Use the fact that the AES-NI instructions can be pipelined to improve

performance... Use SSE2 instructions for calculating the XTS tweek
factor...  Let the compiler do more work and handle register allocation
by using intrinsics, now only the key schedule is in assembly...

Replace .byte hard coded instructions w/ the proper instructions now
that both clang and gcc support them...

On my machine, pulling the code to userland I saw performance go from
~150MB/sec to 2GB/sec in XTS mode.  GELI on GNOP saw a more modest
increase of about 3x due to other system overhead (geom and
opencrypto)...

These changes allow almost full disk io rate w/ geli...

Reviewed by:	-current, -security
Thanks to:	Mike Hamburg for the XTS tweek algorithm

7 files changed, 359 insertions, 422 deletions

diff --git a/sys/crypto/aesni/aesencdec.h b/sys/crypto/aesni/aesencdec.h
new file mode 100644
index 0000000..0c9bf5f
--- /dev/null
+++ b/sys/crypto/aesni/aesencdec.h
@@ -0,0 +1,136 @@
+/*-
+ * Copyright 2013 John-Mark Gurney <jmg@FreeBSD.org>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF 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.
+ *
+ * $FreeBSD$
+ *
+ */
+
+#include <wmmintrin.h>
+
+static inline void
+aesni_enc8(int rounds, const uint8_t *key_schedule, __m128i a,
+    __m128i b, __m128i c, __m128i d, __m128i e, __m128i f, __m128i g,
+    __m128i h, __m128i out[8])
+{
+	const __m128i *keysched = (const __m128i *)key_schedule;
+	int i;
+
+	a ^= keysched[0];
+	b ^= keysched[0];
+	c ^= keysched[0];
+	d ^= keysched[0];
+	e ^= keysched[0];
+	f ^= keysched[0];
+	g ^= keysched[0];
+	h ^= keysched[0];
+
+	for (i = 0; i < rounds; i++) {
+		a = _mm_aesenc_si128(a, keysched[i + 1]);
+		b = _mm_aesenc_si128(b, keysched[i + 1]);
+		c = _mm_aesenc_si128(c, keysched[i + 1]);
+		d = _mm_aesenc_si128(d, keysched[i + 1]);
+		e = _mm_aesenc_si128(e, keysched[i + 1]);
+		f = _mm_aesenc_si128(f, keysched[i + 1]);
+		g = _mm_aesenc_si128(g, keysched[i + 1]);
+		h = _mm_aesenc_si128(h, keysched[i + 1]);
+	}
+
+	out[0] = _mm_aesenclast_si128(a, keysched[i + 1]);
+	out[1] = _mm_aesenclast_si128(b, keysched[i + 1]);
+	out[2] = _mm_aesenclast_si128(c, keysched[i + 1]);
+	out[3] = _mm_aesenclast_si128(d, keysched[i + 1]);
+	out[4] = _mm_aesenclast_si128(e, keysched[i + 1]);
+	out[5] = _mm_aesenclast_si128(f, keysched[i + 1]);
+	out[6] = _mm_aesenclast_si128(g, keysched[i + 1]);
+	out[7] = _mm_aesenclast_si128(h, keysched[i + 1]);
+}
+
+static inline void
+aesni_dec8(int rounds, const uint8_t *key_schedule, __m128i a,
+    __m128i b, __m128i c, __m128i d, __m128i e, __m128i f, __m128i g,
+    __m128i h, __m128i out[8])
+{
+	const __m128i *keysched = (const __m128i *)key_schedule;
+	int i;
+
+	a ^= keysched[0];
+	b ^= keysched[0];
+	c ^= keysched[0];
+	d ^= keysched[0];
+	e ^= keysched[0];
+	f ^= keysched[0];
+	g ^= keysched[0];
+	h ^= keysched[0];
+
+	for (i = 0; i < rounds; i++) {
+		a = _mm_aesdec_si128(a, keysched[i + 1]);
+		b = _mm_aesdec_si128(b, keysched[i + 1]);
+		c = _mm_aesdec_si128(c, keysched[i + 1]);
+		d = _mm_aesdec_si128(d, keysched[i + 1]);
+		e = _mm_aesdec_si128(e, keysched[i + 1]);
+		f = _mm_aesdec_si128(f, keysched[i + 1]);
+		g = _mm_aesdec_si128(g, keysched[i + 1]);
+		h = _mm_aesdec_si128(h, keysched[i + 1]);
+	}
+
+	out[0] = _mm_aesdeclast_si128(a, keysched[i + 1]);
+	out[1] = _mm_aesdeclast_si128(b, keysched[i + 1]);
+	out[2] = _mm_aesdeclast_si128(c, keysched[i + 1]);
+	out[3] = _mm_aesdeclast_si128(d, keysched[i + 1]);
+	out[4] = _mm_aesdeclast_si128(e, keysched[i + 1]);
+	out[5] = _mm_aesdeclast_si128(f, keysched[i + 1]);
+	out[6] = _mm_aesdeclast_si128(g, keysched[i + 1]);
+	out[7] = _mm_aesdeclast_si128(h, keysched[i + 1]);
+}
+
+static inline __m128i
+aesni_enc(int rounds, const uint8_t *key_schedule, const __m128i from)
+{
+	__m128i tmp;
+	const __m128i *keysched = (const __m128i *)key_schedule;
+	int i;
+
+	tmp = from ^ keysched[0];
+
+	for (i = 0; i < rounds; i++)
+		tmp = _mm_aesenc_si128(tmp, keysched[i + 1]);
+
+	return _mm_aesenclast_si128(tmp, keysched[i + 1]);
+}
+
+static inline __m128i
+aesni_dec(int rounds, const uint8_t *key_schedule, const __m128i from)
+{
+	__m128i tmp;
+	const __m128i *keysched = (const __m128i *)key_schedule;
+	int i;
+
+	tmp = from ^ keysched[0];
+
+	for (i = 0; i < rounds; i++)
+		tmp = _mm_aesdec_si128(tmp, keysched[i + 1]);
+
+	return _mm_aesdeclast_si128(tmp, keysched[i + 1]);
+}
diff --git a/sys/crypto/aesni/aesencdec_amd64.S b/sys/crypto/aesni/aesencdec_amd64.S
deleted file mode 100644
index f77918b..0000000
--- a/sys/crypto/aesni/aesencdec_amd64.S
+++ /dev/null
@@ -1,135 +0,0 @@
-/*-
- * Copyright (c) 2010 Konstantin Belousov <kib@FreeBSD.org>
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
- * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
- * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
- * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
- * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
- * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
- * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
- * HOWEVER CAUSED AND ON ANY THEORY OF 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.
- */
-
-#include <machine/asmacros.h>
-
-	.text
-
-ENTRY(aesni_enc)
-	.cfi_startproc
-	movdqu	(%rdx),%xmm0
-	cmpq	$0,%r8
-	je	1f
-	movdqu	(%r8),%xmm1	/* unaligned load into reg */
-	pxor	%xmm1,%xmm0	/* pxor otherwise can fault on iv */
-1:
-	pxor	(%rsi),%xmm0
-2:
-	addq	$0x10,%rsi
-//	aesenc	(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xdc,0x06
-	decl	%edi
-	jne	2b
-	addq	$0x10,%rsi
-//	aesenclast (%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xdd,0x06
-	movdqu	%xmm0,(%rcx)
-	retq
-	.cfi_endproc
-END(aesni_enc)
-
-ENTRY(aesni_dec)
-	.cfi_startproc
-	movdqu	(%rdx),%xmm0
-	pxor	(%rsi),%xmm0
-1:
-	addq	$0x10,%rsi
-//	aesdec	(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x06
-	decl	%edi
-	jne	1b
-	addq	$0x10,%rsi
-//	aesdeclast (%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xdf,0x06
-	cmpq	$0,%r8
-	je	2f
-	movdqu	(%r8),%xmm1
-	pxor	%xmm1,%xmm0
-2:
-	movdqu	%xmm0,(%rcx)
-	retq
-	.cfi_endproc
-END(aesni_dec)
-
-ENTRY(aesni_decrypt_cbc)
-	.cfi_startproc
-	shrq	$4,%rdx
-	movdqu	(%r8),%xmm1
-1:
-	movdqu	(%rcx),%xmm0
-	movdqa	%xmm0,%xmm2
-	pxor	(%rsi),%xmm0
-	cmpl	$12,%edi
-//	aesdec	0x10(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x46,0x10
-//	aesdec	0x20(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x46,0x20
-//	aesdec	0x30(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x46,0x30
-//	aesdec	0x40(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x46,0x40
-//	aesdec	0x50(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x46,0x50
-//	aesdec	0x60(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x46,0x60
-//	aesdec	0x70(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x46,0x70
-//	aesdec	0x80(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x86,0x80,0x00,0x00,0x00
-//	aesdec	0x90(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x86,0x90,0x00,0x00,0x00
-	jge	2f
-//	aesdeclast 0xa0(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xdf,0x86,0xa0,0x00,0x00,0x00
-	jmp	4f
-2:
-//	aesdec	0xa0(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x86,0xa0,0x00,0x00,0x00
-//	aesdec	0xb0(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x86,0xb0,0x00,0x00,0x00
-	jg	3f
-//	aesdeclast 0xc0(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xdf,0x86,0xc0,0x00,0x00,0x00
-	jmp	4f
-3:
-//	aesdec	0xc0(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x86,0xc0,0x00,0x00,0x00
-//	aesdec	0xd0(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x86,0xd0,0x00,0x00,0x00
-//	aesdeclast 0xe0(%rsi),%xmm0
-	.byte	0x66,0x0f,0x38,0xdf,0x86,0xe0,0x00,0x00,0x00
-4:
-	pxor	%xmm1,%xmm0
-	movdqu	%xmm0,(%rcx)
-	movdqa	%xmm2,%xmm1	// iv
-	addq	$0x10,%rcx
-	decq	%rdx
-	jne	1b
-	retq
-	.cfi_endproc
-END(aesni_decrypt_cbc)
-
-	.ident	"$FreeBSD$"
diff --git a/sys/crypto/aesni/aesencdec_i386.S b/sys/crypto/aesni/aesencdec_i386.S
deleted file mode 100644
index 78de311..0000000
--- a/sys/crypto/aesni/aesencdec_i386.S
+++ /dev/null
@@ -1,166 +0,0 @@
-/*-
- * Copyright (c) 2010 Konstantin Belousov <kib@FreeBSD.org>
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- *
- * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
- * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
- * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
- * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
- * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
- * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
- * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
- * HOWEVER CAUSED AND ON ANY THEORY OF 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.
- */
-
-#include <machine/asmacros.h>
-
-ENTRY(aesni_enc)
-	.cfi_startproc
-	pushl	%ebp
-	.cfi_adjust_cfa_offset 4
-	movl	%esp,%ebp
-	movl	8(%ebp),%ecx	/* rounds */
-	movl	16(%ebp),%edx
-	movdqu	(%edx),%xmm0	/* from */
-	movl	24(%ebp),%eax	/* iv */
-	cmpl	$0,%eax
-	je	1f
-	movdqu	(%eax),%xmm1
-	pxor	%xmm1,%xmm0
-1:
-	movl	12(%ebp),%eax	/* key */
-	pxor	(%eax),%xmm0
-2:
-	addl	$0x10,%eax
-//	aesenc	(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xdc,0x00
-	loopne	2b
-	addl	$0x10,%eax
-//	aesenclast (%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xdd,0x00
-	movl	20(%ebp),%eax
-	movdqu	%xmm0,(%eax)	/* to */
-	leave
-	.cfi_adjust_cfa_offset -4
-	retl
-	.cfi_endproc
-END(aesni_enc)
-
-ENTRY(aesni_dec)
-	.cfi_startproc
-	pushl	%ebp
-	.cfi_adjust_cfa_offset 4
-	movl	%esp,%ebp
-	movl	8(%ebp),%ecx	/* rounds */
-	movl	16(%ebp),%edx
-	movdqu	(%edx),%xmm0	/* from */
-	movl	12(%ebp),%eax	/* key */
-	pxor	(%eax),%xmm0
-1:
-	addl	$0x10,%eax
-//	aesdec	(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x00
-	loopne	1b
-	addl	$0x10,%eax
-//	aesdeclast (%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xdf,0x00
-	movl	24(%ebp),%eax
-	cmpl	$0,%eax		/* iv */
-	je	2f
-	movdqu	(%eax),%xmm1
-	pxor	%xmm1,%xmm0
-2:
-	movl	20(%ebp),%eax
-	movdqu	%xmm0,(%eax)	/* to */
-	leave
-	.cfi_adjust_cfa_offset -4
-	retl
-	.cfi_endproc
-END(aesni_dec)
-
-ENTRY(aesni_decrypt_cbc)
-	.cfi_startproc
-	pushl	%ebp
-	.cfi_adjust_cfa_offset 4
-	movl	%esp,%ebp
-	pushl	%ebx
-	pushl	%esi
-	movl	12(%ebp),%eax	/* key */
-	movl	16(%ebp),%ecx	/* length */
-	shrl	$4,%ecx
-	movl	20(%ebp),%ebx	/* buf */
-	movl	24(%ebp),%esi
-	movdqu	(%esi),%xmm1	/* iv */
-	movl	8(%ebp),%esi	/* rounds */
-1:
-	movdqu	(%ebx),%xmm0
-	movdqa	%xmm0,%xmm2
-	pxor	(%eax),%xmm0
-	cmpl	$12,%esi
-//	aesdec	0x10(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x40,0x10
-//	aesdec	0x20(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x40,0x20
-//	aesdec	0x30(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x40,0x30
-//	aesdec	0x40(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x40,0x40
-//	aesdec	0x50(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x40,0x50
-//	aesdec	0x60(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x40,0x60
-//	aesdec	0x70(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x40,0x70
-//	aesdec	0x80(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x80,0x80,0x00,0x00,0x00
-//	aesdec	0x90(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x80,0x90,0x00,0x00,0x00
-	jge	2f
-//	aesdeclast 0xa0(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xdf,0x80,0xa0,0x00,0x00,0x00
-	jmp	4f
-2:
-//	aesdec	0xa0(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x80,0xa0,0x00,0x00,0x00
-//	aesdec	0xb0(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x80,0xb0,0x00,0x00,0x00
-	jg	3f
-//	aesdeclast 0xc0(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xdf,0x80,0xc0,0x00,0x00,0x00
-	jmp	4f
-3:
-//	aesdec	0xc0(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x80,0xc0,0x00,0x00,0x00
-//	aesdec	0xd0(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xde,0x80,0xd0,0x00,0x00,0x00
-//	aesdeclast 0xe0(%eax),%xmm0
-	.byte	0x66,0x0f,0x38,0xdf,0x80,0xe0,0x00,0x00,0x00
-4:
-	pxor	%xmm1,%xmm0
-	movdqu	%xmm0,(%ebx)
-	movdqa	%xmm2,%xmm1
-	addl	$0x10,%ebx
-	decl	%ecx
-	jne	1b
-
-	popl	%esi
-	popl	%ebx
-	leave
-	.cfi_adjust_cfa_offset -4
-	retl
-	.cfi_endproc
-END(aesni_decrypt_cbc)
-
-	.ident	"$FreeBSD$"
diff --git a/sys/crypto/aesni/aeskeys_amd64.S b/sys/crypto/aesni/aeskeys_amd64.S
index 23a4d3d..9b3e98c 100644
--- a/sys/crypto/aesni/aeskeys_amd64.S
+++ b/sys/crypto/aesni/aeskeys_amd64.S
@@ -125,103 +125,72 @@ ENTRY(aesni_set_enckey)
 	movups	0x10(%rdi),%xmm2	# other user key
 	movaps	%xmm2,(%rsi)
 	addq	$0x10,%rsi
-//	aeskeygenassist $0x1,%xmm2,%xmm1	# round 1
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x01
+	aeskeygenassist $0x1,%xmm2,%xmm1	# round 1
 	call	_key_expansion_256a
-//	aeskeygenassist $0x1,%xmm0,%xmm1
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x01
+	aeskeygenassist $0x1,%xmm0,%xmm1
 	call	_key_expansion_256b
-//	aeskeygenassist $0x2,%xmm2,%xmm1	# round 2
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x02
+	aeskeygenassist $0x2,%xmm2,%xmm1	# round 2
 	call	_key_expansion_256a
-//	aeskeygenassist $0x2,%xmm0,%xmm1
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x02
+	aeskeygenassist $0x2,%xmm0,%xmm1
 	call	_key_expansion_256b
-//	aeskeygenassist $0x4,%xmm2,%xmm1	# round 3
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x04
+	aeskeygenassist $0x4,%xmm2,%xmm1	# round 3
 	call	_key_expansion_256a
-//	aeskeygenassist $0x4,%xmm0,%xmm1
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x04
+	aeskeygenassist $0x4,%xmm0,%xmm1
 	call	_key_expansion_256b
-//	aeskeygenassist $0x8,%xmm2,%xmm1	# round 4
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x08
+	aeskeygenassist $0x8,%xmm2,%xmm1	# round 4
 	call	_key_expansion_256a
-//	aeskeygenassist $0x8,%xmm0,%xmm1
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x08
+	aeskeygenassist $0x8,%xmm0,%xmm1
 	call	_key_expansion_256b
-//	aeskeygenassist $0x10,%xmm2,%xmm1	# round 5
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x10
+	aeskeygenassist $0x10,%xmm2,%xmm1	# round 5
 	call	_key_expansion_256a
-//	aeskeygenassist $0x10,%xmm0,%xmm1
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x10
+	aeskeygenassist $0x10,%xmm0,%xmm1
 	call	_key_expansion_256b
-//	aeskeygenassist $0x20,%xmm2,%xmm1	# round 6
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x20
+	aeskeygenassist $0x20,%xmm2,%xmm1	# round 6
 	call	_key_expansion_256a
-//	aeskeygenassist $0x20,%xmm0,%xmm1
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x20
+	aeskeygenassist $0x20,%xmm0,%xmm1
 	call	_key_expansion_256b
-//	aeskeygenassist $0x40,%xmm2,%xmm1	# round 7
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x40
+	aeskeygenassist $0x40,%xmm2,%xmm1	# round 7
 	call	_key_expansion_256a
 	retq
 .Lenc_key192:
 	movq	0x10(%rdi),%xmm2		# other user key
-//	aeskeygenassist $0x1,%xmm2,%xmm1	# round 1
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x01
+	aeskeygenassist $0x1,%xmm2,%xmm1	# round 1
 	call	_key_expansion_192a
-//	aeskeygenassist $0x2,%xmm2,%xmm1	# round 2
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x02
+	aeskeygenassist $0x2,%xmm2,%xmm1	# round 2
 	call	_key_expansion_192b
-//	aeskeygenassist $0x4,%xmm2,%xmm1	# round 3
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x04
+	aeskeygenassist $0x4,%xmm2,%xmm1	# round 3
 	call	_key_expansion_192a
-//	aeskeygenassist $0x8,%xmm2,%xmm1	# round 4
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x08
+	aeskeygenassist $0x8,%xmm2,%xmm1	# round 4
 	call	_key_expansion_192b
-//	aeskeygenassist $0x10,%xmm2,%xmm1	# round 5
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x10
+	aeskeygenassist $0x10,%xmm2,%xmm1	# round 5
 	call	_key_expansion_192a
-//	aeskeygenassist $0x20,%xmm2,%xmm1	# round 6
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x20
+	aeskeygenassist $0x20,%xmm2,%xmm1	# round 6
 	call	_key_expansion_192b
-//	aeskeygenassist $0x40,%xmm2,%xmm1	# round 7
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x40
+	aeskeygenassist $0x40,%xmm2,%xmm1	# round 7
 	call	_key_expansion_192a
-//	aeskeygenassist $0x80,%xmm2,%xmm1	# round 8
-	.byte	0x66,0x0f,0x3a,0xdf,0xca,0x80
+	aeskeygenassist $0x80,%xmm2,%xmm1	# round 8
 	call	_key_expansion_192b
 	retq
 .Lenc_key128:
-//	aeskeygenassist $0x1,%xmm0,%xmm1	# round 1
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x01
+	aeskeygenassist $0x1,%xmm0,%xmm1	# round 1
 	call	_key_expansion_128
-//	aeskeygenassist $0x2,%xmm0,%xmm1	# round 2
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x02
+	aeskeygenassist $0x2,%xmm0,%xmm1	# round 2
 	call	_key_expansion_128
-//	aeskeygenassist $0x4,%xmm0,%xmm1	# round 3
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x04
+	aeskeygenassist $0x4,%xmm0,%xmm1	# round 3
 	call	_key_expansion_128
-//	aeskeygenassist $0x8,%xmm0,%xmm1	# round 4
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x08
+	aeskeygenassist $0x8,%xmm0,%xmm1	# round 4
 	call	_key_expansion_128
-//	aeskeygenassist $0x10,%xmm0,%xmm1	# round 5
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x10
+	aeskeygenassist $0x10,%xmm0,%xmm1	# round 5
 	call	_key_expansion_128
-//	aeskeygenassist $0x20,%xmm0,%xmm1	# round 6
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x20
+	aeskeygenassist $0x20,%xmm0,%xmm1	# round 6
 	call	_key_expansion_128
-//	aeskeygenassist $0x40,%xmm0,%xmm1	# round 7
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x40
+	aeskeygenassist $0x40,%xmm0,%xmm1	# round 7
 	call	_key_expansion_128
-//	aeskeygenassist $0x80,%xmm0,%xmm1	# round 8
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x80
+	aeskeygenassist $0x80,%xmm0,%xmm1	# round 8
 	call	_key_expansion_128
-//	aeskeygenassist $0x1b,%xmm0,%xmm1	# round 9
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x1b
+	aeskeygenassist $0x1b,%xmm0,%xmm1	# round 9
 	call	_key_expansion_128
-//	aeskeygenassist $0x36,%xmm0,%xmm1	# round 10
-	.byte	0x66,0x0f,0x3a,0xdf,0xc8,0x36
+	aeskeygenassist $0x36,%xmm0,%xmm1	# round 10
 	call	_key_expansion_128
 	retq
 	.cfi_endproc
@@ -238,8 +207,7 @@ ENTRY(aesni_set_deckey)
 1:
 	addq	$0x10,%rsi
 	subq	$0x10,%rdi
-//	aesimc	(%rdi),%xmm1
-	.byte	0x66,0x0f,0x38,0xdb,0x0f
+	aesimc	(%rdi),%xmm1
 	movdqa	%xmm1,(%rsi)
 	decl	%edx
 	jne	1b
diff --git a/sys/crypto/aesni/aesni.c b/sys/crypto/aesni/aesni.c
index ca00a57..73eb28a 100644
--- a/sys/crypto/aesni/aesni.c
+++ b/sys/crypto/aesni/aesni.c
@@ -40,7 +40,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/bus.h>
 #include <sys/uio.h>
 #include <crypto/aesni/aesni.h>
-#include "cryptodev_if.h"
+#include <cryptodev_if.h>
 
 struct aesni_softc {
 	int32_t cid;
@@ -74,6 +74,12 @@ aesni_probe(device_t dev)
 		device_printf(dev, "No AESNI support.\n");
 		return (EINVAL);
 	}
+
+	if ((cpu_feature & CPUID_SSE2) == 0) {
+		device_printf(dev, "No SSE2 support but AESNI!?!\n");
+		return (EINVAL);
+	}
+
 	device_set_desc_copy(dev, "AES-CBC,AES-XTS");
 	return (0);
 }
diff --git a/sys/crypto/aesni/aesni.h b/sys/crypto/aesni/aesni.h
index 78255b7..17ca9c5 100644
--- a/sys/crypto/aesni/aesni.h
+++ b/sys/crypto/aesni/aesni.h
@@ -71,12 +71,6 @@ struct aesni_session {
 /*
  * Internal functions, implemented in assembler.
  */
-void aesni_enc(int rounds, const uint8_t *key_schedule,
-    const uint8_t from[AES_BLOCK_LEN], uint8_t to[AES_BLOCK_LEN],
-    const uint8_t iv[AES_BLOCK_LEN]);
-void aesni_dec(int rounds, const uint8_t *key_schedule,
-    const uint8_t from[AES_BLOCK_LEN], uint8_t to[AES_BLOCK_LEN],
-    const uint8_t iv[AES_BLOCK_LEN]);
 void aesni_set_enckey(const uint8_t *userkey, uint8_t *encrypt_schedule,
     int number_of_rounds);
 void aesni_set_deckey(const uint8_t *encrypt_schedule,
@@ -88,12 +82,19 @@ void aesni_set_deckey(const uint8_t *encrypt_schedule,
 void aesni_encrypt_cbc(int rounds, const void *key_schedule, size_t len,
     const uint8_t *from, uint8_t *to, const uint8_t iv[AES_BLOCK_LEN]);
 void aesni_decrypt_cbc(int rounds, const void *key_schedule, size_t len,
-    const uint8_t *from, const uint8_t iv[AES_BLOCK_LEN]);
+    uint8_t *buf, const uint8_t iv[AES_BLOCK_LEN]);
 void aesni_encrypt_ecb(int rounds, const void *key_schedule, size_t len,
     const uint8_t from[AES_BLOCK_LEN], uint8_t to[AES_BLOCK_LEN]);
 void aesni_decrypt_ecb(int rounds, const void *key_schedule, size_t len,
     const uint8_t from[AES_BLOCK_LEN], uint8_t to[AES_BLOCK_LEN]);
 
+void aesni_encrypt_xts(int rounds, const void *data_schedule,
+    const void *tweak_schedule, size_t len, const uint8_t *from, uint8_t *to,
+    const uint8_t iv[AES_BLOCK_LEN]);
+void aesni_decrypt_xts(int rounds, const void *data_schedule,
+    const void *tweak_schedule, size_t len, const uint8_t *from, uint8_t *to,
+    const uint8_t iv[AES_BLOCK_LEN]);
+
 int aesni_cipher_setup(struct aesni_session *ses,
     struct cryptoini *encini);
 int aesni_cipher_process(struct aesni_session *ses,
diff --git a/sys/crypto/aesni/aesni_wrap.c b/sys/crypto/aesni/aesni_wrap.c
index 3340b1f..197baf7 100644
--- a/sys/crypto/aesni/aesni_wrap.c
+++ b/sys/crypto/aesni/aesni_wrap.c
@@ -2,6 +2,7 @@
  * Copyright (C) 2008 Damien Miller <djm@mindrot.org>
  * Copyright (c) 2010 Konstantin Belousov <kib@FreeBSD.org>
  * Copyright (c) 2010-2011 Pawel Jakub Dawidek <pawel@dawidek.net>
+ * Copyright 2012-2013 John-Mark Gurney <jmg@FreeBSD.org>
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
@@ -28,13 +29,15 @@
 
 #include <sys/cdefs.h>
 __FBSDID("$FreeBSD$");
-
+ 
 #include <sys/param.h>
 #include <sys/libkern.h>
 #include <sys/malloc.h>
 #include <sys/proc.h>
 #include <sys/systm.h>
 #include <crypto/aesni/aesni.h>
+ 
+#include "aesencdec.h"
 
 MALLOC_DECLARE(M_AESNI);
 
@@ -42,28 +45,78 @@ void
 aesni_encrypt_cbc(int rounds, const void *key_schedule, size_t len,
     const uint8_t *from, uint8_t *to, const uint8_t iv[AES_BLOCK_LEN])
 {
-	const uint8_t *ivp;
+	__m128i tot, ivreg;
 	size_t i;
 
 	len /= AES_BLOCK_LEN;
-	ivp = iv;
+	ivreg = _mm_loadu_si128((const __m128i *)iv);
 	for (i = 0; i < len; i++) {
-		aesni_enc(rounds - 1, key_schedule, from, to, ivp);
-		ivp = to;
+		tot = aesni_enc(rounds - 1, key_schedule,
+		    _mm_loadu_si128((const __m128i *)from) ^ ivreg);
+		ivreg = tot;
+		_mm_storeu_si128((__m128i *)to, tot);
 		from += AES_BLOCK_LEN;
 		to += AES_BLOCK_LEN;
 	}
 }
 
 void
-aesni_encrypt_ecb(int rounds, const void *key_schedule, size_t len,
-    const uint8_t from[AES_BLOCK_LEN], uint8_t to[AES_BLOCK_LEN])
+aesni_decrypt_cbc(int rounds, const void *key_schedule, size_t len,
+    uint8_t *buf, const uint8_t iv[AES_BLOCK_LEN])
 {
-	size_t i;
+	__m128i blocks[8];
+	__m128i *bufs;
+	__m128i ivreg, nextiv;
+	size_t i, j, cnt;
+
+	ivreg = _mm_loadu_si128((const __m128i *)iv);
+	cnt = len / AES_BLOCK_LEN / 8;
+	for (i = 0; i < cnt; i++) {
+		bufs = (__m128i *)buf;
+		aesni_dec8(rounds - 1, key_schedule, bufs[0], bufs[1],
+		    bufs[2], bufs[3], bufs[4], bufs[5], bufs[6],
+		    bufs[7], &blocks[0]);
+		for (j = 0; j < 8; j++) {
+			nextiv = bufs[j];
+			bufs[j] = blocks[j] ^ ivreg;
+			ivreg = nextiv;
+		}
+		buf += AES_BLOCK_LEN * 8;
+	}
+	i *= 8;
+	cnt = len / AES_BLOCK_LEN;
+	for (; i < cnt; i++) {
+		bufs = (__m128i *)buf;
+		nextiv = bufs[0];
+		bufs[0] = aesni_dec(rounds - 1, key_schedule, bufs[0]) ^ ivreg;
+		ivreg = nextiv;
+		buf += AES_BLOCK_LEN;
+	}
+}
 
-	len /= AES_BLOCK_LEN;
-	for (i = 0; i < len; i++) {
-		aesni_enc(rounds - 1, key_schedule, from, to, NULL);
+void
+aesni_encrypt_ecb(int rounds, const void *key_schedule, size_t len,
+    const uint8_t *from, uint8_t *to)
+{
+	__m128i tot;
+	const __m128i *blocks;
+	size_t i, cnt;
+
+	cnt = len / AES_BLOCK_LEN / 8;
+	for (i = 0; i < cnt; i++) {
+		blocks = (const __m128i *)from;
+		aesni_enc8(rounds - 1, key_schedule, blocks[0], blocks[1],
+		    blocks[2], blocks[3], blocks[4], blocks[5], blocks[6],
+		    blocks[7], (__m128i *)to);
+		from += AES_BLOCK_LEN * 8;
+		to += AES_BLOCK_LEN * 8;
+	}
+	i *= 8;
+	cnt = len / AES_BLOCK_LEN;
+	for (; i < cnt; i++) {
+		tot = aesni_enc(rounds - 1, key_schedule,
+		    _mm_loadu_si128((const __m128i *)from));
+		_mm_storeu_si128((__m128i *)to, tot);
 		from += AES_BLOCK_LEN;
 		to += AES_BLOCK_LEN;
 	}
@@ -73,11 +126,25 @@ void
 aesni_decrypt_ecb(int rounds, const void *key_schedule, size_t len,
     const uint8_t from[AES_BLOCK_LEN], uint8_t to[AES_BLOCK_LEN])
 {
-	size_t i;
-
-	len /= AES_BLOCK_LEN;
-	for (i = 0; i < len; i++) {
-		aesni_dec(rounds - 1, key_schedule, from, to, NULL);
+	__m128i tot;
+	const __m128i *blocks;
+	size_t i, cnt;
+
+	cnt = len / AES_BLOCK_LEN / 8;
+	for (i = 0; i < cnt; i++) {
+		blocks = (const __m128i *)from;
+		aesni_dec8(rounds - 1, key_schedule, blocks[0], blocks[1],
+		    blocks[2], blocks[3], blocks[4], blocks[5], blocks[6],
+		    blocks[7], (__m128i *)to);
+		from += AES_BLOCK_LEN * 8;
+		to += AES_BLOCK_LEN * 8;
+	}
+	i *= 8;
+	cnt = len / AES_BLOCK_LEN;
+	for (; i < cnt; i++) {
+		tot = aesni_dec(rounds - 1, key_schedule,
+		    _mm_loadu_si128((const __m128i *)from));
+		_mm_storeu_si128((__m128i *)to, tot);
 		from += AES_BLOCK_LEN;
 		to += AES_BLOCK_LEN;
 	}
@@ -87,34 +154,88 @@ aesni_decrypt_ecb(int rounds, const void *key_schedule, size_t len,
 #define	AES_XTS_IVSIZE		8
 #define	AES_XTS_ALPHA		0x87	/* GF(2^128) generator polynomial */
 
+static inline __m128i
+xts_crank_lfsr(__m128i inp)
+{
+	const __m128i alphamask = _mm_set_epi32(1, 1, 1, AES_XTS_ALPHA);
+	__m128i xtweak, ret;
+
+	/* set up xor mask */
+	xtweak = _mm_shuffle_epi32(inp, 0x93);
+	xtweak = _mm_srai_epi32(xtweak, 31);
+	xtweak &= alphamask;
+
+	/* next term */
+	ret = _mm_slli_epi32(inp, 1);
+	ret ^= xtweak;
+
+	return ret;
+}
+
 static void
-aesni_crypt_xts_block(int rounds, const void *key_schedule, uint64_t *tweak,
-    const uint64_t *from, uint64_t *to, uint64_t *block, int do_encrypt)
+aesni_crypt_xts_block(int rounds, const void *key_schedule, __m128i *tweak,
+    const __m128i *from, __m128i *to, int do_encrypt)
 {
-	int carry;
+	__m128i block;
 
-	block[0] = from[0] ^ tweak[0];
-	block[1] = from[1] ^ tweak[1];
+	block = *from ^ *tweak;
 
 	if (do_encrypt)
-		aesni_enc(rounds - 1, key_schedule, (uint8_t *)block, (uint8_t *)to, NULL);
+		block = aesni_enc(rounds - 1, key_schedule, block);
 	else
-		aesni_dec(rounds - 1, key_schedule, (uint8_t *)block, (uint8_t *)to, NULL);
+		block = aesni_dec(rounds - 1, key_schedule, block);
 
-	to[0] ^= tweak[0];
-	to[1] ^= tweak[1];
+	*to = block ^ *tweak;
 
-	/* Exponentiate tweak. */
-	carry = ((tweak[0] & 0x8000000000000000ULL) > 0);
-	tweak[0] <<= 1;
-	if (tweak[1] & 0x8000000000000000ULL) {
-		uint8_t *twk = (uint8_t *)tweak;
+	*tweak = xts_crank_lfsr(*tweak);
+}
 
-		twk[0] ^= AES_XTS_ALPHA;
-	}
-	tweak[1] <<= 1;
-	if (carry)
-		tweak[1] |= 1;
+static void
+aesni_crypt_xts_block8(int rounds, const void *key_schedule, __m128i *tweak,
+    const __m128i *from, __m128i *to, int do_encrypt)
+{
+	__m128i tmptweak;
+	__m128i a, b, c, d, e, f, g, h;
+	__m128i tweaks[8];
+	__m128i tmp[8];
+
+	tmptweak = *tweak;
+
+	/*
+	 * unroll the loop.  This lets gcc put values directly in the
+	 * register and saves memory accesses.
+	 */
+#define PREPINP(v, pos) 					\
+		do {						\
+			tweaks[(pos)] = tmptweak;		\
+			(v) = from[(pos)] ^ tmptweak;		\
+			tmptweak = xts_crank_lfsr(tmptweak);	\
+		} while (0)
+	PREPINP(a, 0);
+	PREPINP(b, 1);
+	PREPINP(c, 2);
+	PREPINP(d, 3);
+	PREPINP(e, 4);
+	PREPINP(f, 5);
+	PREPINP(g, 6);
+	PREPINP(h, 7);
+	*tweak = tmptweak;
+
+	if (do_encrypt)
+		aesni_enc8(rounds - 1, key_schedule, a, b, c, d, e, f, g, h,
+		    tmp);
+	else
+		aesni_dec8(rounds - 1, key_schedule, a, b, c, d, e, f, g, h,
+		    tmp);
+
+	to[0] = tmp[0] ^ tweaks[0];
+	to[1] = tmp[1] ^ tweaks[1];
+	to[2] = tmp[2] ^ tweaks[2];
+	to[3] = tmp[3] ^ tweaks[3];
+	to[4] = tmp[4] ^ tweaks[4];
+	to[5] = tmp[5] ^ tweaks[5];
+	to[6] = tmp[6] ^ tweaks[6];
+	to[7] = tmp[7] ^ tweaks[7];
 }
 
 static void
@@ -122,9 +243,9 @@ aesni_crypt_xts(int rounds, const void *data_schedule,
     const void *tweak_schedule, size_t len, const uint8_t *from, uint8_t *to,
     const uint8_t iv[AES_BLOCK_LEN], int do_encrypt)
 {
-	uint64_t block[AES_XTS_BLOCKSIZE / 8];
-	uint8_t tweak[AES_XTS_BLOCKSIZE];
-	size_t i;
+	__m128i tweakreg;
+	uint8_t tweak[AES_XTS_BLOCKSIZE] __aligned(16);
+	size_t i, cnt;
 
 	/*
 	 * Prepare tweak as E_k2(IV). IV is specified as LE representation
@@ -137,21 +258,27 @@ aesni_crypt_xts(int rounds, const void *data_schedule,
 #else
 #error Only LITTLE_ENDIAN architectures are supported.
 #endif
-	aesni_enc(rounds - 1, tweak_schedule, tweak, tweak, NULL);
-
-	len /= AES_XTS_BLOCKSIZE;
-	for (i = 0; i < len; i++) {
-		aesni_crypt_xts_block(rounds, data_schedule, (uint64_t *)tweak,
-		    (const uint64_t *)from, (uint64_t *)to, block, do_encrypt);
+	tweakreg = _mm_loadu_si128((__m128i *)&tweak[0]);
+	tweakreg = aesni_enc(rounds - 1, tweak_schedule, tweakreg);
+
+	cnt = len / AES_XTS_BLOCKSIZE / 8;
+	for (i = 0; i < cnt; i++) {
+		aesni_crypt_xts_block8(rounds, data_schedule, &tweakreg,
+		    (const __m128i *)from, (__m128i *)to, do_encrypt);
+		from += AES_XTS_BLOCKSIZE * 8;
+		to += AES_XTS_BLOCKSIZE * 8;
+	}
+	i *= 8;
+	cnt = len / AES_XTS_BLOCKSIZE;
+	for (; i < cnt; i++) {
+		aesni_crypt_xts_block(rounds, data_schedule, &tweakreg,
+		    (const __m128i *)from, (__m128i *)to, do_encrypt);
 		from += AES_XTS_BLOCKSIZE;
 		to += AES_XTS_BLOCKSIZE;
 	}
-
-	bzero(tweak, sizeof(tweak));
-	bzero(block, sizeof(block));
 }
 
-static void
+void
 aesni_encrypt_xts(int rounds, const void *data_schedule,
     const void *tweak_schedule, size_t len, const uint8_t *from, uint8_t *to,
     const uint8_t iv[AES_BLOCK_LEN])
@@ -161,7 +288,7 @@ aesni_encrypt_xts(int rounds, const void *data_schedule,
 	    iv, 1);
 }
 
-static void
+void
 aesni_decrypt_xts(int rounds, const void *data_schedule,
     const void *tweak_schedule, size_t len, const uint8_t *from, uint8_t *to,
     const uint8_t iv[AES_BLOCK_LEN])