From: Eric Biggers ebiggers@google.com
[ Upstream commit d2f2516a3882c0c6463e33c9b112b39bd483f821 ]
The new ARM BLAKE2s code doesn't work correctly (fails the self-tests) in big endian kernel builds because it doesn't swap the endianness of the message words when loading them. Fix this.
Fixes: 5172d322d34c ("crypto: arm/blake2s - add ARM scalar optimized BLAKE2s") Signed-off-by: Eric Biggers ebiggers@google.com Acked-by: Ard Biesheuvel ardb@kernel.org Signed-off-by: Herbert Xu herbert@gondor.apana.org.au Signed-off-by: Sasha Levin sashal@kernel.org --- arch/arm/crypto/blake2s-core.S | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+)
diff --git a/arch/arm/crypto/blake2s-core.S b/arch/arm/crypto/blake2s-core.S index bed897e9a181..86345751bbf3 100644 --- a/arch/arm/crypto/blake2s-core.S +++ b/arch/arm/crypto/blake2s-core.S @@ -8,6 +8,7 @@ */
#include <linux/linkage.h> +#include <asm/assembler.h>
// Registers used to hold message words temporarily. There aren't // enough ARM registers to hold the whole message block, so we have to @@ -38,6 +39,23 @@ #endif .endm
+.macro _le32_bswap a, tmp +#ifdef __ARMEB__ + rev_l \a, \tmp +#endif +.endm + +.macro _le32_bswap_8x a, b, c, d, e, f, g, h, tmp + _le32_bswap \a, \tmp + _le32_bswap \b, \tmp + _le32_bswap \c, \tmp + _le32_bswap \d, \tmp + _le32_bswap \e, \tmp + _le32_bswap \f, \tmp + _le32_bswap \g, \tmp + _le32_bswap \h, \tmp +.endm + // Execute a quarter-round of BLAKE2s by mixing two columns or two diagonals. // (a0, b0, c0, d0) and (a1, b1, c1, d1) give the registers containing the two // columns/diagonals. s0-s1 are the word offsets to the message words the first @@ -180,8 +198,10 @@ ENTRY(blake2s_compress_arch) tst r1, #3 bne .Lcopy_block_misaligned ldmia r1!, {r2-r9} + _le32_bswap_8x r2, r3, r4, r5, r6, r7, r8, r9, r14 stmia r12!, {r2-r9} ldmia r1!, {r2-r9} + _le32_bswap_8x r2, r3, r4, r5, r6, r7, r8, r9, r14 stmia r12, {r2-r9} .Lcopy_block_done: str r1, [sp, #68] // Update message pointer @@ -268,6 +288,7 @@ ENTRY(blake2s_compress_arch) 1: #ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS ldr r3, [r1], #4 + _le32_bswap r3, r4 #else ldrb r3, [r1, #0] ldrb r4, [r1, #1]