crypto: x86/salsa20 - cleanup and convert to skcipher API

Convert salsa20-asm from the deprecated "blkcipher" API to the
"skcipher" API, in the process fixing it up to use the generic helpers.
This allows removing the salsa20_keysetup() and salsa20_ivsetup()
assembly functions, which aren't performance critical; the C versions do
just fine.

This also fixes the same bug that salsa20-generic had, where the state
array was being maintained directly in the transform context rather than
on the stack or in the request context.  Thus, if multiple threads used
the same Salsa20 transform concurrently they produced the wrong results.

Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

1 files changed, 4 insertions, 180 deletions

diff --git a/arch/x86/crypto/salsa20-i586-asm_32.S b/arch/x86/crypto/salsa20-i586-asm_32.S
index 329452b8f794..6014b7b9e52a 100644
--- a/arch/x86/crypto/salsa20-i586-asm_32.S
+++ b/arch/x86/crypto/salsa20-i586-asm_32.S
@@ -1,6 +1,7 @@
-# salsa20_pm.s version 20051229
-# D. J. Bernstein
-# Public domain.
+# Derived from:
+#	salsa20_pm.s version 20051229
+#	D. J. Bernstein
+#	Public domain.
 
 #include <linux/linkage.h>
 
@@ -935,180 +936,3 @@ ENTRY(salsa20_encrypt_bytes)
 	# goto bytesatleast1
 	jmp	._bytesatleast1
 ENDPROC(salsa20_encrypt_bytes)
-
-# enter salsa20_keysetup
-ENTRY(salsa20_keysetup)
-	mov	%esp,%eax
-	and	$31,%eax
-	add	$256,%eax
-	sub	%eax,%esp
-	#   eax_stack = eax
-	movl	%eax,64(%esp)
-	#   ebx_stack = ebx
-	movl	%ebx,68(%esp)
-	#   esi_stack = esi
-	movl	%esi,72(%esp)
-	#   edi_stack = edi
-	movl	%edi,76(%esp)
-	#   ebp_stack = ebp
-	movl	%ebp,80(%esp)
-	#   k = arg2
-	movl	8(%esp,%eax),%ecx
-	#   kbits = arg3
-	movl	12(%esp,%eax),%edx
-	#   x = arg1
-	movl	4(%esp,%eax),%eax
-	#   in1 = *(uint32 *) (k + 0)
-	movl	0(%ecx),%ebx
-	#   in2 = *(uint32 *) (k + 4)
-	movl	4(%ecx),%esi
-	#   in3 = *(uint32 *) (k + 8)
-	movl	8(%ecx),%edi
-	#   in4 = *(uint32 *) (k + 12)
-	movl	12(%ecx),%ebp
-	#   *(uint32 *) (x + 4) = in1
-	movl	%ebx,4(%eax)
-	#   *(uint32 *) (x + 8) = in2
-	movl	%esi,8(%eax)
-	#   *(uint32 *) (x + 12) = in3
-	movl	%edi,12(%eax)
-	#   *(uint32 *) (x + 16) = in4
-	movl	%ebp,16(%eax)
-	#   kbits - 256
-	cmp	$256,%edx
-	#   goto kbits128 if unsigned<
-	jb	._kbits128
-._kbits256:
-	#     in11 = *(uint32 *) (k + 16)
-	movl	16(%ecx),%edx
-	#     in12 = *(uint32 *) (k + 20)
-	movl	20(%ecx),%ebx
-	#     in13 = *(uint32 *) (k + 24)
-	movl	24(%ecx),%esi
-	#     in14 = *(uint32 *) (k + 28)
-	movl	28(%ecx),%ecx
-	#     *(uint32 *) (x + 44) = in11
-	movl	%edx,44(%eax)
-	#     *(uint32 *) (x + 48) = in12
-	movl	%ebx,48(%eax)
-	#     *(uint32 *) (x + 52) = in13
-	movl	%esi,52(%eax)
-	#     *(uint32 *) (x + 56) = in14
-	movl	%ecx,56(%eax)
-	#     in0 = 1634760805
-	mov	$1634760805,%ecx
-	#     in5 = 857760878
-	mov	$857760878,%edx
-	#     in10 = 2036477234
-	mov	$2036477234,%ebx
-	#     in15 = 1797285236
-	mov	$1797285236,%esi
-	#     *(uint32 *) (x + 0) = in0
-	movl	%ecx,0(%eax)
-	#     *(uint32 *) (x + 20) = in5
-	movl	%edx,20(%eax)
-	#     *(uint32 *) (x + 40) = in10
-	movl	%ebx,40(%eax)
-	#     *(uint32 *) (x + 60) = in15
-	movl	%esi,60(%eax)
-	#   goto keysetupdone
-	jmp	._keysetupdone
-._kbits128:
-	#     in11 = *(uint32 *) (k + 0)
-	movl	0(%ecx),%edx
-	#     in12 = *(uint32 *) (k + 4)
-	movl	4(%ecx),%ebx
-	#     in13 = *(uint32 *) (k + 8)
-	movl	8(%ecx),%esi
-	#     in14 = *(uint32 *) (k + 12)
-	movl	12(%ecx),%ecx
-	#     *(uint32 *) (x + 44) = in11
-	movl	%edx,44(%eax)
-	#     *(uint32 *) (x + 48) = in12
-	movl	%ebx,48(%eax)
-	#     *(uint32 *) (x + 52) = in13
-	movl	%esi,52(%eax)
-	#     *(uint32 *) (x + 56) = in14
-	movl	%ecx,56(%eax)
-	#     in0 = 1634760805
-	mov	$1634760805,%ecx
-	#     in5 = 824206446
-	mov	$824206446,%edx
-	#     in10 = 2036477238
-	mov	$2036477238,%ebx
-	#     in15 = 1797285236
-	mov	$1797285236,%esi
-	#     *(uint32 *) (x + 0) = in0
-	movl	%ecx,0(%eax)
-	#     *(uint32 *) (x + 20) = in5
-	movl	%edx,20(%eax)
-	#     *(uint32 *) (x + 40) = in10
-	movl	%ebx,40(%eax)
-	#     *(uint32 *) (x + 60) = in15
-	movl	%esi,60(%eax)
-._keysetupdone:
-	#   eax = eax_stack
-	movl	64(%esp),%eax
-	#   ebx = ebx_stack
-	movl	68(%esp),%ebx
-	#   esi = esi_stack
-	movl	72(%esp),%esi
-	#   edi = edi_stack
-	movl	76(%esp),%edi
-	#   ebp = ebp_stack
-	movl	80(%esp),%ebp
-	# leave
-	add	%eax,%esp
-	ret
-ENDPROC(salsa20_keysetup)
-
-# enter salsa20_ivsetup
-ENTRY(salsa20_ivsetup)
-	mov	%esp,%eax
-	and	$31,%eax
-	add	$256,%eax
-	sub	%eax,%esp
-	#   eax_stack = eax
-	movl	%eax,64(%esp)
-	#   ebx_stack = ebx
-	movl	%ebx,68(%esp)
-	#   esi_stack = esi
-	movl	%esi,72(%esp)
-	#   edi_stack = edi
-	movl	%edi,76(%esp)
-	#   ebp_stack = ebp
-	movl	%ebp,80(%esp)
-	#   iv = arg2
-	movl	8(%esp,%eax),%ecx
-	#   x = arg1
-	movl	4(%esp,%eax),%eax
-	#   in6 = *(uint32 *) (iv + 0)
-	movl	0(%ecx),%edx
-	#   in7 = *(uint32 *) (iv + 4)
-	movl	4(%ecx),%ecx
-	#   in8 = 0
-	mov	$0,%ebx
-	#   in9 = 0
-	mov	$0,%esi
-	#   *(uint32 *) (x + 24) = in6
-	movl	%edx,24(%eax)
-	#   *(uint32 *) (x + 28) = in7
-	movl	%ecx,28(%eax)
-	#   *(uint32 *) (x + 32) = in8
-	movl	%ebx,32(%eax)
-	#   *(uint32 *) (x + 36) = in9
-	movl	%esi,36(%eax)
-	#   eax = eax_stack
-	movl	64(%esp),%eax
-	#   ebx = ebx_stack
-	movl	68(%esp),%ebx
-	#   esi = esi_stack
-	movl	72(%esp),%esi
-	#   edi = edi_stack
-	movl	76(%esp),%edi
-	#   ebp = ebp_stack
-	movl	80(%esp),%ebp
-	# leave
-	add	%eax,%esp
-	ret
-ENDPROC(salsa20_ivsetup)