CTX),		RT0d; \
+	andl r ## 0d,					RT0d; \
+	roll $1,					RT0d; \
+	shlq $32,					RT0; \
+	xorq RT0,					r ## 0;
+
+#define enc_rounds(i) \
+	roundsm(RAB, i + 2, RCD); \
+	roundsm(RCD, i + 3, RAB); \
+	roundsm(RAB, i + 4, RCD); \
+	roundsm(RCD, i + 5, RAB); \
+	roundsm(RAB, i + 6, RCD); \
+	roundsm(RCD, i + 7, RAB);
+
+#define enc_fls(i) \
+	fls(RAB, RCD, i + 0, i + 1);
+
+#define enc_inpack() \
+	movq (RIO),			RAB0; \
+	bswapq				RAB0; \
+	rolq $32,			RAB0; \
+	movq 4*2(RIO),			RCD0; \
+	bswapq				RCD0; \
+	rorq $32,			RCD0; \
+	xorq key_table(CTX),		RAB0;
+
+#define enc_outunpack(op, max) \
+	xorq key_table(CTX, max, 8),	RCD0; \
+	rorq $32,			RCD0; \
+	bswapq				RCD0; \
+	op ## q RCD0,			(RIO); \
+	rolq $32,			RAB0; \
+	bswapq				RAB0; \
+	op ## q RAB0,			4*2(RIO);
+
+#define dec_rounds(i) \
+	roundsm(RAB, i + 7, RCD); \
+	roundsm(RCD, i + 6, RAB); \
+	roundsm(RAB, i + 5, RCD); \
+	roundsm(RCD, i + 4, RAB); \
+	roundsm(RAB, i + 3, RCD); \
+	roundsm(RCD, i + 2, RAB);
+
+#define dec_fls(i) \
+	fls(RAB, RCD, i + 1, i + 0);
+
+#define dec_inpack(max) \
+	movq (RIO),			RAB0; \
+	bswapq				RAB0; \
+	rolq $32,			RAB0; \
+	movq 4*2(RIO),			RCD0; \
+	bswapq				RCD0; \
+	rorq $32,			RCD0; \
+	xorq key_table(CTX, max, 8),	RAB0;
+
+#define dec_outunpack() \
+	xorq key_table(CTX),		RCD0; \
+	rorq $32,			RCD0; \
+	bswapq				RCD0; \
+	movq RCD0,			(RIO); \
+	rolq $32,			RAB0; \
+	bswapq				RAB0; \
+	movq RAB0,			4*2(RIO);
+
+SYM_TYPED_FUNC_START(__camellia_enc_blk)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 *	%rcx: bool xor
+	 */
+	movq %r12, RR12;
+
+	movq %rcx, RXOR;
+	movq %rsi, RDST;
+	movq %rdx, RIO;
+
+	enc_inpack();
+
+	enc_rounds(0);
+	enc_fls(8);
+	enc_rounds(8);
+	enc_fls(16);
+	enc_rounds(16);
+	movl $24, RT1d; /* max */
+
+	cmpb $16, key_length(CTX);
+	je .L__enc_done;
+
+	enc_fls(24);
+	enc_rounds(24);
+	movl $32, RT1d; /* max */
+
+.L__enc_done:
+	testb RXORbl, RXORbl;
+	movq RDST, RIO;
+
+	jnz .L__enc_xor;
+
+	enc_outunpack(mov, RT1);
+
+	movq RR12, %r12;
+	RET;
+
+.L__enc_xor:
+	enc_outunpack(xor, RT1);
+
+	movq RR12, %r12;
+	RET;
+SYM_FUNC_END(__camellia_enc_blk)
+
+SYM_TYPED_FUNC_START(camellia_dec_blk)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	cmpl $16, key_length(CTX);
+	movl $32, RT2d;
+	movl $24, RXORd;
+	cmovel RXORd, RT2d; /* max */
+
+	movq %r12, RR12;
+	movq %rsi, RDST;
+	movq %rdx, RIO;
+
+	dec_inpack(RT2);
+
+	cmpb $24, RT2bl;
+	je .L__dec_rounds16;
+
+	dec_rounds(24);
+	dec_fls(24);
+
+.L__dec_rounds16:
+	dec_rounds(16);
+	dec_fls(16);
+	dec_rounds(8);
+	dec_fls(8);
+	dec_rounds(0);
+
+	movq RDST, RIO;
+
+	dec_outunpack();
+
+	movq RR12, %r12;
+	RET;
+SYM_FUNC_END(camellia_dec_blk)
+
+/**********************************************************************
+  2-way camellia
+ **********************************************************************/
+#define roundsm2(ab, subkey, cd) \
+	movq (key_table + ((subkey) * 2) * 4)(CTX),	RT2; \
+	xorq RT2,					cd ## 1; \
+	\
+	xor2ror16(sp00444404, sp03303033, RT0, RT1, ab ## 0, cd ## 0); \
+	xor2ror16(sp22000222, sp10011110, RT0, RT1, ab ## 0, RT2); \
+	xor2ror16(sp11101110, sp44044404, RT0, RT1, ab ## 0, cd ## 0); \
+	xor2ror16(sp30333033, sp02220222, RT0, RT1, ab ## 0, RT2); \
+	\
+		xor2ror16(sp00444404, sp03303033, RT0, RT1, ab ## 1, cd ## 1); \
+		xorq RT2,					cd ## 0; \
+		xor2ror16(sp22000222, sp10011110, RT0, RT1, ab ## 1, cd ## 1); \
+		xor2ror16(sp11101110, sp44044404, RT0, RT1, ab ## 1, cd ## 1); \
+		xor2ror16(sp30333033, sp02220222, RT0, RT1, ab ## 1, cd ## 1);
+
+#define fls2(l, r, kl, kr) \
+	movl (key_table + ((kl) * 2) * 4)(CTX),		RT0d; \
+	andl l ## 0d,					RT0d; \
+	roll $1,					RT0d; \
+	shlq $32,					RT0; \
+	xorq RT0,					l ## 0; \
+	movq (key_table + ((kr) * 2) * 4)(CTX),		RT1; \
+	orq r ## 0,					RT1; \
+	shrq $32,					RT1; \
+	xorq RT1,					r ## 0; \
+	\
+		movl (key_table + ((kl) * 2) * 4)(CTX),		RT2d; \
+		andl l ## 1d,					RT2d; \
+		roll $1,					RT2d; \
+		shlq $32,					RT2; \
+		xorq RT2,					l ## 1; \
+		movq (key_table + ((kr) * 2) * 4)(CTX),		RT0; \
+		orq r ## 1,					RT0; \
+		shrq $32,					RT0; \
+		xorq RT0,					r ## 1; \
+	\
+	movq (key_table + ((kl) * 2) * 4)(CTX),		RT1; \
+	orq l ## 0,					RT1; \
+	shrq $32,					RT1; \
+	xorq RT1,					l ## 0; \
+	movl (key_table + ((kr) * 2) * 4)(CTX),		RT2d; \
+	andl r ## 0d,					RT2d; \
+	roll $1,					RT2d; \
+	shlq $32,					RT2; \
+	xorq RT2,					r ## 0; \
+	\
+		movq (key_table + ((kl) * 2) * 4)(CTX),		RT0; \
+		orq l ## 1,					RT0; \
+		shrq $32,					RT0; \
+		xorq RT0,					l ## 1; \
+		movl (key_table + ((kr) * 2) * 4)(CTX),		RT1d; \
+		andl r ## 1d,					RT1d; \
+		roll $1,					RT1d; \
+		shlq $32,					RT1; \
+		xorq RT1,					r ## 1;
+
+#define enc_rounds2(i) \
+	roundsm2(RAB, i + 2, RCD); \
+	roundsm2(RCD, i + 3, RAB); \
+	roundsm2(RAB, i + 4, RCD); \
+	roundsm2(RCD, i + 5, RAB); \
+	roundsm2(RAB, i + 6, RCD); \
+	roundsm2(RCD, i + 7, RAB);
+
+#define enc_fls2(i) \
+	fls2(RAB, RCD, i + 0, i + 1);
+
+#define enc_inpack2() \
+	movq (RIO),			RAB0; \
+	bswapq				RAB0; \
+	rorq $32,			RAB0; \
+	movq 4*2(RIO),			RCD0; \
+	bswapq				RCD0; \
+	rolq $32,			RCD0; \
+	xorq key_table(CTX),		RAB0; \
+	\
+		movq 8*2(RIO),			RAB1; \
+		bswapq				RAB1; \
+		rorq $32,			RAB1; \
+		movq 12*2(RIO),			RCD1; \
+		bswapq				RCD1; \
+		rolq $32,			RCD1; \
+		xorq key_table(CTX),		RAB1;
+
+#define enc_outunpack2(op, max) \
+	xorq key_table(CTX, max, 8),	RCD0; \
+	rolq $32,			RCD0; \
+	bswapq				RCD0; \
+	op ## q RCD0,			(RIO); \
+	rorq $32,			RAB0; \
+	bswapq				RAB0; \
+	op ## q RAB0,			4*2(RIO); \
+	\
+		xorq key_table(CTX, max, 8),	RCD1; \
+		rolq $32,			RCD1; \
+		bswapq				RCD1; \
+		op ## q RCD1,			8*2(RIO); \
+		rorq $32,			RAB1; \
+		bswapq				RAB1; \
+		op ## q RAB1,			12*2(RIO);
+
+#define dec_rounds2(i) \
+	roundsm2(RAB, i + 7, RCD); \
+	roundsm2(RCD, i + 6, RAB); \
+	roundsm2(RAB, i + 5, RCD); \
+	roundsm2(RCD, i + 4, RAB); \
+	roundsm2(RAB, i + 3, RCD); \
+	roundsm2(RCD, i + 2, RAB);
+
+#define dec_fls2(i) \
+	fls2(RAB, RCD, i + 1, i + 0);
+
+#define dec_inpack2(max) \
+	movq (RIO),			RAB0; \
+	bswapq				RAB0; \
+	rorq $32,			RAB0; \
+	movq 4*2(RIO),			RCD0; \
+	bswapq				RCD0; \
+	rolq $32,			RCD0; \
+	xorq key_table(CTX, max, 8),	RAB0; \
+	\
+		movq 8*2(RIO),			RAB1; \
+		bswapq				RAB1; \
+		rorq $32,			RAB1; \
+		movq 12*2(RIO),			RCD1; \
+		bswapq				RCD1; \
+		rolq $32,			RCD1; \
+		xorq key_table(CTX, max, 8),	RAB1;
+
+#define dec_outunpack2() \
+	xorq key_table(CTX),		RCD0; \
+	rolq $32,			RCD0; \
+	bswapq				RCD0; \
+	movq RCD0,			(RIO); \
+	rorq $32,			RAB0; \
+	bswapq				RAB0; \
+	movq RAB0,			4*2(RIO); \
+	\
+		xorq key_table(CTX),		RCD1; \
+		rolq $32,			RCD1; \
+		bswapq				RCD1; \
+		movq RCD1,			8*2(RIO); \
+		rorq $32,			RAB1; \
+		bswapq				RAB1; \
+		movq RAB1,			12*2(RIO);
+
+SYM_TYPED_FUNC_START(__camellia_enc_blk_2way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 *	%rcx: bool xor
+	 */
+	pushq %rbx;
+
+	movq %r12, RR12;
+	movq %rcx, RXOR;
+	movq %rsi, RDST;
+	movq %rdx, RIO;
+
+	enc_inpack2();
+
+	enc_rounds2(0);
+	enc_fls2(8);
+	enc_rounds2(8);
+	enc_fls2(16);
+	enc_rounds2(16);
+	movl $24, RT2d; /* max */
+
+	cmpb $16, key_length(CTX);
+	je .L__enc2_done;
+
+	enc_fls2(24);
+	enc_rounds2(24);
+	movl $32, RT2d; /* max */
+
+.L__enc2_done:
+	test RXORbl, RXORbl;
+	movq RDST, RIO;
+	jnz .L__enc2_xor;
+
+	enc_outunpack2(mov, RT2);
+
+	movq RR12, %r12;
+	popq %rbx;
+	RET;
+
+.L__enc2_xor:
+	enc_outunpack2(xor, RT2);
+
+	movq RR12, %r12;
+	popq %rbx;
+	RET;
+SYM_FUNC_END(__camellia_enc_blk_2way)
+
+SYM_TYPED_FUNC_START(camellia_dec_blk_2way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	cmpl $16, key_length(CTX);
+	movl $32, RT2d;
+	movl $24, RXORd;
+	cmovel RXORd, RT2d; /* max */
+
+	movq %rbx, RXOR;
+	movq %r12, RR12;
+	movq %rsi, RDST;
+	movq %rdx, RIO;
+
+	dec_inpack2(RT2);
+
+	cmpb $24, RT2bl;
+	je .L__dec2_rounds16;
+
+	dec_rounds2(24);
+	dec_fls2(24);
+
+.L__dec2_rounds16:
+	dec_rounds2(16);
+	dec_fls2(16);
+	dec_rounds2(8);
+	dec_fls2(8);
+	dec_rounds2(0);
+
+	movq RDST, RIO;
+
+	dec_outunpack2();
+
+	movq RR12, %r12;
+	movq RXOR, %rbx;
+	RET;
+SYM_FUNC_END(camellia_dec_blk_2way)
diff --git a/arch/x86/crypto/camellia.h b/arch/x86/crypto/camellia.h
new file mode 100644
index 000000000..1dcea79e8
--- /dev/null
+++ b/arch/x86/crypto/camellia.h
@@ -0,0 +1,67 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef ASM_X86_CAMELLIA_H
+#define ASM_X86_CAMELLIA_H
+
+#include <crypto/b128ops.h>
+#include <linux/crypto.h>
+#include <linux/kernel.h>
+
+#define CAMELLIA_MIN_KEY_SIZE	16
+#define CAMELLIA_MAX_KEY_SIZE	32
+#define CAMELLIA_BLOCK_SIZE	16
+#define CAMELLIA_TABLE_BYTE_LEN	272
+#define CAMELLIA_PARALLEL_BLOCKS 2
+
+struct crypto_skcipher;
+
+struct camellia_ctx {
+	u64 key_table[CAMELLIA_TABLE_BYTE_LEN / sizeof(u64)];
+	u32 key_length;
+};
+
+extern int __camellia_setkey(struct camellia_ctx *cctx,
+			     const unsigned char *key,
+			     unsigned int key_len);
+
+/* regular block cipher functions */
+asmlinkage void __camellia_enc_blk(const void *ctx, u8 *dst, const u8 *src,
+				   bool xor);
+asmlinkage void camellia_dec_blk(const void *ctx, u8 *dst, const u8 *src);
+
+/* 2-way parallel cipher functions */
+asmlinkage void __camellia_enc_blk_2way(const void *ctx, u8 *dst, const u8 *src,
+					bool xor);
+asmlinkage void camellia_dec_blk_2way(const void *ctx, u8 *dst, const u8 *src);
+
+/* 16-way parallel cipher functions (avx/aes-ni) */
+asmlinkage void camellia_ecb_enc_16way(const void *ctx, u8 *dst, const u8 *src);
+asmlinkage void camellia_ecb_dec_16way(const void *ctx, u8 *dst, const u8 *src);
+
+asmlinkage void camellia_cbc_dec_16way(const void *ctx, u8 *dst, const u8 *src);
+
+static inline void camellia_enc_blk(const void *ctx, u8 *dst, const u8 *src)
+{
+	__camellia_enc_blk(ctx, dst, src, false);
+}
+
+static inline void camellia_enc_blk_xor(const void *ctx, u8 *dst, const u8 *src)
+{
+	__camellia_enc_blk(ctx, dst, src, true);
+}
+
+static inline void camellia_enc_blk_2way(const void *ctx, u8 *dst,
+					 const u8 *src)
+{
+	__camellia_enc_blk_2way(ctx, dst, src, false);
+}
+
+static inline void camellia_enc_blk_xor_2way(const void *ctx, u8 *dst,
+					     const u8 *src)
+{
+	__camellia_enc_blk_2way(ctx, dst, src, true);
+}
+
+/* glue helpers */
+extern void camellia_decrypt_cbc_2way(const void *ctx, u8 *dst, const u8 *src);
+
+#endif /* ASM_X86_CAMELLIA_H */
diff --git a/arch/x86/crypto/camellia_aesni_avx2_glue.c b/arch/x86/crypto/camellia_aesni_avx2_glue.c
new file mode 100644
index 000000000..2d2f4e165
--- /dev/null
+++ b/arch/x86/crypto/camellia_aesni_avx2_glue.c
@@ -0,0 +1,131 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Glue Code for x86_64/AVX2/AES-NI assembler optimized version of Camellia
+ *
+ * Copyright © 2013 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ */
+
+#include <crypto/algapi.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <linux/module.h>
+#include <linux/types.h>
+
+#include "camellia.h"
+#include "ecb_cbc_helpers.h"
+
+#define CAMELLIA_AESNI_PARALLEL_BLOCKS 16
+#define CAMELLIA_AESNI_AVX2_PARALLEL_BLOCKS 32
+
+/* 32-way AVX2/AES-NI parallel cipher functions */
+asmlinkage void camellia_ecb_enc_32way(const void *ctx, u8 *dst, const u8 *src);
+asmlinkage void camellia_ecb_dec_32way(const void *ctx, u8 *dst, const u8 *src);
+
+asmlinkage void camellia_cbc_dec_32way(const void *ctx, u8 *dst, const u8 *src);
+
+static int camellia_setkey(struct crypto_skcipher *tfm, const u8 *key,
+			   unsigned int keylen)
+{
+	return __camellia_setkey(crypto_skcipher_ctx(tfm), key, keylen);
+}
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, CAMELLIA_BLOCK_SIZE, CAMELLIA_AESNI_PARALLEL_BLOCKS);
+	ECB_BLOCK(CAMELLIA_AESNI_AVX2_PARALLEL_BLOCKS, camellia_ecb_enc_32way);
+	ECB_BLOCK(CAMELLIA_AESNI_PARALLEL_BLOCKS, camellia_ecb_enc_16way);
+	ECB_BLOCK(2, camellia_enc_blk_2way);
+	ECB_BLOCK(1, camellia_enc_blk);
+	ECB_WALK_END();
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, CAMELLIA_BLOCK_SIZE, CAMELLIA_AESNI_PARALLEL_BLOCKS);
+	ECB_BLOCK(CAMELLIA_AESNI_AVX2_PARALLEL_BLOCKS, camellia_ecb_dec_32way);
+	ECB_BLOCK(CAMELLIA_AESNI_PARALLEL_BLOCKS, camellia_ecb_dec_16way);
+	ECB_BLOCK(2, camellia_dec_blk_2way);
+	ECB_BLOCK(1, camellia_dec_blk);
+	ECB_WALK_END();
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, CAMELLIA_BLOCK_SIZE, -1);
+	CBC_ENC_BLOCK(camellia_enc_blk);
+	CBC_WALK_END();
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, CAMELLIA_BLOCK_SIZE, CAMELLIA_AESNI_PARALLEL_BLOCKS);
+	CBC_DEC_BLOCK(CAMELLIA_AESNI_AVX2_PARALLEL_BLOCKS, camellia_cbc_dec_32way);
+	CBC_DEC_BLOCK(CAMELLIA_AESNI_PARALLEL_BLOCKS, camellia_cbc_dec_16way);
+	CBC_DEC_BLOCK(2, camellia_decrypt_cbc_2way);
+	CBC_DEC_BLOCK(1, camellia_dec_blk);
+	CBC_WALK_END();
+}
+
+static struct skcipher_alg camellia_algs[] = {
+	{
+		.base.cra_name		= "ecb(camellia)",
+		.base.cra_driver_name	= "ecb-camellia-aesni-avx2",
+		.base.cra_priority	= 500,
+		.base.cra_blocksize	= CAMELLIA_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct camellia_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= CAMELLIA_MIN_KEY_SIZE,
+		.max_keysize		= CAMELLIA_MAX_KEY_SIZE,
+		.setkey			= camellia_setkey,
+		.encrypt		= ecb_encrypt,
+		.decrypt		= ecb_decrypt,
+	}, {
+		.base.cra_name		= "cbc(camellia)",
+		.base.cra_driver_name	= "cbc-camellia-aesni-avx2",
+		.base.cra_priority	= 500,
+		.base.cra_blocksize	= CAMELLIA_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct camellia_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= CAMELLIA_MIN_KEY_SIZE,
+		.max_keysize		= CAMELLIA_MAX_KEY_SIZE,
+		.ivsize			= CAMELLIA_BLOCK_SIZE,
+		.setkey			= camellia_setkey,
+		.encrypt		= cbc_encrypt,
+		.decrypt		= cbc_decrypt,
+	},
+};
+
+static int __init camellia_aesni_init(void)
+{
+	const char *feature_name;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX) ||
+	    !boot_cpu_has(X86_FEATURE_AVX2) ||
+	    !boot_cpu_has(X86_FEATURE_AES) ||
+	    !boot_cpu_has(X86_FEATURE_OSXSAVE)) {
+		pr_info("AVX2 or AES-NI instructions are not detected.\n");
+		return -ENODEV;
+	}
+
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM,
+				&feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	return crypto_register_skciphers(camellia_algs,
+					 ARRAY_SIZE(camellia_algs));
+}
+
+static void __exit camellia_aesni_fini(void)
+{
+	crypto_unregister_skciphers(camellia_algs, ARRAY_SIZE(camellia_algs));
+}
+
+module_init(camellia_aesni_init);
+module_exit(camellia_aesni_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Camellia Cipher Algorithm, AES-NI/AVX2 optimized");
+MODULE_ALIAS_CRYPTO("camellia");
+MODULE_ALIAS_CRYPTO("camellia-asm");
diff --git a/arch/x86/crypto/camellia_aesni_avx_glue.c b/arch/x86/crypto/camellia_aesni_avx_glue.c
new file mode 100644
index 000000000..5c321f255
--- /dev/null
+++ b/arch/x86/crypto/camellia_aesni_avx_glue.c
@@ -0,0 +1,131 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Glue Code for x86_64/AVX/AES-NI assembler optimized version of Camellia
+ *
+ * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ */
+
+#include <crypto/algapi.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <linux/export.h>
+#include <linux/module.h>
+#include <linux/types.h>
+
+#include "camellia.h"
+#include "ecb_cbc_helpers.h"
+
+#define CAMELLIA_AESNI_PARALLEL_BLOCKS 16
+
+/* 16-way parallel cipher functions (avx/aes-ni) */
+asmlinkage void camellia_ecb_enc_16way(const void *ctx, u8 *dst, const u8 *src);
+EXPORT_SYMBOL_GPL(camellia_ecb_enc_16way);
+
+asmlinkage void camellia_ecb_dec_16way(const void *ctx, u8 *dst, const u8 *src);
+EXPORT_SYMBOL_GPL(camellia_ecb_dec_16way);
+
+asmlinkage void camellia_cbc_dec_16way(const void *ctx, u8 *dst, const u8 *src);
+EXPORT_SYMBOL_GPL(camellia_cbc_dec_16way);
+
+static int camellia_setkey(struct crypto_skcipher *tfm, const u8 *key,
+			   unsigned int keylen)
+{
+	return __camellia_setkey(crypto_skcipher_ctx(tfm), key, keylen);
+}
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, CAMELLIA_BLOCK_SIZE, CAMELLIA_AESNI_PARALLEL_BLOCKS);
+	ECB_BLOCK(CAMELLIA_AESNI_PARALLEL_BLOCKS, camellia_ecb_enc_16way);
+	ECB_BLOCK(2, camellia_enc_blk_2way);
+	ECB_BLOCK(1, camellia_enc_blk);
+	ECB_WALK_END();
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, CAMELLIA_BLOCK_SIZE, CAMELLIA_AESNI_PARALLEL_BLOCKS);
+	ECB_BLOCK(CAMELLIA_AESNI_PARALLEL_BLOCKS, camellia_ecb_dec_16way);
+	ECB_BLOCK(2, camellia_dec_blk_2way);
+	ECB_BLOCK(1, camellia_dec_blk);
+	ECB_WALK_END();
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, CAMELLIA_BLOCK_SIZE, -1);
+	CBC_ENC_BLOCK(camellia_enc_blk);
+	CBC_WALK_END();
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, CAMELLIA_BLOCK_SIZE, CAMELLIA_AESNI_PARALLEL_BLOCKS);
+	CBC_DEC_BLOCK(CAMELLIA_AESNI_PARALLEL_BLOCKS, camellia_cbc_dec_16way);
+	CBC_DEC_BLOCK(2, camellia_decrypt_cbc_2way);
+	CBC_DEC_BLOCK(1, camellia_dec_blk);
+	CBC_WALK_END();
+}
+
+static struct skcipher_alg camellia_algs[] = {
+	{
+		.base.cra_name		= "ecb(camellia)",
+		.base.cra_driver_name	= "ecb-camellia-aesni",
+		.base.cra_priority	= 400,
+		.base.cra_blocksize	= CAMELLIA_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct camellia_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= CAMELLIA_MIN_KEY_SIZE,
+		.max_keysize		= CAMELLIA_MAX_KEY_SIZE,
+		.setkey			= camellia_setkey,
+		.encrypt		= ecb_encrypt,
+		.decrypt		= ecb_decrypt,
+	}, {
+		.base.cra_name		= "cbc(camellia)",
+		.base.cra_driver_name	= "cbc-camellia-aesni",
+		.base.cra_priority	= 400,
+		.base.cra_blocksize	= CAMELLIA_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct camellia_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= CAMELLIA_MIN_KEY_SIZE,
+		.max_keysize		= CAMELLIA_MAX_KEY_SIZE,
+		.ivsize			= CAMELLIA_BLOCK_SIZE,
+		.setkey			= camellia_setkey,
+		.encrypt		= cbc_encrypt,
+		.decrypt		= cbc_decrypt,
+	}
+};
+
+static int __init camellia_aesni_init(void)
+{
+	const char *feature_name;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX) ||
+	    !boot_cpu_has(X86_FEATURE_AES) ||
+	    !boot_cpu_has(X86_FEATURE_OSXSAVE)) {
+		pr_info("AVX or AES-NI instructions are not detected.\n");
+		return -ENODEV;
+	}
+
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM,
+				&feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	return crypto_register_skciphers(camellia_algs,
+					 ARRAY_SIZE(camellia_algs));
+}
+
+static void __exit camellia_aesni_fini(void)
+{
+	crypto_unregister_skciphers(camellia_algs, ARRAY_SIZE(camellia_algs));
+}
+
+module_init(camellia_aesni_init);
+module_exit(camellia_aesni_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Camellia Cipher Algorithm, AES-NI/AVX optimized");
+MODULE_ALIAS_CRYPTO("camellia");
+MODULE_ALIAS_CRYPTO("camellia-asm");
diff --git a/arch/x86/crypto/camellia_glue.c b/arch/x86/crypto/camellia_glue.c
new file mode 100644
index 000000000..cbede120e
--- /dev/null
+++ b/arch/x86/crypto/camellia_glue.c
@@ -0,0 +1,1417 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Glue Code for assembler optimized version of Camellia
+ *
+ * Copyright (c) 2012 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ *
+ * Camellia parts based on code by:
+ *  Copyright (C) 2006 NTT (Nippon Telegraph and Telephone Corporation)
+ */
+
+#include <linux/unaligned.h>
+#include <linux/crypto.h>
+#include <linux/export.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/types.h>
+#include <crypto/algapi.h>
+
+#include "camellia.h"
+#include "ecb_cbc_helpers.h"
+
+/* regular block cipher functions */
+asmlinkage void __camellia_enc_blk(const void *ctx, u8 *dst, const u8 *src,
+				   bool xor);
+EXPORT_SYMBOL_GPL(__camellia_enc_blk);
+asmlinkage void camellia_dec_blk(const void *ctx, u8 *dst, const u8 *src);
+EXPORT_SYMBOL_GPL(camellia_dec_blk);
+
+/* 2-way parallel cipher functions */
+asmlinkage void __camellia_enc_blk_2way(const void *ctx, u8 *dst, const u8 *src,
+					bool xor);
+EXPORT_SYMBOL_GPL(__camellia_enc_blk_2way);
+asmlinkage void camellia_dec_blk_2way(const void *ctx, u8 *dst, const u8 *src);
+EXPORT_SYMBOL_GPL(camellia_dec_blk_2way);
+
+static void camellia_encrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
+{
+	camellia_enc_blk(crypto_tfm_ctx(tfm), dst, src);
+}
+
+static void camellia_decrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
+{
+	camellia_dec_blk(crypto_tfm_ctx(tfm), dst, src);
+}
+
+/* camellia sboxes */
+__visible const u64 camellia_sp10011110[256] = {
+	0x7000007070707000ULL, 0x8200008282828200ULL, 0x2c00002c2c2c2c00ULL,
+	0xec0000ecececec00ULL, 0xb30000b3b3b3b300ULL, 0x2700002727272700ULL,
+	0xc00000c0c0c0c000ULL, 0xe50000e5e5e5e500ULL, 0xe40000e4e4e4e400ULL,
+	0x8500008585858500ULL, 0x5700005757575700ULL, 0x3500003535353500ULL,
+	0xea0000eaeaeaea00ULL, 0x0c00000c0c0c0c00ULL, 0xae0000aeaeaeae00ULL,
+	0x4100004141414100ULL, 0x2300002323232300ULL, 0xef0000efefefef00ULL,
+	0x6b00006b6b6b6b00ULL, 0x9300009393939300ULL, 0x4500004545454500ULL,
+	0x1900001919191900ULL, 0xa50000a5a5a5a500ULL, 0x2100002121212100ULL,
+	0xed0000edededed00ULL, 0x0e00000e0e0e0e00ULL, 0x4f00004f4f4f4f00ULL,
+	0x4e00004e4e4e4e00ULL, 0x1d00001d1d1d1d00ULL, 0x6500006565656500ULL,
+	0x9200009292929200ULL, 0xbd0000bdbdbdbd00ULL, 0x8600008686868600ULL,
+	0xb80000b8b8b8b800ULL, 0xaf0000afafafaf00ULL, 0x8f00008f8f8f8f00ULL,
+	0x7c00007c7c7c7c00ULL, 0xeb0000ebebebeb00ULL, 0x1f00001f1f1f1f00ULL,
+	0xce0000cececece00ULL, 0x3e00003e3e3e3e00ULL, 0x3000003030303000ULL,
+	0xdc0000dcdcdcdc00ULL, 0x5f00005f5f5f5f00ULL, 0x5e00005e5e5e5e00ULL,
+	0xc50000c5c5c5c500ULL, 0x0b00000b0b0b0b00ULL, 0x1a00001a1a1a1a00ULL,
+	0xa60000a6a6a6a600ULL, 0xe10000e1e1e1e100ULL, 0x3900003939393900ULL,
+	0xca0000cacacaca00ULL, 0xd50000d5d5d5d500ULL, 0x4700004747474700ULL,
+	0x5d00005d5d5d5d00ULL, 0x3d00003d3d3d3d00ULL, 0xd90000d9d9d9d900ULL,
+	0x0100000101010100ULL, 0x5a00005a5a5a5a00ULL, 0xd60000d6d6d6d600ULL,
+	0x5100005151515100ULL, 0x5600005656565600ULL, 0x6c00006c6c6c6c00ULL,
+	0x4d00004d4d4d4d00ULL, 0x8b00008b8b8b8b00ULL, 0x0d00000d0d0d0d00ULL,
+	0x9a00009a9a9a9a00ULL, 0x6600006666666600ULL, 0xfb0000fbfbfbfb00ULL,
+	0xcc0000cccccccc00ULL, 0xb00000b0b0b0b000ULL, 0x2d00002d2d2d2d00ULL,
+	0x7400007474747400ULL, 0x1200001212121200ULL, 0x2b00002b2b2b2b00ULL,
+	0x2000002020202000ULL, 0xf00000f0f0f0f000ULL, 0xb10000b1b1b1b100ULL,
+	0x8400008484848400ULL, 0x9900009999999900ULL, 0xdf0000dfdfdfdf00ULL,
+	0x4c00004c4c4c4c00ULL, 0xcb0000cbcbcbcb00ULL, 0xc20000c2c2c2c200ULL,
+	0x3400003434343400ULL, 0x7e00007e7e7e7e00ULL, 0x7600007676767600ULL,
+	0x0500000505050500ULL, 0x6d00006d6d6d6d00ULL, 0xb70000b7b7b7b700ULL,
+	0xa90000a9a9a9a900ULL, 0x3100003131313100ULL, 0xd10000d1d1d1d100ULL,
+	0x1700001717171700ULL, 0x0400000404040400ULL, 0xd70000d7d7d7d700ULL,
+	0x1400001414141400ULL, 0x5800005858585800ULL, 0x3a00003a3a3a3a00ULL,
+	0x6100006161616100ULL, 0xde0000dededede00ULL, 0x1b00001b1b1b1b00ULL,
+	0x1100001111111100ULL, 0x1c00001c1c1c1c00ULL, 0x3200003232323200ULL,
+	0x0f00000f0f0f0f00ULL, 0x9c00009c9c9c9c00ULL, 0x1600001616161600ULL,
+	0x5300005353535300ULL, 0x1800001818181800ULL, 0xf20000f2f2f2f200ULL,
+	0x2200002222222200ULL, 0xfe0000fefefefe00ULL, 0x4400004444444400ULL,
+	0xcf0000cfcfcfcf00ULL, 0xb20000b2b2b2b200ULL, 0xc30000c3c3c3c300ULL,
+	0xb50000b5b5b5b500ULL, 0x7a00007a7a7a7a00ULL, 0x9100009191919100ULL,
+	0x2400002424242400ULL, 0x0800000808080800ULL, 0xe80000e8e8e8e800ULL,
+	0xa80000a8a8a8a800ULL, 0x6000006060606000ULL, 0xfc0000fcfcfcfc00ULL,
+	0x6900006969696900ULL, 0x5000005050505000ULL, 0xaa0000aaaaaaaa00ULL,
+	0xd00000d0d0d0d000ULL, 0xa00000a0a0a0a000ULL, 0x7d00007d7d7d7d00ULL,
+	0xa10000a1a1a1a100ULL, 0x8900008989898900ULL, 0x6200006262626200ULL,
+	0x9700009797979700ULL, 0x5400005454545400ULL, 0x5b00005b5b5b5b00ULL,
+	0x1e00001e1e1e1e00ULL, 0x9500009595959500ULL, 0xe00000e0e0e0e000ULL,
+	0xff0000ffffffff00ULL, 0x6400006464646400ULL, 0xd20000d2d2d2d200ULL,
+	0x1000001010101000ULL, 0xc40000c4c4c4c400ULL, 0x0000000000000000ULL,
+	0x4800004848484800ULL, 0xa30000a3a3a3a300ULL, 0xf70000f7f7f7f700ULL,
+	0x7500007575757500ULL, 0xdb0000dbdbdbdb00ULL, 0x8a00008a8a8a8a00ULL,
+	0x0300000303030300ULL, 0xe60000e6e6e6e600ULL, 0xda0000dadadada00ULL,
+	0x0900000909090900ULL, 0x3f00003f3f3f3f00ULL, 0xdd0000dddddddd00ULL,
+	0x9400009494949400ULL, 0x8700008787878700ULL, 0x5c00005c5c5c5c00ULL,
+	0x8300008383838300ULL, 0x0200000202020200ULL, 0xcd0000cdcdcdcd00ULL,
+	0x4a00004a4a4a4a00ULL, 0x9000009090909000ULL, 0x3300003333333300ULL,
+	0x7300007373737300ULL, 0x6700006767676700ULL, 0xf60000f6f6f6f600ULL,
+	0xf30000f3f3f3f300ULL, 0x9d00009d9d9d9d00ULL, 0x7f00007f7f7f7f00ULL,
+	0xbf0000bfbfbfbf00ULL, 0xe20000e2e2e2e200ULL, 0x5200005252525200ULL,
+	0x9b00009b9b9b9b00ULL, 0xd80000d8d8d8d800ULL, 0x2600002626262600ULL,
+	0xc80000c8c8c8c800ULL, 0x3700003737373700ULL, 0xc60000c6c6c6c600ULL,
+	0x3b00003b3b3b3b00ULL, 0x8100008181818100ULL, 0x9600009696969600ULL,
+	0x6f00006f6f6f6f00ULL, 0x4b00004b4b4b4b00ULL, 0x1300001313131300ULL,
+	0xbe0000bebebebe00ULL, 0x6300006363636300ULL, 0x2e00002e2e2e2e00ULL,
+	0xe90000e9e9e9e900ULL, 0x7900007979797900ULL, 0xa70000a7a7a7a700ULL,
+	0x8c00008c8c8c8c00ULL, 0x9f00009f9f9f9f00ULL, 0x6e00006e6e6e6e00ULL,
+	0xbc0000bcbcbcbc00ULL, 0x8e00008e8e8e8e00ULL, 0x2900002929292900ULL,
+	0xf50000f5f5f5f500ULL, 0xf90000f9f9f9f900ULL, 0xb60000b6b6b6b600ULL,
+	0x2f00002f2f2f2f00ULL, 0xfd0000fdfdfdfd00ULL, 0xb40000b4b4b4b400ULL,
+	0x5900005959595900ULL, 0x7800007878787800ULL, 0x9800009898989800ULL,
+	0x0600000606060600ULL, 0x6a00006a6a6a6a00ULL, 0xe70000e7e7e7e700ULL,
+	0x4600004646464600ULL, 0x7100007171717100ULL, 0xba0000babababa00ULL,
+	0xd40000d4d4d4d400ULL, 0x2500002525252500ULL, 0xab0000abababab00ULL,
+	0x4200004242424200ULL, 0x8800008888888800ULL, 0xa20000a2a2a2a200ULL,
+	0x8d00008d8d8d8d00ULL, 0xfa0000fafafafa00ULL, 0x7200007272727200ULL,
+	0x0700000707070700ULL, 0xb90000b9b9b9b900ULL, 0x5500005555555500ULL,
+	0xf80000f8f8f8f800ULL, 0xee0000eeeeeeee00ULL, 0xac0000acacacac00ULL,
+	0x0a00000a0a0a0a00ULL, 0x3600003636363600ULL, 0x4900004949494900ULL,
+	0x2a00002a2a2a2a00ULL, 0x6800006868686800ULL, 0x3c00003c3c3c3c00ULL,
+	0x3800003838383800ULL, 0xf10000f1f1f1f100ULL, 0xa40000a4a4a4a400ULL,
+	0x4000004040404000ULL, 0x2800002828282800ULL, 0xd30000d3d3d3d300ULL,
+	0x7b00007b7b7b7b00ULL, 0xbb0000bbbbbbbb00ULL, 0xc90000c9c9c9c900ULL,
+	0x4300004343434300ULL, 0xc10000c1c1c1c100ULL, 0x1500001515151500ULL,
+	0xe30000e3e3e3e300ULL, 0xad0000adadadad00ULL, 0xf40000f4f4f4f400ULL,
+	0x7700007777777700ULL, 0xc70000c7c7c7c700ULL, 0x8000008080808000ULL,
+	0x9e00009e9e9e9e00ULL,
+};
+
+__visible const u64 camellia_sp22000222[256] = {
+	0xe0e0000000e0e0e0ULL, 0x0505000000050505ULL, 0x5858000000585858ULL,
+	0xd9d9000000d9d9d9ULL, 0x6767000000676767ULL, 0x4e4e0000004e4e4eULL,
+	0x8181000000818181ULL, 0xcbcb000000cbcbcbULL, 0xc9c9000000c9c9c9ULL,
+	0x0b0b0000000b0b0bULL, 0xaeae000000aeaeaeULL, 0x6a6a0000006a6a6aULL,
+	0xd5d5000000d5d5d5ULL, 0x1818000000181818ULL, 0x5d5d0000005d5d5dULL,
+	0x8282000000828282ULL, 0x4646000000464646ULL, 0xdfdf000000dfdfdfULL,
+	0xd6d6000000d6d6d6ULL, 0x2727000000272727ULL, 0x8a8a0000008a8a8aULL,
+	0x3232000000323232ULL, 0x4b4b0000004b4b4bULL, 0x4242000000424242ULL,
+	0xdbdb000000dbdbdbULL, 0x1c1c0000001c1c1cULL, 0x9e9e0000009e9e9eULL,
+	0x9c9c0000009c9c9cULL, 0x3a3a0000003a3a3aULL, 0xcaca000000cacacaULL,
+	0x2525000000252525ULL, 0x7b7b0000007b7b7bULL, 0x0d0d0000000d0d0dULL,
+	0x7171000000717171ULL, 0x5f5f0000005f5f5fULL, 0x1f1f0000001f1f1fULL,
+	0xf8f8000000f8f8f8ULL, 0xd7d7000000d7d7d7ULL, 0x3e3e0000003e3e3eULL,
+	0x9d9d0000009d9d9dULL, 0x7c7c0000007c7c7cULL, 0x6060000000606060ULL,
+	0xb9b9000000b9b9b9ULL, 0xbebe000000bebebeULL, 0xbcbc000000bcbcbcULL,
+	0x8b8b0000008b8b8bULL, 0x1616000000161616ULL, 0x3434000000343434ULL,
+	0x4d4d0000004d4d4dULL, 0xc3c3000000c3c3c3ULL, 0x7272000000727272ULL,
+	0x9595000000959595ULL, 0xabab000000abababULL, 0x8e8e0000008e8e8eULL,
+	0xbaba000000bababaULL, 0x7a7a0000007a7a7aULL, 0xb3b3000000b3b3b3ULL,
+	0x0202000000020202ULL, 0xb4b4000000b4b4b4ULL, 0xadad000000adadadULL,
+	0xa2a2000000a2a2a2ULL, 0xacac000000acacacULL, 0xd8d8000000d8d8d8ULL,
+	0x9a9a0000009a9a9aULL, 0x1717000000171717ULL, 0x1a1a0000001a1a1aULL,
+	0x3535000000353535ULL, 0xcccc000000ccccccULL, 0xf7f7000000f7f7f7ULL,
+	0x9999000000999999ULL, 0x6161000000616161ULL, 0x5a5a0000005a5a5aULL,
+	0xe8e8000000e8e8e8ULL, 0x2424000000242424ULL, 0x5656000000565656ULL,
+	0x4040000000404040ULL, 0xe1e1000000e1e1e1ULL, 0x6363000000636363ULL,
+	0x0909000000090909ULL, 0x3333000000333333ULL, 0xbfbf000000bfbfbfULL,
+	0x9898000000989898ULL, 0x9797000000979797ULL, 0x8585000000858585ULL,
+	0x6868000000686868ULL, 0xfcfc000000fcfcfcULL, 0xecec000000ecececULL,
+	0x0a0a0000000a0a0aULL, 0xdada000000dadadaULL, 0x6f6f0000006f6f6fULL,
+	0x5353000000535353ULL, 0x6262000000626262ULL, 0xa3a3000000a3a3a3ULL,
+	0x2e2e0000002e2e2eULL, 0x0808000000080808ULL, 0xafaf000000afafafULL,
+	0x2828000000282828ULL, 0xb0b0000000b0b0b0ULL, 0x7474000000747474ULL,
+	0xc2c2000000c2c2c2ULL, 0xbdbd000000bdbdbdULL, 0x3636000000363636ULL,
+	0x2222000000222222ULL, 0x3838000000383838ULL, 0x6464000000646464ULL,
+	0x1e1e0000001e1e1eULL, 0x3939000000393939ULL, 0x2c2c0000002c2c2cULL,
+	0xa6a6000000a6a6a6ULL, 0x3030000000303030ULL, 0xe5e5000000e5e5e5ULL,
+	0x4444000000444444ULL, 0xfdfd000000fdfdfdULL, 0x8888000000888888ULL,
+	0x9f9f0000009f9f9fULL, 0x6565000000656565ULL, 0x8787000000878787ULL,
+	0x6b6b0000006b6b6bULL, 0xf4f4000000f4f4f4ULL, 0x2323000000232323ULL,
+	0x4848000000484848ULL, 0x1010000000101010ULL, 0xd1d1000000d1d1d1ULL,
+	0x5151000000515151ULL, 0xc0c0000000c0c0c0ULL, 0xf9f9000000f9f9f9ULL,
+	0xd2d2000000d2d2d2ULL, 0xa0a0000000a0a0a0ULL, 0x5555000000555555ULL,
+	0xa1a1000000a1a1a1ULL, 0x4141000000414141ULL, 0xfafa000000fafafaULL,
+	0x4343000000434343ULL, 0x1313000000131313ULL, 0xc4c4000000c4c4c4ULL,
+	0x2f2f0000002f2f2fULL, 0xa8a8000000a8a8a8ULL, 0xb6b6000000b6b6b6ULL,
+	0x3c3c0000003c3c3cULL, 0x2b2b0000002b2b2bULL, 0xc1c1000000c1c1c1ULL,
+	0xffff000000ffffffULL, 0xc8c8000000c8c8c8ULL, 0xa5a5000000a5a5a5ULL,
+	0x2020000000202020ULL, 0x8989000000898989ULL, 0x0000000000000000ULL,
+	0x9090000000909090ULL, 0x4747000000474747ULL, 0xefef000000efefefULL,
+	0xeaea000000eaeaeaULL, 0xb7b7000000b7b7b7ULL, 0x1515000000151515ULL,
+	0x0606000000060606ULL, 0xcdcd000000cdcdcdULL, 0xb5b5000000b5b5b5ULL,
+	0x1212000000121212ULL, 0x7e7e0000007e7e7eULL, 0xbbbb000000bbbbbbULL,
+	0x2929000000292929ULL, 0x0f0f0000000f0f0fULL, 0xb8b8000000b8b8b8ULL,
+	0x0707000000070707ULL, 0x0404000000040404ULL, 0x9b9b0000009b9b9bULL,
+	0x9494000000949494ULL, 0x2121000000212121ULL, 0x6666000000666666ULL,
+	0xe6e6000000e6e6e6ULL, 0xcece000000cececeULL, 0xeded000000edededULL,
+	0xe7e7000000e7e7e7ULL, 0x3b3b0000003b3b3bULL, 0xfefe000000fefefeULL,
+	0x7f7f0000007f7f7fULL, 0xc5c5000000c5c5c5ULL, 0xa4a4000000a4a4a4ULL,
+	0x3737000000373737ULL, 0xb1b1000000b1b1b1ULL, 0x4c4c0000004c4c4cULL,
+	0x9191000000919191ULL, 0x6e6e0000006e6e6eULL, 0x8d8d0000008d8d8dULL,
+	0x7676000000767676ULL, 0x0303000000030303ULL, 0x2d2d0000002d2d2dULL,
+	0xdede000000dededeULL, 0x9696000000969696ULL, 0x2626000000262626ULL,
+	0x7d7d0000007d7d7dULL, 0xc6c6000000c6c6c6ULL, 0x5c5c0000005c5c5cULL,
+	0xd3d3000000d3d3d3ULL, 0xf2f2000000f2f2f2ULL, 0x4f4f0000004f4f4fULL,
+	0x1919000000191919ULL, 0x3f3f0000003f3f3fULL, 0xdcdc000000dcdcdcULL,
+	0x7979000000797979ULL, 0x1d1d0000001d1d1dULL, 0x5252000000525252ULL,
+	0xebeb000000ebebebULL, 0xf3f3000000f3f3f3ULL, 0x6d6d0000006d6d6dULL,
+	0x5e5e0000005e5e5eULL, 0xfbfb000000fbfbfbULL, 0x6969000000696969ULL,
+	0xb2b2000000b2b2b2ULL, 0xf0f0000000f0f0f0ULL, 0x3131000000313131ULL,
+	0x0c0c0000000c0c0cULL, 0xd4d4000000d4d4d4ULL, 0xcfcf000000cfcfcfULL,
+	0x8c8c0000008c8c8cULL, 0xe2e2000000e2e2e2ULL, 0x7575000000757575ULL,
+	0xa9a9000000a9a9a9ULL, 0x4a4a0000004a4a4aULL, 0x5757000000575757ULL,
+	0x8484000000848484ULL, 0x1111000000111111ULL, 0x4545000000454545ULL,
+	0x1b1b0000001b1b1bULL, 0xf5f5000000f5f5f5ULL, 0xe4e4000000e4e4e4ULL,
+	0x0e0e0000000e0e0eULL, 0x7373000000737373ULL, 0xaaaa000000aaaaaaULL,
+	0xf1f1000000f1f1f1ULL, 0xdddd000000ddddddULL, 0x5959000000595959ULL,
+	0x1414000000141414ULL, 0x6c6c0000006c6c6cULL, 0x9292000000929292ULL,
+	0x5454000000545454ULL, 0xd0d0000000d0d0d0ULL, 0x7878000000787878ULL,
+	0x7070000000707070ULL, 0xe3e3000000e3e3e3ULL, 0x4949000000494949ULL,
+	0x8080000000808080ULL, 0x5050000000505050ULL, 0xa7a7000000a7a7a7ULL,
+	0xf6f6000000f6f6f6ULL, 0x7777000000777777ULL, 0x9393000000939393ULL,
+	0x8686000000868686ULL, 0x8383000000838383ULL, 0x2a2a0000002a2a2aULL,
+	0xc7c7000000c7c7c7ULL, 0x5b5b0000005b5b5bULL, 0xe9e9000000e9e9e9ULL,
+	0xeeee000000eeeeeeULL, 0x8f8f0000008f8f8fULL, 0x0101000000010101ULL,
+	0x3d3d0000003d3d3dULL,
+};
+
+__visible const u64 camellia_sp03303033[256] = {
+	0x0038380038003838ULL, 0x0041410041004141ULL, 0x0016160016001616ULL,
+	0x0076760076007676ULL, 0x00d9d900d900d9d9ULL, 0x0093930093009393ULL,
+	0x0060600060006060ULL, 0x00f2f200f200f2f2ULL, 0x0072720072007272ULL,
+	0x00c2c200c200c2c2ULL, 0x00abab00ab00ababULL, 0x009a9a009a009a9aULL,
+	0x0075750075007575ULL, 0x0006060006000606ULL, 0x0057570057005757ULL,
+	0x00a0a000a000a0a0ULL, 0x0091910091009191ULL, 0x00f7f700f700f7f7ULL,
+	0x00b5b500b500b5b5ULL, 0x00c9c900c900c9c9ULL, 0x00a2a200a200a2a2ULL,
+	0x008c8c008c008c8cULL, 0x00d2d200d200d2d2ULL, 0x0090900090009090ULL,
+	0x00f6f600f600f6f6ULL, 0x0007070007000707ULL, 0x00a7a700a700a7a7ULL,
+	0x0027270027002727ULL, 0x008e8e008e008e8eULL, 0x00b2b200b200b2b2ULL,
+	0x0049490049004949ULL, 0x00dede00de00dedeULL, 0x0043430043004343ULL,
+	0x005c5c005c005c5cULL, 0x00d7d700d700d7d7ULL, 0x00c7c700c700c7c7ULL,
+	0x003e3e003e003e3eULL, 0x00f5f500f500f5f5ULL, 0x008f8f008f008f8fULL,
+	0x0067670067006767ULL, 0x001f1f001f001f1fULL, 0x0018180018001818ULL,
+	0x006e6e006e006e6eULL, 0x00afaf00af00afafULL, 0x002f2f002f002f2fULL,
+	0x00e2e200e200e2e2ULL, 0x0085850085008585ULL, 0x000d0d000d000d0dULL,
+	0x0053530053005353ULL, 0x00f0f000f000f0f0ULL, 0x009c9c009c009c9cULL,
+	0x0065650065006565ULL, 0x00eaea00ea00eaeaULL, 0x00a3a300a300a3a3ULL,
+	0x00aeae00ae00aeaeULL, 0x009e9e009e009e9eULL, 0x00ecec00ec00ececULL,
+	0x0080800080008080ULL, 0x002d2d002d002d2dULL, 0x006b6b006b006b6bULL,
+	0x00a8a800a800a8a8ULL, 0x002b2b002b002b2bULL, 0x0036360036003636ULL,
+	0x00a6a600a600a6a6ULL, 0x00c5c500c500c5c5ULL, 0x0086860086008686ULL,
+	0x004d4d004d004d4dULL, 0x0033330033003333ULL, 0x00fdfd00fd00fdfdULL,
+	0x0066660066006666ULL, 0x0058580058005858ULL, 0x0096960096009696ULL,
+	0x003a3a003a003a3aULL, 0x0009090009000909ULL, 0x0095950095009595ULL,
+	0x0010100010001010ULL, 0x0078780078007878ULL, 0x00d8d800d800d8d8ULL,
+	0x0042420042004242ULL, 0x00cccc00cc00ccccULL, 0x00efef00ef00efefULL,
+	0x0026260026002626ULL, 0x00e5e500e500e5e5ULL, 0x0061610061006161ULL,
+	0x001a1a001a001a1aULL, 0x003f3f003f003f3fULL, 0x003b3b003b003b3bULL,
+	0x0082820082008282ULL, 0x00b6b600b600b6b6ULL, 0x00dbdb00db00dbdbULL,
+	0x00d4d400d400d4d4ULL, 0x0098980098009898ULL, 0x00e8e800e800e8e8ULL,
+	0x008b8b008b008b8bULL, 0x0002020002000202ULL, 0x00ebeb00eb00ebebULL,
+	0x000a0a000a000a0aULL, 0x002c2c002c002c2cULL, 0x001d1d001d001d1dULL,
+	0x00b0b000b000b0b0ULL, 0x006f6f006f006f6fULL, 0x008d8d008d008d8dULL,
+	0x0088880088008888ULL, 0x000e0e000e000e0eULL, 0x0019190019001919ULL,
+	0x0087870087008787ULL, 0x004e4e004e004e4eULL, 0x000b0b000b000b0bULL,
+	0x00a9a900a900a9a9ULL, 0x000c0c000c000c0cULL, 0x0079790079007979ULL,
+	0x0011110011001111ULL, 0x007f7f007f007f7fULL, 0x0022220022002222ULL,
+	0x00e7e700e700e7e7ULL, 0x0059590059005959ULL, 0x00e1e100e100e1e1ULL,
+	0x00dada00da00dadaULL, 0x003d3d003d003d3dULL, 0x00c8c800c800c8c8ULL,
+	0x0012120012001212ULL, 0x0004040004000404ULL, 0x0074740074007474ULL,
+	0x0054540054005454ULL, 0x0030300030003030ULL, 0x007e7e007e007e7eULL,
+	0x00b4b400b400b4b4ULL, 0x0028280028002828ULL, 0x0055550055005555ULL,
+	0x0068680068006868ULL, 0x0050500050005050ULL, 0x00bebe00be00bebeULL,
+	0x00d0d000d000d0d0ULL, 0x00c4c400c400c4c4ULL, 0x0031310031003131ULL,
+	0x00cbcb00cb00cbcbULL, 0x002a2a002a002a2aULL, 0x00adad00ad00adadULL,
+	0x000f0f000f000f0fULL, 0x00caca00ca00cacaULL, 0x0070700070007070ULL,
+	0x00ffff00ff00ffffULL, 0x0032320032003232ULL, 0x0069690069006969ULL,
+	0x0008080008000808ULL, 0x0062620062006262ULL, 0x0000000000000000ULL,
+	0x0024240024002424ULL, 0x00d1d100d100d1d1ULL, 0x00fbfb00fb00fbfbULL,
+	0x00baba00ba00babaULL, 0x00eded00ed00ededULL, 0x0045450045004545ULL,
+	0x0081810081008181ULL, 0x0073730073007373ULL, 0x006d6d006d006d6dULL,
+	0x0084840084008484ULL, 0x009f9f009f009f9fULL, 0x00eeee00ee00eeeeULL,
+	0x004a4a004a004a4aULL, 0x00c3c300c300c3c3ULL, 0x002e2e002e002e2eULL,
+	0x00c1c100c100c1c1ULL, 0x0001010001000101ULL, 0x00e6e600e600e6e6ULL,
+	0x0025250025002525ULL, 0x0048480048004848ULL, 0x0099990099009999ULL,
+	0x00b9b900b900b9b9ULL, 0x00b3b300b300b3b3ULL, 0x007b7b007b007b7bULL,
+	0x00f9f900f900f9f9ULL, 0x00cece00ce00ceceULL, 0x00bfbf00bf00bfbfULL,
+	0x00dfdf00df00dfdfULL, 0x0071710071007171ULL, 0x0029290029002929ULL,
+	0x00cdcd00cd00cdcdULL, 0x006c6c006c006c6cULL, 0x0013130013001313ULL,
+	0x0064640064006464ULL, 0x009b9b009b009b9bULL, 0x0063630063006363ULL,
+	0x009d9d009d009d9dULL, 0x00c0c000c000c0c0ULL, 0x004b4b004b004b4bULL,
+	0x00b7b700b700b7b7ULL, 0x00a5a500a500a5a5ULL, 0x0089890089008989ULL,
+	0x005f5f005f005f5fULL, 0x00b1b100b100b1b1ULL, 0x0017170017001717ULL,
+	0x00f4f400f400f4f4ULL, 0x00bcbc00bc00bcbcULL, 0x00d3d300d300d3d3ULL,
+	0x0046460046004646ULL, 0x00cfcf00cf00cfcfULL, 0x0037370037003737ULL,
+	0x005e5e005e005e5eULL, 0x0047470047004747ULL, 0x0094940094009494ULL,
+	0x00fafa00fa00fafaULL, 0x00fcfc00fc00fcfcULL, 0x005b5b005b005b5bULL,
+	0x0097970097009797ULL, 0x00fefe00fe00fefeULL, 0x005a5a005a005a5aULL,
+	0x00acac00ac00acacULL, 0x003c3c003c003c3cULL, 0x004c4c004c004c4cULL,
+	0x0003030003000303ULL, 0x0035350035003535ULL, 0x00f3f300f300f3f3ULL,
+	0x0023230023002323ULL, 0x00b8b800b800b8b8ULL, 0x005d5d005d005d5dULL,
+	0x006a6a006a006a6aULL, 0x0092920092009292ULL, 0x00d5d500d500d5d5ULL,
+	0x0021210021002121ULL, 0x0044440044004444ULL, 0x0051510051005151ULL,
+	0x00c6c600c600c6c6ULL, 0x007d7d007d007d7dULL, 0x0039390039003939ULL,
+	0x0083830083008383ULL, 0x00dcdc00dc00dcdcULL, 0x00aaaa00aa00aaaaULL,
+	0x007c7c007c007c7cULL, 0x0077770077007777ULL, 0x0056560056005656ULL,
+	0x0005050005000505ULL, 0x001b1b001b001b1bULL, 0x00a4a400a400a4a4ULL,
+	0x0015150015001515ULL, 0x0034340034003434ULL, 0x001e1e001e001e1eULL,
+	0x001c1c001c001c1cULL, 0x00f8f800f800f8f8ULL, 0x0052520052005252ULL,
+	0x0020200020002020ULL, 0x0014140014001414ULL, 0x00e9e900e900e9e9ULL,
+	0x00bdbd00bd00bdbdULL, 0x00dddd00dd00ddddULL, 0x00e4e400e400e4e4ULL,
+	0x00a1a100a100a1a1ULL, 0x00e0e000e000e0e0ULL, 0x008a8a008a008a8aULL,
+	0x00f1f100f100f1f1ULL, 0x00d6d600d600d6d6ULL, 0x007a7a007a007a7aULL,
+	0x00bbbb00bb00bbbbULL, 0x00e3e300e300e3e3ULL, 0x0040400040004040ULL,
+	0x004f4f004f004f4fULL,
+};
+
+__visible const u64 camellia_sp00444404[256] = {
+	0x0000707070700070ULL, 0x00002c2c2c2c002cULL, 0x0000b3b3b3b300b3ULL,
+	0x0000c0c0c0c000c0ULL, 0x0000e4e4e4e400e4ULL, 0x0000575757570057ULL,
+	0x0000eaeaeaea00eaULL, 0x0000aeaeaeae00aeULL, 0x0000232323230023ULL,
+	0x00006b6b6b6b006bULL, 0x0000454545450045ULL, 0x0000a5a5a5a500a5ULL,
+	0x0000edededed00edULL, 0x00004f4f4f4f004fULL, 0x00001d1d1d1d001dULL,
+	0x0000929292920092ULL, 0x0000868686860086ULL, 0x0000afafafaf00afULL,
+	0x00007c7c7c7c007cULL, 0x00001f1f1f1f001fULL, 0x00003e3e3e3e003eULL,
+	0x0000dcdcdcdc00dcULL, 0x00005e5e5e5e005eULL, 0x00000b0b0b0b000bULL,
+	0x0000a6a6a6a600a6ULL, 0x0000393939390039ULL, 0x0000d5d5d5d500d5ULL,
+	0x00005d5d5d5d005dULL, 0x0000d9d9d9d900d9ULL, 0x00005a5a5a5a005aULL,
+	0x0000515151510051ULL, 0x00006c6c6c6c006cULL, 0x00008b8b8b8b008bULL,
+	0x00009a9a9a9a009aULL, 0x0000fbfbfbfb00fbULL, 0x0000b0b0b0b000b0ULL,
+	0x0000747474740074ULL, 0x00002b2b2b2b002bULL, 0x0000f0f0f0f000f0ULL,
+	0x0000848484840084ULL, 0x0000dfdfdfdf00dfULL, 0x0000cbcbcbcb00cbULL,
+	0x0000343434340034ULL, 0x0000767676760076ULL, 0x00006d6d6d6d006dULL,
+	0x0000a9a9a9a900a9ULL, 0x0000d1d1d1d100d1ULL, 0x0000040404040004ULL,
+	0x0000141414140014ULL, 0x00003a3a3a3a003aULL, 0x0000dededede00deULL,
+	0x0000111111110011ULL, 0x0000323232320032ULL, 0x00009c9c9c9c009cULL,
+	0x0000535353530053ULL, 0x0000f2f2f2f200f2ULL, 0x0000fefefefe00feULL,
+	0x0000cfcfcfcf00cfULL, 0x0000c3c3c3c300c3ULL, 0x00007a7a7a7a007aULL,
+	0x0000242424240024ULL, 0x0000e8e8e8e800e8ULL, 0x0000606060600060ULL,
+	0x0000696969690069ULL, 0x0000aaaaaaaa00aaULL, 0x0000a0a0a0a000a0ULL,
+	0x0000a1a1a1a100a1ULL, 0x0000626262620062ULL, 0x0000545454540054ULL,
+	0x00001e1e1e1e001eULL, 0x0000e0e0e0e000e0ULL, 0x0000646464640064ULL,
+	0x0000101010100010ULL, 0x0000000000000000ULL, 0x0000a3a3a3a300a3ULL,
+	0x0000757575750075ULL, 0x00008a8a8a8a008aULL, 0x0000e6e6e6e600e6ULL,
+	0x0000090909090009ULL, 0x0000dddddddd00ddULL, 0x0000878787870087ULL,
+	0x0000838383830083ULL, 0x0000cdcdcdcd00cdULL, 0x0000909090900090ULL,
+	0x0000737373730073ULL, 0x0000f6f6f6f600f6ULL, 0x00009d9d9d9d009dULL,
+	0x0000bfbfbfbf00bfULL, 0x0000525252520052ULL, 0x0000d8d8d8d800d8ULL,
+	0x0000c8c8c8c800c8ULL, 0x0000c6c6c6c600c6ULL, 0x0000818181810081ULL,
+	0x00006f6f6f6f006fULL, 0x0000131313130013ULL, 0x0000636363630063ULL,
+	0x0000e9e9e9e900e9ULL, 0x0000a7a7a7a700a7ULL, 0x00009f9f9f9f009fULL,
+	0x0000bcbcbcbc00bcULL, 0x0000292929290029ULL, 0x0000f9f9f9f900f9ULL,
+	0x00002f2f2f2f002fULL, 0x0000b4b4b4b400b4ULL, 0x0000787878780078ULL,
+	0x0000060606060006ULL, 0x0000e7e7e7e700e7ULL, 0x0000717171710071ULL,
+	0x0000d4d4d4d400d4ULL, 0x0000abababab00abULL, 0x0000888888880088ULL,
+	0x00008d8d8d8d008dULL, 0x0000727272720072ULL, 0x0000b9b9b9b900b9ULL,
+	0x0000f8f8f8f800f8ULL, 0x0000acacacac00acULL, 0x0000363636360036ULL,
+	0x00002a2a2a2a002aULL, 0x00003c3c3c3c003cULL, 0x0000f1f1f1f100f1ULL,
+	0x0000404040400040ULL, 0x0000d3d3d3d300d3ULL, 0x0000bbbbbbbb00bbULL,
+	0x0000434343430043ULL, 0x0000151515150015ULL, 0x0000adadadad00adULL,
+	0x0000777777770077ULL, 0x0000808080800080ULL, 0x0000828282820082ULL,
+	0x0000ecececec00ecULL, 0x0000272727270027ULL, 0x0000e5e5e5e500e5ULL,
+	0x0000858585850085ULL, 0x0000353535350035ULL, 0x00000c0c0c0c000cULL,
+	0x0000414141410041ULL, 0x0000efefefef00efULL, 0x0000939393930093ULL,
+	0x0000191919190019ULL, 0x0000212121210021ULL, 0x00000e0e0e0e000eULL,
+	0x00004e4e4e4e004eULL, 0x0000656565650065ULL, 0x0000bdbdbdbd00bdULL,
+	0x0000b8b8b8b800b8ULL, 0x00008f8f8f8f008fULL, 0x0000ebebebeb00ebULL,
+	0x0000cececece00ceULL, 0x0000303030300030ULL, 0x00005f5f5f5f005fULL,
+	0x0000c5c5c5c500c5ULL, 0x00001a1a1a1a001aULL, 0x0000e1e1e1e100e1ULL,
+	0x0000cacacaca00caULL, 0x0000474747470047ULL, 0x00003d3d3d3d003dULL,
+	0x0000010101010001ULL, 0x0000d6d6d6d600d6ULL, 0x0000565656560056ULL,
+	0x00004d4d4d4d004dULL, 0x00000d0d0d0d000dULL, 0x0000666666660066ULL,
+	0x0000cccccccc00ccULL, 0x00002d2d2d2d002dULL, 0x0000121212120012ULL,
+	0x0000202020200020ULL, 0x0000b1b1b1b100b1ULL, 0x0000999999990099ULL,
+	0x00004c4c4c4c004cULL, 0x0000c2c2c2c200c2ULL, 0x00007e7e7e7e007eULL,
+	0x0000050505050005ULL, 0x0000b7b7b7b700b7ULL, 0x0000313131310031ULL,
+	0x0000171717170017ULL, 0x0000d7d7d7d700d7ULL, 0x0000585858580058ULL,
+	0x0000616161610061ULL, 0x00001b1b1b1b001bULL, 0x00001c1c1c1c001cULL,
+	0x00000f0f0f0f000fULL, 0x0000161616160016ULL, 0x0000181818180018ULL,
+	0x0000222222220022ULL, 0x0000444444440044ULL, 0x0000b2b2b2b200b2ULL,
+	0x0000b5b5b5b500b5ULL, 0x0000919191910091ULL, 0x0000080808080008ULL,
+	0x0000a8a8a8a800a8ULL, 0x0000fcfcfcfc00fcULL, 0x0000505050500050ULL,
+	0x0000d0d0d0d000d0ULL, 0x00007d7d7d7d007dULL, 0x0000898989890089ULL,
+	0x0000979797970097ULL, 0x00005b5b5b5b005bULL, 0x0000959595950095ULL,
+	0x0000ffffffff00ffULL, 0x0000d2d2d2d200d2ULL, 0x0000c4c4c4c400c4ULL,
+	0x0000484848480048ULL, 0x0000f7f7f7f700f7ULL, 0x0000dbdbdbdb00dbULL,
+	0x0000030303030003ULL, 0x0000dadadada00daULL, 0x00003f3f3f3f003fULL,
+	0x0000949494940094ULL, 0x00005c5c5c5c005cULL, 0x0000020202020002ULL,
+	0x00004a4a4a4a004aULL, 0x0000333333330033ULL, 0x0000676767670067ULL,
+	0x0000f3f3f3f300f3ULL, 0x00007f7f7f7f007fULL, 0x0000e2e2e2e200e2ULL,
+	0x00009b9b9b9b009bULL, 0x0000262626260026ULL, 0x0000373737370037ULL,
+	0x00003b3b3b3b003bULL, 0x0000969696960096ULL, 0x00004b4b4b4b004bULL,
+	0x0000bebebebe00beULL, 0x00002e2e2e2e002eULL, 0x0000797979790079ULL,
+	0x00008c8c8c8c008cULL, 0x00006e6e6e6e006eULL, 0x00008e8e8e8e008eULL,
+	0x0000f5f5f5f500f5ULL, 0x0000b6b6b6b600b6ULL, 0x0000fdfdfdfd00fdULL,
+	0x0000595959590059ULL, 0x0000989898980098ULL, 0x00006a6a6a6a006aULL,
+	0x0000464646460046ULL, 0x0000babababa00baULL, 0x0000252525250025ULL,
+	0x0000424242420042ULL, 0x0000a2a2a2a200a2ULL, 0x0000fafafafa00faULL,
+	0x0000070707070007ULL, 0x0000555555550055ULL, 0x0000eeeeeeee00eeULL,
+	0x00000a0a0a0a000aULL, 0x0000494949490049ULL, 0x0000686868680068ULL,
+	0x0000383838380038ULL, 0x0000a4a4a4a400a4ULL, 0x0000282828280028ULL,
+	0x00007b7b7b7b007bULL, 0x0000c9c9c9c900c9ULL, 0x0000c1c1c1c100c1ULL,
+	0x0000e3e3e3e300e3ULL, 0x0000f4f4f4f400f4ULL, 0x0000c7c7c7c700c7ULL,
+	0x00009e9e9e9e009eULL,
+};
+
+__visible const u64 camellia_sp02220222[256] = {
+	0x00e0e0e000e0e0e0ULL, 0x0005050500050505ULL, 0x0058585800585858ULL,
+	0x00d9d9d900d9d9d9ULL, 0x0067676700676767ULL, 0x004e4e4e004e4e4eULL,
+	0x0081818100818181ULL, 0x00cbcbcb00cbcbcbULL, 0x00c9c9c900c9c9c9ULL,
+	0x000b0b0b000b0b0bULL, 0x00aeaeae00aeaeaeULL, 0x006a6a6a006a6a6aULL,
+	0x00d5d5d500d5d5d5ULL, 0x0018181800181818ULL, 0x005d5d5d005d5d5dULL,
+	0x0082828200828282ULL, 0x0046464600464646ULL, 0x00dfdfdf00dfdfdfULL,
+	0x00d6d6d600d6d6d6ULL, 0x0027272700272727ULL, 0x008a8a8a008a8a8aULL,
+	0x0032323200323232ULL, 0x004b4b4b004b4b4bULL, 0x0042424200424242ULL,
+	0x00dbdbdb00dbdbdbULL, 0x001c1c1c001c1c1cULL, 0x009e9e9e009e9e9eULL,
+	0x009c9c9c009c9c9cULL, 0x003a3a3a003a3a3aULL, 0x00cacaca00cacacaULL,
+	0x0025252500252525ULL, 0x007b7b7b007b7b7bULL, 0x000d0d0d000d0d0dULL,
+	0x0071717100717171ULL, 0x005f5f5f005f5f5fULL, 0x001f1f1f001f1f1fULL,
+	0x00f8f8f800f8f8f8ULL, 0x00d7d7d700d7d7d7ULL, 0x003e3e3e003e3e3eULL,
+	0x009d9d9d009d9d9dULL, 0x007c7c7c007c7c7cULL, 0x0060606000606060ULL,
+	0x00b9b9b900b9b9b9ULL, 0x00bebebe00bebebeULL, 0x00bcbcbc00bcbcbcULL,
+	0x008b8b8b008b8b8bULL, 0x0016161600161616ULL, 0x0034343400343434ULL,
+	0x004d4d4d004d4d4dULL, 0x00c3c3c300c3c3c3ULL, 0x0072727200727272ULL,
+	0x0095959500959595ULL, 0x00ababab00abababULL, 0x008e8e8e008e8e8eULL,
+	0x00bababa00bababaULL, 0x007a7a7a007a7a7aULL, 0x00b3b3b300b3b3b3ULL,
+	0x0002020200020202ULL, 0x00b4b4b400b4b4b4ULL, 0x00adadad00adadadULL,
+	0x00a2a2a200a2a2a2ULL, 0x00acacac00acacacULL, 0x00d8d8d800d8d8d8ULL,
+	0x009a9a9a009a9a9aULL, 0x0017171700171717ULL, 0x001a1a1a001a1a1aULL,
+	0x0035353500353535ULL, 0x00cccccc00ccccccULL, 0x00f7f7f700f7f7f7ULL,
+	0x0099999900999999ULL, 0x0061616100616161ULL, 0x005a5a5a005a5a5aULL,
+	0x00e8e8e800e8e8e8ULL, 0x0024242400242424ULL, 0x0056565600565656ULL,
+	0x0040404000404040ULL, 0x00e1e1e100e1e1e1ULL, 0x0063636300636363ULL,
+	0x0009090900090909ULL, 0x0033333300333333ULL, 0x00bfbfbf00bfbfbfULL,
+	0x0098989800989898ULL, 0x0097979700979797ULL, 0x0085858500858585ULL,
+	0x0068686800686868ULL, 0x00fcfcfc00fcfcfcULL, 0x00ececec00ecececULL,
+	0x000a0a0a000a0a0aULL, 0x00dadada00dadadaULL, 0x006f6f6f006f6f6fULL,
+	0x0053535300535353ULL, 0x0062626200626262ULL, 0x00a3a3a300a3a3a3ULL,
+	0x002e2e2e002e2e2eULL, 0x0008080800080808ULL, 0x00afafaf00afafafULL,
+	0x0028282800282828ULL, 0x00b0b0b000b0b0b0ULL, 0x0074747400747474ULL,
+	0x00c2c2c200c2c2c2ULL, 0x00bdbdbd00bdbdbdULL, 0x0036363600363636ULL,
+	0x0022222200222222ULL, 0x0038383800383838ULL, 0x0064646400646464ULL,
+	0x001e1e1e001e1e1eULL, 0x0039393900393939ULL, 0x002c2c2c002c2c2cULL,
+	0x00a6a6a600a6a6a6ULL, 0x0030303000303030ULL, 0x00e5e5e500e5e5e5ULL,
+	0x0044444400444444ULL, 0x00fdfdfd00fdfdfdULL, 0x0088888800888888ULL,
+	0x009f9f9f009f9f9fULL, 0x0065656500656565ULL, 0x0087878700878787ULL,
+	0x006b6b6b006b6b6bULL, 0x00f4f4f400f4f4f4ULL, 0x0023232300232323ULL,
+	0x0048484800484848ULL, 0x0010101000101010ULL, 0x00d1d1d100d1d1d1ULL,
+	0x0051515100515151ULL, 0x00c0c0c000c0c0c0ULL, 0x00f9f9f900f9f9f9ULL,
+	0x00d2d2d200d2d2d2ULL, 0x00a0a0a000a0a0a0ULL, 0x0055555500555555ULL,
+	0x00a1a1a100a1a1a1ULL, 0x0041414100414141ULL, 0x00fafafa00fafafaULL,
+	0x0043434300434343ULL, 0x0013131300131313ULL, 0x00c4c4c400c4c4c4ULL,
+	0x002f2f2f002f2f2fULL, 0x00a8a8a800a8a8a8ULL, 0x00b6b6b600b6b6b6ULL,
+	0x003c3c3c003c3c3cULL, 0x002b2b2b002b2b2bULL, 0x00c1c1c100c1c1c1ULL,
+	0x00ffffff00ffffffULL, 0x00c8c8c800c8c8c8ULL, 0x00a5a5a500a5a5a5ULL,
+	0x0020202000202020ULL, 0x0089898900898989ULL, 0x0000000000000000ULL,
+	0x0090909000909090ULL, 0x0047474700474747ULL, 0x00efefef00efefefULL,
+	0x00eaeaea00eaeaeaULL, 0x00b7b7b700b7b7b7ULL, 0x0015151500151515ULL,
+	0x0006060600060606ULL, 0x00cdcdcd00cdcdcdULL, 0x00b5b5b500b5b5b5ULL,
+	0x0012121200121212ULL, 0x007e7e7e007e7e7eULL, 0x00bbbbbb00bbbbbbULL,
+	0x0029292900292929ULL, 0x000f0f0f000f0f0fULL, 0x00b8b8b800b8b8b8ULL,
+	0x0007070700070707ULL, 0x0004040400040404ULL, 0x009b9b9b009b9b9bULL,
+	0x0094949400949494ULL, 0x0021212100212121ULL, 0x0066666600666666ULL,
+	0x00e6e6e600e6e6e6ULL, 0x00cecece00cececeULL, 0x00ededed00edededULL,
+	0x00e7e7e700e7e7e7ULL, 0x003b3b3b003b3b3bULL, 0x00fefefe00fefefeULL,
+	0x007f7f7f007f7f7fULL, 0x00c5c5c500c5c5c5ULL, 0x00a4a4a400a4a4a4ULL,
+	0x0037373700373737ULL, 0x00b1b1b100b1b1b1ULL, 0x004c4c4c004c4c4cULL,
+	0x0091919100919191ULL, 0x006e6e6e006e6e6eULL, 0x008d8d8d008d8d8dULL,
+	0x0076767600767676ULL, 0x0003030300030303ULL, 0x002d2d2d002d2d2dULL,
+	0x00dedede00dededeULL, 0x0096969600969696ULL, 0x0026262600262626ULL,
+	0x007d7d7d007d7d7dULL, 0x00c6c6c600c6c6c6ULL, 0x005c5c5c005c5c5cULL,
+	0x00d3d3d300d3d3d3ULL, 0x00f2f2f200f2f2f2ULL, 0x004f4f4f004f4f4fULL,
+	0x0019191900191919ULL, 0x003f3f3f003f3f3fULL, 0x00dcdcdc00dcdcdcULL,
+	0x0079797900797979ULL, 0x001d1d1d001d1d1dULL, 0x0052525200525252ULL,
+	0x00ebebeb00ebebebULL, 0x00f3f3f300f3f3f3ULL, 0x006d6d6d006d6d6dULL,
+	0x005e5e5e005e5e5eULL, 0x00fbfbfb00fbfbfbULL, 0x0069696900696969ULL,
+	0x00b2b2b200b2b2b2ULL, 0x00f0f0f000f0f0f0ULL, 0x0031313100313131ULL,
+	0x000c0c0c000c0c0cULL, 0x00d4d4d400d4d4d4ULL, 0x00cfcfcf00cfcfcfULL,
+	0x008c8c8c008c8c8cULL, 0x00e2e2e200e2e2e2ULL, 0x0075757500757575ULL,
+	0x00a9a9a900a9a9a9ULL, 0x004a4a4a004a4a4aULL, 0x0057575700575757ULL,
+	0x0084848400848484ULL, 0x0011111100111111ULL, 0x0045454500454545ULL,
+	0x001b1b1b001b1b1bULL, 0x00f5f5f500f5f5f5ULL, 0x00e4e4e400e4e4e4ULL,
+	0x000e0e0e000e0e0eULL, 0x0073737300737373ULL, 0x00aaaaaa00aaaaaaULL,
+	0x00f1f1f100f1f1f1ULL, 0x00dddddd00ddddddULL, 0x0059595900595959ULL,
+	0x0014141400141414ULL, 0x006c6c6c006c6c6cULL, 0x0092929200929292ULL,
+	0x0054545400545454ULL, 0x00d0d0d000d0d0d0ULL, 0x0078787800787878ULL,
+	0x0070707000707070ULL, 0x00e3e3e300e3e3e3ULL, 0x0049494900494949ULL,
+	0x0080808000808080ULL, 0x0050505000505050ULL, 0x00a7a7a700a7a7a7ULL,
+	0x00f6f6f600f6f6f6ULL, 0x0077777700777777ULL, 0x0093939300939393ULL,
+	0x0086868600868686ULL, 0x0083838300838383ULL, 0x002a2a2a002a2a2aULL,
+	0x00c7c7c700c7c7c7ULL, 0x005b5b5b005b5b5bULL, 0x00e9e9e900e9e9e9ULL,
+	0x00eeeeee00eeeeeeULL, 0x008f8f8f008f8f8fULL, 0x0001010100010101ULL,
+	0x003d3d3d003d3d3dULL,
+};
+
+__visible const u64 camellia_sp30333033[256] = {
+	0x3800383838003838ULL, 0x4100414141004141ULL, 0x1600161616001616ULL,
+	0x7600767676007676ULL, 0xd900d9d9d900d9d9ULL, 0x9300939393009393ULL,
+	0x6000606060006060ULL, 0xf200f2f2f200f2f2ULL, 0x7200727272007272ULL,
+	0xc200c2c2c200c2c2ULL, 0xab00ababab00ababULL, 0x9a009a9a9a009a9aULL,
+	0x7500757575007575ULL, 0x0600060606000606ULL, 0x5700575757005757ULL,
+	0xa000a0a0a000a0a0ULL, 0x9100919191009191ULL, 0xf700f7f7f700f7f7ULL,
+	0xb500b5b5b500b5b5ULL, 0xc900c9c9c900c9c9ULL, 0xa200a2a2a200a2a2ULL,
+	0x8c008c8c8c008c8cULL, 0xd200d2d2d200d2d2ULL, 0x9000909090009090ULL,
+	0xf600f6f6f600f6f6ULL, 0x0700070707000707ULL, 0xa700a7a7a700a7a7ULL,
+	0x2700272727002727ULL, 0x8e008e8e8e008e8eULL, 0xb200b2b2b200b2b2ULL,
+	0x4900494949004949ULL, 0xde00dedede00dedeULL, 0x4300434343004343ULL,
+	0x5c005c5c5c005c5cULL, 0xd700d7d7d700d7d7ULL, 0xc700c7c7c700c7c7ULL,
+	0x3e003e3e3e003e3eULL, 0xf500f5f5f500f5f5ULL, 0x8f008f8f8f008f8fULL,
+	0x6700676767006767ULL, 0x1f001f1f1f001f1fULL, 0x1800181818001818ULL,
+	0x6e006e6e6e006e6eULL, 0xaf00afafaf00afafULL, 0x2f002f2f2f002f2fULL,
+	0xe200e2e2e200e2e2ULL, 0x8500858585008585ULL, 0x0d000d0d0d000d0dULL,
+	0x5300535353005353ULL, 0xf000f0f0f000f0f0ULL, 0x9c009c9c9c009c9cULL,
+	0x6500656565006565ULL, 0xea00eaeaea00eaeaULL, 0xa300a3a3a300a3a3ULL,
+	0xae00aeaeae00aeaeULL, 0x9e009e9e9e009e9eULL, 0xec00ececec00ececULL,
+	0x8000808080008080ULL, 0x2d002d2d2d002d2dULL, 0x6b006b6b6b006b6bULL,
+	0xa800a8a8a800a8a8ULL, 0x2b002b2b2b002b2bULL, 0x3600363636003636ULL,
+	0xa600a6a6a600a6a6ULL, 0xc500c5c5c500c5c5ULL, 0x8600868686008686ULL,
+	0x4d004d4d4d004d4dULL, 0x3300333333003333ULL, 0xfd00fdfdfd00fdfdULL,
+	0x6600666666006666ULL, 0x5800585858005858ULL, 0x9600969696009696ULL,
+	0x3a003a3a3a003a3aULL, 0x0900090909000909ULL, 0x9500959595009595ULL,
+	0x1000101010001010ULL, 0x7800787878007878ULL, 0xd800d8d8d800d8d8ULL,
+	0x4200424242004242ULL, 0xcc00cccccc00ccccULL, 0xef00efefef00efefULL,
+	0x2600262626002626ULL, 0xe500e5e5e500e5e5ULL, 0x6100616161006161ULL,
+	0x1a001a1a1a001a1aULL, 0x3f003f3f3f003f3fULL, 0x3b003b3b3b003b3bULL,
+	0x8200828282008282ULL, 0xb600b6b6b600b6b6ULL, 0xdb00dbdbdb00dbdbULL,
+	0xd400d4d4d400d4d4ULL, 0x9800989898009898ULL, 0xe800e8e8e800e8e8ULL,
+	0x8b008b8b8b008b8bULL, 0x0200020202000202ULL, 0xeb00ebebeb00ebebULL,
+	0x0a000a0a0a000a0aULL, 0x2c002c2c2c002c2cULL, 0x1d001d1d1d001d1dULL,
+	0xb000b0b0b000b0b0ULL, 0x6f006f6f6f006f6fULL, 0x8d008d8d8d008d8dULL,
+	0x8800888888008888ULL, 0x0e000e0e0e000e0eULL, 0x1900191919001919ULL,
+	0x8700878787008787ULL, 0x4e004e4e4e004e4eULL, 0x0b000b0b0b000b0bULL,
+	0xa900a9a9a900a9a9ULL, 0x0c000c0c0c000c0cULL, 0x7900797979007979ULL,
+	0x1100111111001111ULL, 0x7f007f7f7f007f7fULL, 0x2200222222002222ULL,
+	0xe700e7e7e700e7e7ULL, 0x5900595959005959ULL, 0xe100e1e1e100e1e1ULL,
+	0xda00dadada00dadaULL, 0x3d003d3d3d003d3dULL, 0xc800c8c8c800c8c8ULL,
+	0x1200121212001212ULL, 0x0400040404000404ULL, 0x7400747474007474ULL,
+	0x5400545454005454ULL, 0x3000303030003030ULL, 0x7e007e7e7e007e7eULL,
+	0xb400b4b4b400b4b4ULL, 0x2800282828002828ULL, 0x5500555555005555ULL,
+	0x6800686868006868ULL, 0x5000505050005050ULL, 0xbe00bebebe00bebeULL,
+	0xd000d0d0d000d0d0ULL, 0xc400c4c4c400c4c4ULL, 0x3100313131003131ULL,
+	0xcb00cbcbcb00cbcbULL, 0x2a002a2a2a002a2aULL, 0xad00adadad00adadULL,
+	0x0f000f0f0f000f0fULL, 0xca00cacaca00cacaULL, 0x7000707070007070ULL,
+	0xff00ffffff00ffffULL, 0x3200323232003232ULL, 0x6900696969006969ULL,
+	0x0800080808000808ULL, 0x6200626262006262ULL, 0x0000000000000000ULL,
+	0x2400242424002424ULL, 0xd100d1d1d100d1d1ULL, 0xfb00fbfbfb00fbfbULL,
+	0xba00bababa00babaULL, 0xed00ededed00ededULL, 0x4500454545004545ULL,
+	0x8100818181008181ULL, 0x7300737373007373ULL, 0x6d006d6d6d006d6dULL,
+	0x8400848484008484ULL, 0x9f009f9f9f009f9fULL, 0xee00eeeeee00eeeeULL,
+	0x4a004a4a4a004a4aULL, 0xc300c3c3c300c3c3ULL, 0x2e002e2e2e002e2eULL,
+	0xc100c1c1c100c1c1ULL, 0x0100010101000101ULL, 0xe600e6e6e600e6e6ULL,
+	0x2500252525002525ULL, 0x4800484848004848ULL, 0x9900999999009999ULL,
+	0xb900b9b9b900b9b9ULL, 0xb300b3b3b300b3b3ULL, 0x7b007b7b7b007b7bULL,
+	0xf900f9f9f900f9f9ULL, 0xce00cecece00ceceULL, 0xbf00bfbfbf00bfbfULL,
+	0xdf00dfdfdf00dfdfULL, 0x7100717171007171ULL, 0x2900292929002929ULL,
+	0xcd00cdcdcd00cdcdULL, 0x6c006c6c6c006c6cULL, 0x1300131313001313ULL,
+	0x6400646464006464ULL, 0x9b009b9b9b009b9bULL, 0x6300636363006363ULL,
+	0x9d009d9d9d009d9dULL, 0xc000c0c0c000c0c0ULL, 0x4b004b4b4b004b4bULL,
+	0xb700b7b7b700b7b7ULL, 0xa500a5a5a500a5a5ULL, 0x8900898989008989ULL,
+	0x5f005f5f5f005f5fULL, 0xb100b1b1b100b1b1ULL, 0x1700171717001717ULL,
+	0xf400f4f4f400f4f4ULL, 0xbc00bcbcbc00bcbcULL, 0xd300d3d3d300d3d3ULL,
+	0x4600464646004646ULL, 0xcf00cfcfcf00cfcfULL, 0x3700373737003737ULL,
+	0x5e005e5e5e005e5eULL, 0x4700474747004747ULL, 0x9400949494009494ULL,
+	0xfa00fafafa00fafaULL, 0xfc00fcfcfc00fcfcULL, 0x5b005b5b5b005b5bULL,
+	0x9700979797009797ULL, 0xfe00fefefe00fefeULL, 0x5a005a5a5a005a5aULL,
+	0xac00acacac00acacULL, 0x3c003c3c3c003c3cULL, 0x4c004c4c4c004c4cULL,
+	0x0300030303000303ULL, 0x3500353535003535ULL, 0xf300f3f3f300f3f3ULL,
+	0x2300232323002323ULL, 0xb800b8b8b800b8b8ULL, 0x5d005d5d5d005d5dULL,
+	0x6a006a6a6a006a6aULL, 0x9200929292009292ULL, 0xd500d5d5d500d5d5ULL,
+	0x2100212121002121ULL, 0x4400444444004444ULL, 0x5100515151005151ULL,
+	0xc600c6c6c600c6c6ULL, 0x7d007d7d7d007d7dULL, 0x3900393939003939ULL,
+	0x8300838383008383ULL, 0xdc00dcdcdc00dcdcULL, 0xaa00aaaaaa00aaaaULL,
+	0x7c007c7c7c007c7cULL, 0x7700777777007777ULL, 0x5600565656005656ULL,
+	0x0500050505000505ULL, 0x1b001b1b1b001b1bULL, 0xa400a4a4a400a4a4ULL,
+	0x1500151515001515ULL, 0x3400343434003434ULL, 0x1e001e1e1e001e1eULL,
+	0x1c001c1c1c001c1cULL, 0xf800f8f8f800f8f8ULL, 0x5200525252005252ULL,
+	0x2000202020002020ULL, 0x1400141414001414ULL, 0xe900e9e9e900e9e9ULL,
+	0xbd00bdbdbd00bdbdULL, 0xdd00dddddd00ddddULL, 0xe400e4e4e400e4e4ULL,
+	0xa100a1a1a100a1a1ULL, 0xe000e0e0e000e0e0ULL, 0x8a008a8a8a008a8aULL,
+	0xf100f1f1f100f1f1ULL, 0xd600d6d6d600d6d6ULL, 0x7a007a7a7a007a7aULL,
+	0xbb00bbbbbb00bbbbULL, 0xe300e3e3e300e3e3ULL, 0x4000404040004040ULL,
+	0x4f004f4f4f004f4fULL,
+};
+
+__visible const u64 camellia_sp44044404[256] = {
+	0x7070007070700070ULL, 0x2c2c002c2c2c002cULL, 0xb3b300b3b3b300b3ULL,
+	0xc0c000c0c0c000c0ULL, 0xe4e400e4e4e400e4ULL, 0x5757005757570057ULL,
+	0xeaea00eaeaea00eaULL, 0xaeae00aeaeae00aeULL, 0x2323002323230023ULL,
+	0x6b6b006b6b6b006bULL, 0x4545004545450045ULL, 0xa5a500a5a5a500a5ULL,
+	0xeded00ededed00edULL, 0x4f4f004f4f4f004fULL, 0x1d1d001d1d1d001dULL,
+	0x9292009292920092ULL, 0x8686008686860086ULL, 0xafaf00afafaf00afULL,
+	0x7c7c007c7c7c007cULL, 0x1f1f001f1f1f001fULL, 0x3e3e003e3e3e003eULL,
+	0xdcdc00dcdcdc00dcULL, 0x5e5e005e5e5e005eULL, 0x0b0b000b0b0b000bULL,
+	0xa6a600a6a6a600a6ULL, 0x3939003939390039ULL, 0xd5d500d5d5d500d5ULL,
+	0x5d5d005d5d5d005dULL, 0xd9d900d9d9d900d9ULL, 0x5a5a005a5a5a005aULL,
+	0x5151005151510051ULL, 0x6c6c006c6c6c006cULL, 0x8b8b008b8b8b008bULL,
+	0x9a9a009a9a9a009aULL, 0xfbfb00fbfbfb00fbULL, 0xb0b000b0b0b000b0ULL,
+	0x7474007474740074ULL, 0x2b2b002b2b2b002bULL, 0xf0f000f0f0f000f0ULL,
+	0x8484008484840084ULL, 0xdfdf00dfdfdf00dfULL, 0xcbcb00cbcbcb00cbULL,
+	0x3434003434340034ULL, 0x7676007676760076ULL, 0x6d6d006d6d6d006dULL,
+	0xa9a900a9a9a900a9ULL, 0xd1d100d1d1d100d1ULL, 0x0404000404040004ULL,
+	0x1414001414140014ULL, 0x3a3a003a3a3a003aULL, 0xdede00dedede00deULL,
+	0x1111001111110011ULL, 0x3232003232320032ULL, 0x9c9c009c9c9c009cULL,
+	0x5353005353530053ULL, 0xf2f200f2f2f200f2ULL, 0xfefe00fefefe00feULL,
+	0xcfcf00cfcfcf00cfULL, 0xc3c300c3c3c300c3ULL, 0x7a7a007a7a7a007aULL,
+	0x2424002424240024ULL, 0xe8e800e8e8e800e8ULL, 0x6060006060600060ULL,
+	0x6969006969690069ULL, 0xaaaa00aaaaaa00aaULL, 0xa0a000a0a0a000a0ULL,
+	0xa1a100a1a1a100a1ULL, 0x6262006262620062ULL, 0x5454005454540054ULL,
+	0x1e1e001e1e1e001eULL, 0xe0e000e0e0e000e0ULL, 0x6464006464640064ULL,
+	0x1010001010100010ULL, 0x0000000000000000ULL, 0xa3a300a3a3a300a3ULL,
+	0x7575007575750075ULL, 0x8a8a008a8a8a008aULL, 0xe6e600e6e6e600e6ULL,
+	0x0909000909090009ULL, 0xdddd00dddddd00ddULL, 0x8787008787870087ULL,
+	0x8383008383830083ULL, 0xcdcd00cdcdcd00cdULL, 0x9090009090900090ULL,
+	0x7373007373730073ULL, 0xf6f600f6f6f600f6ULL, 0x9d9d009d9d9d009dULL,
+	0xbfbf00bfbfbf00bfULL, 0x5252005252520052ULL, 0xd8d800d8d8d800d8ULL,
+	0xc8c800c8c8c800c8ULL, 0xc6c600c6c6c600c6ULL, 0x8181008181810081ULL,
+	0x6f6f006f6f6f006fULL, 0x1313001313130013ULL, 0x6363006363630063ULL,
+	0xe9e900e9e9e900e9ULL, 0xa7a700a7a7a700a7ULL, 0x9f9f009f9f9f009fULL,
+	0xbcbc00bcbcbc00bcULL, 0x2929002929290029ULL, 0xf9f900f9f9f900f9ULL,
+	0x2f2f002f2f2f002fULL, 0xb4b400b4b4b400b4ULL, 0x7878007878780078ULL,
+	0x0606000606060006ULL, 0xe7e700e7e7e700e7ULL, 0x7171007171710071ULL,
+	0xd4d400d4d4d400d4ULL, 0xabab00ababab00abULL, 0x8888008888880088ULL,
+	0x8d8d008d8d8d008dULL, 0x7272007272720072ULL, 0xb9b900b9b9b900b9ULL,
+	0xf8f800f8f8f800f8ULL, 0xacac00acacac00acULL, 0x3636003636360036ULL,
+	0x2a2a002a2a2a002aULL, 0x3c3c003c3c3c003cULL, 0xf1f100f1f1f100f1ULL,
+	0x4040004040400040ULL, 0xd3d300d3d3d300d3ULL, 0xbbbb00bbbbbb00bbULL,
+	0x4343004343430043ULL, 0x1515001515150015ULL, 0xadad00adadad00adULL,
+	0x7777007777770077ULL, 0x8080008080800080ULL, 0x8282008282820082ULL,
+	0xecec00ececec00ecULL, 0x2727002727270027ULL, 0xe5e500e5e5e500e5ULL,
+	0x8585008585850085ULL, 0x3535003535350035ULL, 0x0c0c000c0c0c000cULL,
+	0x4141004141410041ULL, 0xefef00efefef00efULL, 0x9393009393930093ULL,
+	0x1919001919190019ULL, 0x2121002121210021ULL, 0x0e0e000e0e0e000eULL,
+	0x4e4e004e4e4e004eULL, 0x6565006565650065ULL, 0xbdbd00bdbdbd00bdULL,
+	0xb8b800b8b8b800b8ULL, 0x8f8f008f8f8f008fULL, 0xebeb00ebebeb00ebULL,
+	0xcece00cecece00ceULL, 0x3030003030300030ULL, 0x5f5f005f5f5f005fULL,
+	0xc5c500c5c5c500c5ULL, 0x1a1a001a1a1a001aULL, 0xe1e100e1e1e100e1ULL,
+	0xcaca00cacaca00caULL, 0x4747004747470047ULL, 0x3d3d003d3d3d003dULL,
+	0x0101000101010001ULL, 0xd6d600d6d6d600d6ULL, 0x5656005656560056ULL,
+	0x4d4d004d4d4d004dULL, 0x0d0d000d0d0d000dULL, 0x6666006666660066ULL,
+	0xcccc00cccccc00ccULL, 0x2d2d002d2d2d002dULL, 0x1212001212120012ULL,
+	0x2020002020200020ULL, 0xb1b100b1b1b100b1ULL, 0x9999009999990099ULL,
+	0x4c4c004c4c4c004cULL, 0xc2c200c2c2c200c2ULL, 0x7e7e007e7e7e007eULL,
+	0x0505000505050005ULL, 0xb7b700b7b7b700b7ULL, 0x3131003131310031ULL,
+	0x1717001717170017ULL, 0xd7d700d7d7d700d7ULL, 0x5858005858580058ULL,
+	0x6161006161610061ULL, 0x1b1b001b1b1b001bULL, 0x1c1c001c1c1c001cULL,
+	0x0f0f000f0f0f000fULL, 0x1616001616160016ULL, 0x1818001818180018ULL,
+	0x2222002222220022ULL, 0x4444004444440044ULL, 0xb2b200b2b2b200b2ULL,
+	0xb5b500b5b5b500b5ULL, 0x9191009191910091ULL, 0x0808000808080008ULL,
+	0xa8a800a8a8a800a8ULL, 0xfcfc00fcfcfc00fcULL, 0x5050005050500050ULL,
+	0xd0d000d0d0d000d0ULL, 0x7d7d007d7d7d007dULL, 0x8989008989890089ULL,
+	0x9797009797970097ULL, 0x5b5b005b5b5b005bULL, 0x9595009595950095ULL,
+	0xffff00ffffff00ffULL, 0xd2d200d2d2d200d2ULL, 0xc4c400c4c4c400c4ULL,
+	0x4848004848480048ULL, 0xf7f700f7f7f700f7ULL, 0xdbdb00dbdbdb00dbULL,
+	0x0303000303030003ULL, 0xdada00dadada00daULL, 0x3f3f003f3f3f003fULL,
+	0x9494009494940094ULL, 0x5c5c005c5c5c005cULL, 0x0202000202020002ULL,
+	0x4a4a004a4a4a004aULL, 0x3333003333330033ULL, 0x6767006767670067ULL,
+	0xf3f300f3f3f300f3ULL, 0x7f7f007f7f7f007fULL, 0xe2e200e2e2e200e2ULL,
+	0x9b9b009b9b9b009bULL, 0x2626002626260026ULL, 0x3737003737370037ULL,
+	0x3b3b003b3b3b003bULL, 0x9696009696960096ULL, 0x4b4b004b4b4b004bULL,
+	0xbebe00bebebe00beULL, 0x2e2e002e2e2e002eULL, 0x7979007979790079ULL,
+	0x8c8c008c8c8c008cULL, 0x6e6e006e6e6e006eULL, 0x8e8e008e8e8e008eULL,
+	0xf5f500f5f5f500f5ULL, 0xb6b600b6b6b600b6ULL, 0xfdfd00fdfdfd00fdULL,
+	0x5959005959590059ULL, 0x9898009898980098ULL, 0x6a6a006a6a6a006aULL,
+	0x4646004646460046ULL, 0xbaba00bababa00baULL, 0x2525002525250025ULL,
+	0x4242004242420042ULL, 0xa2a200a2a2a200a2ULL, 0xfafa00fafafa00faULL,
+	0x0707000707070007ULL, 0x5555005555550055ULL, 0xeeee00eeeeee00eeULL,
+	0x0a0a000a0a0a000aULL, 0x4949004949490049ULL, 0x6868006868680068ULL,
+	0x3838003838380038ULL, 0xa4a400a4a4a400a4ULL, 0x2828002828280028ULL,
+	0x7b7b007b7b7b007bULL, 0xc9c900c9c9c900c9ULL, 0xc1c100c1c1c100c1ULL,
+	0xe3e300e3e3e300e3ULL, 0xf4f400f4f4f400f4ULL, 0xc7c700c7c7c700c7ULL,
+	0x9e9e009e9e9e009eULL,
+};
+
+__visible const u64 camellia_sp11101110[256] = {
+	0x7070700070707000ULL, 0x8282820082828200ULL, 0x2c2c2c002c2c2c00ULL,
+	0xececec00ececec00ULL, 0xb3b3b300b3b3b300ULL, 0x2727270027272700ULL,
+	0xc0c0c000c0c0c000ULL, 0xe5e5e500e5e5e500ULL, 0xe4e4e400e4e4e400ULL,
+	0x8585850085858500ULL, 0x5757570057575700ULL, 0x3535350035353500ULL,
+	0xeaeaea00eaeaea00ULL, 0x0c0c0c000c0c0c00ULL, 0xaeaeae00aeaeae00ULL,
+	0x4141410041414100ULL, 0x2323230023232300ULL, 0xefefef00efefef00ULL,
+	0x6b6b6b006b6b6b00ULL, 0x9393930093939300ULL, 0x4545450045454500ULL,
+	0x1919190019191900ULL, 0xa5a5a500a5a5a500ULL, 0x2121210021212100ULL,
+	0xededed00ededed00ULL, 0x0e0e0e000e0e0e00ULL, 0x4f4f4f004f4f4f00ULL,
+	0x4e4e4e004e4e4e00ULL, 0x1d1d1d001d1d1d00ULL, 0x6565650065656500ULL,
+	0x9292920092929200ULL, 0xbdbdbd00bdbdbd00ULL, 0x8686860086868600ULL,
+	0xb8b8b800b8b8b800ULL, 0xafafaf00afafaf00ULL, 0x8f8f8f008f8f8f00ULL,
+	0x7c7c7c007c7c7c00ULL, 0xebebeb00ebebeb00ULL, 0x1f1f1f001f1f1f00ULL,
+	0xcecece00cecece00ULL, 0x3e3e3e003e3e3e00ULL, 0x3030300030303000ULL,
+	0xdcdcdc00dcdcdc00ULL, 0x5f5f5f005f5f5f00ULL, 0x5e5e5e005e5e5e00ULL,
+	0xc5c5c500c5c5c500ULL, 0x0b0b0b000b0b0b00ULL, 0x1a1a1a001a1a1a00ULL,
+	0xa6a6a600a6a6a600ULL, 0xe1e1e100e1e1e100ULL, 0x3939390039393900ULL,
+	0xcacaca00cacaca00ULL, 0xd5d5d500d5d5d500ULL, 0x4747470047474700ULL,
+	0x5d5d5d005d5d5d00ULL, 0x3d3d3d003d3d3d00ULL, 0xd9d9d900d9d9d900ULL,
+	0x0101010001010100ULL, 0x5a5a5a005a5a5a00ULL, 0xd6d6d600d6d6d600ULL,
+	0x5151510051515100ULL, 0x5656560056565600ULL, 0x6c6c6c006c6c6c00ULL,
+	0x4d4d4d004d4d4d00ULL, 0x8b8b8b008b8b8b00ULL, 0x0d0d0d000d0d0d00ULL,
+	0x9a9a9a009a9a9a00ULL, 0x6666660066666600ULL, 0xfbfbfb00fbfbfb00ULL,
+	0xcccccc00cccccc00ULL, 0xb0b0b000b0b0b000ULL, 0x2d2d2d002d2d2d00ULL,
+	0x7474740074747400ULL, 0x1212120012121200ULL, 0x2b2b2b002b2b2b00ULL,
+	0x2020200020202000ULL, 0xf0f0f000f0f0f000ULL, 0xb1b1b100b1b1b100ULL,
+	0x8484840084848400ULL, 0x9999990099999900ULL, 0xdfdfdf00dfdfdf00ULL,
+	0x4c4c4c004c4c4c00ULL, 0xcbcbcb00cbcbcb00ULL, 0xc2c2c200c2c2c200ULL,
+	0x3434340034343400ULL, 0x7e7e7e007e7e7e00ULL, 0x7676760076767600ULL,
+	0x0505050005050500ULL, 0x6d6d6d006d6d6d00ULL, 0xb7b7b700b7b7b700ULL,
+	0xa9a9a900a9a9a900ULL, 0x3131310031313100ULL, 0xd1d1d100d1d1d100ULL,
+	0x1717170017171700ULL, 0x0404040004040400ULL, 0xd7d7d700d7d7d700ULL,
+	0x1414140014141400ULL, 0x5858580058585800ULL, 0x3a3a3a003a3a3a00ULL,
+	0x6161610061616100ULL, 0xdedede00dedede00ULL, 0x1b1b1b001b1b1b00ULL,
+	0x1111110011111100ULL, 0x1c1c1c001c1c1c00ULL, 0x3232320032323200ULL,
+	0x0f0f0f000f0f0f00ULL, 0x9c9c9c009c9c9c00ULL, 0x1616160016161600ULL,
+	0x5353530053535300ULL, 0x1818180018181800ULL, 0xf2f2f200f2f2f200ULL,
+	0x2222220022222200ULL, 0xfefefe00fefefe00ULL, 0x4444440044444400ULL,
+	0xcfcfcf00cfcfcf00ULL, 0xb2b2b200b2b2b200ULL, 0xc3c3c300c3c3c300ULL,
+	0xb5b5b500b5b5b500ULL, 0x7a7a7a007a7a7a00ULL, 0x9191910091919100ULL,
+	0x2424240024242400ULL, 0x0808080008080800ULL, 0xe8e8e800e8e8e800ULL,
+	0xa8a8a800a8a8a800ULL, 0x6060600060606000ULL, 0xfcfcfc00fcfcfc00ULL,
+	0x6969690069696900ULL, 0x5050500050505000ULL, 0xaaaaaa00aaaaaa00ULL,
+	0xd0d0d000d0d0d000ULL, 0xa0a0a000a0a0a000ULL, 0x7d7d7d007d7d7d00ULL,
+	0xa1a1a100a1a1a100ULL, 0x8989890089898900ULL, 0x6262620062626200ULL,
+	0x9797970097979700ULL, 0x5454540054545400ULL, 0x5b5b5b005b5b5b00ULL,
+	0x1e1e1e001e1e1e00ULL, 0x9595950095959500ULL, 0xe0e0e000e0e0e000ULL,
+	0xffffff00ffffff00ULL, 0x6464640064646400ULL, 0xd2d2d200d2d2d200ULL,
+	0x1010100010101000ULL, 0xc4c4c400c4c4c400ULL, 0x0000000000000000ULL,
+	0x4848480048484800ULL, 0xa3a3a300a3a3a300ULL, 0xf7f7f700f7f7f700ULL,
+	0x7575750075757500ULL, 0xdbdbdb00dbdbdb00ULL, 0x8a8a8a008a8a8a00ULL,
+	0x0303030003030300ULL, 0xe6e6e600e6e6e600ULL, 0xdadada00dadada00ULL,
+	0x0909090009090900ULL, 0x3f3f3f003f3f3f00ULL, 0xdddddd00dddddd00ULL,
+	0x9494940094949400ULL, 0x8787870087878700ULL, 0x5c5c5c005c5c5c00ULL,
+	0x8383830083838300ULL, 0x0202020002020200ULL, 0xcdcdcd00cdcdcd00ULL,
+	0x4a4a4a004a4a4a00ULL, 0x9090900090909000ULL, 0x3333330033333300ULL,
+	0x7373730073737300ULL, 0x6767670067676700ULL, 0xf6f6f600f6f6f600ULL,
+	0xf3f3f300f3f3f300ULL, 0x9d9d9d009d9d9d00ULL, 0x7f7f7f007f7f7f00ULL,
+	0xbfbfbf00bfbfbf00ULL, 0xe2e2e200e2e2e200ULL, 0x5252520052525200ULL,
+	0x9b9b9b009b9b9b00ULL, 0xd8d8d800d8d8d800ULL, 0x2626260026262600ULL,
+	0xc8c8c800c8c8c800ULL, 0x3737370037373700ULL, 0xc6c6c600c6c6c600ULL,
+	0x3b3b3b003b3b3b00ULL, 0x8181810081818100ULL, 0x9696960096969600ULL,
+	0x6f6f6f006f6f6f00ULL, 0x4b4b4b004b4b4b00ULL, 0x1313130013131300ULL,
+	0xbebebe00bebebe00ULL, 0x6363630063636300ULL, 0x2e2e2e002e2e2e00ULL,
+	0xe9e9e900e9e9e900ULL, 0x7979790079797900ULL, 0xa7a7a700a7a7a700ULL,
+	0x8c8c8c008c8c8c00ULL, 0x9f9f9f009f9f9f00ULL, 0x6e6e6e006e6e6e00ULL,
+	0xbcbcbc00bcbcbc00ULL, 0x8e8e8e008e8e8e00ULL, 0x2929290029292900ULL,
+	0xf5f5f500f5f5f500ULL, 0xf9f9f900f9f9f900ULL, 0xb6b6b600b6b6b600ULL,
+	0x2f2f2f002f2f2f00ULL, 0xfdfdfd00fdfdfd00ULL, 0xb4b4b400b4b4b400ULL,
+	0x5959590059595900ULL, 0x7878780078787800ULL, 0x9898980098989800ULL,
+	0x0606060006060600ULL, 0x6a6a6a006a6a6a00ULL, 0xe7e7e700e7e7e700ULL,
+	0x4646460046464600ULL, 0x7171710071717100ULL, 0xbababa00bababa00ULL,
+	0xd4d4d400d4d4d400ULL, 0x2525250025252500ULL, 0xababab00ababab00ULL,
+	0x4242420042424200ULL, 0x8888880088888800ULL, 0xa2a2a200a2a2a200ULL,
+	0x8d8d8d008d8d8d00ULL, 0xfafafa00fafafa00ULL, 0x7272720072727200ULL,
+	0x0707070007070700ULL, 0xb9b9b900b9b9b900ULL, 0x5555550055555500ULL,
+	0xf8f8f800f8f8f800ULL, 0xeeeeee00eeeeee00ULL, 0xacacac00acacac00ULL,
+	0x0a0a0a000a0a0a00ULL, 0x3636360036363600ULL, 0x4949490049494900ULL,
+	0x2a2a2a002a2a2a00ULL, 0x6868680068686800ULL, 0x3c3c3c003c3c3c00ULL,
+	0x3838380038383800ULL, 0xf1f1f100f1f1f100ULL, 0xa4a4a400a4a4a400ULL,
+	0x4040400040404000ULL, 0x2828280028282800ULL, 0xd3d3d300d3d3d300ULL,
+	0x7b7b7b007b7b7b00ULL, 0xbbbbbb00bbbbbb00ULL, 0xc9c9c900c9c9c900ULL,
+	0x4343430043434300ULL, 0xc1c1c100c1c1c100ULL, 0x1515150015151500ULL,
+	0xe3e3e300e3e3e300ULL, 0xadadad00adadad00ULL, 0xf4f4f400f4f4f400ULL,
+	0x7777770077777700ULL, 0xc7c7c700c7c7c700ULL, 0x8080800080808000ULL,
+	0x9e9e9e009e9e9e00ULL,
+};
+
+/* key constants */
+#define CAMELLIA_SIGMA1L (0xA09E667FL)
+#define CAMELLIA_SIGMA1R (0x3BCC908BL)
+#define CAMELLIA_SIGMA2L (0xB67AE858L)
+#define CAMELLIA_SIGMA2R (0x4CAA73B2L)
+#define CAMELLIA_SIGMA3L (0xC6EF372FL)
+#define CAMELLIA_SIGMA3R (0xE94F82BEL)
+#define CAMELLIA_SIGMA4L (0x54FF53A5L)
+#define CAMELLIA_SIGMA4R (0xF1D36F1CL)
+#define CAMELLIA_SIGMA5L (0x10E527FAL)
+#define CAMELLIA_SIGMA5R (0xDE682D1DL)
+#define CAMELLIA_SIGMA6L (0xB05688C2L)
+#define CAMELLIA_SIGMA6R (0xB3E6C1FDL)
+
+/* macros */
+#define ROLDQ(l, r, bits) ({ \
+	u64 t = l;					\
+	l = (l << bits) | (r >> (64 - bits));		\
+	r = (r << bits) | (t >> (64 - bits));		\
+})
+
+#define CAMELLIA_F(x, kl, kr, y) ({ \
+	u64 ii = x ^ (((u64)kl << 32) | kr);				\
+	y = camellia_sp11101110[(uint8_t)ii];				\
+	y ^= camellia_sp44044404[(uint8_t)(ii >> 8)];			\
+	ii >>= 16;							\
+	y ^= camellia_sp30333033[(uint8_t)ii];				\
+	y ^= camellia_sp02220222[(uint8_t)(ii >> 8)];			\
+	ii >>= 16;							\
+	y ^= camellia_sp00444404[(uint8_t)ii];				\
+	y ^= camellia_sp03303033[(uint8_t)(ii >> 8)];			\
+	ii >>= 16;							\
+	y ^= camellia_sp22000222[(uint8_t)ii];				\
+	y ^= camellia_sp10011110[(uint8_t)(ii >> 8)];			\
+	y = ror64(y, 32);						\
+})
+
+#define SET_SUBKEY_LR(INDEX, sRL) (subkey[(INDEX)] = ror64((sRL), 32))
+
+static void camellia_setup_tail(u64 *subkey, u64 *subRL, int max)
+{
+	u64 kw4, tt;
+	u32 dw, tl, tr;
+
+	/* absorb kw2 to other subkeys */
+	/* round 2 */
+	subRL[3] ^= subRL[1];
+	/* round 4 */
+	subRL[5] ^= subRL[1];
+	/* round 6 */
+	subRL[7] ^= subRL[1];
+
+	subRL[1] ^= (subRL[1] & ~subRL[9]) << 32;
+	/* modified for FLinv(kl2) */
+	dw = (subRL[1] & subRL[9]) >> 32;
+	subRL[1] ^= rol32(dw, 1);
+
+	/* round 8 */
+	subRL[11] ^= subRL[1];
+	/* round 10 */
+	subRL[13] ^= subRL[1];
+	/* round 12 */
+	subRL[15] ^= subRL[1];
+
+	subRL[1] ^= (subRL[1] & ~subRL[17]) << 32;
+	/* modified for FLinv(kl4) */
+	dw = (subRL[1] & subRL[17]) >> 32;
+	subRL[1] ^= rol32(dw, 1);
+
+	/* round 14 */
+	subRL[19] ^= subRL[1];
+	/* round 16 */
+	subRL[21] ^= subRL[1];
+	/* round 18 */
+	subRL[23] ^= subRL[1];
+
+	if (max == 24) {
+		/* kw3 */
+		subRL[24] ^= subRL[1];
+
+		/* absorb kw4 to other subkeys */
+		kw4 = subRL[25];
+	} else {
+		subRL[1] ^= (subRL[1] & ~subRL[25]) << 32;
+		/* modified for FLinv(kl6) */
+		dw = (subRL[1] & subRL[25]) >> 32;
+		subRL[1] ^= rol32(dw, 1);
+
+		/* round 20 */
+		subRL[27] ^= subRL[1];
+		/* round 22 */
+		subRL[29] ^= subRL[1];
+		/* round 24 */
+		subRL[31] ^= subRL[1];
+		/* kw3 */
+		subRL[32] ^= subRL[1];
+
+		/* absorb kw4 to other subkeys */
+		kw4 = subRL[33];
+		/* round 23 */
+		subRL[30] ^= kw4;
+		/* round 21 */
+		subRL[28] ^= kw4;
+		/* round 19 */
+		subRL[26] ^= kw4;
+
+		kw4 ^= (kw4 & ~subRL[24]) << 32;
+		/* modified for FL(kl5) */
+		dw = (kw4 & subRL[24]) >> 32;
+		kw4 ^= rol32(dw, 1);
+	}
+
+	/* round 17 */
+	subRL[22] ^= kw4;
+	/* round 15 */
+	subRL[20] ^= kw4;
+	/* round 13 */
+	subRL[18] ^= kw4;
+
+	kw4 ^= (kw4 & ~subRL[16]) << 32;
+	/* modified for FL(kl3) */
+	dw = (kw4 & subRL[16]) >> 32;
+	kw4 ^= rol32(dw, 1);
+
+	/* round 11 */
+	subRL[14] ^= kw4;
+	/* round 9 */
+	subRL[12] ^= kw4;
+	/* round 7 */
+	subRL[10] ^= kw4;
+
+	kw4 ^= (kw4 & ~subRL[8]) << 32;
+	/* modified for FL(kl1) */
+	dw = (kw4 & subRL[8]) >> 32;
+	kw4 ^= rol32(dw, 1);
+
+	/* round 5 */
+	subRL[6] ^= kw4;
+	/* round 3 */
+	subRL[4] ^= kw4;
+	/* round 1 */
+	subRL[2] ^= kw4;
+	/* kw1 */
+	subRL[0] ^= kw4;
+
+	/* key XOR is end of F-function */
+	SET_SUBKEY_LR(0, subRL[0] ^ subRL[2]);			/* kw1 */
+	SET_SUBKEY_LR(2, subRL[3]);				/* round 1 */
+	SET_SUBKEY_LR(3, subRL[2] ^ subRL[4]);			/* round 2 */
+	SET_SUBKEY_LR(4, subRL[3] ^ subRL[5]);			/* round 3 */
+	SET_SUBKEY_LR(5, subRL[4] ^ subRL[6]);			/* round 4 */
+	SET_SUBKEY_LR(6, subRL[5] ^ subRL[7]);			/* round 5 */
+
+	tl = (subRL[10] >> 32) ^ (subRL[10] & ~subRL[8]);
+	dw = tl & (subRL[8] >> 32);				/* FL(kl1) */
+	tr = subRL[10] ^ rol32(dw, 1);
+	tt = (tr | ((u64)tl << 32));
+
+	SET_SUBKEY_LR(7, subRL[6] ^ tt);			/* round 6 */
+	SET_SUBKEY_LR(8, subRL[8]);				/* FL(kl1) */
+	SET_SUBKEY_LR(9, subRL[9]);				/* FLinv(kl2) */
+
+	tl = (subRL[7] >> 32) ^ (subRL[7] & ~subRL[9]);
+	dw = tl & (subRL[9] >> 32);				/* FLinv(kl2) */
+	tr = subRL[7] ^ rol32(dw, 1);
+	tt = (tr | ((u64)tl << 32));
+
+	SET_SUBKEY_LR(10, subRL[11] ^ tt);			/* round 7 */
+	SET_SUBKEY_LR(11, subRL[10] ^ subRL[12]);		/* round 8 */
+	SET_SUBKEY_LR(12, subRL[11] ^ subRL[13]);		/* round 9 */
+	SET_SUBKEY_LR(13, subRL[12] ^ subRL[14]);		/* round 10 */
+	SET_SUBKEY_LR(14, subRL[13] ^ subRL[15]);		/* round 11 */
+
+	tl = (subRL[18] >> 32) ^ (subRL[18] & ~subRL[16]);
+	dw = tl & (subRL[16] >> 32);				/* FL(kl3) */
+	tr = subRL[18] ^ rol32(dw, 1);
+	tt = (tr | ((u64)tl << 32));
+
+	SET_SUBKEY_LR(15, subRL[14] ^ tt);			/* round 12 */
+	SET_SUBKEY_LR(16, subRL[16]);				/* FL(kl3) */
+	SET_SUBKEY_LR(17, subRL[17]);				/* FLinv(kl4) */
+
+	tl = (subRL[15] >> 32) ^ (subRL[15] & ~subRL[17]);
+	dw = tl & (subRL[17] >> 32);				/* FLinv(kl4) */
+	tr = subRL[15] ^ rol32(dw, 1);
+	tt = (tr | ((u64)tl << 32));
+
+	SET_SUBKEY_LR(18, subRL[19] ^ tt);			/* round 13 */
+	SET_SUBKEY_LR(19, subRL[18] ^ subRL[20]);		/* round 14 */
+	SET_SUBKEY_LR(20, subRL[19] ^ subRL[21]);		/* round 15 */
+	SET_SUBKEY_LR(21, subRL[20] ^ subRL[22]);		/* round 16 */
+	SET_SUBKEY_LR(22, subRL[21] ^ subRL[23]);		/* round 17 */
+
+	if (max == 24) {
+		SET_SUBKEY_LR(23, subRL[22]);			/* round 18 */
+		SET_SUBKEY_LR(24, subRL[24] ^ subRL[23]);	/* kw3 */
+	} else {
+		tl = (subRL[26] >> 32) ^ (subRL[26] & ~subRL[24]);
+		dw = tl & (subRL[24] >> 32);			/* FL(kl5) */
+		tr = subRL[26] ^ rol32(dw, 1);
+		tt = (tr | ((u64)tl << 32));
+
+		SET_SUBKEY_LR(23, subRL[22] ^ tt);		/* round 18 */
+		SET_SUBKEY_LR(24, subRL[24]);			/* FL(kl5) */
+		SET_SUBKEY_LR(25, subRL[25]);			/* FLinv(kl6) */
+
+		tl = (subRL[23] >> 32) ^ (subRL[23] & ~subRL[25]);
+		dw = tl & (subRL[25] >> 32);			/* FLinv(kl6) */
+		tr = subRL[23] ^ rol32(dw, 1);
+		tt = (tr | ((u64)tl << 32));
+
+		SET_SUBKEY_LR(26, subRL[27] ^ tt);		/* round 19 */
+		SET_SUBKEY_LR(27, subRL[26] ^ subRL[28]);	/* round 20 */
+		SET_SUBKEY_LR(28, subRL[27] ^ subRL[29]);	/* round 21 */
+		SET_SUBKEY_LR(29, subRL[28] ^ subRL[30]);	/* round 22 */
+		SET_SUBKEY_LR(30, subRL[29] ^ subRL[31]);	/* round 23 */
+		SET_SUBKEY_LR(31, subRL[30]);			/* round 24 */
+		SET_SUBKEY_LR(32, subRL[32] ^ subRL[31]);	/* kw3 */
+	}
+}
+
+static void camellia_setup128(const unsigned char *key, u64 *subkey)
+{
+	u64 kl, kr, ww;
+	u64 subRL[26];
+
+	/**
+	 *  k == kl || kr (|| is concatenation)
+	 */
+	kl = get_unaligned_be64(key);
+	kr = get_unaligned_be64(key + 8);
+
+	/* generate KL dependent subkeys */
+	/* kw1 */
+	subRL[0] = kl;
+	/* kw2 */
+	subRL[1] = kr;
+
+	/* rotation left shift 15bit */
+	ROLDQ(kl, kr, 15);
+
+	/* k3 */
+	subRL[4] = kl;
+	/* k4 */
+	subRL[5] = kr;
+
+	/* rotation left shift 15+30bit */
+	ROLDQ(kl, kr, 30);
+
+	/* k7 */
+	subRL[10] = kl;
+	/* k8 */
+	subRL[11] = kr;
+
+	/* rotation left shift 15+30+15bit */
+	ROLDQ(kl, kr, 15);
+
+	/* k10 */
+	subRL[13] = kr;
+	/* rotation left shift 15+30+15+17 bit */
+	ROLDQ(kl, kr, 17);
+
+	/* kl3 */
+	subRL[16] = kl;
+	/* kl4 */
+	subRL[17] = kr;
+
+	/* rotation left shift 15+30+15+17+17 bit */
+	ROLDQ(kl, kr, 17);
+
+	/* k13 */
+	subRL[18] = kl;
+	/* k14 */
+	subRL[19] = kr;
+
+	/* rotation left shift 15+30+15+17+17+17 bit */
+	ROLDQ(kl, kr, 17);
+
+	/* k17 */
+	subRL[22] = kl;
+	/* k18 */
+	subRL[23] = kr;
+
+	/* generate KA */
+	kl = subRL[0];
+	kr = subRL[1];
+	CAMELLIA_F(kl, CAMELLIA_SIGMA1L, CAMELLIA_SIGMA1R, ww);
+	kr ^= ww;
+	CAMELLIA_F(kr, CAMELLIA_SIGMA2L, CAMELLIA_SIGMA2R, kl);
+
+	/* current status == (kll, klr, w0, w1) */
+	CAMELLIA_F(kl, CAMELLIA_SIGMA3L, CAMELLIA_SIGMA3R, kr);
+	kr ^= ww;
+	CAMELLIA_F(kr, CAMELLIA_SIGMA4L, CAMELLIA_SIGMA4R, ww);
+	kl ^= ww;
+
+	/* generate KA dependent subkeys */
+	/* k1, k2 */
+	subRL[2] = kl;
+	subRL[3] = kr;
+	ROLDQ(kl, kr, 15);
+	/* k5,k6 */
+	subRL[6] = kl;
+	subRL[7] = kr;
+	ROLDQ(kl, kr, 15);
+	/* kl1, kl2 */
+	subRL[8] = kl;
+	subRL[9] = kr;
+	ROLDQ(kl, kr, 15);
+	/* k9 */
+	subRL[12] = kl;
+	ROLDQ(kl, kr, 15);
+	/* k11, k12 */
+	subRL[14] = kl;
+	subRL[15] = kr;
+	ROLDQ(kl, kr, 34);
+	/* k15, k16 */
+	subRL[20] = kl;
+	subRL[21] = kr;
+	ROLDQ(kl, kr, 17);
+	/* kw3, kw4 */
+	subRL[24] = kl;
+	subRL[25] = kr;
+
+	camellia_setup_tail(subkey, subRL, 24);
+}
+
+static void camellia_setup256(const unsigned char *key, u64 *subkey)
+{
+	u64 kl, kr;			/* left half of key */
+	u64 krl, krr;			/* right half of key */
+	u64 ww;				/* temporary variables */
+	u64 subRL[34];
+
+	/**
+	 *  key = (kl || kr || krl || krr) (|| is concatenation)
+	 */
+	kl = get_unaligned_be64(key);
+	kr = get_unaligned_be64(key + 8);
+	krl = get_unaligned_be64(key + 16);
+	krr = get_unaligned_be64(key + 24);
+
+	/* generate KL dependent subkeys */
+	/* kw1 */
+	subRL[0] = kl;
+	/* kw2 */
+	subRL[1] = kr;
+	ROLDQ(kl, kr, 45);
+	/* k9 */
+	subRL[12] = kl;
+	/* k10 */
+	subRL[13] = kr;
+	ROLDQ(kl, kr, 15);
+	/* kl3 */
+	subRL[16] = kl;
+	/* kl4 */
+	subRL[17] = kr;
+	ROLDQ(kl, kr, 17);
+	/* k17 */
+	subRL[22] = kl;
+	/* k18 */
+	subRL[23] = kr;
+	ROLDQ(kl, kr, 34);
+	/* k23 */
+	subRL[30] = kl;
+	/* k24 */
+	subRL[31] = kr;
+
+	/* generate KR dependent subkeys */
+	ROLDQ(krl, krr, 15);
+	/* k3 */
+	subRL[4] = krl;
+	/* k4 */
+	subRL[5] = krr;
+	ROLDQ(krl, krr, 15);
+	/* kl1 */
+	subRL[8] = krl;
+	/* kl2 */
+	subRL[9] = krr;
+	ROLDQ(krl, krr, 30);
+	/* k13 */
+	subRL[18] = krl;
+	/* k14 */
+	subRL[19] = krr;
+	ROLDQ(krl, krr, 34);
+	/* k19 */
+	subRL[26] = krl;
+	/* k20 */
+	subRL[27] = krr;
+	ROLDQ(krl, krr, 34);
+
+	/* generate KA */
+	kl = subRL[0] ^ krl;
+	kr = subRL[1] ^ krr;
+
+	CAMELLIA_F(kl, CAMELLIA_SIGMA1L, CAMELLIA_SIGMA1R, ww);
+	kr ^= ww;
+	CAMELLIA_F(kr, CAMELLIA_SIGMA2L, CAMELLIA_SIGMA2R, kl);
+	kl ^= krl;
+	CAMELLIA_F(kl, CAMELLIA_SIGMA3L, CAMELLIA_SIGMA3R, kr);
+	kr ^= ww ^ krr;
+	CAMELLIA_F(kr, CAMELLIA_SIGMA4L, CAMELLIA_SIGMA4R, ww);
+	kl ^= ww;
+
+	/* generate KB */
+	krl ^= kl;
+	krr ^= kr;
+	CAMELLIA_F(krl, CAMELLIA_SIGMA5L, CAMELLIA_SIGMA5R, ww);
+	krr ^= ww;
+	CAMELLIA_F(krr, CAMELLIA_SIGMA6L, CAMELLIA_SIGMA6R, ww);
+	krl ^= ww;
+
+	/* generate KA dependent subkeys */
+	ROLDQ(kl, kr, 15);
+	/* k5 */
+	subRL[6] = kl;
+	/* k6 */
+	subRL[7] = kr;
+	ROLDQ(kl, kr, 30);
+	/* k11 */
+	subRL[14] = kl;
+	/* k12 */
+	subRL[15] = kr;
+	/* rotation left shift 32bit */
+	ROLDQ(kl, kr, 32);
+	/* kl5 */
+	subRL[24] = kl;
+	/* kl6 */
+	subRL[25] = kr;
+	/* rotation left shift 17 from k11,k12 -> k21,k22 */
+	ROLDQ(kl, kr, 17);
+	/* k21 */
+	subRL[28] = kl;
+	/* k22 */
+	subRL[29] = kr;
+
+	/* generate KB dependent subkeys */
+	/* k1 */
+	subRL[2] = krl;
+	/* k2 */
+	subRL[3] = krr;
+	ROLDQ(krl, krr, 30);
+	/* k7 */
+	subRL[10] = krl;
+	/* k8 */
+	subRL[11] = krr;
+	ROLDQ(krl, krr, 30);
+	/* k15 */
+	subRL[20] = krl;
+	/* k16 */
+	subRL[21] = krr;
+	ROLDQ(krl, krr, 51);
+	/* kw3 */
+	subRL[32] = krl;
+	/* kw4 */
+	subRL[33] = krr;
+
+	camellia_setup_tail(subkey, subRL, 32);
+}
+
+static void camellia_setup192(const unsigned char *key, u64 *subkey)
+{
+	unsigned char kk[32];
+	u64 krl, krr;
+
+	memcpy(kk, key, 24);
+	memcpy((unsigned char *)&krl, key+16, 8);
+	krr = ~krl;
+	memcpy(kk+24, (unsigned char *)&krr, 8);
+	camellia_setup256(kk, subkey);
+}
+
+int __camellia_setkey(struct camellia_ctx *cctx, const unsigned char *key,
+		      unsigned int key_len)
+{
+	if (key_len != 16 && key_len != 24 && key_len != 32)
+		return -EINVAL;
+
+	cctx->key_length = key_len;
+
+	switch (key_len) {
+	case 16:
+		camellia_setup128(key, cctx->key_table);
+		break;
+	case 24:
+		camellia_setup192(key, cctx->key_table);
+		break;
+	case 32:
+		camellia_setup256(key, cctx->key_table);
+		break;
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(__camellia_setkey);
+
+static int camellia_setkey(struct crypto_tfm *tfm, const u8 *key,
+			   unsigned int key_len)
+{
+	return __camellia_setkey(crypto_tfm_ctx(tfm), key, key_len);
+}
+
+static int camellia_setkey_skcipher(struct crypto_skcipher *tfm, const u8 *key,
+				    unsigned int key_len)
+{
+	return camellia_setkey(&tfm->base, key, key_len);
+}
+
+void camellia_decrypt_cbc_2way(const void *ctx, u8 *dst, const u8 *src)
+{
+	u8 buf[CAMELLIA_BLOCK_SIZE];
+	const u8 *iv = src;
+
+	if (dst == src)
+		iv = memcpy(buf, iv, sizeof(buf));
+	camellia_dec_blk_2way(ctx, dst, src);
+	crypto_xor(dst + CAMELLIA_BLOCK_SIZE, iv, CAMELLIA_BLOCK_SIZE);
+}
+EXPORT_SYMBOL_GPL(camellia_decrypt_cbc_2way);
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, CAMELLIA_BLOCK_SIZE, -1);
+	ECB_BLOCK(2, camellia_enc_blk_2way);
+	ECB_BLOCK(1, camellia_enc_blk);
+	ECB_WALK_END();
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, CAMELLIA_BLOCK_SIZE, -1);
+	ECB_BLOCK(2, camellia_dec_blk_2way);
+	ECB_BLOCK(1, camellia_dec_blk);
+	ECB_WALK_END();
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, CAMELLIA_BLOCK_SIZE, -1);
+	CBC_ENC_BLOCK(camellia_enc_blk);
+	CBC_WALK_END();
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, CAMELLIA_BLOCK_SIZE, -1);
+	CBC_DEC_BLOCK(2, camellia_decrypt_cbc_2way);
+	CBC_DEC_BLOCK(1, camellia_dec_blk);
+	CBC_WALK_END();
+}
+
+static struct crypto_alg camellia_cipher_alg = {
+	.cra_name		= "camellia",
+	.cra_driver_name	= "camellia-asm",
+	.cra_priority		= 200,
+	.cra_flags		= CRYPTO_ALG_TYPE_CIPHER,
+	.cra_blocksize		= CAMELLIA_BLOCK_SIZE,
+	.cra_ctxsize		= sizeof(struct camellia_ctx),
+	.cra_module		= THIS_MODULE,
+	.cra_u			= {
+		.cipher = {
+			.cia_min_keysize = CAMELLIA_MIN_KEY_SIZE,
+			.cia_max_keysize = CAMELLIA_MAX_KEY_SIZE,
+			.cia_setkey	 = camellia_setkey,
+			.cia_encrypt	 = camellia_encrypt,
+			.cia_decrypt	 = camellia_decrypt
+		}
+	}
+};
+
+static struct skcipher_alg camellia_skcipher_algs[] = {
+	{
+		.base.cra_name		= "ecb(camellia)",
+		.base.cra_driver_name	= "ecb-camellia-asm",
+		.base.cra_priority	= 300,
+		.base.cra_blocksize	= CAMELLIA_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct camellia_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= CAMELLIA_MIN_KEY_SIZE,
+		.max_keysize		= CAMELLIA_MAX_KEY_SIZE,
+		.setkey			= camellia_setkey_skcipher,
+		.encrypt		= ecb_encrypt,
+		.decrypt		= ecb_decrypt,
+	}, {
+		.base.cra_name		= "cbc(camellia)",
+		.base.cra_driver_name	= "cbc-camellia-asm",
+		.base.cra_priority	= 300,
+		.base.cra_blocksize	= CAMELLIA_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct camellia_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= CAMELLIA_MIN_KEY_SIZE,
+		.max_keysize		= CAMELLIA_MAX_KEY_SIZE,
+		.ivsize			= CAMELLIA_BLOCK_SIZE,
+		.setkey			= camellia_setkey_skcipher,
+		.encrypt		= cbc_encrypt,
+		.decrypt		= cbc_decrypt,
+	}
+};
+
+static bool is_blacklisted_cpu(void)
+{
+	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
+		return false;
+
+	if (boot_cpu_data.x86 == 0x0f) {
+		/*
+		 * On Pentium 4, camellia-asm is slower than original assembler
+		 * implementation because excessive uses of 64bit rotate and
+		 * left-shifts (which are really slow on P4) needed to store and
+		 * handle 128bit block in two 64bit registers.
+		 */
+		return true;
+	}
+
+	return false;
+}
+
+static int force;
+module_param(force, int, 0);
+MODULE_PARM_DESC(force, "Force module load, ignore CPU blacklist");
+
+static int __init camellia_init(void)
+{
+	int err;
+
+	if (!force && is_blacklisted_cpu()) {
+		printk(KERN_INFO
+			"camellia-x86_64: performance on this CPU "
+			"would be suboptimal: disabling "
+			"camellia-x86_64.\n");
+		return -ENODEV;
+	}
+
+	err = crypto_register_alg(&camellia_cipher_alg);
+	if (err)
+		return err;
+
+	err = crypto_register_skciphers(camellia_skcipher_algs,
+					ARRAY_SIZE(camellia_skcipher_algs));
+	if (err)
+		crypto_unregister_alg(&camellia_cipher_alg);
+
+	return err;
+}
+
+static void __exit camellia_fini(void)
+{
+	crypto_unregister_alg(&camellia_cipher_alg);
+	crypto_unregister_skciphers(camellia_skcipher_algs,
+				    ARRAY_SIZE(camellia_skcipher_algs));
+}
+
+module_init(camellia_init);
+module_exit(camellia_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Camellia Cipher Algorithm, asm optimized");
+MODULE_ALIAS_CRYPTO("camellia");
+MODULE_ALIAS_CRYPTO("camellia-asm");
diff --git a/arch/x86/crypto/cast5-avx-x86_64-asm_64.S b/arch/x86/crypto/cast5-avx-x86_64-asm_64.S
new file mode 100644
index 000000000..fb95a6142
--- /dev/null
+++ b/arch/x86/crypto/cast5-avx-x86_64-asm_64.S
@@ -0,0 +1,489 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Cast5 Cipher 16-way parallel algorithm (AVX/x86_64)
+ *
+ * Copyright (C) 2012 Johannes Goetzfried
+ *     <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
+ *
+ * Copyright © 2012 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ */
+
+#include <linux/linkage.h>
+#include <asm/frame.h>
+
+.file "cast5-avx-x86_64-asm_64.S"
+
+.extern cast_s1
+.extern cast_s2
+.extern cast_s3
+.extern cast_s4
+
+/* structure of crypto context */
+#define km	0
+#define kr	(16*4)
+#define rr	((16*4)+16)
+
+/* s-boxes */
+#define s1	cast_s1
+#define s2	cast_s2
+#define s3	cast_s3
+#define s4	cast_s4
+
+/**********************************************************************
+  16-way AVX cast5
+ **********************************************************************/
+#define CTX %r15
+
+#define RL1 %xmm0
+#define RR1 %xmm1
+#define RL2 %xmm2
+#define RR2 %xmm3
+#define RL3 %xmm4
+#define RR3 %xmm5
+#define RL4 %xmm6
+#define RR4 %xmm7
+
+#define RX %xmm8
+
+#define RKM  %xmm9
+#define RKR  %xmm10
+#define RKRF %xmm11
+#define RKRR %xmm12
+
+#define R32  %xmm13
+#define R1ST %xmm14
+
+#define RTMP %xmm15
+
+#define RID1  %rdi
+#define RID1d %edi
+#define RID2  %rsi
+#define RID2d %esi
+
+#define RGI1   %rdx
+#define RGI1bl %dl
+#define RGI1bh %dh
+#define RGI2   %rcx
+#define RGI2bl %cl
+#define RGI2bh %ch
+
+#define RGI3   %rax
+#define RGI3bl %al
+#define RGI3bh %ah
+#define RGI4   %rbx
+#define RGI4bl %bl
+#define RGI4bh %bh
+
+#define RFS1  %r8
+#define RFS1d %r8d
+#define RFS2  %r9
+#define RFS2d %r9d
+#define RFS3  %r10
+#define RFS3d %r10d
+
+
+#define lookup_32bit(src, dst, op1, op2, op3, interleave_op, il_reg) \
+	movzbl		src ## bh,     RID1d;    \
+	leaq		s1(%rip),      RID2;     \
+	movl		(RID2,RID1,4), dst ## d; \
+	movzbl		src ## bl,     RID2d;    \
+	leaq		s2(%rip),      RID1;     \
+	op1		(RID1,RID2,4), dst ## d; \
+	shrq $16,	src;                     \
+	movzbl		src ## bh,     RID1d;    \
+	leaq		s3(%rip),      RID2;     \
+	op2		(RID2,RID1,4), dst ## d; \
+	movzbl		src ## bl,     RID2d;    \
+	interleave_op(il_reg);			 \
+	leaq		s4(%rip),      RID1;     \
+	op3		(RID1,RID2,4), dst ## d;
+
+#define dummy(d) /* do nothing */
+
+#define shr_next(reg) \
+	shrq $16,	reg;
+
+#define F_head(a, x, gi1, gi2, op0) \
+	op0	a,	RKM,  x;                 \
+	vpslld	RKRF,	x,    RTMP;              \
+	vpsrld	RKRR,	x,    x;                 \
+	vpor	RTMP,	x,    x;                 \
+	\
+	vmovq		x,    gi1;               \
+	vpextrq $1,	x,    gi2;
+
+#define F_tail(a, x, gi1, gi2, op1, op2, op3) \
+	lookup_32bit(##gi1, RFS1, op1, op2, op3, shr_next, ##gi1); \
+	lookup_32bit(##gi2, RFS3, op1, op2, op3, shr_next, ##gi2); \
+	\
+	lookup_32bit(##gi1, RFS2, op1, op2, op3, dummy, none);     \
+	shlq $32,	RFS2;                                      \
+	orq		RFS1, RFS2;                                \
+	lookup_32bit(##gi2, RFS1, op1, op2, op3, dummy, none);     \
+	shlq $32,	RFS1;                                      \
+	orq		RFS1, RFS3;                                \
+	\
+	vmovq		RFS2, x;                                   \
+	vpinsrq $1,	RFS3, x, x;
+
+#define F_2(a1, b1, a2, b2, op0, op1, op2, op3) \
+	F_head(b1, RX, RGI1, RGI2, op0);              \
+	F_head(b2, RX, RGI3, RGI4, op0);              \
+	\
+	F_tail(b1, RX, RGI1, RGI2, op1, op2, op3);    \
+	F_tail(b2, RTMP, RGI3, RGI4, op1, op2, op3);  \
+	\
+	vpxor		a1, RX,   a1;                 \
+	vpxor		a2, RTMP, a2;
+
+#define F1_2(a1, b1, a2, b2) \
+	F_2(a1, b1, a2, b2, vpaddd, xorl, subl, addl)
+#define F2_2(a1, b1, a2, b2) \
+	F_2(a1, b1, a2, b2, vpxor, subl, addl, xorl)
+#define F3_2(a1, b1, a2, b2) \
+	F_2(a1, b1, a2, b2, vpsubd, addl, xorl, subl)
+
+#define subround(a1, b1, a2, b2, f) \
+	F ## f ## _2(a1, b1, a2, b2);
+
+#define round(l, r, n, f) \
+	vbroadcastss 	(km+(4*n))(CTX), RKM;        \
+	vpand		R1ST,            RKR,  RKRF; \
+	vpsubq		RKRF,            R32,  RKRR; \
+	vpsrldq $1,	RKR,             RKR;        \
+	subround(l ## 1, r ## 1, l ## 2, r ## 2, f); \
+	subround(l ## 3, r ## 3, l ## 4, r ## 4, f);
+
+#define enc_preload_rkr() \
+	vbroadcastss	.L16_mask(%rip),          RKR;      \
+	/* add 16-bit rotation to key rotations (mod 32) */ \
+	vpxor		kr(CTX),                  RKR, RKR;
+
+#define dec_preload_rkr() \
+	vbroadcastss	.L16_mask(%rip),          RKR;      \
+	/* add 16-bit rotation to key rotations (mod 32) */ \
+	vpxor		kr(CTX),                  RKR, RKR; \
+	vpshufb		.Lbswap128_mask(%rip),    RKR, RKR;
+
+#define transpose_2x4(x0, x1, t0, t1) \
+	vpunpckldq		x1, x0, t0; \
+	vpunpckhdq		x1, x0, t1; \
+	\
+	vpunpcklqdq		t1, t0, x0; \
+	vpunpckhqdq		t1, t0, x1;
+
+#define inpack_blocks(x0, x1, t0, t1, rmask) \
+	vpshufb rmask, 	x0,	x0; \
+	vpshufb rmask, 	x1,	x1; \
+	\
+	transpose_2x4(x0, x1, t0, t1)
+
+#define outunpack_blocks(x0, x1, t0, t1, rmask) \
+	transpose_2x4(x0, x1, t0, t1) \
+	\
+	vpshufb rmask,	x0, x0;           \
+	vpshufb rmask,	x1, x1;
+
+.section	.rodata.cst16.bswap_mask, "aM", @progbits, 16
+.align 16
+.Lbswap_mask:
+	.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
+.section	.rodata.cst16.bswap128_mask, "aM", @progbits, 16
+.align 16
+.Lbswap128_mask:
+	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+.section	.rodata.cst16.bswap_iv_mask, "aM", @progbits, 16
+.align 16
+.Lbswap_iv_mask:
+	.byte 7, 6, 5, 4, 3, 2, 1, 0, 7, 6, 5, 4, 3, 2, 1, 0
+
+.section	.rodata.cst4.16_mask, "aM", @progbits, 4
+.align 4
+.L16_mask:
+	.byte 16, 16, 16, 16
+.section	.rodata.cst4.32_mask, "aM", @progbits, 4
+.align 4
+.L32_mask:
+	.byte 32, 0, 0, 0
+.section	.rodata.cst4.first_mask, "aM", @progbits, 4
+.align 4
+.Lfirst_mask:
+	.byte 0x1f, 0, 0, 0
+
+.text
+
+SYM_FUNC_START_LOCAL(__cast5_enc_blk16)
+	/* input:
+	 *	%rdi: ctx
+	 *	RL1: blocks 1 and 2
+	 *	RR1: blocks 3 and 4
+	 *	RL2: blocks 5 and 6
+	 *	RR2: blocks 7 and 8
+	 *	RL3: blocks 9 and 10
+	 *	RR3: blocks 11 and 12
+	 *	RL4: blocks 13 and 14
+	 *	RR4: blocks 15 and 16
+	 * output:
+	 *	RL1: encrypted blocks 1 and 2
+	 *	RR1: encrypted blocks 3 and 4
+	 *	RL2: encrypted blocks 5 and 6
+	 *	RR2: encrypted blocks 7 and 8
+	 *	RL3: encrypted blocks 9 and 10
+	 *	RR3: encrypted blocks 11 and 12
+	 *	RL4: encrypted blocks 13 and 14
+	 *	RR4: encrypted blocks 15 and 16
+	 */
+
+	pushq %r15;
+	pushq %rbx;
+
+	movq %rdi, CTX;
+
+	vmovdqa .Lbswap_mask(%rip), RKM;
+	vmovd .Lfirst_mask(%rip), R1ST;
+	vmovd .L32_mask(%rip), R32;
+	enc_preload_rkr();
+
+	inpack_blocks(RL1, RR1, RTMP, RX, RKM);
+	inpack_blocks(RL2, RR2, RTMP, RX, RKM);
+	inpack_blocks(RL3, RR3, RTMP, RX, RKM);
+	inpack_blocks(RL4, RR4, RTMP, RX, RKM);
+
+	round(RL, RR, 0, 1);
+	round(RR, RL, 1, 2);
+	round(RL, RR, 2, 3);
+	round(RR, RL, 3, 1);
+	round(RL, RR, 4, 2);
+	round(RR, RL, 5, 3);
+	round(RL, RR, 6, 1);
+	round(RR, RL, 7, 2);
+	round(RL, RR, 8, 3);
+	round(RR, RL, 9, 1);
+	round(RL, RR, 10, 2);
+	round(RR, RL, 11, 3);
+
+	movzbl rr(CTX), %eax;
+	testl %eax, %eax;
+	jnz .L__skip_enc;
+
+	round(RL, RR, 12, 1);
+	round(RR, RL, 13, 2);
+	round(RL, RR, 14, 3);
+	round(RR, RL, 15, 1);
+
+.L__skip_enc:
+	popq %rbx;
+	popq %r15;
+
+	vmovdqa .Lbswap_mask(%rip), RKM;
+
+	outunpack_blocks(RR1, RL1, RTMP, RX, RKM);
+	outunpack_blocks(RR2, RL2, RTMP, RX, RKM);
+	outunpack_blocks(RR3, RL3, RTMP, RX, RKM);
+	outunpack_blocks(RR4, RL4, RTMP, RX, RKM);
+
+	RET;
+SYM_FUNC_END(__cast5_enc_blk16)
+
+SYM_FUNC_START_LOCAL(__cast5_dec_blk16)
+	/* input:
+	 *	%rdi: ctx
+	 *	RL1: encrypted blocks 1 and 2
+	 *	RR1: encrypted blocks 3 and 4
+	 *	RL2: encrypted blocks 5 and 6
+	 *	RR2: encrypted blocks 7 and 8
+	 *	RL3: encrypted blocks 9 and 10
+	 *	RR3: encrypted blocks 11 and 12
+	 *	RL4: encrypted blocks 13 and 14
+	 *	RR4: encrypted blocks 15 and 16
+	 * output:
+	 *	RL1: decrypted blocks 1 and 2
+	 *	RR1: decrypted blocks 3 and 4
+	 *	RL2: decrypted blocks 5 and 6
+	 *	RR2: decrypted blocks 7 and 8
+	 *	RL3: decrypted blocks 9 and 10
+	 *	RR3: decrypted blocks 11 and 12
+	 *	RL4: decrypted blocks 13 and 14
+	 *	RR4: decrypted blocks 15 and 16
+	 */
+
+	pushq %r15;
+	pushq %rbx;
+
+	movq %rdi, CTX;
+
+	vmovdqa .Lbswap_mask(%rip), RKM;
+	vmovd .Lfirst_mask(%rip), R1ST;
+	vmovd .L32_mask(%rip), R32;
+	dec_preload_rkr();
+
+	inpack_blocks(RL1, RR1, RTMP, RX, RKM);
+	inpack_blocks(RL2, RR2, RTMP, RX, RKM);
+	inpack_blocks(RL3, RR3, RTMP, RX, RKM);
+	inpack_blocks(RL4, RR4, RTMP, RX, RKM);
+
+	movzbl rr(CTX), %eax;
+	testl %eax, %eax;
+	jnz .L__skip_dec;
+
+	round(RL, RR, 15, 1);
+	round(RR, RL, 14, 3);
+	round(RL, RR, 13, 2);
+	round(RR, RL, 12, 1);
+
+.L__dec_tail:
+	round(RL, RR, 11, 3);
+	round(RR, RL, 10, 2);
+	round(RL, RR, 9, 1);
+	round(RR, RL, 8, 3);
+	round(RL, RR, 7, 2);
+	round(RR, RL, 6, 1);
+	round(RL, RR, 5, 3);
+	round(RR, RL, 4, 2);
+	round(RL, RR, 3, 1);
+	round(RR, RL, 2, 3);
+	round(RL, RR, 1, 2);
+	round(RR, RL, 0, 1);
+
+	vmovdqa .Lbswap_mask(%rip), RKM;
+	popq %rbx;
+	popq %r15;
+
+	outunpack_blocks(RR1, RL1, RTMP, RX, RKM);
+	outunpack_blocks(RR2, RL2, RTMP, RX, RKM);
+	outunpack_blocks(RR3, RL3, RTMP, RX, RKM);
+	outunpack_blocks(RR4, RL4, RTMP, RX, RKM);
+
+	RET;
+
+.L__skip_dec:
+	vpsrldq $4, RKR, RKR;
+	jmp .L__dec_tail;
+SYM_FUNC_END(__cast5_dec_blk16)
+
+SYM_FUNC_START(cast5_ecb_enc_16way)
+	/* input:
+	 *	%rdi: ctx
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+	pushq %r15;
+
+	movq %rdi, CTX;
+	movq %rsi, %r11;
+
+	vmovdqu (0*4*4)(%rdx), RL1;
+	vmovdqu (1*4*4)(%rdx), RR1;
+	vmovdqu (2*4*4)(%rdx), RL2;
+	vmovdqu (3*4*4)(%rdx), RR2;
+	vmovdqu (4*4*4)(%rdx), RL3;
+	vmovdqu (5*4*4)(%rdx), RR3;
+	vmovdqu (6*4*4)(%rdx), RL4;
+	vmovdqu (7*4*4)(%rdx), RR4;
+
+	call __cast5_enc_blk16;
+
+	vmovdqu RR1, (0*4*4)(%r11);
+	vmovdqu RL1, (1*4*4)(%r11);
+	vmovdqu RR2, (2*4*4)(%r11);
+	vmovdqu RL2, (3*4*4)(%r11);
+	vmovdqu RR3, (4*4*4)(%r11);
+	vmovdqu RL3, (5*4*4)(%r11);
+	vmovdqu RR4, (6*4*4)(%r11);
+	vmovdqu RL4, (7*4*4)(%r11);
+
+	popq %r15;
+	FRAME_END
+	RET;
+SYM_FUNC_END(cast5_ecb_enc_16way)
+
+SYM_FUNC_START(cast5_ecb_dec_16way)
+	/* input:
+	 *	%rdi: ctx
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+
+	FRAME_BEGIN
+	pushq %r15;
+
+	movq %rdi, CTX;
+	movq %rsi, %r11;
+
+	vmovdqu (0*4*4)(%rdx), RL1;
+	vmovdqu (1*4*4)(%rdx), RR1;
+	vmovdqu (2*4*4)(%rdx), RL2;
+	vmovdqu (3*4*4)(%rdx), RR2;
+	vmovdqu (4*4*4)(%rdx), RL3;
+	vmovdqu (5*4*4)(%rdx), RR3;
+	vmovdqu (6*4*4)(%rdx), RL4;
+	vmovdqu (7*4*4)(%rdx), RR4;
+
+	call __cast5_dec_blk16;
+
+	vmovdqu RR1, (0*4*4)(%r11);
+	vmovdqu RL1, (1*4*4)(%r11);
+	vmovdqu RR2, (2*4*4)(%r11);
+	vmovdqu RL2, (3*4*4)(%r11);
+	vmovdqu RR3, (4*4*4)(%r11);
+	vmovdqu RL3, (5*4*4)(%r11);
+	vmovdqu RR4, (6*4*4)(%r11);
+	vmovdqu RL4, (7*4*4)(%r11);
+
+	popq %r15;
+	FRAME_END
+	RET;
+SYM_FUNC_END(cast5_ecb_dec_16way)
+
+SYM_FUNC_START(cast5_cbc_dec_16way)
+	/* input:
+	 *	%rdi: ctx
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+	pushq %r12;
+	pushq %r15;
+
+	movq %rdi, CTX;
+	movq %rsi, %r11;
+	movq %rdx, %r12;
+
+	vmovdqu (0*16)(%rdx), RL1;
+	vmovdqu (1*16)(%rdx), RR1;
+	vmovdqu (2*16)(%rdx), RL2;
+	vmovdqu (3*16)(%rdx), RR2;
+	vmovdqu (4*16)(%rdx), RL3;
+	vmovdqu (5*16)(%rdx), RR3;
+	vmovdqu (6*16)(%rdx), RL4;
+	vmovdqu (7*16)(%rdx), RR4;
+
+	call __cast5_dec_blk16;
+
+	/* xor with src */
+	vmovq (%r12), RX;
+	vpshufd $0x4f, RX, RX;
+	vpxor RX, RR1, RR1;
+	vpxor 0*16+8(%r12), RL1, RL1;
+	vpxor 1*16+8(%r12), RR2, RR2;
+	vpxor 2*16+8(%r12), RL2, RL2;
+	vpxor 3*16+8(%r12), RR3, RR3;
+	vpxor 4*16+8(%r12), RL3, RL3;
+	vpxor 5*16+8(%r12), RR4, RR4;
+	vpxor 6*16+8(%r12), RL4, RL4;
+
+	vmovdqu RR1, (0*16)(%r11);
+	vmovdqu RL1, (1*16)(%r11);
+	vmovdqu RR2, (2*16)(%r11);
+	vmovdqu RL2, (3*16)(%r11);
+	vmovdqu RR3, (4*16)(%r11);
+	vmovdqu RL3, (5*16)(%r11);
+	vmovdqu RR4, (6*16)(%r11);
+	vmovdqu RL4, (7*16)(%r11);
+
+	popq %r15;
+	popq %r12;
+	FRAME_END
+	RET;
+SYM_FUNC_END(cast5_cbc_dec_16way)
diff --git a/arch/x86/crypto/cast5_avx_glue.c b/arch/x86/crypto/cast5_avx_glue.c
new file mode 100644
index 000000000..3aca04d43
--- /dev/null
+++ b/arch/x86/crypto/cast5_avx_glue.c
@@ -0,0 +1,117 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Glue Code for the AVX assembler implementation of the Cast5 Cipher
+ *
+ * Copyright (C) 2012 Johannes Goetzfried
+ *     <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
+ */
+
+#include <crypto/algapi.h>
+#include <crypto/cast5.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <linux/module.h>
+#include <linux/types.h>
+
+#include "ecb_cbc_helpers.h"
+
+#define CAST5_PARALLEL_BLOCKS 16
+
+asmlinkage void cast5_ecb_enc_16way(struct cast5_ctx *ctx, u8 *dst,
+				    const u8 *src);
+asmlinkage void cast5_ecb_dec_16way(struct cast5_ctx *ctx, u8 *dst,
+				    const u8 *src);
+asmlinkage void cast5_cbc_dec_16way(struct cast5_ctx *ctx, u8 *dst,
+				    const u8 *src);
+
+static int cast5_setkey_skcipher(struct crypto_skcipher *tfm, const u8 *key,
+				 unsigned int keylen)
+{
+	return cast5_setkey(&tfm->base, key, keylen);
+}
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, CAST5_BLOCK_SIZE, CAST5_PARALLEL_BLOCKS);
+	ECB_BLOCK(CAST5_PARALLEL_BLOCKS, cast5_ecb_enc_16way);
+	ECB_BLOCK(1, __cast5_encrypt);
+	ECB_WALK_END();
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, CAST5_BLOCK_SIZE, CAST5_PARALLEL_BLOCKS);
+	ECB_BLOCK(CAST5_PARALLEL_BLOCKS, cast5_ecb_dec_16way);
+	ECB_BLOCK(1, __cast5_decrypt);
+	ECB_WALK_END();
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, CAST5_BLOCK_SIZE, -1);
+	CBC_ENC_BLOCK(__cast5_encrypt);
+	CBC_WALK_END();
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, CAST5_BLOCK_SIZE, CAST5_PARALLEL_BLOCKS);
+	CBC_DEC_BLOCK(CAST5_PARALLEL_BLOCKS, cast5_cbc_dec_16way);
+	CBC_DEC_BLOCK(1, __cast5_decrypt);
+	CBC_WALK_END();
+}
+
+static struct skcipher_alg cast5_algs[] = {
+	{
+		.base.cra_name		= "ecb(cast5)",
+		.base.cra_driver_name	= "ecb-cast5-avx",
+		.base.cra_priority	= 200,
+		.base.cra_blocksize	= CAST5_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct cast5_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= CAST5_MIN_KEY_SIZE,
+		.max_keysize		= CAST5_MAX_KEY_SIZE,
+		.setkey			= cast5_setkey_skcipher,
+		.encrypt		= ecb_encrypt,
+		.decrypt		= ecb_decrypt,
+	}, {
+		.base.cra_name		= "cbc(cast5)",
+		.base.cra_driver_name	= "cbc-cast5-avx",
+		.base.cra_priority	= 200,
+		.base.cra_blocksize	= CAST5_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct cast5_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= CAST5_MIN_KEY_SIZE,
+		.max_keysize		= CAST5_MAX_KEY_SIZE,
+		.ivsize			= CAST5_BLOCK_SIZE,
+		.setkey			= cast5_setkey_skcipher,
+		.encrypt		= cbc_encrypt,
+		.decrypt		= cbc_decrypt,
+	}
+};
+
+static int __init cast5_init(void)
+{
+	const char *feature_name;
+
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM,
+				&feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	return crypto_register_skciphers(cast5_algs,
+					 ARRAY_SIZE(cast5_algs));
+}
+
+static void __exit cast5_exit(void)
+{
+	crypto_unregister_skciphers(cast5_algs, ARRAY_SIZE(cast5_algs));
+}
+
+module_init(cast5_init);
+module_exit(cast5_exit);
+
+MODULE_DESCRIPTION("Cast5 Cipher Algorithm, AVX optimized");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_CRYPTO("cast5");
diff --git a/arch/x86/crypto/cast6-avx-x86_64-asm_64.S b/arch/x86/crypto/cast6-avx-x86_64-asm_64.S
new file mode 100644
index 000000000..9e86d460b
--- /dev/null
+++ b/arch/x86/crypto/cast6-avx-x86_64-asm_64.S
@@ -0,0 +1,416 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Cast6 Cipher 8-way parallel algorithm (AVX/x86_64)
+ *
+ * Copyright (C) 2012 Johannes Goetzfried
+ *     <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
+ *
+ * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ */
+
+#include <linux/linkage.h>
+#include <asm/frame.h>
+#include "glue_helper-asm-avx.S"
+
+.file "cast6-avx-x86_64-asm_64.S"
+
+.extern cast_s1
+.extern cast_s2
+.extern cast_s3
+.extern cast_s4
+
+/* structure of crypto context */
+#define km	0
+#define kr	(12*4*4)
+
+/* s-boxes */
+#define s1	cast_s1
+#define s2	cast_s2
+#define s3	cast_s3
+#define s4	cast_s4
+
+/**********************************************************************
+  8-way AVX cast6
+ **********************************************************************/
+#define CTX %r15
+
+#define RA1 %xmm0
+#define RB1 %xmm1
+#define RC1 %xmm2
+#define RD1 %xmm3
+
+#define RA2 %xmm4
+#define RB2 %xmm5
+#define RC2 %xmm6
+#define RD2 %xmm7
+
+#define RX  %xmm8
+
+#define RKM  %xmm9
+#define RKR  %xmm10
+#define RKRF %xmm11
+#define RKRR %xmm12
+#define R32  %xmm13
+#define R1ST %xmm14
+
+#define RTMP %xmm15
+
+#define RID1  %rdi
+#define RID1d %edi
+#define RID2  %rsi
+#define RID2d %esi
+
+#define RGI1   %rdx
+#define RGI1bl %dl
+#define RGI1bh %dh
+#define RGI2   %rcx
+#define RGI2bl %cl
+#define RGI2bh %ch
+
+#define RGI3   %rax
+#define RGI3bl %al
+#define RGI3bh %ah
+#define RGI4   %rbx
+#define RGI4bl %bl
+#define RGI4bh %bh
+
+#define RFS1  %r8
+#define RFS1d %r8d
+#define RFS2  %r9
+#define RFS2d %r9d
+#define RFS3  %r10
+#define RFS3d %r10d
+
+
+#define lookup_32bit(src, dst, op1, op2, op3, interleave_op, il_reg) \
+	movzbl		src ## bh,     RID1d;    \
+	leaq		s1(%rip),      RID2;     \
+	movl		(RID2,RID1,4), dst ## d; \
+	movzbl		src ## bl,     RID2d;    \
+	leaq		s2(%rip),      RID1;     \
+	op1		(RID1,RID2,4), dst ## d; \
+	shrq $16,	src;                     \
+	movzbl		src ## bh,     RID1d;    \
+	leaq		s3(%rip),      RID2;     \
+	op2		(RID2,RID1,4), dst ## d; \
+	movzbl		src ## bl,     RID2d;    \
+	interleave_op(il_reg);			 \
+	leaq		s4(%rip),      RID1;     \
+	op3		(RID1,RID2,4), dst ## d;
+
+#define dummy(d) /* do nothing */
+
+#define shr_next(reg) \
+	shrq $16,	reg;
+
+#define F_head(a, x, gi1, gi2, op0) \
+	op0	a,	RKM,  x;                 \
+	vpslld	RKRF,	x,    RTMP;              \
+	vpsrld	RKRR,	x,    x;                 \
+	vpor	RTMP,	x,    x;                 \
+	\
+	vmovq		x,    gi1;               \
+	vpextrq $1,	x,    gi2;
+
+#define F_tail(a, x, gi1, gi2, op1, op2, op3) \
+	lookup_32bit(##gi1, RFS1, op1, op2, op3, shr_next, ##gi1); \
+	lookup_32bit(##gi2, RFS3, op1, op2, op3, shr_next, ##gi2); \
+	\
+	lookup_32bit(##gi1, RFS2, op1, op2, op3, dummy, none);     \
+	shlq $32,	RFS2;                                      \
+	orq		RFS1, RFS2;                                \
+	lookup_32bit(##gi2, RFS1, op1, op2, op3, dummy, none);     \
+	shlq $32,	RFS1;                                      \
+	orq		RFS1, RFS3;                                \
+	\
+	vmovq		RFS2, x;                                   \
+	vpinsrq $1,	RFS3, x, x;
+
+#define F_2(a1, b1, a2, b2, op0, op1, op2, op3) \
+	F_head(b1, RX, RGI1, RGI2, op0);              \
+	F_head(b2, RX, RGI3, RGI4, op0);              \
+	\
+	F_tail(b1, RX, RGI1, RGI2, op1, op2, op3);    \
+	F_tail(b2, RTMP, RGI3, RGI4, op1, op2, op3);  \
+	\
+	vpxor		a1, RX,   a1;                 \
+	vpxor		a2, RTMP, a2;
+
+#define F1_2(a1, b1, a2, b2) \
+	F_2(a1, b1, a2, b2, vpaddd, xorl, subl, addl)
+#define F2_2(a1, b1, a2, b2) \
+	F_2(a1, b1, a2, b2, vpxor, subl, addl, xorl)
+#define F3_2(a1, b1, a2, b2) \
+	F_2(a1, b1, a2, b2, vpsubd, addl, xorl, subl)
+
+#define qop(in, out, f) \
+	F ## f ## _2(out ## 1, in ## 1, out ## 2, in ## 2);
+
+#define get_round_keys(nn) \
+	vbroadcastss	(km+(4*(nn)))(CTX), RKM;        \
+	vpand		R1ST,               RKR,  RKRF; \
+	vpsubq		RKRF,               R32,  RKRR; \
+	vpsrldq $1,	RKR,                RKR;
+
+#define Q(n) \
+	get_round_keys(4*n+0); \
+	qop(RD, RC, 1);        \
+	\
+	get_round_keys(4*n+1); \
+	qop(RC, RB, 2);        \
+	\
+	get_round_keys(4*n+2); \
+	qop(RB, RA, 3);        \
+	\
+	get_round_keys(4*n+3); \
+	qop(RA, RD, 1);
+
+#define QBAR(n) \
+	get_round_keys(4*n+3); \
+	qop(RA, RD, 1);        \
+	\
+	get_round_keys(4*n+2); \
+	qop(RB, RA, 3);        \
+	\
+	get_round_keys(4*n+1); \
+	qop(RC, RB, 2);        \
+	\
+	get_round_keys(4*n+0); \
+	qop(RD, RC, 1);
+
+#define shuffle(mask) \
+	vpshufb		mask(%rip),            RKR, RKR;
+
+#define preload_rkr(n, do_mask, mask) \
+	vbroadcastss	.L16_mask(%rip),          RKR;      \
+	/* add 16-bit rotation to key rotations (mod 32) */ \
+	vpxor		(kr+n*16)(CTX),           RKR, RKR; \
+	do_mask(mask);
+
+#define transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	vpunpckldq		x1, x0, t0; \
+	vpunpckhdq		x1, x0, t2; \
+	vpunpckldq		x3, x2, t1; \
+	vpunpckhdq		x3, x2, x3; \
+	\
+	vpunpcklqdq		t1, t0, x0; \
+	vpunpckhqdq		t1, t0, x1; \
+	vpunpcklqdq		x3, t2, x2; \
+	vpunpckhqdq		x3, t2, x3;
+
+#define inpack_blocks(x0, x1, x2, x3, t0, t1, t2, rmask) \
+	vpshufb rmask, x0,	x0; \
+	vpshufb rmask, x1,	x1; \
+	vpshufb rmask, x2,	x2; \
+	vpshufb rmask, x3,	x3; \
+	\
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2)
+
+#define outunpack_blocks(x0, x1, x2, x3, t0, t1, t2, rmask) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	\
+	vpshufb rmask,		x0, x0;       \
+	vpshufb rmask,		x1, x1;       \
+	vpshufb rmask,		x2, x2;       \
+	vpshufb rmask,		x3, x3;
+
+.section	.rodata.cst16, "aM", @progbits, 16
+.align 16
+.Lbswap_mask:
+	.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
+.Lbswap128_mask:
+	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+.Lrkr_enc_Q_Q_QBAR_QBAR:
+	.byte 0, 1, 2, 3, 4, 5, 6, 7, 11, 10, 9, 8, 15, 14, 13, 12
+.Lrkr_enc_QBAR_QBAR_QBAR_QBAR:
+	.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
+.Lrkr_dec_Q_Q_Q_Q:
+	.byte 12, 13, 14, 15, 8, 9, 10, 11, 4, 5, 6, 7, 0, 1, 2, 3
+.Lrkr_dec_Q_Q_QBAR_QBAR:
+	.byte 12, 13, 14, 15, 8, 9, 10, 11, 7, 6, 5, 4, 3, 2, 1, 0
+.Lrkr_dec_QBAR_QBAR_QBAR_QBAR:
+	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+
+.section	.rodata.cst4.L16_mask, "aM", @progbits, 4
+.align 4
+.L16_mask:
+	.byte 16, 16, 16, 16
+
+.section	.rodata.cst4.L32_mask, "aM", @progbits, 4
+.align 4
+.L32_mask:
+	.byte 32, 0, 0, 0
+
+.section	.rodata.cst4.first_mask, "aM", @progbits, 4
+.align 4
+.Lfirst_mask:
+	.byte 0x1f, 0, 0, 0
+
+.text
+
+.align 8
+SYM_FUNC_START_LOCAL(__cast6_enc_blk8)
+	/* input:
+	 *	%rdi: ctx
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: blocks
+	 * output:
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: encrypted blocks
+	 */
+
+	pushq %r15;
+	pushq %rbx;
+
+	movq %rdi, CTX;
+
+	vmovdqa .Lbswap_mask(%rip), RKM;
+	vmovd .Lfirst_mask(%rip), R1ST;
+	vmovd .L32_mask(%rip), R32;
+
+	inpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
+	inpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);
+
+	preload_rkr(0, dummy, none);
+	Q(0);
+	Q(1);
+	Q(2);
+	Q(3);
+	preload_rkr(1, shuffle, .Lrkr_enc_Q_Q_QBAR_QBAR);
+	Q(4);
+	Q(5);
+	QBAR(6);
+	QBAR(7);
+	preload_rkr(2, shuffle, .Lrkr_enc_QBAR_QBAR_QBAR_QBAR);
+	QBAR(8);
+	QBAR(9);
+	QBAR(10);
+	QBAR(11);
+
+	popq %rbx;
+	popq %r15;
+
+	vmovdqa .Lbswap_mask(%rip), RKM;
+
+	outunpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
+	outunpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);
+
+	RET;
+SYM_FUNC_END(__cast6_enc_blk8)
+
+.align 8
+SYM_FUNC_START_LOCAL(__cast6_dec_blk8)
+	/* input:
+	 *	%rdi: ctx
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: encrypted blocks
+	 * output:
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: decrypted blocks
+	 */
+
+	pushq %r15;
+	pushq %rbx;
+
+	movq %rdi, CTX;
+
+	vmovdqa .Lbswap_mask(%rip), RKM;
+	vmovd .Lfirst_mask(%rip), R1ST;
+	vmovd .L32_mask(%rip), R32;
+
+	inpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
+	inpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);
+
+	preload_rkr(2, shuffle, .Lrkr_dec_Q_Q_Q_Q);
+	Q(11);
+	Q(10);
+	Q(9);
+	Q(8);
+	preload_rkr(1, shuffle, .Lrkr_dec_Q_Q_QBAR_QBAR);
+	Q(7);
+	Q(6);
+	QBAR(5);
+	QBAR(4);
+	preload_rkr(0, shuffle, .Lrkr_dec_QBAR_QBAR_QBAR_QBAR);
+	QBAR(3);
+	QBAR(2);
+	QBAR(1);
+	QBAR(0);
+
+	popq %rbx;
+	popq %r15;
+
+	vmovdqa .Lbswap_mask(%rip), RKM;
+	outunpack_blocks(RA1, RB1, RC1, RD1, RTMP, RX, RKRF, RKM);
+	outunpack_blocks(RA2, RB2, RC2, RD2, RTMP, RX, RKRF, RKM);
+
+	RET;
+SYM_FUNC_END(__cast6_dec_blk8)
+
+SYM_FUNC_START(cast6_ecb_enc_8way)
+	/* input:
+	 *	%rdi: ctx
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+	pushq %r15;
+
+	movq %rdi, CTX;
+	movq %rsi, %r11;
+
+	load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	call __cast6_enc_blk8;
+
+	store_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	popq %r15;
+	FRAME_END
+	RET;
+SYM_FUNC_END(cast6_ecb_enc_8way)
+
+SYM_FUNC_START(cast6_ecb_dec_8way)
+	/* input:
+	 *	%rdi: ctx
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+	pushq %r15;
+
+	movq %rdi, CTX;
+	movq %rsi, %r11;
+
+	load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	call __cast6_dec_blk8;
+
+	store_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	popq %r15;
+	FRAME_END
+	RET;
+SYM_FUNC_END(cast6_ecb_dec_8way)
+
+SYM_FUNC_START(cast6_cbc_dec_8way)
+	/* input:
+	 *	%rdi: ctx
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+	pushq %r12;
+	pushq %r15;
+
+	movq %rdi, CTX;
+	movq %rsi, %r11;
+	movq %rdx, %r12;
+
+	load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	call __cast6_dec_blk8;
+
+	store_cbc_8way(%r12, %r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	popq %r15;
+	popq %r12;
+	FRAME_END
+	RET;
+SYM_FUNC_END(cast6_cbc_dec_8way)
diff --git a/arch/x86/crypto/cast6_avx_glue.c b/arch/x86/crypto/cast6_avx_glue.c
new file mode 100644
index 000000000..c4dd28c30
--- /dev/null
+++ b/arch/x86/crypto/cast6_avx_glue.c
@@ -0,0 +1,116 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Glue Code for the AVX assembler implementation of the Cast6 Cipher
+ *
+ * Copyright (C) 2012 Johannes Goetzfried
+ *     <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
+ *
+ * Copyright © 2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ */
+
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <crypto/algapi.h>
+#include <crypto/cast6.h>
+
+#include "ecb_cbc_helpers.h"
+
+#define CAST6_PARALLEL_BLOCKS 8
+
+asmlinkage void cast6_ecb_enc_8way(const void *ctx, u8 *dst, const u8 *src);
+asmlinkage void cast6_ecb_dec_8way(const void *ctx, u8 *dst, const u8 *src);
+
+asmlinkage void cast6_cbc_dec_8way(const void *ctx, u8 *dst, const u8 *src);
+
+static int cast6_setkey_skcipher(struct crypto_skcipher *tfm,
+				 const u8 *key, unsigned int keylen)
+{
+	return cast6_setkey(&tfm->base, key, keylen);
+}
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, CAST6_BLOCK_SIZE, CAST6_PARALLEL_BLOCKS);
+	ECB_BLOCK(CAST6_PARALLEL_BLOCKS, cast6_ecb_enc_8way);
+	ECB_BLOCK(1, __cast6_encrypt);
+	ECB_WALK_END();
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, CAST6_BLOCK_SIZE, CAST6_PARALLEL_BLOCKS);
+	ECB_BLOCK(CAST6_PARALLEL_BLOCKS, cast6_ecb_dec_8way);
+	ECB_BLOCK(1, __cast6_decrypt);
+	ECB_WALK_END();
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, CAST6_BLOCK_SIZE, -1);
+	CBC_ENC_BLOCK(__cast6_encrypt);
+	CBC_WALK_END();
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, CAST6_BLOCK_SIZE, CAST6_PARALLEL_BLOCKS);
+	CBC_DEC_BLOCK(CAST6_PARALLEL_BLOCKS, cast6_cbc_dec_8way);
+	CBC_DEC_BLOCK(1, __cast6_decrypt);
+	CBC_WALK_END();
+}
+
+static struct skcipher_alg cast6_algs[] = {
+	{
+		.base.cra_name		= "ecb(cast6)",
+		.base.cra_driver_name	= "ecb-cast6-avx",
+		.base.cra_priority	= 200,
+		.base.cra_blocksize	= CAST6_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct cast6_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= CAST6_MIN_KEY_SIZE,
+		.max_keysize		= CAST6_MAX_KEY_SIZE,
+		.setkey			= cast6_setkey_skcipher,
+		.encrypt		= ecb_encrypt,
+		.decrypt		= ecb_decrypt,
+	}, {
+		.base.cra_name		= "cbc(cast6)",
+		.base.cra_driver_name	= "cbc-cast6-avx",
+		.base.cra_priority	= 200,
+		.base.cra_blocksize	= CAST6_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct cast6_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= CAST6_MIN_KEY_SIZE,
+		.max_keysize		= CAST6_MAX_KEY_SIZE,
+		.ivsize			= CAST6_BLOCK_SIZE,
+		.setkey			= cast6_setkey_skcipher,
+		.encrypt		= cbc_encrypt,
+		.decrypt		= cbc_decrypt,
+	},
+};
+
+static int __init cast6_init(void)
+{
+	const char *feature_name;
+
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM,
+				&feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	return crypto_register_skciphers(cast6_algs, ARRAY_SIZE(cast6_algs));
+}
+
+static void __exit cast6_exit(void)
+{
+	crypto_unregister_skciphers(cast6_algs, ARRAY_SIZE(cast6_algs));
+}
+
+module_init(cast6_init);
+module_exit(cast6_exit);
+
+MODULE_DESCRIPTION("Cast6 Cipher Algorithm, AVX optimized");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_CRYPTO("cast6");
diff --git a/arch/x86/crypto/ecb_cbc_helpers.h b/arch/x86/crypto/ecb_cbc_helpers.h
new file mode 100644
index 000000000..11955bd01
--- /dev/null
+++ b/arch/x86/crypto/ecb_cbc_helpers.h
@@ -0,0 +1,87 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+
+#ifndef _CRYPTO_ECB_CBC_HELPER_H
+#define _CRYPTO_ECB_CBC_HELPER_H
+
+#include <crypto/internal/skcipher.h>
+#include <asm/fpu/api.h>
+
+/*
+ * Mode helpers to instantiate parameterized skcipher ECB/CBC modes without
+ * having to rely on indirect calls and retpolines.
+ */
+
+#define ECB_WALK_START(req, bsize, fpu_blocks) do {			\
+	void *ctx = crypto_skcipher_ctx(crypto_skcipher_reqtfm(req));	\
+	const int __fpu_blocks = (fpu_blocks);				\
+	const int __bsize = (bsize);					\
+	struct skcipher_walk walk;					\
+	int err = skcipher_walk_virt(&walk, (req), false);		\
+	while (walk.nbytes > 0) {					\
+		unsigned int nbytes = walk.nbytes;			\
+		bool do_fpu = __fpu_blocks != -1 &&			\
+			      nbytes >= __fpu_blocks * __bsize;		\
+		const u8 *src = walk.src.virt.addr;			\
+		u8 *dst = walk.dst.virt.addr;				\
+		u8 __maybe_unused buf[(bsize)];				\
+		if (do_fpu) kernel_fpu_begin()
+
+#define CBC_WALK_START(req, bsize, fpu_blocks)				\
+	ECB_WALK_START(req, bsize, fpu_blocks)
+
+#define ECB_WALK_ADVANCE(blocks) do {					\
+	dst += (blocks) * __bsize;					\
+	src += (blocks) * __bsize;					\
+	nbytes -= (blocks) * __bsize;					\
+} while (0)
+
+#define ECB_BLOCK(blocks, func) do {					\
+	const int __blocks = (blocks);					\
+	if (do_fpu && __blocks < __fpu_blocks) {			\
+		kernel_fpu_end();					\
+		do_fpu = false;						\
+	}								\
+	while (nbytes >= __blocks * __bsize) {				\
+		(func)(ctx, dst, src);					\
+		ECB_WALK_ADVANCE(blocks);				\
+	}								\
+} while (0)
+
+#define CBC_ENC_BLOCK(func) do {					\
+	const u8 *__iv = walk.iv;					\
+	while (nbytes >= __bsize) {					\
+		crypto_xor_cpy(dst, src, __iv, __bsize);		\
+		(func)(ctx, dst, dst);					\
+		__iv = dst;						\
+		ECB_WALK_ADVANCE(1);					\
+	}								\
+	memcpy(walk.iv, __iv, __bsize);					\
+} while (0)
+
+#define CBC_DEC_BLOCK(blocks, func) do {				\
+	const int __blocks = (blocks);					\
+	if (do_fpu && __blocks <  __fpu_blocks) {			\
+		kernel_fpu_end();					\
+		do_fpu = false;						\
+	}								\
+	while (nbytes >= __blocks * __bsize) {				\
+		const u8 *__iv = src + ((blocks) - 1) * __bsize;	\
+		if (dst == src)						\
+			__iv = memcpy(buf, __iv, __bsize);		\
+		(func)(ctx, dst, src);					\
+		crypto_xor(dst, walk.iv, __bsize);			\
+		memcpy(walk.iv, __iv, __bsize);				\
+		ECB_WALK_ADVANCE(blocks);				\
+	}								\
+} while (0)
+
+#define ECB_WALK_END()							\
+		if (do_fpu) kernel_fpu_end();				\
+		err = skcipher_walk_done(&walk, nbytes);		\
+	}								\
+	return err;							\
+} while (0)
+
+#define CBC_WALK_END() ECB_WALK_END()
+
+#endif
diff --git a/arch/x86/crypto/glue_helper-asm-avx.S b/arch/x86/crypto/glue_helper-asm-avx.S
new file mode 100644
index 000000000..3da385271
--- /dev/null
+++ b/arch/x86/crypto/glue_helper-asm-avx.S
@@ -0,0 +1,36 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Shared glue code for 128bit block ciphers, AVX assembler macros
+ *
+ * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ */
+
+#define load_8way(src, x0, x1, x2, x3, x4, x5, x6, x7) \
+	vmovdqu (0*16)(src), x0; \
+	vmovdqu (1*16)(src), x1; \
+	vmovdqu (2*16)(src), x2; \
+	vmovdqu (3*16)(src), x3; \
+	vmovdqu (4*16)(src), x4; \
+	vmovdqu (5*16)(src), x5; \
+	vmovdqu (6*16)(src), x6; \
+	vmovdqu (7*16)(src), x7;
+
+#define store_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \
+	vmovdqu x0, (0*16)(dst); \
+	vmovdqu x1, (1*16)(dst); \
+	vmovdqu x2, (2*16)(dst); \
+	vmovdqu x3, (3*16)(dst); \
+	vmovdqu x4, (4*16)(dst); \
+	vmovdqu x5, (5*16)(dst); \
+	vmovdqu x6, (6*16)(dst); \
+	vmovdqu x7, (7*16)(dst);
+
+#define store_cbc_8way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7) \
+	vpxor (0*16)(src), x1, x1; \
+	vpxor (1*16)(src), x2, x2; \
+	vpxor (2*16)(src), x3, x3; \
+	vpxor (3*16)(src), x4, x4; \
+	vpxor (4*16)(src), x5, x5; \
+	vpxor (5*16)(src), x6, x6; \
+	vpxor (6*16)(src), x7, x7; \
+	store_8way(dst, x0, x1, x2, x3, x4, x5, x6, x7);
diff --git a/arch/x86/crypto/glue_helper-asm-avx2.S b/arch/x86/crypto/glue_helper-asm-avx2.S
new file mode 100644
index 000000000..c77e90494
--- /dev/null
+++ b/arch/x86/crypto/glue_helper-asm-avx2.S
@@ -0,0 +1,39 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Shared glue code for 128bit block ciphers, AVX2 assembler macros
+ *
+ * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ */
+
+#define load_16way(src, x0, x1, x2, x3, x4, x5, x6, x7) \
+	vmovdqu (0*32)(src), x0; \
+	vmovdqu (1*32)(src), x1; \
+	vmovdqu (2*32)(src), x2; \
+	vmovdqu (3*32)(src), x3; \
+	vmovdqu (4*32)(src), x4; \
+	vmovdqu (5*32)(src), x5; \
+	vmovdqu (6*32)(src), x6; \
+	vmovdqu (7*32)(src), x7;
+
+#define store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7) \
+	vmovdqu x0, (0*32)(dst); \
+	vmovdqu x1, (1*32)(dst); \
+	vmovdqu x2, (2*32)(dst); \
+	vmovdqu x3, (3*32)(dst); \
+	vmovdqu x4, (4*32)(dst); \
+	vmovdqu x5, (5*32)(dst); \
+	vmovdqu x6, (6*32)(dst); \
+	vmovdqu x7, (7*32)(dst);
+
+#define store_cbc_16way(src, dst, x0, x1, x2, x3, x4, x5, x6, x7, t0) \
+	vpxor t0, t0, t0; \
+	vinserti128 $1, (src), t0, t0; \
+	vpxor t0, x0, x0; \
+	vpxor (0*32+16)(src), x1, x1; \
+	vpxor (1*32+16)(src), x2, x2; \
+	vpxor (2*32+16)(src), x3, x3; \
+	vpxor (3*32+16)(src), x4, x4; \
+	vpxor (4*32+16)(src), x5, x5; \
+	vpxor (5*32+16)(src), x6, x6; \
+	vpxor (6*32+16)(src), x7, x7; \
+	store_16way(dst, x0, x1, x2, x3, x4, x5, x6, x7);
diff --git a/arch/x86/crypto/serpent-avx-x86_64-asm_64.S b/arch/x86/crypto/serpent-avx-x86_64-asm_64.S
new file mode 100644
index 000000000..84e47f7f6
--- /dev/null
+++ b/arch/x86/crypto/serpent-avx-x86_64-asm_64.S
@@ -0,0 +1,712 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Serpent Cipher 8-way parallel algorithm (x86_64/AVX)
+ *
+ * Copyright (C) 2012 Johannes Goetzfried
+ *     <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
+ *
+ * Copyright © 2011-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ */
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+#include <asm/frame.h>
+#include "glue_helper-asm-avx.S"
+
+.file "serpent-avx-x86_64-asm_64.S"
+
+.section	.rodata.cst16.bswap128_mask, "aM", @progbits, 16
+.align 16
+.Lbswap128_mask:
+	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+
+.text
+
+#define CTX %rdi
+
+/**********************************************************************
+  8-way AVX serpent
+ **********************************************************************/
+#define RA1 %xmm0
+#define RB1 %xmm1
+#define RC1 %xmm2
+#define RD1 %xmm3
+#define RE1 %xmm4
+
+#define tp  %xmm5
+
+#define RA2 %xmm6
+#define RB2 %xmm7
+#define RC2 %xmm8
+#define RD2 %xmm9
+#define RE2 %xmm10
+
+#define RNOT %xmm11
+
+#define RK0 %xmm12
+#define RK1 %xmm13
+#define RK2 %xmm14
+#define RK3 %xmm15
+
+
+#define S0_1(x0, x1, x2, x3, x4)      \
+	vpor		x0,   x3, tp; \
+	vpxor		x3,   x0, x0; \
+	vpxor		x2,   x3, x4; \
+	vpxor		RNOT, x4, x4; \
+	vpxor		x1,   tp, x3; \
+	vpand		x0,   x1, x1; \
+	vpxor		x4,   x1, x1; \
+	vpxor		x0,   x2, x2;
+#define S0_2(x0, x1, x2, x3, x4)      \
+	vpxor		x3,   x0, x0; \
+	vpor		x0,   x4, x4; \
+	vpxor		x2,   x0, x0; \
+	vpand		x1,   x2, x2; \
+	vpxor		x2,   x3, x3; \
+	vpxor		RNOT, x1, x1; \
+	vpxor		x4,   x2, x2; \
+	vpxor		x2,   x1, x1;
+
+#define S1_1(x0, x1, x2, x3, x4)      \
+	vpxor		x0,   x1, tp; \
+	vpxor		x3,   x0, x0; \
+	vpxor		RNOT, x3, x3; \
+	vpand		tp,   x1, x4; \
+	vpor		tp,   x0, x0; \
+	vpxor		x2,   x3, x3; \
+	vpxor		x3,   x0, x0; \
+	vpxor		x3,   tp, x1;
+#define S1_2(x0, x1, x2, x3, x4)      \
+	vpxor		x4,   x3, x3; \
+	vpor		x4,   x1, x1; \
+	vpxor		x2,   x4, x4; \
+	vpand		x0,   x2, x2; \
+	vpxor		x1,   x2, x2; \
+	vpor		x0,   x1, x1; \
+	vpxor		RNOT, x0, x0; \
+	vpxor		x2,   x0, x0; \
+	vpxor		x1,   x4, x4;
+
+#define S2_1(x0, x1, x2, x3, x4)      \
+	vpxor		RNOT, x3, x3; \
+	vpxor		x0,   x1, x1; \
+	vpand		x2,   x0, tp; \
+	vpxor		x3,   tp, tp; \
+	vpor		x0,   x3, x3; \
+	vpxor		x1,   x2, x2; \
+	vpxor		x1,   x3, x3; \
+	vpand		tp,   x1, x1;
+#define S2_2(x0, x1, x2, x3, x4)      \
+	vpxor		x2,   tp, tp; \
+	vpand		x3,   x2, x2; \
+	vpor		x1,   x3, x3; \
+	vpxor		RNOT, tp, tp; \
+	vpxor		tp,   x3, x3; \
+	vpxor		tp,   x0, x4; \
+	vpxor		x2,   tp, x0; \
+	vpor		x2,   x1, x1;
+
+#define S3_1(x0, x1, x2, x3, x4)      \
+	vpxor		x3,   x1, tp; \
+	vpor		x0,   x3, x3; \
+	vpand		x0,   x1, x4; \
+	vpxor		x2,   x0, x0; \
+	vpxor		tp,   x2, x2; \
+	vpand		x3,   tp, x1; \
+	vpxor		x3,   x2, x2; \
+	vpor		x4,   x0, x0; \
+	vpxor		x3,   x4, x4;
+#define S3_2(x0, x1, x2, x3, x4)      \
+	vpxor		x0,   x1, x1; \
+	vpand		x3,   x0, x0; \
+	vpand		x4,   x3, x3; \
+	vpxor		x2,   x3, x3; \
+	vpor		x1,   x4, x4; \
+	vpand		x1,   x2, x2; \
+	vpxor		x3,   x4, x4; \
+	vpxor		x3,   x0, x0; \
+	vpxor		x2,   x3, x3;
+
+#define S4_1(x0, x1, x2, x3, x4)      \
+	vpand		x0,   x3, tp; \
+	vpxor		x3,   x0, x0; \
+	vpxor		x2,   tp, tp; \
+	vpor		x3,   x2, x2; \
+	vpxor		x1,   x0, x0; \
+	vpxor		tp,   x3, x4; \
+	vpor		x0,   x2, x2; \
+	vpxor		x1,   x2, x2;
+#define S4_2(x0, x1, x2, x3, x4)      \
+	vpand		x0,   x1, x1; \
+	vpxor		x4,   x1, x1; \
+	vpand		x2,   x4, x4; \
+	vpxor		tp,   x2, x2; \
+	vpxor		x0,   x4, x4; \
+	vpor		x1,   tp, x3; \
+	vpxor		RNOT, x1, x1; \
+	vpxor		x0,   x3, x3;
+
+#define S5_1(x0, x1, x2, x3, x4)      \
+	vpor		x0,   x1, tp; \
+	vpxor		tp,   x2, x2; \
+	vpxor		RNOT, x3, x3; \
+	vpxor		x0,   x1, x4; \
+	vpxor		x2,   x0, x0; \
+	vpand		x4,   tp, x1; \
+	vpor		x3,   x4, x4; \
+	vpxor		x0,   x4, x4;
+#define S5_2(x0, x1, x2, x3, x4)      \
+	vpand		x3,   x0, x0; \
+	vpxor		x3,   x1, x1; \
+	vpxor		x2,   x3, x3; \
+	vpxor		x1,   x0, x0; \
+	vpand		x4,   x2, x2; \
+	vpxor		x2,   x1, x1; \
+	vpand		x0,   x2, x2; \
+	vpxor		x2,   x3, x3;
+
+#define S6_1(x0, x1, x2, x3, x4)      \
+	vpxor		x0,   x3, x3; \
+	vpxor		x2,   x1, tp; \
+	vpxor		x0,   x2, x2; \
+	vpand		x3,   x0, x0; \
+	vpor		x3,   tp, tp; \
+	vpxor		RNOT, x1, x4; \
+	vpxor		tp,   x0, x0; \
+	vpxor		x2,   tp, x1;
+#define S6_2(x0, x1, x2, x3, x4)      \
+	vpxor		x4,   x3, x3; \
+	vpxor		x0,   x4, x4; \
+	vpand		x0,   x2, x2; \
+	vpxor		x1,   x4, x4; \
+	vpxor		x3,   x2, x2; \
+	vpand		x1,   x3, x3; \
+	vpxor		x0,   x3, x3; \
+	vpxor		x2,   x1, x1;
+
+#define S7_1(x0, x1, x2, x3, x4)      \
+	vpxor		RNOT, x1, tp; \
+	vpxor		RNOT, x0, x0; \
+	vpand		x2,   tp, x1; \
+	vpxor		x3,   x1, x1; \
+	vpor		tp,   x3, x3; \
+	vpxor		x2,   tp, x4; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x0,   x3, x3; \
+	vpor		x1,   x0, x0;
+#define S7_2(x0, x1, x2, x3, x4)      \
+	vpand		x0,   x2, x2; \
+	vpxor		x4,   x0, x0; \
+	vpxor		x3,   x4, x4; \
+	vpand		x0,   x3, x3; \
+	vpxor		x1,   x4, x4; \
+	vpxor		x4,   x2, x2; \
+	vpxor		x1,   x3, x3; \
+	vpor		x0,   x4, x4; \
+	vpxor		x1,   x4, x4;
+
+#define SI0_1(x0, x1, x2, x3, x4)     \
+	vpxor		x0,   x1, x1; \
+	vpor		x1,   x3, tp; \
+	vpxor		x1,   x3, x4; \
+	vpxor		RNOT, x0, x0; \
+	vpxor		tp,   x2, x2; \
+	vpxor		x0,   tp, x3; \
+	vpand		x1,   x0, x0; \
+	vpxor		x2,   x0, x0;
+#define SI0_2(x0, x1, x2, x3, x4)     \
+	vpand		x3,   x2, x2; \
+	vpxor		x4,   x3, x3; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x3,   x1, x1; \
+	vpand		x0,   x3, x3; \
+	vpxor		x0,   x1, x1; \
+	vpxor		x2,   x0, x0; \
+	vpxor		x3,   x4, x4;
+
+#define SI1_1(x0, x1, x2, x3, x4)     \
+	vpxor		x3,   x1, x1; \
+	vpxor		x2,   x0, tp; \
+	vpxor		RNOT, x2, x2; \
+	vpor		x1,   x0, x4; \
+	vpxor		x3,   x4, x4; \
+	vpand		x1,   x3, x3; \
+	vpxor		x2,   x1, x1; \
+	vpand		x4,   x2, x2;
+#define SI1_2(x0, x1, x2, x3, x4)     \
+	vpxor		x1,   x4, x4; \
+	vpor		x3,   x1, x1; \
+	vpxor		tp,   x3, x3; \
+	vpxor		tp,   x2, x2; \
+	vpor		x4,   tp, x0; \
+	vpxor		x4,   x2, x2; \
+	vpxor		x0,   x1, x1; \
+	vpxor		x1,   x4, x4;
+
+#define SI2_1(x0, x1, x2, x3, x4)     \
+	vpxor		x1,   x2, x2; \
+	vpxor		RNOT, x3, tp; \
+	vpor		x2,   tp, tp; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x0,   x3, x4; \
+	vpxor		x1,   tp, x3; \
+	vpor		x2,   x1, x1; \
+	vpxor		x0,   x2, x2;
+#define SI2_2(x0, x1, x2, x3, x4)     \
+	vpxor		x4,   x1, x1; \
+	vpor		x3,   x4, x4; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x2,   x4, x4; \
+	vpand		x1,   x2, x2; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x4,   x3, x3; \
+	vpxor		x0,   x4, x4;
+
+#define SI3_1(x0, x1, x2, x3, x4)     \
+	vpxor		x1,   x2, x2; \
+	vpand		x2,   x1, tp; \
+	vpxor		x0,   tp, tp; \
+	vpor		x1,   x0, x0; \
+	vpxor		x3,   x1, x4; \
+	vpxor		x3,   x0, x0; \
+	vpor		tp,   x3, x3; \
+	vpxor		x2,   tp, x1;
+#define SI3_2(x0, x1, x2, x3, x4)     \
+	vpxor		x3,   x1, x1; \
+	vpxor		x2,   x0, x0; \
+	vpxor		x3,   x2, x2; \
+	vpand		x1,   x3, x3; \
+	vpxor		x0,   x1, x1; \
+	vpand		x2,   x0, x0; \
+	vpxor		x3,   x4, x4; \
+	vpxor		x0,   x3, x3; \
+	vpxor		x1,   x0, x0;
+
+#define SI4_1(x0, x1, x2, x3, x4)     \
+	vpxor		x3,   x2, x2; \
+	vpand		x1,   x0, tp; \
+	vpxor		x2,   tp, tp; \
+	vpor		x3,   x2, x2; \
+	vpxor		RNOT, x0, x4; \
+	vpxor		tp,   x1, x1; \
+	vpxor		x2,   tp, x0; \
+	vpand		x4,   x2, x2;
+#define SI4_2(x0, x1, x2, x3, x4)     \
+	vpxor		x0,   x2, x2; \
+	vpor		x4,   x0, x0; \
+	vpxor		x3,   x0, x0; \
+	vpand		x2,   x3, x3; \
+	vpxor		x3,   x4, x4; \
+	vpxor		x1,   x3, x3; \
+	vpand		x0,   x1, x1; \
+	vpxor		x1,   x4, x4; \
+	vpxor		x3,   x0, x0;
+
+#define SI5_1(x0, x1, x2, x3, x4)     \
+	vpor		x2,   x1, tp; \
+	vpxor		x1,   x2, x2; \
+	vpxor		x3,   tp, tp; \
+	vpand		x1,   x3, x3; \
+	vpxor		x3,   x2, x2; \
+	vpor		x0,   x3, x3; \
+	vpxor		RNOT, x0, x0; \
+	vpxor		x2,   x3, x3; \
+	vpor		x0,   x2, x2;
+#define SI5_2(x0, x1, x2, x3, x4)     \
+	vpxor		tp,   x1, x4; \
+	vpxor		x4,   x2, x2; \
+	vpand		x0,   x4, x4; \
+	vpxor		tp,   x0, x0; \
+	vpxor		x3,   tp, x1; \
+	vpand		x2,   x0, x0; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x2,   x0, x0; \
+	vpxor		x4,   x2, x2; \
+	vpxor		x3,   x4, x4;
+
+#define SI6_1(x0, x1, x2, x3, x4)     \
+	vpxor		x2,   x0, x0; \
+	vpand		x3,   x0, tp; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x2,   tp, tp; \
+	vpxor		x1,   x3, x3; \
+	vpor		x0,   x2, x2; \
+	vpxor		x3,   x2, x2; \
+	vpand		tp,   x3, x3;
+#define SI6_2(x0, x1, x2, x3, x4)     \
+	vpxor		RNOT, tp, tp; \
+	vpxor		x1,   x3, x3; \
+	vpand		x2,   x1, x1; \
+	vpxor		tp,   x0, x4; \
+	vpxor		x4,   x3, x3; \
+	vpxor		x2,   x4, x4; \
+	vpxor		x1,   tp, x0; \
+	vpxor		x0,   x2, x2;
+
+#define SI7_1(x0, x1, x2, x3, x4)     \
+	vpand		x0,   x3, tp; \
+	vpxor		x2,   x0, x0; \
+	vpor		x3,   x2, x2; \
+	vpxor		x1,   x3, x4; \
+	vpxor		RNOT, x0, x0; \
+	vpor		tp,   x1, x1; \
+	vpxor		x0,   x4, x4; \
+	vpand		x2,   x0, x0; \
+	vpxor		x1,   x0, x0;
+#define SI7_2(x0, x1, x2, x3, x4)     \
+	vpand		x2,   x1, x1; \
+	vpxor		x2,   tp, x3; \
+	vpxor		x3,   x4, x4; \
+	vpand		x3,   x2, x2; \
+	vpor		x0,   x3, x3; \
+	vpxor		x4,   x1, x1; \
+	vpxor		x4,   x3, x3; \
+	vpand		x0,   x4, x4; \
+	vpxor		x2,   x4, x4;
+
+#define get_key(i, j, t) \
+	vbroadcastss (4*(i)+(j))*4(CTX), t;
+
+#define K2(x0, x1, x2, x3, x4, i) \
+	get_key(i, 0, RK0); \
+	get_key(i, 1, RK1); \
+	get_key(i, 2, RK2); \
+	get_key(i, 3, RK3); \
+	vpxor RK0,	x0 ## 1, x0 ## 1; \
+	vpxor RK1,	x1 ## 1, x1 ## 1; \
+	vpxor RK2,	x2 ## 1, x2 ## 1; \
+	vpxor RK3,	x3 ## 1, x3 ## 1; \
+		vpxor RK0,	x0 ## 2, x0 ## 2; \
+		vpxor RK1,	x1 ## 2, x1 ## 2; \
+		vpxor RK2,	x2 ## 2, x2 ## 2; \
+		vpxor RK3,	x3 ## 2, x3 ## 2;
+
+#define LK2(x0, x1, x2, x3, x4, i) \
+	vpslld $13,		x0 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 13),	x0 ## 1, x0 ## 1;          \
+	vpor			x4 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			x0 ## 1, x1 ## 1, x1 ## 1; \
+	vpslld $3,		x2 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 3),	x2 ## 1, x2 ## 1;          \
+	vpor			x4 ## 1, x2 ## 1, x2 ## 1; \
+	vpxor			x2 ## 1, x1 ## 1, x1 ## 1; \
+		vpslld $13,		x0 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 13),	x0 ## 2, x0 ## 2;          \
+		vpor			x4 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			x0 ## 2, x1 ## 2, x1 ## 2; \
+		vpslld $3,		x2 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 3),	x2 ## 2, x2 ## 2;          \
+		vpor			x4 ## 2, x2 ## 2, x2 ## 2; \
+		vpxor			x2 ## 2, x1 ## 2, x1 ## 2; \
+	vpslld $1,		x1 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 1),	x1 ## 1, x1 ## 1;          \
+	vpor			x4 ## 1, x1 ## 1, x1 ## 1; \
+	vpslld $3,		x0 ## 1, x4 ## 1;          \
+	vpxor			x2 ## 1, x3 ## 1, x3 ## 1; \
+	vpxor			x4 ## 1, x3 ## 1, x3 ## 1; \
+	get_key(i, 1, RK1); \
+		vpslld $1,		x1 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 1),	x1 ## 2, x1 ## 2;          \
+		vpor			x4 ## 2, x1 ## 2, x1 ## 2; \
+		vpslld $3,		x0 ## 2, x4 ## 2;          \
+		vpxor			x2 ## 2, x3 ## 2, x3 ## 2; \
+		vpxor			x4 ## 2, x3 ## 2, x3 ## 2; \
+		get_key(i, 3, RK3); \
+	vpslld $7,		x3 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 7),	x3 ## 1, x3 ## 1;          \
+	vpor			x4 ## 1, x3 ## 1, x3 ## 1; \
+	vpslld $7,		x1 ## 1, x4 ## 1;          \
+	vpxor			x1 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			x3 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			x3 ## 1, x2 ## 1, x2 ## 1; \
+	vpxor			x4 ## 1, x2 ## 1, x2 ## 1; \
+	get_key(i, 0, RK0); \
+		vpslld $7,		x3 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 7),	x3 ## 2, x3 ## 2;          \
+		vpor			x4 ## 2, x3 ## 2, x3 ## 2; \
+		vpslld $7,		x1 ## 2, x4 ## 2;          \
+		vpxor			x1 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			x3 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			x3 ## 2, x2 ## 2, x2 ## 2; \
+		vpxor			x4 ## 2, x2 ## 2, x2 ## 2; \
+		get_key(i, 2, RK2); \
+	vpxor			RK1, x1 ## 1, x1 ## 1;     \
+	vpxor			RK3, x3 ## 1, x3 ## 1;     \
+	vpslld $5,		x0 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 5),	x0 ## 1, x0 ## 1;          \
+	vpor			x4 ## 1, x0 ## 1, x0 ## 1; \
+	vpslld $22,		x2 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 22),	x2 ## 1, x2 ## 1;          \
+	vpor			x4 ## 1, x2 ## 1, x2 ## 1; \
+	vpxor			RK0, x0 ## 1, x0 ## 1;     \
+	vpxor			RK2, x2 ## 1, x2 ## 1;     \
+		vpxor			RK1, x1 ## 2, x1 ## 2;     \
+		vpxor			RK3, x3 ## 2, x3 ## 2;     \
+		vpslld $5,		x0 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 5),	x0 ## 2, x0 ## 2;          \
+		vpor			x4 ## 2, x0 ## 2, x0 ## 2; \
+		vpslld $22,		x2 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 22),	x2 ## 2, x2 ## 2;          \
+		vpor			x4 ## 2, x2 ## 2, x2 ## 2; \
+		vpxor			RK0, x0 ## 2, x0 ## 2;     \
+		vpxor			RK2, x2 ## 2, x2 ## 2;
+
+#define KL2(x0, x1, x2, x3, x4, i) \
+	vpxor			RK0, x0 ## 1, x0 ## 1;     \
+	vpxor			RK2, x2 ## 1, x2 ## 1;     \
+	vpsrld $5,		x0 ## 1, x4 ## 1;          \
+	vpslld $(32 - 5),	x0 ## 1, x0 ## 1;          \
+	vpor			x4 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			RK3, x3 ## 1, x3 ## 1;     \
+	vpxor			RK1, x1 ## 1, x1 ## 1;     \
+	vpsrld $22,		x2 ## 1, x4 ## 1;          \
+	vpslld $(32 - 22),	x2 ## 1, x2 ## 1;          \
+	vpor			x4 ## 1, x2 ## 1, x2 ## 1; \
+	vpxor			x3 ## 1, x2 ## 1, x2 ## 1; \
+		vpxor			RK0, x0 ## 2, x0 ## 2;     \
+		vpxor			RK2, x2 ## 2, x2 ## 2;     \
+		vpsrld $5,		x0 ## 2, x4 ## 2;          \
+		vpslld $(32 - 5),	x0 ## 2, x0 ## 2;          \
+		vpor			x4 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			RK3, x3 ## 2, x3 ## 2;     \
+		vpxor			RK1, x1 ## 2, x1 ## 2;     \
+		vpsrld $22,		x2 ## 2, x4 ## 2;          \
+		vpslld $(32 - 22),	x2 ## 2, x2 ## 2;          \
+		vpor			x4 ## 2, x2 ## 2, x2 ## 2; \
+		vpxor			x3 ## 2, x2 ## 2, x2 ## 2; \
+	vpxor			x3 ## 1, x0 ## 1, x0 ## 1; \
+	vpslld $7,		x1 ## 1, x4 ## 1;          \
+	vpxor			x1 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			x4 ## 1, x2 ## 1, x2 ## 1; \
+	vpsrld $1,		x1 ## 1, x4 ## 1;          \
+	vpslld $(32 - 1),	x1 ## 1, x1 ## 1;          \
+	vpor			x4 ## 1, x1 ## 1, x1 ## 1; \
+		vpxor			x3 ## 2, x0 ## 2, x0 ## 2; \
+		vpslld $7,		x1 ## 2, x4 ## 2;          \
+		vpxor			x1 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			x4 ## 2, x2 ## 2, x2 ## 2; \
+		vpsrld $1,		x1 ## 2, x4 ## 2;          \
+		vpslld $(32 - 1),	x1 ## 2, x1 ## 2;          \
+		vpor			x4 ## 2, x1 ## 2, x1 ## 2; \
+	vpsrld $7,		x3 ## 1, x4 ## 1;          \
+	vpslld $(32 - 7),	x3 ## 1, x3 ## 1;          \
+	vpor			x4 ## 1, x3 ## 1, x3 ## 1; \
+	vpxor			x0 ## 1, x1 ## 1, x1 ## 1; \
+	vpslld $3,		x0 ## 1, x4 ## 1;          \
+	vpxor			x4 ## 1, x3 ## 1, x3 ## 1; \
+		vpsrld $7,		x3 ## 2, x4 ## 2;          \
+		vpslld $(32 - 7),	x3 ## 2, x3 ## 2;          \
+		vpor			x4 ## 2, x3 ## 2, x3 ## 2; \
+		vpxor			x0 ## 2, x1 ## 2, x1 ## 2; \
+		vpslld $3,		x0 ## 2, x4 ## 2;          \
+		vpxor			x4 ## 2, x3 ## 2, x3 ## 2; \
+	vpsrld $13,		x0 ## 1, x4 ## 1;          \
+	vpslld $(32 - 13),	x0 ## 1, x0 ## 1;          \
+	vpor			x4 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			x2 ## 1, x1 ## 1, x1 ## 1; \
+	vpxor			x2 ## 1, x3 ## 1, x3 ## 1; \
+	vpsrld $3,		x2 ## 1, x4 ## 1;          \
+	vpslld $(32 - 3),	x2 ## 1, x2 ## 1;          \
+	vpor			x4 ## 1, x2 ## 1, x2 ## 1; \
+		vpsrld $13,		x0 ## 2, x4 ## 2;          \
+		vpslld $(32 - 13),	x0 ## 2, x0 ## 2;          \
+		vpor			x4 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			x2 ## 2, x1 ## 2, x1 ## 2; \
+		vpxor			x2 ## 2, x3 ## 2, x3 ## 2; \
+		vpsrld $3,		x2 ## 2, x4 ## 2;          \
+		vpslld $(32 - 3),	x2 ## 2, x2 ## 2;          \
+		vpor			x4 ## 2, x2 ## 2, x2 ## 2;
+
+#define S(SBOX, x0, x1, x2, x3, x4) \
+	SBOX ## _1(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	SBOX ## _2(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	SBOX ## _1(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2); \
+	SBOX ## _2(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2);
+
+#define SP(SBOX, x0, x1, x2, x3, x4, i) \
+	get_key(i, 0, RK0); \
+	SBOX ## _1(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	get_key(i, 2, RK2); \
+	SBOX ## _2(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	get_key(i, 3, RK3); \
+	SBOX ## _1(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2); \
+	get_key(i, 1, RK1); \
+	SBOX ## _2(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2); \
+
+#define transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	vpunpckldq		x1, x0, t0; \
+	vpunpckhdq		x1, x0, t2; \
+	vpunpckldq		x3, x2, t1; \
+	vpunpckhdq		x3, x2, x3; \
+	\
+	vpunpcklqdq		t1, t0, x0; \
+	vpunpckhqdq		t1, t0, x1; \
+	vpunpcklqdq		x3, t2, x2; \
+	vpunpckhqdq		x3, t2, x3;
+
+#define read_blocks(x0, x1, x2, x3, t0, t1, t2) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2)
+
+#define write_blocks(x0, x1, x2, x3, t0, t1, t2) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2)
+
+SYM_FUNC_START_LOCAL(__serpent_enc_blk8_avx)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: blocks
+	 * output:
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: encrypted blocks
+	 */
+
+	vpcmpeqd RNOT, RNOT, RNOT;
+
+	read_blocks(RA1, RB1, RC1, RD1, RK0, RK1, RK2);
+	read_blocks(RA2, RB2, RC2, RD2, RK0, RK1, RK2);
+
+						 K2(RA, RB, RC, RD, RE, 0);
+	S(S0, RA, RB, RC, RD, RE);		LK2(RC, RB, RD, RA, RE, 1);
+	S(S1, RC, RB, RD, RA, RE);		LK2(RE, RD, RA, RC, RB, 2);
+	S(S2, RE, RD, RA, RC, RB);		LK2(RB, RD, RE, RC, RA, 3);
+	S(S3, RB, RD, RE, RC, RA);		LK2(RC, RA, RD, RB, RE, 4);
+	S(S4, RC, RA, RD, RB, RE);		LK2(RA, RD, RB, RE, RC, 5);
+	S(S5, RA, RD, RB, RE, RC);		LK2(RC, RA, RD, RE, RB, 6);
+	S(S6, RC, RA, RD, RE, RB);		LK2(RD, RB, RA, RE, RC, 7);
+	S(S7, RD, RB, RA, RE, RC);		LK2(RC, RA, RE, RD, RB, 8);
+	S(S0, RC, RA, RE, RD, RB);		LK2(RE, RA, RD, RC, RB, 9);
+	S(S1, RE, RA, RD, RC, RB);		LK2(RB, RD, RC, RE, RA, 10);
+	S(S2, RB, RD, RC, RE, RA);		LK2(RA, RD, RB, RE, RC, 11);
+	S(S3, RA, RD, RB, RE, RC);		LK2(RE, RC, RD, RA, RB, 12);
+	S(S4, RE, RC, RD, RA, RB);		LK2(RC, RD, RA, RB, RE, 13);
+	S(S5, RC, RD, RA, RB, RE);		LK2(RE, RC, RD, RB, RA, 14);
+	S(S6, RE, RC, RD, RB, RA);		LK2(RD, RA, RC, RB, RE, 15);
+	S(S7, RD, RA, RC, RB, RE);		LK2(RE, RC, RB, RD, RA, 16);
+	S(S0, RE, RC, RB, RD, RA);		LK2(RB, RC, RD, RE, RA, 17);
+	S(S1, RB, RC, RD, RE, RA);		LK2(RA, RD, RE, RB, RC, 18);
+	S(S2, RA, RD, RE, RB, RC);		LK2(RC, RD, RA, RB, RE, 19);
+	S(S3, RC, RD, RA, RB, RE);		LK2(RB, RE, RD, RC, RA, 20);
+	S(S4, RB, RE, RD, RC, RA);		LK2(RE, RD, RC, RA, RB, 21);
+	S(S5, RE, RD, RC, RA, RB);		LK2(RB, RE, RD, RA, RC, 22);
+	S(S6, RB, RE, RD, RA, RC);		LK2(RD, RC, RE, RA, RB, 23);
+	S(S7, RD, RC, RE, RA, RB);		LK2(RB, RE, RA, RD, RC, 24);
+	S(S0, RB, RE, RA, RD, RC);		LK2(RA, RE, RD, RB, RC, 25);
+	S(S1, RA, RE, RD, RB, RC);		LK2(RC, RD, RB, RA, RE, 26);
+	S(S2, RC, RD, RB, RA, RE);		LK2(RE, RD, RC, RA, RB, 27);
+	S(S3, RE, RD, RC, RA, RB);		LK2(RA, RB, RD, RE, RC, 28);
+	S(S4, RA, RB, RD, RE, RC);		LK2(RB, RD, RE, RC, RA, 29);
+	S(S5, RB, RD, RE, RC, RA);		LK2(RA, RB, RD, RC, RE, 30);
+	S(S6, RA, RB, RD, RC, RE);		LK2(RD, RE, RB, RC, RA, 31);
+	S(S7, RD, RE, RB, RC, RA);		 K2(RA, RB, RC, RD, RE, 32);
+
+	write_blocks(RA1, RB1, RC1, RD1, RK0, RK1, RK2);
+	write_blocks(RA2, RB2, RC2, RD2, RK0, RK1, RK2);
+
+	RET;
+SYM_FUNC_END(__serpent_enc_blk8_avx)
+
+SYM_FUNC_START_LOCAL(__serpent_dec_blk8_avx)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: encrypted blocks
+	 * output:
+	 *	RC1, RD1, RB1, RE1, RC2, RD2, RB2, RE2: decrypted blocks
+	 */
+
+	vpcmpeqd RNOT, RNOT, RNOT;
+
+	read_blocks(RA1, RB1, RC1, RD1, RK0, RK1, RK2);
+	read_blocks(RA2, RB2, RC2, RD2, RK0, RK1, RK2);
+
+						 K2(RA, RB, RC, RD, RE, 32);
+	SP(SI7, RA, RB, RC, RD, RE, 31);	KL2(RB, RD, RA, RE, RC, 31);
+	SP(SI6, RB, RD, RA, RE, RC, 30);	KL2(RA, RC, RE, RB, RD, 30);
+	SP(SI5, RA, RC, RE, RB, RD, 29);	KL2(RC, RD, RA, RE, RB, 29);
+	SP(SI4, RC, RD, RA, RE, RB, 28);	KL2(RC, RA, RB, RE, RD, 28);
+	SP(SI3, RC, RA, RB, RE, RD, 27);	KL2(RB, RC, RD, RE, RA, 27);
+	SP(SI2, RB, RC, RD, RE, RA, 26);	KL2(RC, RA, RE, RD, RB, 26);
+	SP(SI1, RC, RA, RE, RD, RB, 25);	KL2(RB, RA, RE, RD, RC, 25);
+	SP(SI0, RB, RA, RE, RD, RC, 24);	KL2(RE, RC, RA, RB, RD, 24);
+	SP(SI7, RE, RC, RA, RB, RD, 23);	KL2(RC, RB, RE, RD, RA, 23);
+	SP(SI6, RC, RB, RE, RD, RA, 22);	KL2(RE, RA, RD, RC, RB, 22);
+	SP(SI5, RE, RA, RD, RC, RB, 21);	KL2(RA, RB, RE, RD, RC, 21);
+	SP(SI4, RA, RB, RE, RD, RC, 20);	KL2(RA, RE, RC, RD, RB, 20);
+	SP(SI3, RA, RE, RC, RD, RB, 19);	KL2(RC, RA, RB, RD, RE, 19);
+	SP(SI2, RC, RA, RB, RD, RE, 18);	KL2(RA, RE, RD, RB, RC, 18);
+	SP(SI1, RA, RE, RD, RB, RC, 17);	KL2(RC, RE, RD, RB, RA, 17);
+	SP(SI0, RC, RE, RD, RB, RA, 16);	KL2(RD, RA, RE, RC, RB, 16);
+	SP(SI7, RD, RA, RE, RC, RB, 15);	KL2(RA, RC, RD, RB, RE, 15);
+	SP(SI6, RA, RC, RD, RB, RE, 14);	KL2(RD, RE, RB, RA, RC, 14);
+	SP(SI5, RD, RE, RB, RA, RC, 13);	KL2(RE, RC, RD, RB, RA, 13);
+	SP(SI4, RE, RC, RD, RB, RA, 12);	KL2(RE, RD, RA, RB, RC, 12);
+	SP(SI3, RE, RD, RA, RB, RC, 11);	KL2(RA, RE, RC, RB, RD, 11);
+	SP(SI2, RA, RE, RC, RB, RD, 10);	KL2(RE, RD, RB, RC, RA, 10);
+	SP(SI1, RE, RD, RB, RC, RA, 9);		KL2(RA, RD, RB, RC, RE, 9);
+	SP(SI0, RA, RD, RB, RC, RE, 8);		KL2(RB, RE, RD, RA, RC, 8);
+	SP(SI7, RB, RE, RD, RA, RC, 7);		KL2(RE, RA, RB, RC, RD, 7);
+	SP(SI6, RE, RA, RB, RC, RD, 6);		KL2(RB, RD, RC, RE, RA, 6);
+	SP(SI5, RB, RD, RC, RE, RA, 5);		KL2(RD, RA, RB, RC, RE, 5);
+	SP(SI4, RD, RA, RB, RC, RE, 4);		KL2(RD, RB, RE, RC, RA, 4);
+	SP(SI3, RD, RB, RE, RC, RA, 3);		KL2(RE, RD, RA, RC, RB, 3);
+	SP(SI2, RE, RD, RA, RC, RB, 2);		KL2(RD, RB, RC, RA, RE, 2);
+	SP(SI1, RD, RB, RC, RA, RE, 1);		KL2(RE, RB, RC, RA, RD, 1);
+	S(SI0, RE, RB, RC, RA, RD);		 K2(RC, RD, RB, RE, RA, 0);
+
+	write_blocks(RC1, RD1, RB1, RE1, RK0, RK1, RK2);
+	write_blocks(RC2, RD2, RB2, RE2, RK0, RK1, RK2);
+
+	RET;
+SYM_FUNC_END(__serpent_dec_blk8_avx)
+
+SYM_TYPED_FUNC_START(serpent_ecb_enc_8way_avx)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+
+	load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	call __serpent_enc_blk8_avx;
+
+	store_8way(%rsi, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(serpent_ecb_enc_8way_avx)
+
+SYM_TYPED_FUNC_START(serpent_ecb_dec_8way_avx)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+
+	load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	call __serpent_dec_blk8_avx;
+
+	store_8way(%rsi, RC1, RD1, RB1, RE1, RC2, RD2, RB2, RE2);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(serpent_ecb_dec_8way_avx)
+
+SYM_TYPED_FUNC_START(serpent_cbc_dec_8way_avx)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+
+	load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	call __serpent_dec_blk8_avx;
+
+	store_cbc_8way(%rdx, %rsi, RC1, RD1, RB1, RE1, RC2, RD2, RB2, RE2);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(serpent_cbc_dec_8way_avx)
diff --git a/arch/x86/crypto/serpent-avx.h b/arch/x86/crypto/serpent-avx.h
new file mode 100644
index 000000000..23f3361a0
--- /dev/null
+++ b/arch/x86/crypto/serpent-avx.h
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef ASM_X86_SERPENT_AVX_H
+#define ASM_X86_SERPENT_AVX_H
+
+#include <crypto/b128ops.h>
+#include <crypto/serpent.h>
+#include <linux/types.h>
+
+struct crypto_skcipher;
+
+#define SERPENT_PARALLEL_BLOCKS 8
+
+asmlinkage void serpent_ecb_enc_8way_avx(const void *ctx, u8 *dst,
+					 const u8 *src);
+asmlinkage void serpent_ecb_dec_8way_avx(const void *ctx, u8 *dst,
+					 const u8 *src);
+
+asmlinkage void serpent_cbc_dec_8way_avx(const void *ctx, u8 *dst,
+					 const u8 *src);
+
+#endif
diff --git a/arch/x86/crypto/serpent-avx2-asm_64.S b/arch/x86/crypto/serpent-avx2-asm_64.S
new file mode 100644
index 000000000..6d60c5059
--- /dev/null
+++ b/arch/x86/crypto/serpent-avx2-asm_64.S
@@ -0,0 +1,724 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * x86_64/AVX2 assembler optimized version of Serpent
+ *
+ * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ *
+ * Based on AVX assembler implementation of Serpent by:
+ *  Copyright © 2012 Johannes Goetzfried
+ *      <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
+ */
+
+#include <linux/linkage.h>
+#include <asm/frame.h>
+#include "glue_helper-asm-avx2.S"
+
+.file "serpent-avx2-asm_64.S"
+
+.section	.rodata.cst16.bswap128_mask, "aM", @progbits, 16
+.align 16
+.Lbswap128_mask:
+	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+
+.text
+
+#define CTX %rdi
+
+#define RNOT %ymm0
+#define tp  %ymm1
+
+#define RA1 %ymm2
+#define RA2 %ymm3
+#define RB1 %ymm4
+#define RB2 %ymm5
+#define RC1 %ymm6
+#define RC2 %ymm7
+#define RD1 %ymm8
+#define RD2 %ymm9
+#define RE1 %ymm10
+#define RE2 %ymm11
+
+#define RK0 %ymm12
+#define RK1 %ymm13
+#define RK2 %ymm14
+#define RK3 %ymm15
+
+#define RK0x %xmm12
+#define RK1x %xmm13
+#define RK2x %xmm14
+#define RK3x %xmm15
+
+#define S0_1(x0, x1, x2, x3, x4)      \
+	vpor		x0,   x3, tp; \
+	vpxor		x3,   x0, x0; \
+	vpxor		x2,   x3, x4; \
+	vpxor		RNOT, x4, x4; \
+	vpxor		x1,   tp, x3; \
+	vpand		x0,   x1, x1; \
+	vpxor		x4,   x1, x1; \
+	vpxor		x0,   x2, x2;
+#define S0_2(x0, x1, x2, x3, x4)      \
+	vpxor		x3,   x0, x0; \
+	vpor		x0,   x4, x4; \
+	vpxor		x2,   x0, x0; \
+	vpand		x1,   x2, x2; \
+	vpxor		x2,   x3, x3; \
+	vpxor		RNOT, x1, x1; \
+	vpxor		x4,   x2, x2; \
+	vpxor		x2,   x1, x1;
+
+#define S1_1(x0, x1, x2, x3, x4)      \
+	vpxor		x0,   x1, tp; \
+	vpxor		x3,   x0, x0; \
+	vpxor		RNOT, x3, x3; \
+	vpand		tp,   x1, x4; \
+	vpor		tp,   x0, x0; \
+	vpxor		x2,   x3, x3; \
+	vpxor		x3,   x0, x0; \
+	vpxor		x3,   tp, x1;
+#define S1_2(x0, x1, x2, x3, x4)      \
+	vpxor		x4,   x3, x3; \
+	vpor		x4,   x1, x1; \
+	vpxor		x2,   x4, x4; \
+	vpand		x0,   x2, x2; \
+	vpxor		x1,   x2, x2; \
+	vpor		x0,   x1, x1; \
+	vpxor		RNOT, x0, x0; \
+	vpxor		x2,   x0, x0; \
+	vpxor		x1,   x4, x4;
+
+#define S2_1(x0, x1, x2, x3, x4)      \
+	vpxor		RNOT, x3, x3; \
+	vpxor		x0,   x1, x1; \
+	vpand		x2,   x0, tp; \
+	vpxor		x3,   tp, tp; \
+	vpor		x0,   x3, x3; \
+	vpxor		x1,   x2, x2; \
+	vpxor		x1,   x3, x3; \
+	vpand		tp,   x1, x1;
+#define S2_2(x0, x1, x2, x3, x4)      \
+	vpxor		x2,   tp, tp; \
+	vpand		x3,   x2, x2; \
+	vpor		x1,   x3, x3; \
+	vpxor		RNOT, tp, tp; \
+	vpxor		tp,   x3, x3; \
+	vpxor		tp,   x0, x4; \
+	vpxor		x2,   tp, x0; \
+	vpor		x2,   x1, x1;
+
+#define S3_1(x0, x1, x2, x3, x4)      \
+	vpxor		x3,   x1, tp; \
+	vpor		x0,   x3, x3; \
+	vpand		x0,   x1, x4; \
+	vpxor		x2,   x0, x0; \
+	vpxor		tp,   x2, x2; \
+	vpand		x3,   tp, x1; \
+	vpxor		x3,   x2, x2; \
+	vpor		x4,   x0, x0; \
+	vpxor		x3,   x4, x4;
+#define S3_2(x0, x1, x2, x3, x4)      \
+	vpxor		x0,   x1, x1; \
+	vpand		x3,   x0, x0; \
+	vpand		x4,   x3, x3; \
+	vpxor		x2,   x3, x3; \
+	vpor		x1,   x4, x4; \
+	vpand		x1,   x2, x2; \
+	vpxor		x3,   x4, x4; \
+	vpxor		x3,   x0, x0; \
+	vpxor		x2,   x3, x3;
+
+#define S4_1(x0, x1, x2, x3, x4)      \
+	vpand		x0,   x3, tp; \
+	vpxor		x3,   x0, x0; \
+	vpxor		x2,   tp, tp; \
+	vpor		x3,   x2, x2; \
+	vpxor		x1,   x0, x0; \
+	vpxor		tp,   x3, x4; \
+	vpor		x0,   x2, x2; \
+	vpxor		x1,   x2, x2;
+#define S4_2(x0, x1, x2, x3, x4)      \
+	vpand		x0,   x1, x1; \
+	vpxor		x4,   x1, x1; \
+	vpand		x2,   x4, x4; \
+	vpxor		tp,   x2, x2; \
+	vpxor		x0,   x4, x4; \
+	vpor		x1,   tp, x3; \
+	vpxor		RNOT, x1, x1; \
+	vpxor		x0,   x3, x3;
+
+#define S5_1(x0, x1, x2, x3, x4)      \
+	vpor		x0,   x1, tp; \
+	vpxor		tp,   x2, x2; \
+	vpxor		RNOT, x3, x3; \
+	vpxor		x0,   x1, x4; \
+	vpxor		x2,   x0, x0; \
+	vpand		x4,   tp, x1; \
+	vpor		x3,   x4, x4; \
+	vpxor		x0,   x4, x4;
+#define S5_2(x0, x1, x2, x3, x4)      \
+	vpand		x3,   x0, x0; \
+	vpxor		x3,   x1, x1; \
+	vpxor		x2,   x3, x3; \
+	vpxor		x1,   x0, x0; \
+	vpand		x4,   x2, x2; \
+	vpxor		x2,   x1, x1; \
+	vpand		x0,   x2, x2; \
+	vpxor		x2,   x3, x3;
+
+#define S6_1(x0, x1, x2, x3, x4)      \
+	vpxor		x0,   x3, x3; \
+	vpxor		x2,   x1, tp; \
+	vpxor		x0,   x2, x2; \
+	vpand		x3,   x0, x0; \
+	vpor		x3,   tp, tp; \
+	vpxor		RNOT, x1, x4; \
+	vpxor		tp,   x0, x0; \
+	vpxor		x2,   tp, x1;
+#define S6_2(x0, x1, x2, x3, x4)      \
+	vpxor		x4,   x3, x3; \
+	vpxor		x0,   x4, x4; \
+	vpand		x0,   x2, x2; \
+	vpxor		x1,   x4, x4; \
+	vpxor		x3,   x2, x2; \
+	vpand		x1,   x3, x3; \
+	vpxor		x0,   x3, x3; \
+	vpxor		x2,   x1, x1;
+
+#define S7_1(x0, x1, x2, x3, x4)      \
+	vpxor		RNOT, x1, tp; \
+	vpxor		RNOT, x0, x0; \
+	vpand		x2,   tp, x1; \
+	vpxor		x3,   x1, x1; \
+	vpor		tp,   x3, x3; \
+	vpxor		x2,   tp, x4; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x0,   x3, x3; \
+	vpor		x1,   x0, x0;
+#define S7_2(x0, x1, x2, x3, x4)      \
+	vpand		x0,   x2, x2; \
+	vpxor		x4,   x0, x0; \
+	vpxor		x3,   x4, x4; \
+	vpand		x0,   x3, x3; \
+	vpxor		x1,   x4, x4; \
+	vpxor		x4,   x2, x2; \
+	vpxor		x1,   x3, x3; \
+	vpor		x0,   x4, x4; \
+	vpxor		x1,   x4, x4;
+
+#define SI0_1(x0, x1, x2, x3, x4)     \
+	vpxor		x0,   x1, x1; \
+	vpor		x1,   x3, tp; \
+	vpxor		x1,   x3, x4; \
+	vpxor		RNOT, x0, x0; \
+	vpxor		tp,   x2, x2; \
+	vpxor		x0,   tp, x3; \
+	vpand		x1,   x0, x0; \
+	vpxor		x2,   x0, x0;
+#define SI0_2(x0, x1, x2, x3, x4)     \
+	vpand		x3,   x2, x2; \
+	vpxor		x4,   x3, x3; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x3,   x1, x1; \
+	vpand		x0,   x3, x3; \
+	vpxor		x0,   x1, x1; \
+	vpxor		x2,   x0, x0; \
+	vpxor		x3,   x4, x4;
+
+#define SI1_1(x0, x1, x2, x3, x4)     \
+	vpxor		x3,   x1, x1; \
+	vpxor		x2,   x0, tp; \
+	vpxor		RNOT, x2, x2; \
+	vpor		x1,   x0, x4; \
+	vpxor		x3,   x4, x4; \
+	vpand		x1,   x3, x3; \
+	vpxor		x2,   x1, x1; \
+	vpand		x4,   x2, x2;
+#define SI1_2(x0, x1, x2, x3, x4)     \
+	vpxor		x1,   x4, x4; \
+	vpor		x3,   x1, x1; \
+	vpxor		tp,   x3, x3; \
+	vpxor		tp,   x2, x2; \
+	vpor		x4,   tp, x0; \
+	vpxor		x4,   x2, x2; \
+	vpxor		x0,   x1, x1; \
+	vpxor		x1,   x4, x4;
+
+#define SI2_1(x0, x1, x2, x3, x4)     \
+	vpxor		x1,   x2, x2; \
+	vpxor		RNOT, x3, tp; \
+	vpor		x2,   tp, tp; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x0,   x3, x4; \
+	vpxor		x1,   tp, x3; \
+	vpor		x2,   x1, x1; \
+	vpxor		x0,   x2, x2;
+#define SI2_2(x0, x1, x2, x3, x4)     \
+	vpxor		x4,   x1, x1; \
+	vpor		x3,   x4, x4; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x2,   x4, x4; \
+	vpand		x1,   x2, x2; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x4,   x3, x3; \
+	vpxor		x0,   x4, x4;
+
+#define SI3_1(x0, x1, x2, x3, x4)     \
+	vpxor		x1,   x2, x2; \
+	vpand		x2,   x1, tp; \
+	vpxor		x0,   tp, tp; \
+	vpor		x1,   x0, x0; \
+	vpxor		x3,   x1, x4; \
+	vpxor		x3,   x0, x0; \
+	vpor		tp,   x3, x3; \
+	vpxor		x2,   tp, x1;
+#define SI3_2(x0, x1, x2, x3, x4)     \
+	vpxor		x3,   x1, x1; \
+	vpxor		x2,   x0, x0; \
+	vpxor		x3,   x2, x2; \
+	vpand		x1,   x3, x3; \
+	vpxor		x0,   x1, x1; \
+	vpand		x2,   x0, x0; \
+	vpxor		x3,   x4, x4; \
+	vpxor		x0,   x3, x3; \
+	vpxor		x1,   x0, x0;
+
+#define SI4_1(x0, x1, x2, x3, x4)     \
+	vpxor		x3,   x2, x2; \
+	vpand		x1,   x0, tp; \
+	vpxor		x2,   tp, tp; \
+	vpor		x3,   x2, x2; \
+	vpxor		RNOT, x0, x4; \
+	vpxor		tp,   x1, x1; \
+	vpxor		x2,   tp, x0; \
+	vpand		x4,   x2, x2;
+#define SI4_2(x0, x1, x2, x3, x4)     \
+	vpxor		x0,   x2, x2; \
+	vpor		x4,   x0, x0; \
+	vpxor		x3,   x0, x0; \
+	vpand		x2,   x3, x3; \
+	vpxor		x3,   x4, x4; \
+	vpxor		x1,   x3, x3; \
+	vpand		x0,   x1, x1; \
+	vpxor		x1,   x4, x4; \
+	vpxor		x3,   x0, x0;
+
+#define SI5_1(x0, x1, x2, x3, x4)     \
+	vpor		x2,   x1, tp; \
+	vpxor		x1,   x2, x2; \
+	vpxor		x3,   tp, tp; \
+	vpand		x1,   x3, x3; \
+	vpxor		x3,   x2, x2; \
+	vpor		x0,   x3, x3; \
+	vpxor		RNOT, x0, x0; \
+	vpxor		x2,   x3, x3; \
+	vpor		x0,   x2, x2;
+#define SI5_2(x0, x1, x2, x3, x4)     \
+	vpxor		tp,   x1, x4; \
+	vpxor		x4,   x2, x2; \
+	vpand		x0,   x4, x4; \
+	vpxor		tp,   x0, x0; \
+	vpxor		x3,   tp, x1; \
+	vpand		x2,   x0, x0; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x2,   x0, x0; \
+	vpxor		x4,   x2, x2; \
+	vpxor		x3,   x4, x4;
+
+#define SI6_1(x0, x1, x2, x3, x4)     \
+	vpxor		x2,   x0, x0; \
+	vpand		x3,   x0, tp; \
+	vpxor		x3,   x2, x2; \
+	vpxor		x2,   tp, tp; \
+	vpxor		x1,   x3, x3; \
+	vpor		x0,   x2, x2; \
+	vpxor		x3,   x2, x2; \
+	vpand		tp,   x3, x3;
+#define SI6_2(x0, x1, x2, x3, x4)     \
+	vpxor		RNOT, tp, tp; \
+	vpxor		x1,   x3, x3; \
+	vpand		x2,   x1, x1; \
+	vpxor		tp,   x0, x4; \
+	vpxor		x4,   x3, x3; \
+	vpxor		x2,   x4, x4; \
+	vpxor		x1,   tp, x0; \
+	vpxor		x0,   x2, x2;
+
+#define SI7_1(x0, x1, x2, x3, x4)     \
+	vpand		x0,   x3, tp; \
+	vpxor		x2,   x0, x0; \
+	vpor		x3,   x2, x2; \
+	vpxor		x1,   x3, x4; \
+	vpxor		RNOT, x0, x0; \
+	vpor		tp,   x1, x1; \
+	vpxor		x0,   x4, x4; \
+	vpand		x2,   x0, x0; \
+	vpxor		x1,   x0, x0;
+#define SI7_2(x0, x1, x2, x3, x4)     \
+	vpand		x2,   x1, x1; \
+	vpxor		x2,   tp, x3; \
+	vpxor		x3,   x4, x4; \
+	vpand		x3,   x2, x2; \
+	vpor		x0,   x3, x3; \
+	vpxor		x4,   x1, x1; \
+	vpxor		x4,   x3, x3; \
+	vpand		x0,   x4, x4; \
+	vpxor		x2,   x4, x4;
+
+#define get_key(i,j,t) \
+	vpbroadcastd (4*(i)+(j))*4(CTX), t;
+
+#define K2(x0, x1, x2, x3, x4, i) \
+	get_key(i, 0, RK0); \
+	get_key(i, 1, RK1); \
+	get_key(i, 2, RK2); \
+	get_key(i, 3, RK3); \
+	vpxor RK0,	x0 ## 1, x0 ## 1; \
+	vpxor RK1,	x1 ## 1, x1 ## 1; \
+	vpxor RK2,	x2 ## 1, x2 ## 1; \
+	vpxor RK3,	x3 ## 1, x3 ## 1; \
+		vpxor RK0,	x0 ## 2, x0 ## 2; \
+		vpxor RK1,	x1 ## 2, x1 ## 2; \
+		vpxor RK2,	x2 ## 2, x2 ## 2; \
+		vpxor RK3,	x3 ## 2, x3 ## 2;
+
+#define LK2(x0, x1, x2, x3, x4, i) \
+	vpslld $13,		x0 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 13),	x0 ## 1, x0 ## 1;          \
+	vpor			x4 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			x0 ## 1, x1 ## 1, x1 ## 1; \
+	vpslld $3,		x2 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 3),	x2 ## 1, x2 ## 1;          \
+	vpor			x4 ## 1, x2 ## 1, x2 ## 1; \
+	vpxor			x2 ## 1, x1 ## 1, x1 ## 1; \
+		vpslld $13,		x0 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 13),	x0 ## 2, x0 ## 2;          \
+		vpor			x4 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			x0 ## 2, x1 ## 2, x1 ## 2; \
+		vpslld $3,		x2 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 3),	x2 ## 2, x2 ## 2;          \
+		vpor			x4 ## 2, x2 ## 2, x2 ## 2; \
+		vpxor			x2 ## 2, x1 ## 2, x1 ## 2; \
+	vpslld $1,		x1 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 1),	x1 ## 1, x1 ## 1;          \
+	vpor			x4 ## 1, x1 ## 1, x1 ## 1; \
+	vpslld $3,		x0 ## 1, x4 ## 1;          \
+	vpxor			x2 ## 1, x3 ## 1, x3 ## 1; \
+	vpxor			x4 ## 1, x3 ## 1, x3 ## 1; \
+	get_key(i, 1, RK1); \
+		vpslld $1,		x1 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 1),	x1 ## 2, x1 ## 2;          \
+		vpor			x4 ## 2, x1 ## 2, x1 ## 2; \
+		vpslld $3,		x0 ## 2, x4 ## 2;          \
+		vpxor			x2 ## 2, x3 ## 2, x3 ## 2; \
+		vpxor			x4 ## 2, x3 ## 2, x3 ## 2; \
+		get_key(i, 3, RK3); \
+	vpslld $7,		x3 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 7),	x3 ## 1, x3 ## 1;          \
+	vpor			x4 ## 1, x3 ## 1, x3 ## 1; \
+	vpslld $7,		x1 ## 1, x4 ## 1;          \
+	vpxor			x1 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			x3 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			x3 ## 1, x2 ## 1, x2 ## 1; \
+	vpxor			x4 ## 1, x2 ## 1, x2 ## 1; \
+	get_key(i, 0, RK0); \
+		vpslld $7,		x3 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 7),	x3 ## 2, x3 ## 2;          \
+		vpor			x4 ## 2, x3 ## 2, x3 ## 2; \
+		vpslld $7,		x1 ## 2, x4 ## 2;          \
+		vpxor			x1 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			x3 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			x3 ## 2, x2 ## 2, x2 ## 2; \
+		vpxor			x4 ## 2, x2 ## 2, x2 ## 2; \
+		get_key(i, 2, RK2); \
+	vpxor			RK1, x1 ## 1, x1 ## 1;     \
+	vpxor			RK3, x3 ## 1, x3 ## 1;     \
+	vpslld $5,		x0 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 5),	x0 ## 1, x0 ## 1;          \
+	vpor			x4 ## 1, x0 ## 1, x0 ## 1; \
+	vpslld $22,		x2 ## 1, x4 ## 1;          \
+	vpsrld $(32 - 22),	x2 ## 1, x2 ## 1;          \
+	vpor			x4 ## 1, x2 ## 1, x2 ## 1; \
+	vpxor			RK0, x0 ## 1, x0 ## 1;     \
+	vpxor			RK2, x2 ## 1, x2 ## 1;     \
+		vpxor			RK1, x1 ## 2, x1 ## 2;     \
+		vpxor			RK3, x3 ## 2, x3 ## 2;     \
+		vpslld $5,		x0 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 5),	x0 ## 2, x0 ## 2;          \
+		vpor			x4 ## 2, x0 ## 2, x0 ## 2; \
+		vpslld $22,		x2 ## 2, x4 ## 2;          \
+		vpsrld $(32 - 22),	x2 ## 2, x2 ## 2;          \
+		vpor			x4 ## 2, x2 ## 2, x2 ## 2; \
+		vpxor			RK0, x0 ## 2, x0 ## 2;     \
+		vpxor			RK2, x2 ## 2, x2 ## 2;
+
+#define KL2(x0, x1, x2, x3, x4, i) \
+	vpxor			RK0, x0 ## 1, x0 ## 1;     \
+	vpxor			RK2, x2 ## 1, x2 ## 1;     \
+	vpsrld $5,		x0 ## 1, x4 ## 1;          \
+	vpslld $(32 - 5),	x0 ## 1, x0 ## 1;          \
+	vpor			x4 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			RK3, x3 ## 1, x3 ## 1;     \
+	vpxor			RK1, x1 ## 1, x1 ## 1;     \
+	vpsrld $22,		x2 ## 1, x4 ## 1;          \
+	vpslld $(32 - 22),	x2 ## 1, x2 ## 1;          \
+	vpor			x4 ## 1, x2 ## 1, x2 ## 1; \
+	vpxor			x3 ## 1, x2 ## 1, x2 ## 1; \
+		vpxor			RK0, x0 ## 2, x0 ## 2;     \
+		vpxor			RK2, x2 ## 2, x2 ## 2;     \
+		vpsrld $5,		x0 ## 2, x4 ## 2;          \
+		vpslld $(32 - 5),	x0 ## 2, x0 ## 2;          \
+		vpor			x4 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			RK3, x3 ## 2, x3 ## 2;     \
+		vpxor			RK1, x1 ## 2, x1 ## 2;     \
+		vpsrld $22,		x2 ## 2, x4 ## 2;          \
+		vpslld $(32 - 22),	x2 ## 2, x2 ## 2;          \
+		vpor			x4 ## 2, x2 ## 2, x2 ## 2; \
+		vpxor			x3 ## 2, x2 ## 2, x2 ## 2; \
+	vpxor			x3 ## 1, x0 ## 1, x0 ## 1; \
+	vpslld $7,		x1 ## 1, x4 ## 1;          \
+	vpxor			x1 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			x4 ## 1, x2 ## 1, x2 ## 1; \
+	vpsrld $1,		x1 ## 1, x4 ## 1;          \
+	vpslld $(32 - 1),	x1 ## 1, x1 ## 1;          \
+	vpor			x4 ## 1, x1 ## 1, x1 ## 1; \
+		vpxor			x3 ## 2, x0 ## 2, x0 ## 2; \
+		vpslld $7,		x1 ## 2, x4 ## 2;          \
+		vpxor			x1 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			x4 ## 2, x2 ## 2, x2 ## 2; \
+		vpsrld $1,		x1 ## 2, x4 ## 2;          \
+		vpslld $(32 - 1),	x1 ## 2, x1 ## 2;          \
+		vpor			x4 ## 2, x1 ## 2, x1 ## 2; \
+	vpsrld $7,		x3 ## 1, x4 ## 1;          \
+	vpslld $(32 - 7),	x3 ## 1, x3 ## 1;          \
+	vpor			x4 ## 1, x3 ## 1, x3 ## 1; \
+	vpxor			x0 ## 1, x1 ## 1, x1 ## 1; \
+	vpslld $3,		x0 ## 1, x4 ## 1;          \
+	vpxor			x4 ## 1, x3 ## 1, x3 ## 1; \
+		vpsrld $7,		x3 ## 2, x4 ## 2;          \
+		vpslld $(32 - 7),	x3 ## 2, x3 ## 2;          \
+		vpor			x4 ## 2, x3 ## 2, x3 ## 2; \
+		vpxor			x0 ## 2, x1 ## 2, x1 ## 2; \
+		vpslld $3,		x0 ## 2, x4 ## 2;          \
+		vpxor			x4 ## 2, x3 ## 2, x3 ## 2; \
+	vpsrld $13,		x0 ## 1, x4 ## 1;          \
+	vpslld $(32 - 13),	x0 ## 1, x0 ## 1;          \
+	vpor			x4 ## 1, x0 ## 1, x0 ## 1; \
+	vpxor			x2 ## 1, x1 ## 1, x1 ## 1; \
+	vpxor			x2 ## 1, x3 ## 1, x3 ## 1; \
+	vpsrld $3,		x2 ## 1, x4 ## 1;          \
+	vpslld $(32 - 3),	x2 ## 1, x2 ## 1;          \
+	vpor			x4 ## 1, x2 ## 1, x2 ## 1; \
+		vpsrld $13,		x0 ## 2, x4 ## 2;          \
+		vpslld $(32 - 13),	x0 ## 2, x0 ## 2;          \
+		vpor			x4 ## 2, x0 ## 2, x0 ## 2; \
+		vpxor			x2 ## 2, x1 ## 2, x1 ## 2; \
+		vpxor			x2 ## 2, x3 ## 2, x3 ## 2; \
+		vpsrld $3,		x2 ## 2, x4 ## 2;          \
+		vpslld $(32 - 3),	x2 ## 2, x2 ## 2;          \
+		vpor			x4 ## 2, x2 ## 2, x2 ## 2;
+
+#define S(SBOX, x0, x1, x2, x3, x4) \
+	SBOX ## _1(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	SBOX ## _2(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	SBOX ## _1(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2); \
+	SBOX ## _2(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2);
+
+#define SP(SBOX, x0, x1, x2, x3, x4, i) \
+	get_key(i, 0, RK0); \
+	SBOX ## _1(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	get_key(i, 2, RK2); \
+	SBOX ## _2(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	get_key(i, 3, RK3); \
+	SBOX ## _1(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2); \
+	get_key(i, 1, RK1); \
+	SBOX ## _2(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2); \
+
+#define transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	vpunpckldq		x1, x0, t0; \
+	vpunpckhdq		x1, x0, t2; \
+	vpunpckldq		x3, x2, t1; \
+	vpunpckhdq		x3, x2, x3; \
+	\
+	vpunpcklqdq		t1, t0, x0; \
+	vpunpckhqdq		t1, t0, x1; \
+	vpunpcklqdq		x3, t2, x2; \
+	vpunpckhqdq		x3, t2, x3;
+
+#define read_blocks(x0, x1, x2, x3, t0, t1, t2) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2)
+
+#define write_blocks(x0, x1, x2, x3, t0, t1, t2) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2)
+
+SYM_FUNC_START_LOCAL(__serpent_enc_blk16)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: plaintext
+	 * output:
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: ciphertext
+	 */
+
+	vpcmpeqd RNOT, RNOT, RNOT;
+
+	read_blocks(RA1, RB1, RC1, RD1, RK0, RK1, RK2);
+	read_blocks(RA2, RB2, RC2, RD2, RK0, RK1, RK2);
+
+						 K2(RA, RB, RC, RD, RE, 0);
+	S(S0, RA, RB, RC, RD, RE);		LK2(RC, RB, RD, RA, RE, 1);
+	S(S1, RC, RB, RD, RA, RE);		LK2(RE, RD, RA, RC, RB, 2);
+	S(S2, RE, RD, RA, RC, RB);		LK2(RB, RD, RE, RC, RA, 3);
+	S(S3, RB, RD, RE, RC, RA);		LK2(RC, RA, RD, RB, RE, 4);
+	S(S4, RC, RA, RD, RB, RE);		LK2(RA, RD, RB, RE, RC, 5);
+	S(S5, RA, RD, RB, RE, RC);		LK2(RC, RA, RD, RE, RB, 6);
+	S(S6, RC, RA, RD, RE, RB);		LK2(RD, RB, RA, RE, RC, 7);
+	S(S7, RD, RB, RA, RE, RC);		LK2(RC, RA, RE, RD, RB, 8);
+	S(S0, RC, RA, RE, RD, RB);		LK2(RE, RA, RD, RC, RB, 9);
+	S(S1, RE, RA, RD, RC, RB);		LK2(RB, RD, RC, RE, RA, 10);
+	S(S2, RB, RD, RC, RE, RA);		LK2(RA, RD, RB, RE, RC, 11);
+	S(S3, RA, RD, RB, RE, RC);		LK2(RE, RC, RD, RA, RB, 12);
+	S(S4, RE, RC, RD, RA, RB);		LK2(RC, RD, RA, RB, RE, 13);
+	S(S5, RC, RD, RA, RB, RE);		LK2(RE, RC, RD, RB, RA, 14);
+	S(S6, RE, RC, RD, RB, RA);		LK2(RD, RA, RC, RB, RE, 15);
+	S(S7, RD, RA, RC, RB, RE);		LK2(RE, RC, RB, RD, RA, 16);
+	S(S0, RE, RC, RB, RD, RA);		LK2(RB, RC, RD, RE, RA, 17);
+	S(S1, RB, RC, RD, RE, RA);		LK2(RA, RD, RE, RB, RC, 18);
+	S(S2, RA, RD, RE, RB, RC);		LK2(RC, RD, RA, RB, RE, 19);
+	S(S3, RC, RD, RA, RB, RE);		LK2(RB, RE, RD, RC, RA, 20);
+	S(S4, RB, RE, RD, RC, RA);		LK2(RE, RD, RC, RA, RB, 21);
+	S(S5, RE, RD, RC, RA, RB);		LK2(RB, RE, RD, RA, RC, 22);
+	S(S6, RB, RE, RD, RA, RC);		LK2(RD, RC, RE, RA, RB, 23);
+	S(S7, RD, RC, RE, RA, RB);		LK2(RB, RE, RA, RD, RC, 24);
+	S(S0, RB, RE, RA, RD, RC);		LK2(RA, RE, RD, RB, RC, 25);
+	S(S1, RA, RE, RD, RB, RC);		LK2(RC, RD, RB, RA, RE, 26);
+	S(S2, RC, RD, RB, RA, RE);		LK2(RE, RD, RC, RA, RB, 27);
+	S(S3, RE, RD, RC, RA, RB);		LK2(RA, RB, RD, RE, RC, 28);
+	S(S4, RA, RB, RD, RE, RC);		LK2(RB, RD, RE, RC, RA, 29);
+	S(S5, RB, RD, RE, RC, RA);		LK2(RA, RB, RD, RC, RE, 30);
+	S(S6, RA, RB, RD, RC, RE);		LK2(RD, RE, RB, RC, RA, 31);
+	S(S7, RD, RE, RB, RC, RA);		 K2(RA, RB, RC, RD, RE, 32);
+
+	write_blocks(RA1, RB1, RC1, RD1, RK0, RK1, RK2);
+	write_blocks(RA2, RB2, RC2, RD2, RK0, RK1, RK2);
+
+	RET;
+SYM_FUNC_END(__serpent_enc_blk16)
+
+SYM_FUNC_START_LOCAL(__serpent_dec_blk16)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: ciphertext
+	 * output:
+	 *	RC1, RD1, RB1, RE1, RC2, RD2, RB2, RE2: plaintext
+	 */
+
+	vpcmpeqd RNOT, RNOT, RNOT;
+
+	read_blocks(RA1, RB1, RC1, RD1, RK0, RK1, RK2);
+	read_blocks(RA2, RB2, RC2, RD2, RK0, RK1, RK2);
+
+						 K2(RA, RB, RC, RD, RE, 32);
+	SP(SI7, RA, RB, RC, RD, RE, 31);	KL2(RB, RD, RA, RE, RC, 31);
+	SP(SI6, RB, RD, RA, RE, RC, 30);	KL2(RA, RC, RE, RB, RD, 30);
+	SP(SI5, RA, RC, RE, RB, RD, 29);	KL2(RC, RD, RA, RE, RB, 29);
+	SP(SI4, RC, RD, RA, RE, RB, 28);	KL2(RC, RA, RB, RE, RD, 28);
+	SP(SI3, RC, RA, RB, RE, RD, 27);	KL2(RB, RC, RD, RE, RA, 27);
+	SP(SI2, RB, RC, RD, RE, RA, 26);	KL2(RC, RA, RE, RD, RB, 26);
+	SP(SI1, RC, RA, RE, RD, RB, 25);	KL2(RB, RA, RE, RD, RC, 25);
+	SP(SI0, RB, RA, RE, RD, RC, 24);	KL2(RE, RC, RA, RB, RD, 24);
+	SP(SI7, RE, RC, RA, RB, RD, 23);	KL2(RC, RB, RE, RD, RA, 23);
+	SP(SI6, RC, RB, RE, RD, RA, 22);	KL2(RE, RA, RD, RC, RB, 22);
+	SP(SI5, RE, RA, RD, RC, RB, 21);	KL2(RA, RB, RE, RD, RC, 21);
+	SP(SI4, RA, RB, RE, RD, RC, 20);	KL2(RA, RE, RC, RD, RB, 20);
+	SP(SI3, RA, RE, RC, RD, RB, 19);	KL2(RC, RA, RB, RD, RE, 19);
+	SP(SI2, RC, RA, RB, RD, RE, 18);	KL2(RA, RE, RD, RB, RC, 18);
+	SP(SI1, RA, RE, RD, RB, RC, 17);	KL2(RC, RE, RD, RB, RA, 17);
+	SP(SI0, RC, RE, RD, RB, RA, 16);	KL2(RD, RA, RE, RC, RB, 16);
+	SP(SI7, RD, RA, RE, RC, RB, 15);	KL2(RA, RC, RD, RB, RE, 15);
+	SP(SI6, RA, RC, RD, RB, RE, 14);	KL2(RD, RE, RB, RA, RC, 14);
+	SP(SI5, RD, RE, RB, RA, RC, 13);	KL2(RE, RC, RD, RB, RA, 13);
+	SP(SI4, RE, RC, RD, RB, RA, 12);	KL2(RE, RD, RA, RB, RC, 12);
+	SP(SI3, RE, RD, RA, RB, RC, 11);	KL2(RA, RE, RC, RB, RD, 11);
+	SP(SI2, RA, RE, RC, RB, RD, 10);	KL2(RE, RD, RB, RC, RA, 10);
+	SP(SI1, RE, RD, RB, RC, RA, 9);		KL2(RA, RD, RB, RC, RE, 9);
+	SP(SI0, RA, RD, RB, RC, RE, 8);		KL2(RB, RE, RD, RA, RC, 8);
+	SP(SI7, RB, RE, RD, RA, RC, 7);		KL2(RE, RA, RB, RC, RD, 7);
+	SP(SI6, RE, RA, RB, RC, RD, 6);		KL2(RB, RD, RC, RE, RA, 6);
+	SP(SI5, RB, RD, RC, RE, RA, 5);		KL2(RD, RA, RB, RC, RE, 5);
+	SP(SI4, RD, RA, RB, RC, RE, 4);		KL2(RD, RB, RE, RC, RA, 4);
+	SP(SI3, RD, RB, RE, RC, RA, 3);		KL2(RE, RD, RA, RC, RB, 3);
+	SP(SI2, RE, RD, RA, RC, RB, 2);		KL2(RD, RB, RC, RA, RE, 2);
+	SP(SI1, RD, RB, RC, RA, RE, 1);		KL2(RE, RB, RC, RA, RD, 1);
+	S(SI0, RE, RB, RC, RA, RD);		 K2(RC, RD, RB, RE, RA, 0);
+
+	write_blocks(RC1, RD1, RB1, RE1, RK0, RK1, RK2);
+	write_blocks(RC2, RD2, RB2, RE2, RK0, RK1, RK2);
+
+	RET;
+SYM_FUNC_END(__serpent_dec_blk16)
+
+SYM_FUNC_START(serpent_ecb_enc_16way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+
+	vzeroupper;
+
+	load_16way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	call __serpent_enc_blk16;
+
+	store_16way(%rsi, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	vzeroupper;
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(serpent_ecb_enc_16way)
+
+SYM_FUNC_START(serpent_ecb_dec_16way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+
+	vzeroupper;
+
+	load_16way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	call __serpent_dec_blk16;
+
+	store_16way(%rsi, RC1, RD1, RB1, RE1, RC2, RD2, RB2, RE2);
+
+	vzeroupper;
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(serpent_ecb_dec_16way)
+
+SYM_FUNC_START(serpent_cbc_dec_16way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+
+	vzeroupper;
+
+	load_16way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	call __serpent_dec_blk16;
+
+	store_cbc_16way(%rdx, %rsi, RC1, RD1, RB1, RE1, RC2, RD2, RB2, RE2,
+			RK0);
+
+	vzeroupper;
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(serpent_cbc_dec_16way)
diff --git a/arch/x86/crypto/serpent-sse2-i586-asm_32.S b/arch/x86/crypto/serpent-sse2-i586-asm_32.S
new file mode 100644
index 000000000..8ccb03ad7
--- /dev/null
+++ b/arch/x86/crypto/serpent-sse2-i586-asm_32.S
@@ -0,0 +1,616 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Serpent Cipher 4-way parallel algorithm (i586/SSE2)
+ *
+ * Copyright (C) 2011 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ *
+ * Based on crypto/serpent.c by
+ *  Copyright (C) 2002 Dag Arne Osvik <osvik@ii.uib.no>
+ *                2003 Herbert Valerio Riedel <hvr@gnu.org>
+ */
+
+#include <linux/linkage.h>
+
+.file "serpent-sse2-i586-asm_32.S"
+.text
+
+#define arg_ctx 4
+#define arg_dst 8
+#define arg_src 12
+#define arg_xor 16
+
+/**********************************************************************
+  4-way SSE2 serpent
+ **********************************************************************/
+#define CTX %edx
+
+#define RA %xmm0
+#define RB %xmm1
+#define RC %xmm2
+#define RD %xmm3
+#define RE %xmm4
+
+#define RT0 %xmm5
+#define RT1 %xmm6
+
+#define RNOT %xmm7
+
+#define get_key(i, j, t) \
+	movd (4*(i)+(j))*4(CTX), t; \
+	pshufd $0, t, t;
+
+#define K(x0, x1, x2, x3, x4, i) \
+	get_key(i, 0, x4); \
+	get_key(i, 1, RT0); \
+	get_key(i, 2, RT1); \
+	pxor x4,		x0; \
+	pxor RT0,		x1; \
+	pxor RT1,		x2; \
+	get_key(i, 3, x4); \
+	pxor x4,		x3;
+
+#define LK(x0, x1, x2, x3, x4, i) \
+	movdqa x0,		x4; \
+	pslld $13,		x0; \
+	psrld $(32 - 13),	x4; \
+	por x4,			x0; \
+	pxor x0,		x1; \
+	movdqa x2,		x4; \
+	pslld $3,		x2; \
+	psrld $(32 - 3),	x4; \
+	por x4,			x2; \
+	pxor x2,		x1; \
+	movdqa x1,		x4; \
+	pslld $1,		x1; \
+	psrld $(32 - 1),	x4; \
+	por x4,			x1; \
+	movdqa x0,		x4; \
+	pslld $3,		x4; \
+	pxor x2,		x3; \
+	pxor x4,		x3; \
+	movdqa x3,		x4; \
+	pslld $7,		x3; \
+	psrld $(32 - 7),	x4; \
+	por x4,			x3; \
+	movdqa x1,		x4; \
+	pslld $7,		x4; \
+	pxor x1,		x0; \
+	pxor x3,		x0; \
+	pxor x3,		x2; \
+	pxor x4,		x2; \
+	movdqa x0,		x4; \
+	get_key(i, 1, RT0); \
+	pxor RT0,		x1; \
+	get_key(i, 3, RT0); \
+	pxor RT0,		x3; \
+	pslld $5,		x0; \
+	psrld $(32 - 5),	x4; \
+	por x4,			x0; \
+	movdqa x2,		x4; \
+	pslld $22,		x2; \
+	psrld $(32 - 22),	x4; \
+	por x4,			x2; \
+	get_key(i, 0, RT0); \
+	pxor RT0,		x0; \
+	get_key(i, 2, RT0); \
+	pxor RT0,		x2;
+
+#define KL(x0, x1, x2, x3, x4, i) \
+	K(x0, x1, x2, x3, x4, i); \
+	movdqa x0,		x4; \
+	psrld $5,		x0; \
+	pslld $(32 - 5),	x4; \
+	por x4,			x0; \
+	movdqa x2,		x4; \
+	psrld $22,		x2; \
+	pslld $(32 - 22),	x4; \
+	por x4,			x2; \
+	pxor x3,		x2; \
+	pxor x3,		x0; \
+	movdqa x1,		x4; \
+	pslld $7,		x4; \
+	pxor x1,		x0; \
+	pxor x4,		x2; \
+	movdqa x1,		x4; \
+	psrld $1,		x1; \
+	pslld $(32 - 1),	x4; \
+	por x4,			x1; \
+	movdqa x3,		x4; \
+	psrld $7,		x3; \
+	pslld $(32 - 7),	x4; \
+	por x4,			x3; \
+	pxor x0,		x1; \
+	movdqa x0,		x4; \
+	pslld $3,		x4; \
+	pxor x4,		x3; \
+	movdqa x0,		x4; \
+	psrld $13,		x0; \
+	pslld $(32 - 13),	x4; \
+	por x4,			x0; \
+	pxor x2,		x1; \
+	pxor x2,		x3; \
+	movdqa x2,		x4; \
+	psrld $3,		x2; \
+	pslld $(32 - 3),	x4; \
+	por x4,			x2;
+
+#define S0(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	por x0,			x3; \
+	pxor x4,		x0; \
+	pxor x2,		x4; \
+	pxor RNOT,		x4; \
+	pxor x1,		x3; \
+	pand x0,		x1; \
+	pxor x4,		x1; \
+	pxor x0,		x2; \
+	pxor x3,		x0; \
+	por x0,			x4; \
+	pxor x2,		x0; \
+	pand x1,		x2; \
+	pxor x2,		x3; \
+	pxor RNOT,		x1; \
+	pxor x4,		x2; \
+	pxor x2,		x1;
+
+#define S1(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	pxor x0,		x1; \
+	pxor x3,		x0; \
+	pxor RNOT,		x3; \
+	pand x1,		x4; \
+	por x1,			x0; \
+	pxor x2,		x3; \
+	pxor x3,		x0; \
+	pxor x3,		x1; \
+	pxor x4,		x3; \
+	por x4,			x1; \
+	pxor x2,		x4; \
+	pand x0,		x2; \
+	pxor x1,		x2; \
+	por x0,			x1; \
+	pxor RNOT,		x0; \
+	pxor x2,		x0; \
+	pxor x1,		x4;
+
+#define S2(x0, x1, x2, x3, x4) \
+	pxor RNOT,		x3; \
+	pxor x0,		x1; \
+	movdqa x0,		x4; \
+	pand x2,		x0; \
+	pxor x3,		x0; \
+	por x4,			x3; \
+	pxor x1,		x2; \
+	pxor x1,		x3; \
+	pand x0,		x1; \
+	pxor x2,		x0; \
+	pand x3,		x2; \
+	por x1,			x3; \
+	pxor RNOT,		x0; \
+	pxor x0,		x3; \
+	pxor x0,		x4; \
+	pxor x2,		x0; \
+	por x2,			x1;
+
+#define S3(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	pxor x3,		x1; \
+	por x0,			x3; \
+	pand x0,		x4; \
+	pxor x2,		x0; \
+	pxor x1,		x2; \
+	pand x3,		x1; \
+	pxor x3,		x2; \
+	por x4,			x0; \
+	pxor x3,		x4; \
+	pxor x0,		x1; \
+	pand x3,		x0; \
+	pand x4,		x3; \
+	pxor x2,		x3; \
+	por x1,			x4; \
+	pand x1,		x2; \
+	pxor x3,		x4; \
+	pxor x3,		x0; \
+	pxor x2,		x3;
+
+#define S4(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	pand x0,		x3; \
+	pxor x4,		x0; \
+	pxor x2,		x3; \
+	por x4,			x2; \
+	pxor x1,		x0; \
+	pxor x3,		x4; \
+	por x0,			x2; \
+	pxor x1,		x2; \
+	pand x0,		x1; \
+	pxor x4,		x1; \
+	pand x2,		x4; \
+	pxor x3,		x2; \
+	pxor x0,		x4; \
+	por x1,			x3; \
+	pxor RNOT,		x1; \
+	pxor x0,		x3;
+
+#define S5(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	por x0,			x1; \
+	pxor x1,		x2; \
+	pxor RNOT,		x3; \
+	pxor x0,		x4; \
+	pxor x2,		x0; \
+	pand x4,		x1; \
+	por x3,			x4; \
+	pxor x0,		x4; \
+	pand x3,		x0; \
+	pxor x3,		x1; \
+	pxor x2,		x3; \
+	pxor x1,		x0; \
+	pand x4,		x2; \
+	pxor x2,		x1; \
+	pand x0,		x2; \
+	pxor x2,		x3;
+
+#define S6(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	pxor x0,		x3; \
+	pxor x2,		x1; \
+	pxor x0,		x2; \
+	pand x3,		x0; \
+	por x3,			x1; \
+	pxor RNOT,		x4; \
+	pxor x1,		x0; \
+	pxor x2,		x1; \
+	pxor x4,		x3; \
+	pxor x0,		x4; \
+	pand x0,		x2; \
+	pxor x1,		x4; \
+	pxor x3,		x2; \
+	pand x1,		x3; \
+	pxor x0,		x3; \
+	pxor x2,		x1;
+
+#define S7(x0, x1, x2, x3, x4) \
+	pxor RNOT,		x1; \
+	movdqa x1,		x4; \
+	pxor RNOT,		x0; \
+	pand x2,		x1; \
+	pxor x3,		x1; \
+	por x4,			x3; \
+	pxor x2,		x4; \
+	pxor x3,		x2; \
+	pxor x0,		x3; \
+	por x1,			x0; \
+	pand x0,		x2; \
+	pxor x4,		x0; \
+	pxor x3,		x4; \
+	pand x0,		x3; \
+	pxor x1,		x4; \
+	pxor x4,		x2; \
+	pxor x1,		x3; \
+	por x0,			x4; \
+	pxor x1,		x4;
+
+#define SI0(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	pxor x0,		x1; \
+	por x1,			x3; \
+	pxor x1,		x4; \
+	pxor RNOT,		x0; \
+	pxor x3,		x2; \
+	pxor x0,		x3; \
+	pand x1,		x0; \
+	pxor x2,		x0; \
+	pand x3,		x2; \
+	pxor x4,		x3; \
+	pxor x3,		x2; \
+	pxor x3,		x1; \
+	pand x0,		x3; \
+	pxor x0,		x1; \
+	pxor x2,		x0; \
+	pxor x3,		x4;
+
+#define SI1(x0, x1, x2, x3, x4) \
+	pxor x3,		x1; \
+	movdqa x0,		x4; \
+	pxor x2,		x0; \
+	pxor RNOT,		x2; \
+	por x1,			x4; \
+	pxor x3,		x4; \
+	pand x1,		x3; \
+	pxor x2,		x1; \
+	pand x4,		x2; \
+	pxor x1,		x4; \
+	por x3,			x1; \
+	pxor x0,		x3; \
+	pxor x0,		x2; \
+	por x4,			x0; \
+	pxor x4,		x2; \
+	pxor x0,		x1; \
+	pxor x1,		x4;
+
+#define SI2(x0, x1, x2, x3, x4) \
+	pxor x1,		x2; \
+	movdqa x3,		x4; \
+	pxor RNOT,		x3; \
+	por x2,			x3; \
+	pxor x4,		x2; \
+	pxor x0,		x4; \
+	pxor x1,		x3; \
+	por x2,			x1; \
+	pxor x0,		x2; \
+	pxor x4,		x1; \
+	por x3,			x4; \
+	pxor x3,		x2; \
+	pxor x2,		x4; \
+	pand x1,		x2; \
+	pxor x3,		x2; \
+	pxor x4,		x3; \
+	pxor x0,		x4;
+
+#define SI3(x0, x1, x2, x3, x4) \
+	pxor x1,		x2; \
+	movdqa x1,		x4; \
+	pand x2,		x1; \
+	pxor x0,		x1; \
+	por x4,			x0; \
+	pxor x3,		x4; \
+	pxor x3,		x0; \
+	por x1,			x3; \
+	pxor x2,		x1; \
+	pxor x3,		x1; \
+	pxor x2,		x0; \
+	pxor x3,		x2; \
+	pand x1,		x3; \
+	pxor x0,		x1; \
+	pand x2,		x0; \
+	pxor x3,		x4; \
+	pxor x0,		x3; \
+	pxor x1,		x0;
+
+#define SI4(x0, x1, x2, x3, x4) \
+	pxor x3,		x2; \
+	movdqa x0,		x4; \
+	pand x1,		x0; \
+	pxor x2,		x0; \
+	por x3,			x2; \
+	pxor RNOT,		x4; \
+	pxor x0,		x1; \
+	pxor x2,		x0; \
+	pand x4,		x2; \
+	pxor x0,		x2; \
+	por x4,			x0; \
+	pxor x3,		x0; \
+	pand x2,		x3; \
+	pxor x3,		x4; \
+	pxor x1,		x3; \
+	pand x0,		x1; \
+	pxor x1,		x4; \
+	pxor x3,		x0;
+
+#define SI5(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	por x2,			x1; \
+	pxor x4,		x2; \
+	pxor x3,		x1; \
+	pand x4,		x3; \
+	pxor x3,		x2; \
+	por x0,			x3; \
+	pxor RNOT,		x0; \
+	pxor x2,		x3; \
+	por x0,			x2; \
+	pxor x1,		x4; \
+	pxor x4,		x2; \
+	pand x0,		x4; \
+	pxor x1,		x0; \
+	pxor x3,		x1; \
+	pand x2,		x0; \
+	pxor x3,		x2; \
+	pxor x2,		x0; \
+	pxor x4,		x2; \
+	pxor x3,		x4;
+
+#define SI6(x0, x1, x2, x3, x4) \
+	pxor x2,		x0; \
+	movdqa x0,		x4; \
+	pand x3,		x0; \
+	pxor x3,		x2; \
+	pxor x2,		x0; \
+	pxor x1,		x3; \
+	por x4,			x2; \
+	pxor x3,		x2; \
+	pand x0,		x3; \
+	pxor RNOT,		x0; \
+	pxor x1,		x3; \
+	pand x2,		x1; \
+	pxor x0,		x4; \
+	pxor x4,		x3; \
+	pxor x2,		x4; \
+	pxor x1,		x0; \
+	pxor x0,		x2;
+
+#define SI7(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	pand x0,		x3; \
+	pxor x2,		x0; \
+	por x4,			x2; \
+	pxor x1,		x4; \
+	pxor RNOT,		x0; \
+	por x3,			x1; \
+	pxor x0,		x4; \
+	pand x2,		x0; \
+	pxor x1,		x0; \
+	pand x2,		x1; \
+	pxor x2,		x3; \
+	pxor x3,		x4; \
+	pand x3,		x2; \
+	por x0,			x3; \
+	pxor x4,		x1; \
+	pxor x4,		x3; \
+	pand x0,		x4; \
+	pxor x2,		x4;
+
+#define transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	movdqa x0,		t2; \
+	punpckldq x1,		x0; \
+	punpckhdq x1,		t2; \
+	movdqa x2,		t1; \
+	punpckhdq x3,		x2; \
+	punpckldq x3,		t1; \
+	movdqa x0,		x1; \
+	punpcklqdq t1,		x0; \
+	punpckhqdq t1,		x1; \
+	movdqa t2,		x3; \
+	punpcklqdq x2,		t2; \
+	punpckhqdq x2,		x3; \
+	movdqa t2,		x2;
+
+#define read_blocks(in, x0, x1, x2, x3, t0, t1, t2) \
+	movdqu (0*4*4)(in),	x0; \
+	movdqu (1*4*4)(in),	x1; \
+	movdqu (2*4*4)(in),	x2; \
+	movdqu (3*4*4)(in),	x3; \
+	\
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2)
+
+#define write_blocks(out, x0, x1, x2, x3, t0, t1, t2) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	\
+	movdqu x0, (0*4*4)(out); \
+	movdqu x1, (1*4*4)(out); \
+	movdqu x2, (2*4*4)(out); \
+	movdqu x3, (3*4*4)(out);
+
+#define xor_blocks(out, x0, x1, x2, x3, t0, t1, t2) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	\
+	movdqu (0*4*4)(out),	t0; \
+	pxor t0,		x0; \
+	movdqu x0,		(0*4*4)(out); \
+	movdqu (1*4*4)(out),	t0; \
+	pxor t0,		x1; \
+	movdqu x1,		(1*4*4)(out); \
+	movdqu (2*4*4)(out),	t0; \
+	pxor t0,		x2; \
+	movdqu x2,		(2*4*4)(out); \
+	movdqu (3*4*4)(out),	t0; \
+	pxor t0,		x3; \
+	movdqu x3,		(3*4*4)(out);
+
+SYM_FUNC_START(__serpent_enc_blk_4way)
+	/* input:
+	 *	arg_ctx(%esp): ctx, CTX
+	 *	arg_dst(%esp): dst
+	 *	arg_src(%esp): src
+	 *	arg_xor(%esp): bool, if true: xor output
+	 */
+
+	pcmpeqd RNOT, RNOT;
+
+	movl arg_ctx(%esp), CTX;
+
+	movl arg_src(%esp), %eax;
+	read_blocks(%eax, RA, RB, RC, RD, RT0, RT1, RE);
+
+					 K(RA, RB, RC, RD, RE, 0);
+	S0(RA, RB, RC, RD, RE);		LK(RC, RB, RD, RA, RE, 1);
+	S1(RC, RB, RD, RA, RE);		LK(RE, RD, RA, RC, RB, 2);
+	S2(RE, RD, RA, RC, RB);		LK(RB, RD, RE, RC, RA, 3);
+	S3(RB, RD, RE, RC, RA);		LK(RC, RA, RD, RB, RE, 4);
+	S4(RC, RA, RD, RB, RE);		LK(RA, RD, RB, RE, RC, 5);
+	S5(RA, RD, RB, RE, RC);		LK(RC, RA, RD, RE, RB, 6);
+	S6(RC, RA, RD, RE, RB);		LK(RD, RB, RA, RE, RC, 7);
+	S7(RD, RB, RA, RE, RC);		LK(RC, RA, RE, RD, RB, 8);
+	S0(RC, RA, RE, RD, RB);		LK(RE, RA, RD, RC, RB, 9);
+	S1(RE, RA, RD, RC, RB);		LK(RB, RD, RC, RE, RA, 10);
+	S2(RB, RD, RC, RE, RA);		LK(RA, RD, RB, RE, RC, 11);
+	S3(RA, RD, RB, RE, RC);		LK(RE, RC, RD, RA, RB, 12);
+	S4(RE, RC, RD, RA, RB);		LK(RC, RD, RA, RB, RE, 13);
+	S5(RC, RD, RA, RB, RE);		LK(RE, RC, RD, RB, RA, 14);
+	S6(RE, RC, RD, RB, RA);		LK(RD, RA, RC, RB, RE, 15);
+	S7(RD, RA, RC, RB, RE);		LK(RE, RC, RB, RD, RA, 16);
+	S0(RE, RC, RB, RD, RA);		LK(RB, RC, RD, RE, RA, 17);
+	S1(RB, RC, RD, RE, RA);		LK(RA, RD, RE, RB, RC, 18);
+	S2(RA, RD, RE, RB, RC);		LK(RC, RD, RA, RB, RE, 19);
+	S3(RC, RD, RA, RB, RE);		LK(RB, RE, RD, RC, RA, 20);
+	S4(RB, RE, RD, RC, RA);		LK(RE, RD, RC, RA, RB, 21);
+	S5(RE, RD, RC, RA, RB);		LK(RB, RE, RD, RA, RC, 22);
+	S6(RB, RE, RD, RA, RC);		LK(RD, RC, RE, RA, RB, 23);
+	S7(RD, RC, RE, RA, RB);		LK(RB, RE, RA, RD, RC, 24);
+	S0(RB, RE, RA, RD, RC);		LK(RA, RE, RD, RB, RC, 25);
+	S1(RA, RE, RD, RB, RC);		LK(RC, RD, RB, RA, RE, 26);
+	S2(RC, RD, RB, RA, RE);		LK(RE, RD, RC, RA, RB, 27);
+	S3(RE, RD, RC, RA, RB);		LK(RA, RB, RD, RE, RC, 28);
+	S4(RA, RB, RD, RE, RC);		LK(RB, RD, RE, RC, RA, 29);
+	S5(RB, RD, RE, RC, RA);		LK(RA, RB, RD, RC, RE, 30);
+	S6(RA, RB, RD, RC, RE);		LK(RD, RE, RB, RC, RA, 31);
+	S7(RD, RE, RB, RC, RA);		 K(RA, RB, RC, RD, RE, 32);
+
+	movl arg_dst(%esp), %eax;
+
+	cmpb $0, arg_xor(%esp);
+	jnz .L__enc_xor4;
+
+	write_blocks(%eax, RA, RB, RC, RD, RT0, RT1, RE);
+
+	RET;
+
+.L__enc_xor4:
+	xor_blocks(%eax, RA, RB, RC, RD, RT0, RT1, RE);
+
+	RET;
+SYM_FUNC_END(__serpent_enc_blk_4way)
+
+SYM_FUNC_START(serpent_dec_blk_4way)
+	/* input:
+	 *	arg_ctx(%esp): ctx, CTX
+	 *	arg_dst(%esp): dst
+	 *	arg_src(%esp): src
+	 */
+
+	pcmpeqd RNOT, RNOT;
+
+	movl arg_ctx(%esp), CTX;
+
+	movl arg_src(%esp), %eax;
+	read_blocks(%eax, RA, RB, RC, RD, RT0, RT1, RE);
+
+					 K(RA, RB, RC, RD, RE, 32);
+	SI7(RA, RB, RC, RD, RE);	KL(RB, RD, RA, RE, RC, 31);
+	SI6(RB, RD, RA, RE, RC);	KL(RA, RC, RE, RB, RD, 30);
+	SI5(RA, RC, RE, RB, RD);	KL(RC, RD, RA, RE, RB, 29);
+	SI4(RC, RD, RA, RE, RB);	KL(RC, RA, RB, RE, RD, 28);
+	SI3(RC, RA, RB, RE, RD);	KL(RB, RC, RD, RE, RA, 27);
+	SI2(RB, RC, RD, RE, RA);	KL(RC, RA, RE, RD, RB, 26);
+	SI1(RC, RA, RE, RD, RB);	KL(RB, RA, RE, RD, RC, 25);
+	SI0(RB, RA, RE, RD, RC);	KL(RE, RC, RA, RB, RD, 24);
+	SI7(RE, RC, RA, RB, RD);	KL(RC, RB, RE, RD, RA, 23);
+	SI6(RC, RB, RE, RD, RA);	KL(RE, RA, RD, RC, RB, 22);
+	SI5(RE, RA, RD, RC, RB);	KL(RA, RB, RE, RD, RC, 21);
+	SI4(RA, RB, RE, RD, RC);	KL(RA, RE, RC, RD, RB, 20);
+	SI3(RA, RE, RC, RD, RB);	KL(RC, RA, RB, RD, RE, 19);
+	SI2(RC, RA, RB, RD, RE);	KL(RA, RE, RD, RB, RC, 18);
+	SI1(RA, RE, RD, RB, RC);	KL(RC, RE, RD, RB, RA, 17);
+	SI0(RC, RE, RD, RB, RA);	KL(RD, RA, RE, RC, RB, 16);
+	SI7(RD, RA, RE, RC, RB);	KL(RA, RC, RD, RB, RE, 15);
+	SI6(RA, RC, RD, RB, RE);	KL(RD, RE, RB, RA, RC, 14);
+	SI5(RD, RE, RB, RA, RC);	KL(RE, RC, RD, RB, RA, 13);
+	SI4(RE, RC, RD, RB, RA);	KL(RE, RD, RA, RB, RC, 12);
+	SI3(RE, RD, RA, RB, RC);	KL(RA, RE, RC, RB, RD, 11);
+	SI2(RA, RE, RC, RB, RD);	KL(RE, RD, RB, RC, RA, 10);
+	SI1(RE, RD, RB, RC, RA);	KL(RA, RD, RB, RC, RE, 9);
+	SI0(RA, RD, RB, RC, RE);	KL(RB, RE, RD, RA, RC, 8);
+	SI7(RB, RE, RD, RA, RC);	KL(RE, RA, RB, RC, RD, 7);
+	SI6(RE, RA, RB, RC, RD);	KL(RB, RD, RC, RE, RA, 6);
+	SI5(RB, RD, RC, RE, RA);	KL(RD, RA, RB, RC, RE, 5);
+	SI4(RD, RA, RB, RC, RE);	KL(RD, RB, RE, RC, RA, 4);
+	SI3(RD, RB, RE, RC, RA);	KL(RE, RD, RA, RC, RB, 3);
+	SI2(RE, RD, RA, RC, RB);	KL(RD, RB, RC, RA, RE, 2);
+	SI1(RD, RB, RC, RA, RE);	KL(RE, RB, RC, RA, RD, 1);
+	SI0(RE, RB, RC, RA, RD);	 K(RC, RD, RB, RE, RA, 0);
+
+	movl arg_dst(%esp), %eax;
+	write_blocks(%eax, RC, RD, RB, RE, RT0, RT1, RA);
+
+	RET;
+SYM_FUNC_END(serpent_dec_blk_4way)
diff --git a/arch/x86/crypto/serpent-sse2-x86_64-asm_64.S b/arch/x86/crypto/serpent-sse2-x86_64-asm_64.S
new file mode 100644
index 000000000..e0998a011
--- /dev/null
+++ b/arch/x86/crypto/serpent-sse2-x86_64-asm_64.S
@@ -0,0 +1,739 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Serpent Cipher 8-way parallel algorithm (x86_64/SSE2)
+ *
+ * Copyright (C) 2011 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ *
+ * Based on crypto/serpent.c by
+ *  Copyright (C) 2002 Dag Arne Osvik <osvik@ii.uib.no>
+ *                2003 Herbert Valerio Riedel <hvr@gnu.org>
+ */
+
+#include <linux/linkage.h>
+
+.file "serpent-sse2-x86_64-asm_64.S"
+.text
+
+#define CTX %rdi
+
+/**********************************************************************
+  8-way SSE2 serpent
+ **********************************************************************/
+#define RA1 %xmm0
+#define RB1 %xmm1
+#define RC1 %xmm2
+#define RD1 %xmm3
+#define RE1 %xmm4
+
+#define RA2 %xmm5
+#define RB2 %xmm6
+#define RC2 %xmm7
+#define RD2 %xmm8
+#define RE2 %xmm9
+
+#define RNOT %xmm10
+
+#define RK0 %xmm11
+#define RK1 %xmm12
+#define RK2 %xmm13
+#define RK3 %xmm14
+
+#define S0_1(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	por x0,			x3; \
+	pxor x4,		x0; \
+	pxor x2,		x4; \
+	pxor RNOT,		x4; \
+	pxor x1,		x3; \
+	pand x0,		x1; \
+	pxor x4,		x1; \
+	pxor x0,		x2;
+#define S0_2(x0, x1, x2, x3, x4) \
+	pxor x3,		x0; \
+	por x0,			x4; \
+	pxor x2,		x0; \
+	pand x1,		x2; \
+	pxor x2,		x3; \
+	pxor RNOT,		x1; \
+	pxor x4,		x2; \
+	pxor x2,		x1;
+
+#define S1_1(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	pxor x0,		x1; \
+	pxor x3,		x0; \
+	pxor RNOT,		x3; \
+	pand x1,		x4; \
+	por x1,			x0; \
+	pxor x2,		x3; \
+	pxor x3,		x0; \
+	pxor x3,		x1;
+#define S1_2(x0, x1, x2, x3, x4) \
+	pxor x4,		x3; \
+	por x4,			x1; \
+	pxor x2,		x4; \
+	pand x0,		x2; \
+	pxor x1,		x2; \
+	por x0,			x1; \
+	pxor RNOT,		x0; \
+	pxor x2,		x0; \
+	pxor x1,		x4;
+
+#define S2_1(x0, x1, x2, x3, x4) \
+	pxor RNOT,		x3; \
+	pxor x0,		x1; \
+	movdqa x0,		x4; \
+	pand x2,		x0; \
+	pxor x3,		x0; \
+	por x4,			x3; \
+	pxor x1,		x2; \
+	pxor x1,		x3; \
+	pand x0,		x1;
+#define S2_2(x0, x1, x2, x3, x4) \
+	pxor x2,		x0; \
+	pand x3,		x2; \
+	por x1,			x3; \
+	pxor RNOT,		x0; \
+	pxor x0,		x3; \
+	pxor x0,		x4; \
+	pxor x2,		x0; \
+	por x2,			x1;
+
+#define S3_1(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	pxor x3,		x1; \
+	por x0,			x3; \
+	pand x0,		x4; \
+	pxor x2,		x0; \
+	pxor x1,		x2; \
+	pand x3,		x1; \
+	pxor x3,		x2; \
+	por x4,			x0; \
+	pxor x3,		x4;
+#define S3_2(x0, x1, x2, x3, x4) \
+	pxor x0,		x1; \
+	pand x3,		x0; \
+	pand x4,		x3; \
+	pxor x2,		x3; \
+	por x1,			x4; \
+	pand x1,		x2; \
+	pxor x3,		x4; \
+	pxor x3,		x0; \
+	pxor x2,		x3;
+
+#define S4_1(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	pand x0,		x3; \
+	pxor x4,		x0; \
+	pxor x2,		x3; \
+	por x4,			x2; \
+	pxor x1,		x0; \
+	pxor x3,		x4; \
+	por x0,			x2; \
+	pxor x1,		x2;
+#define S4_2(x0, x1, x2, x3, x4) \
+	pand x0,		x1; \
+	pxor x4,		x1; \
+	pand x2,		x4; \
+	pxor x3,		x2; \
+	pxor x0,		x4; \
+	por x1,			x3; \
+	pxor RNOT,		x1; \
+	pxor x0,		x3;
+
+#define S5_1(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	por x0,			x1; \
+	pxor x1,		x2; \
+	pxor RNOT,		x3; \
+	pxor x0,		x4; \
+	pxor x2,		x0; \
+	pand x4,		x1; \
+	por x3,			x4; \
+	pxor x0,		x4;
+#define S5_2(x0, x1, x2, x3, x4) \
+	pand x3,		x0; \
+	pxor x3,		x1; \
+	pxor x2,		x3; \
+	pxor x1,		x0; \
+	pand x4,		x2; \
+	pxor x2,		x1; \
+	pand x0,		x2; \
+	pxor x2,		x3;
+
+#define S6_1(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	pxor x0,		x3; \
+	pxor x2,		x1; \
+	pxor x0,		x2; \
+	pand x3,		x0; \
+	por x3,			x1; \
+	pxor RNOT,		x4; \
+	pxor x1,		x0; \
+	pxor x2,		x1;
+#define S6_2(x0, x1, x2, x3, x4) \
+	pxor x4,		x3; \
+	pxor x0,		x4; \
+	pand x0,		x2; \
+	pxor x1,		x4; \
+	pxor x3,		x2; \
+	pand x1,		x3; \
+	pxor x0,		x3; \
+	pxor x2,		x1;
+
+#define S7_1(x0, x1, x2, x3, x4) \
+	pxor RNOT,		x1; \
+	movdqa x1,		x4; \
+	pxor RNOT,		x0; \
+	pand x2,		x1; \
+	pxor x3,		x1; \
+	por x4,			x3; \
+	pxor x2,		x4; \
+	pxor x3,		x2; \
+	pxor x0,		x3; \
+	por x1,			x0;
+#define S7_2(x0, x1, x2, x3, x4) \
+	pand x0,		x2; \
+	pxor x4,		x0; \
+	pxor x3,		x4; \
+	pand x0,		x3; \
+	pxor x1,		x4; \
+	pxor x4,		x2; \
+	pxor x1,		x3; \
+	por x0,			x4; \
+	pxor x1,		x4;
+
+#define SI0_1(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	pxor x0,		x1; \
+	por x1,			x3; \
+	pxor x1,		x4; \
+	pxor RNOT,		x0; \
+	pxor x3,		x2; \
+	pxor x0,		x3; \
+	pand x1,		x0; \
+	pxor x2,		x0;
+#define SI0_2(x0, x1, x2, x3, x4) \
+	pand x3,		x2; \
+	pxor x4,		x3; \
+	pxor x3,		x2; \
+	pxor x3,		x1; \
+	pand x0,		x3; \
+	pxor x0,		x1; \
+	pxor x2,		x0; \
+	pxor x3,		x4;
+
+#define SI1_1(x0, x1, x2, x3, x4) \
+	pxor x3,		x1; \
+	movdqa x0,		x4; \
+	pxor x2,		x0; \
+	pxor RNOT,		x2; \
+	por x1,			x4; \
+	pxor x3,		x4; \
+	pand x1,		x3; \
+	pxor x2,		x1; \
+	pand x4,		x2;
+#define SI1_2(x0, x1, x2, x3, x4) \
+	pxor x1,		x4; \
+	por x3,			x1; \
+	pxor x0,		x3; \
+	pxor x0,		x2; \
+	por x4,			x0; \
+	pxor x4,		x2; \
+	pxor x0,		x1; \
+	pxor x1,		x4;
+
+#define SI2_1(x0, x1, x2, x3, x4) \
+	pxor x1,		x2; \
+	movdqa x3,		x4; \
+	pxor RNOT,		x3; \
+	por x2,			x3; \
+	pxor x4,		x2; \
+	pxor x0,		x4; \
+	pxor x1,		x3; \
+	por x2,			x1; \
+	pxor x0,		x2;
+#define SI2_2(x0, x1, x2, x3, x4) \
+	pxor x4,		x1; \
+	por x3,			x4; \
+	pxor x3,		x2; \
+	pxor x2,		x4; \
+	pand x1,		x2; \
+	pxor x3,		x2; \
+	pxor x4,		x3; \
+	pxor x0,		x4;
+
+#define SI3_1(x0, x1, x2, x3, x4) \
+	pxor x1,		x2; \
+	movdqa x1,		x4; \
+	pand x2,		x1; \
+	pxor x0,		x1; \
+	por x4,			x0; \
+	pxor x3,		x4; \
+	pxor x3,		x0; \
+	por x1,			x3; \
+	pxor x2,		x1;
+#define SI3_2(x0, x1, x2, x3, x4) \
+	pxor x3,		x1; \
+	pxor x2,		x0; \
+	pxor x3,		x2; \
+	pand x1,		x3; \
+	pxor x0,		x1; \
+	pand x2,		x0; \
+	pxor x3,		x4; \
+	pxor x0,		x3; \
+	pxor x1,		x0;
+
+#define SI4_1(x0, x1, x2, x3, x4) \
+	pxor x3,		x2; \
+	movdqa x0,		x4; \
+	pand x1,		x0; \
+	pxor x2,		x0; \
+	por x3,			x2; \
+	pxor RNOT,		x4; \
+	pxor x0,		x1; \
+	pxor x2,		x0; \
+	pand x4,		x2;
+#define SI4_2(x0, x1, x2, x3, x4) \
+	pxor x0,		x2; \
+	por x4,			x0; \
+	pxor x3,		x0; \
+	pand x2,		x3; \
+	pxor x3,		x4; \
+	pxor x1,		x3; \
+	pand x0,		x1; \
+	pxor x1,		x4; \
+	pxor x3,		x0;
+
+#define SI5_1(x0, x1, x2, x3, x4) \
+	movdqa x1,		x4; \
+	por x2,			x1; \
+	pxor x4,		x2; \
+	pxor x3,		x1; \
+	pand x4,		x3; \
+	pxor x3,		x2; \
+	por x0,			x3; \
+	pxor RNOT,		x0; \
+	pxor x2,		x3; \
+	por x0,			x2;
+#define SI5_2(x0, x1, x2, x3, x4) \
+	pxor x1,		x4; \
+	pxor x4,		x2; \
+	pand x0,		x4; \
+	pxor x1,		x0; \
+	pxor x3,		x1; \
+	pand x2,		x0; \
+	pxor x3,		x2; \
+	pxor x2,		x0; \
+	pxor x4,		x2; \
+	pxor x3,		x4;
+
+#define SI6_1(x0, x1, x2, x3, x4) \
+	pxor x2,		x0; \
+	movdqa x0,		x4; \
+	pand x3,		x0; \
+	pxor x3,		x2; \
+	pxor x2,		x0; \
+	pxor x1,		x3; \
+	por x4,			x2; \
+	pxor x3,		x2; \
+	pand x0,		x3;
+#define SI6_2(x0, x1, x2, x3, x4) \
+	pxor RNOT,		x0; \
+	pxor x1,		x3; \
+	pand x2,		x1; \
+	pxor x0,		x4; \
+	pxor x4,		x3; \
+	pxor x2,		x4; \
+	pxor x1,		x0; \
+	pxor x0,		x2;
+
+#define SI7_1(x0, x1, x2, x3, x4) \
+	movdqa x3,		x4; \
+	pand x0,		x3; \
+	pxor x2,		x0; \
+	por x4,			x2; \
+	pxor x1,		x4; \
+	pxor RNOT,		x0; \
+	por x3,			x1; \
+	pxor x0,		x4; \
+	pand x2,		x0; \
+	pxor x1,		x0;
+#define SI7_2(x0, x1, x2, x3, x4) \
+	pand x2,		x1; \
+	pxor x2,		x3; \
+	pxor x3,		x4; \
+	pand x3,		x2; \
+	por x0,			x3; \
+	pxor x4,		x1; \
+	pxor x4,		x3; \
+	pand x0,		x4; \
+	pxor x2,		x4;
+
+#define get_key(i, j, t) \
+	movd (4*(i)+(j))*4(CTX), t; \
+	pshufd $0, t, t;
+
+#define K2(x0, x1, x2, x3, x4, i) \
+	get_key(i, 0, RK0); \
+	get_key(i, 1, RK1); \
+	get_key(i, 2, RK2); \
+	get_key(i, 3, RK3); \
+	pxor RK0,		x0 ## 1; \
+	pxor RK1,		x1 ## 1; \
+	pxor RK2,		x2 ## 1; \
+	pxor RK3,		x3 ## 1; \
+		pxor RK0,		x0 ## 2; \
+		pxor RK1,		x1 ## 2; \
+		pxor RK2,		x2 ## 2; \
+		pxor RK3,		x3 ## 2;
+
+#define LK2(x0, x1, x2, x3, x4, i) \
+	movdqa x0 ## 1,		x4 ## 1; \
+	pslld $13,		x0 ## 1; \
+	psrld $(32 - 13),	x4 ## 1; \
+	por x4 ## 1,		x0 ## 1; \
+	pxor x0 ## 1,		x1 ## 1; \
+	movdqa x2 ## 1,		x4 ## 1; \
+	pslld $3,		x2 ## 1; \
+	psrld $(32 - 3),	x4 ## 1; \
+	por x4 ## 1,		x2 ## 1; \
+	pxor x2 ## 1,		x1 ## 1; \
+		movdqa x0 ## 2,		x4 ## 2; \
+		pslld $13,		x0 ## 2; \
+		psrld $(32 - 13),	x4 ## 2; \
+		por x4 ## 2,		x0 ## 2; \
+		pxor x0 ## 2,		x1 ## 2; \
+		movdqa x2 ## 2,		x4 ## 2; \
+		pslld $3,		x2 ## 2; \
+		psrld $(32 - 3),	x4 ## 2; \
+		por x4 ## 2,		x2 ## 2; \
+		pxor x2 ## 2,		x1 ## 2; \
+	movdqa x1 ## 1,		x4 ## 1; \
+	pslld $1,		x1 ## 1; \
+	psrld $(32 - 1),	x4 ## 1; \
+	por x4 ## 1,		x1 ## 1; \
+	movdqa x0 ## 1,		x4 ## 1; \
+	pslld $3,		x4 ## 1; \
+	pxor x2 ## 1,		x3 ## 1; \
+	pxor x4 ## 1,		x3 ## 1; \
+	movdqa x3 ## 1,		x4 ## 1; \
+	get_key(i, 1, RK1); \
+		movdqa x1 ## 2,		x4 ## 2; \
+		pslld $1,		x1 ## 2; \
+		psrld $(32 - 1),	x4 ## 2; \
+		por x4 ## 2,		x1 ## 2; \
+		movdqa x0 ## 2,		x4 ## 2; \
+		pslld $3,		x4 ## 2; \
+		pxor x2 ## 2,		x3 ## 2; \
+		pxor x4 ## 2,		x3 ## 2; \
+		movdqa x3 ## 2,		x4 ## 2; \
+		get_key(i, 3, RK3); \
+	pslld $7,		x3 ## 1; \
+	psrld $(32 - 7),	x4 ## 1; \
+	por x4 ## 1,		x3 ## 1; \
+	movdqa x1 ## 1,		x4 ## 1; \
+	pslld $7,		x4 ## 1; \
+	pxor x1 ## 1,		x0 ## 1; \
+	pxor x3 ## 1,		x0 ## 1; \
+	pxor x3 ## 1,		x2 ## 1; \
+	pxor x4 ## 1,		x2 ## 1; \
+	get_key(i, 0, RK0); \
+		pslld $7,		x3 ## 2; \
+		psrld $(32 - 7),	x4 ## 2; \
+		por x4 ## 2,		x3 ## 2; \
+		movdqa x1 ## 2,		x4 ## 2; \
+		pslld $7,		x4 ## 2; \
+		pxor x1 ## 2,		x0 ## 2; \
+		pxor x3 ## 2,		x0 ## 2; \
+		pxor x3 ## 2,		x2 ## 2; \
+		pxor x4 ## 2,		x2 ## 2; \
+		get_key(i, 2, RK2); \
+	pxor RK1,		x1 ## 1; \
+	pxor RK3,		x3 ## 1; \
+	movdqa x0 ## 1,		x4 ## 1; \
+	pslld $5,		x0 ## 1; \
+	psrld $(32 - 5),	x4 ## 1; \
+	por x4 ## 1,		x0 ## 1; \
+	movdqa x2 ## 1,		x4 ## 1; \
+	pslld $22,		x2 ## 1; \
+	psrld $(32 - 22),	x4 ## 1; \
+	por x4 ## 1,		x2 ## 1; \
+	pxor RK0,		x0 ## 1; \
+	pxor RK2,		x2 ## 1; \
+		pxor RK1,		x1 ## 2; \
+		pxor RK3,		x3 ## 2; \
+		movdqa x0 ## 2,		x4 ## 2; \
+		pslld $5,		x0 ## 2; \
+		psrld $(32 - 5),	x4 ## 2; \
+		por x4 ## 2,		x0 ## 2; \
+		movdqa x2 ## 2,		x4 ## 2; \
+		pslld $22,		x2 ## 2; \
+		psrld $(32 - 22),	x4 ## 2; \
+		por x4 ## 2,		x2 ## 2; \
+		pxor RK0,		x0 ## 2; \
+		pxor RK2,		x2 ## 2;
+
+#define KL2(x0, x1, x2, x3, x4, i) \
+	pxor RK0,		x0 ## 1; \
+	pxor RK2,		x2 ## 1; \
+	movdqa x0 ## 1,		x4 ## 1; \
+	psrld $5,		x0 ## 1; \
+	pslld $(32 - 5),	x4 ## 1; \
+	por x4 ## 1,		x0 ## 1; \
+	pxor RK3,		x3 ## 1; \
+	pxor RK1,		x1 ## 1; \
+	movdqa x2 ## 1,		x4 ## 1; \
+	psrld $22,		x2 ## 1; \
+	pslld $(32 - 22),	x4 ## 1; \
+	por x4 ## 1,		x2 ## 1; \
+	pxor x3 ## 1,		x2 ## 1; \
+		pxor RK0,		x0 ## 2; \
+		pxor RK2,		x2 ## 2; \
+		movdqa x0 ## 2,		x4 ## 2; \
+		psrld $5,		x0 ## 2; \
+		pslld $(32 - 5),	x4 ## 2; \
+		por x4 ## 2,		x0 ## 2; \
+		pxor RK3,		x3 ## 2; \
+		pxor RK1,		x1 ## 2; \
+		movdqa x2 ## 2,		x4 ## 2; \
+		psrld $22,		x2 ## 2; \
+		pslld $(32 - 22),	x4 ## 2; \
+		por x4 ## 2,		x2 ## 2; \
+		pxor x3 ## 2,		x2 ## 2; \
+	pxor x3 ## 1,		x0 ## 1; \
+	movdqa x1 ## 1,		x4 ## 1; \
+	pslld $7,		x4 ## 1; \
+	pxor x1 ## 1,		x0 ## 1; \
+	pxor x4 ## 1,		x2 ## 1; \
+	movdqa x1 ## 1,		x4 ## 1; \
+	psrld $1,		x1 ## 1; \
+	pslld $(32 - 1),	x4 ## 1; \
+	por x4 ## 1,		x1 ## 1; \
+		pxor x3 ## 2,		x0 ## 2; \
+		movdqa x1 ## 2,		x4 ## 2; \
+		pslld $7,		x4 ## 2; \
+		pxor x1 ## 2,		x0 ## 2; \
+		pxor x4 ## 2,		x2 ## 2; \
+		movdqa x1 ## 2,		x4 ## 2; \
+		psrld $1,		x1 ## 2; \
+		pslld $(32 - 1),	x4 ## 2; \
+		por x4 ## 2,		x1 ## 2; \
+	movdqa x3 ## 1,		x4 ## 1; \
+	psrld $7,		x3 ## 1; \
+	pslld $(32 - 7),	x4 ## 1; \
+	por x4 ## 1,		x3 ## 1; \
+	pxor x0 ## 1,		x1 ## 1; \
+	movdqa x0 ## 1,		x4 ## 1; \
+	pslld $3,		x4 ## 1; \
+	pxor x4 ## 1,		x3 ## 1; \
+	movdqa x0 ## 1,		x4 ## 1; \
+		movdqa x3 ## 2,		x4 ## 2; \
+		psrld $7,		x3 ## 2; \
+		pslld $(32 - 7),	x4 ## 2; \
+		por x4 ## 2,		x3 ## 2; \
+		pxor x0 ## 2,		x1 ## 2; \
+		movdqa x0 ## 2,		x4 ## 2; \
+		pslld $3,		x4 ## 2; \
+		pxor x4 ## 2,		x3 ## 2; \
+		movdqa x0 ## 2,		x4 ## 2; \
+	psrld $13,		x0 ## 1; \
+	pslld $(32 - 13),	x4 ## 1; \
+	por x4 ## 1,		x0 ## 1; \
+	pxor x2 ## 1,		x1 ## 1; \
+	pxor x2 ## 1,		x3 ## 1; \
+	movdqa x2 ## 1,		x4 ## 1; \
+	psrld $3,		x2 ## 1; \
+	pslld $(32 - 3),	x4 ## 1; \
+	por x4 ## 1,		x2 ## 1; \
+		psrld $13,		x0 ## 2; \
+		pslld $(32 - 13),	x4 ## 2; \
+		por x4 ## 2,		x0 ## 2; \
+		pxor x2 ## 2,		x1 ## 2; \
+		pxor x2 ## 2,		x3 ## 2; \
+		movdqa x2 ## 2,		x4 ## 2; \
+		psrld $3,		x2 ## 2; \
+		pslld $(32 - 3),	x4 ## 2; \
+		por x4 ## 2,		x2 ## 2;
+
+#define S(SBOX, x0, x1, x2, x3, x4) \
+	SBOX ## _1(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	SBOX ## _2(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	SBOX ## _1(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2); \
+	SBOX ## _2(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2);
+
+#define SP(SBOX, x0, x1, x2, x3, x4, i) \
+	get_key(i, 0, RK0); \
+	SBOX ## _1(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	get_key(i, 2, RK2); \
+	SBOX ## _1(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2); \
+	get_key(i, 3, RK3); \
+	SBOX ## _2(x0 ## 1, x1 ## 1, x2 ## 1, x3 ## 1, x4 ## 1); \
+	get_key(i, 1, RK1); \
+	SBOX ## _2(x0 ## 2, x1 ## 2, x2 ## 2, x3 ## 2, x4 ## 2); \
+
+#define transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	movdqa x0,		t2; \
+	punpckldq x1,		x0; \
+	punpckhdq x1,		t2; \
+	movdqa x2,		t1; \
+	punpckhdq x3,		x2; \
+	punpckldq x3,		t1; \
+	movdqa x0,		x1; \
+	punpcklqdq t1,		x0; \
+	punpckhqdq t1,		x1; \
+	movdqa t2,		x3; \
+	punpcklqdq x2,		t2; \
+	punpckhqdq x2,		x3; \
+	movdqa t2,		x2;
+
+#define read_blocks(in, x0, x1, x2, x3, t0, t1, t2) \
+	movdqu (0*4*4)(in),	x0; \
+	movdqu (1*4*4)(in),	x1; \
+	movdqu (2*4*4)(in),	x2; \
+	movdqu (3*4*4)(in),	x3; \
+	\
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2)
+
+#define write_blocks(out, x0, x1, x2, x3, t0, t1, t2) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	\
+	movdqu x0,		(0*4*4)(out); \
+	movdqu x1,		(1*4*4)(out); \
+	movdqu x2,		(2*4*4)(out); \
+	movdqu x3,		(3*4*4)(out);
+
+#define xor_blocks(out, x0, x1, x2, x3, t0, t1, t2) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	\
+	movdqu (0*4*4)(out),	t0; \
+	pxor t0,		x0; \
+	movdqu x0,		(0*4*4)(out); \
+	movdqu (1*4*4)(out),	t0; \
+	pxor t0,		x1; \
+	movdqu x1,		(1*4*4)(out); \
+	movdqu (2*4*4)(out),	t0; \
+	pxor t0,		x2; \
+	movdqu x2,		(2*4*4)(out); \
+	movdqu (3*4*4)(out),	t0; \
+	pxor t0,		x3; \
+	movdqu x3,		(3*4*4)(out);
+
+SYM_FUNC_START(__serpent_enc_blk_8way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 *	%rcx: bool, if true: xor output
+	 */
+
+	pcmpeqd RNOT, RNOT;
+
+	leaq (4*4*4)(%rdx), %rax;
+	read_blocks(%rdx, RA1, RB1, RC1, RD1, RK0, RK1, RK2);
+	read_blocks(%rax, RA2, RB2, RC2, RD2, RK0, RK1, RK2);
+
+						 K2(RA, RB, RC, RD, RE, 0);
+	S(S0, RA, RB, RC, RD, RE);		LK2(RC, RB, RD, RA, RE, 1);
+	S(S1, RC, RB, RD, RA, RE);		LK2(RE, RD, RA, RC, RB, 2);
+	S(S2, RE, RD, RA, RC, RB);		LK2(RB, RD, RE, RC, RA, 3);
+	S(S3, RB, RD, RE, RC, RA);		LK2(RC, RA, RD, RB, RE, 4);
+	S(S4, RC, RA, RD, RB, RE);		LK2(RA, RD, RB, RE, RC, 5);
+	S(S5, RA, RD, RB, RE, RC);		LK2(RC, RA, RD, RE, RB, 6);
+	S(S6, RC, RA, RD, RE, RB);		LK2(RD, RB, RA, RE, RC, 7);
+	S(S7, RD, RB, RA, RE, RC);		LK2(RC, RA, RE, RD, RB, 8);
+	S(S0, RC, RA, RE, RD, RB);		LK2(RE, RA, RD, RC, RB, 9);
+	S(S1, RE, RA, RD, RC, RB);		LK2(RB, RD, RC, RE, RA, 10);
+	S(S2, RB, RD, RC, RE, RA);		LK2(RA, RD, RB, RE, RC, 11);
+	S(S3, RA, RD, RB, RE, RC);		LK2(RE, RC, RD, RA, RB, 12);
+	S(S4, RE, RC, RD, RA, RB);		LK2(RC, RD, RA, RB, RE, 13);
+	S(S5, RC, RD, RA, RB, RE);		LK2(RE, RC, RD, RB, RA, 14);
+	S(S6, RE, RC, RD, RB, RA);		LK2(RD, RA, RC, RB, RE, 15);
+	S(S7, RD, RA, RC, RB, RE);		LK2(RE, RC, RB, RD, RA, 16);
+	S(S0, RE, RC, RB, RD, RA);		LK2(RB, RC, RD, RE, RA, 17);
+	S(S1, RB, RC, RD, RE, RA);		LK2(RA, RD, RE, RB, RC, 18);
+	S(S2, RA, RD, RE, RB, RC);		LK2(RC, RD, RA, RB, RE, 19);
+	S(S3, RC, RD, RA, RB, RE);		LK2(RB, RE, RD, RC, RA, 20);
+	S(S4, RB, RE, RD, RC, RA);		LK2(RE, RD, RC, RA, RB, 21);
+	S(S5, RE, RD, RC, RA, RB);		LK2(RB, RE, RD, RA, RC, 22);
+	S(S6, RB, RE, RD, RA, RC);		LK2(RD, RC, RE, RA, RB, 23);
+	S(S7, RD, RC, RE, RA, RB);		LK2(RB, RE, RA, RD, RC, 24);
+	S(S0, RB, RE, RA, RD, RC);		LK2(RA, RE, RD, RB, RC, 25);
+	S(S1, RA, RE, RD, RB, RC);		LK2(RC, RD, RB, RA, RE, 26);
+	S(S2, RC, RD, RB, RA, RE);		LK2(RE, RD, RC, RA, RB, 27);
+	S(S3, RE, RD, RC, RA, RB);		LK2(RA, RB, RD, RE, RC, 28);
+	S(S4, RA, RB, RD, RE, RC);		LK2(RB, RD, RE, RC, RA, 29);
+	S(S5, RB, RD, RE, RC, RA);		LK2(RA, RB, RD, RC, RE, 30);
+	S(S6, RA, RB, RD, RC, RE);		LK2(RD, RE, RB, RC, RA, 31);
+	S(S7, RD, RE, RB, RC, RA);		 K2(RA, RB, RC, RD, RE, 32);
+
+	leaq (4*4*4)(%rsi), %rax;
+
+	testb %cl, %cl;
+	jnz .L__enc_xor8;
+
+	write_blocks(%rsi, RA1, RB1, RC1, RD1, RK0, RK1, RK2);
+	write_blocks(%rax, RA2, RB2, RC2, RD2, RK0, RK1, RK2);
+
+	RET;
+
+.L__enc_xor8:
+	xor_blocks(%rsi, RA1, RB1, RC1, RD1, RK0, RK1, RK2);
+	xor_blocks(%rax, RA2, RB2, RC2, RD2, RK0, RK1, RK2);
+
+	RET;
+SYM_FUNC_END(__serpent_enc_blk_8way)
+
+SYM_FUNC_START(serpent_dec_blk_8way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+
+	pcmpeqd RNOT, RNOT;
+
+	leaq (4*4*4)(%rdx), %rax;
+	read_blocks(%rdx, RA1, RB1, RC1, RD1, RK0, RK1, RK2);
+	read_blocks(%rax, RA2, RB2, RC2, RD2, RK0, RK1, RK2);
+
+						 K2(RA, RB, RC, RD, RE, 32);
+	SP(SI7, RA, RB, RC, RD, RE, 31);	KL2(RB, RD, RA, RE, RC, 31);
+	SP(SI6, RB, RD, RA, RE, RC, 30);	KL2(RA, RC, RE, RB, RD, 30);
+	SP(SI5, RA, RC, RE, RB, RD, 29);	KL2(RC, RD, RA, RE, RB, 29);
+	SP(SI4, RC, RD, RA, RE, RB, 28);	KL2(RC, RA, RB, RE, RD, 28);
+	SP(SI3, RC, RA, RB, RE, RD, 27);	KL2(RB, RC, RD, RE, RA, 27);
+	SP(SI2, RB, RC, RD, RE, RA, 26);	KL2(RC, RA, RE, RD, RB, 26);
+	SP(SI1, RC, RA, RE, RD, RB, 25);	KL2(RB, RA, RE, RD, RC, 25);
+	SP(SI0, RB, RA, RE, RD, RC, 24);	KL2(RE, RC, RA, RB, RD, 24);
+	SP(SI7, RE, RC, RA, RB, RD, 23);	KL2(RC, RB, RE, RD, RA, 23);
+	SP(SI6, RC, RB, RE, RD, RA, 22);	KL2(RE, RA, RD, RC, RB, 22);
+	SP(SI5, RE, RA, RD, RC, RB, 21);	KL2(RA, RB, RE, RD, RC, 21);
+	SP(SI4, RA, RB, RE, RD, RC, 20);	KL2(RA, RE, RC, RD, RB, 20);
+	SP(SI3, RA, RE, RC, RD, RB, 19);	KL2(RC, RA, RB, RD, RE, 19);
+	SP(SI2, RC, RA, RB, RD, RE, 18);	KL2(RA, RE, RD, RB, RC, 18);
+	SP(SI1, RA, RE, RD, RB, RC, 17);	KL2(RC, RE, RD, RB, RA, 17);
+	SP(SI0, RC, RE, RD, RB, RA, 16);	KL2(RD, RA, RE, RC, RB, 16);
+	SP(SI7, RD, RA, RE, RC, RB, 15);	KL2(RA, RC, RD, RB, RE, 15);
+	SP(SI6, RA, RC, RD, RB, RE, 14);	KL2(RD, RE, RB, RA, RC, 14);
+	SP(SI5, RD, RE, RB, RA, RC, 13);	KL2(RE, RC, RD, RB, RA, 13);
+	SP(SI4, RE, RC, RD, RB, RA, 12);	KL2(RE, RD, RA, RB, RC, 12);
+	SP(SI3, RE, RD, RA, RB, RC, 11);	KL2(RA, RE, RC, RB, RD, 11);
+	SP(SI2, RA, RE, RC, RB, RD, 10);	KL2(RE, RD, RB, RC, RA, 10);
+	SP(SI1, RE, RD, RB, RC, RA, 9);		KL2(RA, RD, RB, RC, RE, 9);
+	SP(SI0, RA, RD, RB, RC, RE, 8);		KL2(RB, RE, RD, RA, RC, 8);
+	SP(SI7, RB, RE, RD, RA, RC, 7);		KL2(RE, RA, RB, RC, RD, 7);
+	SP(SI6, RE, RA, RB, RC, RD, 6);		KL2(RB, RD, RC, RE, RA, 6);
+	SP(SI5, RB, RD, RC, RE, RA, 5);		KL2(RD, RA, RB, RC, RE, 5);
+	SP(SI4, RD, RA, RB, RC, RE, 4);		KL2(RD, RB, RE, RC, RA, 4);
+	SP(SI3, RD, RB, RE, RC, RA, 3);		KL2(RE, RD, RA, RC, RB, 3);
+	SP(SI2, RE, RD, RA, RC, RB, 2);		KL2(RD, RB, RC, RA, RE, 2);
+	SP(SI1, RD, RB, RC, RA, RE, 1);		KL2(RE, RB, RC, RA, RD, 1);
+	S(SI0, RE, RB, RC, RA, RD);		 K2(RC, RD, RB, RE, RA, 0);
+
+	leaq (4*4*4)(%rsi), %rax;
+	write_blocks(%rsi, RC1, RD1, RB1, RE1, RK0, RK1, RK2);
+	write_blocks(%rax, RC2, RD2, RB2, RE2, RK0, RK1, RK2);
+
+	RET;
+SYM_FUNC_END(serpent_dec_blk_8way)
diff --git a/arch/x86/crypto/serpent-sse2.h b/arch/x86/crypto/serpent-sse2.h
new file mode 100644
index 000000000..860ca2489
--- /dev/null
+++ b/arch/x86/crypto/serpent-sse2.h
@@ -0,0 +1,60 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef ASM_X86_SERPENT_SSE2_H
+#define ASM_X86_SERPENT_SSE2_H
+
+#include <linux/crypto.h>
+#include <crypto/serpent.h>
+
+#ifdef CONFIG_X86_32
+
+#define SERPENT_PARALLEL_BLOCKS 4
+
+asmlinkage void __serpent_enc_blk_4way(const struct serpent_ctx *ctx, u8 *dst,
+				       const u8 *src, bool xor);
+asmlinkage void serpent_dec_blk_4way(const struct serpent_ctx *ctx, u8 *dst,
+				     const u8 *src);
+
+static inline void serpent_enc_blk_xway(const void *ctx, u8 *dst, const u8 *src)
+{
+	__serpent_enc_blk_4way(ctx, dst, src, false);
+}
+
+static inline void serpent_enc_blk_xway_xor(const struct serpent_ctx *ctx,
+					    u8 *dst, const u8 *src)
+{
+	__serpent_enc_blk_4way(ctx, dst, src, true);
+}
+
+static inline void serpent_dec_blk_xway(const void *ctx, u8 *dst, const u8 *src)
+{
+	serpent_dec_blk_4way(ctx, dst, src);
+}
+
+#else
+
+#define SERPENT_PARALLEL_BLOCKS 8
+
+asmlinkage void __serpent_enc_blk_8way(const struct serpent_ctx *ctx, u8 *dst,
+				       const u8 *src, bool xor);
+asmlinkage void serpent_dec_blk_8way(const struct serpent_ctx *ctx, u8 *dst,
+				     const u8 *src);
+
+static inline void serpent_enc_blk_xway(const void *ctx, u8 *dst, const u8 *src)
+{
+	__serpent_enc_blk_8way(ctx, dst, src, false);
+}
+
+static inline void serpent_enc_blk_xway_xor(const struct serpent_ctx *ctx,
+					    u8 *dst, const u8 *src)
+{
+	__serpent_enc_blk_8way(ctx, dst, src, true);
+}
+
+static inline void serpent_dec_blk_xway(const void *ctx, u8 *dst, const u8 *src)
+{
+	serpent_dec_blk_8way(ctx, dst, src);
+}
+
+#endif
+
+#endif
diff --git a/arch/x86/crypto/serpent_avx2_glue.c b/arch/x86/crypto/serpent_avx2_glue.c
new file mode 100644
index 000000000..f5f2121b7
--- /dev/null
+++ b/arch/x86/crypto/serpent_avx2_glue.c
@@ -0,0 +1,123 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Glue Code for x86_64/AVX2 assembler optimized version of Serpent
+ *
+ * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ */
+
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <crypto/algapi.h>
+#include <crypto/serpent.h>
+
+#include "serpent-avx.h"
+#include "ecb_cbc_helpers.h"
+
+#define SERPENT_AVX2_PARALLEL_BLOCKS 16
+
+/* 16-way AVX2 parallel cipher functions */
+asmlinkage void serpent_ecb_enc_16way(const void *ctx, u8 *dst, const u8 *src);
+asmlinkage void serpent_ecb_dec_16way(const void *ctx, u8 *dst, const u8 *src);
+asmlinkage void serpent_cbc_dec_16way(const void *ctx, u8 *dst, const u8 *src);
+
+static int serpent_setkey_skcipher(struct crypto_skcipher *tfm,
+				   const u8 *key, unsigned int keylen)
+{
+	return __serpent_setkey(crypto_skcipher_ctx(tfm), key, keylen);
+}
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, SERPENT_BLOCK_SIZE, SERPENT_PARALLEL_BLOCKS);
+	ECB_BLOCK(SERPENT_AVX2_PARALLEL_BLOCKS, serpent_ecb_enc_16way);
+	ECB_BLOCK(SERPENT_PARALLEL_BLOCKS, serpent_ecb_enc_8way_avx);
+	ECB_BLOCK(1, __serpent_encrypt);
+	ECB_WALK_END();
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, SERPENT_BLOCK_SIZE, SERPENT_PARALLEL_BLOCKS);
+	ECB_BLOCK(SERPENT_AVX2_PARALLEL_BLOCKS, serpent_ecb_dec_16way);
+	ECB_BLOCK(SERPENT_PARALLEL_BLOCKS, serpent_ecb_dec_8way_avx);
+	ECB_BLOCK(1, __serpent_decrypt);
+	ECB_WALK_END();
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, SERPENT_BLOCK_SIZE, -1);
+	CBC_ENC_BLOCK(__serpent_encrypt);
+	CBC_WALK_END();
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, SERPENT_BLOCK_SIZE, SERPENT_PARALLEL_BLOCKS);
+	CBC_DEC_BLOCK(SERPENT_AVX2_PARALLEL_BLOCKS, serpent_cbc_dec_16way);
+	CBC_DEC_BLOCK(SERPENT_PARALLEL_BLOCKS, serpent_cbc_dec_8way_avx);
+	CBC_DEC_BLOCK(1, __serpent_decrypt);
+	CBC_WALK_END();
+}
+
+static struct skcipher_alg serpent_algs[] = {
+	{
+		.base.cra_name		= "ecb(serpent)",
+		.base.cra_driver_name	= "ecb-serpent-avx2",
+		.base.cra_priority	= 600,
+		.base.cra_blocksize	= SERPENT_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct serpent_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= SERPENT_MIN_KEY_SIZE,
+		.max_keysize		= SERPENT_MAX_KEY_SIZE,
+		.setkey			= serpent_setkey_skcipher,
+		.encrypt		= ecb_encrypt,
+		.decrypt		= ecb_decrypt,
+	}, {
+		.base.cra_name		= "cbc(serpent)",
+		.base.cra_driver_name	= "cbc-serpent-avx2",
+		.base.cra_priority	= 600,
+		.base.cra_blocksize	= SERPENT_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct serpent_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= SERPENT_MIN_KEY_SIZE,
+		.max_keysize		= SERPENT_MAX_KEY_SIZE,
+		.ivsize			= SERPENT_BLOCK_SIZE,
+		.setkey			= serpent_setkey_skcipher,
+		.encrypt		= cbc_encrypt,
+		.decrypt		= cbc_decrypt,
+	},
+};
+
+static int __init serpent_avx2_init(void)
+{
+	const char *feature_name;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX2) || !boot_cpu_has(X86_FEATURE_OSXSAVE)) {
+		pr_info("AVX2 instructions are not detected.\n");
+		return -ENODEV;
+	}
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM,
+				&feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	return crypto_register_skciphers(serpent_algs,
+					 ARRAY_SIZE(serpent_algs));
+}
+
+static void __exit serpent_avx2_fini(void)
+{
+	crypto_unregister_skciphers(serpent_algs, ARRAY_SIZE(serpent_algs));
+}
+
+module_init(serpent_avx2_init);
+module_exit(serpent_avx2_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Serpent Cipher Algorithm, AVX2 optimized");
+MODULE_ALIAS_CRYPTO("serpent");
+MODULE_ALIAS_CRYPTO("serpent-asm");
diff --git a/arch/x86/crypto/serpent_avx_glue.c b/arch/x86/crypto/serpent_avx_glue.c
new file mode 100644
index 000000000..9c8b3a335
--- /dev/null
+++ b/arch/x86/crypto/serpent_avx_glue.c
@@ -0,0 +1,125 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Glue Code for AVX assembler versions of Serpent Cipher
+ *
+ * Copyright (C) 2012 Johannes Goetzfried
+ *     <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
+ *
+ * Copyright © 2011-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ */
+
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <linux/export.h>
+#include <crypto/algapi.h>
+#include <crypto/serpent.h>
+
+#include "serpent-avx.h"
+#include "ecb_cbc_helpers.h"
+
+/* 8-way parallel cipher functions */
+asmlinkage void serpent_ecb_enc_8way_avx(const void *ctx, u8 *dst,
+					 const u8 *src);
+EXPORT_SYMBOL_GPL(serpent_ecb_enc_8way_avx);
+
+asmlinkage void serpent_ecb_dec_8way_avx(const void *ctx, u8 *dst,
+					 const u8 *src);
+EXPORT_SYMBOL_GPL(serpent_ecb_dec_8way_avx);
+
+asmlinkage void serpent_cbc_dec_8way_avx(const void *ctx, u8 *dst,
+					 const u8 *src);
+EXPORT_SYMBOL_GPL(serpent_cbc_dec_8way_avx);
+
+static int serpent_setkey_skcipher(struct crypto_skcipher *tfm,
+				   const u8 *key, unsigned int keylen)
+{
+	return __serpent_setkey(crypto_skcipher_ctx(tfm), key, keylen);
+}
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, SERPENT_BLOCK_SIZE, SERPENT_PARALLEL_BLOCKS);
+	ECB_BLOCK(SERPENT_PARALLEL_BLOCKS, serpent_ecb_enc_8way_avx);
+	ECB_BLOCK(1, __serpent_encrypt);
+	ECB_WALK_END();
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, SERPENT_BLOCK_SIZE, SERPENT_PARALLEL_BLOCKS);
+	ECB_BLOCK(SERPENT_PARALLEL_BLOCKS, serpent_ecb_dec_8way_avx);
+	ECB_BLOCK(1, __serpent_decrypt);
+	ECB_WALK_END();
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, SERPENT_BLOCK_SIZE, -1);
+	CBC_ENC_BLOCK(__serpent_encrypt);
+	CBC_WALK_END();
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, SERPENT_BLOCK_SIZE, SERPENT_PARALLEL_BLOCKS);
+	CBC_DEC_BLOCK(SERPENT_PARALLEL_BLOCKS, serpent_cbc_dec_8way_avx);
+	CBC_DEC_BLOCK(1, __serpent_decrypt);
+	CBC_WALK_END();
+}
+
+static struct skcipher_alg serpent_algs[] = {
+	{
+		.base.cra_name		= "ecb(serpent)",
+		.base.cra_driver_name	= "ecb-serpent-avx",
+		.base.cra_priority	= 500,
+		.base.cra_blocksize	= SERPENT_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct serpent_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= SERPENT_MIN_KEY_SIZE,
+		.max_keysize		= SERPENT_MAX_KEY_SIZE,
+		.setkey			= serpent_setkey_skcipher,
+		.encrypt		= ecb_encrypt,
+		.decrypt		= ecb_decrypt,
+	}, {
+		.base.cra_name		= "cbc(serpent)",
+		.base.cra_driver_name	= "cbc-serpent-avx",
+		.base.cra_priority	= 500,
+		.base.cra_blocksize	= SERPENT_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct serpent_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= SERPENT_MIN_KEY_SIZE,
+		.max_keysize		= SERPENT_MAX_KEY_SIZE,
+		.ivsize			= SERPENT_BLOCK_SIZE,
+		.setkey			= serpent_setkey_skcipher,
+		.encrypt		= cbc_encrypt,
+		.decrypt		= cbc_decrypt,
+	},
+};
+
+static int __init serpent_init(void)
+{
+	const char *feature_name;
+
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM,
+				&feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	return crypto_register_skciphers(serpent_algs,
+					 ARRAY_SIZE(serpent_algs));
+}
+
+static void __exit serpent_exit(void)
+{
+	crypto_unregister_skciphers(serpent_algs, ARRAY_SIZE(serpent_algs));
+}
+
+module_init(serpent_init);
+module_exit(serpent_exit);
+
+MODULE_DESCRIPTION("Serpent Cipher Algorithm, AVX optimized");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_CRYPTO("serpent");
diff --git a/arch/x86/crypto/serpent_sse2_glue.c b/arch/x86/crypto/serpent_sse2_glue.c
new file mode 100644
index 000000000..80ee17ec2
--- /dev/null
+++ b/arch/x86/crypto/serpent_sse2_glue.c
@@ -0,0 +1,124 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Glue Code for SSE2 assembler versions of Serpent Cipher
+ *
+ * Copyright (c) 2011 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ *
+ * Glue code based on aesni-intel_glue.c by:
+ *  Copyright (C) 2008, Intel Corp.
+ *    Author: Huang Ying <ying.huang@intel.com>
+ *
+ * CBC & ECB parts based on code (crypto/cbc.c,ecb.c) by:
+ *   Copyright (c) 2006 Herbert Xu <herbert@gondor.apana.org.au>
+ */
+
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <crypto/algapi.h>
+#include <crypto/b128ops.h>
+#include <crypto/serpent.h>
+
+#include "serpent-sse2.h"
+#include "ecb_cbc_helpers.h"
+
+static int serpent_setkey_skcipher(struct crypto_skcipher *tfm,
+				   const u8 *key, unsigned int keylen)
+{
+	return __serpent_setkey(crypto_skcipher_ctx(tfm), key, keylen);
+}
+
+static void serpent_decrypt_cbc_xway(const void *ctx, u8 *dst, const u8 *src)
+{
+	u8 buf[SERPENT_PARALLEL_BLOCKS - 1][SERPENT_BLOCK_SIZE];
+	const u8 *s = src;
+
+	if (dst == src)
+		s = memcpy(buf, src, sizeof(buf));
+	serpent_dec_blk_xway(ctx, dst, src);
+	crypto_xor(dst + SERPENT_BLOCK_SIZE, s, sizeof(buf));
+}
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, SERPENT_BLOCK_SIZE, SERPENT_PARALLEL_BLOCKS);
+	ECB_BLOCK(SERPENT_PARALLEL_BLOCKS, serpent_enc_blk_xway);
+	ECB_BLOCK(1, __serpent_encrypt);
+	ECB_WALK_END();
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, SERPENT_BLOCK_SIZE, SERPENT_PARALLEL_BLOCKS);
+	ECB_BLOCK(SERPENT_PARALLEL_BLOCKS, serpent_dec_blk_xway);
+	ECB_BLOCK(1, __serpent_decrypt);
+	ECB_WALK_END();
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, SERPENT_BLOCK_SIZE, -1);
+	CBC_ENC_BLOCK(__serpent_encrypt);
+	CBC_WALK_END();
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, SERPENT_BLOCK_SIZE, SERPENT_PARALLEL_BLOCKS);
+	CBC_DEC_BLOCK(SERPENT_PARALLEL_BLOCKS, serpent_decrypt_cbc_xway);
+	CBC_DEC_BLOCK(1, __serpent_decrypt);
+	CBC_WALK_END();
+}
+
+static struct skcipher_alg serpent_algs[] = {
+	{
+		.base.cra_name		= "ecb(serpent)",
+		.base.cra_driver_name	= "ecb-serpent-sse2",
+		.base.cra_priority	= 400,
+		.base.cra_blocksize	= SERPENT_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct serpent_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= SERPENT_MIN_KEY_SIZE,
+		.max_keysize		= SERPENT_MAX_KEY_SIZE,
+		.setkey			= serpent_setkey_skcipher,
+		.encrypt		= ecb_encrypt,
+		.decrypt		= ecb_decrypt,
+	}, {
+		.base.cra_name		= "cbc(serpent)",
+		.base.cra_driver_name	= "cbc-serpent-sse2",
+		.base.cra_priority	= 400,
+		.base.cra_blocksize	= SERPENT_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct serpent_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= SERPENT_MIN_KEY_SIZE,
+		.max_keysize		= SERPENT_MAX_KEY_SIZE,
+		.ivsize			= SERPENT_BLOCK_SIZE,
+		.setkey			= serpent_setkey_skcipher,
+		.encrypt		= cbc_encrypt,
+		.decrypt		= cbc_decrypt,
+	},
+};
+
+static int __init serpent_sse2_init(void)
+{
+	if (!boot_cpu_has(X86_FEATURE_XMM2)) {
+		printk(KERN_INFO "SSE2 instructions are not detected.\n");
+		return -ENODEV;
+	}
+
+	return crypto_register_skciphers(serpent_algs,
+					 ARRAY_SIZE(serpent_algs));
+}
+
+static void __exit serpent_sse2_exit(void)
+{
+	crypto_unregister_skciphers(serpent_algs, ARRAY_SIZE(serpent_algs));
+}
+
+module_init(serpent_sse2_init);
+module_exit(serpent_sse2_exit);
+
+MODULE_DESCRIPTION("Serpent Cipher Algorithm, SSE2 optimized");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_CRYPTO("serpent");
diff --git a/arch/x86/crypto/sm4-aesni-avx-asm_64.S b/arch/x86/crypto/sm4-aesni-avx-asm_64.S
new file mode 100644
index 000000000..2bf611eaa
--- /dev/null
+++ b/arch/x86/crypto/sm4-aesni-avx-asm_64.S
@@ -0,0 +1,536 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * SM4 Cipher Algorithm, AES-NI/AVX optimized.
+ * as specified in
+ * https://tools.ietf.org/id/draft-ribose-cfrg-sm4-10.html
+ *
+ * Copyright (C) 2018 Markku-Juhani O. Saarinen <mjos@iki.fi>
+ * Copyright (C) 2020 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ * Copyright (c) 2021 Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
+ */
+
+/* Based on SM4 AES-NI work by libgcrypt and Markku-Juhani O. Saarinen at:
+ *  https://github.com/mjosaarinen/sm4ni
+ */
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+#include <asm/frame.h>
+
+#define rRIP         (%rip)
+
+#define RX0          %xmm0
+#define RX1          %xmm1
+#define MASK_4BIT    %xmm2
+#define RTMP0        %xmm3
+#define RTMP1        %xmm4
+#define RTMP2        %xmm5
+#define RTMP3        %xmm6
+#define RTMP4        %xmm7
+
+#define RA0          %xmm8
+#define RA1          %xmm9
+#define RA2          %xmm10
+#define RA3          %xmm11
+
+#define RB0          %xmm12
+#define RB1          %xmm13
+#define RB2          %xmm14
+#define RB3          %xmm15
+
+#define RNOT         %xmm0
+#define RBSWAP       %xmm1
+
+
+/* Transpose four 32-bit words between 128-bit vectors. */
+#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
+	vpunpckhdq x1, x0, t2;                \
+	vpunpckldq x1, x0, x0;                \
+	                                      \
+	vpunpckldq x3, x2, t1;                \
+	vpunpckhdq x3, x2, x2;                \
+	                                      \
+	vpunpckhqdq t1, x0, x1;               \
+	vpunpcklqdq t1, x0, x0;               \
+	                                      \
+	vpunpckhqdq x2, t2, x3;               \
+	vpunpcklqdq x2, t2, x2;
+
+/* pre-SubByte transform. */
+#define transform_pre(x, lo_t, hi_t, mask4bit, tmp0) \
+	vpand x, mask4bit, tmp0;                     \
+	vpandn x, mask4bit, x;                       \
+	vpsrld $4, x, x;                             \
+	                                             \
+	vpshufb tmp0, lo_t, tmp0;                    \
+	vpshufb x, hi_t, x;                          \
+	vpxor tmp0, x, x;
+
+/* post-SubByte transform. Note: x has been XOR'ed with mask4bit by
+ * 'vaeslastenc' instruction.
+ */
+#define transform_post(x, lo_t, hi_t, mask4bit, tmp0) \
+	vpandn mask4bit, x, tmp0;                     \
+	vpsrld $4, x, x;                              \
+	vpand x, mask4bit, x;                         \
+	                                              \
+	vpshufb tmp0, lo_t, tmp0;                     \
+	vpshufb x, hi_t, x;                           \
+	vpxor tmp0, x, x;
+
+
+.section	.rodata.cst16, "aM", @progbits, 16
+.align 16
+
+/*
+ * Following four affine transform look-up tables are from work by
+ * Markku-Juhani O. Saarinen, at https://github.com/mjosaarinen/sm4ni
+ *
+ * These allow exposing SM4 S-Box from AES SubByte.
+ */
+
+/* pre-SubByte affine transform, from SM4 field to AES field. */
+.Lpre_tf_lo_s:
+	.quad 0x9197E2E474720701, 0xC7C1B4B222245157
+.Lpre_tf_hi_s:
+	.quad 0xE240AB09EB49A200, 0xF052B91BF95BB012
+
+/* post-SubByte affine transform, from AES field to SM4 field. */
+.Lpost_tf_lo_s:
+	.quad 0x5B67F2CEA19D0834, 0xEDD14478172BBE82
+.Lpost_tf_hi_s:
+	.quad 0xAE7201DD73AFDC00, 0x11CDBE62CC1063BF
+
+/* For isolating SubBytes from AESENCLAST, inverse shift row */
+.Linv_shift_row:
+	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
+	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
+
+/* Inverse shift row + Rotate left by 8 bits on 32-bit words with vpshufb */
+.Linv_shift_row_rol_8:
+	.byte 0x07, 0x00, 0x0d, 0x0a, 0x0b, 0x04, 0x01, 0x0e
+	.byte 0x0f, 0x08, 0x05, 0x02, 0x03, 0x0c, 0x09, 0x06
+
+/* Inverse shift row + Rotate left by 16 bits on 32-bit words with vpshufb */
+.Linv_shift_row_rol_16:
+	.byte 0x0a, 0x07, 0x00, 0x0d, 0x0e, 0x0b, 0x04, 0x01
+	.byte 0x02, 0x0f, 0x08, 0x05, 0x06, 0x03, 0x0c, 0x09
+
+/* Inverse shift row + Rotate left by 24 bits on 32-bit words with vpshufb */
+.Linv_shift_row_rol_24:
+	.byte 0x0d, 0x0a, 0x07, 0x00, 0x01, 0x0e, 0x0b, 0x04
+	.byte 0x05, 0x02, 0x0f, 0x08, 0x09, 0x06, 0x03, 0x0c
+
+/* For CTR-mode IV byteswap */
+.Lbswap128_mask:
+	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+
+/* For input word byte-swap */
+.Lbswap32_mask:
+	.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
+
+.align 4
+/* 4-bit mask */
+.L0f0f0f0f:
+	.long 0x0f0f0f0f
+
+/* 12 bytes, only for padding */
+.Lpadding_deadbeef:
+	.long 0xdeadbeef, 0xdeadbeef, 0xdeadbeef
+
+
+.text
+
+/*
+ * void sm4_aesni_avx_crypt4(const u32 *rk, u8 *dst,
+ *                           const u8 *src, int nblocks)
+ */
+SYM_FUNC_START(sm4_aesni_avx_crypt4)
+	/* input:
+	 *	%rdi: round key array, CTX
+	 *	%rsi: dst (1..4 blocks)
+	 *	%rdx: src (1..4 blocks)
+	 *	%rcx: num blocks (1..4)
+	 */
+	FRAME_BEGIN
+
+	vmovdqu 0*16(%rdx), RA0;
+	vmovdqa RA0, RA1;
+	vmovdqa RA0, RA2;
+	vmovdqa RA0, RA3;
+	cmpq $2, %rcx;
+	jb .Lblk4_load_input_done;
+	vmovdqu 1*16(%rdx), RA1;
+	je .Lblk4_load_input_done;
+	vmovdqu 2*16(%rdx), RA2;
+	cmpq $3, %rcx;
+	je .Lblk4_load_input_done;
+	vmovdqu 3*16(%rdx), RA3;
+
+.Lblk4_load_input_done:
+
+	vmovdqa .Lbswap32_mask rRIP, RTMP2;
+	vpshufb RTMP2, RA0, RA0;
+	vpshufb RTMP2, RA1, RA1;
+	vpshufb RTMP2, RA2, RA2;
+	vpshufb RTMP2, RA3, RA3;
+
+	vbroadcastss .L0f0f0f0f rRIP, MASK_4BIT;
+	vmovdqa .Lpre_tf_lo_s rRIP, RTMP4;
+	vmovdqa .Lpre_tf_hi_s rRIP, RB0;
+	vmovdqa .Lpost_tf_lo_s rRIP, RB1;
+	vmovdqa .Lpost_tf_hi_s rRIP, RB2;
+	vmovdqa .Linv_shift_row rRIP, RB3;
+	vmovdqa .Linv_shift_row_rol_8 rRIP, RTMP2;
+	vmovdqa .Linv_shift_row_rol_16 rRIP, RTMP3;
+	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
+
+#define ROUND(round, s0, s1, s2, s3)                                \
+	vbroadcastss (4*(round))(%rdi), RX0;                        \
+	vpxor s1, RX0, RX0;                                         \
+	vpxor s2, RX0, RX0;                                         \
+	vpxor s3, RX0, RX0; /* s1 ^ s2 ^ s3 ^ rk */                 \
+	                                                            \
+	/* sbox, non-linear part */                                 \
+	transform_pre(RX0, RTMP4, RB0, MASK_4BIT, RTMP0);           \
+	vaesenclast MASK_4BIT, RX0, RX0;                            \
+	transform_post(RX0, RB1, RB2, MASK_4BIT, RTMP0);            \
+	                                                            \
+	/* linear part */                                           \
+	vpshufb RB3, RX0, RTMP0;                                    \
+	vpxor RTMP0, s0, s0; /* s0 ^ x */                           \
+	vpshufb RTMP2, RX0, RTMP1;                                  \
+	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) */               \
+	vpshufb RTMP3, RX0, RTMP1;                                  \
+	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) ^ rol(x,16) */   \
+	vpshufb .Linv_shift_row_rol_24 rRIP, RX0, RTMP1;            \
+	vpxor RTMP1, s0, s0; /* s0 ^ x ^ rol(x,24) */               \
+	vpslld $2, RTMP0, RTMP1;                                    \
+	vpsrld $30, RTMP0, RTMP0;                                   \
+	vpxor RTMP0, s0, s0;                                        \
+	/* s0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */ \
+	vpxor RTMP1, s0, s0;
+
+	leaq (32*4)(%rdi), %rax;
+.align 16
+.Lroundloop_blk4:
+	ROUND(0, RA0, RA1, RA2, RA3);
+	ROUND(1, RA1, RA2, RA3, RA0);
+	ROUND(2, RA2, RA3, RA0, RA1);
+	ROUND(3, RA3, RA0, RA1, RA2);
+	leaq (4*4)(%rdi), %rdi;
+	cmpq %rax, %rdi;
+	jne .Lroundloop_blk4;
+
+#undef ROUND
+
+	vmovdqa .Lbswap128_mask rRIP, RTMP2;
+
+	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
+	vpshufb RTMP2, RA0, RA0;
+	vpshufb RTMP2, RA1, RA1;
+	vpshufb RTMP2, RA2, RA2;
+	vpshufb RTMP2, RA3, RA3;
+
+	vmovdqu RA0, 0*16(%rsi);
+	cmpq $2, %rcx;
+	jb .Lblk4_store_output_done;
+	vmovdqu RA1, 1*16(%rsi);
+	je .Lblk4_store_output_done;
+	vmovdqu RA2, 2*16(%rsi);
+	cmpq $3, %rcx;
+	je .Lblk4_store_output_done;
+	vmovdqu RA3, 3*16(%rsi);
+
+.Lblk4_store_output_done:
+	vzeroall;
+	FRAME_END
+	RET;
+SYM_FUNC_END(sm4_aesni_avx_crypt4)
+
+SYM_FUNC_START_LOCAL(__sm4_crypt_blk8)
+	/* input:
+	 *	%rdi: round key array, CTX
+	 *	RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3: eight parallel
+	 *						plaintext blocks
+	 * output:
+	 *	RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3: eight parallel
+	 * 						ciphertext blocks
+	 */
+	FRAME_BEGIN
+
+	vmovdqa .Lbswap32_mask rRIP, RTMP2;
+	vpshufb RTMP2, RA0, RA0;
+	vpshufb RTMP2, RA1, RA1;
+	vpshufb RTMP2, RA2, RA2;
+	vpshufb RTMP2, RA3, RA3;
+	vpshufb RTMP2, RB0, RB0;
+	vpshufb RTMP2, RB1, RB1;
+	vpshufb RTMP2, RB2, RB2;
+	vpshufb RTMP2, RB3, RB3;
+
+	vbroadcastss .L0f0f0f0f rRIP, MASK_4BIT;
+	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
+	transpose_4x4(RB0, RB1, RB2, RB3, RTMP0, RTMP1);
+
+#define ROUND(round, s0, s1, s2, s3, r0, r1, r2, r3)                \
+	vbroadcastss (4*(round))(%rdi), RX0;                        \
+	vmovdqa .Lpre_tf_lo_s rRIP, RTMP4;                          \
+	vmovdqa .Lpre_tf_hi_s rRIP, RTMP1;                          \
+	vmovdqa RX0, RX1;                                           \
+	vpxor s1, RX0, RX0;                                         \
+	vpxor s2, RX0, RX0;                                         \
+	vpxor s3, RX0, RX0; /* s1 ^ s2 ^ s3 ^ rk */                 \
+	vmovdqa .Lpost_tf_lo_s rRIP, RTMP2;                         \
+	vmovdqa .Lpost_tf_hi_s rRIP, RTMP3;                         \
+	vpxor r1, RX1, RX1;                                         \
+	vpxor r2, RX1, RX1;                                         \
+	vpxor r3, RX1, RX1; /* r1 ^ r2 ^ r3 ^ rk */                 \
+                                                                    \
+	/* sbox, non-linear part */                                 \
+	transform_pre(RX0, RTMP4, RTMP1, MASK_4BIT, RTMP0);         \
+	transform_pre(RX1, RTMP4, RTMP1, MASK_4BIT, RTMP0);         \
+	vmovdqa .Linv_shift_row rRIP, RTMP4;                        \
+	vaesenclast MASK_4BIT, RX0, RX0;                            \
+	vaesenclast MASK_4BIT, RX1, RX1;                            \
+	transform_post(RX0, RTMP2, RTMP3, MASK_4BIT, RTMP0);        \
+	transform_post(RX1, RTMP2, RTMP3, MASK_4BIT, RTMP0);        \
+                                                                    \
+	/* linear part */                                           \
+	vpshufb RTMP4, RX0, RTMP0;                                  \
+	vpxor RTMP0, s0, s0; /* s0 ^ x */                           \
+	vpshufb RTMP4, RX1, RTMP2;                                  \
+	vmovdqa .Linv_shift_row_rol_8 rRIP, RTMP4;                  \
+	vpxor RTMP2, r0, r0; /* r0 ^ x */                           \
+	vpshufb RTMP4, RX0, RTMP1;                                  \
+	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) */               \
+	vpshufb RTMP4, RX1, RTMP3;                                  \
+	vmovdqa .Linv_shift_row_rol_16 rRIP, RTMP4;                 \
+	vpxor RTMP3, RTMP2, RTMP2; /* x ^ rol(x,8) */               \
+	vpshufb RTMP4, RX0, RTMP1;                                  \
+	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) ^ rol(x,16) */   \
+	vpshufb RTMP4, RX1, RTMP3;                                  \
+	vmovdqa .Linv_shift_row_rol_24 rRIP, RTMP4;                 \
+	vpxor RTMP3, RTMP2, RTMP2; /* x ^ rol(x,8) ^ rol(x,16) */   \
+	vpshufb RTMP4, RX0, RTMP1;                                  \
+	vpxor RTMP1, s0, s0; /* s0 ^ x ^ rol(x,24) */               \
+	/* s0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */ \
+	vpslld $2, RTMP0, RTMP1;                                    \
+	vpsrld $30, RTMP0, RTMP0;                                   \
+	vpxor RTMP0, s0, s0;                                        \
+	vpxor RTMP1, s0, s0;                                        \
+	vpshufb RTMP4, RX1, RTMP3;                                  \
+	vpxor RTMP3, r0, r0; /* r0 ^ x ^ rol(x,24) */               \
+	/* r0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */ \
+	vpslld $2, RTMP2, RTMP3;                                    \
+	vpsrld $30, RTMP2, RTMP2;                                   \
+	vpxor RTMP2, r0, r0;                                        \
+	vpxor RTMP3, r0, r0;
+
+	leaq (32*4)(%rdi), %rax;
+.align 16
+.Lroundloop_blk8:
+	ROUND(0, RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3);
+	ROUND(1, RA1, RA2, RA3, RA0, RB1, RB2, RB3, RB0);
+	ROUND(2, RA2, RA3, RA0, RA1, RB2, RB3, RB0, RB1);
+	ROUND(3, RA3, RA0, RA1, RA2, RB3, RB0, RB1, RB2);
+	leaq (4*4)(%rdi), %rdi;
+	cmpq %rax, %rdi;
+	jne .Lroundloop_blk8;
+
+#undef ROUND
+
+	vmovdqa .Lbswap128_mask rRIP, RTMP2;
+
+	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
+	transpose_4x4(RB0, RB1, RB2, RB3, RTMP0, RTMP1);
+	vpshufb RTMP2, RA0, RA0;
+	vpshufb RTMP2, RA1, RA1;
+	vpshufb RTMP2, RA2, RA2;
+	vpshufb RTMP2, RA3, RA3;
+	vpshufb RTMP2, RB0, RB0;
+	vpshufb RTMP2, RB1, RB1;
+	vpshufb RTMP2, RB2, RB2;
+	vpshufb RTMP2, RB3, RB3;
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(__sm4_crypt_blk8)
+
+/*
+ * void sm4_aesni_avx_crypt8(const u32 *rk, u8 *dst,
+ *                           const u8 *src, int nblocks)
+ */
+SYM_FUNC_START(sm4_aesni_avx_crypt8)
+	/* input:
+	 *	%rdi: round key array, CTX
+	 *	%rsi: dst (1..8 blocks)
+	 *	%rdx: src (1..8 blocks)
+	 *	%rcx: num blocks (1..8)
+	 */
+	cmpq $5, %rcx;
+	jb sm4_aesni_avx_crypt4;
+
+	FRAME_BEGIN
+
+	vmovdqu (0 * 16)(%rdx), RA0;
+	vmovdqu (1 * 16)(%rdx), RA1;
+	vmovdqu (2 * 16)(%rdx), RA2;
+	vmovdqu (3 * 16)(%rdx), RA3;
+	vmovdqu (4 * 16)(%rdx), RB0;
+	vmovdqa RB0, RB1;
+	vmovdqa RB0, RB2;
+	vmovdqa RB0, RB3;
+	je .Lblk8_load_input_done;
+	vmovdqu (5 * 16)(%rdx), RB1;
+	cmpq $7, %rcx;
+	jb .Lblk8_load_input_done;
+	vmovdqu (6 * 16)(%rdx), RB2;
+	je .Lblk8_load_input_done;
+	vmovdqu (7 * 16)(%rdx), RB3;
+
+.Lblk8_load_input_done:
+	call __sm4_crypt_blk8;
+
+	cmpq $6, %rcx;
+	vmovdqu RA0, (0 * 16)(%rsi);
+	vmovdqu RA1, (1 * 16)(%rsi);
+	vmovdqu RA2, (2 * 16)(%rsi);
+	vmovdqu RA3, (3 * 16)(%rsi);
+	vmovdqu RB0, (4 * 16)(%rsi);
+	jb .Lblk8_store_output_done;
+	vmovdqu RB1, (5 * 16)(%rsi);
+	je .Lblk8_store_output_done;
+	vmovdqu RB2, (6 * 16)(%rsi);
+	cmpq $7, %rcx;
+	je .Lblk8_store_output_done;
+	vmovdqu RB3, (7 * 16)(%rsi);
+
+.Lblk8_store_output_done:
+	vzeroall;
+	FRAME_END
+	RET;
+SYM_FUNC_END(sm4_aesni_avx_crypt8)
+
+/*
+ * void sm4_aesni_avx_ctr_enc_blk8(const u32 *rk, u8 *dst,
+ *                                 const u8 *src, u8 *iv)
+ */
+SYM_TYPED_FUNC_START(sm4_aesni_avx_ctr_enc_blk8)
+	/* input:
+	 *	%rdi: round key array, CTX
+	 *	%rsi: dst (8 blocks)
+	 *	%rdx: src (8 blocks)
+	 *	%rcx: iv (big endian, 128bit)
+	 */
+	FRAME_BEGIN
+
+	/* load IV and byteswap */
+	vmovdqu (%rcx), RA0;
+
+	vmovdqa .Lbswap128_mask rRIP, RBSWAP;
+	vpshufb RBSWAP, RA0, RTMP0; /* be => le */
+
+	vpcmpeqd RNOT, RNOT, RNOT;
+	vpsrldq $8, RNOT, RNOT; /* low: -1, high: 0 */
+
+#define inc_le128(x, minus_one, tmp) \
+	vpcmpeqq minus_one, x, tmp;  \
+	vpsubq minus_one, x, x;      \
+	vpslldq $8, tmp, tmp;        \
+	vpsubq tmp, x, x;
+
+	/* construct IVs */
+	inc_le128(RTMP0, RNOT, RTMP2); /* +1 */
+	vpshufb RBSWAP, RTMP0, RA1;
+	inc_le128(RTMP0, RNOT, RTMP2); /* +2 */
+	vpshufb RBSWAP, RTMP0, RA2;
+	inc_le128(RTMP0, RNOT, RTMP2); /* +3 */
+	vpshufb RBSWAP, RTMP0, RA3;
+	inc_le128(RTMP0, RNOT, RTMP2); /* +4 */
+	vpshufb RBSWAP, RTMP0, RB0;
+	inc_le128(RTMP0, RNOT, RTMP2); /* +5 */
+	vpshufb RBSWAP, RTMP0, RB1;
+	inc_le128(RTMP0, RNOT, RTMP2); /* +6 */
+	vpshufb RBSWAP, RTMP0, RB2;
+	inc_le128(RTMP0, RNOT, RTMP2); /* +7 */
+	vpshufb RBSWAP, RTMP0, RB3;
+	inc_le128(RTMP0, RNOT, RTMP2); /* +8 */
+	vpshufb RBSWAP, RTMP0, RTMP1;
+
+	/* store new IV */
+	vmovdqu RTMP1, (%rcx);
+
+	call __sm4_crypt_blk8;
+
+	vpxor (0 * 16)(%rdx), RA0, RA0;
+	vpxor (1 * 16)(%rdx), RA1, RA1;
+	vpxor (2 * 16)(%rdx), RA2, RA2;
+	vpxor (3 * 16)(%rdx), RA3, RA3;
+	vpxor (4 * 16)(%rdx), RB0, RB0;
+	vpxor (5 * 16)(%rdx), RB1, RB1;
+	vpxor (6 * 16)(%rdx), RB2, RB2;
+	vpxor (7 * 16)(%rdx), RB3, RB3;
+
+	vmovdqu RA0, (0 * 16)(%rsi);
+	vmovdqu RA1, (1 * 16)(%rsi);
+	vmovdqu RA2, (2 * 16)(%rsi);
+	vmovdqu RA3, (3 * 16)(%rsi);
+	vmovdqu RB0, (4 * 16)(%rsi);
+	vmovdqu RB1, (5 * 16)(%rsi);
+	vmovdqu RB2, (6 * 16)(%rsi);
+	vmovdqu RB3, (7 * 16)(%rsi);
+
+	vzeroall;
+	FRAME_END
+	RET;
+SYM_FUNC_END(sm4_aesni_avx_ctr_enc_blk8)
+
+/*
+ * void sm4_aesni_avx_cbc_dec_blk8(const u32 *rk, u8 *dst,
+ *                                 const u8 *src, u8 *iv)
+ */
+SYM_TYPED_FUNC_START(sm4_aesni_avx_cbc_dec_blk8)
+	/* input:
+	 *	%rdi: round key array, CTX
+	 *	%rsi: dst (8 blocks)
+	 *	%rdx: src (8 blocks)
+	 *	%rcx: iv
+	 */
+	FRAME_BEGIN
+
+	vmovdqu (0 * 16)(%rdx), RA0;
+	vmovdqu (1 * 16)(%rdx), RA1;
+	vmovdqu (2 * 16)(%rdx), RA2;
+	vmovdqu (3 * 16)(%rdx), RA3;
+	vmovdqu (4 * 16)(%rdx), RB0;
+	vmovdqu (5 * 16)(%rdx), RB1;
+	vmovdqu (6 * 16)(%rdx), RB2;
+	vmovdqu (7 * 16)(%rdx), RB3;
+
+	call __sm4_crypt_blk8;
+
+	vmovdqu (7 * 16)(%rdx), RNOT;
+	vpxor (%rcx), RA0, RA0;
+	vpxor (0 * 16)(%rdx), RA1, RA1;
+	vpxor (1 * 16)(%rdx), RA2, RA2;
+	vpxor (2 * 16)(%rdx), RA3, RA3;
+	vpxor (3 * 16)(%rdx), RB0, RB0;
+	vpxor (4 * 16)(%rdx), RB1, RB1;
+	vpxor (5 * 16)(%rdx), RB2, RB2;
+	vpxor (6 * 16)(%rdx), RB3, RB3;
+	vmovdqu RNOT, (%rcx); /* store new IV */
+
+	vmovdqu RA0, (0 * 16)(%rsi);
+	vmovdqu RA1, (1 * 16)(%rsi);
+	vmovdqu RA2, (2 * 16)(%rsi);
+	vmovdqu RA3, (3 * 16)(%rsi);
+	vmovdqu RB0, (4 * 16)(%rsi);
+	vmovdqu RB1, (5 * 16)(%rsi);
+	vmovdqu RB2, (6 * 16)(%rsi);
+	vmovdqu RB3, (7 * 16)(%rsi);
+
+	vzeroall;
+	FRAME_END
+	RET;
+SYM_FUNC_END(sm4_aesni_avx_cbc_dec_blk8)
diff --git a/arch/x86/crypto/sm4-aesni-avx2-asm_64.S b/arch/x86/crypto/sm4-aesni-avx2-asm_64.S
new file mode 100644
index 000000000..9ff5ba075
--- /dev/null
+++ b/arch/x86/crypto/sm4-aesni-avx2-asm_64.S
@@ -0,0 +1,441 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * SM4 Cipher Algorithm, AES-NI/AVX2 optimized.
+ * as specified in
+ * https://tools.ietf.org/id/draft-ribose-cfrg-sm4-10.html
+ *
+ * Copyright (C) 2018 Markku-Juhani O. Saarinen <mjos@iki.fi>
+ * Copyright (C) 2020 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ * Copyright (c) 2021 Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
+ */
+
+/* Based on SM4 AES-NI work by libgcrypt and Markku-Juhani O. Saarinen at:
+ *  https://github.com/mjosaarinen/sm4ni
+ */
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+#include <asm/frame.h>
+
+#define rRIP         (%rip)
+
+/* vector registers */
+#define RX0          %ymm0
+#define RX1          %ymm1
+#define MASK_4BIT    %ymm2
+#define RTMP0        %ymm3
+#define RTMP1        %ymm4
+#define RTMP2        %ymm5
+#define RTMP3        %ymm6
+#define RTMP4        %ymm7
+
+#define RA0          %ymm8
+#define RA1          %ymm9
+#define RA2          %ymm10
+#define RA3          %ymm11
+
+#define RB0          %ymm12
+#define RB1          %ymm13
+#define RB2          %ymm14
+#define RB3          %ymm15
+
+#define RNOT         %ymm0
+#define RBSWAP       %ymm1
+
+#define RX0x         %xmm0
+#define RX1x         %xmm1
+#define MASK_4BITx   %xmm2
+
+#define RNOTx        %xmm0
+#define RBSWAPx      %xmm1
+
+#define RTMP0x       %xmm3
+#define RTMP1x       %xmm4
+#define RTMP2x       %xmm5
+#define RTMP3x       %xmm6
+#define RTMP4x       %xmm7
+
+
+/* helper macros */
+
+/* Transpose four 32-bit words between 128-bit vector lanes. */
+#define transpose_4x4(x0, x1, x2, x3, t1, t2) \
+	vpunpckhdq x1, x0, t2;                \
+	vpunpckldq x1, x0, x0;                \
+	                                      \
+	vpunpckldq x3, x2, t1;                \
+	vpunpckhdq x3, x2, x2;                \
+	                                      \
+	vpunpckhqdq t1, x0, x1;               \
+	vpunpcklqdq t1, x0, x0;               \
+	                                      \
+	vpunpckhqdq x2, t2, x3;               \
+	vpunpcklqdq x2, t2, x2;
+
+/* post-SubByte transform. */
+#define transform_pre(x, lo_t, hi_t, mask4bit, tmp0) \
+	vpand x, mask4bit, tmp0;                     \
+	vpandn x, mask4bit, x;                       \
+	vpsrld $4, x, x;                             \
+	                                             \
+	vpshufb tmp0, lo_t, tmp0;                    \
+	vpshufb x, hi_t, x;                          \
+	vpxor tmp0, x, x;
+
+/* post-SubByte transform. Note: x has been XOR'ed with mask4bit by
+ * 'vaeslastenc' instruction. */
+#define transform_post(x, lo_t, hi_t, mask4bit, tmp0) \
+	vpandn mask4bit, x, tmp0;                     \
+	vpsrld $4, x, x;                              \
+	vpand x, mask4bit, x;                         \
+	                                              \
+	vpshufb tmp0, lo_t, tmp0;                     \
+	vpshufb x, hi_t, x;                           \
+	vpxor tmp0, x, x;
+
+
+.section	.rodata.cst16, "aM", @progbits, 16
+.align 16
+
+/*
+ * Following four affine transform look-up tables are from work by
+ * Markku-Juhani O. Saarinen, at https://github.com/mjosaarinen/sm4ni
+ *
+ * These allow exposing SM4 S-Box from AES SubByte.
+ */
+
+/* pre-SubByte affine transform, from SM4 field to AES field. */
+.Lpre_tf_lo_s:
+	.quad 0x9197E2E474720701, 0xC7C1B4B222245157
+.Lpre_tf_hi_s:
+	.quad 0xE240AB09EB49A200, 0xF052B91BF95BB012
+
+/* post-SubByte affine transform, from AES field to SM4 field. */
+.Lpost_tf_lo_s:
+	.quad 0x5B67F2CEA19D0834, 0xEDD14478172BBE82
+.Lpost_tf_hi_s:
+	.quad 0xAE7201DD73AFDC00, 0x11CDBE62CC1063BF
+
+/* For isolating SubBytes from AESENCLAST, inverse shift row */
+.Linv_shift_row:
+	.byte 0x00, 0x0d, 0x0a, 0x07, 0x04, 0x01, 0x0e, 0x0b
+	.byte 0x08, 0x05, 0x02, 0x0f, 0x0c, 0x09, 0x06, 0x03
+
+/* Inverse shift row + Rotate left by 8 bits on 32-bit words with vpshufb */
+.Linv_shift_row_rol_8:
+	.byte 0x07, 0x00, 0x0d, 0x0a, 0x0b, 0x04, 0x01, 0x0e
+	.byte 0x0f, 0x08, 0x05, 0x02, 0x03, 0x0c, 0x09, 0x06
+
+/* Inverse shift row + Rotate left by 16 bits on 32-bit words with vpshufb */
+.Linv_shift_row_rol_16:
+	.byte 0x0a, 0x07, 0x00, 0x0d, 0x0e, 0x0b, 0x04, 0x01
+	.byte 0x02, 0x0f, 0x08, 0x05, 0x06, 0x03, 0x0c, 0x09
+
+/* Inverse shift row + Rotate left by 24 bits on 32-bit words with vpshufb */
+.Linv_shift_row_rol_24:
+	.byte 0x0d, 0x0a, 0x07, 0x00, 0x01, 0x0e, 0x0b, 0x04
+	.byte 0x05, 0x02, 0x0f, 0x08, 0x09, 0x06, 0x03, 0x0c
+
+/* For CTR-mode IV byteswap */
+.Lbswap128_mask:
+	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+
+/* For input word byte-swap */
+.Lbswap32_mask:
+	.byte 3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12
+
+.align 4
+/* 4-bit mask */
+.L0f0f0f0f:
+	.long 0x0f0f0f0f
+
+/* 12 bytes, only for padding */
+.Lpadding_deadbeef:
+	.long 0xdeadbeef, 0xdeadbeef, 0xdeadbeef
+
+.text
+SYM_FUNC_START_LOCAL(__sm4_crypt_blk16)
+	/* input:
+	 *	%rdi: round key array, CTX
+	 *	RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3: sixteen parallel
+	 *						plaintext blocks
+	 * output:
+	 *	RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3: sixteen parallel
+	 * 						ciphertext blocks
+	 */
+	FRAME_BEGIN
+
+	vbroadcasti128 .Lbswap32_mask rRIP, RTMP2;
+	vpshufb RTMP2, RA0, RA0;
+	vpshufb RTMP2, RA1, RA1;
+	vpshufb RTMP2, RA2, RA2;
+	vpshufb RTMP2, RA3, RA3;
+	vpshufb RTMP2, RB0, RB0;
+	vpshufb RTMP2, RB1, RB1;
+	vpshufb RTMP2, RB2, RB2;
+	vpshufb RTMP2, RB3, RB3;
+
+	vpbroadcastd .L0f0f0f0f rRIP, MASK_4BIT;
+	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
+	transpose_4x4(RB0, RB1, RB2, RB3, RTMP0, RTMP1);
+
+#define ROUND(round, s0, s1, s2, s3, r0, r1, r2, r3)                \
+	vpbroadcastd (4*(round))(%rdi), RX0;                        \
+	vbroadcasti128 .Lpre_tf_lo_s rRIP, RTMP4;                   \
+	vbroadcasti128 .Lpre_tf_hi_s rRIP, RTMP1;                   \
+	vmovdqa RX0, RX1;                                           \
+	vpxor s1, RX0, RX0;                                         \
+	vpxor s2, RX0, RX0;                                         \
+	vpxor s3, RX0, RX0; /* s1 ^ s2 ^ s3 ^ rk */                 \
+	vbroadcasti128 .Lpost_tf_lo_s rRIP, RTMP2;                  \
+	vbroadcasti128 .Lpost_tf_hi_s rRIP, RTMP3;                  \
+	vpxor r1, RX1, RX1;                                         \
+	vpxor r2, RX1, RX1;                                         \
+	vpxor r3, RX1, RX1; /* r1 ^ r2 ^ r3 ^ rk */                 \
+	                                                            \
+	/* sbox, non-linear part */                                 \
+	transform_pre(RX0, RTMP4, RTMP1, MASK_4BIT, RTMP0);         \
+	transform_pre(RX1, RTMP4, RTMP1, MASK_4BIT, RTMP0);         \
+	vextracti128 $1, RX0, RTMP4x;                               \
+	vextracti128 $1, RX1, RTMP0x;                               \
+	vaesenclast MASK_4BITx, RX0x, RX0x;                         \
+	vaesenclast MASK_4BITx, RTMP4x, RTMP4x;                     \
+	vaesenclast MASK_4BITx, RX1x, RX1x;                         \
+	vaesenclast MASK_4BITx, RTMP0x, RTMP0x;                     \
+	vinserti128 $1, RTMP4x, RX0, RX0;                           \
+	vbroadcasti128 .Linv_shift_row rRIP, RTMP4;                 \
+	vinserti128 $1, RTMP0x, RX1, RX1;                           \
+	transform_post(RX0, RTMP2, RTMP3, MASK_4BIT, RTMP0);        \
+	transform_post(RX1, RTMP2, RTMP3, MASK_4BIT, RTMP0);        \
+	                                                            \
+	/* linear part */                                           \
+	vpshufb RTMP4, RX0, RTMP0;                                  \
+	vpxor RTMP0, s0, s0; /* s0 ^ x */                           \
+	vpshufb RTMP4, RX1, RTMP2;                                  \
+	vbroadcasti128 .Linv_shift_row_rol_8 rRIP, RTMP4;           \
+	vpxor RTMP2, r0, r0; /* r0 ^ x */                           \
+	vpshufb RTMP4, RX0, RTMP1;                                  \
+	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) */               \
+	vpshufb RTMP4, RX1, RTMP3;                                  \
+	vbroadcasti128 .Linv_shift_row_rol_16 rRIP, RTMP4;          \
+	vpxor RTMP3, RTMP2, RTMP2; /* x ^ rol(x,8) */               \
+	vpshufb RTMP4, RX0, RTMP1;                                  \
+	vpxor RTMP1, RTMP0, RTMP0; /* x ^ rol(x,8) ^ rol(x,16) */   \
+	vpshufb RTMP4, RX1, RTMP3;                                  \
+	vbroadcasti128 .Linv_shift_row_rol_24 rRIP, RTMP4;          \
+	vpxor RTMP3, RTMP2, RTMP2; /* x ^ rol(x,8) ^ rol(x,16) */   \
+	vpshufb RTMP4, RX0, RTMP1;                                  \
+	vpxor RTMP1, s0, s0; /* s0 ^ x ^ rol(x,24) */               \
+	vpslld $2, RTMP0, RTMP1;                                    \
+	vpsrld $30, RTMP0, RTMP0;                                   \
+	vpxor RTMP0, s0, s0;                                        \
+	/* s0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */ \
+	vpxor RTMP1, s0, s0;                                        \
+	vpshufb RTMP4, RX1, RTMP3;                                  \
+	vpxor RTMP3, r0, r0; /* r0 ^ x ^ rol(x,24) */               \
+	vpslld $2, RTMP2, RTMP3;                                    \
+	vpsrld $30, RTMP2, RTMP2;                                   \
+	vpxor RTMP2, r0, r0;                                        \
+	/* r0 ^ x ^ rol(x,2) ^ rol(x,10) ^ rol(x,18) ^ rol(x,24) */ \
+	vpxor RTMP3, r0, r0;
+
+	leaq (32*4)(%rdi), %rax;
+.align 16
+.Lroundloop_blk8:
+	ROUND(0, RA0, RA1, RA2, RA3, RB0, RB1, RB2, RB3);
+	ROUND(1, RA1, RA2, RA3, RA0, RB1, RB2, RB3, RB0);
+	ROUND(2, RA2, RA3, RA0, RA1, RB2, RB3, RB0, RB1);
+	ROUND(3, RA3, RA0, RA1, RA2, RB3, RB0, RB1, RB2);
+	leaq (4*4)(%rdi), %rdi;
+	cmpq %rax, %rdi;
+	jne .Lroundloop_blk8;
+
+#undef ROUND
+
+	vbroadcasti128 .Lbswap128_mask rRIP, RTMP2;
+
+	transpose_4x4(RA0, RA1, RA2, RA3, RTMP0, RTMP1);
+	transpose_4x4(RB0, RB1, RB2, RB3, RTMP0, RTMP1);
+	vpshufb RTMP2, RA0, RA0;
+	vpshufb RTMP2, RA1, RA1;
+	vpshufb RTMP2, RA2, RA2;
+	vpshufb RTMP2, RA3, RA3;
+	vpshufb RTMP2, RB0, RB0;
+	vpshufb RTMP2, RB1, RB1;
+	vpshufb RTMP2, RB2, RB2;
+	vpshufb RTMP2, RB3, RB3;
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(__sm4_crypt_blk16)
+
+#define inc_le128(x, minus_one, tmp) \
+	vpcmpeqq minus_one, x, tmp;  \
+	vpsubq minus_one, x, x;      \
+	vpslldq $8, tmp, tmp;        \
+	vpsubq tmp, x, x;
+
+/*
+ * void sm4_aesni_avx2_ctr_enc_blk16(const u32 *rk, u8 *dst,
+ *                                   const u8 *src, u8 *iv)
+ */
+SYM_TYPED_FUNC_START(sm4_aesni_avx2_ctr_enc_blk16)
+	/* input:
+	 *	%rdi: round key array, CTX
+	 *	%rsi: dst (16 blocks)
+	 *	%rdx: src (16 blocks)
+	 *	%rcx: iv (big endian, 128bit)
+	 */
+	FRAME_BEGIN
+
+	movq 8(%rcx), %rax;
+	bswapq %rax;
+
+	vzeroupper;
+
+	vbroadcasti128 .Lbswap128_mask rRIP, RTMP3;
+	vpcmpeqd RNOT, RNOT, RNOT;
+	vpsrldq $8, RNOT, RNOT;   /* ab: -1:0 ; cd: -1:0 */
+	vpaddq RNOT, RNOT, RTMP2; /* ab: -2:0 ; cd: -2:0 */
+
+	/* load IV and byteswap */
+	vmovdqu (%rcx), RTMP4x;
+	vpshufb RTMP3x, RTMP4x, RTMP4x;
+	vmovdqa RTMP4x, RTMP0x;
+	inc_le128(RTMP4x, RNOTx, RTMP1x);
+	vinserti128 $1, RTMP4x, RTMP0, RTMP0;
+	vpshufb RTMP3, RTMP0, RA0; /* +1 ; +0 */
+
+	/* check need for handling 64-bit overflow and carry */
+	cmpq $(0xffffffffffffffff - 16), %rax;
+	ja .Lhandle_ctr_carry;
+
+	/* construct IVs */
+	vpsubq RTMP2, RTMP0, RTMP0; /* +3 ; +2 */
+	vpshufb RTMP3, RTMP0, RA1;
+	vpsubq RTMP2, RTMP0, RTMP0; /* +5 ; +4 */
+	vpshufb RTMP3, RTMP0, RA2;
+	vpsubq RTMP2, RTMP0, RTMP0; /* +7 ; +6 */
+	vpshufb RTMP3, RTMP0, RA3;
+	vpsubq RTMP2, RTMP0, RTMP0; /* +9 ; +8 */
+	vpshufb RTMP3, RTMP0, RB0;
+	vpsubq RTMP2, RTMP0, RTMP0; /* +11 ; +10 */
+	vpshufb RTMP3, RTMP0, RB1;
+	vpsubq RTMP2, RTMP0, RTMP0; /* +13 ; +12 */
+	vpshufb RTMP3, RTMP0, RB2;
+	vpsubq RTMP2, RTMP0, RTMP0; /* +15 ; +14 */
+	vpshufb RTMP3, RTMP0, RB3;
+	vpsubq RTMP2, RTMP0, RTMP0; /* +16 */
+	vpshufb RTMP3x, RTMP0x, RTMP0x;
+
+	jmp .Lctr_carry_done;
+
+.Lhandle_ctr_carry:
+	/* construct IVs */
+	inc_le128(RTMP0, RNOT, RTMP1);
+	inc_le128(RTMP0, RNOT, RTMP1);
+	vpshufb RTMP3, RTMP0, RA1; /* +3 ; +2 */
+	inc_le128(RTMP0, RNOT, RTMP1);
+	inc_le128(RTMP0, RNOT, RTMP1);
+	vpshufb RTMP3, RTMP0, RA2; /* +5 ; +4 */
+	inc_le128(RTMP0, RNOT, RTMP1);
+	inc_le128(RTMP0, RNOT, RTMP1);
+	vpshufb RTMP3, RTMP0, RA3; /* +7 ; +6 */
+	inc_le128(RTMP0, RNOT, RTMP1);
+	inc_le128(RTMP0, RNOT, RTMP1);
+	vpshufb RTMP3, RTMP0, RB0; /* +9 ; +8 */
+	inc_le128(RTMP0, RNOT, RTMP1);
+	inc_le128(RTMP0, RNOT, RTMP1);
+	vpshufb RTMP3, RTMP0, RB1; /* +11 ; +10 */
+	inc_le128(RTMP0, RNOT, RTMP1);
+	inc_le128(RTMP0, RNOT, RTMP1);
+	vpshufb RTMP3, RTMP0, RB2; /* +13 ; +12 */
+	inc_le128(RTMP0, RNOT, RTMP1);
+	inc_le128(RTMP0, RNOT, RTMP1);
+	vpshufb RTMP3, RTMP0, RB3; /* +15 ; +14 */
+	inc_le128(RTMP0, RNOT, RTMP1);
+	vextracti128 $1, RTMP0, RTMP0x;
+	vpshufb RTMP3x, RTMP0x, RTMP0x; /* +16 */
+
+.align 4
+.Lctr_carry_done:
+	/* store new IV */
+	vmovdqu RTMP0x, (%rcx);
+
+	call __sm4_crypt_blk16;
+
+	vpxor (0 * 32)(%rdx), RA0, RA0;
+	vpxor (1 * 32)(%rdx), RA1, RA1;
+	vpxor (2 * 32)(%rdx), RA2, RA2;
+	vpxor (3 * 32)(%rdx), RA3, RA3;
+	vpxor (4 * 32)(%rdx), RB0, RB0;
+	vpxor (5 * 32)(%rdx), RB1, RB1;
+	vpxor (6 * 32)(%rdx), RB2, RB2;
+	vpxor (7 * 32)(%rdx), RB3, RB3;
+
+	vmovdqu RA0, (0 * 32)(%rsi);
+	vmovdqu RA1, (1 * 32)(%rsi);
+	vmovdqu RA2, (2 * 32)(%rsi);
+	vmovdqu RA3, (3 * 32)(%rsi);
+	vmovdqu RB0, (4 * 32)(%rsi);
+	vmovdqu RB1, (5 * 32)(%rsi);
+	vmovdqu RB2, (6 * 32)(%rsi);
+	vmovdqu RB3, (7 * 32)(%rsi);
+
+	vzeroall;
+	FRAME_END
+	RET;
+SYM_FUNC_END(sm4_aesni_avx2_ctr_enc_blk16)
+
+/*
+ * void sm4_aesni_avx2_cbc_dec_blk16(const u32 *rk, u8 *dst,
+ *                                   const u8 *src, u8 *iv)
+ */
+SYM_TYPED_FUNC_START(sm4_aesni_avx2_cbc_dec_blk16)
+	/* input:
+	 *	%rdi: round key array, CTX
+	 *	%rsi: dst (16 blocks)
+	 *	%rdx: src (16 blocks)
+	 *	%rcx: iv
+	 */
+	FRAME_BEGIN
+
+	vzeroupper;
+
+	vmovdqu (0 * 32)(%rdx), RA0;
+	vmovdqu (1 * 32)(%rdx), RA1;
+	vmovdqu (2 * 32)(%rdx), RA2;
+	vmovdqu (3 * 32)(%rdx), RA3;
+	vmovdqu (4 * 32)(%rdx), RB0;
+	vmovdqu (5 * 32)(%rdx), RB1;
+	vmovdqu (6 * 32)(%rdx), RB2;
+	vmovdqu (7 * 32)(%rdx), RB3;
+
+	call __sm4_crypt_blk16;
+
+	vmovdqu (%rcx), RNOTx;
+	vinserti128 $1, (%rdx), RNOT, RNOT;
+	vpxor RNOT, RA0, RA0;
+	vpxor (0 * 32 + 16)(%rdx), RA1, RA1;
+	vpxor (1 * 32 + 16)(%rdx), RA2, RA2;
+	vpxor (2 * 32 + 16)(%rdx), RA3, RA3;
+	vpxor (3 * 32 + 16)(%rdx), RB0, RB0;
+	vpxor (4 * 32 + 16)(%rdx), RB1, RB1;
+	vpxor (5 * 32 + 16)(%rdx), RB2, RB2;
+	vpxor (6 * 32 + 16)(%rdx), RB3, RB3;
+	vmovdqu (7 * 32 + 16)(%rdx), RNOTx;
+	vmovdqu RNOTx, (%rcx); /* store new IV */
+
+	vmovdqu RA0, (0 * 32)(%rsi);
+	vmovdqu RA1, (1 * 32)(%rsi);
+	vmovdqu RA2, (2 * 32)(%rsi);
+	vmovdqu RA3, (3 * 32)(%rsi);
+	vmovdqu RB0, (4 * 32)(%rsi);
+	vmovdqu RB1, (5 * 32)(%rsi);
+	vmovdqu RB2, (6 * 32)(%rsi);
+	vmovdqu RB3, (7 * 32)(%rsi);
+
+	vzeroall;
+	FRAME_END
+	RET;
+SYM_FUNC_END(sm4_aesni_avx2_cbc_dec_blk16)
diff --git a/arch/x86/crypto/sm4-avx.h b/arch/x86/crypto/sm4-avx.h
new file mode 100644
index 000000000..b5b5e67e4
--- /dev/null
+++ b/arch/x86/crypto/sm4-avx.h
@@ -0,0 +1,20 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#ifndef ASM_X86_SM4_AVX_H
+#define ASM_X86_SM4_AVX_H
+
+#include <linux/types.h>
+#include <crypto/sm4.h>
+
+typedef void (*sm4_crypt_func)(const u32 *rk, u8 *dst, const u8 *src, u8 *iv);
+
+int sm4_avx_ecb_encrypt(struct skcipher_request *req);
+int sm4_avx_ecb_decrypt(struct skcipher_request *req);
+
+int sm4_cbc_encrypt(struct skcipher_request *req);
+int sm4_avx_cbc_decrypt(struct skcipher_request *req,
+			unsigned int bsize, sm4_crypt_func func);
+
+int sm4_avx_ctr_crypt(struct skcipher_request *req,
+			unsigned int bsize, sm4_crypt_func func);
+
+#endif
diff --git a/arch/x86/crypto/sm4_aesni_avx2_glue.c b/arch/x86/crypto/sm4_aesni_avx2_glue.c
new file mode 100644
index 000000000..fec0ab7a6
--- /dev/null
+++ b/arch/x86/crypto/sm4_aesni_avx2_glue.c
@@ -0,0 +1,134 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * SM4 Cipher Algorithm, AES-NI/AVX2 optimized.
+ * as specified in
+ * https://tools.ietf.org/id/draft-ribose-cfrg-sm4-10.html
+ *
+ * Copyright (c) 2021, Alibaba Group.
+ * Copyright (c) 2021 Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
+ */
+
+#include <asm/fpu/api.h>
+#include <linux/module.h>
+#include <linux/crypto.h>
+#include <linux/kernel.h>
+#include <crypto/internal/skcipher.h>
+#include <crypto/sm4.h>
+#include "sm4-avx.h"
+
+#define SM4_CRYPT16_BLOCK_SIZE	(SM4_BLOCK_SIZE * 16)
+
+asmlinkage void sm4_aesni_avx2_ctr_enc_blk16(const u32 *rk, u8 *dst,
+					const u8 *src, u8 *iv);
+asmlinkage void sm4_aesni_avx2_cbc_dec_blk16(const u32 *rk, u8 *dst,
+					const u8 *src, u8 *iv);
+
+static int sm4_skcipher_setkey(struct crypto_skcipher *tfm, const u8 *key,
+			unsigned int key_len)
+{
+	struct sm4_ctx *ctx = crypto_skcipher_ctx(tfm);
+
+	return sm4_expandkey(ctx, key, key_len);
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	return sm4_avx_cbc_decrypt(req, SM4_CRYPT16_BLOCK_SIZE,
+				sm4_aesni_avx2_cbc_dec_blk16);
+}
+
+
+static int ctr_crypt(struct skcipher_request *req)
+{
+	return sm4_avx_ctr_crypt(req, SM4_CRYPT16_BLOCK_SIZE,
+				sm4_aesni_avx2_ctr_enc_blk16);
+}
+
+static struct skcipher_alg sm4_aesni_avx2_skciphers[] = {
+	{
+		.base = {
+			.cra_name		= "ecb(sm4)",
+			.cra_driver_name	= "ecb-sm4-aesni-avx2",
+			.cra_priority		= 500,
+			.cra_blocksize		= SM4_BLOCK_SIZE,
+			.cra_ctxsize		= sizeof(struct sm4_ctx),
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= SM4_KEY_SIZE,
+		.max_keysize	= SM4_KEY_SIZE,
+		.walksize	= 16 * SM4_BLOCK_SIZE,
+		.setkey		= sm4_skcipher_setkey,
+		.encrypt	= sm4_avx_ecb_encrypt,
+		.decrypt	= sm4_avx_ecb_decrypt,
+	}, {
+		.base = {
+			.cra_name		= "cbc(sm4)",
+			.cra_driver_name	= "cbc-sm4-aesni-avx2",
+			.cra_priority		= 500,
+			.cra_blocksize		= SM4_BLOCK_SIZE,
+			.cra_ctxsize		= sizeof(struct sm4_ctx),
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= SM4_KEY_SIZE,
+		.max_keysize	= SM4_KEY_SIZE,
+		.ivsize		= SM4_BLOCK_SIZE,
+		.walksize	= 16 * SM4_BLOCK_SIZE,
+		.setkey		= sm4_skcipher_setkey,
+		.encrypt	= sm4_cbc_encrypt,
+		.decrypt	= cbc_decrypt,
+	}, {
+		.base = {
+			.cra_name		= "ctr(sm4)",
+			.cra_driver_name	= "ctr-sm4-aesni-avx2",
+			.cra_priority		= 500,
+			.cra_blocksize		= 1,
+			.cra_ctxsize		= sizeof(struct sm4_ctx),
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= SM4_KEY_SIZE,
+		.max_keysize	= SM4_KEY_SIZE,
+		.ivsize		= SM4_BLOCK_SIZE,
+		.chunksize	= SM4_BLOCK_SIZE,
+		.walksize	= 16 * SM4_BLOCK_SIZE,
+		.setkey		= sm4_skcipher_setkey,
+		.encrypt	= ctr_crypt,
+		.decrypt	= ctr_crypt,
+	}
+};
+
+static int __init sm4_init(void)
+{
+	const char *feature_name;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX) ||
+	    !boot_cpu_has(X86_FEATURE_AVX2) ||
+	    !boot_cpu_has(X86_FEATURE_AES) ||
+	    !boot_cpu_has(X86_FEATURE_OSXSAVE)) {
+		pr_info("AVX2 or AES-NI instructions are not detected.\n");
+		return -ENODEV;
+	}
+
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM,
+				&feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	return crypto_register_skciphers(sm4_aesni_avx2_skciphers,
+					 ARRAY_SIZE(sm4_aesni_avx2_skciphers));
+}
+
+static void __exit sm4_exit(void)
+{
+	crypto_unregister_skciphers(sm4_aesni_avx2_skciphers,
+				    ARRAY_SIZE(sm4_aesni_avx2_skciphers));
+}
+
+module_init(sm4_init);
+module_exit(sm4_exit);
+
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Tianjia Zhang <tianjia.zhang@linux.alibaba.com>");
+MODULE_DESCRIPTION("SM4 Cipher Algorithm, AES-NI/AVX2 optimized");
+MODULE_ALIAS_CRYPTO("sm4");
+MODULE_ALIAS_CRYPTO("sm4-aesni-avx2");
diff --git a/arch/x86/crypto/sm4_aesni_avx_glue.c b/arch/x86/crypto/sm4_aesni_avx_glue.c
new file mode 100644
index 000000000..88caf418a
--- /dev/null
+++ b/arch/x86/crypto/sm4_aesni_avx_glue.c
@@ -0,0 +1,349 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * SM4 Cipher Algorithm, AES-NI/AVX optimized.
+ * as specified in
+ * https://tools.ietf.org/id/draft-ribose-cfrg-sm4-10.html
+ *
+ * Copyright (c) 2021, Alibaba Group.
+ * Copyright (c) 2021 Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
+ */
+
+#include <asm/fpu/api.h>
+#include <linux/module.h>
+#include <linux/crypto.h>
+#include <linux/export.h>
+#include <linux/kernel.h>
+#include <crypto/internal/skcipher.h>
+#include <crypto/sm4.h>
+#include "sm4-avx.h"
+
+#define SM4_CRYPT8_BLOCK_SIZE	(SM4_BLOCK_SIZE * 8)
+
+asmlinkage void sm4_aesni_avx_crypt4(const u32 *rk, u8 *dst,
+				const u8 *src, int nblocks);
+asmlinkage void sm4_aesni_avx_crypt8(const u32 *rk, u8 *dst,
+				const u8 *src, int nblocks);
+asmlinkage void sm4_aesni_avx_ctr_enc_blk8(const u32 *rk, u8 *dst,
+				const u8 *src, u8 *iv);
+asmlinkage void sm4_aesni_avx_cbc_dec_blk8(const u32 *rk, u8 *dst,
+				const u8 *src, u8 *iv);
+
+static int sm4_skcipher_setkey(struct crypto_skcipher *tfm, const u8 *key,
+			unsigned int key_len)
+{
+	struct sm4_ctx *ctx = crypto_skcipher_ctx(tfm);
+
+	return sm4_expandkey(ctx, key, key_len);
+}
+
+static int ecb_do_crypt(struct skcipher_request *req, const u32 *rkey)
+{
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes) > 0) {
+		const u8 *src = walk.src.virt.addr;
+		u8 *dst = walk.dst.virt.addr;
+
+		kernel_fpu_begin();
+		while (nbytes >= SM4_CRYPT8_BLOCK_SIZE) {
+			sm4_aesni_avx_crypt8(rkey, dst, src, 8);
+			dst += SM4_CRYPT8_BLOCK_SIZE;
+			src += SM4_CRYPT8_BLOCK_SIZE;
+			nbytes -= SM4_CRYPT8_BLOCK_SIZE;
+		}
+		while (nbytes >= SM4_BLOCK_SIZE) {
+			unsigned int nblocks = min(nbytes >> 4, 4u);
+			sm4_aesni_avx_crypt4(rkey, dst, src, nblocks);
+			dst += nblocks * SM4_BLOCK_SIZE;
+			src += nblocks * SM4_BLOCK_SIZE;
+			nbytes -= nblocks * SM4_BLOCK_SIZE;
+		}
+		kernel_fpu_end();
+
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	return err;
+}
+
+int sm4_avx_ecb_encrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct sm4_ctx *ctx = crypto_skcipher_ctx(tfm);
+
+	return ecb_do_crypt(req, ctx->rkey_enc);
+}
+EXPORT_SYMBOL_GPL(sm4_avx_ecb_encrypt);
+
+int sm4_avx_ecb_decrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct sm4_ctx *ctx = crypto_skcipher_ctx(tfm);
+
+	return ecb_do_crypt(req, ctx->rkey_dec);
+}
+EXPORT_SYMBOL_GPL(sm4_avx_ecb_decrypt);
+
+int sm4_cbc_encrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct sm4_ctx *ctx = crypto_skcipher_ctx(tfm);
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes) > 0) {
+		const u8 *iv = walk.iv;
+		const u8 *src = walk.src.virt.addr;
+		u8 *dst = walk.dst.virt.addr;
+
+		while (nbytes >= SM4_BLOCK_SIZE) {
+			crypto_xor_cpy(dst, src, iv, SM4_BLOCK_SIZE);
+			sm4_crypt_block(ctx->rkey_enc, dst, dst);
+			iv = dst;
+			src += SM4_BLOCK_SIZE;
+			dst += SM4_BLOCK_SIZE;
+			nbytes -= SM4_BLOCK_SIZE;
+		}
+		if (iv != walk.iv)
+			memcpy(walk.iv, iv, SM4_BLOCK_SIZE);
+
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	return err;
+}
+EXPORT_SYMBOL_GPL(sm4_cbc_encrypt);
+
+int sm4_avx_cbc_decrypt(struct skcipher_request *req,
+			unsigned int bsize, sm4_crypt_func func)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct sm4_ctx *ctx = crypto_skcipher_ctx(tfm);
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes) > 0) {
+		const u8 *src = walk.src.virt.addr;
+		u8 *dst = walk.dst.virt.addr;
+
+		kernel_fpu_begin();
+
+		while (nbytes >= bsize) {
+			func(ctx->rkey_dec, dst, src, walk.iv);
+			dst += bsize;
+			src += bsize;
+			nbytes -= bsize;
+		}
+
+		while (nbytes >= SM4_BLOCK_SIZE) {
+			u8 keystream[SM4_BLOCK_SIZE * 8];
+			u8 iv[SM4_BLOCK_SIZE];
+			unsigned int nblocks = min(nbytes >> 4, 8u);
+			int i;
+
+			sm4_aesni_avx_crypt8(ctx->rkey_dec, keystream,
+						src, nblocks);
+
+			src += ((int)nblocks - 2) * SM4_BLOCK_SIZE;
+			dst += (nblocks - 1) * SM4_BLOCK_SIZE;
+			memcpy(iv, src + SM4_BLOCK_SIZE, SM4_BLOCK_SIZE);
+
+			for (i = nblocks - 1; i > 0; i--) {
+				crypto_xor_cpy(dst, src,
+					&keystream[i * SM4_BLOCK_SIZE],
+					SM4_BLOCK_SIZE);
+				src -= SM4_BLOCK_SIZE;
+				dst -= SM4_BLOCK_SIZE;
+			}
+			crypto_xor_cpy(dst, walk.iv, keystream, SM4_BLOCK_SIZE);
+			memcpy(walk.iv, iv, SM4_BLOCK_SIZE);
+			dst += nblocks * SM4_BLOCK_SIZE;
+			src += (nblocks + 1) * SM4_BLOCK_SIZE;
+			nbytes -= nblocks * SM4_BLOCK_SIZE;
+		}
+
+		kernel_fpu_end();
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	return err;
+}
+EXPORT_SYMBOL_GPL(sm4_avx_cbc_decrypt);
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	return sm4_avx_cbc_decrypt(req, SM4_CRYPT8_BLOCK_SIZE,
+				sm4_aesni_avx_cbc_dec_blk8);
+}
+
+int sm4_avx_ctr_crypt(struct skcipher_request *req,
+			unsigned int bsize, sm4_crypt_func func)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct sm4_ctx *ctx = crypto_skcipher_ctx(tfm);
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while ((nbytes = walk.nbytes) > 0) {
+		const u8 *src = walk.src.virt.addr;
+		u8 *dst = walk.dst.virt.addr;
+
+		kernel_fpu_begin();
+
+		while (nbytes >= bsize) {
+			func(ctx->rkey_enc, dst, src, walk.iv);
+			dst += bsize;
+			src += bsize;
+			nbytes -= bsize;
+		}
+
+		while (nbytes >= SM4_BLOCK_SIZE) {
+			u8 keystream[SM4_BLOCK_SIZE * 8];
+			unsigned int nblocks = min(nbytes >> 4, 8u);
+			int i;
+
+			for (i = 0; i < nblocks; i++) {
+				memcpy(&keystream[i * SM4_BLOCK_SIZE],
+					walk.iv, SM4_BLOCK_SIZE);
+				crypto_inc(walk.iv, SM4_BLOCK_SIZE);
+			}
+			sm4_aesni_avx_crypt8(ctx->rkey_enc, keystream,
+					keystream, nblocks);
+
+			crypto_xor_cpy(dst, src, keystream,
+					nblocks * SM4_BLOCK_SIZE);
+			dst += nblocks * SM4_BLOCK_SIZE;
+			src += nblocks * SM4_BLOCK_SIZE;
+			nbytes -= nblocks * SM4_BLOCK_SIZE;
+		}
+
+		kernel_fpu_end();
+
+		/* tail */
+		if (walk.nbytes == walk.total && nbytes > 0) {
+			u8 keystream[SM4_BLOCK_SIZE];
+
+			memcpy(keystream, walk.iv, SM4_BLOCK_SIZE);
+			crypto_inc(walk.iv, SM4_BLOCK_SIZE);
+
+			sm4_crypt_block(ctx->rkey_enc, keystream, keystream);
+
+			crypto_xor_cpy(dst, src, keystream, nbytes);
+			dst += nbytes;
+			src += nbytes;
+			nbytes = 0;
+		}
+
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	return err;
+}
+EXPORT_SYMBOL_GPL(sm4_avx_ctr_crypt);
+
+static int ctr_crypt(struct skcipher_request *req)
+{
+	return sm4_avx_ctr_crypt(req, SM4_CRYPT8_BLOCK_SIZE,
+				sm4_aesni_avx_ctr_enc_blk8);
+}
+
+static struct skcipher_alg sm4_aesni_avx_skciphers[] = {
+	{
+		.base = {
+			.cra_name		= "ecb(sm4)",
+			.cra_driver_name	= "ecb-sm4-aesni-avx",
+			.cra_priority		= 400,
+			.cra_blocksize		= SM4_BLOCK_SIZE,
+			.cra_ctxsize		= sizeof(struct sm4_ctx),
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= SM4_KEY_SIZE,
+		.max_keysize	= SM4_KEY_SIZE,
+		.walksize	= 8 * SM4_BLOCK_SIZE,
+		.setkey		= sm4_skcipher_setkey,
+		.encrypt	= sm4_avx_ecb_encrypt,
+		.decrypt	= sm4_avx_ecb_decrypt,
+	}, {
+		.base = {
+			.cra_name		= "cbc(sm4)",
+			.cra_driver_name	= "cbc-sm4-aesni-avx",
+			.cra_priority		= 400,
+			.cra_blocksize		= SM4_BLOCK_SIZE,
+			.cra_ctxsize		= sizeof(struct sm4_ctx),
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= SM4_KEY_SIZE,
+		.max_keysize	= SM4_KEY_SIZE,
+		.ivsize		= SM4_BLOCK_SIZE,
+		.walksize	= 8 * SM4_BLOCK_SIZE,
+		.setkey		= sm4_skcipher_setkey,
+		.encrypt	= sm4_cbc_encrypt,
+		.decrypt	= cbc_decrypt,
+	}, {
+		.base = {
+			.cra_name		= "ctr(sm4)",
+			.cra_driver_name	= "ctr-sm4-aesni-avx",
+			.cra_priority		= 400,
+			.cra_blocksize		= 1,
+			.cra_ctxsize		= sizeof(struct sm4_ctx),
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= SM4_KEY_SIZE,
+		.max_keysize	= SM4_KEY_SIZE,
+		.ivsize		= SM4_BLOCK_SIZE,
+		.chunksize	= SM4_BLOCK_SIZE,
+		.walksize	= 8 * SM4_BLOCK_SIZE,
+		.setkey		= sm4_skcipher_setkey,
+		.encrypt	= ctr_crypt,
+		.decrypt	= ctr_crypt,
+	}
+};
+
+static int __init sm4_init(void)
+{
+	const char *feature_name;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX) ||
+	    !boot_cpu_has(X86_FEATURE_AES) ||
+	    !boot_cpu_has(X86_FEATURE_OSXSAVE)) {
+		pr_info("AVX or AES-NI instructions are not detected.\n");
+		return -ENODEV;
+	}
+
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM,
+				&feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	return crypto_register_skciphers(sm4_aesni_avx_skciphers,
+					 ARRAY_SIZE(sm4_aesni_avx_skciphers));
+}
+
+static void __exit sm4_exit(void)
+{
+	crypto_unregister_skciphers(sm4_aesni_avx_skciphers,
+				    ARRAY_SIZE(sm4_aesni_avx_skciphers));
+}
+
+module_init(sm4_init);
+module_exit(sm4_exit);
+
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Tianjia Zhang <tianjia.zhang@linux.alibaba.com>");
+MODULE_DESCRIPTION("SM4 Cipher Algorithm, AES-NI/AVX optimized");
+MODULE_ALIAS_CRYPTO("sm4");
+MODULE_ALIAS_CRYPTO("sm4-aesni-avx");
diff --git a/arch/x86/crypto/twofish-avx-x86_64-asm_64.S b/arch/x86/crypto/twofish-avx-x86_64-asm_64.S
new file mode 100644
index 000000000..12fde271c
--- /dev/null
+++ b/arch/x86/crypto/twofish-avx-x86_64-asm_64.S
@@ -0,0 +1,374 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Twofish Cipher 8-way parallel algorithm (AVX/x86_64)
+ *
+ * Copyright (C) 2012 Johannes Goetzfried
+ *     <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
+ *
+ * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ */
+
+#include <linux/linkage.h>
+#include <asm/frame.h>
+#include "glue_helper-asm-avx.S"
+
+.file "twofish-avx-x86_64-asm_64.S"
+
+.section	.rodata.cst16.bswap128_mask, "aM", @progbits, 16
+.align 16
+.Lbswap128_mask:
+	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
+
+.text
+
+/* structure of crypto context */
+#define s0	0
+#define s1	1024
+#define s2	2048
+#define s3	3072
+#define w	4096
+#define k	4128
+
+/**********************************************************************
+  8-way AVX twofish
+ **********************************************************************/
+#define CTX %rdi
+
+#define RA1 %xmm0
+#define RB1 %xmm1
+#define RC1 %xmm2
+#define RD1 %xmm3
+
+#define RA2 %xmm4
+#define RB2 %xmm5
+#define RC2 %xmm6
+#define RD2 %xmm7
+
+#define RX0 %xmm8
+#define RY0 %xmm9
+
+#define RX1 %xmm10
+#define RY1 %xmm11
+
+#define RK1 %xmm12
+#define RK2 %xmm13
+
+#define RT %xmm14
+#define RR %xmm15
+
+#define RID1  %r13
+#define RID1d %r13d
+#define RID2  %rsi
+#define RID2d %esi
+
+#define RGI1   %rdx
+#define RGI1bl %dl
+#define RGI1bh %dh
+#define RGI2   %rcx
+#define RGI2bl %cl
+#define RGI2bh %ch
+
+#define RGI3   %rax
+#define RGI3bl %al
+#define RGI3bh %ah
+#define RGI4   %rbx
+#define RGI4bl %bl
+#define RGI4bh %bh
+
+#define RGS1  %r8
+#define RGS1d %r8d
+#define RGS2  %r9
+#define RGS2d %r9d
+#define RGS3  %r10
+#define RGS3d %r10d
+
+
+#define lookup_32bit(t0, t1, t2, t3, src, dst, interleave_op, il_reg) \
+	movzbl		src ## bl,        RID1d;     \
+	movzbl		src ## bh,        RID2d;     \
+	shrq $16,	src;                         \
+	movl		t0(CTX, RID1, 4), dst ## d;  \
+	movl		t1(CTX, RID2, 4), RID2d;     \
+	movzbl		src ## bl,        RID1d;     \
+	xorl		RID2d,            dst ## d;  \
+	movzbl		src ## bh,        RID2d;     \
+	interleave_op(il_reg);			     \
+	xorl		t2(CTX, RID1, 4), dst ## d;  \
+	xorl		t3(CTX, RID2, 4), dst ## d;
+
+#define dummy(d) /* do nothing */
+
+#define shr_next(reg) \
+	shrq $16,	reg;
+
+#define G(gi1, gi2, x, t0, t1, t2, t3) \
+	lookup_32bit(t0, t1, t2, t3, ##gi1, RGS1, shr_next, ##gi1);  \
+	lookup_32bit(t0, t1, t2, t3, ##gi2, RGS3, shr_next, ##gi2);  \
+	\
+	lookup_32bit(t0, t1, t2, t3, ##gi1, RGS2, dummy, none);      \
+	shlq $32,	RGS2;                                        \
+	orq		RGS1, RGS2;                                  \
+	lookup_32bit(t0, t1, t2, t3, ##gi2, RGS1, dummy, none);      \
+	shlq $32,	RGS1;                                        \
+	orq		RGS1, RGS3;
+
+#define round_head_2(a, b, x1, y1, x2, y2) \
+	vmovq		b ## 1, RGI3;           \
+	vpextrq $1,	b ## 1, RGI4;           \
+	\
+	G(RGI1, RGI2, x1, s0, s1, s2, s3);      \
+	vmovq		a ## 2, RGI1;           \
+	vpextrq $1,	a ## 2, RGI2;           \
+	vmovq		RGS2, x1;               \
+	vpinsrq $1,	RGS3, x1, x1;           \
+	\
+	G(RGI3, RGI4, y1, s1, s2, s3, s0);      \
+	vmovq		b ## 2, RGI3;           \
+	vpextrq $1,	b ## 2, RGI4;           \
+	vmovq		RGS2, y1;               \
+	vpinsrq $1,	RGS3, y1, y1;           \
+	\
+	G(RGI1, RGI2, x2, s0, s1, s2, s3);      \
+	vmovq		RGS2, x2;               \
+	vpinsrq $1,	RGS3, x2, x2;           \
+	\
+	G(RGI3, RGI4, y2, s1, s2, s3, s0);      \
+	vmovq		RGS2, y2;               \
+	vpinsrq $1,	RGS3, y2, y2;
+
+#define encround_tail(a, b, c, d, x, y, prerotate) \
+	vpaddd			x, y,   x; \
+	vpaddd			x, RK1, RT;\
+	prerotate(b);			   \
+	vpxor			RT, c,  c; \
+	vpaddd			y, x,   y; \
+	vpaddd			y, RK2, y; \
+	vpsrld $1,		c, RT;     \
+	vpslld $(32 - 1),	c, c;      \
+	vpor			c, RT,  c; \
+	vpxor			d, y,   d; \
+
+#define decround_tail(a, b, c, d, x, y, prerotate) \
+	vpaddd			x, y,   x; \
+	vpaddd			x, RK1, RT;\
+	prerotate(a);			   \
+	vpxor			RT, c,  c; \
+	vpaddd			y, x,   y; \
+	vpaddd			y, RK2, y; \
+	vpxor			d, y,   d; \
+	vpsrld $1,		d, y;      \
+	vpslld $(32 - 1),	d, d;      \
+	vpor			d, y,   d; \
+
+#define rotate_1l(x) \
+	vpslld $1,		x, RR;     \
+	vpsrld $(32 - 1),	x, x;      \
+	vpor			x, RR,  x;
+
+#define preload_rgi(c) \
+	vmovq			c, RGI1; \
+	vpextrq $1,		c, RGI2;
+
+#define encrypt_round(n, a, b, c, d, preload, prerotate) \
+	vbroadcastss (k+4*(2*(n)))(CTX),   RK1;                  \
+	vbroadcastss (k+4*(2*(n)+1))(CTX), RK2;                  \
+	round_head_2(a, b, RX0, RY0, RX1, RY1);                  \
+	encround_tail(a ## 1, b ## 1, c ## 1, d ## 1, RX0, RY0, prerotate); \
+	preload(c ## 1);                                         \
+	encround_tail(a ## 2, b ## 2, c ## 2, d ## 2, RX1, RY1, prerotate);
+
+#define decrypt_round(n, a, b, c, d, preload, prerotate) \
+	vbroadcastss (k+4*(2*(n)))(CTX),   RK1;                  \
+	vbroadcastss (k+4*(2*(n)+1))(CTX), RK2;                  \
+	round_head_2(a, b, RX0, RY0, RX1, RY1);                  \
+	decround_tail(a ## 1, b ## 1, c ## 1, d ## 1, RX0, RY0, prerotate); \
+	preload(c ## 1);                                         \
+	decround_tail(a ## 2, b ## 2, c ## 2, d ## 2, RX1, RY1, prerotate);
+
+#define encrypt_cycle(n) \
+	encrypt_round((2*n), RA, RB, RC, RD, preload_rgi, rotate_1l); \
+	encrypt_round(((2*n) + 1), RC, RD, RA, RB, preload_rgi, rotate_1l);
+
+#define encrypt_cycle_last(n) \
+	encrypt_round((2*n), RA, RB, RC, RD, preload_rgi, rotate_1l); \
+	encrypt_round(((2*n) + 1), RC, RD, RA, RB, dummy, dummy);
+
+#define decrypt_cycle(n) \
+	decrypt_round(((2*n) + 1), RC, RD, RA, RB, preload_rgi, rotate_1l); \
+	decrypt_round((2*n), RA, RB, RC, RD, preload_rgi, rotate_1l);
+
+#define decrypt_cycle_last(n) \
+	decrypt_round(((2*n) + 1), RC, RD, RA, RB, preload_rgi, rotate_1l); \
+	decrypt_round((2*n), RA, RB, RC, RD, dummy, dummy);
+
+#define transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	vpunpckldq		x1, x0, t0; \
+	vpunpckhdq		x1, x0, t2; \
+	vpunpckldq		x3, x2, t1; \
+	vpunpckhdq		x3, x2, x3; \
+	\
+	vpunpcklqdq		t1, t0, x0; \
+	vpunpckhqdq		t1, t0, x1; \
+	vpunpcklqdq		x3, t2, x2; \
+	vpunpckhqdq		x3, t2, x3;
+
+#define inpack_blocks(x0, x1, x2, x3, wkey, t0, t1, t2) \
+	vpxor		x0, wkey, x0; \
+	vpxor		x1, wkey, x1; \
+	vpxor		x2, wkey, x2; \
+	vpxor		x3, wkey, x3; \
+	\
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2)
+
+#define outunpack_blocks(x0, x1, x2, x3, wkey, t0, t1, t2) \
+	transpose_4x4(x0, x1, x2, x3, t0, t1, t2) \
+	\
+	vpxor		x0, wkey, x0; \
+	vpxor		x1, wkey, x1; \
+	vpxor		x2, wkey, x2; \
+	vpxor		x3, wkey, x3;
+
+SYM_FUNC_START_LOCAL(__twofish_enc_blk8)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: blocks
+	 * output:
+	 *	RC1, RD1, RA1, RB1, RC2, RD2, RA2, RB2: encrypted blocks
+	 */
+
+	vmovdqu w(CTX), RK1;
+
+	pushq %r13;
+	pushq %rbx;
+	pushq %rcx;
+
+	inpack_blocks(RA1, RB1, RC1, RD1, RK1, RX0, RY0, RK2);
+	preload_rgi(RA1);
+	rotate_1l(RD1);
+	inpack_blocks(RA2, RB2, RC2, RD2, RK1, RX0, RY0, RK2);
+	rotate_1l(RD2);
+
+	encrypt_cycle(0);
+	encrypt_cycle(1);
+	encrypt_cycle(2);
+	encrypt_cycle(3);
+	encrypt_cycle(4);
+	encrypt_cycle(5);
+	encrypt_cycle(6);
+	encrypt_cycle_last(7);
+
+	vmovdqu (w+4*4)(CTX), RK1;
+
+	popq %rcx;
+	popq %rbx;
+	popq %r13;
+
+	outunpack_blocks(RC1, RD1, RA1, RB1, RK1, RX0, RY0, RK2);
+	outunpack_blocks(RC2, RD2, RA2, RB2, RK1, RX0, RY0, RK2);
+
+	RET;
+SYM_FUNC_END(__twofish_enc_blk8)
+
+SYM_FUNC_START_LOCAL(__twofish_dec_blk8)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	RC1, RD1, RA1, RB1, RC2, RD2, RA2, RB2: encrypted blocks
+	 * output:
+	 *	RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2: decrypted blocks
+	 */
+
+	vmovdqu (w+4*4)(CTX), RK1;
+
+	pushq %r13;
+	pushq %rbx;
+
+	inpack_blocks(RC1, RD1, RA1, RB1, RK1, RX0, RY0, RK2);
+	preload_rgi(RC1);
+	rotate_1l(RA1);
+	inpack_blocks(RC2, RD2, RA2, RB2, RK1, RX0, RY0, RK2);
+	rotate_1l(RA2);
+
+	decrypt_cycle(7);
+	decrypt_cycle(6);
+	decrypt_cycle(5);
+	decrypt_cycle(4);
+	decrypt_cycle(3);
+	decrypt_cycle(2);
+	decrypt_cycle(1);
+	decrypt_cycle_last(0);
+
+	vmovdqu (w)(CTX), RK1;
+
+	popq %rbx;
+	popq %r13;
+
+	outunpack_blocks(RA1, RB1, RC1, RD1, RK1, RX0, RY0, RK2);
+	outunpack_blocks(RA2, RB2, RC2, RD2, RK1, RX0, RY0, RK2);
+
+	RET;
+SYM_FUNC_END(__twofish_dec_blk8)
+
+SYM_FUNC_START(twofish_ecb_enc_8way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+
+	movq %rsi, %r11;
+
+	load_8way(%rdx, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	call __twofish_enc_blk8;
+
+	store_8way(%r11, RC1, RD1, RA1, RB1, RC2, RD2, RA2, RB2);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(twofish_ecb_enc_8way)
+
+SYM_FUNC_START(twofish_ecb_dec_8way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+
+	movq %rsi, %r11;
+
+	load_8way(%rdx, RC1, RD1, RA1, RB1, RC2, RD2, RA2, RB2);
+
+	call __twofish_dec_blk8;
+
+	store_8way(%r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(twofish_ecb_dec_8way)
+
+SYM_FUNC_START(twofish_cbc_dec_8way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src
+	 */
+	FRAME_BEGIN
+
+	pushq %r12;
+
+	movq %rsi, %r11;
+	movq %rdx, %r12;
+
+	load_8way(%rdx, RC1, RD1, RA1, RB1, RC2, RD2, RA2, RB2);
+
+	call __twofish_dec_blk8;
+
+	store_cbc_8way(%r12, %r11, RA1, RB1, RC1, RD1, RA2, RB2, RC2, RD2);
+
+	popq %r12;
+
+	FRAME_END
+	RET;
+SYM_FUNC_END(twofish_cbc_dec_8way)
diff --git a/arch/x86/crypto/twofish-i586-asm_32.S b/arch/x86/crypto/twofish-i586-asm_32.S
new file mode 100644
index 000000000..3abcad661
--- /dev/null
+++ b/arch/x86/crypto/twofish-i586-asm_32.S
@@ -0,0 +1,321 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/***************************************************************************
+*   Copyright (C) 2006 by Joachim Fritschi, <jfritschi@freenet.de>        *
+*                                                                         *
+***************************************************************************/
+
+.file "twofish-i586-asm.S"
+.text
+
+#include <linux/linkage.h>
+#include <asm/asm-offsets.h>
+
+/* return address at 0 */
+
+#define in_blk    12  /* input byte array address parameter*/
+#define out_blk   8  /* output byte array address parameter*/
+#define ctx       4  /* Twofish context structure */
+
+#define a_offset	0
+#define b_offset	4
+#define c_offset	8
+#define d_offset	12
+
+/* Structure of the crypto context struct*/
+
+#define s0	0	/* S0 Array 256 Words each */
+#define s1	1024	/* S1 Array */
+#define s2	2048	/* S2 Array */
+#define s3	3072	/* S3 Array */
+#define w	4096	/* 8 whitening keys (word) */
+#define k	4128	/* key 1-32 ( word ) */
+
+/* define a few register aliases to allow macro substitution */
+
+#define R0D    %eax
+#define R0B    %al
+#define R0H    %ah
+
+#define R1D    %ebx
+#define R1B    %bl
+#define R1H    %bh
+
+#define R2D    %ecx
+#define R2B    %cl
+#define R2H    %ch
+
+#define R3D    %edx
+#define R3B    %dl
+#define R3H    %dh
+
+
+/* performs input whitening */
+#define input_whitening(src,context,offset)\
+	xor	w+offset(context),	src;
+
+/* performs input whitening */
+#define output_whitening(src,context,offset)\
+	xor	w+16+offset(context),	src;
+
+/*
+ * a input register containing a (rotated 16)
+ * b input register containing b
+ * c input register containing c
+ * d input register containing d (already rol $1)
+ * operations on a and b are interleaved to increase performance
+ */
+#define encrypt_round(a,b,c,d,round)\
+	push	d ## D;\
+	movzx	b ## B,		%edi;\
+	mov	s1(%ebp,%edi,4),d ## D;\
+	movzx	a ## B,		%edi;\
+	mov	s2(%ebp,%edi,4),%esi;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	s2(%ebp,%edi,4),d ## D;\
+	movzx	a ## H,		%edi;\
+	ror	$16,		a ## D;\
+	xor	s3(%ebp,%edi,4),%esi;\
+	movzx	b ## B,		%edi;\
+	xor	s3(%ebp,%edi,4),d ## D;\
+	movzx	a ## B,		%edi;\
+	xor	(%ebp,%edi,4),	%esi;\
+	movzx	b ## H,		%edi;\
+	ror	$15,		b ## D;\
+	xor	(%ebp,%edi,4),	d ## D;\
+	movzx	a ## H,		%edi;\
+	xor	s1(%ebp,%edi,4),%esi;\
+	pop	%edi;\
+	add	d ## D,		%esi;\
+	add	%esi,		d ## D;\
+	add	k+round(%ebp),	%esi;\
+	xor	%esi,		c ## D;\
+	rol	$15,		c ## D;\
+	add	k+4+round(%ebp),d ## D;\
+	xor	%edi,		d ## D;
+
+/*
+ * a input register containing a (rotated 16)
+ * b input register containing b
+ * c input register containing c
+ * d input register containing d (already rol $1)
+ * operations on a and b are interleaved to increase performance
+ * last round has different rotations for the output preparation
+ */
+#define encrypt_last_round(a,b,c,d,round)\
+	push	d ## D;\
+	movzx	b ## B,		%edi;\
+	mov	s1(%ebp,%edi,4),d ## D;\
+	movzx	a ## B,		%edi;\
+	mov	s2(%ebp,%edi,4),%esi;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	s2(%ebp,%edi,4),d ## D;\
+	movzx	a ## H,		%edi;\
+	ror	$16,		a ## D;\
+	xor	s3(%ebp,%edi,4),%esi;\
+	movzx	b ## B,		%edi;\
+	xor	s3(%ebp,%edi,4),d ## D;\
+	movzx	a ## B,		%edi;\
+	xor	(%ebp,%edi,4),	%esi;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	(%ebp,%edi,4),	d ## D;\
+	movzx	a ## H,		%edi;\
+	xor	s1(%ebp,%edi,4),%esi;\
+	pop	%edi;\
+	add	d ## D,		%esi;\
+	add	%esi,		d ## D;\
+	add	k+round(%ebp),	%esi;\
+	xor	%esi,		c ## D;\
+	ror	$1,		c ## D;\
+	add	k+4+round(%ebp),d ## D;\
+	xor	%edi,		d ## D;
+
+/*
+ * a input register containing a
+ * b input register containing b (rotated 16)
+ * c input register containing c
+ * d input register containing d (already rol $1)
+ * operations on a and b are interleaved to increase performance
+ */
+#define decrypt_round(a,b,c,d,round)\
+	push	c ## D;\
+	movzx	a ## B,		%edi;\
+	mov	(%ebp,%edi,4),	c ## D;\
+	movzx	b ## B,		%edi;\
+	mov	s3(%ebp,%edi,4),%esi;\
+	movzx	a ## H,		%edi;\
+	ror	$16,		a ## D;\
+	xor	s1(%ebp,%edi,4),c ## D;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	(%ebp,%edi,4),	%esi;\
+	movzx	a ## B,		%edi;\
+	xor	s2(%ebp,%edi,4),c ## D;\
+	movzx	b ## B,		%edi;\
+	xor	s1(%ebp,%edi,4),%esi;\
+	movzx	a ## H,		%edi;\
+	ror	$15,		a ## D;\
+	xor	s3(%ebp,%edi,4),c ## D;\
+	movzx	b ## H,		%edi;\
+	xor	s2(%ebp,%edi,4),%esi;\
+	pop	%edi;\
+	add	%esi,		c ## D;\
+	add	c ## D,		%esi;\
+	add	k+round(%ebp),	c ## D;\
+	xor	%edi,		c ## D;\
+	add	k+4+round(%ebp),%esi;\
+	xor	%esi,		d ## D;\
+	rol	$15,		d ## D;
+
+/*
+ * a input register containing a
+ * b input register containing b (rotated 16)
+ * c input register containing c
+ * d input register containing d (already rol $1)
+ * operations on a and b are interleaved to increase performance
+ * last round has different rotations for the output preparation
+ */
+#define decrypt_last_round(a,b,c,d,round)\
+	push	c ## D;\
+	movzx	a ## B,		%edi;\
+	mov	(%ebp,%edi,4),	c ## D;\
+	movzx	b ## B,		%edi;\
+	mov	s3(%ebp,%edi,4),%esi;\
+	movzx	a ## H,		%edi;\
+	ror	$16,		a ## D;\
+	xor	s1(%ebp,%edi,4),c ## D;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	(%ebp,%edi,4),	%esi;\
+	movzx	a ## B,		%edi;\
+	xor	s2(%ebp,%edi,4),c ## D;\
+	movzx	b ## B,		%edi;\
+	xor	s1(%ebp,%edi,4),%esi;\
+	movzx	a ## H,		%edi;\
+	ror	$16,		a ## D;\
+	xor	s3(%ebp,%edi,4),c ## D;\
+	movzx	b ## H,		%edi;\
+	xor	s2(%ebp,%edi,4),%esi;\
+	pop	%edi;\
+	add	%esi,		c ## D;\
+	add	c ## D,		%esi;\
+	add	k+round(%ebp),	c ## D;\
+	xor	%edi,		c ## D;\
+	add	k+4+round(%ebp),%esi;\
+	xor	%esi,		d ## D;\
+	ror	$1,		d ## D;
+
+SYM_FUNC_START(twofish_enc_blk)
+	push	%ebp			/* save registers according to calling convention*/
+	push    %ebx
+	push    %esi
+	push    %edi
+
+	mov	ctx + 16(%esp),	%ebp	/* abuse the base pointer: set new base
+					 * pointer to the ctx address */
+	mov     in_blk+16(%esp),%edi	/* input address in edi */
+
+	mov	(%edi),		%eax
+	mov	b_offset(%edi),	%ebx
+	mov	c_offset(%edi),	%ecx
+	mov	d_offset(%edi),	%edx
+	input_whitening(%eax,%ebp,a_offset)
+	ror	$16,	%eax
+	input_whitening(%ebx,%ebp,b_offset)
+	input_whitening(%ecx,%ebp,c_offset)
+	input_whitening(%edx,%ebp,d_offset)
+	rol	$1,	%edx
+
+	encrypt_round(R0,R1,R2,R3,0);
+	encrypt_round(R2,R3,R0,R1,8);
+	encrypt_round(R0,R1,R2,R3,2*8);
+	encrypt_round(R2,R3,R0,R1,3*8);
+	encrypt_round(R0,R1,R2,R3,4*8);
+	encrypt_round(R2,R3,R0,R1,5*8);
+	encrypt_round(R0,R1,R2,R3,6*8);
+	encrypt_round(R2,R3,R0,R1,7*8);
+	encrypt_round(R0,R1,R2,R3,8*8);
+	encrypt_round(R2,R3,R0,R1,9*8);
+	encrypt_round(R0,R1,R2,R3,10*8);
+	encrypt_round(R2,R3,R0,R1,11*8);
+	encrypt_round(R0,R1,R2,R3,12*8);
+	encrypt_round(R2,R3,R0,R1,13*8);
+	encrypt_round(R0,R1,R2,R3,14*8);
+	encrypt_last_round(R2,R3,R0,R1,15*8);
+
+	output_whitening(%eax,%ebp,c_offset)
+	output_whitening(%ebx,%ebp,d_offset)
+	output_whitening(%ecx,%ebp,a_offset)
+	output_whitening(%edx,%ebp,b_offset)
+	mov	out_blk+16(%esp),%edi;
+	mov	%eax,		c_offset(%edi)
+	mov	%ebx,		d_offset(%edi)
+	mov	%ecx,		(%edi)
+	mov	%edx,		b_offset(%edi)
+
+	pop	%edi
+	pop	%esi
+	pop	%ebx
+	pop	%ebp
+	mov	$1,	%eax
+	RET
+SYM_FUNC_END(twofish_enc_blk)
+
+SYM_FUNC_START(twofish_dec_blk)
+	push	%ebp			/* save registers according to calling convention*/
+	push    %ebx
+	push    %esi
+	push    %edi
+
+
+	mov	ctx + 16(%esp),	%ebp	/* abuse the base pointer: set new base
+					 * pointer to the ctx address */
+	mov     in_blk+16(%esp),%edi	/* input address in edi */
+
+	mov	(%edi),		%eax
+	mov	b_offset(%edi),	%ebx
+	mov	c_offset(%edi),	%ecx
+	mov	d_offset(%edi),	%edx
+	output_whitening(%eax,%ebp,a_offset)
+	output_whitening(%ebx,%ebp,b_offset)
+	ror	$16,	%ebx
+	output_whitening(%ecx,%ebp,c_offset)
+	output_whitening(%edx,%ebp,d_offset)
+	rol	$1,	%ecx
+
+	decrypt_round(R0,R1,R2,R3,15*8);
+	decrypt_round(R2,R3,R0,R1,14*8);
+	decrypt_round(R0,R1,R2,R3,13*8);
+	decrypt_round(R2,R3,R0,R1,12*8);
+	decrypt_round(R0,R1,R2,R3,11*8);
+	decrypt_round(R2,R3,R0,R1,10*8);
+	decrypt_round(R0,R1,R2,R3,9*8);
+	decrypt_round(R2,R3,R0,R1,8*8);
+	decrypt_round(R0,R1,R2,R3,7*8);
+	decrypt_round(R2,R3,R0,R1,6*8);
+	decrypt_round(R0,R1,R2,R3,5*8);
+	decrypt_round(R2,R3,R0,R1,4*8);
+	decrypt_round(R0,R1,R2,R3,3*8);
+	decrypt_round(R2,R3,R0,R1,2*8);
+	decrypt_round(R0,R1,R2,R3,1*8);
+	decrypt_last_round(R2,R3,R0,R1,0);
+
+	input_whitening(%eax,%ebp,c_offset)
+	input_whitening(%ebx,%ebp,d_offset)
+	input_whitening(%ecx,%ebp,a_offset)
+	input_whitening(%edx,%ebp,b_offset)
+	mov	out_blk+16(%esp),%edi;
+	mov	%eax,		c_offset(%edi)
+	mov	%ebx,		d_offset(%edi)
+	mov	%ecx,		(%edi)
+	mov	%edx,		b_offset(%edi)
+
+	pop	%edi
+	pop	%esi
+	pop	%ebx
+	pop	%ebp
+	mov	$1,	%eax
+	RET
+SYM_FUNC_END(twofish_dec_blk)
diff --git a/arch/x86/crypto/twofish-x86_64-asm_64-3way.S b/arch/x86/crypto/twofish-x86_64-asm_64-3way.S
new file mode 100644
index 000000000..071e90e7f
--- /dev/null
+++ b/arch/x86/crypto/twofish-x86_64-asm_64-3way.S
@@ -0,0 +1,306 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Twofish Cipher 3-way parallel algorithm (x86_64)
+ *
+ * Copyright (C) 2011 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ */
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+
+.file "twofish-x86_64-asm-3way.S"
+.text
+
+/* structure of crypto context */
+#define s0	0
+#define s1	1024
+#define s2	2048
+#define s3	3072
+#define w	4096
+#define k	4128
+
+/**********************************************************************
+  3-way twofish
+ **********************************************************************/
+#define CTX %rdi
+#define RIO %rdx
+
+#define RAB0 %rax
+#define RAB1 %rbx
+#define RAB2 %rcx
+
+#define RAB0d %eax
+#define RAB1d %ebx
+#define RAB2d %ecx
+
+#define RAB0bh %ah
+#define RAB1bh %bh
+#define RAB2bh %ch
+
+#define RAB0bl %al
+#define RAB1bl %bl
+#define RAB2bl %cl
+
+#define CD0 0x0(%rsp)
+#define CD1 0x8(%rsp)
+#define CD2 0x10(%rsp)
+
+# used only before/after all rounds
+#define RCD0 %r8
+#define RCD1 %r9
+#define RCD2 %r10
+
+# used only during rounds
+#define RX0 %r8
+#define RX1 %r9
+#define RX2 %r10
+
+#define RX0d %r8d
+#define RX1d %r9d
+#define RX2d %r10d
+
+#define RY0 %r11
+#define RY1 %r12
+#define RY2 %r13
+
+#define RY0d %r11d
+#define RY1d %r12d
+#define RY2d %r13d
+
+#define RT0 %rdx
+#define RT1 %rsi
+
+#define RT0d %edx
+#define RT1d %esi
+
+#define RT1bl %sil
+
+#define do16bit_ror(rot, op1, op2, T0, T1, tmp1, tmp2, ab, dst) \
+	movzbl ab ## bl,		tmp2 ## d; \
+	movzbl ab ## bh,		tmp1 ## d; \
+	rorq $(rot),			ab; \
+	op1##l T0(CTX, tmp2, 4),	dst ## d; \
+	op2##l T1(CTX, tmp1, 4),	dst ## d;
+
+#define swap_ab_with_cd(ab, cd, tmp)	\
+	movq cd, tmp;			\
+	movq ab, cd;			\
+	movq tmp, ab;
+
+/*
+ * Combined G1 & G2 function. Reordered with help of rotates to have moves
+ * at beginning.
+ */
+#define g1g2_3(ab, cd, Tx0, Tx1, Tx2, Tx3, Ty0, Ty1, Ty2, Ty3, x, y) \
+	/* G1,1 && G2,1 */ \
+	do16bit_ror(32, mov, xor, Tx0, Tx1, RT0, x ## 0, ab ## 0, x ## 0); \
+	do16bit_ror(48, mov, xor, Ty1, Ty2, RT0, y ## 0, ab ## 0, y ## 0); \
+	\
+	do16bit_ror(32, mov, xor, Tx0, Tx1, RT0, x ## 1, ab ## 1, x ## 1); \
+	do16bit_ror(48, mov, xor, Ty1, Ty2, RT0, y ## 1, ab ## 1, y ## 1); \
+	\
+	do16bit_ror(32, mov, xor, Tx0, Tx1, RT0, x ## 2, ab ## 2, x ## 2); \
+	do16bit_ror(48, mov, xor, Ty1, Ty2, RT0, y ## 2, ab ## 2, y ## 2); \
+	\
+	/* G1,2 && G2,2 */ \
+	do16bit_ror(32, xor, xor, Tx2, Tx3, RT0, RT1, ab ## 0, x ## 0); \
+	do16bit_ror(16, xor, xor, Ty3, Ty0, RT0, RT1, ab ## 0, y ## 0); \
+	swap_ab_with_cd(ab ## 0, cd ## 0, RT0); \
+	\
+	do16bit_ror(32, xor, xor, Tx2, Tx3, RT0, RT1, ab ## 1, x ## 1); \
+	do16bit_ror(16, xor, xor, Ty3, Ty0, RT0, RT1, ab ## 1, y ## 1); \
+	swap_ab_with_cd(ab ## 1, cd ## 1, RT0); \
+	\
+	do16bit_ror(32, xor, xor, Tx2, Tx3, RT0, RT1, ab ## 2, x ## 2); \
+	do16bit_ror(16, xor, xor, Ty3, Ty0, RT0, RT1, ab ## 2, y ## 2); \
+	swap_ab_with_cd(ab ## 2, cd ## 2, RT0);
+
+#define enc_round_end(ab, x, y, n) \
+	addl y ## d,			x ## d; \
+	addl x ## d,			y ## d; \
+	addl k+4*(2*(n))(CTX),		x ## d; \
+	xorl ab ## d,			x ## d; \
+	addl k+4*(2*(n)+1)(CTX),	y ## d; \
+	shrq $32,			ab; \
+	roll $1,			ab ## d; \
+	xorl y ## d,			ab ## d; \
+	shlq $32,			ab; \
+	rorl $1,			x ## d; \
+	orq x,				ab;
+
+#define dec_round_end(ba, x, y, n) \
+	addl y ## d,			x ## d; \
+	addl x ## d,			y ## d; \
+	addl k+4*(2*(n))(CTX),		x ## d; \
+	addl k+4*(2*(n)+1)(CTX),	y ## d; \
+	xorl ba ## d,			y ## d; \
+	shrq $32,			ba; \
+	roll $1,			ba ## d; \
+	xorl x ## d,			ba ## d; \
+	shlq $32,			ba; \
+	rorl $1,			y ## d; \
+	orq y,				ba;
+
+#define encrypt_round3(ab, cd, n) \
+	g1g2_3(ab, cd, s0, s1, s2, s3, s0, s1, s2, s3, RX, RY); \
+	\
+	enc_round_end(ab ## 0, RX0, RY0, n); \
+	enc_round_end(ab ## 1, RX1, RY1, n); \
+	enc_round_end(ab ## 2, RX2, RY2, n);
+
+#define decrypt_round3(ba, dc, n) \
+	g1g2_3(ba, dc, s1, s2, s3, s0, s3, s0, s1, s2, RY, RX); \
+	\
+	dec_round_end(ba ## 0, RX0, RY0, n); \
+	dec_round_end(ba ## 1, RX1, RY1, n); \
+	dec_round_end(ba ## 2, RX2, RY2, n);
+
+#define encrypt_cycle3(ab, cd, n) \
+	encrypt_round3(ab, cd, n*2); \
+	encrypt_round3(ab, cd, (n*2)+1);
+
+#define decrypt_cycle3(ba, dc, n) \
+	decrypt_round3(ba, dc, (n*2)+1); \
+	decrypt_round3(ba, dc, (n*2));
+
+#define push_cd()	\
+	pushq RCD2;	\
+	pushq RCD1;	\
+	pushq RCD0;
+
+#define pop_cd()	\
+	popq RCD0;	\
+	popq RCD1;	\
+	popq RCD2;
+
+#define inpack3(in, n, xy, m) \
+	movq 4*(n)(in),			xy ## 0; \
+	xorq w+4*m(CTX),		xy ## 0; \
+	\
+	movq 4*(4+(n))(in),		xy ## 1; \
+	xorq w+4*m(CTX),		xy ## 1; \
+	\
+	movq 4*(8+(n))(in),		xy ## 2; \
+	xorq w+4*m(CTX),		xy ## 2;
+
+#define outunpack3(op, out, n, xy, m) \
+	xorq w+4*m(CTX),		xy ## 0; \
+	op ## q xy ## 0,		4*(n)(out); \
+	\
+	xorq w+4*m(CTX),		xy ## 1; \
+	op ## q xy ## 1,		4*(4+(n))(out); \
+	\
+	xorq w+4*m(CTX),		xy ## 2; \
+	op ## q xy ## 2,		4*(8+(n))(out);
+
+#define inpack_enc3() \
+	inpack3(RIO, 0, RAB, 0); \
+	inpack3(RIO, 2, RCD, 2);
+
+#define outunpack_enc3(op) \
+	outunpack3(op, RIO, 2, RAB, 6); \
+	outunpack3(op, RIO, 0, RCD, 4);
+
+#define inpack_dec3() \
+	inpack3(RIO, 0, RAB, 4); \
+	rorq $32,			RAB0; \
+	rorq $32,			RAB1; \
+	rorq $32,			RAB2; \
+	inpack3(RIO, 2, RCD, 6); \
+	rorq $32,			RCD0; \
+	rorq $32,			RCD1; \
+	rorq $32,			RCD2;
+
+#define outunpack_dec3() \
+	rorq $32,			RCD0; \
+	rorq $32,			RCD1; \
+	rorq $32,			RCD2; \
+	outunpack3(mov, RIO, 0, RCD, 0); \
+	rorq $32,			RAB0; \
+	rorq $32,			RAB1; \
+	rorq $32,			RAB2; \
+	outunpack3(mov, RIO, 2, RAB, 2);
+
+SYM_TYPED_FUNC_START(__twofish_enc_blk_3way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src, RIO
+	 *	%rcx: bool, if true: xor output
+	 */
+	pushq %r13;
+	pushq %r12;
+	pushq %rbx;
+
+	pushq %rcx; /* bool xor */
+	pushq %rsi; /* dst */
+
+	inpack_enc3();
+
+	push_cd();
+	encrypt_cycle3(RAB, CD, 0);
+	encrypt_cycle3(RAB, CD, 1);
+	encrypt_cycle3(RAB, CD, 2);
+	encrypt_cycle3(RAB, CD, 3);
+	encrypt_cycle3(RAB, CD, 4);
+	encrypt_cycle3(RAB, CD, 5);
+	encrypt_cycle3(RAB, CD, 6);
+	encrypt_cycle3(RAB, CD, 7);
+	pop_cd();
+
+	popq RIO; /* dst */
+	popq RT1; /* bool xor */
+
+	testb RT1bl, RT1bl;
+	jnz .L__enc_xor3;
+
+	outunpack_enc3(mov);
+
+	popq %rbx;
+	popq %r12;
+	popq %r13;
+	RET;
+
+.L__enc_xor3:
+	outunpack_enc3(xor);
+
+	popq %rbx;
+	popq %r12;
+	popq %r13;
+	RET;
+SYM_FUNC_END(__twofish_enc_blk_3way)
+
+SYM_TYPED_FUNC_START(twofish_dec_blk_3way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst
+	 *	%rdx: src, RIO
+	 */
+	pushq %r13;
+	pushq %r12;
+	pushq %rbx;
+
+	pushq %rsi; /* dst */
+
+	inpack_dec3();
+
+	push_cd();
+	decrypt_cycle3(RAB, CD, 7);
+	decrypt_cycle3(RAB, CD, 6);
+	decrypt_cycle3(RAB, CD, 5);
+	decrypt_cycle3(RAB, CD, 4);
+	decrypt_cycle3(RAB, CD, 3);
+	decrypt_cycle3(RAB, CD, 2);
+	decrypt_cycle3(RAB, CD, 1);
+	decrypt_cycle3(RAB, CD, 0);
+	pop_cd();
+
+	popq RIO; /* dst */
+
+	outunpack_dec3();
+
+	popq %rbx;
+	popq %r12;
+	popq %r13;
+	RET;
+SYM_FUNC_END(twofish_dec_blk_3way)
diff --git a/arch/x86/crypto/twofish-x86_64-asm_64.S b/arch/x86/crypto/twofish-x86_64-asm_64.S
new file mode 100644
index 000000000..e08b4ba07
--- /dev/null
+++ b/arch/x86/crypto/twofish-x86_64-asm_64.S
@@ -0,0 +1,309 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/***************************************************************************
+*   Copyright (C) 2006 by Joachim Fritschi, <jfritschi@freenet.de>        *
+*                                                                         *
+***************************************************************************/
+
+.file "twofish-x86_64-asm.S"
+.text
+
+#include <linux/linkage.h>
+#include <linux/cfi_types.h>
+#include <asm/asm-offsets.h>
+
+#define a_offset	0
+#define b_offset	4
+#define c_offset	8
+#define d_offset	12
+
+/* Structure of the crypto context struct*/
+
+#define s0	0	/* S0 Array 256 Words each */
+#define s1	1024	/* S1 Array */
+#define s2	2048	/* S2 Array */
+#define s3	3072	/* S3 Array */
+#define w	4096	/* 8 whitening keys (word) */
+#define k	4128	/* key 1-32 ( word ) */
+
+/* define a few register aliases to allow macro substitution */
+
+#define R0     %rax
+#define R0D    %eax
+#define R0B    %al
+#define R0H    %ah
+
+#define R1     %rbx
+#define R1D    %ebx
+#define R1B    %bl
+#define R1H    %bh
+
+#define R2     %rcx
+#define R2D    %ecx
+#define R2B    %cl
+#define R2H    %ch
+
+#define R3     %rdx
+#define R3D    %edx
+#define R3B    %dl
+#define R3H    %dh
+
+
+/* performs input whitening */
+#define input_whitening(src,context,offset)\
+	xor	w+offset(context),	src;
+
+/* performs input whitening */
+#define output_whitening(src,context,offset)\
+	xor	w+16+offset(context),	src;
+
+
+/*
+ * a input register containing a (rotated 16)
+ * b input register containing b
+ * c input register containing c
+ * d input register containing d (already rol $1)
+ * operations on a and b are interleaved to increase performance
+ */
+#define encrypt_round(a,b,c,d,round)\
+	movzx	b ## B,		%edi;\
+	mov	s1(%r11,%rdi,4),%r8d;\
+	movzx	a ## B,		%edi;\
+	mov	s2(%r11,%rdi,4),%r9d;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	s2(%r11,%rdi,4),%r8d;\
+	movzx	a ## H,		%edi;\
+	ror	$16,		a ## D;\
+	xor	s3(%r11,%rdi,4),%r9d;\
+	movzx	b ## B,		%edi;\
+	xor	s3(%r11,%rdi,4),%r8d;\
+	movzx	a ## B,		%edi;\
+	xor	(%r11,%rdi,4),	%r9d;\
+	movzx	b ## H,		%edi;\
+	ror	$15,		b ## D;\
+	xor	(%r11,%rdi,4),	%r8d;\
+	movzx	a ## H,		%edi;\
+	xor	s1(%r11,%rdi,4),%r9d;\
+	add	%r8d,		%r9d;\
+	add	%r9d,		%r8d;\
+	add	k+round(%r11),	%r9d;\
+	xor	%r9d,		c ## D;\
+	rol	$15,		c ## D;\
+	add	k+4+round(%r11),%r8d;\
+	xor	%r8d,		d ## D;
+
+/*
+ * a input register containing a(rotated 16)
+ * b input register containing b
+ * c input register containing c
+ * d input register containing d (already rol $1)
+ * operations on a and b are interleaved to increase performance
+ * during the round a and b are prepared for the output whitening
+ */
+#define encrypt_last_round(a,b,c,d,round)\
+	mov	b ## D,		%r10d;\
+	shl	$32,		%r10;\
+	movzx	b ## B,		%edi;\
+	mov	s1(%r11,%rdi,4),%r8d;\
+	movzx	a ## B,		%edi;\
+	mov	s2(%r11,%rdi,4),%r9d;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	s2(%r11,%rdi,4),%r8d;\
+	movzx	a ## H,		%edi;\
+	ror	$16,		a ## D;\
+	xor	s3(%r11,%rdi,4),%r9d;\
+	movzx	b ## B,		%edi;\
+	xor	s3(%r11,%rdi,4),%r8d;\
+	movzx	a ## B,		%edi;\
+	xor	(%r11,%rdi,4),	%r9d;\
+	xor	a,		%r10;\
+	movzx	b ## H,		%edi;\
+	xor	(%r11,%rdi,4),	%r8d;\
+	movzx	a ## H,		%edi;\
+	xor	s1(%r11,%rdi,4),%r9d;\
+	add	%r8d,		%r9d;\
+	add	%r9d,		%r8d;\
+	add	k+round(%r11),	%r9d;\
+	xor	%r9d,		c ## D;\
+	ror	$1,		c ## D;\
+	add	k+4+round(%r11),%r8d;\
+	xor	%r8d,		d ## D
+
+/*
+ * a input register containing a
+ * b input register containing b (rotated 16)
+ * c input register containing c (already rol $1)
+ * d input register containing d
+ * operations on a and b are interleaved to increase performance
+ */
+#define decrypt_round(a,b,c,d,round)\
+	movzx	a ## B,		%edi;\
+	mov	(%r11,%rdi,4),	%r9d;\
+	movzx	b ## B,		%edi;\
+	mov	s3(%r11,%rdi,4),%r8d;\
+	movzx	a ## H,		%edi;\
+	ror	$16,		a ## D;\
+	xor	s1(%r11,%rdi,4),%r9d;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	(%r11,%rdi,4),	%r8d;\
+	movzx	a ## B,		%edi;\
+	xor	s2(%r11,%rdi,4),%r9d;\
+	movzx	b ## B,		%edi;\
+	xor	s1(%r11,%rdi,4),%r8d;\
+	movzx	a ## H,		%edi;\
+	ror	$15,		a ## D;\
+	xor	s3(%r11,%rdi,4),%r9d;\
+	movzx	b ## H,		%edi;\
+	xor	s2(%r11,%rdi,4),%r8d;\
+	add	%r8d,		%r9d;\
+	add	%r9d,		%r8d;\
+	add	k+round(%r11),	%r9d;\
+	xor	%r9d,		c ## D;\
+	add	k+4+round(%r11),%r8d;\
+	xor	%r8d,		d ## D;\
+	rol	$15,		d ## D;
+
+/*
+ * a input register containing a
+ * b input register containing b
+ * c input register containing c (already rol $1)
+ * d input register containing d
+ * operations on a and b are interleaved to increase performance
+ * during the round a and b are prepared for the output whitening
+ */
+#define decrypt_last_round(a,b,c,d,round)\
+	movzx	a ## B,		%edi;\
+	mov	(%r11,%rdi,4),	%r9d;\
+	movzx	b ## B,		%edi;\
+	mov	s3(%r11,%rdi,4),%r8d;\
+	movzx	b ## H,		%edi;\
+	ror	$16,		b ## D;\
+	xor	(%r11,%rdi,4),	%r8d;\
+	movzx	a ## H,		%edi;\
+	mov	b ## D,		%r10d;\
+	shl	$32,		%r10;\
+	xor	a,		%r10;\
+	ror	$16,		a ## D;\
+	xor	s1(%r11,%rdi,4),%r9d;\
+	movzx	b ## B,		%edi;\
+	xor	s1(%r11,%rdi,4),%r8d;\
+	movzx	a ## B,		%edi;\
+	xor	s2(%r11,%rdi,4),%r9d;\
+	movzx	b ## H,		%edi;\
+	xor	s2(%r11,%rdi,4),%r8d;\
+	movzx	a ## H,		%edi;\
+	xor	s3(%r11,%rdi,4),%r9d;\
+	add	%r8d,		%r9d;\
+	add	%r9d,		%r8d;\
+	add	k+round(%r11),	%r9d;\
+	xor	%r9d,		c ## D;\
+	add	k+4+round(%r11),%r8d;\
+	xor	%r8d,		d ## D;\
+	ror	$1,		d ## D;
+
+SYM_TYPED_FUNC_START(twofish_enc_blk)
+	pushq    R1
+
+	/* %rdi contains the ctx address */
+	/* %rsi contains the output address */
+	/* %rdx contains the input address */
+	/* ctx address is moved to free one non-rex register
+	as target for the 8bit high operations */
+	mov	%rdi,		%r11
+
+	movq	(R3),	R1
+	movq	8(R3),	R3
+	input_whitening(R1,%r11,a_offset)
+	input_whitening(R3,%r11,c_offset)
+	mov	R1D,	R0D
+	rol	$16,	R0D
+	shr	$32,	R1
+	mov	R3D,	R2D
+	shr	$32,	R3
+	rol	$1,	R3D
+
+	encrypt_round(R0,R1,R2,R3,0);
+	encrypt_round(R2,R3,R0,R1,8);
+	encrypt_round(R0,R1,R2,R3,2*8);
+	encrypt_round(R2,R3,R0,R1,3*8);
+	encrypt_round(R0,R1,R2,R3,4*8);
+	encrypt_round(R2,R3,R0,R1,5*8);
+	encrypt_round(R0,R1,R2,R3,6*8);
+	encrypt_round(R2,R3,R0,R1,7*8);
+	encrypt_round(R0,R1,R2,R3,8*8);
+	encrypt_round(R2,R3,R0,R1,9*8);
+	encrypt_round(R0,R1,R2,R3,10*8);
+	encrypt_round(R2,R3,R0,R1,11*8);
+	encrypt_round(R0,R1,R2,R3,12*8);
+	encrypt_round(R2,R3,R0,R1,13*8);
+	encrypt_round(R0,R1,R2,R3,14*8);
+	encrypt_last_round(R2,R3,R0,R1,15*8);
+
+
+	output_whitening(%r10,%r11,a_offset)
+	movq	%r10,	(%rsi)
+
+	shl	$32,	R1
+	xor	R0,	R1
+
+	output_whitening(R1,%r11,c_offset)
+	movq	R1,	8(%rsi)
+
+	popq	R1
+	movl	$1,%eax
+	RET
+SYM_FUNC_END(twofish_enc_blk)
+
+SYM_TYPED_FUNC_START(twofish_dec_blk)
+	pushq    R1
+
+	/* %rdi contains the ctx address */
+	/* %rsi contains the output address */
+	/* %rdx contains the input address */
+	/* ctx address is moved to free one non-rex register
+	as target for the 8bit high operations */
+	mov	%rdi,		%r11
+
+	movq	(R3),	R1
+	movq	8(R3),	R3
+	output_whitening(R1,%r11,a_offset)
+	output_whitening(R3,%r11,c_offset)
+	mov	R1D,	R0D
+	shr	$32,	R1
+	rol	$16,	R1D
+	mov	R3D,	R2D
+	shr	$32,	R3
+	rol	$1,	R2D
+
+	decrypt_round(R0,R1,R2,R3,15*8);
+	decrypt_round(R2,R3,R0,R1,14*8);
+	decrypt_round(R0,R1,R2,R3,13*8);
+	decrypt_round(R2,R3,R0,R1,12*8);
+	decrypt_round(R0,R1,R2,R3,11*8);
+	decrypt_round(R2,R3,R0,R1,10*8);
+	decrypt_round(R0,R1,R2,R3,9*8);
+	decrypt_round(R2,R3,R0,R1,8*8);
+	decrypt_round(R0,R1,R2,R3,7*8);
+	decrypt_round(R2,R3,R0,R1,6*8);
+	decrypt_round(R0,R1,R2,R3,5*8);
+	decrypt_round(R2,R3,R0,R1,4*8);
+	decrypt_round(R0,R1,R2,R3,3*8);
+	decrypt_round(R2,R3,R0,R1,2*8);
+	decrypt_round(R0,R1,R2,R3,1*8);
+	decrypt_last_round(R2,R3,R0,R1,0);
+
+	input_whitening(%r10,%r11,a_offset)
+	movq	%r10,	(%rsi)
+
+	shl	$32,	R1
+	xor	R0,	R1
+
+	input_whitening(R1,%r11,c_offset)
+	movq	R1,	8(%rsi)
+
+	popq	R1
+	movl	$1,%eax
+	RET
+SYM_FUNC_END(twofish_dec_blk)
diff --git a/arch/x86/crypto/twofish.h b/arch/x86/crypto/twofish.h
new file mode 100644
index 000000000..12df400e6
--- /dev/null
+++ b/arch/x86/crypto/twofish.h
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef ASM_X86_TWOFISH_H
+#define ASM_X86_TWOFISH_H
+
+#include <linux/crypto.h>
+#include <crypto/twofish.h>
+#include <crypto/b128ops.h>
+
+/* regular block cipher functions from twofish_x86_64 module */
+asmlinkage void twofish_enc_blk(const void *ctx, u8 *dst, const u8 *src);
+asmlinkage void twofish_dec_blk(const void *ctx, u8 *dst, const u8 *src);
+
+/* 3-way parallel cipher functions */
+asmlinkage void __twofish_enc_blk_3way(const void *ctx, u8 *dst, const u8 *src,
+				       bool xor);
+asmlinkage void twofish_dec_blk_3way(const void *ctx, u8 *dst, const u8 *src);
+
+/* helpers from twofish_x86_64-3way module */
+extern void twofish_dec_blk_cbc_3way(const void *ctx, u8 *dst, const u8 *src);
+
+#endif /* ASM_X86_TWOFISH_H */
diff --git a/arch/x86/crypto/twofish_avx_glue.c b/arch/x86/crypto/twofish_avx_glue.c
new file mode 100644
index 000000000..9e20db013
--- /dev/null
+++ b/arch/x86/crypto/twofish_avx_glue.c
@@ -0,0 +1,126 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Glue Code for AVX assembler version of Twofish Cipher
+ *
+ * Copyright (C) 2012 Johannes Goetzfried
+ *     <Johannes.Goetzfried@informatik.stud.uni-erlangen.de>
+ *
+ * Copyright © 2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
+ */
+
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/crypto.h>
+#include <linux/err.h>
+#include <crypto/algapi.h>
+#include <crypto/twofish.h>
+
+#include "twofish.h"
+#include "ecb_cbc_helpers.h"
+
+#define TWOFISH_PARALLEL_BLOCKS 8
+
+/* 8-way parallel cipher functions */
+asmlinkage void twofish_ecb_enc_8way(const void *ctx, u8 *dst, const u8 *src);
+asmlinkage void twofish_ecb_dec_8way(const void *ctx, u8 *dst, const u8 *src);
+
+asmlinkage void twofish_cbc_dec_8way(const void *ctx, u8 *dst, const u8 *src);
+
+static int twofish_setkey_skcipher(struct crypto_skcipher *tfm,
+				   const u8 *key, unsigned int keylen)
+{
+	return twofish_setkey(&tfm->base, key, keylen);
+}
+
+static inline void twofish_enc_blk_3way(const void *ctx, u8 *dst, const u8 *src)
+{
+	__twofish_enc_blk_3way(ctx, dst, src, false);
+}
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, TF_BLOCK_SIZE, TWOFISH_PARALLEL_BLOCKS);
+	ECB_BLOCK(TWOFISH_PARALLEL_BLOCKS, twofish_ecb_enc_8way);
+	ECB_BLOCK(3, twofish_enc_blk_3way);
+	ECB_BLOCK(1, twofish_enc_blk);
+	ECB_WALK_END();
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, TF_BLOCK_SIZE, TWOFISH_PARALLEL_BLOCKS);
+	ECB_BLOCK(TWOFISH_PARALLEL_BLOCKS, twofish_ecb_dec_8way);
+	ECB_BLOCK(3, twofish_dec_blk_3way);
+	ECB_BLOCK(1, twofish_dec_blk);
+	ECB_WALK_END();
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, TF_BLOCK_SIZE, -1);
+	CBC_ENC_BLOCK(twofish_enc_blk);
+	CBC_WALK_END();
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, TF_BLOCK_SIZE, TWOFISH_PARALLEL_BLOCKS);
+	CBC_DEC_BLOCK(TWOFISH_PARALLEL_BLOCKS, twofish_cbc_dec_8way);
+	CBC_DEC_BLOCK(3, twofish_dec_blk_cbc_3way);
+	CBC_DEC_BLOCK(1, twofish_dec_blk);
+	CBC_WALK_END();
+}
+
+static struct skcipher_alg twofish_algs[] = {
+	{
+		.base.cra_name		= "ecb(twofish)",
+		.base.cra_driver_name	= "ecb-twofish-avx",
+		.base.cra_priority	= 400,
+		.base.cra_blocksize	= TF_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct twofish_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= TF_MIN_KEY_SIZE,
+		.max_keysize		= TF_MAX_KEY_SIZE,
+		.setkey			= twofish_setkey_skcipher,
+		.encrypt		= ecb_encrypt,
+		.decrypt		= ecb_decrypt,
+	}, {
+		.base.cra_name		= "cbc(twofish)",
+		.base.cra_driver_name	= "cbc-twofish-avx",
+		.base.cra_priority	= 400,
+		.base.cra_blocksize	= TF_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct twofish_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= TF_MIN_KEY_SIZE,
+		.max_keysize		= TF_MAX_KEY_SIZE,
+		.ivsize			= TF_BLOCK_SIZE,
+		.setkey			= twofish_setkey_skcipher,
+		.encrypt		= cbc_encrypt,
+		.decrypt		= cbc_decrypt,
+	},
+};
+
+static int __init twofish_init(void)
+{
+	const char *feature_name;
+
+	if (!cpu_has_xfeatures(XFEATURE_MASK_SSE | XFEATURE_MASK_YMM, &feature_name)) {
+		pr_info("CPU feature '%s' is not supported.\n", feature_name);
+		return -ENODEV;
+	}
+
+	return crypto_register_skciphers(twofish_algs,
+					 ARRAY_SIZE(twofish_algs));
+}
+
+static void __exit twofish_exit(void)
+{
+	crypto_unregister_skciphers(twofish_algs, ARRAY_SIZE(twofish_algs));
+}
+
+module_init(twofish_init);
+module_exit(twofish_exit);
+
+MODULE_DESCRIPTION("Twofish Cipher Algorithm, AVX optimized");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_CRYPTO("twofish");
diff --git a/arch/x86/crypto/twofish_glue.c b/arch/x86/crypto/twofish_glue.c
new file mode 100644
index 000000000..8e9906d36
--- /dev/null
+++ b/arch/x86/crypto/twofish_glue.c
@@ -0,0 +1,100 @@
+/*
+ * Glue Code for assembler optimized version of TWOFISH
+ *
+ * Originally Twofish for GPG
+ * By Matthew Skala <mskala@ansuz.sooke.bc.ca>, July 26, 1998
+ * 256-bit key length added March 20, 1999
+ * Some modifications to reduce the text size by Werner Koch, April, 1998
+ * Ported to the kerneli patch by Marc Mutz <Marc@Mutz.com>
+ * Ported to CryptoAPI by Colin Slater <hoho@tacomeat.net>
+ *
+ * The original author has disclaimed all copyright interest in this
+ * code and thus put it in the public domain. The subsequent authors
+ * have put this under the GNU General Public License.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307
+ * USA
+ *
+ * This code is a "clean room" implementation, written from the paper
+ * _Twofish: A 128-Bit Block Cipher_ by Bruce Schneier, John Kelsey,
+ * Doug Whiting, David Wagner, Chris Hall, and Niels Ferguson, available
+ * through http://www.counterpane.com/twofish.html
+ *
+ * For background information on multiplication in finite fields, used for
+ * the matrix operations in the key schedule, see the book _Contemporary
+ * Abstract Algebra_ by Joseph A. Gallian, especially chapter 22 in the
+ * Third Edition.
+ */
+
+#include <crypto/algapi.h>
+#include <crypto/twofish.h>
+#include <linux/export.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/types.h>
+
+asmlinkage void twofish_enc_blk(struct twofish_ctx *ctx, u8 *dst,
+				const u8 *src);
+EXPORT_SYMBOL_GPL(twofish_enc_blk);
+asmlinkage void twofish_dec_blk(struct twofish_ctx *ctx, u8 *dst,
+				const u8 *src);
+EXPORT_SYMBOL_GPL(twofish_dec_blk);
+
+static void twofish_encrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
+{
+	twofish_enc_blk(crypto_tfm_ctx(tfm), dst, src);
+}
+
+static void twofish_decrypt(struct crypto_tfm *tfm, u8 *dst, const u8 *src)
+{
+	twofish_dec_blk(crypto_tfm_ctx(tfm), dst, src);
+}
+
+static struct crypto_alg alg = {
+	.cra_name		=	"twofish",
+	.cra_driver_name	=	"twofish-asm",
+	.cra_priority		=	200,
+	.cra_flags		=	CRYPTO_ALG_TYPE_CIPHER,
+	.cra_blocksize		=	TF_BLOCK_SIZE,
+	.cra_ctxsize		=	sizeof(struct twofish_ctx),
+	.cra_module		=	THIS_MODULE,
+	.cra_u			=	{
+		.cipher = {
+			.cia_min_keysize	=	TF_MIN_KEY_SIZE,
+			.cia_max_keysize	=	TF_MAX_KEY_SIZE,
+			.cia_setkey		=	twofish_setkey,
+			.cia_encrypt		=	twofish_encrypt,
+			.cia_decrypt		=	twofish_decrypt
+		}
+	}
+};
+
+static int __init twofish_glue_init(void)
+{
+	return crypto_register_alg(&alg);
+}
+
+static void __exit twofish_glue_fini(void)
+{
+	crypto_unregister_alg(&alg);
+}
+
+module_init(twofish_glue_init);
+module_exit(twofish_glue_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION ("Twofish Cipher Algorithm, asm optimized");
+MODULE_ALIAS_CRYPTO("twofish");
+MODULE_ALIAS_CRYPTO("twofish-asm");
diff --git a/arch/x86/crypto/twofish_glue_3way.c b/arch/x86/crypto/twofish_glue_3way.c
new file mode 100644
index 000000000..8ad77725b
--- /dev/null
+++ b/arch/x86/crypto/twofish_glue_3way.c
@@ -0,0 +1,170 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Glue Code for 3-way parallel assembler optimized version of Twofish
+ *
+ * Copyright (c) 2011 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ */
+
+#include <asm/cpu_device_id.h>
+#include <crypto/algapi.h>
+#include <crypto/twofish.h>
+#include <linux/crypto.h>
+#include <linux/export.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/types.h>
+
+#include "twofish.h"
+#include "ecb_cbc_helpers.h"
+
+EXPORT_SYMBOL_GPL(__twofish_enc_blk_3way);
+EXPORT_SYMBOL_GPL(twofish_dec_blk_3way);
+
+static int twofish_setkey_skcipher(struct crypto_skcipher *tfm,
+				   const u8 *key, unsigned int keylen)
+{
+	return twofish_setkey(&tfm->base, key, keylen);
+}
+
+static inline void twofish_enc_blk_3way(const void *ctx, u8 *dst, const u8 *src)
+{
+	__twofish_enc_blk_3way(ctx, dst, src, false);
+}
+
+void twofish_dec_blk_cbc_3way(const void *ctx, u8 *dst, const u8 *src)
+{
+	u8 buf[2][TF_BLOCK_SIZE];
+	const u8 *s = src;
+
+	if (dst == src)
+		s = memcpy(buf, src, sizeof(buf));
+	twofish_dec_blk_3way(ctx, dst, src);
+	crypto_xor(dst + TF_BLOCK_SIZE, s, sizeof(buf));
+
+}
+EXPORT_SYMBOL_GPL(twofish_dec_blk_cbc_3way);
+
+static int ecb_encrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, TF_BLOCK_SIZE, -1);
+	ECB_BLOCK(3, twofish_enc_blk_3way);
+	ECB_BLOCK(1, twofish_enc_blk);
+	ECB_WALK_END();
+}
+
+static int ecb_decrypt(struct skcipher_request *req)
+{
+	ECB_WALK_START(req, TF_BLOCK_SIZE, -1);
+	ECB_BLOCK(3, twofish_dec_blk_3way);
+	ECB_BLOCK(1, twofish_dec_blk);
+	ECB_WALK_END();
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, TF_BLOCK_SIZE, -1);
+	CBC_ENC_BLOCK(twofish_enc_blk);
+	CBC_WALK_END();
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	CBC_WALK_START(req, TF_BLOCK_SIZE, -1);
+	CBC_DEC_BLOCK(3, twofish_dec_blk_cbc_3way);
+	CBC_DEC_BLOCK(1, twofish_dec_blk);
+	CBC_WALK_END();
+}
+
+static struct skcipher_alg tf_skciphers[] = {
+	{
+		.base.cra_name		= "ecb(twofish)",
+		.base.cra_driver_name	= "ecb-twofish-3way",
+		.base.cra_priority	= 300,
+		.base.cra_blocksize	= TF_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct twofish_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= TF_MIN_KEY_SIZE,
+		.max_keysize		= TF_MAX_KEY_SIZE,
+		.setkey			= twofish_setkey_skcipher,
+		.encrypt		= ecb_encrypt,
+		.decrypt		= ecb_decrypt,
+	}, {
+		.base.cra_name		= "cbc(twofish)",
+		.base.cra_driver_name	= "cbc-twofish-3way",
+		.base.cra_priority	= 300,
+		.base.cra_blocksize	= TF_BLOCK_SIZE,
+		.base.cra_ctxsize	= sizeof(struct twofish_ctx),
+		.base.cra_module	= THIS_MODULE,
+		.min_keysize		= TF_MIN_KEY_SIZE,
+		.max_keysize		= TF_MAX_KEY_SIZE,
+		.ivsize			= TF_BLOCK_SIZE,
+		.setkey			= twofish_setkey_skcipher,
+		.encrypt		= cbc_encrypt,
+		.decrypt		= cbc_decrypt,
+	},
+};
+
+static bool is_blacklisted_cpu(void)
+{
+	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
+		return false;
+
+	switch (boot_cpu_data.x86_vfm) {
+	case INTEL_ATOM_BONNELL:
+	case INTEL_ATOM_BONNELL_MID:
+	case INTEL_ATOM_SALTWELL:
+		/*
+		 * On Atom, twofish-3way is slower than original assembler
+		 * implementation. Twofish-3way trades off some performance in
+		 * storing blocks in 64bit registers to allow three blocks to
+		 * be processed parallel. Parallel operation then allows gaining
+		 * more performance than was trade off, on out-of-order CPUs.
+		 * However Atom does not benefit from this parallelism and
+		 * should be blacklisted.
+		 */
+		return true;
+	}
+
+	if (boot_cpu_data.x86 == 0x0f) {
+		/*
+		 * On Pentium 4, twofish-3way is slower than original assembler
+		 * implementation because excessive uses of 64bit rotate and
+		 * left-shifts (which are really slow on P4) needed to store and
+		 * handle 128bit block in two 64bit registers.
+		 */
+		return true;
+	}
+
+	return false;
+}
+
+static int force;
+module_param(force, int, 0);
+MODULE_PARM_DESC(force, "Force module load, ignore CPU blacklist");
+
+static int __init twofish_3way_init(void)
+{
+	if (!force && is_blacklisted_cpu()) {
+		printk(KERN_INFO
+			"twofish-x86_64-3way: performance on this CPU "
+			"would be suboptimal: disabling "
+			"twofish-x86_64-3way.\n");
+		return -ENODEV;
+	}
+
+	return crypto_register_skciphers(tf_skciphers,
+					 ARRAY_SIZE(tf_skciphers));
+}
+
+static void __exit twofish_3way_fini(void)
+{
+	crypto_unregister_skciphers(tf_skciphers, ARRAY_SIZE(tf_skciphers));
+}
+
+module_init(twofish_3way_init);
+module_exit(twofish_3way_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("Twofish Cipher Algorithm, 3-way parallel asm optimized");
+MODULE_ALIAS_CRYPTO("twofish");
+MODULE_ALIAS_CRYPTO("twofish-asm");