From 5d0124396423acb08617bdbbc745b60f54d60b57 Mon Sep 17 00:00:00 2001 From: Rich Ercolani <214141+rincebrain@users.noreply.github.com> Date: Fri, 20 Sep 2024 11:16:44 -0400 Subject: [PATCH] Add SIMD metadata in /proc on Linux Too many times, people's performance problems have amounted to "somehow your SIMD support isn't working", and determining that at runtime is difficult to describe to people. This adds a /proc/spl/kstat/zfs/simd node, which exposes metadata about which instructions ZFS thinks it can use, on AArch64 and x86_64 Linux, to make investigating things like this much easier. Reviewed-by: Brian Behlendorf Signed-off-by: Rich Ercolani Closes #16530 --- include/os/freebsd/spl/sys/simd.h | 3 + include/os/linux/kernel/linux/simd.h | 4 + lib/libspl/include/sys/simd.h | 3 + lib/libzpool/Makefile.am | 1 + module/Kbuild.in | 1 + module/zcommon/simd_stat.c | 189 +++++++++++++++++++++++++++ module/zcommon/zfs_prop.c | 6 + 7 files changed, 207 insertions(+) create mode 100644 module/zcommon/simd_stat.c diff --git a/include/os/freebsd/spl/sys/simd.h b/include/os/freebsd/spl/sys/simd.h index 4560bb05e..6bc46755c 100644 --- a/include/os/freebsd/spl/sys/simd.h +++ b/include/os/freebsd/spl/sys/simd.h @@ -50,4 +50,7 @@ #define kfpu_fini() do {} while (0) #endif +#define simd_stat_init() 0 +#define simd_stat_fini() 0 + #endif diff --git a/include/os/linux/kernel/linux/simd.h b/include/os/linux/kernel/linux/simd.h index f4376b218..e7d7a4f22 100644 --- a/include/os/linux/kernel/linux/simd.h +++ b/include/os/linux/kernel/linux/simd.h @@ -45,4 +45,8 @@ #define kfpu_fini() ((void) 0) #endif + +void simd_stat_init(void); +void simd_stat_fini(void); + #endif /* _LINUX_SIMD_H */ diff --git a/lib/libspl/include/sys/simd.h b/lib/libspl/include/sys/simd.h index 41f9df506..2926dc680 100644 --- a/lib/libspl/include/sys/simd.h +++ b/lib/libspl/include/sys/simd.h @@ -592,4 +592,7 @@ zfs_isa207_available(void) #endif +extern void simd_stat_init(void); +extern void simd_stat_fini(void); + #endif /* _LIBSPL_SYS_SIMD_H */ diff --git a/lib/libzpool/Makefile.am b/lib/libzpool/Makefile.am index a93e9b3ad..397959c67 100644 --- a/lib/libzpool/Makefile.am +++ b/lib/libzpool/Makefile.am @@ -50,6 +50,7 @@ nodist_libzpool_la_SOURCES = \ module/os/linux/zfs/zio_crypt.c \ \ module/zcommon/cityhash.c \ + module/zcommon/simd_stat.c \ module/zcommon/zfeature_common.c \ module/zcommon/zfs_comutil.c \ module/zcommon/zfs_deleg.c \ diff --git a/module/Kbuild.in b/module/Kbuild.in index b43b72a0c..dcbdbc912 100644 --- a/module/Kbuild.in +++ b/module/Kbuild.in @@ -232,6 +232,7 @@ zfs-objs += $(addprefix unicode/,$(UNICODE_OBJS)) ZCOMMON_OBJS := \ cityhash.o \ + simd_stat.o \ zfeature_common.o \ zfs_comutil.o \ zfs_deleg.o \ diff --git a/module/zcommon/simd_stat.c b/module/zcommon/simd_stat.c new file mode 100644 index 000000000..538f8317a --- /dev/null +++ b/module/zcommon/simd_stat.c @@ -0,0 +1,189 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or https://opensource.org/licenses/CDDL-1.0. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2024 Google, Inc. All rights reserved. + */ +#include +#include +#include + + +#ifdef _KERNEL +#ifdef __linux__ +#include +#endif /* __linux__ */ +kstat_t *simd_stat_kstat; +#endif /* _KERNEL */ + +#ifdef _KERNEL +#define SIMD_STAT_PRINT(s, feat, val) \ + kmem_scnprintf(s + off, MAX(4095-off, 0), "%-16s\t%1d\n", feat, (val)) + +static int +simd_stat_kstat_data(char *buf, size_t size, void *data) +{ + (void) data; + + static char simd_stat_kstat_payload[4096] = {0}; + static int off = 0; +#ifdef __linux__ + if (off == 0) { + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "kfpu_allowed", kfpu_allowed()); +#ifdef __x86__ + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "kfpu", HAVE_KERNEL_FPU); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "kfpu_internal", HAVE_KERNEL_FPU_INTERNAL); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "__kernel_fpu", HAVE_UNDERSCORE_KERNEL_FPU); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "sse", zfs_sse_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "sse2", zfs_sse2_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "sse3", zfs_sse3_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "ssse3", zfs_ssse3_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "sse41", zfs_sse4_1_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "sse42", zfs_sse4_2_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "avx", zfs_avx_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "avx2", zfs_avx2_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "avx512f", zfs_avx512f_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "avx512cd", zfs_avx512cd_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "avx512er", zfs_avx512er_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "avx512pf", zfs_avx512pf_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "avx512bw", zfs_avx512bw_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "avx512dq", zfs_avx512dq_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "avx512vl", zfs_avx512vl_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "avx512ifma", zfs_avx512ifma_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "avx512vbmi", zfs_avx512vbmi_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "ymm", __ymm_enabled()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "zmm", __zmm_enabled()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "bmi1", zfs_bmi1_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "bmi2", zfs_bmi2_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "aes", zfs_aes_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "pclmulqdq", zfs_pclmulqdq_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "movbe", zfs_movbe_available()); + + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "osxsave", boot_cpu_has(X86_FEATURE_OSXSAVE)); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "xsaves", static_cpu_has(X86_FEATURE_XSAVES)); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "xsaveopt", static_cpu_has(X86_FEATURE_XSAVEOPT)); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "xsave", static_cpu_has(X86_FEATURE_XSAVE)); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "fxsr", static_cpu_has(X86_FEATURE_FXSR)); +#endif /* __x86__ */ +#if defined(__arm__) || defined(__aarch64__) + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "kernel_neon", HAVE_KERNEL_NEON); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "kernel_mode_neon", CONFIG_KERNEL_MODE_NEON); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "neon", zfs_neon_available()); + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "sha256", zfs_sha256_available()); +#if defined(__aarch64__) + /* + * This technically can exist on 32b ARM but we don't + * define hooks to check for it and I didn't want to + * learn enough ARM ASM to add one. + */ + off += SIMD_STAT_PRINT(simd_stat_kstat_payload, + "sha512", zfs_sha512_available()); +#endif /* __aarch64__ */ +#endif /* __arm__ */ + /* We want to short-circuit this on unsupported platforms. */ + off += 1; + } + + kmem_scnprintf(buf, MIN(off, size), "%s", simd_stat_kstat_payload); +#endif /* __linux__ */ + return (0); +} +#endif /* _KERNEL */ + +void +simd_stat_init(void) +{ + static boolean_t simd_stat_initialized = B_FALSE; + + if (!simd_stat_initialized) { +#if defined(_KERNEL) + /* Install kstats for all implementations */ + simd_stat_kstat = kstat_create("zfs", 0, "simd", "misc", + KSTAT_TYPE_RAW, 0, KSTAT_FLAG_VIRTUAL); + + + if (simd_stat_kstat != NULL) { + simd_stat_kstat->ks_data = (void*)(uintptr_t)1; + simd_stat_kstat->ks_ndata = 1; + simd_stat_kstat->ks_flags |= KSTAT_FLAG_NO_HEADERS; + kstat_set_raw_ops(simd_stat_kstat, + NULL, + simd_stat_kstat_data, + NULL); + kstat_install(simd_stat_kstat); + } +#endif /* _KERNEL */ + } + /* Finish initialization */ + simd_stat_initialized = B_TRUE; +} + +void +simd_stat_fini(void) +{ +#if defined(_KERNEL) + if (simd_stat_kstat != NULL) { + kstat_delete(simd_stat_kstat); + simd_stat_kstat = NULL; + } +#endif +} + +#ifdef _KERNEL +EXPORT_SYMBOL(simd_stat_init); +EXPORT_SYMBOL(simd_stat_fini); +#endif diff --git a/module/zcommon/zfs_prop.c b/module/zcommon/zfs_prop.c index 10ac13a89..1206210a9 100644 --- a/module/zcommon/zfs_prop.c +++ b/module/zcommon/zfs_prop.c @@ -41,6 +41,7 @@ #include #include #include +#include #include "zfs_prop.h" #include "zfs_deleg.h" @@ -1074,6 +1075,9 @@ EXPORT_SYMBOL(zfs_kfpu_fpregs); extern int __init zcommon_init(void); extern void zcommon_fini(void); +extern void simd_stat_init(void); +extern void simd_stat_fini(void); + int __init zcommon_init(void) { @@ -1082,6 +1086,7 @@ zcommon_init(void) return (error); fletcher_4_init(); + simd_stat_init(); return (0); } @@ -1089,6 +1094,7 @@ zcommon_init(void) void zcommon_fini(void) { + simd_stat_fini(); fletcher_4_fini(); kfpu_fini(); }