[ARM PATCH] 1110/1: fixes to the ARM checksum code

The current code is, well, broken, for both little and big endian.  The 
corner cases are not all right, and the person to blame is myself.

This patch fixes all bugs.  The resulting code was tested with all alignment 
and length combinations to test all code paths.

arch/arm/lib/csumpartialcopygeneric.S

@@ -137,19 +137,19 @@ FN_ENTRY
 
 4:		ands	len, len, #3
 		beq	.done
-		load1l	r5
+		load1l	r4
 		tst	len, #2
-		mov	r4, r5, lsr #byte(0)
+		mov	r5, r4, lsr #byte(0)
 		beq	.exit
-		adcs	sum, sum, r5, push #16
-		strb	r4, [dst], #1
-		mov	r4, r5, lsr #byte(1)
-		strb	r4, [dst], #1
-		mov	r4, r5, lsr #byte(2)
+		adcs	sum, sum, r4, push #16
+		strb	r5, [dst], #1
+		mov	r5, r4, lsr #byte(1)
+		strb	r5, [dst], #1
+		mov	r5, r4, lsr #byte(2)
 .exit:		tst	len, #1
-		strneb	r4, [dst], #1
-		andne	r4, r4, #255
-		adcnes	sum, sum, r4, lsl #byte(0)
+		strneb	r5, [dst], #1
+		andne	r5, r5, #255
+		adcnes	sum, sum, r5, lsl #byte(0)
 
 		/*
 		 * If the dst pointer was not 16-bit aligned, we
@@ -210,17 +210,17 @@ FN_ENTRY
 		orr	r4, r4, r5, push #24
 		str	r4, [dst], #4
 		adcs	sum, sum, r4
+		mov	r4, r5, pull #8
 4:		ands	len, len, #3
 		beq	.done
-		mov	r4, r5, lsr #byte(1)
+		mov	r5, r4, lsr #byte(0)
 		tst	len, #2
 		beq	.exit
-		bic	r5, r5, #0xff << byte(0)
-		adcs	sum, sum, r5, push #8
-		strb	r4, [dst], #1
-		mov	r4, r5, lsr #byte(2)
-		strb	r4, [dst], #1
-		mov	r4, r5, lsr #byte(3)
+		adcs	sum, sum, r4, push #16
+		strb	r5, [dst], #1
+		mov	r5, r4, lsr #byte(1)
+		strb	r5, [dst], #1
+		mov	r5, r4, lsr #byte(2)
 		b	.exit
 
 .src2_aligned:	mov	r4, r5, pull #16
@@ -262,18 +262,19 @@ FN_ENTRY
 		orr	r4, r4, r5, push #16
 		str	r4, [dst], #4
 		adcs	sum, sum, r4
+		mov	r4, r5, pull #16
 4:		ands	len, len, #3
 		beq	.done
-		mov	r4, r5, lsr #byte(2)
+		mov	r5, r4, lsr #byte(0)
 		tst	len, #2
 		beq	.exit
-		adcs	sum, sum, r5, pull #16
-		strb	r4, [dst], #1
-		mov	r4, r5, lsr #byte(3)
-		strb	r4, [dst], #1
+		adcs	sum, sum, r4
+		strb	r5, [dst], #1
+		mov	r5, r4, lsr #byte(1)
+		strb	r5, [dst], #1
 		tst	len, #1
 		beq	.done
-		load1b	r4
+		load1b	r5
 		b	.exit
 
 .src3_aligned:	mov	r4, r5, pull #24
@@ -315,16 +316,17 @@ FN_ENTRY
 		orr	r4, r4, r5, push #8
 		str	r4, [dst], #4
 		adcs	sum, sum, r4
+		mov	r4, r5, pull #24
 4:		ands	len, len, #3
 		beq	.done
-		mov	r4, r5, lsr #byte(3)
+		mov	r5, r4, lsr #byte(0)
 		tst	len, #2
 		beq	.exit
-		adcs	sum, sum, r5, pull #24
-		strb	r4, [dst], #1
-		load1l	r5
-		mov	r4, r5, lsr #byte(0)
-		strb	r4, [dst], #1
+		strb	r5, [dst], #1
+		adcs	sum, sum, r4
+		load1l	r4
+		mov	r5, r4, lsr #byte(0)
+		strb	r5, [dst], #1
 		adcs	sum, sum, r4, push #24
-		mov	r4, r5, lsr #byte(1)
+		mov	r5, r4, lsr #byte(1)
 		b	.exit