x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

When using eBPF to probe CS CALL instructions within a function,
a crash can be triggered.

The eBPF tool probes offset 257 of the __hrtimer_run_queues()
function:

<__hrtimer_run_queues+249>:  nopl   0x0(%rax,%rax,1)
<__hrtimer_run_queues+254>:  mov    %r14,%rdi
<__hrtimer_run_queues+257>:  cs call <__x86_indirect_thunk_r12>
<__hrtimer_run_queues+263>:  mov    %eax,%r12d
<__hrtimer_run_queues+266>:  xchg   %ax,%ax
<__hrtimer_run_queues+268>:  mov    %r13,%rdi

Which triggers this crash:

  BUG: unable to handle page fault for address: 00000000000f41c9
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] SMP NOPTI
  CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P
  RIP: 0010:__hrtimer_run_queues+0x106/0x230

Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is
at the 6th byte of the above CS CALL instruction. Since the CS CALL
instruction occupies 6 bytes, the exception occurred in the middle of that
call instruction.

The root cause is that when using eBPF tools to probe in the middle of a
function, a kprobe with INT3 is used as the underlying implementation.

During single-step emulation of the original CALL instruction,
int3_emulate_call() assumes that the probed CALL instruction is 5 bytes
long. However, the actual CS-prefixed CALL instruction occupies 6 bytes,
so it constructs an incorrect exception return address. When the CPU
returns from the kprobe handler, the next instruction to be executed is at
the address of the last byte of that CS CALL instruction. Coincidentally,
starting from that address, the CPU fetches and decodes a completely
different instruction, which ultimately triggers a kernel crash.

Fix the issue by using the actual instruction length obtained from
the instruction decoder when constructing the exception return
address, rather than relying on the hardcoded CALL_INSN_SIZE macro.

[ mingo: Refined the changelog ]

Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

6 files changed, 2694 insertions, 0 deletions

diff --git a/arch/sparc/net/Makefile b/arch/sparc/net/Makefile
new file mode 100644
index 000000000..806267de3
--- /dev/null
+++ b/arch/sparc/net/Makefile
@@ -0,0 +1,8 @@
+# SPDX-License-Identifier: GPL-2.0-only
+#
+# Arch-specific network modules
+#
+obj-$(CONFIG_BPF_JIT) += bpf_jit_comp_$(BITS).o
+ifeq ($(BITS),32)
+obj-$(CONFIG_BPF_JIT) += bpf_jit_asm_32.o
+endif
diff --git a/arch/sparc/net/bpf_jit_32.h b/arch/sparc/net/bpf_jit_32.h
new file mode 100644
index 000000000..cfd6a8be0
--- /dev/null
+++ b/arch/sparc/net/bpf_jit_32.h
@@ -0,0 +1,69 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BPF_JIT_H
+#define _BPF_JIT_H
+
+/* Conventions:
+ *  %g1 : temporary
+ *  %g2 : Secondary temporary used by SKB data helper stubs.
+ *  %g3 : packet offset passed into SKB data helper stubs.
+ *  %o0 : pointer to skb (first argument given to JIT function)
+ *  %o1 : BPF A accumulator
+ *  %o2 : BPF X accumulator
+ *  %o3 : Holds saved %o7 so we can call helper functions without needing
+ *        to allocate a register window.
+ *  %o4 : skb->len - skb->data_len
+ *  %o5 : skb->data
+ */
+
+#ifndef __ASSEMBLER__
+#define G0		0x00
+#define G1		0x01
+#define G3		0x03
+#define G6		0x06
+#define O0		0x08
+#define O1		0x09
+#define O2		0x0a
+#define O3		0x0b
+#define O4		0x0c
+#define O5		0x0d
+#define SP		0x0e
+#define O7		0x0f
+#define FP		0x1e
+
+#define r_SKB		O0
+#define r_A		O1
+#define r_X		O2
+#define r_saved_O7	O3
+#define r_HEADLEN	O4
+#define r_SKB_DATA	O5
+#define r_TMP		G1
+#define r_TMP2		G2
+#define r_OFF		G3
+
+/* assembly code in arch/sparc/net/bpf_jit_asm_32.S */
+extern u32 bpf_jit_load_word[];
+extern u32 bpf_jit_load_half[];
+extern u32 bpf_jit_load_byte[];
+extern u32 bpf_jit_load_byte_msh[];
+extern u32 bpf_jit_load_word_positive_offset[];
+extern u32 bpf_jit_load_half_positive_offset[];
+extern u32 bpf_jit_load_byte_positive_offset[];
+extern u32 bpf_jit_load_byte_msh_positive_offset[];
+extern u32 bpf_jit_load_word_negative_offset[];
+extern u32 bpf_jit_load_half_negative_offset[];
+extern u32 bpf_jit_load_byte_negative_offset[];
+extern u32 bpf_jit_load_byte_msh_negative_offset[];
+
+#else
+#define r_SKB		%o0
+#define r_A		%o1
+#define r_X		%o2
+#define r_saved_O7	%o3
+#define r_HEADLEN	%o4
+#define r_SKB_DATA	%o5
+#define r_TMP		%g1
+#define r_TMP2		%g2
+#define r_OFF		%g3
+#endif
+
+#endif /* _BPF_JIT_H */
diff --git a/arch/sparc/net/bpf_jit_64.h b/arch/sparc/net/bpf_jit_64.h
new file mode 100644
index 000000000..fbc836f1c
--- /dev/null
+++ b/arch/sparc/net/bpf_jit_64.h
@@ -0,0 +1,38 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BPF_JIT_H
+#define _BPF_JIT_H
+
+#ifndef __ASSEMBLER__
+#define G0		0x00
+#define G1		0x01
+#define G2		0x02
+#define G3		0x03
+#define G6		0x06
+#define G7		0x07
+#define O0		0x08
+#define O1		0x09
+#define O2		0x0a
+#define O3		0x0b
+#define O4		0x0c
+#define O5		0x0d
+#define SP		0x0e
+#define O7		0x0f
+#define L0		0x10
+#define L1		0x11
+#define L2		0x12
+#define L3		0x13
+#define L4		0x14
+#define L5		0x15
+#define L6		0x16
+#define L7		0x17
+#define I0		0x18
+#define I1		0x19
+#define I2		0x1a
+#define I3		0x1b
+#define I4		0x1c
+#define I5		0x1d
+#define FP		0x1e
+#define I7		0x1f
+#endif
+
+#endif /* _BPF_JIT_H */
diff --git a/arch/sparc/net/bpf_jit_asm_32.S b/arch/sparc/net/bpf_jit_asm_32.S
new file mode 100644
index 000000000..a2e28e046
--- /dev/null
+++ b/arch/sparc/net/bpf_jit_asm_32.S
@@ -0,0 +1,202 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#include <asm/ptrace.h>
+
+#include "bpf_jit_32.h"
+
+#define SAVE_SZ		96
+#define SCRATCH_OFF	72
+#define BE_PTR(label)	be label
+#define SIGN_EXTEND(reg)
+
+#define SKF_MAX_NEG_OFF	(-0x200000) /* SKF_LL_OFF from filter.h */
+
+	.text
+	.globl	bpf_jit_load_word
+bpf_jit_load_word:
+	cmp	r_OFF, 0
+	bl	bpf_slow_path_word_neg
+	 nop
+	.globl	bpf_jit_load_word_positive_offset
+bpf_jit_load_word_positive_offset:
+	sub	r_HEADLEN, r_OFF, r_TMP
+	cmp	r_TMP, 3
+	ble	bpf_slow_path_word
+	 add	r_SKB_DATA, r_OFF, r_TMP
+	andcc	r_TMP, 3, %g0
+	bne	load_word_unaligned
+	 nop
+	retl
+	 ld	[r_TMP], r_A
+load_word_unaligned:
+	ldub	[r_TMP + 0x0], r_OFF
+	ldub	[r_TMP + 0x1], r_TMP2
+	sll	r_OFF, 8, r_OFF
+	or	r_OFF, r_TMP2, r_OFF
+	ldub	[r_TMP + 0x2], r_TMP2
+	sll	r_OFF, 8, r_OFF
+	or	r_OFF, r_TMP2, r_OFF
+	ldub	[r_TMP + 0x3], r_TMP2
+	sll	r_OFF, 8, r_OFF
+	retl
+	 or	r_OFF, r_TMP2, r_A
+
+	.globl	bpf_jit_load_half
+bpf_jit_load_half:
+	cmp	r_OFF, 0
+	bl	bpf_slow_path_half_neg
+	 nop
+	.globl	bpf_jit_load_half_positive_offset
+bpf_jit_load_half_positive_offset:
+	sub	r_HEADLEN, r_OFF, r_TMP
+	cmp	r_TMP, 1
+	ble	bpf_slow_path_half
+	 add	r_SKB_DATA, r_OFF, r_TMP
+	andcc	r_TMP, 1, %g0
+	bne	load_half_unaligned
+	 nop
+	retl
+	 lduh	[r_TMP], r_A
+load_half_unaligned:
+	ldub	[r_TMP + 0x0], r_OFF
+	ldub	[r_TMP + 0x1], r_TMP2
+	sll	r_OFF, 8, r_OFF
+	retl
+	 or	r_OFF, r_TMP2, r_A
+
+	.globl	bpf_jit_load_byte
+bpf_jit_load_byte:
+	cmp	r_OFF, 0
+	bl	bpf_slow_path_byte_neg
+	 nop
+	.globl	bpf_jit_load_byte_positive_offset
+bpf_jit_load_byte_positive_offset:
+	cmp	r_OFF, r_HEADLEN
+	bge	bpf_slow_path_byte
+	 nop
+	retl
+	 ldub	[r_SKB_DATA + r_OFF], r_A
+
+	.globl	bpf_jit_load_byte_msh
+bpf_jit_load_byte_msh:
+	cmp	r_OFF, 0
+	bl	bpf_slow_path_byte_msh_neg
+	 nop
+	.globl	bpf_jit_load_byte_msh_positive_offset
+bpf_jit_load_byte_msh_positive_offset:
+	cmp	r_OFF, r_HEADLEN
+	bge	bpf_slow_path_byte_msh
+	 nop
+	ldub	[r_SKB_DATA + r_OFF], r_OFF
+	and	r_OFF, 0xf, r_OFF
+	retl
+	 sll	r_OFF, 2, r_X
+
+#define bpf_slow_path_common(LEN)	\
+	save	%sp, -SAVE_SZ, %sp;	\
+	mov	%i0, %o0;		\
+	mov	r_OFF, %o1;		\
+	add	%fp, SCRATCH_OFF, %o2;	\
+	call	skb_copy_bits;		\
+	 mov	(LEN), %o3;		\
+	cmp	%o0, 0;			\
+	restore;
+
+bpf_slow_path_word:
+	bpf_slow_path_common(4)
+	bl	bpf_error
+	 ld	[%sp + SCRATCH_OFF], r_A
+	retl
+	 nop
+bpf_slow_path_half:
+	bpf_slow_path_common(2)
+	bl	bpf_error
+	 lduh	[%sp + SCRATCH_OFF], r_A
+	retl
+	 nop
+bpf_slow_path_byte:
+	bpf_slow_path_common(1)
+	bl	bpf_error
+	 ldub	[%sp + SCRATCH_OFF], r_A
+	retl
+	 nop
+bpf_slow_path_byte_msh:
+	bpf_slow_path_common(1)
+	bl	bpf_error
+	 ldub	[%sp + SCRATCH_OFF], r_A
+	and	r_OFF, 0xf, r_OFF
+	retl
+	 sll	r_OFF, 2, r_X
+
+#define bpf_negative_common(LEN)			\
+	save	%sp, -SAVE_SZ, %sp;			\
+	mov	%i0, %o0;				\
+	mov	r_OFF, %o1;				\
+	SIGN_EXTEND(%o1);				\
+	call	bpf_internal_load_pointer_neg_helper;	\
+	 mov	(LEN), %o2;				\
+	mov	%o0, r_TMP;				\
+	cmp	%o0, 0;					\
+	BE_PTR(bpf_error);				\
+	 restore;
+
+bpf_slow_path_word_neg:
+	sethi	%hi(SKF_MAX_NEG_OFF), r_TMP
+	cmp	r_OFF, r_TMP
+	bl	bpf_error
+	 nop
+	.globl	bpf_jit_load_word_negative_offset
+bpf_jit_load_word_negative_offset:
+	bpf_negative_common(4)
+	andcc	r_TMP, 3, %g0
+	bne	load_word_unaligned
+	 nop
+	retl
+	 ld	[r_TMP], r_A
+
+bpf_slow_path_half_neg:
+	sethi	%hi(SKF_MAX_NEG_OFF), r_TMP
+	cmp	r_OFF, r_TMP
+	bl	bpf_error
+	 nop
+	.globl	bpf_jit_load_half_negative_offset
+bpf_jit_load_half_negative_offset:
+	bpf_negative_common(2)
+	andcc	r_TMP, 1, %g0
+	bne	load_half_unaligned
+	 nop
+	retl
+	 lduh	[r_TMP], r_A
+
+bpf_slow_path_byte_neg:
+	sethi	%hi(SKF_MAX_NEG_OFF), r_TMP
+	cmp	r_OFF, r_TMP
+	bl	bpf_error
+	 nop
+	.globl	bpf_jit_load_byte_negative_offset
+bpf_jit_load_byte_negative_offset:
+	bpf_negative_common(1)
+	retl
+	 ldub	[r_TMP], r_A
+
+bpf_slow_path_byte_msh_neg:
+	sethi	%hi(SKF_MAX_NEG_OFF), r_TMP
+	cmp	r_OFF, r_TMP
+	bl	bpf_error
+	 nop
+	.globl	bpf_jit_load_byte_msh_negative_offset
+bpf_jit_load_byte_msh_negative_offset:
+	bpf_negative_common(1)
+	ldub	[r_TMP], r_OFF
+	and	r_OFF, 0xf, r_OFF
+	retl
+	 sll	r_OFF, 2, r_X
+
+bpf_error:
+	/* Make the JIT program return zero.  The JIT epilogue
+	 * stores away the original %o7 into r_saved_O7.  The
+	 * normal leaf function return is to use "retl" which
+	 * would evalute to "jmpl %o7 + 8, %g0" but we want to
+	 * use the saved value thus the sequence you see here.
+	 */
+	jmpl	r_saved_O7 + 8, %g0
+	 clr	%o0
diff --git a/arch/sparc/net/bpf_jit_comp_32.c b/arch/sparc/net/bpf_jit_comp_32.c
new file mode 100644
index 000000000..bda2dbd3f
--- /dev/null
+++ b/arch/sparc/net/bpf_jit_comp_32.c
@@ -0,0 +1,764 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/workqueue.h>
+#include <linux/netdevice.h>
+#include <linux/filter.h>
+#include <linux/cache.h>
+#include <linux/if_vlan.h>
+#include <linux/execmem.h>
+
+#include <asm/cacheflush.h>
+#include <asm/ptrace.h>
+
+#include "bpf_jit_32.h"
+
+static inline bool is_simm13(unsigned int value)
+{
+	return value + 0x1000 < 0x2000;
+}
+
+#define SEEN_DATAREF 1 /* might call external helpers */
+#define SEEN_XREG    2 /* ebx is used */
+#define SEEN_MEM     4 /* use mem[] for temporary storage */
+
+#define S13(X)		((X) & 0x1fff)
+#define IMMED		0x00002000
+#define RD(X)		((X) << 25)
+#define RS1(X)		((X) << 14)
+#define RS2(X)		((X))
+#define OP(X)		((X) << 30)
+#define OP2(X)		((X) << 22)
+#define OP3(X)		((X) << 19)
+#define COND(X)		((X) << 25)
+#define F1(X)		OP(X)
+#define F2(X, Y)	(OP(X) | OP2(Y))
+#define F3(X, Y)	(OP(X) | OP3(Y))
+
+#define CONDN		COND(0x0)
+#define CONDE		COND(0x1)
+#define CONDLE		COND(0x2)
+#define CONDL		COND(0x3)
+#define CONDLEU		COND(0x4)
+#define CONDCS		COND(0x5)
+#define CONDNEG		COND(0x6)
+#define CONDVC		COND(0x7)
+#define CONDA		COND(0x8)
+#define CONDNE		COND(0x9)
+#define CONDG		COND(0xa)
+#define CONDGE		COND(0xb)
+#define CONDGU		COND(0xc)
+#define CONDCC		COND(0xd)
+#define CONDPOS		COND(0xe)
+#define CONDVS		COND(0xf)
+
+#define CONDGEU		CONDCC
+#define CONDLU		CONDCS
+
+#define WDISP22(X)	(((X) >> 2) & 0x3fffff)
+
+#define BA		(F2(0, 2) | CONDA)
+#define BGU		(F2(0, 2) | CONDGU)
+#define BLEU		(F2(0, 2) | CONDLEU)
+#define BGEU		(F2(0, 2) | CONDGEU)
+#define BLU		(F2(0, 2) | CONDLU)
+#define BE		(F2(0, 2) | CONDE)
+#define BNE		(F2(0, 2) | CONDNE)
+
+#define BE_PTR		BE
+
+#define SETHI(K, REG)	\
+	(F2(0, 0x4) | RD(REG) | (((K) >> 10) & 0x3fffff))
+#define OR_LO(K, REG)	\
+	(F3(2, 0x02) | IMMED | RS1(REG) | ((K) & 0x3ff) | RD(REG))
+
+#define ADD		F3(2, 0x00)
+#define AND		F3(2, 0x01)
+#define ANDCC		F3(2, 0x11)
+#define OR		F3(2, 0x02)
+#define XOR		F3(2, 0x03)
+#define SUB		F3(2, 0x04)
+#define SUBCC		F3(2, 0x14)
+#define MUL		F3(2, 0x0a)	/* umul */
+#define DIV		F3(2, 0x0e)	/* udiv */
+#define SLL		F3(2, 0x25)
+#define SRL		F3(2, 0x26)
+#define JMPL		F3(2, 0x38)
+#define CALL		F1(1)
+#define BR		F2(0, 0x01)
+#define RD_Y		F3(2, 0x28)
+#define WR_Y		F3(2, 0x30)
+
+#define LD32		F3(3, 0x00)
+#define LD8		F3(3, 0x01)
+#define LD16		F3(3, 0x02)
+#define LD64		F3(3, 0x0b)
+#define ST32		F3(3, 0x04)
+
+#define LDPTR		LD32
+#define BASE_STACKFRAME	96
+
+#define LD32I		(LD32 | IMMED)
+#define LD8I		(LD8 | IMMED)
+#define LD16I		(LD16 | IMMED)
+#define LD64I		(LD64 | IMMED)
+#define LDPTRI		(LDPTR | IMMED)
+#define ST32I		(ST32 | IMMED)
+
+#define emit_nop()		\
+do {				\
+	*prog++ = SETHI(0, G0);	\
+} while (0)
+
+#define emit_neg()					\
+do {	/* sub %g0, r_A, r_A */				\
+	*prog++ = SUB | RS1(G0) | RS2(r_A) | RD(r_A);	\
+} while (0)
+
+#define emit_reg_move(FROM, TO)				\
+do {	/* or %g0, FROM, TO */				\
+	*prog++ = OR | RS1(G0) | RS2(FROM) | RD(TO);	\
+} while (0)
+
+#define emit_clear(REG)					\
+do {	/* or %g0, %g0, REG */				\
+	*prog++ = OR | RS1(G0) | RS2(G0) | RD(REG);	\
+} while (0)
+
+#define emit_set_const(K, REG)					\
+do {	/* sethi %hi(K), REG */					\
+	*prog++ = SETHI(K, REG);				\
+	/* or REG, %lo(K), REG */				\
+	*prog++ = OR_LO(K, REG);				\
+} while (0)
+
+	/* Emit
+	 *
+	 *	OP	r_A, r_X, r_A
+	 */
+#define emit_alu_X(OPCODE)					\
+do {								\
+	seen |= SEEN_XREG;					\
+	*prog++ = OPCODE | RS1(r_A) | RS2(r_X) | RD(r_A);	\
+} while (0)
+
+	/* Emit either:
+	 *
+	 *	OP	r_A, K, r_A
+	 *
+	 * or
+	 *
+	 *	sethi	%hi(K), r_TMP
+	 *	or	r_TMP, %lo(K), r_TMP
+	 *	OP	r_A, r_TMP, r_A
+	 *
+	 * depending upon whether K fits in a signed 13-bit
+	 * immediate instruction field.  Emit nothing if K
+	 * is zero.
+	 */
+#define emit_alu_K(OPCODE, K)					\
+do {								\
+	if (K || OPCODE == AND || OPCODE == MUL) {		\
+		unsigned int _insn = OPCODE;			\
+		_insn |= RS1(r_A) | RD(r_A);			\
+		if (is_simm13(K)) {				\
+			*prog++ = _insn | IMMED | S13(K);	\
+		} else {					\
+			emit_set_const(K, r_TMP);		\
+			*prog++ = _insn | RS2(r_TMP);		\
+		}						\
+	}							\
+} while (0)
+
+#define emit_loadimm(K, DEST)						\
+do {									\
+	if (is_simm13(K)) {						\
+		/* or %g0, K, DEST */					\
+		*prog++ = OR | IMMED | RS1(G0) | S13(K) | RD(DEST);	\
+	} else {							\
+		emit_set_const(K, DEST);				\
+	}								\
+} while (0)
+
+#define emit_loadptr(BASE, STRUCT, FIELD, DEST)				\
+do {	unsigned int _off = offsetof(STRUCT, FIELD);			\
+	BUILD_BUG_ON(sizeof_field(STRUCT, FIELD) != sizeof(void *));	\
+	*prog++ = LDPTRI | RS1(BASE) | S13(_off) | RD(DEST);		\
+} while (0)
+
+#define emit_load32(BASE, STRUCT, FIELD, DEST)				\
+do {	unsigned int _off = offsetof(STRUCT, FIELD);			\
+	BUILD_BUG_ON(sizeof_field(STRUCT, FIELD) != sizeof(u32));	\
+	*prog++ = LD32I | RS1(BASE) | S13(_off) | RD(DEST);		\
+} while (0)
+
+#define emit_load16(BASE, STRUCT, FIELD, DEST)				\
+do {	unsigned int _off = offsetof(STRUCT, FIELD);			\
+	BUILD_BUG_ON(sizeof_field(STRUCT, FIELD) != sizeof(u16));	\
+	*prog++ = LD16I | RS1(BASE) | S13(_off) | RD(DEST);		\
+} while (0)
+
+#define __emit_load8(BASE, STRUCT, FIELD, DEST)				\
+do {	unsigned int _off = offsetof(STRUCT, FIELD);			\
+	*prog++ = LD8I | RS1(BASE) | S13(_off) | RD(DEST);		\
+} while (0)
+
+#define emit_load8(BASE, STRUCT, FIELD, DEST)				\
+do {	BUILD_BUG_ON(sizeof_field(STRUCT, FIELD) != sizeof(u8));	\
+	__emit_load8(BASE, STRUCT, FIELD, DEST);			\
+} while (0)
+
+#define BIAS (-4)
+
+#define emit_ldmem(OFF, DEST)						\
+do {	*prog++ = LD32I | RS1(SP) | S13(BIAS - (OFF)) | RD(DEST);	\
+} while (0)
+
+#define emit_stmem(OFF, SRC)						\
+do {	*prog++ = ST32I | RS1(SP) | S13(BIAS - (OFF)) | RD(SRC);	\
+} while (0)
+
+#ifdef CONFIG_SMP
+#define emit_load_cpu(REG)						\
+	emit_load32(G6, struct thread_info, cpu, REG)
+#else
+#define emit_load_cpu(REG)	emit_clear(REG)
+#endif
+
+#define emit_skb_loadptr(FIELD, DEST) \
+	emit_loadptr(r_SKB, struct sk_buff, FIELD, DEST)
+#define emit_skb_load32(FIELD, DEST) \
+	emit_load32(r_SKB, struct sk_buff, FIELD, DEST)
+#define emit_skb_load16(FIELD, DEST) \
+	emit_load16(r_SKB, struct sk_buff, FIELD, DEST)
+#define __emit_skb_load8(FIELD, DEST) \
+	__emit_load8(r_SKB, struct sk_buff, FIELD, DEST)
+#define emit_skb_load8(FIELD, DEST) \
+	emit_load8(r_SKB, struct sk_buff, FIELD, DEST)
+
+#define emit_jmpl(BASE, IMM_OFF, LREG) \
+	*prog++ = (JMPL | IMMED | RS1(BASE) | S13(IMM_OFF) | RD(LREG))
+
+#define emit_call(FUNC)					\
+do {	void *_here = image + addrs[i] - 8;		\
+	unsigned int _off = (void *)(FUNC) - _here;	\
+	*prog++ = CALL | (((_off) >> 2) & 0x3fffffff);	\
+	emit_nop();					\
+} while (0)
+
+#define emit_branch(BR_OPC, DEST)			\
+do {	unsigned int _here = addrs[i] - 8;		\
+	*prog++ = BR_OPC | WDISP22((DEST) - _here);	\
+} while (0)
+
+#define emit_branch_off(BR_OPC, OFF)			\
+do {	*prog++ = BR_OPC | WDISP22(OFF);		\
+} while (0)
+
+#define emit_jump(DEST)		emit_branch(BA, DEST)
+
+#define emit_read_y(REG)	*prog++ = RD_Y | RD(REG)
+#define emit_write_y(REG)	*prog++ = WR_Y | IMMED | RS1(REG) | S13(0)
+
+#define emit_cmp(R1, R2) \
+	*prog++ = (SUBCC | RS1(R1) | RS2(R2) | RD(G0))
+
+#define emit_cmpi(R1, IMM) \
+	*prog++ = (SUBCC | IMMED | RS1(R1) | S13(IMM) | RD(G0));
+
+#define emit_btst(R1, R2) \
+	*prog++ = (ANDCC | RS1(R1) | RS2(R2) | RD(G0))
+
+#define emit_btsti(R1, IMM) \
+	*prog++ = (ANDCC | IMMED | RS1(R1) | S13(IMM) | RD(G0));
+
+#define emit_sub(R1, R2, R3) \
+	*prog++ = (SUB | RS1(R1) | RS2(R2) | RD(R3))
+
+#define emit_subi(R1, IMM, R3) \
+	*prog++ = (SUB | IMMED | RS1(R1) | S13(IMM) | RD(R3))
+
+#define emit_add(R1, R2, R3) \
+	*prog++ = (ADD | RS1(R1) | RS2(R2) | RD(R3))
+
+#define emit_addi(R1, IMM, R3) \
+	*prog++ = (ADD | IMMED | RS1(R1) | S13(IMM) | RD(R3))
+
+#define emit_and(R1, R2, R3) \
+	*prog++ = (AND | RS1(R1) | RS2(R2) | RD(R3))
+
+#define emit_andi(R1, IMM, R3) \
+	*prog++ = (AND | IMMED | RS1(R1) | S13(IMM) | RD(R3))
+
+#define emit_alloc_stack(SZ) \
+	*prog++ = (SUB | IMMED | RS1(SP) | S13(SZ) | RD(SP))
+
+#define emit_release_stack(SZ) \
+	*prog++ = (ADD | IMMED | RS1(SP) | S13(SZ) | RD(SP))
+
+/* A note about branch offset calculations.  The addrs[] array,
+ * indexed by BPF instruction, records the address after all the
+ * sparc instructions emitted for that BPF instruction.
+ *
+ * The most common case is to emit a branch at the end of such
+ * a code sequence.  So this would be two instructions, the
+ * branch and its delay slot.
+ *
+ * Therefore by default the branch emitters calculate the branch
+ * offset field as:
+ *
+ *	destination - (addrs[i] - 8)
+ *
+ * This "addrs[i] - 8" is the address of the branch itself or
+ * what "." would be in assembler notation.  The "8" part is
+ * how we take into consideration the branch and its delay
+ * slot mentioned above.
+ *
+ * Sometimes we need to emit a branch earlier in the code
+ * sequence.  And in these situations we adjust "destination"
+ * to accommodate this difference.  For example, if we needed
+ * to emit a branch (and its delay slot) right before the
+ * final instruction emitted for a BPF opcode, we'd use
+ * "destination + 4" instead of just plain "destination" above.
+ *
+ * This is why you see all of these funny emit_branch() and
+ * emit_jump() calls with adjusted offsets.
+ */
+
+void bpf_jit_compile(struct bpf_prog *fp)
+{
+	unsigned int cleanup_addr, proglen, oldproglen = 0;
+	u32 temp[8], *prog, *func, seen = 0, pass;
+	const struct sock_filter *filter = fp->insns;
+	int i, flen = fp->len, pc_ret0 = -1;
+	unsigned int *addrs;
+	void *image;
+
+	if (!bpf_jit_enable)
+		return;
+
+	addrs = kmalloc_array(flen, sizeof(*addrs), GFP_KERNEL);
+	if (addrs == NULL)
+		return;
+
+	/* Before first pass, make a rough estimation of addrs[]
+	 * each bpf instruction is translated to less than 64 bytes
+	 */
+	for (proglen = 0, i = 0; i < flen; i++) {
+		proglen += 64;
+		addrs[i] = proglen;
+	}
+	cleanup_addr = proglen; /* epilogue address */
+	image = NULL;
+	for (pass = 0; pass < 10; pass++) {
+		u8 seen_or_pass0 = (pass == 0) ? (SEEN_XREG | SEEN_DATAREF | SEEN_MEM) : seen;
+
+		/* no prologue/epilogue for trivial filters (RET something) */
+		proglen = 0;
+		prog = temp;
+
+		/* Prologue */
+		if (seen_or_pass0) {
+			if (seen_or_pass0 & SEEN_MEM) {
+				unsigned int sz = BASE_STACKFRAME;
+				sz += BPF_MEMWORDS * sizeof(u32);
+				emit_alloc_stack(sz);
+			}
+
+			/* Make sure we dont leek kernel memory. */
+			if (seen_or_pass0 & SEEN_XREG)
+				emit_clear(r_X);
+
+			/* If this filter needs to access skb data,
+			 * load %o4 and %o5 with:
+			 *  %o4 = skb->len - skb->data_len
+			 *  %o5 = skb->data
+			 * And also back up %o7 into r_saved_O7 so we can
+			 * invoke the stubs using 'call'.
+			 */
+			if (seen_or_pass0 & SEEN_DATAREF) {
+				emit_load32(r_SKB, struct sk_buff, len, r_HEADLEN);
+				emit_load32(r_SKB, struct sk_buff, data_len, r_TMP);
+				emit_sub(r_HEADLEN, r_TMP, r_HEADLEN);
+				emit_loadptr(r_SKB, struct sk_buff, data, r_SKB_DATA);
+			}
+		}
+		emit_reg_move(O7, r_saved_O7);
+
+		/* Make sure we dont leak kernel information to the user. */
+		if (bpf_needs_clear_a(&filter[0]))
+			emit_clear(r_A); /* A = 0 */
+
+		for (i = 0; i < flen; i++) {
+			unsigned int K = filter[i].k;
+			unsigned int t_offset;
+			unsigned int f_offset;
+			u32 t_op, f_op;
+			u16 code = bpf_anc_helper(&filter[i]);
+			int ilen;
+
+			switch (code) {
+			case BPF_ALU | BPF_ADD | BPF_X:	/* A += X; */
+				emit_alu_X(ADD);
+				break;
+			case BPF_ALU | BPF_ADD | BPF_K:	/* A += K; */
+				emit_alu_K(ADD, K);
+				break;
+			case BPF_ALU | BPF_SUB | BPF_X:	/* A -= X; */
+				emit_alu_X(SUB);
+				break;
+			case BPF_ALU | BPF_SUB | BPF_K:	/* A -= K */
+				emit_alu_K(SUB, K);
+				break;
+			case BPF_ALU | BPF_AND | BPF_X:	/* A &= X */
+				emit_alu_X(AND);
+				break;
+			case BPF_ALU | BPF_AND | BPF_K:	/* A &= K */
+				emit_alu_K(AND, K);
+				break;
+			case BPF_ALU | BPF_OR | BPF_X:	/* A |= X */
+				emit_alu_X(OR);
+				break;
+			case BPF_ALU | BPF_OR | BPF_K:	/* A |= K */
+				emit_alu_K(OR, K);
+				break;
+			case BPF_ANC | SKF_AD_ALU_XOR_X: /* A ^= X; */
+			case BPF_ALU | BPF_XOR | BPF_X:
+				emit_alu_X(XOR);
+				break;
+			case BPF_ALU | BPF_XOR | BPF_K:	/* A ^= K */
+				emit_alu_K(XOR, K);
+				break;
+			case BPF_ALU | BPF_LSH | BPF_X:	/* A <<= X */
+				emit_alu_X(SLL);
+				break;
+			case BPF_ALU | BPF_LSH | BPF_K:	/* A <<= K */
+				emit_alu_K(SLL, K);
+				break;
+			case BPF_ALU | BPF_RSH | BPF_X:	/* A >>= X */
+				emit_alu_X(SRL);
+				break;
+			case BPF_ALU | BPF_RSH | BPF_K:	/* A >>= K */
+				emit_alu_K(SRL, K);
+				break;
+			case BPF_ALU | BPF_MUL | BPF_X:	/* A *= X; */
+				emit_alu_X(MUL);
+				break;
+			case BPF_ALU | BPF_MUL | BPF_K:	/* A *= K */
+				emit_alu_K(MUL, K);
+				break;
+			case BPF_ALU | BPF_DIV | BPF_K:	/* A /= K with K != 0*/
+				if (K == 1)
+					break;
+				emit_write_y(G0);
+				/* The Sparc v8 architecture requires
+				 * three instructions between a %y
+				 * register write and the first use.
+				 */
+				emit_nop();
+				emit_nop();
+				emit_nop();
+				emit_alu_K(DIV, K);
+				break;
+			case BPF_ALU | BPF_DIV | BPF_X:	/* A /= X; */
+				emit_cmpi(r_X, 0);
+				if (pc_ret0 > 0) {
+					t_offset = addrs[pc_ret0 - 1];
+					emit_branch(BE, t_offset + 20);
+					emit_nop(); /* delay slot */
+				} else {
+					emit_branch_off(BNE, 16);
+					emit_nop();
+					emit_jump(cleanup_addr + 20);
+					emit_clear(r_A);
+				}
+				emit_write_y(G0);
+				/* The Sparc v8 architecture requires
+				 * three instructions between a %y
+				 * register write and the first use.
+				 */
+				emit_nop();
+				emit_nop();
+				emit_nop();
+				emit_alu_X(DIV);
+				break;
+			case BPF_ALU | BPF_NEG:
+				emit_neg();
+				break;
+			case BPF_RET | BPF_K:
+				if (!K) {
+					if (pc_ret0 == -1)
+						pc_ret0 = i;
+					emit_clear(r_A);
+				} else {
+					emit_loadimm(K, r_A);
+				}
+				fallthrough;
+			case BPF_RET | BPF_A:
+				if (seen_or_pass0) {
+					if (i != flen - 1) {
+						emit_jump(cleanup_addr);
+						emit_nop();
+						break;
+					}
+					if (seen_or_pass0 & SEEN_MEM) {
+						unsigned int sz = BASE_STACKFRAME;
+						sz += BPF_MEMWORDS * sizeof(u32);
+						emit_release_stack(sz);
+					}
+				}
+				/* jmpl %r_saved_O7 + 8, %g0 */
+				emit_jmpl(r_saved_O7, 8, G0);
+				emit_reg_move(r_A, O0); /* delay slot */
+				break;
+			case BPF_MISC | BPF_TAX:
+				seen |= SEEN_XREG;
+				emit_reg_move(r_A, r_X);
+				break;
+			case BPF_MISC | BPF_TXA:
+				seen |= SEEN_XREG;
+				emit_reg_move(r_X, r_A);
+				break;
+			case BPF_ANC | SKF_AD_CPU:
+				emit_load_cpu(r_A);
+				break;
+			case BPF_ANC | SKF_AD_PROTOCOL:
+				emit_skb_load16(protocol, r_A);
+				break;
+			case BPF_ANC | SKF_AD_PKTTYPE:
+				__emit_skb_load8(__pkt_type_offset, r_A);
+				emit_andi(r_A, PKT_TYPE_MAX, r_A);
+				emit_alu_K(SRL, 5);
+				break;
+			case BPF_ANC | SKF_AD_IFINDEX:
+				emit_skb_loadptr(dev, r_A);
+				emit_cmpi(r_A, 0);
+				emit_branch(BE_PTR, cleanup_addr + 4);
+				emit_nop();
+				emit_load32(r_A, struct net_device, ifindex, r_A);
+				break;
+			case BPF_ANC | SKF_AD_MARK:
+				emit_skb_load32(mark, r_A);
+				break;
+			case BPF_ANC | SKF_AD_QUEUE:
+				emit_skb_load16(queue_mapping, r_A);
+				break;
+			case BPF_ANC | SKF_AD_HATYPE:
+				emit_skb_loadptr(dev, r_A);
+				emit_cmpi(r_A, 0);
+				emit_branch(BE_PTR, cleanup_addr + 4);
+				emit_nop();
+				emit_load16(r_A, struct net_device, type, r_A);
+				break;
+			case BPF_ANC | SKF_AD_RXHASH:
+				emit_skb_load32(hash, r_A);
+				break;
+			case BPF_ANC | SKF_AD_VLAN_TAG:
+				emit_skb_load16(vlan_tci, r_A);
+				break;
+			case BPF_ANC | SKF_AD_VLAN_TAG_PRESENT:
+				emit_skb_load32(vlan_all, r_A);
+				emit_cmpi(r_A, 0);
+				emit_branch_off(BE, 12);
+				emit_nop();
+				emit_loadimm(1, r_A);
+				break;
+			case BPF_LD | BPF_W | BPF_LEN:
+				emit_skb_load32(len, r_A);
+				break;
+			case BPF_LDX | BPF_W | BPF_LEN:
+				emit_skb_load32(len, r_X);
+				break;
+			case BPF_LD | BPF_IMM:
+				emit_loadimm(K, r_A);
+				break;
+			case BPF_LDX | BPF_IMM:
+				emit_loadimm(K, r_X);
+				break;
+			case BPF_LD | BPF_MEM:
+				seen |= SEEN_MEM;
+				emit_ldmem(K * 4, r_A);
+				break;
+			case BPF_LDX | BPF_MEM:
+				seen |= SEEN_MEM | SEEN_XREG;
+				emit_ldmem(K * 4, r_X);
+				break;
+			case BPF_ST:
+				seen |= SEEN_MEM;
+				emit_stmem(K * 4, r_A);
+				break;
+			case BPF_STX:
+				seen |= SEEN_MEM | SEEN_XREG;
+				emit_stmem(K * 4, r_X);
+				break;
+
+#define CHOOSE_LOAD_FUNC(K, func) \
+	((int)K < 0 ? ((int)K >= SKF_LL_OFF ? func##_negative_offset : func) : func##_positive_offset)
+
+			case BPF_LD | BPF_W | BPF_ABS:
+				func = CHOOSE_LOAD_FUNC(K, bpf_jit_load_word);
+common_load:			seen |= SEEN_DATAREF;
+				emit_loadimm(K, r_OFF);
+				emit_call(func);
+				break;
+			case BPF_LD | BPF_H | BPF_ABS:
+				func = CHOOSE_LOAD_FUNC(K, bpf_jit_load_half);
+				goto common_load;
+			case BPF_LD | BPF_B | BPF_ABS:
+				func = CHOOSE_LOAD_FUNC(K, bpf_jit_load_byte);
+				goto common_load;
+			case BPF_LDX | BPF_B | BPF_MSH:
+				func = CHOOSE_LOAD_FUNC(K, bpf_jit_load_byte_msh);
+				goto common_load;
+			case BPF_LD | BPF_W | BPF_IND:
+				func = bpf_jit_load_word;
+common_load_ind:		seen |= SEEN_DATAREF | SEEN_XREG;
+				if (K) {
+					if (is_simm13(K)) {
+						emit_addi(r_X, K, r_OFF);
+					} else {
+						emit_loadimm(K, r_TMP);
+						emit_add(r_X, r_TMP, r_OFF);
+					}
+				} else {
+					emit_reg_move(r_X, r_OFF);
+				}
+				emit_call(func);
+				break;
+			case BPF_LD | BPF_H | BPF_IND:
+				func = bpf_jit_load_half;
+				goto common_load_ind;
+			case BPF_LD | BPF_B | BPF_IND:
+				func = bpf_jit_load_byte;
+				goto common_load_ind;
+			case BPF_JMP | BPF_JA:
+				emit_jump(addrs[i + K]);
+				emit_nop();
+				break;
+
+#define COND_SEL(CODE, TOP, FOP)	\
+	case CODE:			\
+		t_op = TOP;		\
+		f_op = FOP;		\
+		goto cond_branch
+
+			COND_SEL(BPF_JMP | BPF_JGT | BPF_K, BGU, BLEU);
+			COND_SEL(BPF_JMP | BPF_JGE | BPF_K, BGEU, BLU);
+			COND_SEL(BPF_JMP | BPF_JEQ | BPF_K, BE, BNE);
+			COND_SEL(BPF_JMP | BPF_JSET | BPF_K, BNE, BE);
+			COND_SEL(BPF_JMP | BPF_JGT | BPF_X, BGU, BLEU);
+			COND_SEL(BPF_JMP | BPF_JGE | BPF_X, BGEU, BLU);
+			COND_SEL(BPF_JMP | BPF_JEQ | BPF_X, BE, BNE);
+			COND_SEL(BPF_JMP | BPF_JSET | BPF_X, BNE, BE);
+
+cond_branch:			f_offset = addrs[i + filter[i].jf];
+				t_offset = addrs[i + filter[i].jt];
+
+				/* same targets, can avoid doing the test :) */
+				if (filter[i].jt == filter[i].jf) {
+					emit_jump(t_offset);
+					emit_nop();
+					break;
+				}
+
+				switch (code) {
+				case BPF_JMP | BPF_JGT | BPF_X:
+				case BPF_JMP | BPF_JGE | BPF_X:
+				case BPF_JMP | BPF_JEQ | BPF_X:
+					seen |= SEEN_XREG;
+					emit_cmp(r_A, r_X);
+					break;
+				case BPF_JMP | BPF_JSET | BPF_X:
+					seen |= SEEN_XREG;
+					emit_btst(r_A, r_X);
+					break;
+				case BPF_JMP | BPF_JEQ | BPF_K:
+				case BPF_JMP | BPF_JGT | BPF_K:
+				case BPF_JMP | BPF_JGE | BPF_K:
+					if (is_simm13(K)) {
+						emit_cmpi(r_A, K);
+					} else {
+						emit_loadimm(K, r_TMP);
+						emit_cmp(r_A, r_TMP);
+					}
+					break;
+				case BPF_JMP | BPF_JSET | BPF_K:
+					if (is_simm13(K)) {
+						emit_btsti(r_A, K);
+					} else {
+						emit_loadimm(K, r_TMP);
+						emit_btst(r_A, r_TMP);
+					}
+					break;
+				}
+				if (filter[i].jt != 0) {
+					if (filter[i].jf)
+						t_offset += 8;
+					emit_branch(t_op, t_offset);
+					emit_nop(); /* delay slot */
+					if (filter[i].jf) {
+						emit_jump(f_offset);
+						emit_nop();
+					}
+					break;
+				}
+				emit_branch(f_op, f_offset);
+				emit_nop(); /* delay slot */
+				break;
+
+			default:
+				/* hmm, too complex filter, give up with jit compiler */
+				goto out;
+			}
+			ilen = (void *) prog - (void *) temp;
+			if (image) {
+				if (unlikely(proglen + ilen > oldproglen)) {
+					pr_err("bpb_jit_compile fatal error\n");
+					kfree(addrs);
+					execmem_free(image);
+					return;
+				}
+				memcpy(image + proglen, temp, ilen);
+			}
+			proglen += ilen;
+			addrs[i] = proglen;
+			prog = temp;
+		}
+		/* last bpf instruction is always a RET :
+		 * use it to give the cleanup instruction(s) addr
+		 */
+		cleanup_addr = proglen - 8; /* jmpl; mov r_A,%o0; */
+		if (seen_or_pass0 & SEEN_MEM)
+			cleanup_addr -= 4; /* add %sp, X, %sp; */
+
+		if (image) {
+			if (proglen != oldproglen)
+				pr_err("bpb_jit_compile proglen=%u != oldproglen=%u\n",
+				       proglen, oldproglen);
+			break;
+		}
+		if (proglen == oldproglen) {
+			image = execmem_alloc(EXECMEM_BPF, proglen);
+			if (!image)
+				goto out;
+		}
+		oldproglen = proglen;
+	}
+
+	if (bpf_jit_enable > 1)
+		bpf_jit_dump(flen, proglen, pass + 1, image);
+
+	if (image) {
+		fp->bpf_func = (void *)image;
+		fp->jited = 1;
+	}
+out:
+	kfree(addrs);
+	return;
+}
+
+void bpf_jit_free(struct bpf_prog *fp)
+{
+	if (fp->jited)
+		execmem_free(fp->bpf_func);
+
+	bpf_prog_unlock_free(fp);
+}
diff --git a/arch/sparc/net/bpf_jit_comp_64.c b/arch/sparc/net/bpf_jit_comp_64.c
new file mode 100644
index 000000000..2fa0e9375
--- /dev/null
+++ b/arch/sparc/net/bpf_jit_comp_64.c
@@ -0,0 +1,1613 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/moduleloader.h>
+#include <linux/workqueue.h>
+#include <linux/netdevice.h>
+#include <linux/filter.h>
+#include <linux/bpf.h>
+#include <linux/cache.h>
+#include <linux/if_vlan.h>
+
+#include <asm/cacheflush.h>
+#include <asm/ptrace.h>
+
+#include "bpf_jit_64.h"
+
+static inline bool is_simm13(unsigned int value)
+{
+	return value + 0x1000 < 0x2000;
+}
+
+static inline bool is_simm10(unsigned int value)
+{
+	return value + 0x200 < 0x400;
+}
+
+static inline bool is_simm5(unsigned int value)
+{
+	return value + 0x10 < 0x20;
+}
+
+static inline bool is_sethi(unsigned int value)
+{
+	return (value & ~0x3fffff) == 0;
+}
+
+static void bpf_flush_icache(void *start_, void *end_)
+{
+	/* Cheetah's I-cache is fully coherent.  */
+	if (tlb_type == spitfire) {
+		unsigned long start = (unsigned long) start_;
+		unsigned long end = (unsigned long) end_;
+
+		start &= ~7UL;
+		end = (end + 7UL) & ~7UL;
+		while (start < end) {
+			flushi(start);
+			start += 32;
+		}
+	}
+}
+
+#define S13(X)		((X) & 0x1fff)
+#define S5(X)		((X) & 0x1f)
+#define IMMED		0x00002000
+#define RD(X)		((X) << 25)
+#define RS1(X)		((X) << 14)
+#define RS2(X)		((X))
+#define OP(X)		((X) << 30)
+#define OP2(X)		((X) << 22)
+#define OP3(X)		((X) << 19)
+#define COND(X)		(((X) & 0xf) << 25)
+#define CBCOND(X)	(((X) & 0x1f) << 25)
+#define F1(X)		OP(X)
+#define F2(X, Y)	(OP(X) | OP2(Y))
+#define F3(X, Y)	(OP(X) | OP3(Y))
+#define ASI(X)		(((X) & 0xff) << 5)
+
+#define CONDN		COND(0x0)
+#define CONDE		COND(0x1)
+#define CONDLE		COND(0x2)
+#define CONDL		COND(0x3)
+#define CONDLEU		COND(0x4)
+#define CONDCS		COND(0x5)
+#define CONDNEG		COND(0x6)
+#define CONDVC		COND(0x7)
+#define CONDA		COND(0x8)
+#define CONDNE		COND(0x9)
+#define CONDG		COND(0xa)
+#define CONDGE		COND(0xb)
+#define CONDGU		COND(0xc)
+#define CONDCC		COND(0xd)
+#define CONDPOS		COND(0xe)
+#define CONDVS		COND(0xf)
+
+#define CONDGEU		CONDCC
+#define CONDLU		CONDCS
+
+#define WDISP22(X)	(((X) >> 2) & 0x3fffff)
+#define WDISP19(X)	(((X) >> 2) & 0x7ffff)
+
+/* The 10-bit branch displacement for CBCOND is split into two fields */
+static u32 WDISP10(u32 off)
+{
+	u32 ret = ((off >> 2) & 0xff) << 5;
+
+	ret |= ((off >> (2 + 8)) & 0x03) << 19;
+
+	return ret;
+}
+
+#define CBCONDE		CBCOND(0x09)
+#define CBCONDLE	CBCOND(0x0a)
+#define CBCONDL		CBCOND(0x0b)
+#define CBCONDLEU	CBCOND(0x0c)
+#define CBCONDCS	CBCOND(0x0d)
+#define CBCONDN		CBCOND(0x0e)
+#define CBCONDVS	CBCOND(0x0f)
+#define CBCONDNE	CBCOND(0x19)
+#define CBCONDG		CBCOND(0x1a)
+#define CBCONDGE	CBCOND(0x1b)
+#define CBCONDGU	CBCOND(0x1c)
+#define CBCONDCC	CBCOND(0x1d)
+#define CBCONDPOS	CBCOND(0x1e)
+#define CBCONDVC	CBCOND(0x1f)
+
+#define CBCONDGEU	CBCONDCC
+#define CBCONDLU	CBCONDCS
+
+#define ANNUL		(1 << 29)
+#define XCC		(1 << 21)
+
+#define BRANCH		(F2(0, 1) | XCC)
+#define CBCOND_OP	(F2(0, 3) | XCC)
+
+#define BA		(BRANCH | CONDA)
+#define BG		(BRANCH | CONDG)
+#define BL		(BRANCH | CONDL)
+#define BLE		(BRANCH | CONDLE)
+#define BGU		(BRANCH | CONDGU)
+#define BLEU		(BRANCH | CONDLEU)
+#define BGE		(BRANCH | CONDGE)
+#define BGEU		(BRANCH | CONDGEU)
+#define BLU		(BRANCH | CONDLU)
+#define BE		(BRANCH | CONDE)
+#define BNE		(BRANCH | CONDNE)
+
+#define SETHI(K, REG)	\
+	(F2(0, 0x4) | RD(REG) | (((K) >> 10) & 0x3fffff))
+#define OR_LO(K, REG)	\
+	(F3(2, 0x02) | IMMED | RS1(REG) | ((K) & 0x3ff) | RD(REG))
+
+#define ADD		F3(2, 0x00)
+#define AND		F3(2, 0x01)
+#define ANDCC		F3(2, 0x11)
+#define OR		F3(2, 0x02)
+#define XOR		F3(2, 0x03)
+#define SUB		F3(2, 0x04)
+#define SUBCC		F3(2, 0x14)
+#define MUL		F3(2, 0x0a)
+#define MULX		F3(2, 0x09)
+#define UDIVX		F3(2, 0x0d)
+#define DIV		F3(2, 0x0e)
+#define SLL		F3(2, 0x25)
+#define SLLX		(F3(2, 0x25)|(1<<12))
+#define SRA		F3(2, 0x27)
+#define SRAX		(F3(2, 0x27)|(1<<12))
+#define SRL		F3(2, 0x26)
+#define SRLX		(F3(2, 0x26)|(1<<12))
+#define JMPL		F3(2, 0x38)
+#define SAVE		F3(2, 0x3c)
+#define RESTORE		F3(2, 0x3d)
+#define CALL		F1(1)
+#define BR		F2(0, 0x01)
+#define RD_Y		F3(2, 0x28)
+#define WR_Y		F3(2, 0x30)
+
+#define LD32		F3(3, 0x00)
+#define LD8		F3(3, 0x01)
+#define LD16		F3(3, 0x02)
+#define LD64		F3(3, 0x0b)
+#define LD64A		F3(3, 0x1b)
+#define ST8		F3(3, 0x05)
+#define ST16		F3(3, 0x06)
+#define ST32		F3(3, 0x04)
+#define ST64		F3(3, 0x0e)
+
+#define CAS		F3(3, 0x3c)
+#define CASX		F3(3, 0x3e)
+
+#define LDPTR		LD64
+#define BASE_STACKFRAME	176
+
+#define LD32I		(LD32 | IMMED)
+#define LD8I		(LD8 | IMMED)
+#define LD16I		(LD16 | IMMED)
+#define LD64I		(LD64 | IMMED)
+#define LDPTRI		(LDPTR | IMMED)
+#define ST32I		(ST32 | IMMED)
+
+struct jit_ctx {
+	struct bpf_prog		*prog;
+	unsigned int		*offset;
+	int			idx;
+	int			epilogue_offset;
+	bool 			tmp_1_used;
+	bool 			tmp_2_used;
+	bool 			tmp_3_used;
+	bool			saw_frame_pointer;
+	bool			saw_call;
+	bool			saw_tail_call;
+	u32			*image;
+};
+
+#define TMP_REG_1	(MAX_BPF_JIT_REG + 0)
+#define TMP_REG_2	(MAX_BPF_JIT_REG + 1)
+#define TMP_REG_3	(MAX_BPF_JIT_REG + 2)
+
+/* Map BPF registers to SPARC registers */
+static const int bpf2sparc[] = {
+	/* return value from in-kernel function, and exit value from eBPF */
+	[BPF_REG_0] = O5,
+
+	/* arguments from eBPF program to in-kernel function */
+	[BPF_REG_1] = O0,
+	[BPF_REG_2] = O1,
+	[BPF_REG_3] = O2,
+	[BPF_REG_4] = O3,
+	[BPF_REG_5] = O4,
+
+	/* callee saved registers that in-kernel function will preserve */
+	[BPF_REG_6] = L0,
+	[BPF_REG_7] = L1,
+	[BPF_REG_8] = L2,
+	[BPF_REG_9] = L3,
+
+	/* read-only frame pointer to access stack */
+	[BPF_REG_FP] = L6,
+
+	[BPF_REG_AX] = G7,
+
+	/* temporary register for BPF JIT */
+	[TMP_REG_1] = G1,
+	[TMP_REG_2] = G2,
+	[TMP_REG_3] = G3,
+};
+
+static void emit(const u32 insn, struct jit_ctx *ctx)
+{
+	if (ctx->image != NULL)
+		ctx->image[ctx->idx] = insn;
+
+	ctx->idx++;
+}
+
+static void emit_call(u32 *func, struct jit_ctx *ctx)
+{
+	if (ctx->image != NULL) {
+		void *here = &ctx->image[ctx->idx];
+		unsigned int off;
+
+		off = (void *)func - here;
+		ctx->image[ctx->idx] = CALL | ((off >> 2) & 0x3fffffff);
+	}
+	ctx->idx++;
+}
+
+static void emit_nop(struct jit_ctx *ctx)
+{
+	emit(SETHI(0, G0), ctx);
+}
+
+static void emit_reg_move(u32 from, u32 to, struct jit_ctx *ctx)
+{
+	emit(OR | RS1(G0) | RS2(from) | RD(to), ctx);
+}
+
+/* Emit 32-bit constant, zero extended. */
+static void emit_set_const(s32 K, u32 reg, struct jit_ctx *ctx)
+{
+	emit(SETHI(K, reg), ctx);
+	emit(OR_LO(K, reg), ctx);
+}
+
+/* Emit 32-bit constant, sign extended. */
+static void emit_set_const_sext(s32 K, u32 reg, struct jit_ctx *ctx)
+{
+	if (K >= 0) {
+		emit(SETHI(K, reg), ctx);
+		emit(OR_LO(K, reg), ctx);
+	} else {
+		u32 hbits = ~(u32) K;
+		u32 lbits = -0x400 | (u32) K;
+
+		emit(SETHI(hbits, reg), ctx);
+		emit(XOR | IMMED | RS1(reg) | S13(lbits) | RD(reg), ctx);
+	}
+}
+
+static void emit_alu(u32 opcode, u32 src, u32 dst, struct jit_ctx *ctx)
+{
+	emit(opcode | RS1(dst) | RS2(src) | RD(dst), ctx);
+}
+
+static void emit_alu3(u32 opcode, u32 a, u32 b, u32 c, struct jit_ctx *ctx)
+{
+	emit(opcode | RS1(a) | RS2(b) | RD(c), ctx);
+}
+
+static void emit_alu_K(unsigned int opcode, unsigned int dst, unsigned int imm,
+		       struct jit_ctx *ctx)
+{
+	bool small_immed = is_simm13(imm);
+	unsigned int insn = opcode;
+
+	insn |= RS1(dst) | RD(dst);
+	if (small_immed) {
+		emit(insn | IMMED | S13(imm), ctx);
+	} else {
+		unsigned int tmp = bpf2sparc[TMP_REG_1];
+
+		ctx->tmp_1_used = true;
+
+		emit_set_const_sext(imm, tmp, ctx);
+		emit(insn | RS2(tmp), ctx);
+	}
+}
+
+static void emit_alu3_K(unsigned int opcode, unsigned int src, unsigned int imm,
+			unsigned int dst, struct jit_ctx *ctx)
+{
+	bool small_immed = is_simm13(imm);
+	unsigned int insn = opcode;
+
+	insn |= RS1(src) | RD(dst);
+	if (small_immed) {
+		emit(insn | IMMED | S13(imm), ctx);
+	} else {
+		unsigned int tmp = bpf2sparc[TMP_REG_1];
+
+		ctx->tmp_1_used = true;
+
+		emit_set_const_sext(imm, tmp, ctx);
+		emit(insn | RS2(tmp), ctx);
+	}
+}
+
+static void emit_loadimm32(s32 K, unsigned int dest, struct jit_ctx *ctx)
+{
+	if (K >= 0 && is_simm13(K)) {
+		/* or %g0, K, DEST */
+		emit(OR | IMMED | RS1(G0) | S13(K) | RD(dest), ctx);
+	} else {
+		emit_set_const(K, dest, ctx);
+	}
+}
+
+static void emit_loadimm(s32 K, unsigned int dest, struct jit_ctx *ctx)
+{
+	if (is_simm13(K)) {
+		/* or %g0, K, DEST */
+		emit(OR | IMMED | RS1(G0) | S13(K) | RD(dest), ctx);
+	} else {
+		emit_set_const(K, dest, ctx);
+	}
+}
+
+static void emit_loadimm_sext(s32 K, unsigned int dest, struct jit_ctx *ctx)
+{
+	if (is_simm13(K)) {
+		/* or %g0, K, DEST */
+		emit(OR | IMMED | RS1(G0) | S13(K) | RD(dest), ctx);
+	} else {
+		emit_set_const_sext(K, dest, ctx);
+	}
+}
+
+static void analyze_64bit_constant(u32 high_bits, u32 low_bits,
+				   int *hbsp, int *lbsp, int *abbasp)
+{
+	int lowest_bit_set, highest_bit_set, all_bits_between_are_set;
+	int i;
+
+	lowest_bit_set = highest_bit_set = -1;
+	i = 0;
+	do {
+		if ((lowest_bit_set == -1) && ((low_bits >> i) & 1))
+			lowest_bit_set = i;
+		if ((highest_bit_set == -1) && ((high_bits >> (32 - i - 1)) & 1))
+			highest_bit_set = (64 - i - 1);
+	}  while (++i < 32 && (highest_bit_set == -1 ||
+			       lowest_bit_set == -1));
+	if (i == 32) {
+		i = 0;
+		do {
+			if (lowest_bit_set == -1 && ((high_bits >> i) & 1))
+				lowest_bit_set = i + 32;
+			if (highest_bit_set == -1 &&
+			    ((low_bits >> (32 - i - 1)) & 1))
+				highest_bit_set = 32 - i - 1;
+		} while (++i < 32 && (highest_bit_set == -1 ||
+				      lowest_bit_set == -1));
+	}
+
+	all_bits_between_are_set = 1;
+	for (i = lowest_bit_set; i <= highest_bit_set; i++) {
+		if (i < 32) {
+			if ((low_bits & (1 << i)) != 0)
+				continue;
+		} else {
+			if ((high_bits & (1 << (i - 32))) != 0)
+				continue;
+		}
+		all_bits_between_are_set = 0;
+		break;
+	}
+	*hbsp = highest_bit_set;
+	*lbsp = lowest_bit_set;
+	*abbasp = all_bits_between_are_set;
+}
+
+static unsigned long create_simple_focus_bits(unsigned long high_bits,
+					      unsigned long low_bits,
+					      int lowest_bit_set, int shift)
+{
+	long hi, lo;
+
+	if (lowest_bit_set < 32) {
+		lo = (low_bits >> lowest_bit_set) << shift;
+		hi = ((high_bits << (32 - lowest_bit_set)) << shift);
+	} else {
+		lo = 0;
+		hi = ((high_bits >> (lowest_bit_set - 32)) << shift);
+	}
+	return hi | lo;
+}
+
+static bool const64_is_2insns(unsigned long high_bits,
+			      unsigned long low_bits)
+{
+	int highest_bit_set, lowest_bit_set, all_bits_between_are_set;
+
+	if (high_bits == 0 || high_bits == 0xffffffff)
+		return true;
+
+	analyze_64bit_constant(high_bits, low_bits,
+			       &highest_bit_set, &lowest_bit_set,
+			       &all_bits_between_are_set);
+
+	if ((highest_bit_set == 63 || lowest_bit_set == 0) &&
+	    all_bits_between_are_set != 0)
+		return true;
+
+	if (highest_bit_set - lowest_bit_set < 21)
+		return true;
+
+	return false;
+}
+
+static void sparc_emit_set_const64_quick2(unsigned long high_bits,
+					  unsigned long low_imm,
+					  unsigned int dest,
+					  int shift_count, struct jit_ctx *ctx)
+{
+	emit_loadimm32(high_bits, dest, ctx);
+
+	/* Now shift it up into place.  */
+	emit_alu_K(SLLX, dest, shift_count, ctx);
+
+	/* If there is a low immediate part piece, finish up by
+	 * putting that in as well.
+	 */
+	if (low_imm != 0)
+		emit(OR | IMMED | RS1(dest) | S13(low_imm) | RD(dest), ctx);
+}
+
+static void emit_loadimm64(u64 K, unsigned int dest, struct jit_ctx *ctx)
+{
+	int all_bits_between_are_set, lowest_bit_set, highest_bit_set;
+	unsigned int tmp = bpf2sparc[TMP_REG_1];
+	u32 low_bits = (K & 0xffffffff);
+	u32 high_bits = (K >> 32);
+
+	/* These two tests also take care of all of the one
+	 * instruction cases.
+	 */
+	if (high_bits == 0xffffffff && (low_bits & 0x80000000))
+		return emit_loadimm_sext(K, dest, ctx);
+	if (high_bits == 0x00000000)
+		return emit_loadimm32(K, dest, ctx);
+
+	analyze_64bit_constant(high_bits, low_bits, &highest_bit_set,
+			       &lowest_bit_set, &all_bits_between_are_set);
+
+	/* 1) mov	-1, %reg
+	 *    sllx	%reg, shift, %reg
+	 * 2) mov	-1, %reg
+	 *    srlx	%reg, shift, %reg
+	 * 3) mov	some_small_const, %reg
+	 *    sllx	%reg, shift, %reg
+	 */
+	if (((highest_bit_set == 63 || lowest_bit_set == 0) &&
+	     all_bits_between_are_set != 0) ||
+	    ((highest_bit_set - lowest_bit_set) < 12)) {
+		int shift = lowest_bit_set;
+		long the_const = -1;
+
+		if ((highest_bit_set != 63 && lowest_bit_set != 0) ||
+		    all_bits_between_are_set == 0) {
+			the_const =
+				create_simple_focus_bits(high_bits, low_bits,
+							 lowest_bit_set, 0);
+		} else if (lowest_bit_set == 0)
+			shift = -(63 - highest_bit_set);
+
+		emit(OR | IMMED | RS1(G0) | S13(the_const) | RD(dest), ctx);
+		if (shift > 0)
+			emit_alu_K(SLLX, dest, shift, ctx);
+		else if (shift < 0)
+			emit_alu_K(SRLX, dest, -shift, ctx);
+
+		return;
+	}
+
+	/* Now a range of 22 or less bits set somewhere.
+	 * 1) sethi	%hi(focus_bits), %reg
+	 *    sllx	%reg, shift, %reg
+	 * 2) sethi	%hi(focus_bits), %reg
+	 *    srlx	%reg, shift, %reg
+	 */
+	if ((highest_bit_set - lowest_bit_set) < 21) {
+		unsigned long focus_bits =
+			create_simple_focus_bits(high_bits, low_bits,
+						 lowest_bit_set, 10);
+
+		emit(SETHI(focus_bits, dest), ctx);
+
+		/* If lowest_bit_set == 10 then a sethi alone could
+		 * have done it.
+		 */
+		if (lowest_bit_set < 10)
+			emit_alu_K(SRLX, dest, 10 - lowest_bit_set, ctx);
+		else if (lowest_bit_set > 10)
+			emit_alu_K(SLLX, dest, lowest_bit_set - 10, ctx);
+		return;
+	}
+
+	/* Ok, now 3 instruction sequences.  */
+	if (low_bits == 0) {
+		emit_loadimm32(high_bits, dest, ctx);
+		emit_alu_K(SLLX, dest, 32, ctx);
+		return;
+	}
+
+	/* We may be able to do something quick
+	 * when the constant is negated, so try that.
+	 */
+	if (const64_is_2insns((~high_bits) & 0xffffffff,
+			      (~low_bits) & 0xfffffc00)) {
+		/* NOTE: The trailing bits get XOR'd so we need the
+		 * non-negated bits, not the negated ones.
+		 */
+		unsigned long trailing_bits = low_bits & 0x3ff;
+
+		if ((((~high_bits) & 0xffffffff) == 0 &&
+		     ((~low_bits) & 0x80000000) == 0) ||
+		    (((~high_bits) & 0xffffffff) == 0xffffffff &&
+		     ((~low_bits) & 0x80000000) != 0)) {
+			unsigned long fast_int = (~low_bits & 0xffffffff);
+
+			if ((is_sethi(fast_int) &&
+			     (~high_bits & 0xffffffff) == 0)) {
+				emit(SETHI(fast_int, dest), ctx);
+			} else if (is_simm13(fast_int)) {
+				emit(OR | IMMED | RS1(G0) | S13(fast_int) | RD(dest), ctx);
+			} else {
+				emit_loadimm64(fast_int, dest, ctx);
+			}
+		} else {
+			u64 n = ((~low_bits) & 0xfffffc00) |
+				(((unsigned long)((~high_bits) & 0xffffffff))<<32);
+			emit_loadimm64(n, dest, ctx);
+		}
+
+		low_bits = -0x400 | trailing_bits;
+
+		emit(XOR | IMMED | RS1(dest) | S13(low_bits) | RD(dest), ctx);
+		return;
+	}
+
+	/* 1) sethi	%hi(xxx), %reg
+	 *    or	%reg, %lo(xxx), %reg
+	 *    sllx	%reg, yyy, %reg
+	 */
+	if ((highest_bit_set - lowest_bit_set) < 32) {
+		unsigned long focus_bits =
+			create_simple_focus_bits(high_bits, low_bits,
+						 lowest_bit_set, 0);
+
+		/* So what we know is that the set bits straddle the
+		 * middle of the 64-bit word.
+		 */
+		sparc_emit_set_const64_quick2(focus_bits, 0, dest,
+					      lowest_bit_set, ctx);
+		return;
+	}
+
+	/* 1) sethi	%hi(high_bits), %reg
+	 *    or	%reg, %lo(high_bits), %reg
+	 *    sllx	%reg, 32, %reg
+	 *    or	%reg, low_bits, %reg
+	 */
+	if (is_simm13(low_bits) && ((int)low_bits > 0)) {
+		sparc_emit_set_const64_quick2(high_bits, low_bits,
+					      dest, 32, ctx);
+		return;
+	}
+
+	/* Oh well, we tried... Do a full 64-bit decomposition.  */
+	ctx->tmp_1_used = true;
+
+	emit_loadimm32(high_bits, tmp, ctx);
+	emit_loadimm32(low_bits, dest, ctx);
+	emit_alu_K(SLLX, tmp, 32, ctx);
+	emit(OR | RS1(dest) | RS2(tmp) | RD(dest), ctx);
+}
+
+static void emit_branch(unsigned int br_opc, unsigned int from_idx, unsigned int to_idx,
+			struct jit_ctx *ctx)
+{
+	unsigned int off = to_idx - from_idx;
+
+	if (br_opc & XCC)
+		emit(br_opc | WDISP19(off << 2), ctx);
+	else
+		emit(br_opc | WDISP22(off << 2), ctx);
+}
+
+static void emit_cbcond(unsigned int cb_opc, unsigned int from_idx, unsigned int to_idx,
+			const u8 dst, const u8 src, struct jit_ctx *ctx)
+{
+	unsigned int off = to_idx - from_idx;
+
+	emit(cb_opc | WDISP10(off << 2) | RS1(dst) | RS2(src), ctx);
+}
+
+static void emit_cbcondi(unsigned int cb_opc, unsigned int from_idx, unsigned int to_idx,
+			 const u8 dst, s32 imm, struct jit_ctx *ctx)
+{
+	unsigned int off = to_idx - from_idx;
+
+	emit(cb_opc | IMMED | WDISP10(off << 2) | RS1(dst) | S5(imm), ctx);
+}
+
+#define emit_read_y(REG, CTX)	emit(RD_Y | RD(REG), CTX)
+#define emit_write_y(REG, CTX)	emit(WR_Y | IMMED | RS1(REG) | S13(0), CTX)
+
+#define emit_cmp(R1, R2, CTX)				\
+	emit(SUBCC | RS1(R1) | RS2(R2) | RD(G0), CTX)
+
+#define emit_cmpi(R1, IMM, CTX)				\
+	emit(SUBCC | IMMED | RS1(R1) | S13(IMM) | RD(G0), CTX)
+
+#define emit_btst(R1, R2, CTX)				\
+	emit(ANDCC | RS1(R1) | RS2(R2) | RD(G0), CTX)
+
+#define emit_btsti(R1, IMM, CTX)			\
+	emit(ANDCC | IMMED | RS1(R1) | S13(IMM) | RD(G0), CTX)
+
+static int emit_compare_and_branch(const u8 code, const u8 dst, u8 src,
+				   const s32 imm, bool is_imm, int branch_dst,
+				   struct jit_ctx *ctx)
+{
+	bool use_cbcond = (sparc64_elf_hwcap & AV_SPARC_CBCOND) != 0;
+	const u8 tmp = bpf2sparc[TMP_REG_1];
+
+	branch_dst = ctx->offset[branch_dst];
+
+	if (!is_simm10(branch_dst - ctx->idx) ||
+	    BPF_OP(code) == BPF_JSET)
+		use_cbcond = false;
+
+	if (is_imm) {
+		bool fits = true;
+
+		if (use_cbcond) {
+			if (!is_simm5(imm))
+				fits = false;
+		} else if (!is_simm13(imm)) {
+			fits = false;
+		}
+		if (!fits) {
+			ctx->tmp_1_used = true;
+			emit_loadimm_sext(imm, tmp, ctx);
+			src = tmp;
+			is_imm = false;
+		}
+	}
+
+	if (!use_cbcond) {
+		u32 br_opcode;
+
+		if (BPF_OP(code) == BPF_JSET) {
+			if (is_imm)
+				emit_btsti(dst, imm, ctx);
+			else
+				emit_btst(dst, src, ctx);
+		} else {
+			if (is_imm)
+				emit_cmpi(dst, imm, ctx);
+			else
+				emit_cmp(dst, src, ctx);
+		}
+		switch (BPF_OP(code)) {
+		case BPF_JEQ:
+			br_opcode = BE;
+			break;
+		case BPF_JGT:
+			br_opcode = BGU;
+			break;
+		case BPF_JLT:
+			br_opcode = BLU;
+			break;
+		case BPF_JGE:
+			br_opcode = BGEU;
+			break;
+		case BPF_JLE:
+			br_opcode = BLEU;
+			break;
+		case BPF_JSET:
+		case BPF_JNE:
+			br_opcode = BNE;
+			break;
+		case BPF_JSGT:
+			br_opcode = BG;
+			break;
+		case BPF_JSLT:
+			br_opcode = BL;
+			break;
+		case BPF_JSGE:
+			br_opcode = BGE;
+			break;
+		case BPF_JSLE:
+			br_opcode = BLE;
+			break;
+		default:
+			/* Make sure we dont leak kernel information to the
+			 * user.
+			 */
+			return -EFAULT;
+		}
+		emit_branch(br_opcode, ctx->idx, branch_dst, ctx);
+		emit_nop(ctx);
+	} else {
+		u32 cbcond_opcode;
+
+		switch (BPF_OP(code)) {
+		case BPF_JEQ:
+			cbcond_opcode = CBCONDE;
+			break;
+		case BPF_JGT:
+			cbcond_opcode = CBCONDGU;
+			break;
+		case BPF_JLT:
+			cbcond_opcode = CBCONDLU;
+			break;
+		case BPF_JGE:
+			cbcond_opcode = CBCONDGEU;
+			break;
+		case BPF_JLE:
+			cbcond_opcode = CBCONDLEU;
+			break;
+		case BPF_JNE:
+			cbcond_opcode = CBCONDNE;
+			break;
+		case BPF_JSGT:
+			cbcond_opcode = CBCONDG;
+			break;
+		case BPF_JSLT:
+			cbcond_opcode = CBCONDL;
+			break;
+		case BPF_JSGE:
+			cbcond_opcode = CBCONDGE;
+			break;
+		case BPF_JSLE:
+			cbcond_opcode = CBCONDLE;
+			break;
+		default:
+			/* Make sure we dont leak kernel information to the
+			 * user.
+			 */
+			return -EFAULT;
+		}
+		cbcond_opcode |= CBCOND_OP;
+		if (is_imm)
+			emit_cbcondi(cbcond_opcode, ctx->idx, branch_dst,
+				     dst, imm, ctx);
+		else
+			emit_cbcond(cbcond_opcode, ctx->idx, branch_dst,
+				    dst, src, ctx);
+	}
+	return 0;
+}
+
+/* Just skip the save instruction and the ctx register move.  */
+#define BPF_TAILCALL_PROLOGUE_SKIP	32
+#define BPF_TAILCALL_CNT_SP_OFF		(STACK_BIAS + 128)
+
+static void build_prologue(struct jit_ctx *ctx)
+{
+	s32 stack_needed = BASE_STACKFRAME;
+
+	if (ctx->saw_frame_pointer || ctx->saw_tail_call) {
+		struct bpf_prog *prog = ctx->prog;
+		u32 stack_depth;
+
+		stack_depth = prog->aux->stack_depth;
+		stack_needed += round_up(stack_depth, 16);
+	}
+
+	if (ctx->saw_tail_call)
+		stack_needed += 8;
+
+	/* save %sp, -176, %sp */
+	emit(SAVE | IMMED | RS1(SP) | S13(-stack_needed) | RD(SP), ctx);
+
+	/* tail_call_cnt = 0 */
+	if (ctx->saw_tail_call) {
+		u32 off = BPF_TAILCALL_CNT_SP_OFF;
+
+		emit(ST32 | IMMED | RS1(SP) | S13(off) | RD(G0), ctx);
+	} else {
+		emit_nop(ctx);
+	}
+	if (ctx->saw_frame_pointer) {
+		const u8 vfp = bpf2sparc[BPF_REG_FP];
+
+		emit(ADD | IMMED | RS1(FP) | S13(STACK_BIAS) | RD(vfp), ctx);
+	} else {
+		emit_nop(ctx);
+	}
+
+	emit_reg_move(I0, O0, ctx);
+	emit_reg_move(I1, O1, ctx);
+	emit_reg_move(I2, O2, ctx);
+	emit_reg_move(I3, O3, ctx);
+	emit_reg_move(I4, O4, ctx);
+	/* If you add anything here, adjust BPF_TAILCALL_PROLOGUE_SKIP above. */
+}
+
+static void build_epilogue(struct jit_ctx *ctx)
+{
+	ctx->epilogue_offset = ctx->idx;
+
+	/* ret (jmpl %i7 + 8, %g0) */
+	emit(JMPL | IMMED | RS1(I7) | S13(8) | RD(G0), ctx);
+
+	/* restore %i5, %g0, %o0 */
+	emit(RESTORE | RS1(bpf2sparc[BPF_REG_0]) | RS2(G0) | RD(O0), ctx);
+}
+
+static void emit_tail_call(struct jit_ctx *ctx)
+{
+	const u8 bpf_array = bpf2sparc[BPF_REG_2];
+	const u8 bpf_index = bpf2sparc[BPF_REG_3];
+	const u8 tmp = bpf2sparc[TMP_REG_1];
+	u32 off;
+
+	ctx->saw_tail_call = true;
+
+	off = offsetof(struct bpf_array, map.max_entries);
+	emit(LD32 | IMMED | RS1(bpf_array) | S13(off) | RD(tmp), ctx);
+	emit_cmp(bpf_index, tmp, ctx);
+#define OFFSET1 17
+	emit_branch(BGEU, ctx->idx, ctx->idx + OFFSET1, ctx);
+	emit_nop(ctx);
+
+	off = BPF_TAILCALL_CNT_SP_OFF;
+	emit(LD32 | IMMED | RS1(SP) | S13(off) | RD(tmp), ctx);
+	emit_cmpi(tmp, MAX_TAIL_CALL_CNT, ctx);
+#define OFFSET2 13
+	emit_branch(BGEU, ctx->idx, ctx->idx + OFFSET2, ctx);
+	emit_nop(ctx);
+
+	emit_alu_K(ADD, tmp, 1, ctx);
+	off = BPF_TAILCALL_CNT_SP_OFF;
+	emit(ST32 | IMMED | RS1(SP) | S13(off) | RD(tmp), ctx);
+
+	emit_alu3_K(SLL, bpf_index, 3, tmp, ctx);
+	emit_alu(ADD, bpf_array, tmp, ctx);
+	off = offsetof(struct bpf_array, ptrs);
+	emit(LD64 | IMMED | RS1(tmp) | S13(off) | RD(tmp), ctx);
+
+	emit_cmpi(tmp, 0, ctx);
+#define OFFSET3 5
+	emit_branch(BE, ctx->idx, ctx->idx + OFFSET3, ctx);
+	emit_nop(ctx);
+
+	off = offsetof(struct bpf_prog, bpf_func);
+	emit(LD64 | IMMED | RS1(tmp) | S13(off) | RD(tmp), ctx);
+
+	off = BPF_TAILCALL_PROLOGUE_SKIP;
+	emit(JMPL | IMMED | RS1(tmp) | S13(off) | RD(G0), ctx);
+	emit_nop(ctx);
+}
+
+static int build_insn(const struct bpf_insn *insn, struct jit_ctx *ctx)
+{
+	const u8 code = insn->code;
+	const u8 dst = bpf2sparc[insn->dst_reg];
+	const u8 src = bpf2sparc[insn->src_reg];
+	const int i = insn - ctx->prog->insnsi;
+	const s16 off = insn->off;
+	const s32 imm = insn->imm;
+
+	if (insn->src_reg == BPF_REG_FP)
+		ctx->saw_frame_pointer = true;
+
+	switch (code) {
+	/* dst = src */
+	case BPF_ALU | BPF_MOV | BPF_X:
+		emit_alu3_K(SRL, src, 0, dst, ctx);
+		if (insn_is_zext(&insn[1]))
+			return 1;
+		break;
+	case BPF_ALU64 | BPF_MOV | BPF_X:
+		emit_reg_move(src, dst, ctx);
+		break;
+	/* dst = dst OP src */
+	case BPF_ALU | BPF_ADD | BPF_X:
+	case BPF_ALU64 | BPF_ADD | BPF_X:
+		emit_alu(ADD, src, dst, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU | BPF_SUB | BPF_X:
+	case BPF_ALU64 | BPF_SUB | BPF_X:
+		emit_alu(SUB, src, dst, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU | BPF_AND | BPF_X:
+	case BPF_ALU64 | BPF_AND | BPF_X:
+		emit_alu(AND, src, dst, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU | BPF_OR | BPF_X:
+	case BPF_ALU64 | BPF_OR | BPF_X:
+		emit_alu(OR, src, dst, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU | BPF_XOR | BPF_X:
+	case BPF_ALU64 | BPF_XOR | BPF_X:
+		emit_alu(XOR, src, dst, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU | BPF_MUL | BPF_X:
+		emit_alu(MUL, src, dst, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU64 | BPF_MUL | BPF_X:
+		emit_alu(MULX, src, dst, ctx);
+		break;
+	case BPF_ALU | BPF_DIV | BPF_X:
+		emit_write_y(G0, ctx);
+		emit_alu(DIV, src, dst, ctx);
+		if (insn_is_zext(&insn[1]))
+			return 1;
+		break;
+	case BPF_ALU64 | BPF_DIV | BPF_X:
+		emit_alu(UDIVX, src, dst, ctx);
+		break;
+	case BPF_ALU | BPF_MOD | BPF_X: {
+		const u8 tmp = bpf2sparc[TMP_REG_1];
+
+		ctx->tmp_1_used = true;
+
+		emit_write_y(G0, ctx);
+		emit_alu3(DIV, dst, src, tmp, ctx);
+		emit_alu3(MULX, tmp, src, tmp, ctx);
+		emit_alu3(SUB, dst, tmp, dst, ctx);
+		goto do_alu32_trunc;
+	}
+	case BPF_ALU64 | BPF_MOD | BPF_X: {
+		const u8 tmp = bpf2sparc[TMP_REG_1];
+
+		ctx->tmp_1_used = true;
+
+		emit_alu3(UDIVX, dst, src, tmp, ctx);
+		emit_alu3(MULX, tmp, src, tmp, ctx);
+		emit_alu3(SUB, dst, tmp, dst, ctx);
+		break;
+	}
+	case BPF_ALU | BPF_LSH | BPF_X:
+		emit_alu(SLL, src, dst, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU64 | BPF_LSH | BPF_X:
+		emit_alu(SLLX, src, dst, ctx);
+		break;
+	case BPF_ALU | BPF_RSH | BPF_X:
+		emit_alu(SRL, src, dst, ctx);
+		if (insn_is_zext(&insn[1]))
+			return 1;
+		break;
+	case BPF_ALU64 | BPF_RSH | BPF_X:
+		emit_alu(SRLX, src, dst, ctx);
+		break;
+	case BPF_ALU | BPF_ARSH | BPF_X:
+		emit_alu(SRA, src, dst, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU64 | BPF_ARSH | BPF_X:
+		emit_alu(SRAX, src, dst, ctx);
+		break;
+
+	/* dst = -dst */
+	case BPF_ALU | BPF_NEG:
+	case BPF_ALU64 | BPF_NEG:
+		emit(SUB | RS1(0) | RS2(dst) | RD(dst), ctx);
+		goto do_alu32_trunc;
+
+	case BPF_ALU | BPF_END | BPF_FROM_BE:
+		switch (imm) {
+		case 16:
+			emit_alu_K(SLL, dst, 16, ctx);
+			emit_alu_K(SRL, dst, 16, ctx);
+			if (insn_is_zext(&insn[1]))
+				return 1;
+			break;
+		case 32:
+			if (!ctx->prog->aux->verifier_zext)
+				emit_alu_K(SRL, dst, 0, ctx);
+			break;
+		case 64:
+			/* nop */
+			break;
+
+		}
+		break;
+
+	/* dst = BSWAP##imm(dst) */
+	case BPF_ALU | BPF_END | BPF_FROM_LE: {
+		const u8 tmp = bpf2sparc[TMP_REG_1];
+		const u8 tmp2 = bpf2sparc[TMP_REG_2];
+
+		ctx->tmp_1_used = true;
+		switch (imm) {
+		case 16:
+			emit_alu3_K(AND, dst, 0xff, tmp, ctx);
+			emit_alu3_K(SRL, dst, 8, dst, ctx);
+			emit_alu3_K(AND, dst, 0xff, dst, ctx);
+			emit_alu3_K(SLL, tmp, 8, tmp, ctx);
+			emit_alu(OR, tmp, dst, ctx);
+			if (insn_is_zext(&insn[1]))
+				return 1;
+			break;
+
+		case 32:
+			ctx->tmp_2_used = true;
+			emit_alu3_K(SRL, dst, 24, tmp, ctx);	/* tmp  = dst >> 24 */
+			emit_alu3_K(SRL, dst, 16, tmp2, ctx);	/* tmp2 = dst >> 16 */
+			emit_alu3_K(AND, tmp2, 0xff, tmp2, ctx);/* tmp2 = tmp2 & 0xff */
+			emit_alu3_K(SLL, tmp2, 8, tmp2, ctx);	/* tmp2 = tmp2 << 8 */
+			emit_alu(OR, tmp2, tmp, ctx);		/* tmp  = tmp | tmp2 */
+			emit_alu3_K(SRL, dst, 8, tmp2, ctx);	/* tmp2 = dst >> 8 */
+			emit_alu3_K(AND, tmp2, 0xff, tmp2, ctx);/* tmp2 = tmp2 & 0xff */
+			emit_alu3_K(SLL, tmp2, 16, tmp2, ctx);	/* tmp2 = tmp2 << 16 */
+			emit_alu(OR, tmp2, tmp, ctx);		/* tmp  = tmp | tmp2 */
+			emit_alu3_K(AND, dst, 0xff, dst, ctx);	/* dst	= dst & 0xff */
+			emit_alu3_K(SLL, dst, 24, dst, ctx);	/* dst  = dst << 24 */
+			emit_alu(OR, tmp, dst, ctx);		/* dst  = dst | tmp */
+			if (insn_is_zext(&insn[1]))
+				return 1;
+			break;
+
+		case 64:
+			emit_alu3_K(ADD, SP, STACK_BIAS + 128, tmp, ctx);
+			emit(ST64 | RS1(tmp) | RS2(G0) | RD(dst), ctx);
+			emit(LD64A | ASI(ASI_PL) | RS1(tmp) | RS2(G0) | RD(dst), ctx);
+			break;
+		}
+		break;
+	}
+	/* dst = imm */
+	case BPF_ALU | BPF_MOV | BPF_K:
+		emit_loadimm32(imm, dst, ctx);
+		if (insn_is_zext(&insn[1]))
+			return 1;
+		break;
+	case BPF_ALU64 | BPF_MOV | BPF_K:
+		emit_loadimm_sext(imm, dst, ctx);
+		break;
+	/* dst = dst OP imm */
+	case BPF_ALU | BPF_ADD | BPF_K:
+	case BPF_ALU64 | BPF_ADD | BPF_K:
+		emit_alu_K(ADD, dst, imm, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU | BPF_SUB | BPF_K:
+	case BPF_ALU64 | BPF_SUB | BPF_K:
+		emit_alu_K(SUB, dst, imm, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU | BPF_AND | BPF_K:
+	case BPF_ALU64 | BPF_AND | BPF_K:
+		emit_alu_K(AND, dst, imm, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU | BPF_OR | BPF_K:
+	case BPF_ALU64 | BPF_OR | BPF_K:
+		emit_alu_K(OR, dst, imm, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU | BPF_XOR | BPF_K:
+	case BPF_ALU64 | BPF_XOR | BPF_K:
+		emit_alu_K(XOR, dst, imm, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU | BPF_MUL | BPF_K:
+		emit_alu_K(MUL, dst, imm, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU64 | BPF_MUL | BPF_K:
+		emit_alu_K(MULX, dst, imm, ctx);
+		break;
+	case BPF_ALU | BPF_DIV | BPF_K:
+		if (imm == 0)
+			return -EINVAL;
+
+		emit_write_y(G0, ctx);
+		emit_alu_K(DIV, dst, imm, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU64 | BPF_DIV | BPF_K:
+		if (imm == 0)
+			return -EINVAL;
+
+		emit_alu_K(UDIVX, dst, imm, ctx);
+		break;
+	case BPF_ALU64 | BPF_MOD | BPF_K:
+	case BPF_ALU | BPF_MOD | BPF_K: {
+		const u8 tmp = bpf2sparc[TMP_REG_2];
+		unsigned int div;
+
+		if (imm == 0)
+			return -EINVAL;
+
+		div = (BPF_CLASS(code) == BPF_ALU64) ? UDIVX : DIV;
+
+		ctx->tmp_2_used = true;
+
+		if (BPF_CLASS(code) != BPF_ALU64)
+			emit_write_y(G0, ctx);
+		if (is_simm13(imm)) {
+			emit(div | IMMED | RS1(dst) | S13(imm) | RD(tmp), ctx);
+			emit(MULX | IMMED | RS1(tmp) | S13(imm) | RD(tmp), ctx);
+			emit(SUB | RS1(dst) | RS2(tmp) | RD(dst), ctx);
+		} else {
+			const u8 tmp1 = bpf2sparc[TMP_REG_1];
+
+			ctx->tmp_1_used = true;
+
+			emit_set_const_sext(imm, tmp1, ctx);
+			emit(div | RS1(dst) | RS2(tmp1) | RD(tmp), ctx);
+			emit(MULX | RS1(tmp) | RS2(tmp1) | RD(tmp), ctx);
+			emit(SUB | RS1(dst) | RS2(tmp) | RD(dst), ctx);
+		}
+		goto do_alu32_trunc;
+	}
+	case BPF_ALU | BPF_LSH | BPF_K:
+		emit_alu_K(SLL, dst, imm, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU64 | BPF_LSH | BPF_K:
+		emit_alu_K(SLLX, dst, imm, ctx);
+		break;
+	case BPF_ALU | BPF_RSH | BPF_K:
+		emit_alu_K(SRL, dst, imm, ctx);
+		if (insn_is_zext(&insn[1]))
+			return 1;
+		break;
+	case BPF_ALU64 | BPF_RSH | BPF_K:
+		emit_alu_K(SRLX, dst, imm, ctx);
+		break;
+	case BPF_ALU | BPF_ARSH | BPF_K:
+		emit_alu_K(SRA, dst, imm, ctx);
+		goto do_alu32_trunc;
+	case BPF_ALU64 | BPF_ARSH | BPF_K:
+		emit_alu_K(SRAX, dst, imm, ctx);
+		break;
+
+	do_alu32_trunc:
+		if (BPF_CLASS(code) == BPF_ALU &&
+		    !ctx->prog->aux->verifier_zext)
+			emit_alu_K(SRL, dst, 0, ctx);
+		break;
+
+	/* JUMP off */
+	case BPF_JMP | BPF_JA:
+		emit_branch(BA, ctx->idx, ctx->offset[i + off], ctx);
+		emit_nop(ctx);
+		break;
+	/* IF (dst COND src) JUMP off */
+	case BPF_JMP | BPF_JEQ | BPF_X:
+	case BPF_JMP | BPF_JGT | BPF_X:
+	case BPF_JMP | BPF_JLT | BPF_X:
+	case BPF_JMP | BPF_JGE | BPF_X:
+	case BPF_JMP | BPF_JLE | BPF_X:
+	case BPF_JMP | BPF_JNE | BPF_X:
+	case BPF_JMP | BPF_JSGT | BPF_X:
+	case BPF_JMP | BPF_JSLT | BPF_X:
+	case BPF_JMP | BPF_JSGE | BPF_X:
+	case BPF_JMP | BPF_JSLE | BPF_X:
+	case BPF_JMP | BPF_JSET | BPF_X: {
+		int err;
+
+		err = emit_compare_and_branch(code, dst, src, 0, false, i + off, ctx);
+		if (err)
+			return err;
+		break;
+	}
+	/* IF (dst COND imm) JUMP off */
+	case BPF_JMP | BPF_JEQ | BPF_K:
+	case BPF_JMP | BPF_JGT | BPF_K:
+	case BPF_JMP | BPF_JLT | BPF_K:
+	case BPF_JMP | BPF_JGE | BPF_K:
+	case BPF_JMP | BPF_JLE | BPF_K:
+	case BPF_JMP | BPF_JNE | BPF_K:
+	case BPF_JMP | BPF_JSGT | BPF_K:
+	case BPF_JMP | BPF_JSLT | BPF_K:
+	case BPF_JMP | BPF_JSGE | BPF_K:
+	case BPF_JMP | BPF_JSLE | BPF_K:
+	case BPF_JMP | BPF_JSET | BPF_K: {
+		int err;
+
+		err = emit_compare_and_branch(code, dst, 0, imm, true, i + off, ctx);
+		if (err)
+			return err;
+		break;
+	}
+
+	/* function call */
+	case BPF_JMP | BPF_CALL:
+	{
+		u8 *func = ((u8 *)__bpf_call_base) + imm;
+
+		ctx->saw_call = true;
+
+		emit_call((u32 *)func, ctx);
+		emit_nop(ctx);
+
+		emit_reg_move(O0, bpf2sparc[BPF_REG_0], ctx);
+		break;
+	}
+
+	/* tail call */
+	case BPF_JMP | BPF_TAIL_CALL:
+		emit_tail_call(ctx);
+		break;
+
+	/* function return */
+	case BPF_JMP | BPF_EXIT:
+		/* Optimization: when last instruction is EXIT,
+		   simply fallthrough to epilogue. */
+		if (i == ctx->prog->len - 1)
+			break;
+		emit_branch(BA, ctx->idx, ctx->epilogue_offset, ctx);
+		emit_nop(ctx);
+		break;
+
+	/* dst = imm64 */
+	case BPF_LD | BPF_IMM | BPF_DW:
+	{
+		const struct bpf_insn insn1 = insn[1];
+		u64 imm64;
+
+		imm64 = (u64)insn1.imm << 32 | (u32)imm;
+		emit_loadimm64(imm64, dst, ctx);
+
+		return 1;
+	}
+
+	/* LDX: dst = *(size *)(src + off) */
+	case BPF_LDX | BPF_MEM | BPF_W:
+	case BPF_LDX | BPF_MEM | BPF_H:
+	case BPF_LDX | BPF_MEM | BPF_B:
+	case BPF_LDX | BPF_MEM | BPF_DW: {
+		const u8 tmp = bpf2sparc[TMP_REG_1];
+		u32 opcode = 0, rs2;
+
+		ctx->tmp_1_used = true;
+		switch (BPF_SIZE(code)) {
+		case BPF_W:
+			opcode = LD32;
+			break;
+		case BPF_H:
+			opcode = LD16;
+			break;
+		case BPF_B:
+			opcode = LD8;
+			break;
+		case BPF_DW:
+			opcode = LD64;
+			break;
+		}
+
+		if (is_simm13(off)) {
+			opcode |= IMMED;
+			rs2 = S13(off);
+		} else {
+			emit_loadimm(off, tmp, ctx);
+			rs2 = RS2(tmp);
+		}
+		emit(opcode | RS1(src) | rs2 | RD(dst), ctx);
+		if (opcode != LD64 && insn_is_zext(&insn[1]))
+			return 1;
+		break;
+	}
+	/* speculation barrier */
+	case BPF_ST | BPF_NOSPEC:
+		break;
+	/* ST: *(size *)(dst + off) = imm */
+	case BPF_ST | BPF_MEM | BPF_W:
+	case BPF_ST | BPF_MEM | BPF_H:
+	case BPF_ST | BPF_MEM | BPF_B:
+	case BPF_ST | BPF_MEM | BPF_DW: {
+		const u8 tmp = bpf2sparc[TMP_REG_1];
+		const u8 tmp2 = bpf2sparc[TMP_REG_2];
+		u32 opcode = 0, rs2;
+
+		if (insn->dst_reg == BPF_REG_FP)
+			ctx->saw_frame_pointer = true;
+
+		ctx->tmp_2_used = true;
+		emit_loadimm(imm, tmp2, ctx);
+
+		switch (BPF_SIZE(code)) {
+		case BPF_W:
+			opcode = ST32;
+			break;
+		case BPF_H:
+			opcode = ST16;
+			break;
+		case BPF_B:
+			opcode = ST8;
+			break;
+		case BPF_DW:
+			opcode = ST64;
+			break;
+		}
+
+		if (is_simm13(off)) {
+			opcode |= IMMED;
+			rs2 = S13(off);
+		} else {
+			ctx->tmp_1_used = true;
+			emit_loadimm(off, tmp, ctx);
+			rs2 = RS2(tmp);
+		}
+		emit(opcode | RS1(dst) | rs2 | RD(tmp2), ctx);
+		break;
+	}
+
+	/* STX: *(size *)(dst + off) = src */
+	case BPF_STX | BPF_MEM | BPF_W:
+	case BPF_STX | BPF_MEM | BPF_H:
+	case BPF_STX | BPF_MEM | BPF_B:
+	case BPF_STX | BPF_MEM | BPF_DW: {
+		const u8 tmp = bpf2sparc[TMP_REG_1];
+		u32 opcode = 0, rs2;
+
+		if (insn->dst_reg == BPF_REG_FP)
+			ctx->saw_frame_pointer = true;
+
+		switch (BPF_SIZE(code)) {
+		case BPF_W:
+			opcode = ST32;
+			break;
+		case BPF_H:
+			opcode = ST16;
+			break;
+		case BPF_B:
+			opcode = ST8;
+			break;
+		case BPF_DW:
+			opcode = ST64;
+			break;
+		}
+		if (is_simm13(off)) {
+			opcode |= IMMED;
+			rs2 = S13(off);
+		} else {
+			ctx->tmp_1_used = true;
+			emit_loadimm(off, tmp, ctx);
+			rs2 = RS2(tmp);
+		}
+		emit(opcode | RS1(dst) | rs2 | RD(src), ctx);
+		break;
+	}
+
+	case BPF_STX | BPF_ATOMIC | BPF_W: {
+		const u8 tmp = bpf2sparc[TMP_REG_1];
+		const u8 tmp2 = bpf2sparc[TMP_REG_2];
+		const u8 tmp3 = bpf2sparc[TMP_REG_3];
+
+		if (insn->imm != BPF_ADD) {
+			pr_err_once("unknown atomic op %02x\n", insn->imm);
+			return -EINVAL;
+		}
+
+		/* lock *(u32 *)(dst + off) += src */
+
+		if (insn->dst_reg == BPF_REG_FP)
+			ctx->saw_frame_pointer = true;
+
+		ctx->tmp_1_used = true;
+		ctx->tmp_2_used = true;
+		ctx->tmp_3_used = true;
+		emit_loadimm(off, tmp, ctx);
+		emit_alu3(ADD, dst, tmp, tmp, ctx);
+
+		emit(LD32 | RS1(tmp) | RS2(G0) | RD(tmp2), ctx);
+		emit_alu3(ADD, tmp2, src, tmp3, ctx);
+		emit(CAS | ASI(ASI_P) | RS1(tmp) | RS2(tmp2) | RD(tmp3), ctx);
+		emit_cmp(tmp2, tmp3, ctx);
+		emit_branch(BNE, 4, 0, ctx);
+		emit_nop(ctx);
+		break;
+	}
+	/* STX XADD: lock *(u64 *)(dst + off) += src */
+	case BPF_STX | BPF_ATOMIC | BPF_DW: {
+		const u8 tmp = bpf2sparc[TMP_REG_1];
+		const u8 tmp2 = bpf2sparc[TMP_REG_2];
+		const u8 tmp3 = bpf2sparc[TMP_REG_3];
+
+		if (insn->imm != BPF_ADD) {
+			pr_err_once("unknown atomic op %02x\n", insn->imm);
+			return -EINVAL;
+		}
+
+		if (insn->dst_reg == BPF_REG_FP)
+			ctx->saw_frame_pointer = true;
+
+		ctx->tmp_1_used = true;
+		ctx->tmp_2_used = true;
+		ctx->tmp_3_used = true;
+		emit_loadimm(off, tmp, ctx);
+		emit_alu3(ADD, dst, tmp, tmp, ctx);
+
+		emit(LD64 | RS1(tmp) | RS2(G0) | RD(tmp2), ctx);
+		emit_alu3(ADD, tmp2, src, tmp3, ctx);
+		emit(CASX | ASI(ASI_P) | RS1(tmp) | RS2(tmp2) | RD(tmp3), ctx);
+		emit_cmp(tmp2, tmp3, ctx);
+		emit_branch(BNE, 4, 0, ctx);
+		emit_nop(ctx);
+		break;
+	}
+
+	default:
+		pr_err_once("unknown opcode %02x\n", code);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int build_body(struct jit_ctx *ctx)
+{
+	const struct bpf_prog *prog = ctx->prog;
+	int i;
+
+	for (i = 0; i < prog->len; i++) {
+		const struct bpf_insn *insn = &prog->insnsi[i];
+		int ret;
+
+		ret = build_insn(insn, ctx);
+
+		if (ret > 0) {
+			i++;
+			ctx->offset[i] = ctx->idx;
+			continue;
+		}
+		ctx->offset[i] = ctx->idx;
+		if (ret)
+			return ret;
+	}
+	return 0;
+}
+
+static void jit_fill_hole(void *area, unsigned int size)
+{
+	u32 *ptr;
+	/* We are guaranteed to have aligned memory. */
+	for (ptr = area; size >= sizeof(u32); size -= sizeof(u32))
+		*ptr++ = 0x91d02005; /* ta 5 */
+}
+
+bool bpf_jit_needs_zext(void)
+{
+	return true;
+}
+
+struct sparc64_jit_data {
+	struct bpf_binary_header *header;
+	u8 *image;
+	struct jit_ctx ctx;
+};
+
+struct bpf_prog *bpf_int_jit_compile(struct bpf_verifier_env *env, struct bpf_prog *prog)
+{
+	struct sparc64_jit_data *jit_data;
+	struct bpf_binary_header *header;
+	u32 prev_image_size, image_size;
+	bool extra_pass = false;
+	struct jit_ctx ctx;
+	u8 *image_ptr;
+	int pass, i;
+
+	if (!prog->jit_requested)
+		return prog;
+
+	jit_data = prog->aux->jit_data;
+	if (!jit_data) {
+		jit_data = kzalloc_obj(*jit_data);
+		if (!jit_data)
+			return prog;
+		prog->aux->jit_data = jit_data;
+	}
+	if (jit_data->ctx.offset) {
+		ctx = jit_data->ctx;
+		image_ptr = jit_data->image;
+		header = jit_data->header;
+		extra_pass = true;
+		image_size = sizeof(u32) * ctx.idx;
+		prev_image_size = image_size;
+		pass = 1;
+		goto skip_init_ctx;
+	}
+
+	memset(&ctx, 0, sizeof(ctx));
+	ctx.prog = prog;
+
+	ctx.offset = kmalloc_array(prog->len, sizeof(unsigned int), GFP_KERNEL);
+	if (ctx.offset == NULL)
+		goto out_err;
+
+	/* Longest sequence emitted is for bswap32, 12 instructions.  Pre-cook
+	 * the offset array so that we converge faster.
+	 */
+	for (i = 0; i < prog->len; i++)
+		ctx.offset[i] = i * (12 * 4);
+
+	prev_image_size = ~0U;
+	for (pass = 1; pass < 40; pass++) {
+		ctx.idx = 0;
+
+		build_prologue(&ctx);
+		if (build_body(&ctx))
+			goto out_err;
+		build_epilogue(&ctx);
+
+		if (bpf_jit_enable > 1)
+			pr_info("Pass %d: size = %u, seen = [%c%c%c%c%c%c]\n", pass,
+				ctx.idx * 4,
+				ctx.tmp_1_used ? '1' : ' ',
+				ctx.tmp_2_used ? '2' : ' ',
+				ctx.tmp_3_used ? '3' : ' ',
+				ctx.saw_frame_pointer ? 'F' : ' ',
+				ctx.saw_call ? 'C' : ' ',
+				ctx.saw_tail_call ? 'T' : ' ');
+
+		if (ctx.idx * 4 == prev_image_size)
+			break;
+		prev_image_size = ctx.idx * 4;
+		cond_resched();
+	}
+
+	/* Now we know the actual image size. */
+	image_size = sizeof(u32) * ctx.idx;
+	header = bpf_jit_binary_alloc(image_size, &image_ptr,
+				      sizeof(u32), jit_fill_hole);
+	if (header == NULL)
+		goto out_err;
+
+	ctx.image = (u32 *)image_ptr;
+skip_init_ctx:
+	ctx.idx = 0;
+
+	build_prologue(&ctx);
+
+	if (build_body(&ctx)) {
+		bpf_jit_binary_free(header);
+		goto out_err;
+	}
+
+	build_epilogue(&ctx);
+
+	if (ctx.idx * 4 != prev_image_size) {
+		pr_err("bpf_jit: Failed to converge, prev_size=%u size=%d\n",
+		       prev_image_size, ctx.idx * 4);
+		bpf_jit_binary_free(header);
+		goto out_err;
+	}
+
+	if (bpf_jit_enable > 1)
+		bpf_jit_dump(prog->len, image_size, pass, ctx.image);
+
+	bpf_flush_icache(header, (u8 *)header + header->size);
+
+	if (!prog->is_func || extra_pass) {
+		if (bpf_jit_binary_lock_ro(header)) {
+			bpf_jit_binary_free(header);
+			goto out_err;
+		}
+	} else {
+		jit_data->ctx = ctx;
+		jit_data->image = image_ptr;
+		jit_data->header = header;
+	}
+
+	prog->bpf_func = (void *)ctx.image;
+	prog->jited = 1;
+	prog->jited_len = image_size;
+
+	if (!prog->is_func || extra_pass) {
+		bpf_prog_fill_jited_linfo(prog, ctx.offset);
+out_off:
+		kfree(ctx.offset);
+		kfree(jit_data);
+		prog->aux->jit_data = NULL;
+	}
+
+	return prog;
+
+out_err:
+	if (extra_pass) {
+		prog->bpf_func = NULL;
+		prog->jited = 0;
+		prog->jited_len = 0;
+	}
+	goto out_off;
+}