diff mbox series

[v2,3/6] crypto: x86/aes-xts - wire up AESNI + AVX implementation

Message ID 20240329080355.2871-4-ebiggers@kernel.org (mailing list archive)
State Accepted
Delegated to: Herbert Xu
Headers show
Series Faster AES-XTS on modern x86_64 CPUs | expand

Commit Message

Eric Biggers March 29, 2024, 8:03 a.m. UTC
From: Eric Biggers <ebiggers@google.com>

Add an AES-XTS implementation "xts-aes-aesni-avx" for x86_64 CPUs that
have the AES-NI and AVX extensions but not VAES.  It's similar to the
existing xts-aes-aesni in that uses xmm registers to operate on one AES
block at a time.  It differs from xts-aes-aesni in the following ways:

- It uses the VEX-coded (non-destructive) instructions from AVX.
  This improves performance slightly.
- It incorporates some additional optimizations such as interleaving the
  tweak computation with AES en/decryption, handling single-page
  messages more efficiently, and caching the first round key.
- It supports only 64-bit (x86_64).
- It's generated by an assembly macro that will also be used to generate
  VAES-based implementations.

The performance improvement over xts-aes-aesni varies from small to
large, depending on the CPU and other factors such as the size of the
messages en/decrypted.  For example, the following increases in
AES-256-XTS decryption throughput are seen on the following CPUs:

                          | 4096-byte messages | 512-byte messages |
    ----------------------+--------------------+-------------------+
    Intel Skylake         |        6%          |       31%         |
    Intel Cascade Lake    |        4%          |       26%         |
    AMD Zen 1             |        61%         |       73%         |
    AMD Zen 2             |        36%         |       59%         |

(The above CPUs don't support VAES, so they can't use VAES instead.)

While this isn't as large an improvement as what VAES provides, this
still seems worthwhile.  This implementation is fairly easy to provide
based on the assembly macro that's needed for VAES anyway, and it will
be the best implementation on a large number of CPUs (very roughly, the
CPUs launched by Intel and AMD from 2011 to 2018).

This makes the existing xts-aes-aesni *mostly* obsolete.  For now, leave
it in place to support 32-bit kernels and also CPUs like Intel Westmere
that support AES-NI but not AVX.  (We could potentially remove it anyway
and just rely on the indirect acceleration via ecb-aes-aesni in those
cases, but that change will need to be considered separately.)

Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 arch/x86/crypto/aes-xts-avx-x86_64.S |   9 ++
 arch/x86/crypto/aesni-intel_glue.c   | 202 ++++++++++++++++++++++++++-
 2 files changed, 209 insertions(+), 2 deletions(-)
diff mbox series

Patch

diff --git a/arch/x86/crypto/aes-xts-avx-x86_64.S b/arch/x86/crypto/aes-xts-avx-x86_64.S
index a5e2783c46ec..32e26f562cf0 100644
--- a/arch/x86/crypto/aes-xts-avx-x86_64.S
+++ b/arch/x86/crypto/aes-xts-avx-x86_64.S
@@ -796,5 +796,14 @@  SYM_FUNC_END(aes_xts_encrypt_iv)
 // |key| is the data key.  |tweak| contains the next tweak; the encryption of
 // the original IV with the tweak key was already done.  This function supports
 // incremental computation, but |len| must always be >= 16 (AES_BLOCK_SIZE), and
 // |len| must be a multiple of 16 except on the last call.  If |len| is a
 // multiple of 16, then this function updates |tweak| to contain the next tweak.
+
+.set	VL, 16
+.set	USE_AVX10, 0
+SYM_TYPED_FUNC_START(aes_xts_encrypt_aesni_avx)
+	_aes_xts_crypt	1
+SYM_FUNC_END(aes_xts_encrypt_aesni_avx)
+SYM_TYPED_FUNC_START(aes_xts_decrypt_aesni_avx)
+	_aes_xts_crypt	0
+SYM_FUNC_END(aes_xts_decrypt_aesni_avx)
diff --git a/arch/x86/crypto/aesni-intel_glue.c b/arch/x86/crypto/aesni-intel_glue.c
index b1d90c25975a..10e283721a85 100644
--- a/arch/x86/crypto/aesni-intel_glue.c
+++ b/arch/x86/crypto/aesni-intel_glue.c
@@ -1135,11 +1135,200 @@  static struct skcipher_alg aesni_xctr = {
 	.encrypt	= xctr_crypt,
 	.decrypt	= xctr_crypt,
 };
 
 static struct simd_skcipher_alg *aesni_simd_xctr;
-#endif /* CONFIG_X86_64 */
+
+asmlinkage void aes_xts_encrypt_iv(const struct crypto_aes_ctx *tweak_key,
+				   u8 iv[AES_BLOCK_SIZE]);
+
+typedef void (*xts_asm_func)(const struct crypto_aes_ctx *key,
+			     const u8 *src, u8 *dst, size_t len,
+			     u8 tweak[AES_BLOCK_SIZE]);
+
+/* This handles cases where the source and/or destination span pages. */
+static noinline int
+xts_crypt_slowpath(struct skcipher_request *req, xts_asm_func asm_func)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	const struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
+	int tail = req->cryptlen % AES_BLOCK_SIZE;
+	struct scatterlist sg_src[2], sg_dst[2];
+	struct skcipher_request subreq;
+	struct skcipher_walk walk;
+	struct scatterlist *src, *dst;
+	int err;
+
+	/*
+	 * If the message length isn't divisible by the AES block size, then
+	 * separate off the last full block and the partial block.  This ensures
+	 * that they are processed in the same call to the assembly function,
+	 * which is required for ciphertext stealing.
+	 */
+	if (tail) {
+		skcipher_request_set_tfm(&subreq, tfm);
+		skcipher_request_set_callback(&subreq,
+					      skcipher_request_flags(req),
+					      NULL, NULL);
+		skcipher_request_set_crypt(&subreq, req->src, req->dst,
+					   req->cryptlen - tail - AES_BLOCK_SIZE,
+					   req->iv);
+		req = &subreq;
+	}
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while (walk.nbytes) {
+		unsigned int nbytes = walk.nbytes;
+
+		if (nbytes < walk.total)
+			nbytes = round_down(nbytes, AES_BLOCK_SIZE);
+
+		kernel_fpu_begin();
+		(*asm_func)(&ctx->crypt_ctx, walk.src.virt.addr,
+			    walk.dst.virt.addr, nbytes, req->iv);
+		kernel_fpu_end();
+		err = skcipher_walk_done(&walk, walk.nbytes - nbytes);
+	}
+
+	if (err || !tail)
+		return err;
+
+	/* Do ciphertext stealing with the last full block and partial block. */
+
+	dst = src = scatterwalk_ffwd(sg_src, req->src, req->cryptlen);
+	if (req->dst != req->src)
+		dst = scatterwalk_ffwd(sg_dst, req->dst, req->cryptlen);
+
+	skcipher_request_set_crypt(req, src, dst, AES_BLOCK_SIZE + tail,
+				   req->iv);
+
+	err = skcipher_walk_virt(&walk, req, false);
+	if (err)
+		return err;
+
+	kernel_fpu_begin();
+	(*asm_func)(&ctx->crypt_ctx, walk.src.virt.addr, walk.dst.virt.addr,
+		    walk.nbytes, req->iv);
+	kernel_fpu_end();
+
+	return skcipher_walk_done(&walk, 0);
+}
+
+/* __always_inline to avoid indirect call in fastpath */
+static __always_inline int
+xts_crypt2(struct skcipher_request *req, xts_asm_func asm_func)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	const struct aesni_xts_ctx *ctx = aes_xts_ctx(tfm);
+	const unsigned int cryptlen = req->cryptlen;
+	struct scatterlist *src = req->src;
+	struct scatterlist *dst = req->dst;
+
+	if (unlikely(cryptlen < AES_BLOCK_SIZE))
+		return -EINVAL;
+
+	kernel_fpu_begin();
+	aes_xts_encrypt_iv(&ctx->tweak_ctx, req->iv);
+
+	/*
+	 * In practice, virtually all XTS plaintexts and ciphertexts are either
+	 * 512 or 4096 bytes, aligned such that they don't span page boundaries.
+	 * To optimize the performance of these cases, and also any other case
+	 * where no page boundary is spanned, the below fast-path handles
+	 * single-page sources and destinations as efficiently as possible.
+	 */
+	if (likely(src->length >= cryptlen && dst->length >= cryptlen &&
+		   src->offset + cryptlen <= PAGE_SIZE &&
+		   dst->offset + cryptlen <= PAGE_SIZE)) {
+		struct page *src_page = sg_page(src);
+		struct page *dst_page = sg_page(dst);
+		void *src_virt = kmap_local_page(src_page) + src->offset;
+		void *dst_virt = kmap_local_page(dst_page) + dst->offset;
+
+		(*asm_func)(&ctx->crypt_ctx, src_virt, dst_virt, cryptlen,
+			    req->iv);
+		kunmap_local(dst_virt);
+		kunmap_local(src_virt);
+		kernel_fpu_end();
+		return 0;
+	}
+	kernel_fpu_end();
+	return xts_crypt_slowpath(req, asm_func);
+}
+
+#define DEFINE_XTS_ALG(suffix, driver_name, priority)			       \
+									       \
+asmlinkage void aes_xts_encrypt_##suffix(const struct crypto_aes_ctx *key,     \
+					 const u8 *src, u8 *dst, size_t len,   \
+					 u8 tweak[AES_BLOCK_SIZE]);	       \
+asmlinkage void aes_xts_decrypt_##suffix(const struct crypto_aes_ctx *key,     \
+					 const u8 *src, u8 *dst, size_t len,   \
+					 u8 tweak[AES_BLOCK_SIZE]);	       \
+									       \
+static int xts_encrypt_##suffix(struct skcipher_request *req)		       \
+{									       \
+	return xts_crypt2(req, aes_xts_encrypt_##suffix);		       \
+}									       \
+									       \
+static int xts_decrypt_##suffix(struct skcipher_request *req)		       \
+{									       \
+	return xts_crypt2(req, aes_xts_decrypt_##suffix);		       \
+}									       \
+									       \
+static struct skcipher_alg aes_xts_alg_##suffix = {			       \
+	.base = {							       \
+		.cra_name		= "__xts(aes)",			       \
+		.cra_driver_name	= "__" driver_name,		       \
+		.cra_priority		= priority,			       \
+		.cra_flags		= CRYPTO_ALG_INTERNAL,		       \
+		.cra_blocksize		= AES_BLOCK_SIZE,		       \
+		.cra_ctxsize		= XTS_AES_CTX_SIZE,		       \
+		.cra_module		= THIS_MODULE,			       \
+	},								       \
+	.min_keysize	= 2 * AES_MIN_KEY_SIZE,				       \
+	.max_keysize	= 2 * AES_MAX_KEY_SIZE,				       \
+	.ivsize		= AES_BLOCK_SIZE,				       \
+	.walksize	= 2 * AES_BLOCK_SIZE,				       \
+	.setkey		= xts_aesni_setkey,				       \
+	.encrypt	= xts_encrypt_##suffix,				       \
+	.decrypt	= xts_decrypt_##suffix,				       \
+};									       \
+									       \
+static struct simd_skcipher_alg *aes_xts_simdalg_##suffix
+
+DEFINE_XTS_ALG(aesni_avx, "xts-aes-aesni-avx", 500);
+
+static int __init register_xts_algs(void)
+{
+	int err;
+
+	if (!boot_cpu_has(X86_FEATURE_AVX))
+		return 0;
+	err = simd_register_skciphers_compat(&aes_xts_alg_aesni_avx, 1,
+					     &aes_xts_simdalg_aesni_avx);
+	if (err)
+		return err;
+	return 0;
+}
+
+static void unregister_xts_algs(void)
+{
+	if (aes_xts_simdalg_aesni_avx)
+		simd_unregister_skciphers(&aes_xts_alg_aesni_avx, 1,
+					  &aes_xts_simdalg_aesni_avx);
+}
+#else /* CONFIG_X86_64 */
+static int __init register_xts_algs(void)
+{
+	return 0;
+}
+
+static void unregister_xts_algs(void)
+{
+}
+#endif /* !CONFIG_X86_64 */
 
 #ifdef CONFIG_X86_64
 static int generic_gcmaes_set_key(struct crypto_aead *aead, const u8 *key,
 				  unsigned int key_len)
 {
@@ -1274,17 +1463,25 @@  static int __init aesni_init(void)
 						     &aesni_simd_xctr);
 	if (err)
 		goto unregister_aeads;
 #endif /* CONFIG_X86_64 */
 
+	err = register_xts_algs();
+	if (err)
+		goto unregister_xts;
+
 	return 0;
 
+unregister_xts:
+	unregister_xts_algs();
 #ifdef CONFIG_X86_64
+	if (aesni_simd_xctr)
+		simd_unregister_skciphers(&aesni_xctr, 1, &aesni_simd_xctr);
 unregister_aeads:
+#endif /* CONFIG_X86_64 */
 	simd_unregister_aeads(aesni_aeads, ARRAY_SIZE(aesni_aeads),
 				aesni_simd_aeads);
-#endif /* CONFIG_X86_64 */
 
 unregister_skciphers:
 	simd_unregister_skciphers(aesni_skciphers, ARRAY_SIZE(aesni_skciphers),
 				  aesni_simd_skciphers);
 unregister_cipher:
@@ -1301,10 +1498,11 @@  static void __exit aesni_exit(void)
 	crypto_unregister_alg(&aesni_cipher_alg);
 #ifdef CONFIG_X86_64
 	if (boot_cpu_has(X86_FEATURE_AVX))
 		simd_unregister_skciphers(&aesni_xctr, 1, &aesni_simd_xctr);
 #endif /* CONFIG_X86_64 */
+	unregister_xts_algs();
 }
 
 late_initcall(aesni_init);
 module_exit(aesni_exit);