Merge tag 'io_uring-7.3-20261002' of git://git.kernel.org/pub/scm/linux/kernel/git/axboe/linuxgrafted

Pull io_uring fixes from Jens Axboe:

 - Fix a task_work add use-after-free with SQPOLL.

   The sqpoll thread could pop and complete the last request while
   io_req_normal_work_add() was still looking at them after the mpscq
   push.

   Use the same approach as DEFER_TASKRUN to protect from that, holding
   an RCU read lock across the add, and have exit wait for an RCU grace
   period for SQPOLL rings as well.

 - CQE32 ring fixes: correct the free entry check for 32b CQEs, zero the
   big_cqe for aux CQEs, and only post the dummy skip CQE on CQE_MIXED
   rings

 - Mark the source filter table as COW when cloning bpf filters, so
   registering another filter on the source doesn't modify the shared
   table in place

 - Initialize the task context before running the BPF loop

 - Requeue zcrx multishot receives stopped by a local resource

 - End a TX_TIMESTAMP multishot cmd when the CQ is full (lollipopkit)

* tag 'io_uring-7.3-20261002' of git://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux:
  io_uring: fix task_work add use-after-free with SQPOLL
  io_uring/cmd_net: end TX_TIMESTAMP multishot when the CQ is full
  io_uring/zcrx: requeue multishot receives stopped by a local resource
  io_uring: initialize task context before running the BPF loop
  io_uring: zero big_cqe for aux CQEs on CQE32 rings
  io_uring: fix free entry check for 32b CQEs on CQE32 rings
  io_uring: only post the dummy skip CQE on CQE_MIXED rings
  io_uring/bpf_filter: mark source as COW when cloning filters

3 files changed, 934 insertions, 0 deletions

diff --git a/lib/raid/raid6/loongarch/loongarch_simd.c b/lib/raid/raid6/loongarch/loongarch_simd.c
new file mode 100644
index 000000000..c1eb53faf
--- /dev/null
+++ b/lib/raid/raid6/loongarch/loongarch_simd.c
@@ -0,0 +1,409 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * RAID6 syndrome calculations in LoongArch SIMD (LSX & LASX)
+ *
+ * Copyright 2023 WANG Xuerui <git@xen0n.name>
+ *
+ * Based on the generic RAID-6 code (int.uc):
+ *
+ * Copyright 2002-2004 H. Peter Anvin
+ */
+
+#include <asm/cpu-features.h>
+#include <asm/fpu.h>
+#include "algos.h"
+
+/*
+ * The vector algorithms are currently priority 0, which means the generic
+ * scalar algorithms are not being disabled if vector support is present.
+ * This is like the similar LoongArch RAID5 XOR code, with the main reason
+ * repeated here: it cannot be ruled out at this point of time, that some
+ * future (maybe reduced) models could run the vector algorithms slower than
+ * the scalar ones, maybe for errata or micro-op reasons. It may be
+ * appropriate to revisit this after one or two more uarch generations.
+ */
+
+#ifdef CONFIG_CPU_HAS_LSX
+#define NSIZE 16
+
+static void raid6_lsx_gen_syndrome(int disks, size_t bytes, void **ptrs)
+{
+	u8 **dptr = (u8 **)ptrs;
+	u8 *p, *q;
+	int d, z, z0;
+
+	z0 = disks - 3;		/* Highest data disk */
+	p = dptr[z0+1];		/* XOR parity */
+	q = dptr[z0+2];		/* RS syndrome */
+
+	kernel_fpu_begin();
+
+	/*
+	 * $vr0, $vr1, $vr2, $vr3: wp
+	 * $vr4, $vr5, $vr6, $vr7: wq
+	 * $vr8, $vr9, $vr10, $vr11: wd
+	 * $vr12, $vr13, $vr14, $vr15: w2
+	 * $vr16, $vr17, $vr18, $vr19: w1
+	 */
+	for (d = 0; d < bytes; d += NSIZE*4) {
+		/* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
+		asm volatile("vld $vr0, %0" : : "m"(dptr[z0][d+0*NSIZE]));
+		asm volatile("vld $vr1, %0" : : "m"(dptr[z0][d+1*NSIZE]));
+		asm volatile("vld $vr2, %0" : : "m"(dptr[z0][d+2*NSIZE]));
+		asm volatile("vld $vr3, %0" : : "m"(dptr[z0][d+3*NSIZE]));
+		asm volatile("vori.b $vr4, $vr0, 0");
+		asm volatile("vori.b $vr5, $vr1, 0");
+		asm volatile("vori.b $vr6, $vr2, 0");
+		asm volatile("vori.b $vr7, $vr3, 0");
+		for (z = z0-1; z >= 0; z--) {
+			/* wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE]; */
+			asm volatile("vld $vr8, %0" : : "m"(dptr[z][d+0*NSIZE]));
+			asm volatile("vld $vr9, %0" : : "m"(dptr[z][d+1*NSIZE]));
+			asm volatile("vld $vr10, %0" : : "m"(dptr[z][d+2*NSIZE]));
+			asm volatile("vld $vr11, %0" : : "m"(dptr[z][d+3*NSIZE]));
+			/* wp$$ ^= wd$$; */
+			asm volatile("vxor.v $vr0, $vr0, $vr8");
+			asm volatile("vxor.v $vr1, $vr1, $vr9");
+			asm volatile("vxor.v $vr2, $vr2, $vr10");
+			asm volatile("vxor.v $vr3, $vr3, $vr11");
+			/* w2$$ = MASK(wq$$); */
+			asm volatile("vslti.b $vr12, $vr4, 0");
+			asm volatile("vslti.b $vr13, $vr5, 0");
+			asm volatile("vslti.b $vr14, $vr6, 0");
+			asm volatile("vslti.b $vr15, $vr7, 0");
+			/* w1$$ = SHLBYTE(wq$$); */
+			asm volatile("vslli.b $vr16, $vr4, 1");
+			asm volatile("vslli.b $vr17, $vr5, 1");
+			asm volatile("vslli.b $vr18, $vr6, 1");
+			asm volatile("vslli.b $vr19, $vr7, 1");
+			/* w2$$ &= NBYTES(0x1d); */
+			asm volatile("vandi.b $vr12, $vr12, 0x1d");
+			asm volatile("vandi.b $vr13, $vr13, 0x1d");
+			asm volatile("vandi.b $vr14, $vr14, 0x1d");
+			asm volatile("vandi.b $vr15, $vr15, 0x1d");
+			/* w1$$ ^= w2$$; */
+			asm volatile("vxor.v $vr16, $vr16, $vr12");
+			asm volatile("vxor.v $vr17, $vr17, $vr13");
+			asm volatile("vxor.v $vr18, $vr18, $vr14");
+			asm volatile("vxor.v $vr19, $vr19, $vr15");
+			/* wq$$ = w1$$ ^ wd$$; */
+			asm volatile("vxor.v $vr4, $vr16, $vr8");
+			asm volatile("vxor.v $vr5, $vr17, $vr9");
+			asm volatile("vxor.v $vr6, $vr18, $vr10");
+			asm volatile("vxor.v $vr7, $vr19, $vr11");
+		}
+		/* *(unative_t *)&p[d+NSIZE*$$] = wp$$; */
+		asm volatile("vst $vr0, %0" : "=m"(p[d+NSIZE*0]));
+		asm volatile("vst $vr1, %0" : "=m"(p[d+NSIZE*1]));
+		asm volatile("vst $vr2, %0" : "=m"(p[d+NSIZE*2]));
+		asm volatile("vst $vr3, %0" : "=m"(p[d+NSIZE*3]));
+		/* *(unative_t *)&q[d+NSIZE*$$] = wq$$; */
+		asm volatile("vst $vr4, %0" : "=m"(q[d+NSIZE*0]));
+		asm volatile("vst $vr5, %0" : "=m"(q[d+NSIZE*1]));
+		asm volatile("vst $vr6, %0" : "=m"(q[d+NSIZE*2]));
+		asm volatile("vst $vr7, %0" : "=m"(q[d+NSIZE*3]));
+	}
+
+	kernel_fpu_end();
+}
+
+static void raid6_lsx_xor_syndrome(int disks, int start, int stop,
+				   size_t bytes, void **ptrs)
+{
+	u8 **dptr = (u8 **)ptrs;
+	u8 *p, *q;
+	int d, z, z0;
+
+	z0 = stop;		/* P/Q right side optimization */
+	p = dptr[disks-2];	/* XOR parity */
+	q = dptr[disks-1];	/* RS syndrome */
+
+	kernel_fpu_begin();
+
+	/*
+	 * $vr0, $vr1, $vr2, $vr3: wp
+	 * $vr4, $vr5, $vr6, $vr7: wq
+	 * $vr8, $vr9, $vr10, $vr11: wd
+	 * $vr12, $vr13, $vr14, $vr15: w2
+	 * $vr16, $vr17, $vr18, $vr19: w1
+	 */
+	for (d = 0; d < bytes; d += NSIZE*4) {
+		/* P/Q data pages */
+		/* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
+		asm volatile("vld $vr0, %0" : : "m"(dptr[z0][d+0*NSIZE]));
+		asm volatile("vld $vr1, %0" : : "m"(dptr[z0][d+1*NSIZE]));
+		asm volatile("vld $vr2, %0" : : "m"(dptr[z0][d+2*NSIZE]));
+		asm volatile("vld $vr3, %0" : : "m"(dptr[z0][d+3*NSIZE]));
+		asm volatile("vori.b $vr4, $vr0, 0");
+		asm volatile("vori.b $vr5, $vr1, 0");
+		asm volatile("vori.b $vr6, $vr2, 0");
+		asm volatile("vori.b $vr7, $vr3, 0");
+		for (z = z0-1; z >= start; z--) {
+			/* wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE]; */
+			asm volatile("vld $vr8, %0" : : "m"(dptr[z][d+0*NSIZE]));
+			asm volatile("vld $vr9, %0" : : "m"(dptr[z][d+1*NSIZE]));
+			asm volatile("vld $vr10, %0" : : "m"(dptr[z][d+2*NSIZE]));
+			asm volatile("vld $vr11, %0" : : "m"(dptr[z][d+3*NSIZE]));
+			/* wp$$ ^= wd$$; */
+			asm volatile("vxor.v $vr0, $vr0, $vr8");
+			asm volatile("vxor.v $vr1, $vr1, $vr9");
+			asm volatile("vxor.v $vr2, $vr2, $vr10");
+			asm volatile("vxor.v $vr3, $vr3, $vr11");
+			/* w2$$ = MASK(wq$$); */
+			asm volatile("vslti.b $vr12, $vr4, 0");
+			asm volatile("vslti.b $vr13, $vr5, 0");
+			asm volatile("vslti.b $vr14, $vr6, 0");
+			asm volatile("vslti.b $vr15, $vr7, 0");
+			/* w1$$ = SHLBYTE(wq$$); */
+			asm volatile("vslli.b $vr16, $vr4, 1");
+			asm volatile("vslli.b $vr17, $vr5, 1");
+			asm volatile("vslli.b $vr18, $vr6, 1");
+			asm volatile("vslli.b $vr19, $vr7, 1");
+			/* w2$$ &= NBYTES(0x1d); */
+			asm volatile("vandi.b $vr12, $vr12, 0x1d");
+			asm volatile("vandi.b $vr13, $vr13, 0x1d");
+			asm volatile("vandi.b $vr14, $vr14, 0x1d");
+			asm volatile("vandi.b $vr15, $vr15, 0x1d");
+			/* w1$$ ^= w2$$; */
+			asm volatile("vxor.v $vr16, $vr16, $vr12");
+			asm volatile("vxor.v $vr17, $vr17, $vr13");
+			asm volatile("vxor.v $vr18, $vr18, $vr14");
+			asm volatile("vxor.v $vr19, $vr19, $vr15");
+			/* wq$$ = w1$$ ^ wd$$; */
+			asm volatile("vxor.v $vr4, $vr16, $vr8");
+			asm volatile("vxor.v $vr5, $vr17, $vr9");
+			asm volatile("vxor.v $vr6, $vr18, $vr10");
+			asm volatile("vxor.v $vr7, $vr19, $vr11");
+		}
+
+		/* P/Q left side optimization */
+		for (z = start-1; z >= 0; z--) {
+			/* w2$$ = MASK(wq$$); */
+			asm volatile("vslti.b $vr12, $vr4, 0");
+			asm volatile("vslti.b $vr13, $vr5, 0");
+			asm volatile("vslti.b $vr14, $vr6, 0");
+			asm volatile("vslti.b $vr15, $vr7, 0");
+			/* w1$$ = SHLBYTE(wq$$); */
+			asm volatile("vslli.b $vr16, $vr4, 1");
+			asm volatile("vslli.b $vr17, $vr5, 1");
+			asm volatile("vslli.b $vr18, $vr6, 1");
+			asm volatile("vslli.b $vr19, $vr7, 1");
+			/* w2$$ &= NBYTES(0x1d); */
+			asm volatile("vandi.b $vr12, $vr12, 0x1d");
+			asm volatile("vandi.b $vr13, $vr13, 0x1d");
+			asm volatile("vandi.b $vr14, $vr14, 0x1d");
+			asm volatile("vandi.b $vr15, $vr15, 0x1d");
+			/* wq$$ = w1$$ ^ w2$$; */
+			asm volatile("vxor.v $vr4, $vr16, $vr12");
+			asm volatile("vxor.v $vr5, $vr17, $vr13");
+			asm volatile("vxor.v $vr6, $vr18, $vr14");
+			asm volatile("vxor.v $vr7, $vr19, $vr15");
+		}
+		/*
+		 * *(unative_t *)&p[d+NSIZE*$$] ^= wp$$;
+		 * *(unative_t *)&q[d+NSIZE*$$] ^= wq$$;
+		 */
+		asm volatile(
+			"vld $vr20, %0\n\t"
+			"vld $vr21, %1\n\t"
+			"vld $vr22, %2\n\t"
+			"vld $vr23, %3\n\t"
+			"vld $vr24, %4\n\t"
+			"vld $vr25, %5\n\t"
+			"vld $vr26, %6\n\t"
+			"vld $vr27, %7\n\t"
+			"vxor.v $vr20, $vr20, $vr0\n\t"
+			"vxor.v $vr21, $vr21, $vr1\n\t"
+			"vxor.v $vr22, $vr22, $vr2\n\t"
+			"vxor.v $vr23, $vr23, $vr3\n\t"
+			"vxor.v $vr24, $vr24, $vr4\n\t"
+			"vxor.v $vr25, $vr25, $vr5\n\t"
+			"vxor.v $vr26, $vr26, $vr6\n\t"
+			"vxor.v $vr27, $vr27, $vr7\n\t"
+			"vst $vr20, %0\n\t"
+			"vst $vr21, %1\n\t"
+			"vst $vr22, %2\n\t"
+			"vst $vr23, %3\n\t"
+			"vst $vr24, %4\n\t"
+			"vst $vr25, %5\n\t"
+			"vst $vr26, %6\n\t"
+			"vst $vr27, %7\n\t"
+			: "+m"(p[d+NSIZE*0]), "+m"(p[d+NSIZE*1]),
+			  "+m"(p[d+NSIZE*2]), "+m"(p[d+NSIZE*3]),
+			  "+m"(q[d+NSIZE*0]), "+m"(q[d+NSIZE*1]),
+			  "+m"(q[d+NSIZE*2]), "+m"(q[d+NSIZE*3])
+		);
+	}
+
+	kernel_fpu_end();
+}
+
+const struct raid6_calls raid6_lsx = {
+	.gen_syndrome	= raid6_lsx_gen_syndrome,
+	.xor_syndrome	= raid6_lsx_xor_syndrome,
+	.name		= "lsx",
+};
+
+#undef NSIZE
+#endif /* CONFIG_CPU_HAS_LSX */
+
+#ifdef CONFIG_CPU_HAS_LASX
+#define NSIZE 32
+
+static void raid6_lasx_gen_syndrome(int disks, size_t bytes, void **ptrs)
+{
+	u8 **dptr = (u8 **)ptrs;
+	u8 *p, *q;
+	int d, z, z0;
+
+	z0 = disks - 3;		/* Highest data disk */
+	p = dptr[z0+1];		/* XOR parity */
+	q = dptr[z0+2];		/* RS syndrome */
+
+	kernel_fpu_begin();
+
+	/*
+	 * $xr0, $xr1: wp
+	 * $xr2, $xr3: wq
+	 * $xr4, $xr5: wd
+	 * $xr6, $xr7: w2
+	 * $xr8, $xr9: w1
+	 */
+	for (d = 0; d < bytes; d += NSIZE*2) {
+		/* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
+		asm volatile("xvld $xr0, %0" : : "m"(dptr[z0][d+0*NSIZE]));
+		asm volatile("xvld $xr1, %0" : : "m"(dptr[z0][d+1*NSIZE]));
+		asm volatile("xvori.b $xr2, $xr0, 0");
+		asm volatile("xvori.b $xr3, $xr1, 0");
+		for (z = z0-1; z >= 0; z--) {
+			/* wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE]; */
+			asm volatile("xvld $xr4, %0" : : "m"(dptr[z][d+0*NSIZE]));
+			asm volatile("xvld $xr5, %0" : : "m"(dptr[z][d+1*NSIZE]));
+			/* wp$$ ^= wd$$; */
+			asm volatile("xvxor.v $xr0, $xr0, $xr4");
+			asm volatile("xvxor.v $xr1, $xr1, $xr5");
+			/* w2$$ = MASK(wq$$); */
+			asm volatile("xvslti.b $xr6, $xr2, 0");
+			asm volatile("xvslti.b $xr7, $xr3, 0");
+			/* w1$$ = SHLBYTE(wq$$); */
+			asm volatile("xvslli.b $xr8, $xr2, 1");
+			asm volatile("xvslli.b $xr9, $xr3, 1");
+			/* w2$$ &= NBYTES(0x1d); */
+			asm volatile("xvandi.b $xr6, $xr6, 0x1d");
+			asm volatile("xvandi.b $xr7, $xr7, 0x1d");
+			/* w1$$ ^= w2$$; */
+			asm volatile("xvxor.v $xr8, $xr8, $xr6");
+			asm volatile("xvxor.v $xr9, $xr9, $xr7");
+			/* wq$$ = w1$$ ^ wd$$; */
+			asm volatile("xvxor.v $xr2, $xr8, $xr4");
+			asm volatile("xvxor.v $xr3, $xr9, $xr5");
+		}
+		/* *(unative_t *)&p[d+NSIZE*$$] = wp$$; */
+		asm volatile("xvst $xr0, %0" : "=m"(p[d+NSIZE*0]));
+		asm volatile("xvst $xr1, %0" : "=m"(p[d+NSIZE*1]));
+		/* *(unative_t *)&q[d+NSIZE*$$] = wq$$; */
+		asm volatile("xvst $xr2, %0" : "=m"(q[d+NSIZE*0]));
+		asm volatile("xvst $xr3, %0" : "=m"(q[d+NSIZE*1]));
+	}
+
+	kernel_fpu_end();
+}
+
+static void raid6_lasx_xor_syndrome(int disks, int start, int stop,
+				    size_t bytes, void **ptrs)
+{
+	u8 **dptr = (u8 **)ptrs;
+	u8 *p, *q;
+	int d, z, z0;
+
+	z0 = stop;		/* P/Q right side optimization */
+	p = dptr[disks-2];	/* XOR parity */
+	q = dptr[disks-1];	/* RS syndrome */
+
+	kernel_fpu_begin();
+
+	/*
+	 * $xr0, $xr1: wp
+	 * $xr2, $xr3: wq
+	 * $xr4, $xr5: wd
+	 * $xr6, $xr7: w2
+	 * $xr8, $xr9: w1
+	 */
+	for (d = 0; d < bytes; d += NSIZE*2) {
+		/* P/Q data pages */
+		/* wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE]; */
+		asm volatile("xvld $xr0, %0" : : "m"(dptr[z0][d+0*NSIZE]));
+		asm volatile("xvld $xr1, %0" : : "m"(dptr[z0][d+1*NSIZE]));
+		asm volatile("xvori.b $xr2, $xr0, 0");
+		asm volatile("xvori.b $xr3, $xr1, 0");
+		for (z = z0-1; z >= start; z--) {
+			/* wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE]; */
+			asm volatile("xvld $xr4, %0" : : "m"(dptr[z][d+0*NSIZE]));
+			asm volatile("xvld $xr5, %0" : : "m"(dptr[z][d+1*NSIZE]));
+			/* wp$$ ^= wd$$; */
+			asm volatile("xvxor.v $xr0, $xr0, $xr4");
+			asm volatile("xvxor.v $xr1, $xr1, $xr5");
+			/* w2$$ = MASK(wq$$); */
+			asm volatile("xvslti.b $xr6, $xr2, 0");
+			asm volatile("xvslti.b $xr7, $xr3, 0");
+			/* w1$$ = SHLBYTE(wq$$); */
+			asm volatile("xvslli.b $xr8, $xr2, 1");
+			asm volatile("xvslli.b $xr9, $xr3, 1");
+			/* w2$$ &= NBYTES(0x1d); */
+			asm volatile("xvandi.b $xr6, $xr6, 0x1d");
+			asm volatile("xvandi.b $xr7, $xr7, 0x1d");
+			/* w1$$ ^= w2$$; */
+			asm volatile("xvxor.v $xr8, $xr8, $xr6");
+			asm volatile("xvxor.v $xr9, $xr9, $xr7");
+			/* wq$$ = w1$$ ^ wd$$; */
+			asm volatile("xvxor.v $xr2, $xr8, $xr4");
+			asm volatile("xvxor.v $xr3, $xr9, $xr5");
+		}
+
+		/* P/Q left side optimization */
+		for (z = start-1; z >= 0; z--) {
+			/* w2$$ = MASK(wq$$); */
+			asm volatile("xvslti.b $xr6, $xr2, 0");
+			asm volatile("xvslti.b $xr7, $xr3, 0");
+			/* w1$$ = SHLBYTE(wq$$); */
+			asm volatile("xvslli.b $xr8, $xr2, 1");
+			asm volatile("xvslli.b $xr9, $xr3, 1");
+			/* w2$$ &= NBYTES(0x1d); */
+			asm volatile("xvandi.b $xr6, $xr6, 0x1d");
+			asm volatile("xvandi.b $xr7, $xr7, 0x1d");
+			/* wq$$ = w1$$ ^ w2$$; */
+			asm volatile("xvxor.v $xr2, $xr8, $xr6");
+			asm volatile("xvxor.v $xr3, $xr9, $xr7");
+		}
+		/*
+		 * *(unative_t *)&p[d+NSIZE*$$] ^= wp$$;
+		 * *(unative_t *)&q[d+NSIZE*$$] ^= wq$$;
+		 */
+		asm volatile(
+			"xvld $xr10, %0\n\t"
+			"xvld $xr11, %1\n\t"
+			"xvld $xr12, %2\n\t"
+			"xvld $xr13, %3\n\t"
+			"xvxor.v $xr10, $xr10, $xr0\n\t"
+			"xvxor.v $xr11, $xr11, $xr1\n\t"
+			"xvxor.v $xr12, $xr12, $xr2\n\t"
+			"xvxor.v $xr13, $xr13, $xr3\n\t"
+			"xvst $xr10, %0\n\t"
+			"xvst $xr11, %1\n\t"
+			"xvst $xr12, %2\n\t"
+			"xvst $xr13, %3\n\t"
+			: "+m"(p[d+NSIZE*0]), "+m"(p[d+NSIZE*1]),
+			  "+m"(q[d+NSIZE*0]), "+m"(q[d+NSIZE*1])
+		);
+	}
+
+	kernel_fpu_end();
+}
+
+const struct raid6_calls raid6_lasx = {
+	.gen_syndrome	= raid6_lasx_gen_syndrome,
+	.xor_syndrome	= raid6_lasx_xor_syndrome,
+	.name		= "lasx",
+};
+#undef NSIZE
+#endif /* CONFIG_CPU_HAS_LASX */
diff --git a/lib/raid/raid6/loongarch/pq_arch.h b/lib/raid/raid6/loongarch/pq_arch.h
new file mode 100644
index 000000000..ae443a4d7
--- /dev/null
+++ b/lib/raid/raid6/loongarch/pq_arch.h
@@ -0,0 +1,23 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+
+#include <asm/cpu-features.h>
+
+extern const struct raid6_calls raid6_lsx;
+extern const struct raid6_calls raid6_lasx;
+
+extern const struct raid6_recov_calls raid6_recov_lsx;
+extern const struct raid6_recov_calls raid6_recov_lasx;
+
+static __always_inline void __init arch_raid6_init(void)
+{
+	raid6_algo_add_default();
+	if (IS_ENABLED(CONFIG_CPU_HAS_LSX) && cpu_has_lsx)
+		raid6_algo_add(&raid6_lsx);
+	if (IS_ENABLED(CONFIG_CPU_HAS_LASX) && cpu_has_lasx)
+		raid6_algo_add(&raid6_lasx);
+
+	if (IS_ENABLED(CONFIG_CPU_HAS_LASX) && cpu_has_lasx)
+		raid6_recov_algo_add(&raid6_recov_lasx);
+	else if (IS_ENABLED(CONFIG_CPU_HAS_LSX) && cpu_has_lsx)
+		raid6_recov_algo_add(&raid6_recov_lsx);
+}
diff --git a/lib/raid/raid6/loongarch/recov_loongarch_simd.c b/lib/raid/raid6/loongarch/recov_loongarch_simd.c
new file mode 100644
index 000000000..87a2313bb
--- /dev/null
+++ b/lib/raid/raid6/loongarch/recov_loongarch_simd.c
@@ -0,0 +1,502 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * RAID6 recovery algorithms in LoongArch SIMD (LSX & LASX)
+ *
+ * Copyright (C) 2023 WANG Xuerui <git@xen0n.name>
+ *
+ * Originally based on recov_avx2.c and recov_ssse3.c:
+ *
+ * Copyright (C) 2012 Intel Corporation
+ * Author: Jim Kukunas <james.t.kukunas@linux.intel.com>
+ */
+
+#include <linux/mm.h>
+#include <linux/raid/pq.h>
+#include <asm/cpu-features.h>
+#include <asm/fpu.h>
+#include "algos.h"
+
+/*
+ * Unlike with the syndrome calculation algorithms, there's no boot-time
+ * selection of recovery algorithms by benchmarking, so we have to specify
+ * the priorities and hope the future cores will all have decent vector
+ * support (i.e. no LASX slower than LSX, or even scalar code).
+ */
+
+#ifdef CONFIG_CPU_HAS_LSX
+static void raid6_2data_recov_lsx(int disks, size_t bytes, int faila,
+				  int failb, void **ptrs)
+{
+	u8 *p, *q, *dp, *dq;
+	const u8 *pbmul;	/* P multiplier table for B data */
+	const u8 *qmul;		/* Q multiplier table (for both) */
+
+	p = (u8 *)ptrs[disks - 2];
+	q = (u8 *)ptrs[disks - 1];
+
+	/*
+	 * Compute syndrome with zero for the missing data pages
+	 * Use the dead data pages as temporary storage for
+	 * delta p and delta q
+	 */
+	dp = (u8 *)ptrs[faila];
+	ptrs[faila] = page_address(ZERO_PAGE(0));
+	ptrs[disks - 2] = dp;
+	dq = (u8 *)ptrs[failb];
+	ptrs[failb] = page_address(ZERO_PAGE(0));
+	ptrs[disks - 1] = dq;
+
+	raid6_gen_syndrome(disks, bytes, ptrs);
+
+	/* Restore pointer table */
+	ptrs[faila] = dp;
+	ptrs[failb] = dq;
+	ptrs[disks - 2] = p;
+	ptrs[disks - 1] = q;
+
+	/* Now, pick the proper data tables */
+	pbmul = raid6_vgfmul[raid6_gfexi[failb - faila]];
+	qmul  = raid6_vgfmul[raid6_gfinv[raid6_gfexp[faila] ^ raid6_gfexp[failb]]];
+
+	kernel_fpu_begin();
+
+	/*
+	 * vr20, vr21: qmul
+	 * vr22, vr23: pbmul
+	 */
+	asm volatile("vld $vr20, %0" : : "m" (qmul[0]));
+	asm volatile("vld $vr21, %0" : : "m" (qmul[16]));
+	asm volatile("vld $vr22, %0" : : "m" (pbmul[0]));
+	asm volatile("vld $vr23, %0" : : "m" (pbmul[16]));
+
+	while (bytes) {
+		/* vr4 - vr7: Q */
+		asm volatile("vld $vr4, %0" : : "m" (q[0]));
+		asm volatile("vld $vr5, %0" : : "m" (q[16]));
+		asm volatile("vld $vr6, %0" : : "m" (q[32]));
+		asm volatile("vld $vr7, %0" : : "m" (q[48]));
+		/*  vr4 - vr7: Q + Qxy */
+		asm volatile("vld $vr8, %0" : : "m" (dq[0]));
+		asm volatile("vld $vr9, %0" : : "m" (dq[16]));
+		asm volatile("vld $vr10, %0" : : "m" (dq[32]));
+		asm volatile("vld $vr11, %0" : : "m" (dq[48]));
+		asm volatile("vxor.v $vr4, $vr4, $vr8");
+		asm volatile("vxor.v $vr5, $vr5, $vr9");
+		asm volatile("vxor.v $vr6, $vr6, $vr10");
+		asm volatile("vxor.v $vr7, $vr7, $vr11");
+		/* vr0 - vr3: P */
+		asm volatile("vld $vr0, %0" : : "m" (p[0]));
+		asm volatile("vld $vr1, %0" : : "m" (p[16]));
+		asm volatile("vld $vr2, %0" : : "m" (p[32]));
+		asm volatile("vld $vr3, %0" : : "m" (p[48]));
+		/* vr0 - vr3: P + Pxy */
+		asm volatile("vld $vr8, %0" : : "m" (dp[0]));
+		asm volatile("vld $vr9, %0" : : "m" (dp[16]));
+		asm volatile("vld $vr10, %0" : : "m" (dp[32]));
+		asm volatile("vld $vr11, %0" : : "m" (dp[48]));
+		asm volatile("vxor.v $vr0, $vr0, $vr8");
+		asm volatile("vxor.v $vr1, $vr1, $vr9");
+		asm volatile("vxor.v $vr2, $vr2, $vr10");
+		asm volatile("vxor.v $vr3, $vr3, $vr11");
+
+		/* vr8 - vr11: higher 4 bits of each byte of (Q + Qxy) */
+		asm volatile("vsrli.b $vr8, $vr4, 4");
+		asm volatile("vsrli.b $vr9, $vr5, 4");
+		asm volatile("vsrli.b $vr10, $vr6, 4");
+		asm volatile("vsrli.b $vr11, $vr7, 4");
+		/* vr4 - vr7: lower 4 bits of each byte of (Q + Qxy) */
+		asm volatile("vandi.b $vr4, $vr4, 0x0f");
+		asm volatile("vandi.b $vr5, $vr5, 0x0f");
+		asm volatile("vandi.b $vr6, $vr6, 0x0f");
+		asm volatile("vandi.b $vr7, $vr7, 0x0f");
+		/* lookup from qmul[0] */
+		asm volatile("vshuf.b $vr4, $vr20, $vr20, $vr4");
+		asm volatile("vshuf.b $vr5, $vr20, $vr20, $vr5");
+		asm volatile("vshuf.b $vr6, $vr20, $vr20, $vr6");
+		asm volatile("vshuf.b $vr7, $vr20, $vr20, $vr7");
+		/* lookup from qmul[16] */
+		asm volatile("vshuf.b $vr8, $vr21, $vr21, $vr8");
+		asm volatile("vshuf.b $vr9, $vr21, $vr21, $vr9");
+		asm volatile("vshuf.b $vr10, $vr21, $vr21, $vr10");
+		asm volatile("vshuf.b $vr11, $vr21, $vr21, $vr11");
+		/* vr16 - vr19: B(Q + Qxy) */
+		asm volatile("vxor.v $vr16, $vr8, $vr4");
+		asm volatile("vxor.v $vr17, $vr9, $vr5");
+		asm volatile("vxor.v $vr18, $vr10, $vr6");
+		asm volatile("vxor.v $vr19, $vr11, $vr7");
+
+		/* vr4 - vr7: higher 4 bits of each byte of (P + Pxy) */
+		asm volatile("vsrli.b $vr4, $vr0, 4");
+		asm volatile("vsrli.b $vr5, $vr1, 4");
+		asm volatile("vsrli.b $vr6, $vr2, 4");
+		asm volatile("vsrli.b $vr7, $vr3, 4");
+		/* vr12 - vr15: lower 4 bits of each byte of (P + Pxy) */
+		asm volatile("vandi.b $vr12, $vr0, 0x0f");
+		asm volatile("vandi.b $vr13, $vr1, 0x0f");
+		asm volatile("vandi.b $vr14, $vr2, 0x0f");
+		asm volatile("vandi.b $vr15, $vr3, 0x0f");
+		/* lookup from pbmul[0] */
+		asm volatile("vshuf.b $vr12, $vr22, $vr22, $vr12");
+		asm volatile("vshuf.b $vr13, $vr22, $vr22, $vr13");
+		asm volatile("vshuf.b $vr14, $vr22, $vr22, $vr14");
+		asm volatile("vshuf.b $vr15, $vr22, $vr22, $vr15");
+		/* lookup from pbmul[16] */
+		asm volatile("vshuf.b $vr4, $vr23, $vr23, $vr4");
+		asm volatile("vshuf.b $vr5, $vr23, $vr23, $vr5");
+		asm volatile("vshuf.b $vr6, $vr23, $vr23, $vr6");
+		asm volatile("vshuf.b $vr7, $vr23, $vr23, $vr7");
+		/* vr4 - vr7: A(P + Pxy) */
+		asm volatile("vxor.v $vr4, $vr4, $vr12");
+		asm volatile("vxor.v $vr5, $vr5, $vr13");
+		asm volatile("vxor.v $vr6, $vr6, $vr14");
+		asm volatile("vxor.v $vr7, $vr7, $vr15");
+
+		/* vr4 - vr7: A(P + Pxy) + B(Q + Qxy) = Dx */
+		asm volatile("vxor.v $vr4, $vr4, $vr16");
+		asm volatile("vxor.v $vr5, $vr5, $vr17");
+		asm volatile("vxor.v $vr6, $vr6, $vr18");
+		asm volatile("vxor.v $vr7, $vr7, $vr19");
+		asm volatile("vst $vr4, %0" : "=m" (dq[0]));
+		asm volatile("vst $vr5, %0" : "=m" (dq[16]));
+		asm volatile("vst $vr6, %0" : "=m" (dq[32]));
+		asm volatile("vst $vr7, %0" : "=m" (dq[48]));
+
+		/* vr0 - vr3: P + Pxy + Dx = Dy */
+		asm volatile("vxor.v $vr0, $vr0, $vr4");
+		asm volatile("vxor.v $vr1, $vr1, $vr5");
+		asm volatile("vxor.v $vr2, $vr2, $vr6");
+		asm volatile("vxor.v $vr3, $vr3, $vr7");
+		asm volatile("vst $vr0, %0" : "=m" (dp[0]));
+		asm volatile("vst $vr1, %0" : "=m" (dp[16]));
+		asm volatile("vst $vr2, %0" : "=m" (dp[32]));
+		asm volatile("vst $vr3, %0" : "=m" (dp[48]));
+
+		bytes -= 64;
+		p += 64;
+		q += 64;
+		dp += 64;
+		dq += 64;
+	}
+
+	kernel_fpu_end();
+}
+
+static void raid6_datap_recov_lsx(int disks, size_t bytes, int faila,
+				  void **ptrs)
+{
+	u8 *p, *q, *dq;
+	const u8 *qmul;		/* Q multiplier table */
+
+	p = (u8 *)ptrs[disks - 2];
+	q = (u8 *)ptrs[disks - 1];
+
+	/*
+	 * Compute syndrome with zero for the missing data page
+	 * Use the dead data page as temporary storage for delta q
+	 */
+	dq = (u8 *)ptrs[faila];
+	ptrs[faila] = page_address(ZERO_PAGE(0));
+	ptrs[disks - 1] = dq;
+
+	raid6_gen_syndrome(disks, bytes, ptrs);
+
+	/* Restore pointer table */
+	ptrs[faila] = dq;
+	ptrs[disks - 1] = q;
+
+	/* Now, pick the proper data tables */
+	qmul  = raid6_vgfmul[raid6_gfinv[raid6_gfexp[faila]]];
+
+	kernel_fpu_begin();
+
+	/* vr22, vr23: qmul */
+	asm volatile("vld $vr22, %0" : : "m" (qmul[0]));
+	asm volatile("vld $vr23, %0" : : "m" (qmul[16]));
+
+	while (bytes) {
+		/* vr0 - vr3: P + Dx */
+		asm volatile("vld $vr0, %0" : : "m" (p[0]));
+		asm volatile("vld $vr1, %0" : : "m" (p[16]));
+		asm volatile("vld $vr2, %0" : : "m" (p[32]));
+		asm volatile("vld $vr3, %0" : : "m" (p[48]));
+		/* vr4 - vr7: Qx */
+		asm volatile("vld $vr4, %0" : : "m" (dq[0]));
+		asm volatile("vld $vr5, %0" : : "m" (dq[16]));
+		asm volatile("vld $vr6, %0" : : "m" (dq[32]));
+		asm volatile("vld $vr7, %0" : : "m" (dq[48]));
+		/* vr4 - vr7: Q + Qx */
+		asm volatile("vld $vr8, %0" : : "m" (q[0]));
+		asm volatile("vld $vr9, %0" : : "m" (q[16]));
+		asm volatile("vld $vr10, %0" : : "m" (q[32]));
+		asm volatile("vld $vr11, %0" : : "m" (q[48]));
+		asm volatile("vxor.v $vr4, $vr4, $vr8");
+		asm volatile("vxor.v $vr5, $vr5, $vr9");
+		asm volatile("vxor.v $vr6, $vr6, $vr10");
+		asm volatile("vxor.v $vr7, $vr7, $vr11");
+
+		/* vr8 - vr11: higher 4 bits of each byte of (Q + Qx) */
+		asm volatile("vsrli.b $vr8, $vr4, 4");
+		asm volatile("vsrli.b $vr9, $vr5, 4");
+		asm volatile("vsrli.b $vr10, $vr6, 4");
+		asm volatile("vsrli.b $vr11, $vr7, 4");
+		/* vr4 - vr7: lower 4 bits of each byte of (Q + Qx) */
+		asm volatile("vandi.b $vr4, $vr4, 0x0f");
+		asm volatile("vandi.b $vr5, $vr5, 0x0f");
+		asm volatile("vandi.b $vr6, $vr6, 0x0f");
+		asm volatile("vandi.b $vr7, $vr7, 0x0f");
+		/* lookup from qmul[0] */
+		asm volatile("vshuf.b $vr4, $vr22, $vr22, $vr4");
+		asm volatile("vshuf.b $vr5, $vr22, $vr22, $vr5");
+		asm volatile("vshuf.b $vr6, $vr22, $vr22, $vr6");
+		asm volatile("vshuf.b $vr7, $vr22, $vr22, $vr7");
+		/* lookup from qmul[16] */
+		asm volatile("vshuf.b $vr8, $vr23, $vr23, $vr8");
+		asm volatile("vshuf.b $vr9, $vr23, $vr23, $vr9");
+		asm volatile("vshuf.b $vr10, $vr23, $vr23, $vr10");
+		asm volatile("vshuf.b $vr11, $vr23, $vr23, $vr11");
+		/* vr4 - vr7: qmul(Q + Qx) = Dx */
+		asm volatile("vxor.v $vr4, $vr4, $vr8");
+		asm volatile("vxor.v $vr5, $vr5, $vr9");
+		asm volatile("vxor.v $vr6, $vr6, $vr10");
+		asm volatile("vxor.v $vr7, $vr7, $vr11");
+		asm volatile("vst $vr4, %0" : "=m" (dq[0]));
+		asm volatile("vst $vr5, %0" : "=m" (dq[16]));
+		asm volatile("vst $vr6, %0" : "=m" (dq[32]));
+		asm volatile("vst $vr7, %0" : "=m" (dq[48]));
+
+		/* vr0 - vr3: P + Dx + Dx = P */
+		asm volatile("vxor.v $vr0, $vr0, $vr4");
+		asm volatile("vxor.v $vr1, $vr1, $vr5");
+		asm volatile("vxor.v $vr2, $vr2, $vr6");
+		asm volatile("vxor.v $vr3, $vr3, $vr7");
+		asm volatile("vst $vr0, %0" : "=m" (p[0]));
+		asm volatile("vst $vr1, %0" : "=m" (p[16]));
+		asm volatile("vst $vr2, %0" : "=m" (p[32]));
+		asm volatile("vst $vr3, %0" : "=m" (p[48]));
+
+		bytes -= 64;
+		p += 64;
+		q += 64;
+		dq += 64;
+	}
+
+	kernel_fpu_end();
+}
+
+const struct raid6_recov_calls raid6_recov_lsx = {
+	.data2 = raid6_2data_recov_lsx,
+	.datap = raid6_datap_recov_lsx,
+	.name = "lsx",
+};
+#endif /* CONFIG_CPU_HAS_LSX */
+
+#ifdef CONFIG_CPU_HAS_LASX
+static void raid6_2data_recov_lasx(int disks, size_t bytes, int faila,
+				   int failb, void **ptrs)
+{
+	u8 *p, *q, *dp, *dq;
+	const u8 *pbmul;	/* P multiplier table for B data */
+	const u8 *qmul;		/* Q multiplier table (for both) */
+
+	p = (u8 *)ptrs[disks - 2];
+	q = (u8 *)ptrs[disks - 1];
+
+	/*
+	 * Compute syndrome with zero for the missing data pages
+	 * Use the dead data pages as temporary storage for
+	 * delta p and delta q
+	 */
+	dp = (u8 *)ptrs[faila];
+	ptrs[faila] = page_address(ZERO_PAGE(0));
+	ptrs[disks - 2] = dp;
+	dq = (u8 *)ptrs[failb];
+	ptrs[failb] = page_address(ZERO_PAGE(0));
+	ptrs[disks - 1] = dq;
+
+	raid6_gen_syndrome(disks, bytes, ptrs);
+
+	/* Restore pointer table */
+	ptrs[faila] = dp;
+	ptrs[failb] = dq;
+	ptrs[disks - 2] = p;
+	ptrs[disks - 1] = q;
+
+	/* Now, pick the proper data tables */
+	pbmul = raid6_vgfmul[raid6_gfexi[failb - faila]];
+	qmul  = raid6_vgfmul[raid6_gfinv[raid6_gfexp[faila] ^ raid6_gfexp[failb]]];
+
+	kernel_fpu_begin();
+
+	/*
+	 * xr20, xr21: qmul
+	 * xr22, xr23: pbmul
+	 */
+	asm volatile("vld $vr20, %0" : : "m" (qmul[0]));
+	asm volatile("vld $vr21, %0" : : "m" (qmul[16]));
+	asm volatile("vld $vr22, %0" : : "m" (pbmul[0]));
+	asm volatile("vld $vr23, %0" : : "m" (pbmul[16]));
+	asm volatile("xvreplve0.q $xr20, $xr20");
+	asm volatile("xvreplve0.q $xr21, $xr21");
+	asm volatile("xvreplve0.q $xr22, $xr22");
+	asm volatile("xvreplve0.q $xr23, $xr23");
+
+	while (bytes) {
+		/* xr0, xr1: Q */
+		asm volatile("xvld $xr0, %0" : : "m" (q[0]));
+		asm volatile("xvld $xr1, %0" : : "m" (q[32]));
+		/* xr0, xr1: Q + Qxy */
+		asm volatile("xvld $xr4, %0" : : "m" (dq[0]));
+		asm volatile("xvld $xr5, %0" : : "m" (dq[32]));
+		asm volatile("xvxor.v $xr0, $xr0, $xr4");
+		asm volatile("xvxor.v $xr1, $xr1, $xr5");
+		/* xr2, xr3: P */
+		asm volatile("xvld $xr2, %0" : : "m" (p[0]));
+		asm volatile("xvld $xr3, %0" : : "m" (p[32]));
+		/* xr2, xr3: P + Pxy */
+		asm volatile("xvld $xr4, %0" : : "m" (dp[0]));
+		asm volatile("xvld $xr5, %0" : : "m" (dp[32]));
+		asm volatile("xvxor.v $xr2, $xr2, $xr4");
+		asm volatile("xvxor.v $xr3, $xr3, $xr5");
+
+		/* xr4, xr5: higher 4 bits of each byte of (Q + Qxy) */
+		asm volatile("xvsrli.b $xr4, $xr0, 4");
+		asm volatile("xvsrli.b $xr5, $xr1, 4");
+		/* xr0, xr1: lower 4 bits of each byte of (Q + Qxy) */
+		asm volatile("xvandi.b $xr0, $xr0, 0x0f");
+		asm volatile("xvandi.b $xr1, $xr1, 0x0f");
+		/* lookup from qmul[0] */
+		asm volatile("xvshuf.b $xr0, $xr20, $xr20, $xr0");
+		asm volatile("xvshuf.b $xr1, $xr20, $xr20, $xr1");
+		/* lookup from qmul[16] */
+		asm volatile("xvshuf.b $xr4, $xr21, $xr21, $xr4");
+		asm volatile("xvshuf.b $xr5, $xr21, $xr21, $xr5");
+		/* xr6, xr7: B(Q + Qxy) */
+		asm volatile("xvxor.v $xr6, $xr4, $xr0");
+		asm volatile("xvxor.v $xr7, $xr5, $xr1");
+
+		/* xr4, xr5: higher 4 bits of each byte of (P + Pxy) */
+		asm volatile("xvsrli.b $xr4, $xr2, 4");
+		asm volatile("xvsrli.b $xr5, $xr3, 4");
+		/* xr0, xr1: lower 4 bits of each byte of (P + Pxy) */
+		asm volatile("xvandi.b $xr0, $xr2, 0x0f");
+		asm volatile("xvandi.b $xr1, $xr3, 0x0f");
+		/* lookup from pbmul[0] */
+		asm volatile("xvshuf.b $xr0, $xr22, $xr22, $xr0");
+		asm volatile("xvshuf.b $xr1, $xr22, $xr22, $xr1");
+		/* lookup from pbmul[16] */
+		asm volatile("xvshuf.b $xr4, $xr23, $xr23, $xr4");
+		asm volatile("xvshuf.b $xr5, $xr23, $xr23, $xr5");
+		/* xr0, xr1: A(P + Pxy) */
+		asm volatile("xvxor.v $xr0, $xr0, $xr4");
+		asm volatile("xvxor.v $xr1, $xr1, $xr5");
+
+		/* xr0, xr1: A(P + Pxy) + B(Q + Qxy) = Dx */
+		asm volatile("xvxor.v $xr0, $xr0, $xr6");
+		asm volatile("xvxor.v $xr1, $xr1, $xr7");
+
+		/* xr2, xr3: P + Pxy + Dx = Dy */
+		asm volatile("xvxor.v $xr2, $xr2, $xr0");
+		asm volatile("xvxor.v $xr3, $xr3, $xr1");
+
+		asm volatile("xvst $xr0, %0" : "=m" (dq[0]));
+		asm volatile("xvst $xr1, %0" : "=m" (dq[32]));
+		asm volatile("xvst $xr2, %0" : "=m" (dp[0]));
+		asm volatile("xvst $xr3, %0" : "=m" (dp[32]));
+
+		bytes -= 64;
+		p += 64;
+		q += 64;
+		dp += 64;
+		dq += 64;
+	}
+
+	kernel_fpu_end();
+}
+
+static void raid6_datap_recov_lasx(int disks, size_t bytes, int faila,
+				   void **ptrs)
+{
+	u8 *p, *q, *dq;
+	const u8 *qmul;		/* Q multiplier table */
+
+	p = (u8 *)ptrs[disks - 2];
+	q = (u8 *)ptrs[disks - 1];
+
+	/*
+	 * Compute syndrome with zero for the missing data page
+	 * Use the dead data page as temporary storage for delta q
+	 */
+	dq = (u8 *)ptrs[faila];
+	ptrs[faila] = page_address(ZERO_PAGE(0));
+	ptrs[disks - 1] = dq;
+
+	raid6_gen_syndrome(disks, bytes, ptrs);
+
+	/* Restore pointer table */
+	ptrs[faila] = dq;
+	ptrs[disks - 1] = q;
+
+	/* Now, pick the proper data tables */
+	qmul  = raid6_vgfmul[raid6_gfinv[raid6_gfexp[faila]]];
+
+	kernel_fpu_begin();
+
+	/* xr22, xr23: qmul */
+	asm volatile("vld $vr22, %0" : : "m" (qmul[0]));
+	asm volatile("xvreplve0.q $xr22, $xr22");
+	asm volatile("vld $vr23, %0" : : "m" (qmul[16]));
+	asm volatile("xvreplve0.q $xr23, $xr23");
+
+	while (bytes) {
+		/* xr0, xr1: P + Dx */
+		asm volatile("xvld $xr0, %0" : : "m" (p[0]));
+		asm volatile("xvld $xr1, %0" : : "m" (p[32]));
+		/* xr2, xr3: Qx */
+		asm volatile("xvld $xr2, %0" : : "m" (dq[0]));
+		asm volatile("xvld $xr3, %0" : : "m" (dq[32]));
+		/* xr2, xr3: Q + Qx */
+		asm volatile("xvld $xr4, %0" : : "m" (q[0]));
+		asm volatile("xvld $xr5, %0" : : "m" (q[32]));
+		asm volatile("xvxor.v $xr2, $xr2, $xr4");
+		asm volatile("xvxor.v $xr3, $xr3, $xr5");
+
+		/* xr4, xr5: higher 4 bits of each byte of (Q + Qx) */
+		asm volatile("xvsrli.b $xr4, $xr2, 4");
+		asm volatile("xvsrli.b $xr5, $xr3, 4");
+		/* xr2, xr3: lower 4 bits of each byte of (Q + Qx) */
+		asm volatile("xvandi.b $xr2, $xr2, 0x0f");
+		asm volatile("xvandi.b $xr3, $xr3, 0x0f");
+		/* lookup from qmul[0] */
+		asm volatile("xvshuf.b $xr2, $xr22, $xr22, $xr2");
+		asm volatile("xvshuf.b $xr3, $xr22, $xr22, $xr3");
+		/* lookup from qmul[16] */
+		asm volatile("xvshuf.b $xr4, $xr23, $xr23, $xr4");
+		asm volatile("xvshuf.b $xr5, $xr23, $xr23, $xr5");
+		/* xr2, xr3: qmul(Q + Qx) = Dx */
+		asm volatile("xvxor.v $xr2, $xr2, $xr4");
+		asm volatile("xvxor.v $xr3, $xr3, $xr5");
+
+		/* xr0, xr1: P + Dx + Dx = P */
+		asm volatile("xvxor.v $xr0, $xr0, $xr2");
+		asm volatile("xvxor.v $xr1, $xr1, $xr3");
+
+		asm volatile("xvst $xr2, %0" : "=m" (dq[0]));
+		asm volatile("xvst $xr3, %0" : "=m" (dq[32]));
+		asm volatile("xvst $xr0, %0" : "=m" (p[0]));
+		asm volatile("xvst $xr1, %0" : "=m" (p[32]));
+
+		bytes -= 64;
+		p += 64;
+		q += 64;
+		dq += 64;
+	}
+
+	kernel_fpu_end();
+}
+
+const struct raid6_recov_calls raid6_recov_lasx = {
+	.data2 = raid6_2data_recov_lasx,
+	.datap = raid6_datap_recov_lasx,
+	.name = "lasx",
+};
+#endif /* CONFIG_CPU_HAS_LASX */