[CRYPTO] salsa20: Add x86-64 assembly version

This is the x86-64 version of the Salsa20 stream cipher algorithm. The
original assembly code came from
<http://cr.yp.to/snuffle/salsa20/amd64-3/salsa20.s>. It has been
reformatted for clarity.
Signed-off-by: NTan Swee Heng <thesweeheng@gmail.com>
Signed-off-by: NHerbert Xu <herbert@gondor.apana.org.au>

arch/x86/crypto/Makefile

@@ -8,6 +8,7 @@ obj-$(CONFIG_CRYPTO_SALSA20_586) += salsa20-i586.o
 
 obj-$(CONFIG_CRYPTO_AES_X86_64) += aes-x86_64.o
 obj-$(CONFIG_CRYPTO_TWOFISH_X86_64) += twofish-x86_64.o
+obj-$(CONFIG_CRYPTO_SALSA20_X86_64) += salsa20-x86_64.o
 
 aes-i586-y := aes-i586-asm_32.o aes_glue.o
 twofish-i586-y := twofish-i586-asm_32.o twofish_32.o
@@ -15,3 +16,4 @@ salsa20-i586-y := salsa20-i586-asm_32.o salsa20_glue.o
 
 aes-x86_64-y := aes-x86_64-asm_64.o aes_glue.o
 twofish-x86_64-y := twofish-x86_64-asm_64.o twofish_64.o
+salsa20-x86_64-y := salsa20-x86_64-asm_64.o salsa20_glue.o

arch/x86/crypto/salsa20-x86_64-asm_64.S

+# enter ECRYPT_encrypt_bytes
+.text
+.p2align 5
+.globl ECRYPT_encrypt_bytes
+ECRYPT_encrypt_bytes:
+	mov	%rsp,%r11
+	and	$31,%r11
+	add	$256,%r11
+	sub	%r11,%rsp
+	# x = arg1
+	mov	%rdi,%r8
+	# m = arg2
+	mov	%rsi,%rsi
+	# out = arg3
+	mov	%rdx,%rdi
+	# bytes = arg4
+	mov	%rcx,%rdx
+	#               unsigned>? bytes - 0
+	cmp	$0,%rdx
+	# comment:fp stack unchanged by jump
+	# goto done if !unsigned>
+	jbe	._done
+	# comment:fp stack unchanged by fallthrough
+# start:
+._start:
+	# r11_stack = r11
+	movq	%r11,0(%rsp)
+	# r12_stack = r12
+	movq	%r12,8(%rsp)
+	# r13_stack = r13
+	movq	%r13,16(%rsp)
+	# r14_stack = r14
+	movq	%r14,24(%rsp)
+	# r15_stack = r15
+	movq	%r15,32(%rsp)
+	# rbx_stack = rbx
+	movq	%rbx,40(%rsp)
+	# rbp_stack = rbp
+	movq	%rbp,48(%rsp)
+	# in0 = *(uint64 *) (x + 0)
+	movq	0(%r8),%rcx
+	# in2 = *(uint64 *) (x + 8)
+	movq	8(%r8),%r9
+	# in4 = *(uint64 *) (x + 16)
+	movq	16(%r8),%rax
+	# in6 = *(uint64 *) (x + 24)
+	movq	24(%r8),%r10
+	# in8 = *(uint64 *) (x + 32)
+	movq	32(%r8),%r11
+	# in10 = *(uint64 *) (x + 40)
+	movq	40(%r8),%r12
+	# in12 = *(uint64 *) (x + 48)
+	movq	48(%r8),%r13
+	# in14 = *(uint64 *) (x + 56)
+	movq	56(%r8),%r14
+	# j0 = in0
+	movq	%rcx,56(%rsp)
+	# j2 = in2
+	movq	%r9,64(%rsp)
+	# j4 = in4
+	movq	%rax,72(%rsp)
+	# j6 = in6
+	movq	%r10,80(%rsp)
+	# j8 = in8
+	movq	%r11,88(%rsp)
+	# j10 = in10
+	movq	%r12,96(%rsp)
+	# j12 = in12
+	movq	%r13,104(%rsp)
+	# j14 = in14
+	movq	%r14,112(%rsp)
+	# x_backup = x
+	movq	%r8,120(%rsp)
+# bytesatleast1:
+._bytesatleast1:
+	#                   unsigned<? bytes - 64
+	cmp	$64,%rdx
+	# comment:fp stack unchanged by jump
+	#   goto nocopy if !unsigned<
+	jae	._nocopy
+	#     ctarget = out
+	movq	%rdi,128(%rsp)
+	#     out = &tmp
+	leaq	192(%rsp),%rdi
+	#     i = bytes
+	mov	%rdx,%rcx
+	#     while (i) { *out++ = *m++; --i }
+	rep	movsb
+	#     out = &tmp
+	leaq	192(%rsp),%rdi
+	#     m = &tmp
+	leaq	192(%rsp),%rsi
+	# comment:fp stack unchanged by fallthrough
+#   nocopy:
+._nocopy:
+	#   out_backup = out
+	movq	%rdi,136(%rsp)
+	#   m_backup = m
+	movq	%rsi,144(%rsp)
+	#   bytes_backup = bytes
+	movq	%rdx,152(%rsp)
+	#   x1 = j0
+	movq	56(%rsp),%rdi
+	#   x0 = x1
+	mov	%rdi,%rdx
+	#   (uint64) x1 >>= 32
+	shr	$32,%rdi
+	#   		x3 = j2
+	movq	64(%rsp),%rsi
+	#   		x2 = x3
+	mov	%rsi,%rcx
+	#   		(uint64) x3 >>= 32
+	shr	$32,%rsi
+	#   x5 = j4
+	movq	72(%rsp),%r8
+	#   x4 = x5
+	mov	%r8,%r9
+	#   (uint64) x5 >>= 32
+	shr	$32,%r8
+	#   x5_stack = x5
+	movq	%r8,160(%rsp)
+	#   		x7 = j6
+	movq	80(%rsp),%r8
+	#   		x6 = x7
+	mov	%r8,%rax
+	#   		(uint64) x7 >>= 32
+	shr	$32,%r8
+	#   x9 = j8
+	movq	88(%rsp),%r10
+	#   x8 = x9
+	mov	%r10,%r11
+	#   (uint64) x9 >>= 32
+	shr	$32,%r10
+	#   		x11 = j10
+	movq	96(%rsp),%r12
+	#   		x10 = x11
+	mov	%r12,%r13
+	#   		x10_stack = x10
+	movq	%r13,168(%rsp)
+	#   		(uint64) x11 >>= 32
+	shr	$32,%r12
+	#   x13 = j12
+	movq	104(%rsp),%r13
+	#   x12 = x13
+	mov	%r13,%r14
+	#   (uint64) x13 >>= 32
+	shr	$32,%r13
+	#   		x15 = j14
+	movq	112(%rsp),%r15
+	#   		x14 = x15
+	mov	%r15,%rbx
+	#   		(uint64) x15 >>= 32
+	shr	$32,%r15
+	#   		x15_stack = x15
+	movq	%r15,176(%rsp)
+	#   i = 20
+	mov	$20,%r15
+#   mainloop:
+._mainloop:
+	#   i_backup = i
+	movq	%r15,184(%rsp)
+	# 		x5 = x5_stack
+	movq	160(%rsp),%r15
+	# a = x12 + x0
+	lea	(%r14,%rdx),%rbp
+	# (uint32) a <<<= 7
+	rol	$7,%ebp
+	# x4 ^= a
+	xor	%rbp,%r9
+	# 		b = x1 + x5
+	lea	(%rdi,%r15),%rbp
+	# 		(uint32) b <<<= 7
+	rol	$7,%ebp
+	# 		x9 ^= b
+	xor	%rbp,%r10
+	# a = x0 + x4
+	lea	(%rdx,%r9),%rbp
+	# (uint32) a <<<= 9
+	rol	$9,%ebp
+	# x8 ^= a
+	xor	%rbp,%r11
+	# 		b = x5 + x9
+	lea	(%r15,%r10),%rbp
+	# 		(uint32) b <<<= 9
+	rol	$9,%ebp
+	# 		x13 ^= b
+	xor	%rbp,%r13
+	# a = x4 + x8
+	lea	(%r9,%r11),%rbp
+	# (uint32) a <<<= 13
+	rol	$13,%ebp
+	# x12 ^= a
+	xor	%rbp,%r14
+	# 		b = x9 + x13
+	lea	(%r10,%r13),%rbp
+	# 		(uint32) b <<<= 13
+	rol	$13,%ebp
+	# 		x1 ^= b
+	xor	%rbp,%rdi
+	# a = x8 + x12
+	lea	(%r11,%r14),%rbp
+	# (uint32) a <<<= 18
+	rol	$18,%ebp
+	# x0 ^= a
+	xor	%rbp,%rdx
+	# 		b = x13 + x1
+	lea	(%r13,%rdi),%rbp
+	# 		(uint32) b <<<= 18
+	rol	$18,%ebp
+	# 		x5 ^= b
+	xor	%rbp,%r15
+	# 				x10 = x10_stack
+	movq	168(%rsp),%rbp
+	# 		x5_stack = x5
+	movq	%r15,160(%rsp)
+	# 				c = x6 + x10
+	lea	(%rax,%rbp),%r15
+	# 				(uint32) c <<<= 7
+	rol	$7,%r15d
+	# 				x14 ^= c
+	xor	%r15,%rbx
+	# 				c = x10 + x14
+	lea	(%rbp,%rbx),%r15
+	# 				(uint32) c <<<= 9
+	rol	$9,%r15d
+	# 				x2 ^= c
+	xor	%r15,%rcx
+	# 				c = x14 + x2
+	lea	(%rbx,%rcx),%r15
+	# 				(uint32) c <<<= 13
+	rol	$13,%r15d
+	# 				x6 ^= c
+	xor	%r15,%rax
+	# 				c = x2 + x6
+	lea	(%rcx,%rax),%r15
+	# 				(uint32) c <<<= 18
+	rol	$18,%r15d
+	# 				x10 ^= c
+	xor	%r15,%rbp
+	# 						x15 = x15_stack
+	movq	176(%rsp),%r15
+	# 				x10_stack = x10
+	movq	%rbp,168(%rsp)
+	# 						d = x11 + x15
+	lea	(%r12,%r15),%rbp
+	# 						(uint32) d <<<= 7
+	rol	$7,%ebp
+	# 						x3 ^= d
+	xor	%rbp,%rsi
+	# 						d = x15 + x3
+	lea	(%r15,%rsi),%rbp
+	# 						(uint32) d <<<= 9
+	rol	$9,%ebp
+	# 						x7 ^= d
+	xor	%rbp,%r8
+	# 						d = x3 + x7
+	lea	(%rsi,%r8),%rbp
+	# 						(uint32) d <<<= 13
+	rol	$13,%ebp
+	# 						x11 ^= d
+	xor	%rbp,%r12
+	# 						d = x7 + x11
+	lea	(%r8,%r12),%rbp
+	# 						(uint32) d <<<= 18
+	rol	$18,%ebp
+	# 						x15 ^= d
+	xor	%rbp,%r15
+	# 						x15_stack = x15
+	movq	%r15,176(%rsp)
+	# 		x5 = x5_stack
+	movq	160(%rsp),%r15
+	# a = x3 + x0
+	lea	(%rsi,%rdx),%rbp
+	# (uint32) a <<<= 7
+	rol	$7,%ebp
+	# x1 ^= a
+	xor	%rbp,%rdi
+	# 		b = x4 + x5
+	lea	(%r9,%r15),%rbp
+	# 		(uint32) b <<<= 7
+	rol	$7,%ebp
+	# 		x6 ^= b
+	xor	%rbp,%rax
+	# a = x0 + x1
+	lea	(%rdx,%rdi),%rbp
+	# (uint32) a <<<= 9
+	rol	$9,%ebp
+	# x2 ^= a
+	xor	%rbp,%rcx
+	# 		b = x5 + x6
+	lea	(%r15,%rax),%rbp
+	# 		(uint32) b <<<= 9
+	rol	$9,%ebp
+	# 		x7 ^= b
+	xor	%rbp,%r8
+	# a = x1 + x2
+	lea	(%rdi,%rcx),%rbp
+	# (uint32) a <<<= 13
+	rol	$13,%ebp
+	# x3 ^= a
+	xor	%rbp,%rsi
+	# 		b = x6 + x7
+	lea	(%rax,%r8),%rbp
+	# 		(uint32) b <<<= 13
+	rol	$13,%ebp
+	# 		x4 ^= b
+	xor	%rbp,%r9
+	# a = x2 + x3
+	lea	(%rcx,%rsi),%rbp
+	# (uint32) a <<<= 18
+	rol	$18,%ebp
+	# x0 ^= a
+	xor	%rbp,%rdx
+	# 		b = x7 + x4
+	lea	(%r8,%r9),%rbp
+	# 		(uint32) b <<<= 18
+	rol	$18,%ebp
+	# 		x5 ^= b
+	xor	%rbp,%r15
+	# 				x10 = x10_stack
+	movq	168(%rsp),%rbp
+	# 		x5_stack = x5
+	movq	%r15,160(%rsp)
+	# 				c = x9 + x10
+	lea	(%r10,%rbp),%r15
+	# 				(uint32) c <<<= 7
+	rol	$7,%r15d
+	# 				x11 ^= c
+	xor	%r15,%r12
+	# 				c = x10 + x11
+	lea	(%rbp,%r12),%r15
+	# 				(uint32) c <<<= 9
+	rol	$9,%r15d
+	# 				x8 ^= c
+	xor	%r15,%r11
+	# 				c = x11 + x8
+	lea	(%r12,%r11),%r15
+	# 				(uint32) c <<<= 13
+	rol	$13,%r15d
+	# 				x9 ^= c
+	xor	%r15,%r10
+	# 				c = x8 + x9
+	lea	(%r11,%r10),%r15
+	# 				(uint32) c <<<= 18
+	rol	$18,%r15d
+	# 				x10 ^= c
+	xor	%r15,%rbp
+	# 						x15 = x15_stack
+	movq	176(%rsp),%r15
+	# 				x10_stack = x10
+	movq	%rbp,168(%rsp)
+	# 						d = x14 + x15
+	lea	(%rbx,%r15),%rbp
+	# 						(uint32) d <<<= 7
+	rol	$7,%ebp
+	# 						x12 ^= d
+	xor	%rbp,%r14
+	# 						d = x15 + x12
+	lea	(%r15,%r14),%rbp
+	# 						(uint32) d <<<= 9
+	rol	$9,%ebp
+	# 						x13 ^= d
+	xor	%rbp,%r13
+	# 						d = x12 + x13
+	lea	(%r14,%r13),%rbp
+	# 						(uint32) d <<<= 13
+	rol	$13,%ebp
+	# 						x14 ^= d
+	xor	%rbp,%rbx
+	# 						d = x13 + x14
+	lea	(%r13,%rbx),%rbp
+	# 						(uint32) d <<<= 18
+	rol	$18,%ebp
+	# 						x15 ^= d
+	xor	%rbp,%r15
+	# 						x15_stack = x15
+	movq	%r15,176(%rsp)
+	# 		x5 = x5_stack
+	movq	160(%rsp),%r15
+	# a = x12 + x0
+	lea	(%r14,%rdx),%rbp
+	# (uint32) a <<<= 7
+	rol	$7,%ebp
+	# x4 ^= a
+	xor	%rbp,%r9
+	# 		b = x1 + x5
+	lea	(%rdi,%r15),%rbp
+	# 		(uint32) b <<<= 7
+	rol	$7,%ebp
+	# 		x9 ^= b
+	xor	%rbp,%r10
+	# a = x0 + x4
+	lea	(%rdx,%r9),%rbp
+	# (uint32) a <<<= 9
+	rol	$9,%ebp
+	# x8 ^= a
+	xor	%rbp,%r11
+	# 		b = x5 + x9
+	lea	(%r15,%r10),%rbp
+	# 		(uint32) b <<<= 9
+	rol	$9,%ebp
+	# 		x13 ^= b
+	xor	%rbp,%r13
+	# a = x4 + x8
+	lea	(%r9,%r11),%rbp
+	# (uint32) a <<<= 13
+	rol	$13,%ebp
+	# x12 ^= a
+	xor	%rbp,%r14
+	# 		b = x9 + x13
+	lea	(%r10,%r13),%rbp
+	# 		(uint32) b <<<= 13
+	rol	$13,%ebp
+	# 		x1 ^= b
+	xor	%rbp,%rdi
+	# a = x8 + x12
+	lea	(%r11,%r14),%rbp
+	# (uint32) a <<<= 18
+	rol	$18,%ebp
+	# x0 ^= a
+	xor	%rbp,%rdx
+	# 		b = x13 + x1
+	lea	(%r13,%rdi),%rbp
+	# 		(uint32) b <<<= 18
+	rol	$18,%ebp
+	# 		x5 ^= b
+	xor	%rbp,%r15
+	# 				x10 = x10_stack
+	movq	168(%rsp),%rbp
+	# 		x5_stack = x5
+	movq	%r15,160(%rsp)
+	# 				c = x6 + x10
+	lea	(%rax,%rbp),%r15
+	# 				(uint32) c <<<= 7
+	rol	$7,%r15d
+	# 				x14 ^= c
+	xor	%r15,%rbx
+	# 				c = x10 + x14
+	lea	(%rbp,%rbx),%r15
+	# 				(uint32) c <<<= 9
+	rol	$9,%r15d
+	# 				x2 ^= c
+	xor	%r15,%rcx
+	# 				c = x14 + x2
+	lea	(%rbx,%rcx),%r15
+	# 				(uint32) c <<<= 13
+	rol	$13,%r15d
+	# 				x6 ^= c
+	xor	%r15,%rax
+	# 				c = x2 + x6
+	lea	(%rcx,%rax),%r15
+	# 				(uint32) c <<<= 18
+	rol	$18,%r15d
+	# 				x10 ^= c
+	xor	%r15,%rbp
+	# 						x15 = x15_stack
+	movq	176(%rsp),%r15
+	# 				x10_stack = x10
+	movq	%rbp,168(%rsp)
+	# 						d = x11 + x15
+	lea	(%r12,%r15),%rbp
+	# 						(uint32) d <<<= 7
+	rol	$7,%ebp
+	# 						x3 ^= d
+	xor	%rbp,%rsi
+	# 						d = x15 + x3
+	lea	(%r15,%rsi),%rbp
+	# 						(uint32) d <<<= 9
+	rol	$9,%ebp
+	# 						x7 ^= d
+	xor	%rbp,%r8
+	# 						d = x3 + x7
+	lea	(%rsi,%r8),%rbp
+	# 						(uint32) d <<<= 13
+	rol	$13,%ebp
+	# 						x11 ^= d
+	xor	%rbp,%r12
+	# 						d = x7 + x11
+	lea	(%r8,%r12),%rbp
+	# 						(uint32) d <<<= 18
+	rol	$18,%ebp
+	# 						x15 ^= d
+	xor	%rbp,%r15
+	# 						x15_stack = x15
+	movq	%r15,176(%rsp)
+	# 		x5 = x5_stack
+	movq	160(%rsp),%r15
+	# a = x3 + x0
+	lea	(%rsi,%rdx),%rbp
+	# (uint32) a <<<= 7
+	rol	$7,%ebp
+	# x1 ^= a
+	xor	%rbp,%rdi
+	# 		b = x4 + x5
+	lea	(%r9,%r15),%rbp
+	# 		(uint32) b <<<= 7
+	rol	$7,%ebp
+	# 		x6 ^= b
+	xor	%rbp,%rax
+	# a = x0 + x1
+	lea	(%rdx,%rdi),%rbp
+	# (uint32) a <<<= 9
+	rol	$9,%ebp
+	# x2 ^= a
+	xor	%rbp,%rcx
+	# 		b = x5 + x6
+	lea	(%r15,%rax),%rbp
+	# 		(uint32) b <<<= 9
+	rol	$9,%ebp
+	# 		x7 ^= b
+	xor	%rbp,%r8
+	# a = x1 + x2
+	lea	(%rdi,%rcx),%rbp
+	# (uint32) a <<<= 13
+	rol	$13,%ebp
+	# x3 ^= a
+	xor	%rbp,%rsi
+	# 		b = x6 + x7
+	lea	(%rax,%r8),%rbp
+	# 		(uint32) b <<<= 13
+	rol	$13,%ebp
+	# 		x4 ^= b
+	xor	%rbp,%r9
+	# a = x2 + x3
+	lea	(%rcx,%rsi),%rbp
+	# (uint32) a <<<= 18
+	rol	$18,%ebp
+	# x0 ^= a
+	xor	%rbp,%rdx
+	# 		b = x7 + x4
+	lea	(%r8,%r9),%rbp
+	# 		(uint32) b <<<= 18
+	rol	$18,%ebp
+	# 		x5 ^= b
+	xor	%rbp,%r15
+	# 				x10 = x10_stack
+	movq	168(%rsp),%rbp
+	# 		x5_stack = x5
+	movq	%r15,160(%rsp)
+	# 				c = x9 + x10
+	lea	(%r10,%rbp),%r15
+	# 				(uint32) c <<<= 7
+	rol	$7,%r15d
+	# 				x11 ^= c
+	xor	%r15,%r12
+	# 				c = x10 + x11
+	lea	(%rbp,%r12),%r15
+	# 				(uint32) c <<<= 9
+	rol	$9,%r15d
+	# 				x8 ^= c
+	xor	%r15,%r11
+	# 				c = x11 + x8
+	lea	(%r12,%r11),%r15
+	# 				(uint32) c <<<= 13
+	rol	$13,%r15d
+	# 				x9 ^= c
+	xor	%r15,%r10
+	# 				c = x8 + x9
+	lea	(%r11,%r10),%r15
+	# 				(uint32) c <<<= 18
+	rol	$18,%r15d
+	# 				x10 ^= c
+	xor	%r15,%rbp
+	# 						x15 = x15_stack
+	movq	176(%rsp),%r15
+	# 				x10_stack = x10
+	movq	%rbp,168(%rsp)
+	# 						d = x14 + x15
+	lea	(%rbx,%r15),%rbp
+	# 						(uint32) d <<<= 7
+	rol	$7,%ebp
+	# 						x12 ^= d
+	xor	%rbp,%r14
+	# 						d = x15 + x12
+	lea	(%r15,%r14),%rbp
+	# 						(uint32) d <<<= 9
+	rol	$9,%ebp
+	# 						x13 ^= d
+	xor	%rbp,%r13
+	# 						d = x12 + x13
+	lea	(%r14,%r13),%rbp
+	# 						(uint32) d <<<= 13
+	rol	$13,%ebp
+	# 						x14 ^= d
+	xor	%rbp,%rbx
+	# 						d = x13 + x14
+	lea	(%r13,%rbx),%rbp
+	# 						(uint32) d <<<= 18
+	rol	$18,%ebp
+	# 						x15 ^= d
+	xor	%rbp,%r15
+	# 						x15_stack = x15
+	movq	%r15,176(%rsp)
+	#   i = i_backup
+	movq	184(%rsp),%r15
+	#                  unsigned>? i -= 4
+	sub	$4,%r15
+	# comment:fp stack unchanged by jump
+	# goto mainloop if unsigned>
+	ja	._mainloop
+	#   (uint32) x2 += j2
+	addl	64(%rsp),%ecx
+	#   x3 <<= 32
+	shl	$32,%rsi
+	#   x3 += j2
+	addq	64(%rsp),%rsi
+	#   (uint64) x3 >>= 32
+	shr	$32,%rsi
+	#   x3 <<= 32
+	shl	$32,%rsi
+	#   x2 += x3
+	add	%rsi,%rcx
+	#   (uint32) x6 += j6
+	addl	80(%rsp),%eax
+	#   x7 <<= 32
+	shl	$32,%r8
+	#   x7 += j6
+	addq	80(%rsp),%r8
+	#   (uint64) x7 >>= 32
+	shr	$32,%r8
+	#   x7 <<= 32
+	shl	$32,%r8
+	#   x6 += x7
+	add	%r8,%rax
+	#   (uint32) x8 += j8
+	addl	88(%rsp),%r11d
+	#   x9 <<= 32
+	shl	$32,%r10
+	#   x9 += j8
+	addq	88(%rsp),%r10
+	#   (uint64) x9 >>= 32
+	shr	$32,%r10
+	#   x9 <<= 32
+	shl	$32,%r10
+	#   x8 += x9
+	add	%r10,%r11
+	#   (uint32) x12 += j12
+	addl	104(%rsp),%r14d
+	#   x13 <<= 32
+	shl	$32,%r13
+	#   x13 += j12
+	addq	104(%rsp),%r13
+	#   (uint64) x13 >>= 32
+	shr	$32,%r13
+	#   x13 <<= 32
+	shl	$32,%r13
+	#   x12 += x13
+	add	%r13,%r14
+	#   (uint32) x0 += j0
+	addl	56(%rsp),%edx
+	#   x1 <<= 32
+	shl	$32,%rdi
+	#   x1 += j0
+	addq	56(%rsp),%rdi
+	#   (uint64) x1 >>= 32
+	shr	$32,%rdi
+	#   x1 <<= 32
+	shl	$32,%rdi
+	#   x0 += x1
+	add	%rdi,%rdx
+	#   x5 = x5_stack
+	movq	160(%rsp),%rdi
+	#   (uint32) x4 += j4
+	addl	72(%rsp),%r9d
+	#   x5 <<= 32
+	shl	$32,%rdi
+	#   x5 += j4
+	addq	72(%rsp),%rdi
+	#   (uint64) x5 >>= 32
+	shr	$32,%rdi
+	#   x5 <<= 32
+	shl	$32,%rdi
+	#   x4 += x5
+	add	%rdi,%r9
+	#   x10 = x10_stack
+	movq	168(%rsp),%r8
+	#   (uint32) x10 += j10
+	addl	96(%rsp),%r8d
+	#   x11 <<= 32
+	shl	$32,%r12
+	#   x11 += j10
+	addq	96(%rsp),%r12
+	#   (uint64) x11 >>= 32
+	shr	$32,%r12
+	#   x11 <<= 32
+	shl	$32,%r12
+	#   x10 += x11
+	add	%r12,%r8
+	#   x15 = x15_stack
+	movq	176(%rsp),%rdi
+	#   (uint32) x14 += j14
+	addl	112(%rsp),%ebx
+	#   x15 <<= 32
+	shl	$32,%rdi
+	#   x15 += j14
+	addq	112(%rsp),%rdi
+	#   (uint64) x15 >>= 32
+	shr	$32,%rdi
+	#   x15 <<= 32
+	shl	$32,%rdi
+	#   x14 += x15
+	add	%rdi,%rbx
+	#   out = out_backup
+	movq	136(%rsp),%rdi
+	#   m = m_backup
+	movq	144(%rsp),%rsi
+	#   x0 ^= *(uint64 *) (m + 0)
+	xorq	0(%rsi),%rdx
+	#   *(uint64 *) (out + 0) = x0
+	movq	%rdx,0(%rdi)
+	#   x2 ^= *(uint64 *) (m + 8)
+	xorq	8(%rsi),%rcx
+	#   *(uint64 *) (out + 8) = x2
+	movq	%rcx,8(%rdi)
+	#   x4 ^= *(uint64 *) (m + 16)
+	xorq	16(%rsi),%r9
+	#   *(uint64 *) (out + 16) = x4
+	movq	%r9,16(%rdi)
+	#   x6 ^= *(uint64 *) (m + 24)
+	xorq	24(%rsi),%rax
+	#   *(uint64 *) (out + 24) = x6
+	movq	%rax,24(%rdi)
+	#   x8 ^= *(uint64 *) (m + 32)
+	xorq	32(%rsi),%r11
+	#   *(uint64 *) (out + 32) = x8
+	movq	%r11,32(%rdi)
+	#   x10 ^= *(uint64 *) (m + 40)
+	xorq	40(%rsi),%r8
+	#   *(uint64 *) (out + 40) = x10
+	movq	%r8,40(%rdi)
+	#   x12 ^= *(uint64 *) (m + 48)
+	xorq	48(%rsi),%r14
+	#   *(uint64 *) (out + 48) = x12
+	movq	%r14,48(%rdi)
+	#   x14 ^= *(uint64 *) (m + 56)
+	xorq	56(%rsi),%rbx
+	#   *(uint64 *) (out + 56) = x14
+	movq	%rbx,56(%rdi)
+	#   bytes = bytes_backup
+	movq	152(%rsp),%rdx
+	#   in8 = j8
+	movq	88(%rsp),%rcx
+	#   in8 += 1
+	add	$1,%rcx
+	#   j8 = in8
+	movq	%rcx,88(%rsp)
+	#                          unsigned>? unsigned<? bytes - 64
+	cmp	$64,%rdx
+	# comment:fp stack unchanged by jump
+	#   goto bytesatleast65 if unsigned>
+	ja	._bytesatleast65
+	# comment:fp stack unchanged by jump
+	#     goto bytesatleast64 if !unsigned<
+	jae	._bytesatleast64
+	#       m = out
+	mov	%rdi,%rsi
+	#       out = ctarget
+	movq	128(%rsp),%rdi
+	#       i = bytes
+	mov	%rdx,%rcx
+	#       while (i) { *out++ = *m++; --i }
+	rep	movsb
+	# comment:fp stack unchanged by fallthrough
+#     bytesatleast64:
+._bytesatleast64:
+	#     x = x_backup
+	movq	120(%rsp),%rdi
+	#     in8 = j8
+	movq	88(%rsp),%rsi
+	#     *(uint64 *) (x + 32) = in8
+	movq	%rsi,32(%rdi)
+	#     r11 = r11_stack
+	movq	0(%rsp),%r11
+	#     r12 = r12_stack
+	movq	8(%rsp),%r12
+	#     r13 = r13_stack
+	movq	16(%rsp),%r13
+	#     r14 = r14_stack
+	movq	24(%rsp),%r14
+	#     r15 = r15_stack
+	movq	32(%rsp),%r15
+	#     rbx = rbx_stack
+	movq	40(%rsp),%rbx
+	#     rbp = rbp_stack
+	movq	48(%rsp),%rbp
+	# comment:fp stack unchanged by fallthrough
+#     done:
+._done:
+	#     leave
+	add	%r11,%rsp
+	mov	%rdi,%rax
+	mov	%rsi,%rdx
+	ret
+#   bytesatleast65:
+._bytesatleast65:
+	#   bytes -= 64
+	sub	$64,%rdx
+	#   out += 64
+	add	$64,%rdi
+	#   m += 64
+	add	$64,%rsi
+	# comment:fp stack unchanged by jump
+	# goto bytesatleast1
+	jmp	._bytesatleast1
+# enter ECRYPT_keysetup
+.text
+.p2align 5
+.globl ECRYPT_keysetup
+ECRYPT_keysetup:
+	mov	%rsp,%r11
+	and	$31,%r11
+	add	$256,%r11
+	sub	%r11,%rsp
+	#   k = arg2
+	mov	%rsi,%rsi
+	#   kbits = arg3
+	mov	%rdx,%rdx
+	#   x = arg1
+	mov	%rdi,%rdi
+	#   in0 = *(uint64 *) (k + 0)
+	movq	0(%rsi),%r8
+	#   in2 = *(uint64 *) (k + 8)
+	movq	8(%rsi),%r9
+	#   *(uint64 *) (x + 4) = in0
+	movq	%r8,4(%rdi)
+	#   *(uint64 *) (x + 12) = in2
+	movq	%r9,12(%rdi)
+	#                    unsigned<? kbits - 256
+	cmp	$256,%rdx
+	# comment:fp stack unchanged by jump
+	#   goto kbits128 if unsigned<
+	jb	._kbits128
+#   kbits256:
+._kbits256:
+	#     in10 = *(uint64 *) (k + 16)
+	movq	16(%rsi),%rdx
+	#     in12 = *(uint64 *) (k + 24)
+	movq	24(%rsi),%rsi
+	#     *(uint64 *) (x + 44) = in10
+	movq	%rdx,44(%rdi)
+	#     *(uint64 *) (x + 52) = in12
+	movq	%rsi,52(%rdi)
+	#     in0 = 1634760805
+	mov	$1634760805,%rsi
+	#     in4 = 857760878
+	mov	$857760878,%rdx
+	#     in10 = 2036477234
+	mov	$2036477234,%rcx
+	#     in14 = 1797285236
+	mov	$1797285236,%r8
+	#     *(uint32 *) (x + 0) = in0
+	movl	%esi,0(%rdi)
+	#     *(uint32 *) (x + 20) = in4
+	movl	%edx,20(%rdi)
+	#     *(uint32 *) (x + 40) = in10
+	movl	%ecx,40(%rdi)
+	#     *(uint32 *) (x + 60) = in14
+	movl	%r8d,60(%rdi)
+	# comment:fp stack unchanged by jump
+	#   goto keysetupdone
+	jmp	._keysetupdone
+#   kbits128:
+._kbits128:
+	#     in10 = *(uint64 *) (k + 0)
+	movq	0(%rsi),%rdx
+	#     in12 = *(uint64 *) (k + 8)
+	movq	8(%rsi),%rsi
+	#     *(uint64 *) (x + 44) = in10
+	movq	%rdx,44(%rdi)
+	#     *(uint64 *) (x + 52) = in12
+	movq	%rsi,52(%rdi)
+	#     in0 = 1634760805
+	mov	$1634760805,%rsi
+	#     in4 = 824206446
+	mov	$824206446,%rdx
+	#     in10 = 2036477238
+	mov	$2036477238,%rcx
+	#     in14 = 1797285236
+	mov	$1797285236,%r8
+	#     *(uint32 *) (x + 0) = in0
+	movl	%esi,0(%rdi)
+	#     *(uint32 *) (x + 20) = in4
+	movl	%edx,20(%rdi)
+	#     *(uint32 *) (x + 40) = in10
+	movl	%ecx,40(%rdi)
+	#     *(uint32 *) (x + 60) = in14
+	movl	%r8d,60(%rdi)
+#   keysetupdone:
+._keysetupdone:
+	# leave
+	add	%r11,%rsp
+	mov	%rdi,%rax
+	mov	%rsi,%rdx
+	ret
+# enter ECRYPT_ivsetup
+.text
+.p2align 5
+.globl ECRYPT_ivsetup
+ECRYPT_ivsetup:
+	mov	%rsp,%r11
+	and	$31,%r11
+	add	$256,%r11
+	sub	%r11,%rsp
+	#   iv = arg2
+	mov	%rsi,%rsi
+	#   x = arg1
+	mov	%rdi,%rdi
+	#   in6 = *(uint64 *) (iv + 0)
+	movq	0(%rsi),%rsi
+	#   in8 = 0
+	mov	$0,%r8
+	#   *(uint64 *) (x + 24) = in6
+	movq	%rsi,24(%rdi)
+	#   *(uint64 *) (x + 32) = in8
+	movq	%r8,32(%rdi)
+	# leave
+	add	%r11,%rsp
+	mov	%rdi,%rax
+	mov	%rsi,%rdx
+	ret

arch/x86/crypto/salsa20_glue.c

@@ -8,6 +8,8 @@
  * and to remove extraneous comments and functions that are not needed.
  * - i586 version, renamed as salsa20-i586-asm_32.S
  *   available from <http://cr.yp.to/snuffle/salsa20/x86-pm/salsa20.s>
+ * - x86-64 version, renamed as salsa20-x86_64-asm_64.S
+ *   available from <http://cr.yp.to/snuffle/salsa20/amd64-3/salsa20.s>
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of the GNU General Public License as published by the Free

crypto/Kconfig

@@ -504,6 +504,21 @@ config CRYPTO_SALSA20_586
 	  The Salsa20 stream cipher algorithm is designed by Daniel J.
 	  Bernstein <djb@cr.yp.to>. See <http://cr.yp.to/snuffle.html>
 
+config CRYPTO_SALSA20_X86_64
+	tristate "Salsa20 stream cipher algorithm (x86_64) (EXPERIMENTAL)"
+	depends on (X86 || UML_X86) && 64BIT
+	depends on EXPERIMENTAL
+	select CRYPTO_BLKCIPHER
+	select CRYPTO_SALSA20
+	help
+	  Salsa20 stream cipher algorithm.
+
+	  Salsa20 is a stream cipher submitted to eSTREAM, the ECRYPT
+	  Stream Cipher Project. See <http://www.ecrypt.eu.org/stream/>
+
+	  The Salsa20 stream cipher algorithm is designed by Daniel J.
+	  Bernstein <djb@cr.yp.to>. See <http://cr.yp.to/snuffle.html>
+
 config CRYPTO_DEFLATE
 	tristate "Deflate compression algorithm"
 	select CRYPTO_ALGAPI