Add PPC cpu feature tests for FreeBSD and Linux

Add needed cpu feature tests for powerpc architecture.

Overview:
zfs_altivec_available() - needed by RAID-Z
zfs_vsx_available()     - needed by BLAKE3
zfs_isa207_available()  - needed by SHA2

Part 1 - Userspace
- use getauxval() for Linux and elf_aux_info() for FreeBSD
- direct including <sys/auxv.h> fails with double definitions
- so we self define the needed functions and definitions

Part 2 - Kernel space FreeBSD
- use exported cpu_features of <powerpc/cpu.h>

Part 3 - Kernel space Linux
- use cpu_has_feature() function of <asm/cpufeature.h>

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Ryan Moeller <ryan@iXsystems.com>
Signed-off-by: Tino Reichardt <milky-zfs@mcmilk.de>
Closes #13725

include/os/freebsd/Makefile.am

@@ -50,6 +50,7 @@ noinst_HEADERS = \
 	%D%/spl/sys/sid.h \
 	%D%/spl/sys/sig.h \
 	%D%/spl/sys/simd.h \
+	%D%/spl/sys/simd_powerpc.h \
 	%D%/spl/sys/simd_x86.h \
 	%D%/spl/sys/spl_condvar.h \
 	%D%/spl/sys/string.h \

include/os/freebsd/spl/sys/simd.h

@@ -26,13 +26,16 @@
  * $FreeBSD$
  */
 
-
 #ifndef _FREEBSD_SIMD_H
 #define	_FREEBSD_SIMD_H
+
 #if defined(__amd64__) || defined(__i386__)
 #include <sys/simd_x86.h>
-#else
 
+#elif defined(__powerpc__)
+#include <sys/simd_powerpc.h>
+
+#else
 #define	kfpu_allowed()		0
 #define	kfpu_initialize(tsk)	do {} while (0)
 #define	kfpu_begin()		do {} while (0)
@@ -40,4 +43,5 @@
 #define	kfpu_init()		(0)
 #define	kfpu_fini()		do {} while (0)
 #endif
+
 #endif

include/os/freebsd/spl/sys/simd_powerpc.h

@@ -0,0 +1,90 @@
+/*
+ * 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 (C) 2022 Tino Reichardt <milky-zfs@mcmilk.de>
+ */
+
+/*
+ * USER API:
+ *
+ * Kernel fpu methods:
+ *	kfpu_allowed()
+ *	kfpu_begin()
+ *	kfpu_end()
+ *	kfpu_init()
+ *	kfpu_fini()
+ *
+ * SIMD support:
+ *
+ * Following functions should be called to determine whether CPU feature
+ * is supported. All functions are usable in kernel and user space.
+ * If a SIMD algorithm is using more than one instruction set
+ * all relevant feature test functions should be called.
+ *
+ * Supported features:
+ *   zfs_altivec_available()
+ *   zfs_vsx_available()
+ *   zfs_isa207_available()
+ */
+
+#ifndef _FREEBSD_SIMD_POWERPC_H
+#define	_FREEBSD_SIMD_POWERPC_H
+
+#include <sys/types.h>
+#include <sys/cdefs.h>
+
+#include <machine/pcb.h>
+#include <powerpc/cpu.h>
+
+#define	kfpu_allowed()		1
+#define	kfpu_initialize(tsk)	do {} while (0)
+#define	kfpu_begin()		do {} while (0)
+#define	kfpu_end()		do {} while (0)
+#define	kfpu_init()		(0)
+#define	kfpu_fini()		do {} while (0)
+
+/*
+ * Check if Altivec is available
+ */
+static inline boolean_t
+zfs_altivec_available(void)
+{
+	return ((cpu_features & PPC_FEATURE_HAS_ALTIVEC) != 0);
+}
+
+/*
+ * Check if VSX is available
+ */
+static inline boolean_t
+zfs_vsx_available(void)
+{
+	return ((cpu_features & PPC_FEATURE_HAS_VSX) != 0);
+}
+
+/*
+ * Check if POWER ISA 2.07 is available (SHA2)
+ */
+static inline boolean_t
+zfs_isa207_available(void)
+{
+	return ((cpu_features2 & PPC_FEATURE2_ARCH_2_07) != 0);
+}

include/os/freebsd/spl/sys/simd_x86.h

@@ -77,7 +77,7 @@ __simd_state_enabled(const uint64_t state)
 	boolean_t has_osxsave;
 	uint64_t xcr0;
 
-	has_osxsave = !!(cpu_feature2 & CPUID2_OSXSAVE);
+	has_osxsave = (cpu_feature2 & CPUID2_OSXSAVE) != 0;
 
 	if (!has_osxsave)
 		return (B_FALSE);
@@ -99,7 +99,7 @@ __simd_state_enabled(const uint64_t state)
 static inline boolean_t
 zfs_sse_available(void)
 {
-	return (!!(cpu_feature & CPUID_SSE));
+	return ((cpu_feature & CPUID_SSE) != 0);
 }
 
 /*
@@ -108,7 +108,7 @@ zfs_sse_available(void)
 static inline boolean_t
 zfs_sse2_available(void)
 {
-	return (!!(cpu_feature & CPUID_SSE2));
+	return ((cpu_feature & CPUID_SSE2) != 0);
 }
 
 /*
@@ -117,7 +117,7 @@ zfs_sse2_available(void)
 static inline boolean_t
 zfs_sse3_available(void)
 {
-	return (!!(cpu_feature2 & CPUID2_SSE3));
+	return ((cpu_feature2 & CPUID2_SSE3) != 0);
 }
 
 /*
@@ -126,7 +126,7 @@ zfs_sse3_available(void)
 static inline boolean_t
 zfs_ssse3_available(void)
 {
-	return (!!(cpu_feature2 & CPUID2_SSSE3));
+	return ((cpu_feature2 & CPUID2_SSSE3) != 0);
 }
 
 /*
@@ -135,7 +135,7 @@ zfs_ssse3_available(void)
 static inline boolean_t
 zfs_sse4_1_available(void)
 {
-	return (!!(cpu_feature2 & CPUID2_SSE41));
+	return ((cpu_feature2 & CPUID2_SSE41) != 0);
 }
 
 /*
@@ -144,7 +144,7 @@ zfs_sse4_1_available(void)
 static inline boolean_t
 zfs_sse4_2_available(void)
 {
-	return (!!(cpu_feature2 & CPUID2_SSE42));
+	return ((cpu_feature2 & CPUID2_SSE42) != 0);
 }
 
 /*
@@ -155,7 +155,7 @@ zfs_avx_available(void)
 {
 	boolean_t has_avx;
 
-	has_avx = !!(cpu_feature2 & CPUID2_AVX);
+	has_avx = (cpu_feature2 & CPUID2_AVX) != 0;
 
 	return (has_avx && __ymm_enabled());
 }
@@ -168,7 +168,7 @@ zfs_avx2_available(void)
 {
 	boolean_t has_avx2;
 
-	has_avx2 = !!(cpu_stdext_feature & CPUID_STDEXT_AVX2);
+	has_avx2 = (cpu_stdext_feature & CPUID_STDEXT_AVX2) != 0;
 
 	return (has_avx2 && __ymm_enabled());
 }
@@ -196,7 +196,7 @@ zfs_avx512f_available(void)
 {
 	boolean_t has_avx512;
 
-	has_avx512 = !!(cpu_stdext_feature & CPUID_STDEXT_AVX512F);
+	has_avx512 = (cpu_stdext_feature & CPUID_STDEXT_AVX512F) != 0;
 
 	return (has_avx512 && __zmm_enabled());
 }
@@ -207,8 +207,8 @@ zfs_avx512cd_available(void)
 {
 	boolean_t has_avx512;
 
-	has_avx512 = !!(cpu_stdext_feature & CPUID_STDEXT_AVX512F) &&
+	has_avx512 = (cpu_stdext_feature & CPUID_STDEXT_AVX512F) != 0 &&
-	    !!(cpu_stdext_feature & CPUID_STDEXT_AVX512CD);
+	    (cpu_stdext_feature & CPUID_STDEXT_AVX512CD) != 0;
 
 	return (has_avx512 && __zmm_enabled());
 }
@@ -219,8 +219,8 @@ zfs_avx512er_available(void)
 {
 	boolean_t has_avx512;
 
-	has_avx512 = !!(cpu_stdext_feature & CPUID_STDEXT_AVX512F) &&
+	has_avx512 = (cpu_stdext_feature & CPUID_STDEXT_AVX512F) != 0 &&
-	    !!(cpu_stdext_feature & CPUID_STDEXT_AVX512CD);
+	    (cpu_stdext_feature & CPUID_STDEXT_AVX512CD) != 0;
 
 	return (has_avx512 && __zmm_enabled());
 }
@@ -231,8 +231,8 @@ zfs_avx512pf_available(void)
 {
 	boolean_t has_avx512;
 
-	has_avx512 = !!(cpu_stdext_feature & CPUID_STDEXT_AVX512F) &&
+	has_avx512 = (cpu_stdext_feature & CPUID_STDEXT_AVX512F) != 0 &&
-	    !!(cpu_stdext_feature & CPUID_STDEXT_AVX512PF);
+	    (cpu_stdext_feature & CPUID_STDEXT_AVX512PF) != 0;
 
 	return (has_avx512 && __zmm_enabled());
 }
@@ -243,7 +243,7 @@ zfs_avx512bw_available(void)
 {
 	boolean_t has_avx512 = B_FALSE;
 
-	has_avx512 = !!(cpu_stdext_feature & CPUID_STDEXT_AVX512BW);
+	has_avx512 = (cpu_stdext_feature & CPUID_STDEXT_AVX512BW) != 0;
 
 	return (has_avx512 && __zmm_enabled());
 }
@@ -254,8 +254,8 @@ zfs_avx512dq_available(void)
 {
 	boolean_t has_avx512;
 
-	has_avx512 = !!(cpu_stdext_feature & CPUID_STDEXT_AVX512F) &&
+	has_avx512 = (cpu_stdext_feature & CPUID_STDEXT_AVX512F) != 0 &&
-	    !!(cpu_stdext_feature & CPUID_STDEXT_AVX512DQ);
+	    (cpu_stdext_feature & CPUID_STDEXT_AVX512DQ) != 0;
 
 	return (has_avx512 && __zmm_enabled());
 }
@@ -266,8 +266,8 @@ zfs_avx512vl_available(void)
 {
 	boolean_t has_avx512;
 
-	has_avx512 = !!(cpu_stdext_feature & CPUID_STDEXT_AVX512F) &&
+	has_avx512 = (cpu_stdext_feature & CPUID_STDEXT_AVX512F) != 0 &&
-	    !!(cpu_stdext_feature & CPUID_STDEXT_AVX512VL);
+	    (cpu_stdext_feature & CPUID_STDEXT_AVX512VL) != 0;
 
 	return (has_avx512 && __zmm_enabled());
 }
@@ -278,8 +278,8 @@ zfs_avx512ifma_available(void)
 {
 	boolean_t has_avx512;
 
-	has_avx512 = !!(cpu_stdext_feature & CPUID_STDEXT_AVX512F) &&
+	has_avx512 = (cpu_stdext_feature & CPUID_STDEXT_AVX512F) != 0 &&
-	    !!(cpu_stdext_feature & CPUID_STDEXT_AVX512IFMA);
+	    (cpu_stdext_feature & CPUID_STDEXT_AVX512IFMA) != 0;
 
 	return (has_avx512 && __zmm_enabled());
 }
@@ -290,8 +290,8 @@ zfs_avx512vbmi_available(void)
 {
 	boolean_t has_avx512;
 
-	has_avx512 = !!(cpu_stdext_feature & CPUID_STDEXT_AVX512F) &&
+	has_avx512 = (cpu_stdext_feature & CPUID_STDEXT_AVX512F) != 0 &&
-	    !!(cpu_stdext_feature & CPUID_STDEXT_BMI1);
+	    (cpu_stdext_feature & CPUID_STDEXT_BMI1) != 0;
 
 	return (has_avx512 && __zmm_enabled());
 }

include/os/linux/kernel/linux/simd_powerpc.h

@@ -21,6 +21,7 @@
 /*
  * Copyright (C) 2019 Romain Dolbeau
  *           <romain.dolbeau@european-processor-initiative.eu>
+ * Copyright (C) 2022 Tino Reichardt <milky-zfs@mcmilk.de>
  */
 
 /*
@@ -41,7 +42,9 @@
  * all relevant feature test functions should be called.
  *
  * Supported features:
- *	zfs_altivec_available()
+ *   zfs_altivec_available()
+ *   zfs_vsx_available()
+ *   zfs_isa207_available()
  */
 
 #ifndef _LINUX_SIMD_POWERPC_H
@@ -57,73 +60,65 @@
 #include <sys/types.h>
 #include <linux/version.h>
 
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 7, 0)
+#include <asm/cpufeature.h>
+#else
+#include <asm/cputable.h>
+#endif
+
 #define	kfpu_allowed()			1
 
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 5, 0)
-#define	kfpu_end()				\
-	{					\
-		disable_kernel_vsx();		\
-		disable_kernel_altivec();	\
-		preempt_enable();		\
-	}
 #define	kfpu_begin()				\
 	{					\
 		preempt_disable();		\
 		enable_kernel_altivec();	\
 		enable_kernel_vsx();		\
+		enable_kernel_spe();		\
+	}
+#define	kfpu_end()				\
+	{					\
+		disable_kernel_spe();		\
+		disable_kernel_vsx();		\
+		disable_kernel_altivec();	\
+		preempt_enable();		\
 	}
 #else
 /* seems that before 4.5 no-one bothered */
 #define	kfpu_begin()
 #define	kfpu_end()		preempt_enable()
 #endif
+
 #define	kfpu_init()		0
 #define	kfpu_fini()		((void) 0)
 
-static inline boolean_t
-zfs_vsx_available(void)
-{
-	boolean_t res;
-#if defined(__powerpc64__)
-	u64 msr;
-#else
-	u32 msr;
-#endif
-	kfpu_begin();
-	__asm volatile("mfmsr %0" : "=r"(msr));
-	res = (msr & 0x800000) != 0;
-	kfpu_end();
-	return (res);
-}
-
 /*
  * Check if AltiVec instruction set is available
  */
 static inline boolean_t
 zfs_altivec_available(void)
 {
-	boolean_t res;
+	return (cpu_has_feature(CPU_FTR_ALTIVEC));
-	/* suggested by macallan at netbsd dot org */
-#if defined(__powerpc64__)
-	u64 msr;
-#else
-	u32 msr;
-#endif
-	kfpu_begin();
-	__asm volatile("mfmsr %0" : "=r"(msr));
-	/*
-	 * 64 bits -> need to check bit 38
-	 * Power ISA Version 3.0B
-	 * p944
-	 * 32 bits -> Need to check bit 6
-	 * AltiVec Technology Programming Environments Manual
-	 * p49 (2-9)
-	 * They are the same, as ppc counts 'backward' ...
-	 */
-	res = (msr & 0x2000000) != 0;
-	kfpu_end();
-	return (res);
 }
+
+/*
+ * Check if VSX is available
+ */
+static inline boolean_t
+zfs_vsx_available(void)
+{
+	return (cpu_has_feature(CPU_FTR_VSX));
+}
+
+/*
+ * Check if POWER ISA 2.07 is available (SHA2)
+ */
+static inline boolean_t
+zfs_isa207_available(void)
+{
+	return (cpu_has_feature(CPU_FTR_ARCH_207S));
+}
+
 #endif /* defined(__powerpc) */
 
 #endif /* _LINUX_SIMD_POWERPC_H */

lib/libspl/include/sys/simd.h

@@ -20,8 +20,8 @@
  * CDDL HEADER END
  */
 /*
- * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
+ * Copyright (c) 2006 Sun Microsystems, Inc.  All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 2022 Tino Reichardt <milky-zfs@mcmilk.de>
  */
 
 #ifndef _LIBSPL_SYS_SIMD_H
@@ -452,63 +452,60 @@ zfs_avx512vbmi_available(void)
 
 #elif defined(__powerpc__)
 
+/* including <sys/auxv.h> clashes with AT_UID and others */
+extern unsigned long getauxval(unsigned long type);
+#if defined(__FreeBSD__)
+#define	AT_HWCAP	25	/* CPU feature flags. */
+#define	AT_HWCAP2	26	/* CPU feature flags 2. */
+extern int elf_aux_info(int aux, void *buf, int buflen);
+static unsigned long getauxval(unsigned long key)
+{
+	unsigned long val = 0UL;
+
+	if (elf_aux_info((int)key, &val, sizeof (val)) != 0)
+		return (0UL);
+
+	return (val);
+}
+#elif defined(__linux__)
+#define	AT_HWCAP	16	/* CPU feature flags. */
+#define	AT_HWCAP2	26	/* CPU feature flags 2. */
+#endif
+
 #define	kfpu_allowed()		1
 #define	kfpu_initialize(tsk)	do {} while (0)
 #define	kfpu_begin()		do {} while (0)
 #define	kfpu_end()		do {} while (0)
 
-/*
+#define	PPC_FEATURE_HAS_ALTIVEC	0x10000000
- * Check if AltiVec instruction set is available
- * No easy way beyond 'altivec works' :-(
- */
-#include <signal.h>
-#include <setjmp.h>
-
-#if defined(__ALTIVEC__) && !defined(__FreeBSD__)
-static jmp_buf env;
-static void sigillhandler(int x)
-{
-	(void) x;
-	longjmp(env, 1);
-}
-#endif
-
 static inline boolean_t
 zfs_altivec_available(void)
 {
-	boolean_t has_altivec = B_FALSE;
+	unsigned long hwcap = getauxval(AT_HWCAP);
-#if defined(__ALTIVEC__) && !defined(__FreeBSD__)
+
-	sighandler_t savesig;
+	return (hwcap & PPC_FEATURE_HAS_ALTIVEC);
-	savesig = signal(SIGILL, sigillhandler);
-	if (setjmp(env)) {
-		signal(SIGILL, savesig);
-		has_altivec = B_FALSE;
-	} else {
-		__asm__ __volatile__("vor 0,0,0\n" : : : "v0");
-		signal(SIGILL, savesig);
-		has_altivec = B_TRUE;
-	}
-#endif
-	return (has_altivec);
 }
+
+#define	PPC_FEATURE_HAS_VSX	0x00000080
 static inline boolean_t
 zfs_vsx_available(void)
 {
-	boolean_t has_vsx = B_FALSE;
+	unsigned long hwcap = getauxval(AT_HWCAP);
-#if defined(__ALTIVEC__) && !defined(__FreeBSD__)
+
-	sighandler_t savesig;
+	return (hwcap & PPC_FEATURE_HAS_VSX);
-	savesig = signal(SIGILL, sigillhandler);
-	if (setjmp(env)) {
-		signal(SIGILL, savesig);
-		has_vsx = B_FALSE;
-	} else {
-		__asm__ __volatile__("xssubsp 0,0,0\n");
-		signal(SIGILL, savesig);
-		has_vsx = B_TRUE;
-	}
-#endif
-	return (has_vsx);
 }
+
+#define	PPC_FEATURE2_ARCH_2_07	0x80000000
+static inline boolean_t
+zfs_isa207_available(void)
+{
+	unsigned long hwcap = getauxval(AT_HWCAP);
+	unsigned long hwcap2 = getauxval(AT_HWCAP2);
+
+	return ((hwcap & PPC_FEATURE_HAS_VSX) &&
+	    (hwcap2 & PPC_FEATURE2_ARCH_2_07));
+}
+
 #else
 
 #define	kfpu_allowed()		0