module/icp/asm-arm/sha2: enable non-SIMD asm kernels on armv5/6

My merged pull request openzfs#15557 fixes compilation of sha2 kernels on arm v5/6. However, the compiler guards only allows sha256/512_armv7_impl to be used when __ARM_ARCH > 6. This patch enables these ASM kernels on all arm architectures. Some compiler guards are adjusted accordingly to avoid the unnecessary compilation of SIMD (e.g., neon, armv8ce) kernels on old architectures. Signed-off-by: Shengqi Chen <harry-chen@outlook.com>
Harry-Chen · Dec 2, 2023 · 070fd50 · 070fd50
1 parent 014265f
commit 070fd50
Show file tree

Hide file tree

Showing 4 changed files with 27 additions and 26 deletions.
diff --git a/module/icp/algs/sha2/sha256_impl.c b/module/icp/algs/sha2/sha256_impl.c
@@ -118,7 +118,15 @@ const sha256_ops_t sha256_shani_impl = {
 };
 #endif
 
-#elif defined(__aarch64__) || (defined(__arm__) && __ARM_ARCH > 6)
+#elif defined(__aarch64__) || defined(__arm__)
+extern void zfs_sha256_block_armv7(uint32_t s[8], const void *, size_t);
+const sha256_ops_t sha256_armv7_impl = {
+	.is_supported = sha2_is_supported,
+	.transform = zfs_sha256_block_armv7,
+	.name = "armv7"
+};
+
+#if __ARM_ARCH > 6
 static boolean_t sha256_have_neon(void)
 {
 	return (kfpu_allowed() && zfs_neon_available());
@@ -129,13 +137,6 @@ static boolean_t sha256_have_armv8ce(void)
 	return (kfpu_allowed() && zfs_sha256_available());
 }
 
-extern void zfs_sha256_block_armv7(uint32_t s[8], const void *, size_t);
-const sha256_ops_t sha256_armv7_impl = {
-	.is_supported = sha2_is_supported,
-	.transform = zfs_sha256_block_armv7,
-	.name = "armv7"
-};
-
 TF(zfs_sha256_block_neon, tf_sha256_neon);
 const sha256_ops_t sha256_neon_impl = {
 	.is_supported = sha256_have_neon,
@@ -149,6 +150,7 @@ const sha256_ops_t sha256_armv8_impl = {
 	.transform = tf_sha256_armv8ce,
 	.name = "armv8-ce"
 };
+#endif
 
 #elif defined(__PPC64__)
 static boolean_t sha256_have_isa207(void)
@@ -192,11 +194,13 @@ static const sha256_ops_t *const sha256_impls[] = {
 #if defined(__x86_64) && defined(HAVE_SSE4_1)
 	&sha256_shani_impl,
 #endif
-#if defined(__aarch64__) || (defined(__arm__) && __ARM_ARCH > 6)
+#if defined(__aarch64__) || defined(__arm__)
 	&sha256_armv7_impl,
+#if __ARM_ARCH > 6
 	&sha256_neon_impl,
 	&sha256_armv8_impl,
 #endif
+#endif
 #if defined(__PPC64__)
 	&sha256_ppc_impl,
 	&sha256_power8_impl,

diff --git a/module/icp/algs/sha2/sha512_impl.c b/module/icp/algs/sha2/sha512_impl.c
@@ -88,14 +88,15 @@ const sha512_ops_t sha512_avx2_impl = {
 };
 #endif
 
-#elif defined(__aarch64__)
+#elif defined(__aarch64__) || defined(__arm__)
 extern void zfs_sha512_block_armv7(uint64_t s[8], const void *, size_t);
 const sha512_ops_t sha512_armv7_impl = {
 	.is_supported = sha2_is_supported,
 	.transform = zfs_sha512_block_armv7,
 	.name = "armv7"
 };
 
+#if defined(__aarch64__)
 static boolean_t sha512_have_armv8ce(void)
 {
 	return (kfpu_allowed() && zfs_sha512_available());
@@ -107,15 +108,9 @@ const sha512_ops_t sha512_armv8_impl = {
 	.transform = tf_sha512_armv8ce,
 	.name = "armv8-ce"
 };
+#endif
 
-#elif defined(__arm__) && __ARM_ARCH > 6
-extern void zfs_sha512_block_armv7(uint64_t s[8], const void *, size_t);
-const sha512_ops_t sha512_armv7_impl = {
-	.is_supported = sha2_is_supported,
-	.transform = zfs_sha512_block_armv7,
-	.name = "armv7"
-};
-
+#if defined(__arm__) && __ARM_ARCH > 6
 static boolean_t sha512_have_neon(void)
 {
 	return (kfpu_allowed() && zfs_neon_available());
@@ -127,6 +122,7 @@ const sha512_ops_t sha512_neon_impl = {
 	.transform = tf_sha512_neon,
 	.name = "neon"
 };
+#endif
 
 #elif defined(__PPC64__)
 TF(zfs_sha512_ppc, tf_sha512_ppc);
@@ -164,14 +160,15 @@ static const sha512_ops_t *const sha512_impls[] = {
 #if defined(__x86_64) && defined(HAVE_AVX2)
 	&sha512_avx2_impl,
 #endif
-#if defined(__aarch64__)
+#if defined(__aarch64__) || defined(__arm__)
 	&sha512_armv7_impl,
+#if defined(__aarch64__)
 	&sha512_armv8_impl,
 #endif
 #if defined(__arm__) && __ARM_ARCH > 6
-	&sha512_armv7_impl,
 	&sha512_neon_impl,
 #endif
+#endif
 #if defined(__PPC64__)
 	&sha512_ppc_impl,
 	&sha512_power8_impl,

diff --git a/module/icp/asm-arm/sha2/sha256-armv7.S b/module/icp/asm-arm/sha2/sha256-armv7.S
@@ -1837,6 +1837,7 @@ zfs_sha256_block_armv7:
 #endif
 .size	zfs_sha256_block_armv7,.-zfs_sha256_block_armv7
 
+#if __ARM_ARCH__ >= 7
 .arch	armv7-a
 .fpu	neon
 
@@ -1849,11 +1850,7 @@ zfs_sha256_block_neon:
 	stmdb	sp!,{r4-r12,lr}
 
 	sub	r11,sp,#16*4+16
-#if __ARM_ARCH__ >=7
 	adr	r14,K256
-#else
-	ldr	r14,=K256
-#endif
 	bic	r11,r11,#15		@ align for 128-bit stores
 	mov	r12,sp
 	mov	sp,r11			@ alloca
@@ -2773,4 +2770,5 @@ zfs_sha256_block_armv8:
 	bx	lr		@ bx lr
 .size	zfs_sha256_block_armv8,.-zfs_sha256_block_armv8
 
-#endif
+#endif // #if __ARM_ARCH__ >= 7
+#endif // #if defined(__arm__)
diff --git a/module/icp/asm-arm/sha2/sha512-armv7.S b/module/icp/asm-arm/sha2/sha512-armv7.S
@@ -493,6 +493,7 @@ zfs_sha512_block_armv7:
 #endif
 .size	zfs_sha512_block_armv7,.-zfs_sha512_block_armv7
 
+#if __ARM_ARCH__>= 7
 .arch	armv7-a
 .fpu	neon
 
@@ -1822,4 +1823,5 @@ zfs_sha512_block_neon:
 	VFP_ABI_POP
 	bx	lr				@ .word	0xe12fff1e
 .size	zfs_sha512_block_neon,.-zfs_sha512_block_neon
-#endif
+#endif // #if __ARM_ARCH__ >= 7
+#endif // #if defined(__arm__)