[PATCH v2 08/10] crypto: x86/aegis128 - take advantage of block-aligned len

[Eric Biggers <ebiggers@kernel.org>]

From: Eric Biggers <ebiggers@google.com>

Update a caller of aegis128_aesni_ad() to round down the length to a
block boundary.  After that, aegis128_aesni_ad(), aegis128_aesni_enc(),
and aegis128_aesni_dec() are only passed whole blocks.  Update the
assembly code to take advantage of that, which eliminates some unneeded
instructions.  For aegis128_aesni_enc() and aegis128_aesni_dec(), the
length is also always nonzero, so stop checking for zero length.

Reviewed-by: Ondrej Mosnacek <omosnace@redhat.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 arch/x86/crypto/aegis128-aesni-asm.S  | 37 +++++++++++----------------
 arch/x86/crypto/aegis128-aesni-glue.c |  4 +--
 2 files changed, 17 insertions(+), 24 deletions(-)

diff --git a/arch/x86/crypto/aegis128-aesni-asm.S b/arch/x86/crypto/aegis128-aesni-asm.S
index e650330ef6951..345b1eafe45af 100644
--- a/arch/x86/crypto/aegis128-aesni-asm.S
+++ b/arch/x86/crypto/aegis128-aesni-asm.S
@@ -188,19 +188,21 @@ SYM_FUNC_START(aegis128_aesni_init)
 SYM_FUNC_END(aegis128_aesni_init)
 
 /*
  * void aegis128_aesni_ad(struct aegis_state *state, const u8 *data,
  *			  unsigned int len);
+ *
+ * len must be a multiple of 16.
  */
 SYM_FUNC_START(aegis128_aesni_ad)
 	.set STATEP, %rdi
 	.set SRC, %rsi
 	.set LEN, %edx
 	FRAME_BEGIN
 
-	cmp $0x10, LEN
-	jb .Lad_out
+	test LEN, LEN
+	jz .Lad_out
 
 	/* load the state: */
 	movdqu 0x00(STATEP), STATE0
 	movdqu 0x10(STATEP), STATE1
 	movdqu 0x20(STATEP), STATE2
@@ -211,40 +213,35 @@ SYM_FUNC_START(aegis128_aesni_ad)
 .Lad_loop:
 	movdqu 0x00(SRC), MSG
 	aegis128_update
 	pxor MSG, STATE4
 	sub $0x10, LEN
-	cmp $0x10, LEN
-	jl .Lad_out_1
+	jz .Lad_out_1
 
 	movdqu 0x10(SRC), MSG
 	aegis128_update
 	pxor MSG, STATE3
 	sub $0x10, LEN
-	cmp $0x10, LEN
-	jl .Lad_out_2
+	jz .Lad_out_2
 
 	movdqu 0x20(SRC), MSG
 	aegis128_update
 	pxor MSG, STATE2
 	sub $0x10, LEN
-	cmp $0x10, LEN
-	jl .Lad_out_3
+	jz .Lad_out_3
 
 	movdqu 0x30(SRC), MSG
 	aegis128_update
 	pxor MSG, STATE1
 	sub $0x10, LEN
-	cmp $0x10, LEN
-	jl .Lad_out_4
+	jz .Lad_out_4
 
 	movdqu 0x40(SRC), MSG
 	aegis128_update
 	pxor MSG, STATE0
 	sub $0x10, LEN
-	cmp $0x10, LEN
-	jl .Lad_out_0
+	jz .Lad_out_0
 
 	add $0x50, SRC
 	jmp .Lad_loop
 
 	/* store the state: */
@@ -310,28 +307,26 @@ SYM_FUNC_END(aegis128_aesni_ad)
 
 	aegis128_update
 	pxor MSG, \s4
 
 	sub $0x10, LEN
-	cmp $0x10, LEN
-	jl .Lenc_out_\i
+	jz .Lenc_out_\i
 .endm
 
 /*
  * void aegis128_aesni_enc(struct aegis_state *state, const u8 *src, u8 *dst,
  *			   unsigned int len);
+ *
+ * len must be nonzero and a multiple of 16.
  */
 SYM_FUNC_START(aegis128_aesni_enc)
 	.set STATEP, %rdi
 	.set SRC, %rsi
 	.set DST, %rdx
 	.set LEN, %ecx
 	FRAME_BEGIN
 
-	cmp $0x10, LEN
-	jb .Lenc_out
-
 	/* load the state: */
 	movdqu 0x00(STATEP), STATE0
 	movdqu 0x10(STATEP), STATE1
 	movdqu 0x20(STATEP), STATE2
 	movdqu 0x30(STATEP), STATE3
@@ -457,28 +452,26 @@ SYM_FUNC_END(aegis128_aesni_enc_tail)
 
 	aegis128_update
 	pxor MSG, \s4
 
 	sub $0x10, LEN
-	cmp $0x10, LEN
-	jl .Ldec_out_\i
+	jz .Ldec_out_\i
 .endm
 
 /*
  * void aegis128_aesni_dec(struct aegis_state *state, const u8 *src, u8 *dst,
  *			   unsigned int len);
+ *
+ * len must be nonzero and a multiple of 16.
  */
 SYM_FUNC_START(aegis128_aesni_dec)
 	.set STATEP, %rdi
 	.set SRC, %rsi
 	.set DST, %rdx
 	.set LEN, %ecx
 	FRAME_BEGIN
 
-	cmp $0x10, LEN
-	jb .Ldec_out
-
 	/* load the state: */
 	movdqu 0x00(STATEP), STATE0
 	movdqu 0x10(STATEP), STATE1
 	movdqu 0x20(STATEP), STATE2
 	movdqu 0x30(STATEP), STATE3
diff --git a/arch/x86/crypto/aegis128-aesni-glue.c b/arch/x86/crypto/aegis128-aesni-glue.c
index 9555958e4089d..c19d8e3d96a35 100644
--- a/arch/x86/crypto/aegis128-aesni-glue.c
+++ b/arch/x86/crypto/aegis128-aesni-glue.c
@@ -85,12 +85,12 @@ static void crypto_aegis128_aesni_process_ad(
 				pos = 0;
 				left -= fill;
 				src += fill;
 			}
 
-			aegis128_aesni_ad(state, src, left);
-
+			aegis128_aesni_ad(state, src,
+					  left & ~(AEGIS128_BLOCK_SIZE - 1));
 			src += left & ~(AEGIS128_BLOCK_SIZE - 1);
 			left &= AEGIS128_BLOCK_SIZE - 1;
 		}
 
 		memcpy(buf.bytes + pos, src, left);
-- 
2.47.0