diff mbox series

[v4,12/13] crypto: x86/aes-kl - Support CBC mode

Message ID 20211214005212.20588-13-chang.seok.bae@intel.com (mailing list archive)
State Changes Requested
Delegated to: Herbert Xu
Headers show
Series x86: Support Key Locker | expand

Commit Message

Chang S. Bae Dec. 14, 2021, 12:52 a.m. UTC 
Implement CBC using AES-KL. Export the methods with a lower priority than
AES-NI to avoid from selected by default.

Signed-off-by: Chang S. Bae <chang.seok.bae@intel.com>
Acked-by: Dan Williams <dan.j.williams@intel.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
Cc: x86@kernel.org
Cc: linux-crypto@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
---
Changes from v3:
* Adjust the code as ECB mode was dropped. (Eric Biggers)

Changes from RFC v2:
* Separate out the code as a new patch.
---
 arch/x86/crypto/aeskl-intel_asm.S  | 381 +++++++++++++++++++++++++++++
 arch/x86/crypto/aeskl-intel_glue.c | 101 +++++++-
 2 files changed, 479 insertions(+), 3 deletions(-)
diff mbox series

Patch

diff --git a/arch/x86/crypto/aeskl-intel_asm.S b/arch/x86/crypto/aeskl-intel_asm.S
index d56ec8dd6644..5ee7b24ee3c8 100644
--- a/arch/x86/crypto/aeskl-intel_asm.S
+++ b/arch/x86/crypto/aeskl-intel_asm.S
@@ -182,3 +182,384 @@  SYM_FUNC_START(_aeskl_dec)
 	ret
 SYM_FUNC_END(_aeskl_dec)
 
+/*
+ * int _aeskl_ecb_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
+ *		      size_t len)
+ */
+SYM_FUNC_START(_aeskl_ecb_enc)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl LEN
+	pushl HANDLEP
+	pushl KLEN
+	movl (FRAME_OFFSET+16)(%esp), HANDLEP	# ctx
+	movl (FRAME_OFFSET+20)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+24)(%esp), INP	# src
+	movl (FRAME_OFFSET+28)(%esp), LEN	# len
+#endif
+	test LEN, LEN
+	jz .Lecb_enc_noerr
+	mov 480(HANDLEP), KLEN
+	cmp $16, LEN
+	jb .Lecb_enc_noerr
+	cmp $128, LEN
+	jb .Lecb_enc1
+
+.align 4
+.Lecb_enc8:
+	movdqu (INP), STATE1
+	movdqu 0x10(INP), STATE2
+	movdqu 0x20(INP), STATE3
+	movdqu 0x30(INP), STATE4
+	movdqu 0x40(INP), STATE5
+	movdqu 0x50(INP), STATE6
+	movdqu 0x60(INP), STATE7
+	movdqu 0x70(INP), STATE8
+
+	cmp $16, KLEN
+	je .Lecb_enc8_128
+	aesencwide256kl (HANDLEP)
+	jz .Lecb_enc_err
+	jmp .Lecb_enc8_end
+.Lecb_enc8_128:
+	aesencwide128kl (HANDLEP)
+	jz .Lecb_enc_err
+
+.Lecb_enc8_end:
+	movdqu STATE1, (OUTP)
+	movdqu STATE2, 0x10(OUTP)
+	movdqu STATE3, 0x20(OUTP)
+	movdqu STATE4, 0x30(OUTP)
+	movdqu STATE5, 0x40(OUTP)
+	movdqu STATE6, 0x50(OUTP)
+	movdqu STATE7, 0x60(OUTP)
+	movdqu STATE8, 0x70(OUTP)
+
+	sub $128, LEN
+	add $128, INP
+	add $128, OUTP
+	cmp $128, LEN
+	jge .Lecb_enc8
+	cmp $16, LEN
+	jb .Lecb_enc_noerr
+
+.align 4
+.Lecb_enc1:
+	movdqu (INP), STATE1
+	cmp $16, KLEN
+	je .Lecb_enc1_128
+	aesenc256kl (HANDLEP), STATE
+	jz .Lecb_enc_err
+	jmp .Lecb_enc1_end
+.Lecb_enc1_128:
+	aesenc128kl (HANDLEP), STATE
+	jz .Lecb_enc_err
+
+.Lecb_enc1_end:
+	movdqu STATE1, (OUTP)
+	sub $16, LEN
+	add $16, INP
+	add $16, OUTP
+	cmp $16, LEN
+	jge .Lecb_enc1
+
+.Lecb_enc_noerr:
+	xor AREG, AREG
+	jmp .Lecb_enc_end
+.Lecb_enc_err:
+	mov $1, AREG
+.Lecb_enc_end:
+#ifndef __x86_64__
+	popl KLEN
+	popl HANDLEP
+	popl LEN
+#endif
+	FRAME_END
+	ret
+SYM_FUNC_END(_aeskl_ecb_enc)
+
+/*
+ * int _aeskl_ecb_dec(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
+ *		      size_t len)
+ */
+SYM_FUNC_START(_aeskl_ecb_dec)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl LEN
+	pushl HANDLEP
+	pushl KLEN
+	movl (FRAME_OFFSET+16)(%esp), HANDLEP	# ctx
+	movl (FRAME_OFFSET+20)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+24)(%esp), INP	# src
+	movl (FRAME_OFFSET+28)(%esp), LEN	# len
+#endif
+
+	test LEN, LEN
+	jz .Lecb_dec_noerr
+	mov 480(HANDLEP), KLEN
+	cmp $16, LEN
+	jb .Lecb_dec_noerr
+	cmp $128, LEN
+	jb .Lecb_dec1
+
+.align 4
+.Lecb_dec8:
+	movdqu (INP), STATE1
+	movdqu 0x10(INP), STATE2
+	movdqu 0x20(INP), STATE3
+	movdqu 0x30(INP), STATE4
+	movdqu 0x40(INP), STATE5
+	movdqu 0x50(INP), STATE6
+	movdqu 0x60(INP), STATE7
+	movdqu 0x70(INP), STATE8
+
+	cmp $16, KLEN
+	je .Lecb_dec8_128
+	aesdecwide256kl (HANDLEP)
+	jz .Lecb_dec_err
+	jmp .Lecb_dec8_end
+.Lecb_dec8_128:
+	aesdecwide128kl (HANDLEP)
+	jz .Lecb_dec_err
+
+.Lecb_dec8_end:
+	movdqu STATE1, (OUTP)
+	movdqu STATE2, 0x10(OUTP)
+	movdqu STATE3, 0x20(OUTP)
+	movdqu STATE4, 0x30(OUTP)
+	movdqu STATE5, 0x40(OUTP)
+	movdqu STATE6, 0x50(OUTP)
+	movdqu STATE7, 0x60(OUTP)
+	movdqu STATE8, 0x70(OUTP)
+
+	sub $128, LEN
+	add $128, INP
+	add $128, OUTP
+	cmp $128, LEN
+	jge .Lecb_dec8
+	cmp $16, LEN
+	jb .Lecb_dec_noerr
+
+.align 4
+.Lecb_dec1:
+	movdqu (INP), STATE1
+	cmp $16, KLEN
+	je .Lecb_dec1_128
+	aesdec256kl (HANDLEP), STATE
+	jz .Lecb_dec_err
+	jmp .Lecb_dec1_end
+.Lecb_dec1_128:
+	aesdec128kl (HANDLEP), STATE
+	jz .Lecb_dec_err
+
+.Lecb_dec1_end:
+	movdqu STATE1, (OUTP)
+	sub $16, LEN
+	add $16, INP
+	add $16, OUTP
+	cmp $16, LEN
+	jge .Lecb_dec1
+
+.Lecb_dec_noerr:
+	xor AREG, AREG
+	jmp .Lecb_dec_end
+.Lecb_dec_err:
+	mov $1, AREG
+.Lecb_dec_end:
+#ifndef __x86_64__
+	popl KLEN
+	popl HANDLEP
+	popl LEN
+#endif
+	FRAME_END
+	ret
+SYM_FUNC_END(_aeskl_ecb_dec)
+
+/*
+ * int _aeskl_cbc_enc(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
+ *		      size_t len, u8 *iv)
+ */
+SYM_FUNC_START(_aeskl_cbc_enc)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl IVP
+	pushl LEN
+	pushl HANDLEP
+	pushl KLEN
+	movl (FRAME_OFFSET+20)(%esp), HANDLEP	# ctx
+	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+28)(%esp), INP	# src
+	movl (FRAME_OFFSET+32)(%esp), LEN	# len
+	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
+#endif
+
+	cmp $16, LEN
+	jb .Lcbc_enc_noerr
+	mov 480(HANDLEP), KLEN
+	movdqu (IVP), STATE
+
+.align 4
+.Lcbc_enc1:
+	movdqu (INP), IN
+	pxor IN, STATE
+
+	cmp $16, KLEN
+	je .Lcbc_enc1_128
+	aesenc256kl (HANDLEP), STATE
+	jz .Lcbc_enc_err
+	jmp .Lcbc_enc1_end
+.Lcbc_enc1_128:
+	aesenc128kl (HANDLEP), STATE
+	jz .Lcbc_enc_err
+
+.Lcbc_enc1_end:
+	movdqu STATE, (OUTP)
+	sub $16, LEN
+	add $16, INP
+	add $16, OUTP
+	cmp $16, LEN
+	jge .Lcbc_enc1
+	movdqu STATE, (IVP)
+
+.Lcbc_enc_noerr:
+	xor AREG, AREG
+	jmp .Lcbc_enc_end
+.Lcbc_enc_err:
+	mov $1, AREG
+.Lcbc_enc_end:
+#ifndef __x86_64__
+	popl KLEN
+	popl HANDLEP
+	popl LEN
+	popl IVP
+#endif
+	FRAME_END
+	ret
+SYM_FUNC_END(_aeskl_cbc_enc)
+
+/*
+ * int _aeskl_cbc_dec(struct crypto_aes_ctx *ctx, const u8 *dst, u8 *src,
+ *		      size_t len, u8 *iv)
+ */
+SYM_FUNC_START(_aeskl_cbc_dec)
+	FRAME_BEGIN
+#ifndef __x86_64__
+	pushl IVP
+	pushl LEN
+	pushl HANDLEP
+	pushl KLEN
+	movl (FRAME_OFFSET+20)(%esp), HANDLEP	# ctx
+	movl (FRAME_OFFSET+24)(%esp), OUTP	# dst
+	movl (FRAME_OFFSET+28)(%esp), INP	# src
+	movl (FRAME_OFFSET+32)(%esp), LEN	# len
+	movl (FRAME_OFFSET+36)(%esp), IVP	# iv
+#endif
+
+	cmp $16, LEN
+	jb .Lcbc_dec_noerr
+	mov 480(HANDLEP), KLEN
+#ifdef __x86_64__
+	cmp $128, LEN
+	jb .Lcbc_dec1_pre
+
+.align 4
+.Lcbc_dec8:
+	movdqu 0x0(INP), STATE1
+	movdqu 0x10(INP), STATE2
+	movdqu 0x20(INP), STATE3
+	movdqu 0x30(INP), STATE4
+	movdqu 0x40(INP), STATE5
+	movdqu 0x50(INP), STATE6
+	movdqu 0x60(INP), STATE7
+	movdqu 0x70(INP), STATE8
+
+	movdqu (IVP), IN1
+	movdqa STATE1, IN2
+	movdqa STATE2, IN3
+	movdqa STATE3, IN4
+	movdqa STATE4, IN5
+	movdqa STATE5, IN6
+	movdqa STATE6, IN7
+	movdqa STATE7, IN8
+	movdqu STATE8, (IVP)
+
+	cmp $16, KLEN
+	je .Lcbc_dec8_128
+	aesdecwide256kl (HANDLEP)
+	jz .Lcbc_dec_err
+	jmp .Lcbc_dec8_end
+.Lcbc_dec8_128:
+	aesdecwide128kl (HANDLEP)
+	jz .Lcbc_dec_err
+
+.Lcbc_dec8_end:
+	pxor IN1, STATE1
+	pxor IN2, STATE2
+	pxor IN3, STATE3
+	pxor IN4, STATE4
+	pxor IN5, STATE5
+	pxor IN6, STATE6
+	pxor IN7, STATE7
+	pxor IN8, STATE8
+
+	movdqu STATE1, 0x0(OUTP)
+	movdqu STATE2, 0x10(OUTP)
+	movdqu STATE3, 0x20(OUTP)
+	movdqu STATE4, 0x30(OUTP)
+	movdqu STATE5, 0x40(OUTP)
+	movdqu STATE6, 0x50(OUTP)
+	movdqu STATE7, 0x60(OUTP)
+	movdqu STATE8, 0x70(OUTP)
+
+	sub $128, LEN
+	add $128, INP
+	add $128, OUTP
+	cmp $128, LEN
+	jge .Lcbc_dec8
+	cmp $16, LEN
+	jb .Lcbc_dec_noerr
+#endif
+
+.align 4
+.Lcbc_dec1_pre:
+	movdqu (IVP), STATE3
+.Lcbc_dec1:
+	movdqu (INP), STATE2
+	movdqa STATE2, STATE1
+
+	cmp $16, KLEN
+	je .Lcbc_dec1_128
+	aesdec256kl (HANDLEP), STATE1
+	jz .Lcbc_dec_err
+	jmp .Lcbc_dec1_end
+.Lcbc_dec1_128:
+	aesdec128kl (HANDLEP), STATE1
+	jz .Lcbc_dec_err
+
+.Lcbc_dec1_end:
+	pxor STATE3, STATE1
+	movdqu STATE1, (OUTP)
+	movdqa STATE2, STATE3
+	sub $16, LEN
+	add $16, INP
+	add $16, OUTP
+	cmp $16, LEN
+	jge .Lcbc_dec1
+	movdqu STATE3, (IVP)
+
+.Lcbc_dec_noerr:
+	xor AREG, AREG
+	jmp .Lcbc_dec_end
+.Lcbc_dec_err:
+	mov $1, AREG
+.Lcbc_dec_end:
+#ifndef __x86_64__
+	popl KLEN
+	popl HANDLEP
+	popl LEN
+	popl IVP
+#endif
+	FRAME_END
+	ret
+SYM_FUNC_END(_aeskl_cbc_dec)
+
diff --git a/arch/x86/crypto/aeskl-intel_glue.c b/arch/x86/crypto/aeskl-intel_glue.c
index 0062baaaf7b2..22b74d13150c 100644
--- a/arch/x86/crypto/aeskl-intel_glue.c
+++ b/arch/x86/crypto/aeskl-intel_glue.c
@@ -27,8 +27,13 @@  asmlinkage int aeskl_setkey(struct crypto_aes_ctx *ctx, const u8 *in_key, unsign
 asmlinkage int _aeskl_enc(const void *ctx, u8 *out, const u8 *in);
 asmlinkage int _aeskl_dec(const void *ctx, u8 *out, const u8 *in);
 
-static int __maybe_unused aeskl_setkey_common(struct crypto_tfm *tfm, void *raw_ctx,
-					      const u8 *in_key, unsigned int key_len)
+asmlinkage int _aeskl_cbc_enc(struct crypto_aes_ctx *ctx, u8 *out, const u8 *in, unsigned int len,
+			      u8 *iv);
+asmlinkage int _aeskl_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out, const u8 *in, unsigned int len,
+			      u8 *iv);
+
+static int aeskl_setkey_common(struct crypto_tfm *tfm, void *raw_ctx, const u8 *in_key,
+			       unsigned int key_len)
 {
 	struct crypto_aes_ctx *ctx = aes_ctx(raw_ctx);
 	int err;
@@ -86,11 +91,95 @@  static inline int aeskl_dec(const void *ctx, u8 *out, const u8 *in)
 		return 0;
 }
 
+static int aeskl_cbc_enc(struct crypto_aes_ctx *ctx, u8 *out, const u8 *in, unsigned int len,
+			 u8 *iv)
+{
+	if (unlikely(ctx->key_length == AES_KEYSIZE_192))
+		return -EINVAL;
+	else if (!valid_keylocker())
+		return -ENODEV;
+	else if (_aeskl_cbc_enc(ctx, out, in, len, iv))
+		return -EINVAL;
+	else
+		return 0;
+}
+
+static int aeskl_cbc_dec(struct crypto_aes_ctx *ctx, u8 *out, const u8 *in, unsigned int len,
+			 u8 *iv)
+{
+	if (unlikely(ctx->key_length == AES_KEYSIZE_192))
+		return -EINVAL;
+	else if (!valid_keylocker())
+		return -ENODEV;
+	else if (_aeskl_cbc_dec(ctx, out, in, len, iv))
+		return -EINVAL;
+	else
+		return 0;
+}
+
+static int aeskl_skcipher_setkey(struct crypto_skcipher *tfm, const u8 *key,
+				 unsigned int len)
+{
+	struct crypto_tfm *crypto_tfm = crypto_skcipher_tfm(tfm);
+	void *raw_ctx = crypto_skcipher_ctx(tfm);
+
+	return aeskl_setkey_common(crypto_tfm, raw_ctx, key, len);
+}
+
+static int cbc_encrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+
+	if (likely(keylength(crypto_skcipher_ctx(tfm)) != AES_KEYSIZE_192))
+		return cbc_crypt_common(req, aeskl_cbc_enc);
+	else
+		return cbc_crypt_common(req, aesni_cbc_enc);
+}
+
+static int cbc_decrypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+
+	if (likely(keylength(crypto_skcipher_ctx(tfm)) != AES_KEYSIZE_192))
+		return cbc_crypt_common(req, aeskl_cbc_dec);
+	else
+		return cbc_crypt_common(req, aesni_cbc_dec);
+}
+
+static struct skcipher_alg aeskl_skciphers[] = {
+	{
+		.base = {
+			.cra_name		= "__cbc(aes)",
+			.cra_driver_name	= "__cbc-aes-aeskl",
+			.cra_priority		= 200,
+			.cra_flags		= CRYPTO_ALG_INTERNAL,
+			.cra_blocksize		= AES_BLOCK_SIZE,
+			.cra_ctxsize		= CRYPTO_AES_CTX_SIZE,
+			.cra_module		= THIS_MODULE,
+		},
+		.min_keysize	= AES_MIN_KEY_SIZE,
+		.max_keysize	= AES_MAX_KEY_SIZE,
+		.ivsize		= AES_BLOCK_SIZE,
+		.setkey		= aeskl_skcipher_setkey,
+		.encrypt	= cbc_encrypt,
+		.decrypt	= cbc_decrypt,
+	}
+};
+
+static struct simd_skcipher_alg *aeskl_simd_skciphers[ARRAY_SIZE(aeskl_skciphers)];
+
 static int __init aeskl_init(void)
 {
+	u32 eax, ebx, ecx, edx;
+	int err;
+
 	if (!valid_keylocker())
 		return -ENODEV;
 
+	cpuid_count(KEYLOCKER_CPUID, 0, &eax, &ebx, &ecx, &edx);
+	if (!(ebx & KEYLOCKER_CPUID_EBX_WIDE))
+		return -ENODEV;
+
 	/*
 	 * AES-KL itself does not depend on AES-NI. But AES-KL does not
 	 * support 192-bit keys. To make itself AES-compliant, it falls
@@ -99,12 +188,18 @@  static int __init aeskl_init(void)
 	if (!boot_cpu_has(X86_FEATURE_AES))
 		return -ENODEV;
 
+	err = simd_register_skciphers_compat(aeskl_skciphers, ARRAY_SIZE(aeskl_skciphers),
+					     aeskl_simd_skciphers);
+	if (err)
+		return err;
+
 	return 0;
 }
 
 static void __exit aeskl_exit(void)
 {
-	return;
+	simd_unregister_skciphers(aeskl_skciphers, ARRAY_SIZE(aeskl_skciphers),
+				  aeskl_simd_skciphers);
 }
 
 late_initcall(aeskl_init);