From 793ff5ffc105a5962ab4c978d08988e3f8dc93aa Mon Sep 17 00:00:00 2001
From: Eric Biggers <ebiggers@google.com>
Date: Thu, 10 Jan 2019 12:17:57 -0800
Subject: crypto: x86/aesni-gcm - make 'struct aesni_gcm_tfm_s' static const

Add missing static keywords to fix the following sparse warnings:

    arch/x86/crypto/aesni-intel_glue.c:197:24: warning: symbol 'aesni_gcm_tfm_sse' was not declared. Should it be static?
    arch/x86/crypto/aesni-intel_glue.c:246:24: warning: symbol 'aesni_gcm_tfm_avx_gen2' was not declared. Should it be static?
    arch/x86/crypto/aesni-intel_glue.c:291:24: warning: symbol 'aesni_gcm_tfm_avx_gen4' was not declared. Should it be static?

I also made the affected structures 'const', and adjusted the
indentation in the struct definition to not be insane.

Cc: Dave Watson <davejwatson@fb.com>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/x86/crypto/aesni-intel_glue.c | 34 +++++++++++++---------------------
 1 file changed, 13 insertions(+), 21 deletions(-)

(limited to 'arch')

diff --git a/arch/x86/crypto/aesni-intel_glue.c b/arch/x86/crypto/aesni-intel_glue.c
index 1321700d6647..9b5ccde3ef31 100644
--- a/arch/x86/crypto/aesni-intel_glue.c
+++ b/arch/x86/crypto/aesni-intel_glue.c
@@ -175,26 +175,18 @@ asmlinkage void aesni_gcm_finalize(void *ctx,
 				   struct gcm_context_data *gdata,
 				   u8 *auth_tag, unsigned long auth_tag_len);
 
-static struct aesni_gcm_tfm_s {
-void (*init)(void *ctx,
-				struct gcm_context_data *gdata,
-				u8 *iv,
-				u8 *hash_subkey, const u8 *aad,
-				unsigned long aad_len);
-void (*enc_update)(void *ctx,
-					struct gcm_context_data *gdata, u8 *out,
-					const u8 *in,
-					unsigned long plaintext_len);
-void (*dec_update)(void *ctx,
-					struct gcm_context_data *gdata, u8 *out,
-					const u8 *in,
-					unsigned long ciphertext_len);
-void (*finalize)(void *ctx,
-				struct gcm_context_data *gdata,
-				u8 *auth_tag, unsigned long auth_tag_len);
+static const struct aesni_gcm_tfm_s {
+	void (*init)(void *ctx, struct gcm_context_data *gdata, u8 *iv,
+		     u8 *hash_subkey, const u8 *aad, unsigned long aad_len);
+	void (*enc_update)(void *ctx, struct gcm_context_data *gdata, u8 *out,
+			   const u8 *in, unsigned long plaintext_len);
+	void (*dec_update)(void *ctx, struct gcm_context_data *gdata, u8 *out,
+			   const u8 *in, unsigned long ciphertext_len);
+	void (*finalize)(void *ctx, struct gcm_context_data *gdata,
+			 u8 *auth_tag, unsigned long auth_tag_len);
 } *aesni_gcm_tfm;
 
-struct aesni_gcm_tfm_s aesni_gcm_tfm_sse = {
+static const struct aesni_gcm_tfm_s aesni_gcm_tfm_sse = {
 	.init = &aesni_gcm_init,
 	.enc_update = &aesni_gcm_enc_update,
 	.dec_update = &aesni_gcm_dec_update,
@@ -243,7 +235,7 @@ asmlinkage void aesni_gcm_dec_avx_gen2(void *ctx,
 			const u8 *aad, unsigned long aad_len,
 			u8 *auth_tag, unsigned long auth_tag_len);
 
-struct aesni_gcm_tfm_s aesni_gcm_tfm_avx_gen2 = {
+static const struct aesni_gcm_tfm_s aesni_gcm_tfm_avx_gen2 = {
 	.init = &aesni_gcm_init_avx_gen2,
 	.enc_update = &aesni_gcm_enc_update_avx_gen2,
 	.dec_update = &aesni_gcm_dec_update_avx_gen2,
@@ -288,7 +280,7 @@ asmlinkage void aesni_gcm_dec_avx_gen4(void *ctx,
 			const u8 *aad, unsigned long aad_len,
 			u8 *auth_tag, unsigned long auth_tag_len);
 
-struct aesni_gcm_tfm_s aesni_gcm_tfm_avx_gen4 = {
+static const struct aesni_gcm_tfm_s aesni_gcm_tfm_avx_gen4 = {
 	.init = &aesni_gcm_init_avx_gen4,
 	.enc_update = &aesni_gcm_enc_update_avx_gen4,
 	.dec_update = &aesni_gcm_dec_update_avx_gen4,
@@ -778,7 +770,7 @@ static int gcmaes_crypt_by_sg(bool enc, struct aead_request *req,
 {
 	struct crypto_aead *tfm = crypto_aead_reqtfm(req);
 	unsigned long auth_tag_len = crypto_aead_authsize(tfm);
-	struct aesni_gcm_tfm_s *gcm_tfm = aesni_gcm_tfm;
+	const struct aesni_gcm_tfm_s *gcm_tfm = aesni_gcm_tfm;
 	struct gcm_context_data data AESNI_ALIGN_ATTR;
 	struct scatter_walk dst_sg_walk = {};
 	unsigned long left = req->cryptlen;
-- 
cgit v1.2.3


From 231baecdef7a906579925ccf1bd45aa734f32320 Mon Sep 17 00:00:00 2001
From: Eric Biggers <ebiggers@google.com>
Date: Fri, 18 Jan 2019 22:48:00 -0800
Subject: crypto: clarify name of WEAK_KEY request flag

CRYPTO_TFM_REQ_WEAK_KEY confuses newcomers to the crypto API because it
sounds like it is requesting a weak key.  Actually, it is requesting
that weak keys be forbidden (for algorithms that have the notion of
"weak keys"; currently only DES and XTS do).

Also it is only one letter away from CRYPTO_TFM_RES_WEAK_KEY, with which
it can be easily confused.  (This in fact happened in the UX500 driver,
though just in some debugging messages.)

Therefore, make the intent clear by renaming it to
CRYPTO_TFM_REQ_FORBID_WEAK_KEYS.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/s390/crypto/des_s390.c  | 4 ++--
 arch/sparc/crypto/des_glue.c | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

(limited to 'arch')

diff --git a/arch/s390/crypto/des_s390.c b/arch/s390/crypto/des_s390.c
index 5346b5a80bb6..0d15383d0ff1 100644
--- a/arch/s390/crypto/des_s390.c
+++ b/arch/s390/crypto/des_s390.c
@@ -38,7 +38,7 @@ static int des_setkey(struct crypto_tfm *tfm, const u8 *key,
 
 	/* check for weak keys */
 	if (!des_ekey(tmp, key) &&
-	    (tfm->crt_flags & CRYPTO_TFM_REQ_WEAK_KEY)) {
+	    (tfm->crt_flags & CRYPTO_TFM_REQ_FORBID_WEAK_KEYS)) {
 		tfm->crt_flags |= CRYPTO_TFM_RES_WEAK_KEY;
 		return -EINVAL;
 	}
@@ -228,7 +228,7 @@ static int des3_setkey(struct crypto_tfm *tfm, const u8 *key,
 	if (!(crypto_memneq(key, &key[DES_KEY_SIZE], DES_KEY_SIZE) &&
 	    crypto_memneq(&key[DES_KEY_SIZE], &key[DES_KEY_SIZE * 2],
 			  DES_KEY_SIZE)) &&
-	    (tfm->crt_flags & CRYPTO_TFM_REQ_WEAK_KEY)) {
+	    (tfm->crt_flags & CRYPTO_TFM_REQ_FORBID_WEAK_KEYS)) {
 		tfm->crt_flags |= CRYPTO_TFM_RES_WEAK_KEY;
 		return -EINVAL;
 	}
diff --git a/arch/sparc/crypto/des_glue.c b/arch/sparc/crypto/des_glue.c
index 56499ea39fd3..4884315daff4 100644
--- a/arch/sparc/crypto/des_glue.c
+++ b/arch/sparc/crypto/des_glue.c
@@ -53,7 +53,7 @@ static int des_set_key(struct crypto_tfm *tfm, const u8 *key,
 	 * weak key detection code.
 	 */
 	ret = des_ekey(tmp, key);
-	if (unlikely(ret == 0) && (*flags & CRYPTO_TFM_REQ_WEAK_KEY)) {
+	if (unlikely(ret == 0) && (*flags & CRYPTO_TFM_REQ_FORBID_WEAK_KEYS)) {
 		*flags |= CRYPTO_TFM_RES_WEAK_KEY;
 		return -EINVAL;
 	}
@@ -209,7 +209,7 @@ static int des3_ede_set_key(struct crypto_tfm *tfm, const u8 *key,
 
 	if (unlikely(!((K[0] ^ K[2]) | (K[1] ^ K[3])) ||
 		     !((K[2] ^ K[4]) | (K[3] ^ K[5]))) &&
-		     (*flags & CRYPTO_TFM_REQ_WEAK_KEY)) {
+		     (*flags & CRYPTO_TFM_REQ_FORBID_WEAK_KEYS)) {
 		*flags |= CRYPTO_TFM_RES_WEAK_KEY;
 		return -EINVAL;
 	}
-- 
cgit v1.2.3


From eaf46edf6ea89675bd36245369c8de5063a0272c Mon Sep 17 00:00:00 2001
From: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Date: Thu, 24 Jan 2019 17:33:45 +0100
Subject: crypto: arm64/aes-ccm - fix logical bug in AAD MAC handling

The NEON MAC calculation routine fails to handle the case correctly
where there is some data in the buffer, and the input fills it up
exactly. In this case, we enter the loop at the end with w8 == 0,
while a negative value is assumed, and so the loop carries on until
the increment of the 32-bit counter wraps around, which is quite
obviously wrong.

So omit the loop altogether in this case, and exit right away.

Reported-by: Eric Biggers <ebiggers@kernel.org>
Fixes: a3fd82105b9d1 ("arm64/crypto: AES in CCM mode using ARMv8 Crypto ...")
Cc: stable@vger.kernel.org
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/arm64/crypto/aes-ce-ccm-core.S | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

(limited to 'arch')

diff --git a/arch/arm64/crypto/aes-ce-ccm-core.S b/arch/arm64/crypto/aes-ce-ccm-core.S
index e3a375c4cb83..1b151442dac1 100644
--- a/arch/arm64/crypto/aes-ce-ccm-core.S
+++ b/arch/arm64/crypto/aes-ce-ccm-core.S
@@ -74,12 +74,13 @@ ENTRY(ce_aes_ccm_auth_data)
 	beq	10f
 	ext	v0.16b, v0.16b, v0.16b, #1	/* rotate out the mac bytes */
 	b	7b
-8:	mov	w7, w8
+8:	cbz	w8, 91f
+	mov	w7, w8
 	add	w8, w8, #16
 9:	ext	v1.16b, v1.16b, v1.16b, #1
 	adds	w7, w7, #1
 	bne	9b
-	eor	v0.16b, v0.16b, v1.16b
+91:	eor	v0.16b, v0.16b, v1.16b
 	st1	{v0.16b}, [x0]
 10:	str	w8, [x3]
 	ret
-- 
cgit v1.2.3


From 969e2f59d589c15f6aaf306e590dde16f12ea4b3 Mon Sep 17 00:00:00 2001
From: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Date: Thu, 24 Jan 2019 17:33:46 +0100
Subject: crypto: arm64/aes-ccm - fix bugs in non-NEON fallback routine

Commit 5092fcf34908 ("crypto: arm64/aes-ce-ccm: add non-SIMD generic
fallback") introduced C fallback code to replace the NEON routines
when invoked from a context where the NEON is not available (i.e.,
from the context of a softirq taken while the NEON is already being
used in kernel process context)

Fix two logical flaws in the MAC calculation of the associated data.

Reported-by: Eric Biggers <ebiggers@kernel.org>
Fixes: 5092fcf34908 ("crypto: arm64/aes-ce-ccm: add non-SIMD generic fallback")
Cc: stable@vger.kernel.org
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/arm64/crypto/aes-ce-ccm-glue.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

(limited to 'arch')

diff --git a/arch/arm64/crypto/aes-ce-ccm-glue.c b/arch/arm64/crypto/aes-ce-ccm-glue.c
index 68b11aa690e4..986191e8c058 100644
--- a/arch/arm64/crypto/aes-ce-ccm-glue.c
+++ b/arch/arm64/crypto/aes-ce-ccm-glue.c
@@ -125,7 +125,7 @@ static void ccm_update_mac(struct crypto_aes_ctx *key, u8 mac[], u8 const in[],
 			abytes -= added;
 		}
 
-		while (abytes > AES_BLOCK_SIZE) {
+		while (abytes >= AES_BLOCK_SIZE) {
 			__aes_arm64_encrypt(key->key_enc, mac, mac,
 					    num_rounds(key));
 			crypto_xor(mac, in, AES_BLOCK_SIZE);
@@ -139,8 +139,6 @@ static void ccm_update_mac(struct crypto_aes_ctx *key, u8 mac[], u8 const in[],
 					    num_rounds(key));
 			crypto_xor(mac, in, abytes);
 			*macp = abytes;
-		} else {
-			*macp = 0;
 		}
 	}
 }
-- 
cgit v1.2.3


From f9352900e064a159942fbf92d3840d6c8eafea56 Mon Sep 17 00:00:00 2001
From: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Date: Thu, 24 Jan 2019 17:33:47 +0100
Subject: crypto: arm64/aes-ccm - don't use an atomic walk needlessly

When the AES-CCM code was first added, the NEON register were saved
and restored eagerly, and so the code avoided doing so, and executed
the scatterwalk in atomic context inside the kernel_neon_begin/end
section.

This has been changed in the meantime, so switch to non-atomic
scatterwalks.

Fixes: bd2ad885e30d ("crypto: arm64/aes-ce-ccm - move kernel mode neon ...")
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/arm64/crypto/aes-ce-ccm-glue.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

(limited to 'arch')

diff --git a/arch/arm64/crypto/aes-ce-ccm-glue.c b/arch/arm64/crypto/aes-ce-ccm-glue.c
index 986191e8c058..5fc6f51908fd 100644
--- a/arch/arm64/crypto/aes-ce-ccm-glue.c
+++ b/arch/arm64/crypto/aes-ce-ccm-glue.c
@@ -253,7 +253,7 @@ static int ccm_encrypt(struct aead_request *req)
 	/* preserve the original iv for the final round */
 	memcpy(buf, req->iv, AES_BLOCK_SIZE);
 
-	err = skcipher_walk_aead_encrypt(&walk, req, true);
+	err = skcipher_walk_aead_encrypt(&walk, req, false);
 
 	if (may_use_simd()) {
 		while (walk.nbytes) {
@@ -311,7 +311,7 @@ static int ccm_decrypt(struct aead_request *req)
 	/* preserve the original iv for the final round */
 	memcpy(buf, req->iv, AES_BLOCK_SIZE);
 
-	err = skcipher_walk_aead_decrypt(&walk, req, true);
+	err = skcipher_walk_aead_decrypt(&walk, req, false);
 
 	if (may_use_simd()) {
 		while (walk.nbytes) {
-- 
cgit v1.2.3


From 5a22b198cd5274470d637d45f0e3bcdc0175a349 Mon Sep 17 00:00:00 2001
From: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Date: Fri, 25 Jan 2019 10:36:27 +0100
Subject: crypto: arm64/ghash - register PMULL variants as separate algos

The arm64 GHASH implementation either uses 8-bit or 64-bit
polynomial multiplication instructions, since the latter are
faster but not mandatory in the architecture.

Since that prevents us from testing both implementations on the
same system, let's expose both implementations to the crypto API,
with the priorities reflecting that the P64 version is the
preferred one if available.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/arm64/crypto/ghash-ce-glue.c | 118 +++++++++++++++++++++++++++++---------
 1 file changed, 90 insertions(+), 28 deletions(-)

(limited to 'arch')

diff --git a/arch/arm64/crypto/ghash-ce-glue.c b/arch/arm64/crypto/ghash-ce-glue.c
index 067d8937d5af..791ad422c427 100644
--- a/arch/arm64/crypto/ghash-ce-glue.c
+++ b/arch/arm64/crypto/ghash-ce-glue.c
@@ -60,10 +60,6 @@ asmlinkage void pmull_ghash_update_p8(int blocks, u64 dg[], const char *src,
 				      struct ghash_key const *k,
 				      const char *head);
 
-static void (*pmull_ghash_update)(int blocks, u64 dg[], const char *src,
-				  struct ghash_key const *k,
-				  const char *head);
-
 asmlinkage void pmull_gcm_encrypt(int blocks, u64 dg[], u8 dst[],
 				  const u8 src[], struct ghash_key const *k,
 				  u8 ctr[], u32 const rk[], int rounds,
@@ -87,11 +83,15 @@ static int ghash_init(struct shash_desc *desc)
 }
 
 static void ghash_do_update(int blocks, u64 dg[], const char *src,
-			    struct ghash_key *key, const char *head)
+			    struct ghash_key *key, const char *head,
+			    void (*simd_update)(int blocks, u64 dg[],
+						const char *src,
+						struct ghash_key const *k,
+						const char *head))
 {
 	if (likely(may_use_simd())) {
 		kernel_neon_begin();
-		pmull_ghash_update(blocks, dg, src, key, head);
+		simd_update(blocks, dg, src, key, head);
 		kernel_neon_end();
 	} else {
 		be128 dst = { cpu_to_be64(dg[1]), cpu_to_be64(dg[0]) };
@@ -119,8 +119,12 @@ static void ghash_do_update(int blocks, u64 dg[], const char *src,
 /* avoid hogging the CPU for too long */
 #define MAX_BLOCKS	(SZ_64K / GHASH_BLOCK_SIZE)
 
-static int ghash_update(struct shash_desc *desc, const u8 *src,
-			unsigned int len)
+static int __ghash_update(struct shash_desc *desc, const u8 *src,
+			  unsigned int len,
+			  void (*simd_update)(int blocks, u64 dg[],
+					      const char *src,
+					      struct ghash_key const *k,
+					      const char *head))
 {
 	struct ghash_desc_ctx *ctx = shash_desc_ctx(desc);
 	unsigned int partial = ctx->count % GHASH_BLOCK_SIZE;
@@ -146,7 +150,8 @@ static int ghash_update(struct shash_desc *desc, const u8 *src,
 			int chunk = min(blocks, MAX_BLOCKS);
 
 			ghash_do_update(chunk, ctx->digest, src, key,
-					partial ? ctx->buf : NULL);
+					partial ? ctx->buf : NULL,
+					simd_update);
 
 			blocks -= chunk;
 			src += chunk * GHASH_BLOCK_SIZE;
@@ -158,7 +163,19 @@ static int ghash_update(struct shash_desc *desc, const u8 *src,
 	return 0;
 }
 
-static int ghash_final(struct shash_desc *desc, u8 *dst)
+static int ghash_update_p8(struct shash_desc *desc, const u8 *src,
+			   unsigned int len)
+{
+	return __ghash_update(desc, src, len, pmull_ghash_update_p8);
+}
+
+static int ghash_update_p64(struct shash_desc *desc, const u8 *src,
+			    unsigned int len)
+{
+	return __ghash_update(desc, src, len, pmull_ghash_update_p64);
+}
+
+static int ghash_final_p8(struct shash_desc *desc, u8 *dst)
 {
 	struct ghash_desc_ctx *ctx = shash_desc_ctx(desc);
 	unsigned int partial = ctx->count % GHASH_BLOCK_SIZE;
@@ -168,7 +185,28 @@ static int ghash_final(struct shash_desc *desc, u8 *dst)
 
 		memset(ctx->buf + partial, 0, GHASH_BLOCK_SIZE - partial);
 
-		ghash_do_update(1, ctx->digest, ctx->buf, key, NULL);
+		ghash_do_update(1, ctx->digest, ctx->buf, key, NULL,
+				pmull_ghash_update_p8);
+	}
+	put_unaligned_be64(ctx->digest[1], dst);
+	put_unaligned_be64(ctx->digest[0], dst + 8);
+
+	*ctx = (struct ghash_desc_ctx){};
+	return 0;
+}
+
+static int ghash_final_p64(struct shash_desc *desc, u8 *dst)
+{
+	struct ghash_desc_ctx *ctx = shash_desc_ctx(desc);
+	unsigned int partial = ctx->count % GHASH_BLOCK_SIZE;
+
+	if (partial) {
+		struct ghash_key *key = crypto_shash_ctx(desc->tfm);
+
+		memset(ctx->buf + partial, 0, GHASH_BLOCK_SIZE - partial);
+
+		ghash_do_update(1, ctx->digest, ctx->buf, key, NULL,
+				pmull_ghash_update_p64);
 	}
 	put_unaligned_be64(ctx->digest[1], dst);
 	put_unaligned_be64(ctx->digest[0], dst + 8);
@@ -224,7 +262,21 @@ static int ghash_setkey(struct crypto_shash *tfm,
 	return __ghash_setkey(key, inkey, keylen);
 }
 
-static struct shash_alg ghash_alg = {
+static struct shash_alg ghash_alg[] = {{
+	.base.cra_name		= "ghash",
+	.base.cra_driver_name	= "ghash-neon",
+	.base.cra_priority	= 100,
+	.base.cra_blocksize	= GHASH_BLOCK_SIZE,
+	.base.cra_ctxsize	= sizeof(struct ghash_key),
+	.base.cra_module	= THIS_MODULE,
+
+	.digestsize		= GHASH_DIGEST_SIZE,
+	.init			= ghash_init,
+	.update			= ghash_update_p8,
+	.final			= ghash_final_p8,
+	.setkey			= ghash_setkey,
+	.descsize		= sizeof(struct ghash_desc_ctx),
+}, {
 	.base.cra_name		= "ghash",
 	.base.cra_driver_name	= "ghash-ce",
 	.base.cra_priority	= 200,
@@ -234,11 +286,11 @@ static struct shash_alg ghash_alg = {
 
 	.digestsize		= GHASH_DIGEST_SIZE,
 	.init			= ghash_init,
-	.update			= ghash_update,
-	.final			= ghash_final,
+	.update			= ghash_update_p64,
+	.final			= ghash_final_p64,
 	.setkey			= ghash_setkey,
 	.descsize		= sizeof(struct ghash_desc_ctx),
-};
+}};
 
 static int num_rounds(struct crypto_aes_ctx *ctx)
 {
@@ -301,7 +353,8 @@ static void gcm_update_mac(u64 dg[], const u8 *src, int count, u8 buf[],
 		int blocks = count / GHASH_BLOCK_SIZE;
 
 		ghash_do_update(blocks, dg, src, &ctx->ghash_key,
-				*buf_count ? buf : NULL);
+				*buf_count ? buf : NULL,
+				pmull_ghash_update_p64);
 
 		src += blocks * GHASH_BLOCK_SIZE;
 		count %= GHASH_BLOCK_SIZE;
@@ -345,7 +398,8 @@ static void gcm_calculate_auth_mac(struct aead_request *req, u64 dg[])
 
 	if (buf_count) {
 		memset(&buf[buf_count], 0, GHASH_BLOCK_SIZE - buf_count);
-		ghash_do_update(1, dg, buf, &ctx->ghash_key, NULL);
+		ghash_do_update(1, dg, buf, &ctx->ghash_key, NULL,
+				pmull_ghash_update_p64);
 	}
 }
 
@@ -358,7 +412,8 @@ static void gcm_final(struct aead_request *req, struct gcm_aes_ctx *ctx,
 	lengths.a = cpu_to_be64(req->assoclen * 8);
 	lengths.b = cpu_to_be64(cryptlen * 8);
 
-	ghash_do_update(1, dg, (void *)&lengths, &ctx->ghash_key, NULL);
+	ghash_do_update(1, dg, (void *)&lengths, &ctx->ghash_key, NULL,
+			pmull_ghash_update_p64);
 
 	put_unaligned_be64(dg[1], mac);
 	put_unaligned_be64(dg[0], mac + 8);
@@ -434,7 +489,7 @@ static int gcm_encrypt(struct aead_request *req)
 
 			ghash_do_update(walk.nbytes / AES_BLOCK_SIZE, dg,
 					walk.dst.virt.addr, &ctx->ghash_key,
-					NULL);
+					NULL, pmull_ghash_update_p64);
 
 			err = skcipher_walk_done(&walk,
 						 walk.nbytes % (2 * AES_BLOCK_SIZE));
@@ -469,7 +524,8 @@ static int gcm_encrypt(struct aead_request *req)
 
 		memcpy(buf, dst, nbytes);
 		memset(buf + nbytes, 0, GHASH_BLOCK_SIZE - nbytes);
-		ghash_do_update(!!nbytes, dg, buf, &ctx->ghash_key, head);
+		ghash_do_update(!!nbytes, dg, buf, &ctx->ghash_key, head,
+				pmull_ghash_update_p64);
 
 		err = skcipher_walk_done(&walk, 0);
 	}
@@ -558,7 +614,8 @@ static int gcm_decrypt(struct aead_request *req)
 			u8 *src = walk.src.virt.addr;
 
 			ghash_do_update(blocks, dg, walk.src.virt.addr,
-					&ctx->ghash_key, NULL);
+					&ctx->ghash_key, NULL,
+					pmull_ghash_update_p64);
 
 			do {
 				__aes_arm64_encrypt(ctx->aes_key.key_enc,
@@ -602,7 +659,8 @@ static int gcm_decrypt(struct aead_request *req)
 
 		memcpy(buf, src, nbytes);
 		memset(buf + nbytes, 0, GHASH_BLOCK_SIZE - nbytes);
-		ghash_do_update(!!nbytes, dg, buf, &ctx->ghash_key, head);
+		ghash_do_update(!!nbytes, dg, buf, &ctx->ghash_key, head,
+				pmull_ghash_update_p64);
 
 		crypto_xor_cpy(walk.dst.virt.addr, walk.src.virt.addr, iv,
 			       walk.nbytes);
@@ -650,26 +708,30 @@ static int __init ghash_ce_mod_init(void)
 		return -ENODEV;
 
 	if (elf_hwcap & HWCAP_PMULL)
-		pmull_ghash_update = pmull_ghash_update_p64;
-
+		ret = crypto_register_shashes(ghash_alg,
+					      ARRAY_SIZE(ghash_alg));
 	else
-		pmull_ghash_update = pmull_ghash_update_p8;
+		/* only register the first array element */
+		ret = crypto_register_shash(ghash_alg);
 
-	ret = crypto_register_shash(&ghash_alg);
 	if (ret)
 		return ret;
 
 	if (elf_hwcap & HWCAP_PMULL) {
 		ret = crypto_register_aead(&gcm_aes_alg);
 		if (ret)
-			crypto_unregister_shash(&ghash_alg);
+			crypto_unregister_shashes(ghash_alg,
+						  ARRAY_SIZE(ghash_alg));
 	}
 	return ret;
 }
 
 static void __exit ghash_ce_mod_exit(void)
 {
-	crypto_unregister_shash(&ghash_alg);
+	if (elf_hwcap & HWCAP_PMULL)
+		crypto_unregister_shashes(ghash_alg, ARRAY_SIZE(ghash_alg));
+	else
+		crypto_unregister_shash(ghash_alg);
 	crypto_unregister_aead(&gcm_aes_alg);
 }
 
-- 
cgit v1.2.3


From 62fecf295e3c48be1b5f17c440b93875b9adb4d6 Mon Sep 17 00:00:00 2001
From: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Date: Sun, 27 Jan 2019 10:16:52 +0100
Subject: crypto: arm/crct10dif - revert to C code for short inputs

The SIMD routine ported from x86 used to have a special code path
for inputs < 16 bytes, which got lost somewhere along the way.
Instead, the current glue code aligns the input pointer to permit
the NEON routine to use special versions of the vld1 instructions
that assume 16 byte alignment, but this could result in inputs of
less than 16 bytes to be passed in. This not only fails the new
extended tests that Eric has implemented, it also results in the
code reading past the end of the input, which could potentially
result in crashes when dealing with less than 16 bytes of input
at the end of a page which is followed by an unmapped page.

So update the glue code to only invoke the NEON routine if the
input is at least 16 bytes.

Reported-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Fixes: 1d481f1cd892 ("crypto: arm/crct10dif - port x86 SSE implementation to ARM")
Cc: <stable@vger.kernel.org> # v4.10+
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/arm/crypto/crct10dif-ce-core.S | 14 +++++++-------
 arch/arm/crypto/crct10dif-ce-glue.c | 23 ++++++-----------------
 2 files changed, 13 insertions(+), 24 deletions(-)

(limited to 'arch')

diff --git a/arch/arm/crypto/crct10dif-ce-core.S b/arch/arm/crypto/crct10dif-ce-core.S
index ce45ba0c0687..16019b5961e7 100644
--- a/arch/arm/crypto/crct10dif-ce-core.S
+++ b/arch/arm/crypto/crct10dif-ce-core.S
@@ -124,10 +124,10 @@ ENTRY(crc_t10dif_pmull)
 	vext.8		q10, qzr, q0, #4
 
 	// receive the initial 64B data, xor the initial crc value
-	vld1.64		{q0-q1}, [arg2, :128]!
-	vld1.64		{q2-q3}, [arg2, :128]!
-	vld1.64		{q4-q5}, [arg2, :128]!
-	vld1.64		{q6-q7}, [arg2, :128]!
+	vld1.64		{q0-q1}, [arg2]!
+	vld1.64		{q2-q3}, [arg2]!
+	vld1.64		{q4-q5}, [arg2]!
+	vld1.64		{q6-q7}, [arg2]!
 CPU_LE(	vrev64.8	q0, q0			)
 CPU_LE(	vrev64.8	q1, q1			)
 CPU_LE(	vrev64.8	q2, q2			)
@@ -167,7 +167,7 @@ CPU_LE(	vrev64.8	q7, q7			)
 _fold_64_B_loop:
 
 	.macro		fold64, reg1, reg2
-	vld1.64		{q11-q12}, [arg2, :128]!
+	vld1.64		{q11-q12}, [arg2]!
 
 	vmull.p64	q8, \reg1\()h, d21
 	vmull.p64	\reg1, \reg1\()l, d20
@@ -238,7 +238,7 @@ _16B_reduction_loop:
 	vmull.p64	q7, d15, d21
 	veor.8		q7, q7, q8
 
-	vld1.64		{q0}, [arg2, :128]!
+	vld1.64		{q0}, [arg2]!
 CPU_LE(	vrev64.8	q0, q0		)
 	vswp		d0, d1
 	veor.8		q7, q7, q0
@@ -335,7 +335,7 @@ _less_than_128:
 	vmov.i8		q0, #0
 	vmov		s3, arg1_low32		// get the initial crc value
 
-	vld1.64		{q7}, [arg2, :128]!
+	vld1.64		{q7}, [arg2]!
 CPU_LE(	vrev64.8	q7, q7		)
 	vswp		d14, d15
 	veor.8		q7, q7, q0
diff --git a/arch/arm/crypto/crct10dif-ce-glue.c b/arch/arm/crypto/crct10dif-ce-glue.c
index d428355cf38d..14c19c70a841 100644
--- a/arch/arm/crypto/crct10dif-ce-glue.c
+++ b/arch/arm/crypto/crct10dif-ce-glue.c
@@ -35,26 +35,15 @@ static int crct10dif_update(struct shash_desc *desc, const u8 *data,
 			    unsigned int length)
 {
 	u16 *crc = shash_desc_ctx(desc);
-	unsigned int l;
 
-	if (!may_use_simd()) {
-		*crc = crc_t10dif_generic(*crc, data, length);
+	if (length >= CRC_T10DIF_PMULL_CHUNK_SIZE && may_use_simd()) {
+		kernel_neon_begin();
+		*crc = crc_t10dif_pmull(*crc, data, length);
+		kernel_neon_end();
 	} else {
-		if (unlikely((u32)data % CRC_T10DIF_PMULL_CHUNK_SIZE)) {
-			l = min_t(u32, length, CRC_T10DIF_PMULL_CHUNK_SIZE -
-				  ((u32)data % CRC_T10DIF_PMULL_CHUNK_SIZE));
-
-			*crc = crc_t10dif_generic(*crc, data, l);
-
-			length -= l;
-			data += l;
-		}
-		if (length > 0) {
-			kernel_neon_begin();
-			*crc = crc_t10dif_pmull(*crc, data, length);
-			kernel_neon_end();
-		}
+		*crc = crc_t10dif_generic(*crc, data, length);
 	}
+
 	return 0;
 }
 
-- 
cgit v1.2.3


From d72b9d4acd548251f55b16843fc7a05dc5c80de8 Mon Sep 17 00:00:00 2001
From: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Date: Sun, 27 Jan 2019 10:16:53 +0100
Subject: crypto: arm64/crct10dif - revert to C code for short inputs

The SIMD routine ported from x86 used to have a special code path
for inputs < 16 bytes, which got lost somewhere along the way.
Instead, the current glue code aligns the input pointer to 16 bytes,
which is not really necessary on this architecture (although it
could be beneficial to performance to expose aligned data to the
the NEON routine), but this could result in inputs of less than
16 bytes to be passed in. This not only fails the new extended
tests that Eric has implemented, it also results in the code
reading past the end of the input, which could potentially result
in crashes when dealing with less than 16 bytes of input at the
end of a page which is followed by an unmapped page.

So update the glue code to only invoke the NEON routine if the
input is at least 16 bytes.

Reported-by: Eric Biggers <ebiggers@kernel.org>
Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Fixes: 6ef5737f3931 ("crypto: arm64/crct10dif - port x86 SSE implementation to arm64")
Cc: <stable@vger.kernel.org> # v4.10+
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/arm64/crypto/crct10dif-ce-glue.c | 25 ++++++-------------------
 1 file changed, 6 insertions(+), 19 deletions(-)

(limited to 'arch')

diff --git a/arch/arm64/crypto/crct10dif-ce-glue.c b/arch/arm64/crypto/crct10dif-ce-glue.c
index b461d62023f2..567c24f3d224 100644
--- a/arch/arm64/crypto/crct10dif-ce-glue.c
+++ b/arch/arm64/crypto/crct10dif-ce-glue.c
@@ -39,26 +39,13 @@ static int crct10dif_update(struct shash_desc *desc, const u8 *data,
 			    unsigned int length)
 {
 	u16 *crc = shash_desc_ctx(desc);
-	unsigned int l;
 
-	if (unlikely((u64)data % CRC_T10DIF_PMULL_CHUNK_SIZE)) {
-		l = min_t(u32, length, CRC_T10DIF_PMULL_CHUNK_SIZE -
-			  ((u64)data % CRC_T10DIF_PMULL_CHUNK_SIZE));
-
-		*crc = crc_t10dif_generic(*crc, data, l);
-
-		length -= l;
-		data += l;
-	}
-
-	if (length > 0) {
-		if (may_use_simd()) {
-			kernel_neon_begin();
-			*crc = crc_t10dif_pmull(*crc, data, length);
-			kernel_neon_end();
-		} else {
-			*crc = crc_t10dif_generic(*crc, data, length);
-		}
+	if (length >= CRC_T10DIF_PMULL_CHUNK_SIZE && may_use_simd()) {
+		kernel_neon_begin();
+		*crc = crc_t10dif_pmull(*crc, data, length);
+		kernel_neon_end();
+	} else {
+		*crc = crc_t10dif_generic(*crc, data, length);
 	}
 
 	return 0;
-- 
cgit v1.2.3


From c03f3cb40ba93a9f49fc44a71ddaa46f93f90a15 Mon Sep 17 00:00:00 2001
From: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Date: Sun, 27 Jan 2019 10:16:54 +0100
Subject: crypto: arm/crct10dif - remove dead code

Remove some code that is no longer called now that we make sure never
to invoke the SIMD routine with less that 16 bytes of input.

Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/arm/crypto/crct10dif-ce-core.S | 13 -------------
 1 file changed, 13 deletions(-)

(limited to 'arch')

diff --git a/arch/arm/crypto/crct10dif-ce-core.S b/arch/arm/crypto/crct10dif-ce-core.S
index 16019b5961e7..d058fad423c2 100644
--- a/arch/arm/crypto/crct10dif-ce-core.S
+++ b/arch/arm/crypto/crct10dif-ce-core.S
@@ -342,7 +342,6 @@ CPU_LE(	vrev64.8	q7, q7		)
 
 	cmp		arg3, #16
 	beq		_128_done		// exactly 16 left
-	blt		_less_than_16_left
 
 	// now if there is, load the constants
 	vldr		d20, rk1
@@ -353,18 +352,6 @@ CPU_LE(	vrev64.8	q7, q7		)
 	addlt		arg3, arg3, #16
 	blt		_get_last_two_regs
 	b		_16B_reduction_loop
-
-_less_than_16_left:
-	// shl r9, 4
-	adr		ip, tbl_shf_table + 16
-	sub		ip, ip, arg3
-	vld1.8		{q0}, [ip]
-	vmov.i8		q9, #0x80
-	veor.8		q0, q0, q9
-	vtbl.8		d18, {d14-d15}, d0
-	vtbl.8		d15, {d14-d15}, d1
-	vmov		d14, d18
-	b		_128_done
 ENDPROC(crc_t10dif_pmull)
 
 // precomputed constants
-- 
cgit v1.2.3


From 1b2ca568ca23a6cfa6aa28e8675b54940807231d Mon Sep 17 00:00:00 2001
From: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Date: Sun, 27 Jan 2019 10:16:55 +0100
Subject: crypto: arm64/crct10dif - remove dead code

Remove some code that is no longer called now that we make sure never
to invoke the SIMD routine with less than 16 bytes of input.

Reviewed-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/arm64/crypto/crct10dif-ce-core.S | 11 -----------
 1 file changed, 11 deletions(-)

(limited to 'arch')

diff --git a/arch/arm64/crypto/crct10dif-ce-core.S b/arch/arm64/crypto/crct10dif-ce-core.S
index 9e82e8e8ed05..f7326259c40d 100644
--- a/arch/arm64/crypto/crct10dif-ce-core.S
+++ b/arch/arm64/crypto/crct10dif-ce-core.S
@@ -497,7 +497,6 @@ CPU_LE(	ext		v7.16b, v7.16b, v7.16b, #8	)
 
 	cmp		arg3, #16
 	b.eq		.L_128_done_\@		// exactly 16 left
-	b.lt		.L_less_than_16_left_\@
 
 	ldr_l		q10, rk1, x8		// rk1 and rk2 in xmm10
 	__pmull_pre_\p	v10
@@ -509,16 +508,6 @@ CPU_LE(	ext		v7.16b, v7.16b, v7.16b, #8	)
 
 	add		arg3, arg3, #16
 	b		.L_get_last_two_regs_\@
-
-.L_less_than_16_left_\@:
-	// shl r9, 4
-	adr_l		x0, tbl_shf_table + 16
-	sub		x0, x0, arg3
-	ld1		{v0.16b}, [x0]
-	movi		v9.16b, #0x80
-	eor		v0.16b, v0.16b, v9.16b
-	tbl		v7.16b, {v7.16b}, v0.16b
-	b		.L_128_done_\@
 	.endm
 
 ENTRY(crc_t10dif_pmull_p8)
-- 
cgit v1.2.3


From 8336bdf12a9ea5bb77bb32d215a34485fb66245c Mon Sep 17 00:00:00 2001
From: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Date: Fri, 25 Jan 2019 10:36:26 +0100
Subject: crypto: arm64/crct10dif - register PMULL variants as separate algos

The arm64 CRC-T10DIF implementation either uses 8-bit or 64-bit
polynomial multiplication instructions, since the latter are
faster but not mandatory in the architecture.

Since that prevents us from testing both implementations on the
same system, let's expose both implementations to the crypto API,
with the priorities reflecting that the P64 version is the
preferred one if available.

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/arm64/crypto/crct10dif-ce-glue.c | 54 +++++++++++++++++++++++++++--------
 1 file changed, 42 insertions(+), 12 deletions(-)

(limited to 'arch')

diff --git a/arch/arm64/crypto/crct10dif-ce-glue.c b/arch/arm64/crypto/crct10dif-ce-glue.c
index 567c24f3d224..242757cc6da9 100644
--- a/arch/arm64/crypto/crct10dif-ce-glue.c
+++ b/arch/arm64/crypto/crct10dif-ce-glue.c
@@ -25,8 +25,6 @@
 asmlinkage u16 crc_t10dif_pmull_p64(u16 init_crc, const u8 buf[], u64 len);
 asmlinkage u16 crc_t10dif_pmull_p8(u16 init_crc, const u8 buf[], u64 len);
 
-static u16 (*crc_t10dif_pmull)(u16 init_crc, const u8 buf[], u64 len);
-
 static int crct10dif_init(struct shash_desc *desc)
 {
 	u16 *crc = shash_desc_ctx(desc);
@@ -35,14 +33,30 @@ static int crct10dif_init(struct shash_desc *desc)
 	return 0;
 }
 
-static int crct10dif_update(struct shash_desc *desc, const u8 *data,
+static int crct10dif_update_pmull_p8(struct shash_desc *desc, const u8 *data,
+			    unsigned int length)
+{
+	u16 *crc = shash_desc_ctx(desc);
+
+	if (length >= CRC_T10DIF_PMULL_CHUNK_SIZE && may_use_simd()) {
+		kernel_neon_begin();
+		*crc = crc_t10dif_pmull_p8(*crc, data, length);
+		kernel_neon_end();
+	} else {
+		*crc = crc_t10dif_generic(*crc, data, length);
+	}
+
+	return 0;
+}
+
+static int crct10dif_update_pmull_p64(struct shash_desc *desc, const u8 *data,
 			    unsigned int length)
 {
 	u16 *crc = shash_desc_ctx(desc);
 
 	if (length >= CRC_T10DIF_PMULL_CHUNK_SIZE && may_use_simd()) {
 		kernel_neon_begin();
-		*crc = crc_t10dif_pmull(*crc, data, length);
+		*crc = crc_t10dif_pmull_p64(*crc, data, length);
 		kernel_neon_end();
 	} else {
 		*crc = crc_t10dif_generic(*crc, data, length);
@@ -59,10 +73,22 @@ static int crct10dif_final(struct shash_desc *desc, u8 *out)
 	return 0;
 }
 
-static struct shash_alg crc_t10dif_alg = {
+static struct shash_alg crc_t10dif_alg[] = {{
 	.digestsize		= CRC_T10DIF_DIGEST_SIZE,
 	.init			= crct10dif_init,
-	.update			= crct10dif_update,
+	.update			= crct10dif_update_pmull_p8,
+	.final			= crct10dif_final,
+	.descsize		= CRC_T10DIF_DIGEST_SIZE,
+
+	.base.cra_name		= "crct10dif",
+	.base.cra_driver_name	= "crct10dif-arm64-neon",
+	.base.cra_priority	= 100,
+	.base.cra_blocksize	= CRC_T10DIF_BLOCK_SIZE,
+	.base.cra_module	= THIS_MODULE,
+}, {
+	.digestsize		= CRC_T10DIF_DIGEST_SIZE,
+	.init			= crct10dif_init,
+	.update			= crct10dif_update_pmull_p64,
 	.final			= crct10dif_final,
 	.descsize		= CRC_T10DIF_DIGEST_SIZE,
 
@@ -71,21 +97,25 @@ static struct shash_alg crc_t10dif_alg = {
 	.base.cra_priority	= 200,
 	.base.cra_blocksize	= CRC_T10DIF_BLOCK_SIZE,
 	.base.cra_module	= THIS_MODULE,
-};
+}};
 
 static int __init crc_t10dif_mod_init(void)
 {
 	if (elf_hwcap & HWCAP_PMULL)
-		crc_t10dif_pmull = crc_t10dif_pmull_p64;
+		return crypto_register_shashes(crc_t10dif_alg,
+					       ARRAY_SIZE(crc_t10dif_alg));
 	else
-		crc_t10dif_pmull = crc_t10dif_pmull_p8;
-
-	return crypto_register_shash(&crc_t10dif_alg);
+		/* only register the first array element */
+		return crypto_register_shash(crc_t10dif_alg);
 }
 
 static void __exit crc_t10dif_mod_exit(void)
 {
-	crypto_unregister_shash(&crc_t10dif_alg);
+	if (elf_hwcap & HWCAP_PMULL)
+		crypto_unregister_shashes(crc_t10dif_alg,
+					  ARRAY_SIZE(crc_t10dif_alg));
+	else
+		crypto_unregister_shash(crc_t10dif_alg);
 }
 
 module_cpu_feature_match(ASIMD, crc_t10dif_mod_init);
-- 
cgit v1.2.3


From 0974037fc55cdf27a0297d3487f4fafa57b4a4a9 Mon Sep 17 00:00:00 2001
From: Eric Biggers <ebiggers@google.com>
Date: Wed, 30 Jan 2019 20:42:40 -0800
Subject: crypto: x86/crct10dif-pcl - cleanup and optimizations

The x86, arm, and arm64 asm implementations of crct10dif are very
difficult to understand partly because many of the comments, labels, and
macros are named incorrectly: the lengths mentioned are usually off by a
factor of two from the actual code.  Many other things are unnecessarily
convoluted as well, e.g. there are many more fold constants than
actually needed and some aren't fully reduced.

This series therefore cleans up all these implementations to be much
more maintainable.  I also made some small optimizations where I saw
opportunities, resulting in slightly better performance.

This patch cleans up the x86 version.

As part of this, I removed support for len < 16 from the x86 assembly;
now the glue code falls back to the generic table-based implementation
in this case.  Due to the overhead of kernel_fpu_begin(), this actually
significantly improves performance on these lengths.  (And even if
kernel_fpu_begin() were free, the generic code is still faster for about
len < 11.)  This removal also eliminates error-prone special cases and
makes the x86, arm32, and arm64 ports of the code match more closely.

Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/x86/crypto/crct10dif-pcl-asm_64.S  | 782 ++++++++++----------------------
 arch/x86/crypto/crct10dif-pclmul_glue.c |  12 +-
 2 files changed, 235 insertions(+), 559 deletions(-)

(limited to 'arch')

diff --git a/arch/x86/crypto/crct10dif-pcl-asm_64.S b/arch/x86/crypto/crct10dif-pcl-asm_64.S
index de04d3e98d8d..3d873e67749d 100644
--- a/arch/x86/crypto/crct10dif-pcl-asm_64.S
+++ b/arch/x86/crypto/crct10dif-pcl-asm_64.S
@@ -43,609 +43,291 @@
 # LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
 # NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 # SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-########################################################################
-#       Function API:
-#       UINT16 crc_t10dif_pcl(
-#               UINT16 init_crc, //initial CRC value, 16 bits
-#               const unsigned char *buf, //buffer pointer to calculate CRC on
-#               UINT64 len //buffer length in bytes (64-bit data)
-#       );
 #
 #       Reference paper titled "Fast CRC Computation for Generic
 #	Polynomials Using PCLMULQDQ Instruction"
 #       URL: http://www.intel.com/content/dam/www/public/us/en/documents
 #  /white-papers/fast-crc-computation-generic-polynomials-pclmulqdq-paper.pdf
 #
-#
 
 #include <linux/linkage.h>
 
 .text
 
-#define        arg1 %rdi
-#define        arg2 %rsi
-#define        arg3 %rdx
-
-#define        arg1_low32 %edi
+#define		init_crc	%edi
+#define		buf		%rsi
+#define		len		%rdx
+
+#define		FOLD_CONSTS	%xmm10
+#define		BSWAP_MASK	%xmm11
+
+# Fold reg1, reg2 into the next 32 data bytes, storing the result back into
+# reg1, reg2.
+.macro	fold_32_bytes	offset, reg1, reg2
+	movdqu	\offset(buf), %xmm9
+	movdqu	\offset+16(buf), %xmm12
+	pshufb	BSWAP_MASK, %xmm9
+	pshufb	BSWAP_MASK, %xmm12
+	movdqa	\reg1, %xmm8
+	movdqa	\reg2, %xmm13
+	pclmulqdq	$0x00, FOLD_CONSTS, \reg1
+	pclmulqdq	$0x11, FOLD_CONSTS, %xmm8
+	pclmulqdq	$0x00, FOLD_CONSTS, \reg2
+	pclmulqdq	$0x11, FOLD_CONSTS, %xmm13
+	pxor	%xmm9 , \reg1
+	xorps	%xmm8 , \reg1
+	pxor	%xmm12, \reg2
+	xorps	%xmm13, \reg2
+.endm
+
+# Fold src_reg into dst_reg.
+.macro	fold_16_bytes	src_reg, dst_reg
+	movdqa	\src_reg, %xmm8
+	pclmulqdq	$0x11, FOLD_CONSTS, \src_reg
+	pclmulqdq	$0x00, FOLD_CONSTS, %xmm8
+	pxor	%xmm8, \dst_reg
+	xorps	\src_reg, \dst_reg
+.endm
 
-ENTRY(crc_t10dif_pcl)
+#
+# u16 crc_t10dif_pcl(u16 init_crc, const *u8 buf, size_t len);
+#
+# Assumes len >= 16.
+#
 .align 16
+ENTRY(crc_t10dif_pcl)
 
-	# adjust the 16-bit initial_crc value, scale it to 32 bits
-	shl	$16, arg1_low32
-
-	# Allocate Stack Space
-	mov     %rsp, %rcx
-	sub	$16*2, %rsp
-	# align stack to 16 byte boundary
-	and     $~(0x10 - 1), %rsp
-
-	# check if smaller than 256
-	cmp	$256, arg3
-
-	# for sizes less than 128, we can't fold 64B at a time...
-	jl	_less_than_128
-
-
-	# load the initial crc value
-	movd	arg1_low32, %xmm10	# initial crc
-
-	# crc value does not need to be byte-reflected, but it needs
-	# to be moved to the high part of the register.
-	# because data will be byte-reflected and will align with
-	# initial crc at correct place.
-	pslldq	$12, %xmm10
-
-	movdqa  SHUF_MASK(%rip), %xmm11
-	# receive the initial 64B data, xor the initial crc value
-	movdqu	16*0(arg2), %xmm0
-	movdqu	16*1(arg2), %xmm1
-	movdqu	16*2(arg2), %xmm2
-	movdqu	16*3(arg2), %xmm3
-	movdqu	16*4(arg2), %xmm4
-	movdqu	16*5(arg2), %xmm5
-	movdqu	16*6(arg2), %xmm6
-	movdqu	16*7(arg2), %xmm7
-
-	pshufb	%xmm11, %xmm0
-	# XOR the initial_crc value
-	pxor	%xmm10, %xmm0
-	pshufb	%xmm11, %xmm1
-	pshufb	%xmm11, %xmm2
-	pshufb	%xmm11, %xmm3
-	pshufb	%xmm11, %xmm4
-	pshufb	%xmm11, %xmm5
-	pshufb	%xmm11, %xmm6
-	pshufb	%xmm11, %xmm7
-
-	movdqa	rk3(%rip), %xmm10	#xmm10 has rk3 and rk4
-					#imm value of pclmulqdq instruction
-					#will determine which constant to use
-
-	#################################################################
-	# we subtract 256 instead of 128 to save one instruction from the loop
-	sub	$256, arg3
-
-	# at this section of the code, there is 64*x+y (0<=y<64) bytes of
-	# buffer. The _fold_64_B_loop will fold 64B at a time
-	# until we have 64+y Bytes of buffer
-
-
-	# fold 64B at a time. This section of the code folds 4 xmm
-	# registers in parallel
-_fold_64_B_loop:
-
-	# update the buffer pointer
-	add	$128, arg2		#    buf += 64#
-
-	movdqu	16*0(arg2), %xmm9
-	movdqu	16*1(arg2), %xmm12
-	pshufb	%xmm11, %xmm9
-	pshufb	%xmm11, %xmm12
-	movdqa	%xmm0, %xmm8
-	movdqa	%xmm1, %xmm13
-	pclmulqdq	$0x0 , %xmm10, %xmm0
-	pclmulqdq	$0x11, %xmm10, %xmm8
-	pclmulqdq	$0x0 , %xmm10, %xmm1
-	pclmulqdq	$0x11, %xmm10, %xmm13
-	pxor	%xmm9 , %xmm0
-	xorps	%xmm8 , %xmm0
-	pxor	%xmm12, %xmm1
-	xorps	%xmm13, %xmm1
-
-	movdqu	16*2(arg2), %xmm9
-	movdqu	16*3(arg2), %xmm12
-	pshufb	%xmm11, %xmm9
-	pshufb	%xmm11, %xmm12
-	movdqa	%xmm2, %xmm8
-	movdqa	%xmm3, %xmm13
-	pclmulqdq	$0x0, %xmm10, %xmm2
-	pclmulqdq	$0x11, %xmm10, %xmm8
-	pclmulqdq	$0x0, %xmm10, %xmm3
-	pclmulqdq	$0x11, %xmm10, %xmm13
-	pxor	%xmm9 , %xmm2
-	xorps	%xmm8 , %xmm2
-	pxor	%xmm12, %xmm3
-	xorps	%xmm13, %xmm3
-
-	movdqu	16*4(arg2), %xmm9
-	movdqu	16*5(arg2), %xmm12
-	pshufb	%xmm11, %xmm9
-	pshufb	%xmm11, %xmm12
-	movdqa	%xmm4, %xmm8
-	movdqa	%xmm5, %xmm13
-	pclmulqdq	$0x0,  %xmm10, %xmm4
-	pclmulqdq	$0x11, %xmm10, %xmm8
-	pclmulqdq	$0x0,  %xmm10, %xmm5
-	pclmulqdq	$0x11, %xmm10, %xmm13
-	pxor	%xmm9 ,  %xmm4
-	xorps	%xmm8 ,  %xmm4
-	pxor	%xmm12,  %xmm5
-	xorps	%xmm13,  %xmm5
-
-	movdqu	16*6(arg2), %xmm9
-	movdqu	16*7(arg2), %xmm12
-	pshufb	%xmm11, %xmm9
-	pshufb	%xmm11, %xmm12
-	movdqa	%xmm6 , %xmm8
-	movdqa	%xmm7 , %xmm13
-	pclmulqdq	$0x0 , %xmm10, %xmm6
-	pclmulqdq	$0x11, %xmm10, %xmm8
-	pclmulqdq	$0x0 , %xmm10, %xmm7
-	pclmulqdq	$0x11, %xmm10, %xmm13
-	pxor	%xmm9 , %xmm6
-	xorps	%xmm8 , %xmm6
-	pxor	%xmm12, %xmm7
-	xorps	%xmm13, %xmm7
-
-	sub	$128, arg3
-
-	# check if there is another 64B in the buffer to be able to fold
-	jge	_fold_64_B_loop
-	##################################################################
-
-
-	add	$128, arg2
-	# at this point, the buffer pointer is pointing at the last y Bytes
-	# of the buffer the 64B of folded data is in 4 of the xmm
-	# registers: xmm0, xmm1, xmm2, xmm3
-
-
-	# fold the 8 xmm registers to 1 xmm register with different constants
-
-	movdqa	rk9(%rip), %xmm10
-	movdqa	%xmm0, %xmm8
-	pclmulqdq	$0x11, %xmm10, %xmm0
-	pclmulqdq	$0x0 , %xmm10, %xmm8
-	pxor	%xmm8, %xmm7
-	xorps	%xmm0, %xmm7
-
-	movdqa	rk11(%rip), %xmm10
-	movdqa	%xmm1, %xmm8
-	pclmulqdq	 $0x11, %xmm10, %xmm1
-	pclmulqdq	 $0x0 , %xmm10, %xmm8
-	pxor	%xmm8, %xmm7
-	xorps	%xmm1, %xmm7
-
-	movdqa	rk13(%rip), %xmm10
-	movdqa	%xmm2, %xmm8
-	pclmulqdq	 $0x11, %xmm10, %xmm2
-	pclmulqdq	 $0x0 , %xmm10, %xmm8
-	pxor	%xmm8, %xmm7
-	pxor	%xmm2, %xmm7
-
-	movdqa	rk15(%rip), %xmm10
-	movdqa	%xmm3, %xmm8
-	pclmulqdq	$0x11, %xmm10, %xmm3
-	pclmulqdq	$0x0 , %xmm10, %xmm8
-	pxor	%xmm8, %xmm7
-	xorps	%xmm3, %xmm7
-
-	movdqa	rk17(%rip), %xmm10
-	movdqa	%xmm4, %xmm8
-	pclmulqdq	$0x11, %xmm10, %xmm4
-	pclmulqdq	$0x0 , %xmm10, %xmm8
-	pxor	%xmm8, %xmm7
-	pxor	%xmm4, %xmm7
-
-	movdqa	rk19(%rip), %xmm10
-	movdqa	%xmm5, %xmm8
-	pclmulqdq	$0x11, %xmm10, %xmm5
-	pclmulqdq	$0x0 , %xmm10, %xmm8
-	pxor	%xmm8, %xmm7
-	xorps	%xmm5, %xmm7
-
-	movdqa	rk1(%rip), %xmm10	#xmm10 has rk1 and rk2
-					#imm value of pclmulqdq instruction
-					#will determine which constant to use
-	movdqa	%xmm6, %xmm8
-	pclmulqdq	$0x11, %xmm10, %xmm6
-	pclmulqdq	$0x0 , %xmm10, %xmm8
-	pxor	%xmm8, %xmm7
-	pxor	%xmm6, %xmm7
-
-
-	# instead of 64, we add 48 to the loop counter to save 1 instruction
-	# from the loop instead of a cmp instruction, we use the negative
-	# flag with the jl instruction
-	add	$128-16, arg3
-	jl	_final_reduction_for_128
-
-	# now we have 16+y bytes left to reduce. 16 Bytes is in register xmm7
-	# and the rest is in memory. We can fold 16 bytes at a time if y>=16
-	# continue folding 16B at a time
-
-_16B_reduction_loop:
+	movdqa	.Lbswap_mask(%rip), BSWAP_MASK
+
+	# For sizes less than 256 bytes, we can't fold 128 bytes at a time.
+	cmp	$256, len
+	jl	.Lless_than_256_bytes
+
+	# Load the first 128 data bytes.  Byte swapping is necessary to make the
+	# bit order match the polynomial coefficient order.
+	movdqu	16*0(buf), %xmm0
+	movdqu	16*1(buf), %xmm1
+	movdqu	16*2(buf), %xmm2
+	movdqu	16*3(buf), %xmm3
+	movdqu	16*4(buf), %xmm4
+	movdqu	16*5(buf), %xmm5
+	movdqu	16*6(buf), %xmm6
+	movdqu	16*7(buf), %xmm7
+	add	$128, buf
+	pshufb	BSWAP_MASK, %xmm0
+	pshufb	BSWAP_MASK, %xmm1
+	pshufb	BSWAP_MASK, %xmm2
+	pshufb	BSWAP_MASK, %xmm3
+	pshufb	BSWAP_MASK, %xmm4
+	pshufb	BSWAP_MASK, %xmm5
+	pshufb	BSWAP_MASK, %xmm6
+	pshufb	BSWAP_MASK, %xmm7
+
+	# XOR the first 16 data *bits* with the initial CRC value.
+	pxor	%xmm8, %xmm8
+	pinsrw	$7, init_crc, %xmm8
+	pxor	%xmm8, %xmm0
+
+	movdqa	.Lfold_across_128_bytes_consts(%rip), FOLD_CONSTS
+
+	# Subtract 128 for the 128 data bytes just consumed.  Subtract another
+	# 128 to simplify the termination condition of the following loop.
+	sub	$256, len
+
+	# While >= 128 data bytes remain (not counting xmm0-7), fold the 128
+	# bytes xmm0-7 into them, storing the result back into xmm0-7.
+.Lfold_128_bytes_loop:
+	fold_32_bytes	0, %xmm0, %xmm1
+	fold_32_bytes	32, %xmm2, %xmm3
+	fold_32_bytes	64, %xmm4, %xmm5
+	fold_32_bytes	96, %xmm6, %xmm7
+	add	$128, buf
+	sub	$128, len
+	jge	.Lfold_128_bytes_loop
+
+	# Now fold the 112 bytes in xmm0-xmm6 into the 16 bytes in xmm7.
+
+	# Fold across 64 bytes.
+	movdqa	.Lfold_across_64_bytes_consts(%rip), FOLD_CONSTS
+	fold_16_bytes	%xmm0, %xmm4
+	fold_16_bytes	%xmm1, %xmm5
+	fold_16_bytes	%xmm2, %xmm6
+	fold_16_bytes	%xmm3, %xmm7
+	# Fold across 32 bytes.
+	movdqa	.Lfold_across_32_bytes_consts(%rip), FOLD_CONSTS
+	fold_16_bytes	%xmm4, %xmm6
+	fold_16_bytes	%xmm5, %xmm7
+	# Fold across 16 bytes.
+	movdqa	.Lfold_across_16_bytes_consts(%rip), FOLD_CONSTS
+	fold_16_bytes	%xmm6, %xmm7
+
+	# Add 128 to get the correct number of data bytes remaining in 0...127
+	# (not counting xmm7), following the previous extra subtraction by 128.
+	# Then subtract 16 to simplify the termination condition of the
+	# following loop.
+	add	$128-16, len
+
+	# While >= 16 data bytes remain (not counting xmm7), fold the 16 bytes
+	# xmm7 into them, storing the result back into xmm7.
+	jl	.Lfold_16_bytes_loop_done
+.Lfold_16_bytes_loop:
 	movdqa	%xmm7, %xmm8
-	pclmulqdq	$0x11, %xmm10, %xmm7
-	pclmulqdq	$0x0 , %xmm10, %xmm8
+	pclmulqdq	$0x11, FOLD_CONSTS, %xmm7
+	pclmulqdq	$0x00, FOLD_CONSTS, %xmm8
 	pxor	%xmm8, %xmm7
-	movdqu	(arg2), %xmm0
-	pshufb	%xmm11, %xmm0
+	movdqu	(buf), %xmm0
+	pshufb	BSWAP_MASK, %xmm0
 	pxor	%xmm0 , %xmm7
-	add	$16, arg2
-	sub	$16, arg3
-	# instead of a cmp instruction, we utilize the flags with the
-	# jge instruction equivalent of: cmp arg3, 16-16
-	# check if there is any more 16B in the buffer to be able to fold
-	jge	_16B_reduction_loop
-
-	#now we have 16+z bytes left to reduce, where 0<= z < 16.
-	#first, we reduce the data in the xmm7 register
-
-
-_final_reduction_for_128:
-	# check if any more data to fold. If not, compute the CRC of
-	# the final 128 bits
-	add	$16, arg3
-	je	_128_done
-
-	# here we are getting data that is less than 16 bytes.
-	# since we know that there was data before the pointer, we can
-	# offset the input pointer before the actual point, to receive
-	# exactly 16 bytes. after that the registers need to be adjusted.
-_get_last_two_xmms:
+	add	$16, buf
+	sub	$16, len
+	jge	.Lfold_16_bytes_loop
+
+.Lfold_16_bytes_loop_done:
+	# Add 16 to get the correct number of data bytes remaining in 0...15
+	# (not counting xmm7), following the previous extra subtraction by 16.
+	add	$16, len
+	je	.Lreduce_final_16_bytes
+
+.Lhandle_partial_segment:
+	# Reduce the last '16 + len' bytes where 1 <= len <= 15 and the first 16
+	# bytes are in xmm7 and the rest are the remaining data in 'buf'.  To do
+	# this without needing a fold constant for each possible 'len', redivide
+	# the bytes into a first chunk of 'len' bytes and a second chunk of 16
+	# bytes, then fold the first chunk into the second.
+
 	movdqa	%xmm7, %xmm2
 
-	movdqu	-16(arg2, arg3), %xmm1
-	pshufb	%xmm11, %xmm1
+	# xmm1 = last 16 original data bytes
+	movdqu	-16(buf, len), %xmm1
+	pshufb	BSWAP_MASK, %xmm1
 
-	# get rid of the extra data that was loaded before
-	# load the shift constant
-	lea	pshufb_shf_table+16(%rip), %rax
-	sub	arg3, %rax
+	# xmm2 = high order part of second chunk: xmm7 left-shifted by 'len' bytes.
+	lea	.Lbyteshift_table+16(%rip), %rax
+	sub	len, %rax
 	movdqu	(%rax), %xmm0
-
-	# shift xmm2 to the left by arg3 bytes
 	pshufb	%xmm0, %xmm2
 
-	# shift xmm7 to the right by 16-arg3 bytes
-	pxor	mask1(%rip), %xmm0
+	# xmm7 = first chunk: xmm7 right-shifted by '16-len' bytes.
+	pxor	.Lmask1(%rip), %xmm0
 	pshufb	%xmm0, %xmm7
+
+	# xmm1 = second chunk: 'len' bytes from xmm1 (low-order bytes),
+	# then '16-len' bytes from xmm2 (high-order bytes).
 	pblendvb	%xmm2, %xmm1	#xmm0 is implicit
 
-	# fold 16 Bytes
-	movdqa	%xmm1, %xmm2
+	# Fold the first chunk into the second chunk, storing the result in xmm7.
 	movdqa	%xmm7, %xmm8
-	pclmulqdq	$0x11, %xmm10, %xmm7
-	pclmulqdq	$0x0 , %xmm10, %xmm8
+	pclmulqdq	$0x11, FOLD_CONSTS, %xmm7
+	pclmulqdq	$0x00, FOLD_CONSTS, %xmm8
 	pxor	%xmm8, %xmm7
-	pxor	%xmm2, %xmm7
+	pxor	%xmm1, %xmm7
 
-_128_done:
-	# compute crc of a 128-bit value
-	movdqa	rk5(%rip), %xmm10	# rk5 and rk6 in xmm10
-	movdqa	%xmm7, %xmm0
+.Lreduce_final_16_bytes:
+	# Reduce the 128-bit value M(x), stored in xmm7, to the final 16-bit CRC
 
-	#64b fold
-	pclmulqdq	$0x1, %xmm10, %xmm7
-	pslldq	$8   ,  %xmm0
-	pxor	%xmm0,  %xmm7
+	# Load 'x^48 * (x^48 mod G(x))' and 'x^48 * (x^80 mod G(x))'.
+	movdqa	.Lfinal_fold_consts(%rip), FOLD_CONSTS
 
-	#32b fold
+	# Fold the high 64 bits into the low 64 bits, while also multiplying by
+	# x^64.  This produces a 128-bit value congruent to x^64 * M(x) and
+	# whose low 48 bits are 0.
 	movdqa	%xmm7, %xmm0
+	pclmulqdq	$0x11, FOLD_CONSTS, %xmm7 # high bits * x^48 * (x^80 mod G(x))
+	pslldq	$8, %xmm0
+	pxor	%xmm0, %xmm7			  # + low bits * x^64
 
-	pand	mask2(%rip), %xmm0
-
-	psrldq	$12, %xmm7
-	pclmulqdq	$0x10, %xmm10, %xmm7
-	pxor	%xmm0, %xmm7
-
-	#barrett reduction
-_barrett:
-	movdqa	rk7(%rip), %xmm10	# rk7 and rk8 in xmm10
+	# Fold the high 32 bits into the low 96 bits.  This produces a 96-bit
+	# value congruent to x^64 * M(x) and whose low 48 bits are 0.
 	movdqa	%xmm7, %xmm0
-	pclmulqdq	$0x01, %xmm10, %xmm7
-	pslldq	$4, %xmm7
-	pclmulqdq	$0x11, %xmm10, %xmm7
+	pand	.Lmask2(%rip), %xmm0		  # zero high 32 bits
+	psrldq	$12, %xmm7			  # extract high 32 bits
+	pclmulqdq	$0x00, FOLD_CONSTS, %xmm7 # high 32 bits * x^48 * (x^48 mod G(x))
+	pxor	%xmm0, %xmm7			  # + low bits
 
-	pslldq	$4, %xmm7
-	pxor	%xmm0, %xmm7
-	pextrd	$1, %xmm7, %eax
+	# Load G(x) and floor(x^48 / G(x)).
+	movdqa	.Lbarrett_reduction_consts(%rip), FOLD_CONSTS
 
-_cleanup:
-	# scale the result back to 16 bits
-	shr	$16, %eax
-	mov     %rcx, %rsp
+	# Use Barrett reduction to compute the final CRC value.
+	movdqa	%xmm7, %xmm0
+	pclmulqdq	$0x11, FOLD_CONSTS, %xmm7 # high 32 bits * floor(x^48 / G(x))
+	psrlq	$32, %xmm7			  # /= x^32
+	pclmulqdq	$0x00, FOLD_CONSTS, %xmm7 # *= G(x)
+	psrlq	$48, %xmm0
+	pxor	%xmm7, %xmm0		     # + low 16 nonzero bits
+	# Final CRC value (x^16 * M(x)) mod G(x) is in low 16 bits of xmm0.
+
+	pextrw	$0, %xmm0, %eax
 	ret
 
-########################################################################
-
 .align 16
-_less_than_128:
-
-	# check if there is enough buffer to be able to fold 16B at a time
-	cmp	$32, arg3
-	jl	_less_than_32
-	movdqa  SHUF_MASK(%rip), %xmm11
+.Lless_than_256_bytes:
+	# Checksumming a buffer of length 16...255 bytes
 
-	# now if there is, load the constants
-	movdqa	rk1(%rip), %xmm10	# rk1 and rk2 in xmm10
+	# Load the first 16 data bytes.
+	movdqu	(buf), %xmm7
+	pshufb	BSWAP_MASK, %xmm7
+	add	$16, buf
 
-	movd	arg1_low32, %xmm0	# get the initial crc value
-	pslldq	$12, %xmm0	# align it to its correct place
-	movdqu	(arg2), %xmm7	# load the plaintext
-	pshufb	%xmm11, %xmm7	# byte-reflect the plaintext
+	# XOR the first 16 data *bits* with the initial CRC value.
+	pxor	%xmm0, %xmm0
+	pinsrw	$7, init_crc, %xmm0
 	pxor	%xmm0, %xmm7
 
-
-	# update the buffer pointer
-	add	$16, arg2
-
-	# update the counter. subtract 32 instead of 16 to save one
-	# instruction from the loop
-	sub	$32, arg3
-
-	jmp	_16B_reduction_loop
-
-
-.align 16
-_less_than_32:
-	# mov initial crc to the return value. this is necessary for
-	# zero-length buffers.
-	mov	arg1_low32, %eax
-	test	arg3, arg3
-	je	_cleanup
-
-	movdqa  SHUF_MASK(%rip), %xmm11
-
-	movd	arg1_low32, %xmm0	# get the initial crc value
-	pslldq	$12, %xmm0	# align it to its correct place
-
-	cmp	$16, arg3
-	je	_exact_16_left
-	jl	_less_than_16_left
-
-	movdqu	(arg2), %xmm7	# load the plaintext
-	pshufb	%xmm11, %xmm7	# byte-reflect the plaintext
-	pxor	%xmm0 , %xmm7	# xor the initial crc value
-	add	$16, arg2
-	sub	$16, arg3
-	movdqa	rk1(%rip), %xmm10	# rk1 and rk2 in xmm10
-	jmp	_get_last_two_xmms
-
-
-.align 16
-_less_than_16_left:
-	# use stack space to load data less than 16 bytes, zero-out
-	# the 16B in memory first.
-
-	pxor	%xmm1, %xmm1
-	mov	%rsp, %r11
-	movdqa	%xmm1, (%r11)
-
-	cmp	$4, arg3
-	jl	_only_less_than_4
-
-	# backup the counter value
-	mov	arg3, %r9
-	cmp	$8, arg3
-	jl	_less_than_8_left
-
-	# load 8 Bytes
-	mov	(arg2), %rax
-	mov	%rax, (%r11)
-	add	$8, %r11
-	sub	$8, arg3
-	add	$8, arg2
-_less_than_8_left:
-
-	cmp	$4, arg3
-	jl	_less_than_4_left
-
-	# load 4 Bytes
-	mov	(arg2), %eax
-	mov	%eax, (%r11)
-	add	$4, %r11
-	sub	$4, arg3
-	add	$4, arg2
-_less_than_4_left:
-
-	cmp	$2, arg3
-	jl	_less_than_2_left
-
-	# load 2 Bytes
-	mov	(arg2), %ax
-	mov	%ax, (%r11)
-	add	$2, %r11
-	sub	$2, arg3
-	add	$2, arg2
-_less_than_2_left:
-	cmp     $1, arg3
-        jl      _zero_left
-
-	# load 1 Byte
-	mov	(arg2), %al
-	mov	%al, (%r11)
-_zero_left:
-	movdqa	(%rsp), %xmm7
-	pshufb	%xmm11, %xmm7
-	pxor	%xmm0 , %xmm7	# xor the initial crc value
-
-	# shl r9, 4
-	lea	pshufb_shf_table+16(%rip), %rax
-	sub	%r9, %rax
-	movdqu	(%rax), %xmm0
-	pxor	mask1(%rip), %xmm0
-
-	pshufb	%xmm0, %xmm7
-	jmp	_128_done
-
-.align 16
-_exact_16_left:
-	movdqu	(arg2), %xmm7
-	pshufb	%xmm11, %xmm7
-	pxor	%xmm0 , %xmm7   # xor the initial crc value
-
-	jmp	_128_done
-
-_only_less_than_4:
-	cmp	$3, arg3
-	jl	_only_less_than_3
-
-	# load 3 Bytes
-	mov	(arg2), %al
-	mov	%al, (%r11)
-
-	mov	1(arg2), %al
-	mov	%al, 1(%r11)
-
-	mov	2(arg2), %al
-	mov	%al, 2(%r11)
-
-	movdqa	 (%rsp), %xmm7
-	pshufb	 %xmm11, %xmm7
-	pxor	 %xmm0 , %xmm7  # xor the initial crc value
-
-	psrldq	$5, %xmm7
-
-	jmp	_barrett
-_only_less_than_3:
-	cmp	$2, arg3
-	jl	_only_less_than_2
-
-	# load 2 Bytes
-	mov	(arg2), %al
-	mov	%al, (%r11)
-
-	mov	1(arg2), %al
-	mov	%al, 1(%r11)
-
-	movdqa	(%rsp), %xmm7
-	pshufb	%xmm11, %xmm7
-	pxor	%xmm0 , %xmm7   # xor the initial crc value
-
-	psrldq	$6, %xmm7
-
-	jmp	_barrett
-_only_less_than_2:
-
-	# load 1 Byte
-	mov	(arg2), %al
-	mov	%al, (%r11)
-
-	movdqa	(%rsp), %xmm7
-	pshufb	%xmm11, %xmm7
-	pxor	%xmm0 , %xmm7   # xor the initial crc value
-
-	psrldq	$7, %xmm7
-
-	jmp	_barrett
-
+	movdqa	.Lfold_across_16_bytes_consts(%rip), FOLD_CONSTS
+	cmp	$16, len
+	je	.Lreduce_final_16_bytes		# len == 16
+	sub	$32, len
+	jge	.Lfold_16_bytes_loop		# 32 <= len <= 255
+	add	$16, len
+	jmp	.Lhandle_partial_segment	# 17 <= len <= 31
 ENDPROC(crc_t10dif_pcl)
 
 .section	.rodata, "a", @progbits
 .align 16
-# precomputed constants
-# these constants are precomputed from the poly:
-# 0x8bb70000 (0x8bb7 scaled to 32 bits)
-# Q = 0x18BB70000
-# rk1 = 2^(32*3) mod Q << 32
-# rk2 = 2^(32*5) mod Q << 32
-# rk3 = 2^(32*15) mod Q << 32
-# rk4 = 2^(32*17) mod Q << 32
-# rk5 = 2^(32*3) mod Q << 32
-# rk6 = 2^(32*2) mod Q << 32
-# rk7 = floor(2^64/Q)
-# rk8 = Q
-rk1:
-.quad 0x2d56000000000000
-rk2:
-.quad 0x06df000000000000
-rk3:
-.quad 0x9d9d000000000000
-rk4:
-.quad 0x7cf5000000000000
-rk5:
-.quad 0x2d56000000000000
-rk6:
-.quad 0x1368000000000000
-rk7:
-.quad 0x00000001f65a57f8
-rk8:
-.quad 0x000000018bb70000
-
-rk9:
-.quad 0xceae000000000000
-rk10:
-.quad 0xbfd6000000000000
-rk11:
-.quad 0x1e16000000000000
-rk12:
-.quad 0x713c000000000000
-rk13:
-.quad 0xf7f9000000000000
-rk14:
-.quad 0x80a6000000000000
-rk15:
-.quad 0x044c000000000000
-rk16:
-.quad 0xe658000000000000
-rk17:
-.quad 0xad18000000000000
-rk18:
-.quad 0xa497000000000000
-rk19:
-.quad 0x6ee3000000000000
-rk20:
-.quad 0xe7b5000000000000
-
 
+# Fold constants precomputed from the polynomial 0x18bb7
+# G(x) = x^16 + x^15 + x^11 + x^9 + x^8 + x^7 + x^5 + x^4 + x^2 + x^1 + x^0
+.Lfold_across_128_bytes_consts:
+	.quad		0x0000000000006123	# x^(8*128)	mod G(x)
+	.quad		0x0000000000002295	# x^(8*128+64)	mod G(x)
+.Lfold_across_64_bytes_consts:
+	.quad		0x0000000000001069	# x^(4*128)	mod G(x)
+	.quad		0x000000000000dd31	# x^(4*128+64)	mod G(x)
+.Lfold_across_32_bytes_consts:
+	.quad		0x000000000000857d	# x^(2*128)	mod G(x)
+	.quad		0x0000000000007acc	# x^(2*128+64)	mod G(x)
+.Lfold_across_16_bytes_consts:
+	.quad		0x000000000000a010	# x^(1*128)	mod G(x)
+	.quad		0x0000000000001faa	# x^(1*128+64)	mod G(x)
+.Lfinal_fold_consts:
+	.quad		0x1368000000000000	# x^48 * (x^48 mod G(x))
+	.quad		0x2d56000000000000	# x^48 * (x^80 mod G(x))
+.Lbarrett_reduction_consts:
+	.quad		0x0000000000018bb7	# G(x)
+	.quad		0x00000001f65a57f8	# floor(x^48 / G(x))
 
 .section	.rodata.cst16.mask1, "aM", @progbits, 16
 .align 16
-mask1:
-.octa 0x80808080808080808080808080808080
+.Lmask1:
+	.octa	0x80808080808080808080808080808080
 
 .section	.rodata.cst16.mask2, "aM", @progbits, 16
 .align 16
-mask2:
-.octa 0x00000000FFFFFFFFFFFFFFFFFFFFFFFF
+.Lmask2:
+	.octa	0x00000000FFFFFFFFFFFFFFFFFFFFFFFF
+
+.section	.rodata.cst16.bswap_mask, "aM", @progbits, 16
+.align 16
+.Lbswap_mask:
+	.octa	0x000102030405060708090A0B0C0D0E0F
 
-.section	.rodata.cst16.SHUF_MASK, "aM", @progbits, 16
+.section	.rodata.cst32.byteshift_table, "aM", @progbits, 32
 .align 16
-SHUF_MASK:
-.octa 0x000102030405060708090A0B0C0D0E0F
-
-.section	.rodata.cst32.pshufb_shf_table, "aM", @progbits, 32
-.align 32
-pshufb_shf_table:
-# use these values for shift constants for the pshufb instruction
-# different alignments result in values as shown:
-#	DDQ 0x008f8e8d8c8b8a898887868584838281 # shl 15 (16-1) / shr1
-#	DDQ 0x01008f8e8d8c8b8a8988878685848382 # shl 14 (16-3) / shr2
-#	DDQ 0x0201008f8e8d8c8b8a89888786858483 # shl 13 (16-4) / shr3
-#	DDQ 0x030201008f8e8d8c8b8a898887868584 # shl 12 (16-4) / shr4
-#	DDQ 0x04030201008f8e8d8c8b8a8988878685 # shl 11 (16-5) / shr5
-#	DDQ 0x0504030201008f8e8d8c8b8a89888786 # shl 10 (16-6) / shr6
-#	DDQ 0x060504030201008f8e8d8c8b8a898887 # shl 9  (16-7) / shr7
-#	DDQ 0x07060504030201008f8e8d8c8b8a8988 # shl 8  (16-8) / shr8
-#	DDQ 0x0807060504030201008f8e8d8c8b8a89 # shl 7  (16-9) / shr9
-#	DDQ 0x090807060504030201008f8e8d8c8b8a # shl 6  (16-10) / shr10
-#	DDQ 0x0a090807060504030201008f8e8d8c8b # shl 5  (16-11) / shr11
-#	DDQ 0x0b0a090807060504030201008f8e8d8c # shl 4  (16-12) / shr12
-#	DDQ 0x0c0b0a090807060504030201008f8e8d # shl 3  (16-13) / shr13
-#	DDQ 0x0d0c0b0a090807060504030201008f8e # shl 2  (16-14) / shr14
-#	DDQ 0x0e0d0c0b0a090807060504030201008f # shl 1  (16-15) / shr15
-.octa 0x8f8e8d8c8b8a89888786858483828100
-.octa 0x000e0d0c0b0a09080706050403020100
+# For 1 <= len <= 15, the 16-byte vector beginning at &byteshift_table[16 - len]
+# is the index vector to shift left by 'len' bytes, and is also {0x80, ...,
+# 0x80} XOR the index vector to shift right by '16 - len' bytes.
+.Lbyteshift_table:
+	.byte		 0x0, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87
+	.byte		0x88, 0x89, 0x8a, 0x8b, 0x8c, 0x8d, 0x8e, 0x8f
+	.byte		 0x0,  0x1,  0x2,  0x3,  0x4,  0x5,  0x6,  0x7
+	.byte		 0x8,  0x9,  0xa,  0xb,  0xc,  0xd,  0xe , 0x0
diff --git a/arch/x86/crypto/crct10dif-pclmul_glue.c b/arch/x86/crypto/crct10dif-pclmul_glue.c
index cd4df9322501..0e785c0b2354 100644
--- a/arch/x86/crypto/crct10dif-pclmul_glue.c
+++ b/arch/x86/crypto/crct10dif-pclmul_glue.c
@@ -33,18 +33,12 @@
 #include <asm/cpufeatures.h>
 #include <asm/cpu_device_id.h>
 
-asmlinkage __u16 crc_t10dif_pcl(__u16 crc, const unsigned char *buf,
-				size_t len);
+asmlinkage u16 crc_t10dif_pcl(u16 init_crc, const u8 *buf, size_t len);
 
 struct chksum_desc_ctx {
 	__u16 crc;
 };
 
-/*
- * Steps through buffer one byte at at time, calculates reflected
- * crc using table.
- */
-
 static int chksum_init(struct shash_desc *desc)
 {
 	struct chksum_desc_ctx *ctx = shash_desc_ctx(desc);
@@ -59,7 +53,7 @@ static int chksum_update(struct shash_desc *desc, const u8 *data,
 {
 	struct chksum_desc_ctx *ctx = shash_desc_ctx(desc);
 
-	if (irq_fpu_usable()) {
+	if (length >= 16 && irq_fpu_usable()) {
 		kernel_fpu_begin();
 		ctx->crc = crc_t10dif_pcl(ctx->crc, data, length);
 		kernel_fpu_end();
@@ -79,7 +73,7 @@ static int chksum_final(struct shash_desc *desc, u8 *out)
 static int __chksum_finup(__u16 *crcp, const u8 *data, unsigned int len,
 			u8 *out)
 {
-	if (irq_fpu_usable()) {
+	if (len >= 16 && irq_fpu_usable()) {
 		kernel_fpu_begin();
 		*(__u16 *)out = crc_t10dif_pcl(*crcp, data, len);
 		kernel_fpu_end();
-- 
cgit v1.2.3


From e7b3ed338097b1237ca03b05fe36ec1517d328d5 Mon Sep 17 00:00:00 2001
From: Eric Biggers <ebiggers@google.com>
Date: Wed, 30 Jan 2019 20:42:41 -0800
Subject: crypto: arm/crct10dif-ce - cleanup and optimizations

The x86, arm, and arm64 asm implementations of crct10dif are very
difficult to understand partly because many of the comments, labels, and
macros are named incorrectly: the lengths mentioned are usually off by a
factor of two from the actual code.  Many other things are unnecessarily
convoluted as well, e.g. there are many more fold constants than
actually needed and some aren't fully reduced.

This series therefore cleans up all these implementations to be much
more maintainable.  I also made some small optimizations where I saw
opportunities, resulting in slightly better performance.

This patch cleans up the arm version.

(Also moved the constants to .rodata as suggested by Ard Biesheuvel.)

Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
---
 arch/arm/crypto/crct10dif-ce-core.S | 553 +++++++++++++++++-------------------
 arch/arm/crypto/crct10dif-ce-glue.c |   2 +-
 2 files changed, 261 insertions(+), 294 deletions(-)

(limited to 'arch')

diff --git a/arch/arm/crypto/crct10dif-ce-core.S b/arch/arm/crypto/crct10dif-ce-core.S
index d058fad423c2..86be258a803f 100644
--- a/arch/arm/crypto/crct10dif-ce-core.S
+++ b/arch/arm/crypto/crct10dif-ce-core.S
@@ -2,12 +2,14 @@
 // Accelerated CRC-T10DIF using ARM NEON and Crypto Extensions instructions
 //
 // Copyright (C) 2016 Linaro Ltd <ard.biesheuvel@linaro.org>
+// Copyright (C) 2019 Google LLC <ebiggers@google.com>
 //
 // 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.
 //
 
+// Derived from the x86 version:
 //
 // Implement fast CRC-T10DIF computation with SSE and PCLMULQDQ instructions
 //
@@ -54,19 +56,11 @@
 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 //
-//       Function API:
-//       UINT16 crc_t10dif_pcl(
-//               UINT16 init_crc, //initial CRC value, 16 bits
-//               const unsigned char *buf, //buffer pointer to calculate CRC on
-//               UINT64 len //buffer length in bytes (64-bit data)
-//       );
-//
 //       Reference paper titled "Fast CRC Computation for Generic
 //	Polynomials Using PCLMULQDQ Instruction"
 //       URL: http://www.intel.com/content/dam/www/public/us/en/documents
 //  /white-papers/fast-crc-computation-generic-polynomials-pclmulqdq-paper.pdf
 //
-//
 
 #include <linux/linkage.h>
 #include <asm/assembler.h>
@@ -78,13 +72,14 @@
 #endif
 
 	.text
+	.arch		armv7-a
 	.fpu		crypto-neon-fp-armv8
 
-	arg1_low32	.req	r0
-	arg2		.req	r1
-	arg3		.req	r2
+	init_crc	.req	r0
+	buf		.req	r1
+	len		.req	r2
 
-	qzr		.req	q13
+	fold_consts_ptr	.req	ip
 
 	q0l		.req	d0
 	q0h		.req	d1
@@ -102,82 +97,35 @@
 	q6h		.req	d13
 	q7l		.req	d14
 	q7h		.req	d15
-
-ENTRY(crc_t10dif_pmull)
-	vmov.i8		qzr, #0			// init zero register
-
-	// adjust the 16-bit initial_crc value, scale it to 32 bits
-	lsl		arg1_low32, arg1_low32, #16
-
-	// check if smaller than 256
-	cmp		arg3, #256
-
-	// for sizes less than 128, we can't fold 64B at a time...
-	blt		_less_than_128
-
-	// load the initial crc value
-	// crc value does not need to be byte-reflected, but it needs
-	// to be moved to the high part of the register.
-	// because data will be byte-reflected and will align with
-	// initial crc at correct place.
-	vmov		s0, arg1_low32		// initial crc
-	vext.8		q10, qzr, q0, #4
-
-	// receive the initial 64B data, xor the initial crc value
-	vld1.64		{q0-q1}