crypto: x86/camellia-aesni-avx - add more optimized XTS code

Add more optimized XTS code for camellia-aesni-avx, for smaller stack usage
and small boost for speed.

tcrypt results, with Intel i5-2450M:
        enc     dec
16B     1.10x   1.01x
64B     0.82x   0.77x
256B    1.14x   1.10x
1024B   1.17x   1.16x
8192B   1.10x   1.11x

Since XTS is practically always used with data blocks of size 512 bytes or
more, I chose to not make use of camellia-2way for block sized smaller than
256 bytes. This causes slower result in tcrypt for 64 bytes.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

arch/x86/crypto/camellia-aesni-avx-asm_64.S

@@ -1,7 +1,7 @@
 /*
  * x86_64/AVX/AES-NI assembler implementation of Camellia
  *
- * Copyright © 2012 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -589,6 +589,10 @@ ENDPROC(roundsm16_x4_x5_x6_x7_x0_x1_x2_x3_y4_y5_y6_y7_y0_y1_y2_y3_ab)
 .Lbswap128_mask:
 	.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
 
+/* For XTS mode IV generation */
+.Lxts_gf128mul_and_shl1_mask:
+	.byte 0x87, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0
+
 /*
  * pre-SubByte transform
  *
@@ -1090,3 +1094,177 @@ ENTRY(camellia_ctr_16way)
 
 	ret;
 ENDPROC(camellia_ctr_16way)
+
+#define gf128mul_x_ble(iv, mask, tmp) \
+	vpsrad $31, iv, tmp; \
+	vpaddq iv, iv, iv; \
+	vpshufd $0x13, tmp, tmp; \
+	vpand mask, tmp, tmp; \
+	vpxor tmp, iv, iv;
+
+.align 8
+camellia_xts_crypt_16way:
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst (16 blocks)
+	 *	%rdx: src (16 blocks)
+	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
+	 *	%r8: index for input whitening key
+	 *	%r9: pointer to  __camellia_enc_blk16 or __camellia_dec_blk16
+	 */
+
+	subq $(16 * 16), %rsp;
+	movq %rsp, %rax;
+
+	vmovdqa .Lxts_gf128mul_and_shl1_mask, %xmm14;
+
+	/* load IV */
+	vmovdqu (%rcx), %xmm0;
+	vpxor 0 * 16(%rdx), %xmm0, %xmm15;
+	vmovdqu %xmm15, 15 * 16(%rax);
+	vmovdqu %xmm0, 0 * 16(%rsi);
+
+	/* construct IVs */
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 1 * 16(%rdx), %xmm0, %xmm15;
+	vmovdqu %xmm15, 14 * 16(%rax);
+	vmovdqu %xmm0, 1 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 2 * 16(%rdx), %xmm0, %xmm13;
+	vmovdqu %xmm0, 2 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 3 * 16(%rdx), %xmm0, %xmm12;
+	vmovdqu %xmm0, 3 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 4 * 16(%rdx), %xmm0, %xmm11;
+	vmovdqu %xmm0, 4 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 5 * 16(%rdx), %xmm0, %xmm10;
+	vmovdqu %xmm0, 5 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 6 * 16(%rdx), %xmm0, %xmm9;
+	vmovdqu %xmm0, 6 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 7 * 16(%rdx), %xmm0, %xmm8;
+	vmovdqu %xmm0, 7 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 8 * 16(%rdx), %xmm0, %xmm7;
+	vmovdqu %xmm0, 8 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 9 * 16(%rdx), %xmm0, %xmm6;
+	vmovdqu %xmm0, 9 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 10 * 16(%rdx), %xmm0, %xmm5;
+	vmovdqu %xmm0, 10 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 11 * 16(%rdx), %xmm0, %xmm4;
+	vmovdqu %xmm0, 11 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 12 * 16(%rdx), %xmm0, %xmm3;
+	vmovdqu %xmm0, 12 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 13 * 16(%rdx), %xmm0, %xmm2;
+	vmovdqu %xmm0, 13 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 14 * 16(%rdx), %xmm0, %xmm1;
+	vmovdqu %xmm0, 14 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vpxor 15 * 16(%rdx), %xmm0, %xmm15;
+	vmovdqu %xmm15, 0 * 16(%rax);
+	vmovdqu %xmm0, 15 * 16(%rsi);
+
+	gf128mul_x_ble(%xmm0, %xmm14, %xmm15);
+	vmovdqu %xmm0, (%rcx);
+
+	/* inpack16_pre: */
+	vmovq (key_table)(CTX, %r8, 8), %xmm15;
+	vpshufb .Lpack_bswap, %xmm15, %xmm15;
+	vpxor 0 * 16(%rax), %xmm15, %xmm0;
+	vpxor %xmm1, %xmm15, %xmm1;
+	vpxor %xmm2, %xmm15, %xmm2;
+	vpxor %xmm3, %xmm15, %xmm3;
+	vpxor %xmm4, %xmm15, %xmm4;
+	vpxor %xmm5, %xmm15, %xmm5;
+	vpxor %xmm6, %xmm15, %xmm6;
+	vpxor %xmm7, %xmm15, %xmm7;
+	vpxor %xmm8, %xmm15, %xmm8;
+	vpxor %xmm9, %xmm15, %xmm9;
+	vpxor %xmm10, %xmm15, %xmm10;
+	vpxor %xmm11, %xmm15, %xmm11;
+	vpxor %xmm12, %xmm15, %xmm12;
+	vpxor %xmm13, %xmm15, %xmm13;
+	vpxor 14 * 16(%rax), %xmm15, %xmm14;
+	vpxor 15 * 16(%rax), %xmm15, %xmm15;
+
+	call *%r9;
+
+	addq $(16 * 16), %rsp;
+
+	vpxor 0 * 16(%rsi), %xmm7, %xmm7;
+	vpxor 1 * 16(%rsi), %xmm6, %xmm6;
+	vpxor 2 * 16(%rsi), %xmm5, %xmm5;
+	vpxor 3 * 16(%rsi), %xmm4, %xmm4;
+	vpxor 4 * 16(%rsi), %xmm3, %xmm3;
+	vpxor 5 * 16(%rsi), %xmm2, %xmm2;
+	vpxor 6 * 16(%rsi), %xmm1, %xmm1;
+	vpxor 7 * 16(%rsi), %xmm0, %xmm0;
+	vpxor 8 * 16(%rsi), %xmm15, %xmm15;
+	vpxor 9 * 16(%rsi), %xmm14, %xmm14;
+	vpxor 10 * 16(%rsi), %xmm13, %xmm13;
+	vpxor 11 * 16(%rsi), %xmm12, %xmm12;
+	vpxor 12 * 16(%rsi), %xmm11, %xmm11;
+	vpxor 13 * 16(%rsi), %xmm10, %xmm10;
+	vpxor 14 * 16(%rsi), %xmm9, %xmm9;
+	vpxor 15 * 16(%rsi), %xmm8, %xmm8;
+	write_output(%xmm7, %xmm6, %xmm5, %xmm4, %xmm3, %xmm2, %xmm1, %xmm0,
+		     %xmm15, %xmm14, %xmm13, %xmm12, %xmm11, %xmm10, %xmm9,
+		     %xmm8, %rsi);
+
+	ret;
+ENDPROC(camellia_xts_crypt_16way)
+
+ENTRY(camellia_xts_enc_16way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst (16 blocks)
+	 *	%rdx: src (16 blocks)
+	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
+	 */
+	xorl %r8d, %r8d; /* input whitening key, 0 for enc */
+
+	leaq __camellia_enc_blk16, %r9;
+
+	jmp camellia_xts_crypt_16way;
+ENDPROC(camellia_xts_enc_16way)
+
+ENTRY(camellia_xts_dec_16way)
+	/* input:
+	 *	%rdi: ctx, CTX
+	 *	%rsi: dst (16 blocks)
+	 *	%rdx: src (16 blocks)
+	 *	%rcx: iv (t ⊕ αⁿ ∈ GF(2¹²⁸))
+	 */
+
+	cmpl $16, key_length(CTX);
+	movl $32, %r8d;
+	movl $24, %eax;
+	cmovel %eax, %r8d;  /* input whitening key, last for dec */
+
+	leaq __camellia_dec_blk16, %r9;
+
+	jmp camellia_xts_crypt_16way;
+ENDPROC(camellia_xts_dec_16way)

arch/x86/crypto/camellia_aesni_avx_glue.c

@@ -1,7 +1,7 @@
 /*
  * Glue Code for x86_64/AVX/AES-NI assembler optimized version of Camellia
  *
- * Copyright © 2012 Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
+ * Copyright © 2012-2013 Jussi Kivilinna <jussi.kivilinna@iki.fi>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -37,6 +37,23 @@ asmlinkage void camellia_cbc_dec_16way(struct camellia_ctx *ctx, u8 *dst,
 asmlinkage void camellia_ctr_16way(struct camellia_ctx *ctx, u8 *dst,
 				   const u8 *src, le128 *iv);
 
+asmlinkage void camellia_xts_enc_16way(struct camellia_ctx *ctx, u8 *dst,
+				       const u8 *src, le128 *iv);
+asmlinkage void camellia_xts_dec_16way(struct camellia_ctx *ctx, u8 *dst,
+				       const u8 *src, le128 *iv);
+
+static void camellia_xts_enc(void *ctx, u128 *dst, const u128 *src, le128 *iv)
+{
+	glue_xts_crypt_128bit_one(ctx, dst, src, iv,
+				  GLUE_FUNC_CAST(camellia_enc_blk));
+}
+
+static void camellia_xts_dec(void *ctx, u128 *dst, const u128 *src, le128 *iv)
+{
+	glue_xts_crypt_128bit_one(ctx, dst, src, iv,
+				  GLUE_FUNC_CAST(camellia_dec_blk));
+}
+
 static const struct common_glue_ctx camellia_enc = {
 	.num_funcs = 3,
 	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,
@@ -69,6 +86,19 @@ static const struct common_glue_ctx camellia_ctr = {
 	} }
 };
 
+static const struct common_glue_ctx camellia_enc_xts = {
+	.num_funcs = 2,
+	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,
+
+	.funcs = { {
+		.num_blocks = CAMELLIA_AESNI_PARALLEL_BLOCKS,
+		.fn_u = { .xts = GLUE_XTS_FUNC_CAST(camellia_xts_enc_16way) }
+	}, {
+		.num_blocks = 1,
+		.fn_u = { .xts = GLUE_XTS_FUNC_CAST(camellia_xts_enc) }
+	} }
+};
+
 static const struct common_glue_ctx camellia_dec = {
 	.num_funcs = 3,
 	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,
@@ -101,6 +131,19 @@ static const struct common_glue_ctx camellia_dec_cbc = {
 	} }
 };
 
+static const struct common_glue_ctx camellia_dec_xts = {
+	.num_funcs = 2,
+	.fpu_blocks_limit = CAMELLIA_AESNI_PARALLEL_BLOCKS,
+
+	.funcs = { {
+		.num_blocks = CAMELLIA_AESNI_PARALLEL_BLOCKS,
+		.fn_u = { .xts = GLUE_XTS_FUNC_CAST(camellia_xts_dec_16way) }
+	}, {
+		.num_blocks = 1,
+		.fn_u = { .xts = GLUE_XTS_FUNC_CAST(camellia_xts_dec) }
+	} }
+};
+
 static int ecb_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
 		       struct scatterlist *src, unsigned int nbytes)
 {
@@ -261,54 +304,20 @@ static int xts_encrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
 		       struct scatterlist *src, unsigned int nbytes)
 {
 	struct camellia_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
-	be128 buf[CAMELLIA_AESNI_PARALLEL_BLOCKS];
-	struct crypt_priv crypt_ctx = {
-		.ctx = &ctx->crypt_ctx,
-		.fpu_enabled = false,
-	};
-	struct xts_crypt_req req = {
-		.tbuf = buf,
-		.tbuflen = sizeof(buf),
-
-		.tweak_ctx = &ctx->tweak_ctx,
-		.tweak_fn = XTS_TWEAK_CAST(camellia_enc_blk),
-		.crypt_ctx = &crypt_ctx,
-		.crypt_fn = encrypt_callback,
-	};
-	int ret;
-
-	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
-	ret = xts_crypt(desc, dst, src, nbytes, &req);
-	camellia_fpu_end(crypt_ctx.fpu_enabled);
 
-	return ret;
+	return glue_xts_crypt_128bit(&camellia_enc_xts, desc, dst, src, nbytes,
+				     XTS_TWEAK_CAST(camellia_enc_blk),
+				     &ctx->tweak_ctx, &ctx->crypt_ctx);
 }
 
 static int xts_decrypt(struct blkcipher_desc *desc, struct scatterlist *dst,
 		       struct scatterlist *src, unsigned int nbytes)
 {
 	struct camellia_xts_ctx *ctx = crypto_blkcipher_ctx(desc->tfm);
-	be128 buf[CAMELLIA_AESNI_PARALLEL_BLOCKS];
-	struct crypt_priv crypt_ctx = {
-		.ctx = &ctx->crypt_ctx,
-		.fpu_enabled = false,
-	};
-	struct xts_crypt_req req = {
-		.tbuf = buf,
-		.tbuflen = sizeof(buf),
 
-		.tweak_ctx = &ctx->tweak_ctx,
-		.tweak_fn = XTS_TWEAK_CAST(camellia_enc_blk),
-		.crypt_ctx = &crypt_ctx,
-		.crypt_fn = decrypt_callback,
-	};
-	int ret;
-
-	desc->flags &= ~CRYPTO_TFM_REQ_MAY_SLEEP;
-	ret = xts_crypt(desc, dst, src, nbytes, &req);
-	camellia_fpu_end(crypt_ctx.fpu_enabled);
-
-	return ret;
+	return glue_xts_crypt_128bit(&camellia_dec_xts, desc, dst, src, nbytes,
+				     XTS_TWEAK_CAST(camellia_enc_blk),
+				     &ctx->tweak_ctx, &ctx->crypt_ctx);
 }
 
 static struct crypto_alg cmll_algs[10] = { {