crypto: x86/aegis128 - access 32-bit arguments as 32-bit

Fix the AEGIS assembly code to access 'unsigned int' arguments as 32-bit
values instead of 64-bit, since the upper bits of the corresponding
64-bit registers are not guaranteed to be zero.

Note: there haven't been any reports of this bug actually causing
incorrect behavior.  Neither gcc nor clang guarantee zero-extension to
64 bits, but zero-extension is likely to happen in practice because most
instructions that operate on 32-bit registers zero-extend to 64 bits.

Fixes: 1d373d4 ("crypto: x86 - Add optimized AEGIS implementations")
Cc: stable@vger.kernel.org
Reviewed-by: Ondrej Mosnacek <omosnace@redhat.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Author: ebiggers

arch/x86/crypto/aegis128-aesni-asm.S

@@ -21,7 +21,7 @@
 #define T1	%xmm7
 
 #define STATEP	%rdi
-#define LEN	%rsi
+#define LEN	%esi
 #define SRC	%rdx
 #define DST	%rcx
 
@@ -76,32 +76,32 @@ SYM_FUNC_START_LOCAL(__load_partial)
 	xor %r9d, %r9d
 	pxor MSG, MSG
 
-	mov LEN, %r8
+	mov LEN, %r8d
 	and $0x1, %r8
 	jz .Lld_partial_1
 
-	mov LEN, %r8
+	mov LEN, %r8d
 	and $0x1E, %r8
 	add SRC, %r8
 	mov (%r8), %r9b
 
 .Lld_partial_1:
-	mov LEN, %r8
+	mov LEN, %r8d
 	and $0x2, %r8
 	jz .Lld_partial_2
 
-	mov LEN, %r8
+	mov LEN, %r8d
 	and $0x1C, %r8
 	add SRC, %r8
 	shl $0x10, %r9
 	mov (%r8), %r9w
 
 .Lld_partial_2:
-	mov LEN, %r8
+	mov LEN, %r8d
 	and $0x4, %r8
 	jz .Lld_partial_4
 
-	mov LEN, %r8
+	mov LEN, %r8d
 	and $0x18, %r8
 	add SRC, %r8
 	shl $32, %r9
@@ -111,11 +111,11 @@ SYM_FUNC_START_LOCAL(__load_partial)
 .Lld_partial_4:
 	movq %r9, MSG
 
-	mov LEN, %r8
+	mov LEN, %r8d
 	and $0x8, %r8
 	jz .Lld_partial_8
 
-	mov LEN, %r8
+	mov LEN, %r8d
 	and $0x10, %r8
 	add SRC, %r8
 	pslldq $8, MSG
@@ -139,7 +139,7 @@ SYM_FUNC_END(__load_partial)
  *   %r10
  */
 SYM_FUNC_START_LOCAL(__store_partial)
-	mov LEN, %r8
+	mov LEN, %r8d
 	mov DST, %r9
 
 	movq T0, %r10
@@ -677,7 +677,7 @@ SYM_TYPED_FUNC_START(crypto_aegis128_aesni_dec_tail)
 	call __store_partial
 
 	/* mask with byte count: */
-	movq LEN, T0
+	movd LEN, T0
 	punpcklbw T0, T0
 	punpcklbw T0, T0
 	punpcklbw T0, T0
@@ -702,7 +702,8 @@ SYM_FUNC_END(crypto_aegis128_aesni_dec_tail)
 
 /*
  * void crypto_aegis128_aesni_final(void *state, void *tag_xor,
- *                                  u64 assoclen, u64 cryptlen);
+ *                                  unsigned int assoclen,
+ *                                  unsigned int cryptlen);
  */
 SYM_FUNC_START(crypto_aegis128_aesni_final)
 	FRAME_BEGIN
@@ -715,8 +716,8 @@ SYM_FUNC_START(crypto_aegis128_aesni_final)
 	movdqu 0x40(STATEP), STATE4
 
 	/* prepare length block: */
-	movq %rdx, MSG
-	movq %rcx, T0
+	movd %edx, MSG
+	movd %ecx, T0
 	pslldq $8, T0
 	pxor T0, MSG
 	psllq $3, MSG /* multiply by 8 (to get bit count) */