btrfs-progs: fix detection of accelerated implementation.

The build fails with crypto backends other than builtin, the
initializers cannot be reached as they're ifdef-ed out.  Move
hash_init_accel under the right condition and delete the
algorithm-specific initializers as they're used only by the hash test
and that can simply call hash_init_accel to set the implementation.

All the -m flags need to be detected at configure time and the flag used
for ifdef (HAVE_CFLAG_m*), not the actual feature defined by compiler as
the dispatcher function is not built with the -m flags.

The uname check for x86_64 must be dropped so on i386/i586 we can still
build accelerated version.

Signed-off-by: David Sterba <dsterba@suse.com>

Makefile

@@ -130,14 +130,18 @@ LIBBTRFSUTIL_LDFLAGS = $(SUBST_LDFLAGS) \
 # Default implementation
 CRYPTO_OBJECTS =
 
-ifeq ($(shell uname -m),x86_64)
+ifeq ($(HAVE_CFLAG_msse2),1)
 crypto_blake2b_sse2_cflags = -msse2
+endif
+ifeq ($(HAVE_CFLAG_msse41),1)
 crypto_blake2b_sse41_cflags = -msse4.1
+endif
+ifeq ($(HAVE_CFLAG_mavx2),1)
 crypto_blake2b_avx2_cflags = -mavx2
+endif
 ifeq ($(HAVE_CFLAG_msha),1)
 crypto_sha256_x86_cflags = -msse4.1 -msha
 endif
-endif
 
 LIBS = $(LIBS_BASE) $(LIBS_CRYPTO)
 LIBBTRFS_LIBS = $(LIBS_BASE) $(LIBS_CRYPTO)

Makefile.inc.in

@@ -24,6 +24,9 @@ PYTHON_CFLAGS = @PYTHON_CFLAGS@
 CRYPTOPROVIDER_BUILTIN = @CRYPTOPROVIDER_BUILTIN@
 CRYPTO_CFLAGS = @GCRYPT_CFLAGS@ @SODIUM_CFLAGS@ @KCAPI_CFLAGS@
 
+HAVE_CFLAG_msse2 = @HAVE_CFLAG_msse2@
+HAVE_CFLAG_msse41 = @HAVE_CFLAG_msse41@
+HAVE_CFLAG_mavx2 = @HAVE_CFLAG_mavx2@
 HAVE_CFLAG_msha = @HAVE_CFLAG_msha@
 
 SUBST_CFLAGS = @CFLAGS@

configure.ac

@@ -43,6 +43,18 @@ AC_C_CONST
 AC_C_VOLATILE
 AC_C_BIGENDIAN
 
+AX_CHECK_COMPILE_FLAG([-msse2], [HAVE_CFLAG_msse2=1], [HAVE_CFLAG_msse2=0])
+AC_SUBST([HAVE_CFLAG_msse2])
+AC_DEFINE_UNQUOTED([HAVE_CFLAG_msse2], [$HAVE_CFLAG_msse2], [Compiler supports -msse2])
+
+AX_CHECK_COMPILE_FLAG([-msse4.1], [HAVE_CFLAG_msse41=1], [HAVE_CFLAG_msse41=0])
+AC_SUBST([HAVE_CFLAG_msse41])
+AC_DEFINE_UNQUOTED([HAVE_CFLAG_msse41], [$HAVE_CFLAG_msse41], [Compiler supports -msse4.1])
+
+AX_CHECK_COMPILE_FLAG([-mavx2], [HAVE_CFLAG_mavx2=1], [HAVE_CFLAG_mavx2=0])
+AC_SUBST([HAVE_CFLAG_mavx2])
+AC_DEFINE_UNQUOTED([HAVE_CFLAG_mavx2], [$HAVE_CFLAG_mavx2], [Compiler supports -mavx2])
+
 AX_CHECK_COMPILE_FLAG([-msha], [HAVE_CFLAG_msha=1], [HAVE_CFLAG_msha=0])
 AC_SUBST([HAVE_CFLAG_msha])
 AC_DEFINE_UNQUOTED([HAVE_CFLAG_msha], [$HAVE_CFLAG_msha], [Compiler supports -msha])

crypto/blake2b-ref.c

@@ -229,15 +229,15 @@ static void (*blake2b_compress)( blake2b_state *S, const uint8_t block[BLAKE2B_B
 void blake2_init_accel(void)
 {
 	if (0);
-#if HAVE_AVX2
+#if HAVE_CFLAG_mavx2 == 1
 	else if (cpu_has_feature(CPU_FLAG_AVX2))
 		blake2b_compress = blake2b_compress_avx2;
 #endif
-#if HAVE_SSE41
+#if HAVE_CFLAG_msse41 == 1
 	else if (cpu_has_feature(CPU_FLAG_SSE41))
 		blake2b_compress = blake2b_compress_sse41;
 #endif
-#if HAVE_SSE2
+#if HAVE_CFLAG_msse2 == 1
 	else if (cpu_has_feature(CPU_FLAG_SSE2))
 		blake2b_compress = blake2b_compress_sse2;
 #endif

crypto/hash-speedtest.c

@@ -183,27 +183,26 @@ int main(int argc, char **argv) {
 		u64 cycles;
 		u64 time;
 		unsigned long cpu_flag;
-		void (*init_accel)(void);
 	} contestants[] = {
 		{ .name = "NULL-NOP", .digest = hash_null_nop, .digest_size = 32 },
 		{ .name = "NULL-MEMCPY", .digest = hash_null_memcpy, .digest_size = 32 },
 		{ .name = "CRC32C-ref", .digest = hash_crc32c, .digest_size = 4,
-		  .cpu_flag = CPU_FLAG_NONE, .init_accel = hash_init_crc32c },
+		  .cpu_flag = CPU_FLAG_NONE },
 		{ .name = "CRC32C-NI", .digest = hash_crc32c, .digest_size = 4,
-		  .cpu_flag = CPU_FLAG_SSE42, .init_accel = hash_init_crc32c },
+		  .cpu_flag = CPU_FLAG_SSE42 },
 		{ .name = "XXHASH", .digest = hash_xxhash, .digest_size = 8 },
 		{ .name = "SHA256-ref", .digest = hash_sha256, .digest_size = 32,
-		  .cpu_flag = CPU_FLAG_NONE, .init_accel = hash_init_sha256 },
+		  .cpu_flag = CPU_FLAG_NONE },
 		{ .name = "SHA256-NI", .digest = hash_sha256, .digest_size = 32,
-		  .cpu_flag = CPU_FLAG_SHA, .init_accel = hash_init_sha256 },
+		  .cpu_flag = CPU_FLAG_SHA },
 		{ .name = "BLAKE2-ref", .digest = hash_blake2b, .digest_size = 32,
-		  .cpu_flag = CPU_FLAG_NONE, .init_accel = hash_init_blake2 },
+		  .cpu_flag = CPU_FLAG_NONE },
 		{ .name = "BLAKE2-SSE2", .digest = hash_blake2b, .digest_size = 32,
-		  .cpu_flag = CPU_FLAG_SSE2, .init_accel = hash_init_blake2 },
+		  .cpu_flag = CPU_FLAG_SSE2 },
 		{ .name = "BLAKE2-SSE41", .digest = hash_blake2b, .digest_size = 32,
-		  .cpu_flag = CPU_FLAG_SSE41, .init_accel = hash_init_blake2 },
+		  .cpu_flag = CPU_FLAG_SSE41 },
 		{ .name = "BLAKE2-AVX2", .digest = hash_blake2b, .digest_size = 32,
-		  .cpu_flag = CPU_FLAG_AVX2, .init_accel = hash_init_blake2 },
+		  .cpu_flag = CPU_FLAG_AVX2 },
 	};
 	int units = UNITS_CYCLES;
 
@@ -278,7 +277,7 @@ int main(int argc, char **argv) {
 
 		if (c->cpu_flag) {
 			cpu_set_level(c->cpu_flag);
-			c->init_accel();
+			hash_init_accel();
 		}
 		tstart = get_time();
 		start = get_cycles(units);

crypto/hash.c

@@ -20,28 +20,11 @@
 #include "crypto/sha.h"
 #include "crypto/blake2.h"
 
-void hash_init_accel(void)
-{
-	crc32c_init_accel();
-	blake2_init_accel();
-	sha256_init_accel();
-}
-
 void hash_init_crc32c(void)
 {
 	crc32c_init_accel();
 }
 
-void hash_init_blake2(void)
-{
-	blake2_init_accel();
-}
-
-void hash_init_sha256(void)
-{
-	sha256_init_accel();
-}
-
 /*
  * Default builtin implementations
  */
@@ -70,6 +53,13 @@ int hash_xxhash(const u8 *buf, size_t length, u8 *out)
  */
 #if CRYPTOPROVIDER_BUILTIN == 1
 
+void hash_init_accel(void)
+{
+	crc32c_init_accel();
+	blake2_init_accel();
+	sha256_init_accel();
+}
+
 int hash_sha256(const u8 *buf, size_t len, u8 *out)
 {
 	SHA256Context context;
@@ -98,6 +88,11 @@ int hash_blake2b(const u8 *buf, size_t len, u8 *out)
 
 #include <gcrypt.h>
 
+void hash_init_accel(void)
+{
+	crc32c_init_accel();
+}
+
 int hash_sha256(const u8 *buf, size_t len, u8 *out)
 {
 	gcry_md_hash_buffer(GCRY_MD_SHA256, out, buf, len);
@@ -117,6 +112,11 @@ int hash_blake2b(const u8 *buf, size_t len, u8 *out)
 #include <sodium/crypto_hash_sha256.h>
 #include <sodium/crypto_generichash_blake2b.h>
 
+void hash_init_accel(void)
+{
+	crc32c_init_accel();
+}
+
 int hash_sha256(const u8 *buf, size_t len, u8 *out)
 {
 	return crypto_hash_sha256(out, buf, len);
@@ -134,6 +134,11 @@ int hash_blake2b(const u8 *buf, size_t len, u8 *out)
 
 #include <kcapi.h>
 
+void hash_init_accel(void)
+{
+	crc32c_init_accel();
+}
+
 int hash_sha256(const u8 *buf, size_t len, u8 *out)
 {
 	static struct kcapi_handle *handle = NULL;

crypto/hash.h

@@ -28,7 +28,5 @@ int hash_blake2b(const u8 *buf, size_t length, u8 *out);
 
 void hash_init_accel(void);
 void hash_init_crc32c(void);
-void hash_init_blake2(void);
-void hash_init_sha256(void);
 
 #endif

crypto/sha224-256.c

@@ -97,7 +97,7 @@ static uint32_t SHA256_H0[SHA256HashSize/4] = {
 
 static void (*sha256_process_message_block)(SHA256Context *context) = SHA224_256ProcessMessageBlock;
 
-#ifdef __SHA__
+#if HAVE_CFLAG_msha == 1
 void sha256_process_x86(uint32_t state[8], const uint8_t data[], uint32_t length);
 
 static void sha256_process_x86_dispatch(SHA256Context *context)
@@ -110,7 +110,7 @@ static void sha256_process_x86_dispatch(SHA256Context *context)
 
 void sha256_init_accel(void)
 {
-#ifdef __SHA__
+#if HAVE_CFLAG_msha == 1
 	if (cpu_has_feature(CPU_FLAG_SHA))
 		sha256_process_message_block = sha256_process_x86_dispatch;
 	else