[PATCH 2/2] arm64/crypto: improve performance of GHASH algorithm

[Ard Biesheuvel]

This patches modifies the GHASH secure hash implementation to switch to a
faster, polynomial multiplication based reduction instead of one that uses
shifts and rotates.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel at linaro.org>
---
 arch/arm64/crypto/ghash-ce-core.S | 92 ++++++++++++++++-----------------------
 arch/arm64/crypto/ghash-ce-glue.c |  4 +-
 2 files changed, 40 insertions(+), 56 deletions(-)

diff --git a/arch/arm64/crypto/ghash-ce-core.S b/arch/arm64/crypto/ghash-ce-core.S
index b9e6eaf41c9b..dc457015884e 100644
--- a/arch/arm64/crypto/ghash-ce-core.S
+++ b/arch/arm64/crypto/ghash-ce-core.S
@@ -3,14 +3,6 @@
  *
  * Copyright (C) 2014 Linaro Ltd. <ard.biesheuvel at linaro.org>
  *
- * Based on arch/x86/crypto/ghash-pmullni-intel_asm.S
- *
- * Copyright (c) 2009 Intel Corp.
- *   Author: Huang Ying <ying.huang at intel.com>
- *           Vinodh Gopal
- *           Erdinc Ozturk
- *           Deniz Karakoyunlu
- *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of the GNU General Public License version 2 as published
  * by the Free Software Foundation.
@@ -19,13 +11,15 @@
 #include <linux/linkage.h>
 #include <asm/assembler.h>
 
-	DATA	.req	v0
-	SHASH	.req	v1
-	IN1	.req	v2
+	SHASH	.req	v0
+	SHASH2	.req	v1
 	T1	.req	v2
 	T2	.req	v3
-	T3	.req	v4
-	VZR	.req	v5
+	MASK	.req	v4
+	XL	.req	v5
+	XM	.req	v6
+	XH	.req	v7
+	IN1	.req	v7
 
 	.text
 	.arch		armv8-a+crypto
@@ -35,61 +29,51 @@
 	 *			   struct ghash_key const *k, const char *head)
 	 */
 ENTRY(pmull_ghash_update)
-	ld1		{DATA.16b}, [x1]
 	ld1		{SHASH.16b}, [x3]
-	eor		VZR.16b, VZR.16b, VZR.16b
+	ld1		{XL.16b}, [x1]
+	movi		MASK.16b, #0xe1
+	ext		SHASH2.16b, SHASH.16b, SHASH.16b, #8
+	shl		MASK.2d, MASK.2d, #57
+	eor		SHASH2.16b, SHASH2.16b, SHASH.16b
 
 	/* do the head block first, if supplied */
 	cbz		x4, 0f
-	ld1		{IN1.2d}, [x4]
+	ld1		{T1.2d}, [x4]
 	b		1f
 
-0:	ld1		{IN1.2d}, [x2], #16
+0:	ld1		{T1.2d}, [x2], #16
 	sub		w0, w0, #1
-1:	ext		IN1.16b, IN1.16b, IN1.16b, #8
-CPU_LE(	rev64		IN1.16b, IN1.16b	)
-	eor		DATA.16b, DATA.16b, IN1.16b
 
-	/* multiply DATA by SHASH in GF(2^128) */
-	ext		T2.16b, DATA.16b, DATA.16b, #8
-	ext		T3.16b, SHASH.16b, SHASH.16b, #8
-	eor		T2.16b, T2.16b, DATA.16b
-	eor		T3.16b, T3.16b, SHASH.16b
+1:	/* multiply XL by SHASH in GF(2^128) */
+CPU_LE(	rev64		T1.16b, T1.16b	)
 
-	pmull2		T1.1q, SHASH.2d, DATA.2d	// a1 * b1
-	pmull		DATA.1q, SHASH.1d, DATA.1d	// a0 * b0
-	pmull		T2.1q, T2.1d, T3.1d		// (a1 + a0)(b1 + b0)
-	eor		T2.16b, T2.16b, T1.16b		// (a0 * b1) + (a1 * b0)
-	eor		T2.16b, T2.16b, DATA.16b
+	ext		T2.16b, XL.16b, XL.16b, #8
+	ext		IN1.16b, T1.16b, T1.16b, #8
+	eor		T1.16b, T1.16b, T2.16b
+	eor		XL.16b, XL.16b, IN1.16b
 
-	ext		T3.16b, VZR.16b, T2.16b, #8
-	ext		T2.16b, T2.16b, VZR.16b, #8
-	eor		DATA.16b, DATA.16b, T3.16b
-	eor		T1.16b, T1.16b, T2.16b	// <T1:DATA> is result of
-						// carry-less multiplication
+	pmull2		XH.1q, SHASH.2d, XL.2d		// a1 * b1
+	eor		T1.16b, T1.16b, XL.16b
+	pmull		XL.1q, SHASH.1d, XL.1d		// a0 * b0
+	pmull		XM.1q, SHASH2.1d, T1.1d		// (a1 + a0)(b1 + b0)
 
-	/* first phase of the reduction */
-	shl		T3.2d, DATA.2d, #1
-	eor		T3.16b, T3.16b, DATA.16b
-	shl		T3.2d, T3.2d, #5
-	eor		T3.16b, T3.16b, DATA.16b
-	shl		T3.2d, T3.2d, #57
-	ext		T2.16b, VZR.16b, T3.16b, #8
-	ext		T3.16b, T3.16b, VZR.16b, #8
-	eor		DATA.16b, DATA.16b, T2.16b
-	eor		T1.16b, T1.16b, T3.16b
+	ext		T1.16b, XL.16b, XH.16b, #8
+	eor		T2.16b, XL.16b, XH.16b
+	eor		XM.16b, XM.16b, T1.16b
+	eor		XM.16b, XM.16b, T2.16b
+	pmull		T2.1q, XL.1d, MASK.1d
 
-	/* second phase of the reduction */
-	ushr		T2.2d, DATA.2d, #5
-	eor		T2.16b, T2.16b, DATA.16b
-	ushr		T2.2d, T2.2d, #1
-	eor		T2.16b, T2.16b, DATA.16b
-	ushr		T2.2d, T2.2d, #1
-	eor		T1.16b, T1.16b, T2.16b
-	eor		DATA.16b, DATA.16b, T1.16b
+	mov		XH.d[0], XM.d[1]
+	mov		XM.d[1], XL.d[0]
+
+	eor		XL.16b, XM.16b, T2.16b
+	ext		T2.16b, XL.16b, XL.16b, #8
+	pmull		XL.1q, XL.1d, MASK.1d
+	eor		T2.16b, T2.16b, XH.16b
+	eor		XL.16b, XL.16b, T2.16b
 
 	cbnz		w0, 0b
 
-	st1		{DATA.16b}, [x1]
+	st1		{XL.16b}, [x1]
 	ret
 ENDPROC(pmull_ghash_update)
diff --git a/arch/arm64/crypto/ghash-ce-glue.c b/arch/arm64/crypto/ghash-ce-glue.c
index ef6aa69c4e0c..833ec1e3f3e9 100644
--- a/arch/arm64/crypto/ghash-ce-glue.c
+++ b/arch/arm64/crypto/ghash-ce-glue.c
@@ -67,7 +67,7 @@ static int ghash_update(struct shash_desc *desc, const u8 *src,
 		blocks = len / GHASH_BLOCK_SIZE;
 		len %= GHASH_BLOCK_SIZE;
 
-		kernel_neon_begin_partial(6);
+		kernel_neon_begin_partial(8);
 		pmull_ghash_update(blocks, ctx->digest, src, key,
 				   partial ? ctx->buf : NULL);
 		kernel_neon_end();
@@ -89,7 +89,7 @@ static int ghash_final(struct shash_desc *desc, u8 *dst)
 
 		memset(ctx->buf + partial, 0, GHASH_BLOCK_SIZE - partial);
 
-		kernel_neon_begin_partial(6);
+		kernel_neon_begin_partial(8);
 		pmull_ghash_update(1, ctx->digest, ctx->buf, key, NULL);
 		kernel_neon_end();
 	}
-- 
1.8.3.2