From 8f3f06dfd6873135068ccf1a0b386308e8c4da38 Mon Sep 17 00:00:00 2001 From: WANG Xuerui Date: Wed, 6 Sep 2023 22:53:55 +0800 Subject: raid6: Add LoongArch SIMD syndrome calculation The algorithms work on 64 bytes at a time, which is the L1 cache line size of all current and future LoongArch cores (that we care about), as confirmed by Huacai. The code is based on the generic int.uc algorithm, unrolled 4 times for LSX and 2 times for LASX. Further unrolling does not meaningfully improve the performance according to experiments. Performance numbers measured during system boot on a 3A5000 @ 2.5GHz: > raid6: lasx gen() 12726 MB/s > raid6: lsx gen() 10001 MB/s > raid6: int64x8 gen() 2876 MB/s > raid6: int64x4 gen() 3867 MB/s > raid6: int64x2 gen() 2531 MB/s > raid6: int64x1 gen() 1945 MB/s Comparison of xor() speeds (from different boots but meaningful anyway): > lasx: 11226 MB/s > lsx: 6395 MB/s > int64x4: 2147 MB/s Performance as measured by raid6test: > raid6: lasx gen() 25109 MB/s > raid6: lsx gen() 13233 MB/s > raid6: int64x8 gen() 4164 MB/s > raid6: int64x4 gen() 6005 MB/s > raid6: int64x2 gen() 5781 MB/s > raid6: int64x1 gen() 4119 MB/s > raid6: using algorithm lasx gen() 25109 MB/s > raid6: .... xor() 14439 MB/s, rmw enabled Acked-by: Song Liu Signed-off-by: WANG Xuerui Signed-off-by: Huacai Chen --- include/linux/raid/pq.h | 2 ++ 1 file changed, 2 insertions(+) (limited to 'include/linux') diff --git a/include/linux/raid/pq.h b/include/linux/raid/pq.h index f29aaaf2eb21..874447485848 100644 --- a/include/linux/raid/pq.h +++ b/include/linux/raid/pq.h @@ -108,6 +108,8 @@ extern const struct raid6_calls raid6_vpermxor1; extern const struct raid6_calls raid6_vpermxor2; extern const struct raid6_calls raid6_vpermxor4; extern const struct raid6_calls raid6_vpermxor8; +extern const struct raid6_calls raid6_lsx; +extern const struct raid6_calls raid6_lasx; struct raid6_recov_calls { void (*data2)(int, size_t, int, int, void **); -- cgit v1.2.3 From f2091321044d9fbcadb93dfc1c9cf23e563ea40c Mon Sep 17 00:00:00 2001 From: WANG Xuerui Date: Wed, 6 Sep 2023 22:53:55 +0800 Subject: raid6: Add LoongArch SIMD recovery implementation Similar to the syndrome calculation, the recovery algorithms also work on 64 bytes at a time to align with the L1 cache line size of current and future LoongArch cores (that we care about). Which means unrolled-by-4 LSX and unrolled-by-2 LASX code. The assembly is originally based on the x86 SSSE3/AVX2 ports, but register allocation has been redone to take advantage of LSX/LASX's 32 vector registers, and instruction sequence has been optimized to suit (e.g. LoongArch can perform per-byte srl and andi on vectors, but x86 cannot). Performance numbers measured by instrumenting the raid6test code, on a 3A5000 system clocked at 2.5GHz: > lasx 2data: 354.987 MiB/s > lasx datap: 350.430 MiB/s > lsx 2data: 340.026 MiB/s > lsx datap: 337.318 MiB/s > intx1 2data: 164.280 MiB/s > intx1 datap: 187.966 MiB/s Because recovery algorithms are chosen solely based on priority and availability, lasx is marked as priority 2 and lsx priority 1. At least for the current generation of LoongArch micro-architectures, LASX should always be faster than LSX whenever supported, and have similar power consumption characteristics (because the only known LASX-capable uarch, the LA464, always compute the full 256-bit result for vector ops). Acked-by: Song Liu Signed-off-by: WANG Xuerui Signed-off-by: Huacai Chen --- include/linux/raid/pq.h | 2 + lib/raid6/Makefile | 2 +- lib/raid6/algos.c | 8 + lib/raid6/recov_loongarch_simd.c | 513 +++++++++++++++++++++++++++++++++++++++ lib/raid6/test/Makefile | 2 +- 5 files changed, 525 insertions(+), 2 deletions(-) create mode 100644 lib/raid6/recov_loongarch_simd.c (limited to 'include/linux') diff --git a/include/linux/raid/pq.h b/include/linux/raid/pq.h index 874447485848..006e18decfad 100644 --- a/include/linux/raid/pq.h +++ b/include/linux/raid/pq.h @@ -125,6 +125,8 @@ extern const struct raid6_recov_calls raid6_recov_avx2; extern const struct raid6_recov_calls raid6_recov_avx512; extern const struct raid6_recov_calls raid6_recov_s390xc; extern const struct raid6_recov_calls raid6_recov_neon; +extern const struct raid6_recov_calls raid6_recov_lsx; +extern const struct raid6_recov_calls raid6_recov_lasx; extern const struct raid6_calls raid6_neonx1; extern const struct raid6_calls raid6_neonx2; diff --git a/lib/raid6/Makefile b/lib/raid6/Makefile index 2b9ebe105480..035b0a4db476 100644 --- a/lib/raid6/Makefile +++ b/lib/raid6/Makefile @@ -9,7 +9,7 @@ raid6_pq-$(CONFIG_ALTIVEC) += altivec1.o altivec2.o altivec4.o altivec8.o \ vpermxor1.o vpermxor2.o vpermxor4.o vpermxor8.o raid6_pq-$(CONFIG_KERNEL_MODE_NEON) += neon.o neon1.o neon2.o neon4.o neon8.o recov_neon.o recov_neon_inner.o raid6_pq-$(CONFIG_S390) += s390vx8.o recov_s390xc.o -raid6_pq-$(CONFIG_LOONGARCH) += loongarch_simd.o +raid6_pq-$(CONFIG_LOONGARCH) += loongarch_simd.o recov_loongarch_simd.o hostprogs += mktables diff --git a/lib/raid6/algos.c b/lib/raid6/algos.c index 739c7ebcae1a..0ec534faf019 100644 --- a/lib/raid6/algos.c +++ b/lib/raid6/algos.c @@ -111,6 +111,14 @@ const struct raid6_recov_calls *const raid6_recov_algos[] = { #endif #if defined(CONFIG_KERNEL_MODE_NEON) &raid6_recov_neon, +#endif +#ifdef CONFIG_LOONGARCH +#ifdef CONFIG_CPU_HAS_LASX + &raid6_recov_lasx, +#endif +#ifdef CONFIG_CPU_HAS_LSX + &raid6_recov_lsx, +#endif #endif &raid6_recov_intx1, NULL diff --git a/lib/raid6/recov_loongarch_simd.c b/lib/raid6/recov_loongarch_simd.c new file mode 100644 index 000000000000..94aeac85e6f7 --- /dev/null +++ b/lib/raid6/recov_loongarch_simd.c @@ -0,0 +1,513 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * RAID6 recovery algorithms in LoongArch SIMD (LSX & LASX) + * + * Copyright (C) 2023 WANG Xuerui + * + * Originally based on recov_avx2.c and recov_ssse3.c: + * + * Copyright (C) 2012 Intel Corporation + * Author: Jim Kukunas + */ + +#include +#include "loongarch.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 int raid6_has_lsx(void) +{ + return 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] = (void *)raid6_empty_zero_page; + ptrs[disks - 2] = dp; + dq = (u8 *)ptrs[failb]; + ptrs[failb] = (void *)raid6_empty_zero_page; + ptrs[disks - 1] = dq; + + raid6_call.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] = (void *)raid6_empty_zero_page; + ptrs[disks - 1] = dq; + + raid6_call.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, + .valid = raid6_has_lsx, + .name = "lsx", + .priority = 1, +}; +#endif /* CONFIG_CPU_HAS_LSX */ + +#ifdef CONFIG_CPU_HAS_LASX +static int raid6_has_lasx(void) +{ + return 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] = (void *)raid6_empty_zero_page; + ptrs[disks - 2] = dp; + dq = (u8 *)ptrs[failb]; + ptrs[failb] = (void *)raid6_empty_zero_page; + ptrs[disks - 1] = dq; + + raid6_call.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] = (void *)raid6_empty_zero_page; + ptrs[disks - 1] = dq; + + raid6_call.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 $xr1, %0" : "=m" (p[32])); + + bytes -= 64; + p += 64; + q += 64; + dq += 64; + } + + kernel_fpu_end(); +} + +const struct raid6_recov_calls raid6_recov_lasx = { + .data2 = raid6_2data_recov_lasx, + .datap = raid6_datap_recov_lasx, + .valid = raid6_has_lasx, + .name = "lasx", + .priority = 2, +}; +#endif /* CONFIG_CPU_HAS_LASX */ diff --git a/lib/raid6/test/Makefile b/lib/raid6/test/Makefile index 7b244bce32b3..2abe0076a636 100644 --- a/lib/raid6/test/Makefile +++ b/lib/raid6/test/Makefile @@ -65,7 +65,7 @@ else ifeq ($(HAS_ALTIVEC),yes) OBJS += altivec1.o altivec2.o altivec4.o altivec8.o \ vpermxor1.o vpermxor2.o vpermxor4.o vpermxor8.o else ifeq ($(ARCH),loongarch64) - OBJS += loongarch_simd.o + OBJS += loongarch_simd.o recov_loongarch_simd.o endif .c.o: -- cgit v1.2.3 From 9b04c764af18a1dab6d48ca0671f70cdcccf90a2 Mon Sep 17 00:00:00 2001 From: Qing Zhang Date: Wed, 6 Sep 2023 22:54:16 +0800 Subject: kasan: Add __HAVE_ARCH_SHADOW_MAP to support arch specific mapping MIPS, LoongArch and some other architectures have many holes between different segments and the valid address space (256T available) is insufficient to map all these segments to kasan shadow memory with the common formula provided by kasan core. So we need architecture specific mapping formulas to ensure different segments are mapped individually, and only limited space lengths of those specific segments are mapped to shadow. Therefore, when the incoming address is converted to a shadow, we need to add a condition to determine whether it is valid. Reviewed-by: Andrey Konovalov Signed-off-by: Qing Zhang Signed-off-by: Huacai Chen --- include/linux/kasan.h | 2 ++ mm/kasan/kasan.h | 6 ++++++ 2 files changed, 8 insertions(+) (limited to 'include/linux') diff --git a/include/linux/kasan.h b/include/linux/kasan.h index 819b6bc8ac08..3df5499f7936 100644 --- a/include/linux/kasan.h +++ b/include/linux/kasan.h @@ -54,11 +54,13 @@ extern p4d_t kasan_early_shadow_p4d[MAX_PTRS_PER_P4D]; int kasan_populate_early_shadow(const void *shadow_start, const void *shadow_end); +#ifndef __HAVE_ARCH_SHADOW_MAP static inline void *kasan_mem_to_shadow(const void *addr) { return (void *)((unsigned long)addr >> KASAN_SHADOW_SCALE_SHIFT) + KASAN_SHADOW_OFFSET; } +#endif int kasan_add_zero_shadow(void *start, unsigned long size); void kasan_remove_zero_shadow(void *start, unsigned long size); diff --git a/mm/kasan/kasan.h b/mm/kasan/kasan.h index 2e973b36fe07..f70e3d7a602e 100644 --- a/mm/kasan/kasan.h +++ b/mm/kasan/kasan.h @@ -291,16 +291,22 @@ struct kasan_stack_ring { #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS) +#ifndef __HAVE_ARCH_SHADOW_MAP static inline const void *kasan_shadow_to_mem(const void *shadow_addr) { return (void *)(((unsigned long)shadow_addr - KASAN_SHADOW_OFFSET) << KASAN_SHADOW_SCALE_SHIFT); } +#endif static __always_inline bool addr_has_metadata(const void *addr) { +#ifdef __HAVE_ARCH_SHADOW_MAP + return (kasan_mem_to_shadow((void *)addr) != NULL); +#else return (kasan_reset_tag(addr) >= kasan_shadow_to_mem((void *)KASAN_SHADOW_START)); +#endif } /** -- cgit v1.2.3