[MIPS] Fix 64-bit IP checksum code

Use unsigned loads to avoid possible misscalculation of IP checksums.  This
bug was instruced in f761106cd728bcf65b7fe161b10221ee00cf7132 (lmo) /
ed99e2b (kernel.org).

[Original fix by Atsushi.  Improved instruction scheduling and fix for
unaligned unsigned load by me -- Ralf]

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Author: atsushi-nemoto

arch/mips/lib/csum_partial.S

@@ -39,12 +39,14 @@
 #ifdef USE_DOUBLE
 
 #define LOAD   ld
+#define LOAD32 lwu
 #define ADD    daddu
 #define NBYTES 8
 
 #else
 
 #define LOAD   lw
+#define LOAD32 lw
 #define ADD    addu
 #define NBYTES 4
 
@@ -60,6 +62,14 @@
 	ADD	sum, v1;					\
 	.set	pop
 
+#define ADDC32(sum,reg)						\
+	.set	push;						\
+	.set	noat;						\
+	addu	sum, reg;					\
+	sltu	v1, sum, reg;					\
+	addu	sum, v1;					\
+	.set	pop
+
 #define CSUM_BIGCHUNK1(src, offset, sum, _t0, _t1, _t2, _t3)	\
 	LOAD	_t0, (offset + UNIT(0))(src);			\
 	LOAD	_t1, (offset + UNIT(1))(src);			\
@@ -132,7 +142,7 @@ LEAF(csum_partial)
 	beqz	t8, .Lqword_align
 	 andi	t8, src, 0x8
 
-	lw	t0, 0x00(src)
+	LOAD32	t0, 0x00(src)
 	LONG_SUBU	a1, a1, 0x4
 	ADDC(sum, t0)
 	PTR_ADDU	src, src, 0x4
@@ -211,7 +221,7 @@ LEAF(csum_partial)
 	LONG_SRL	t8, t8, 0x2
 
 .Lend_words:
-	lw	t0, (src)
+	LOAD32	t0, (src)
 	LONG_SUBU	t8, t8, 0x1
 	ADDC(sum, t0)
 	.set	reorder				/* DADDI_WAR */
@@ -230,6 +240,9 @@ LEAF(csum_partial)
 	/* Still a full word to go  */
 	ulw	t1, (src)
 	PTR_ADDIU	src, 4
+#ifdef USE_DOUBLE
+	dsll	t1, t1, 32			/* clear lower 32bit */
+#endif
 	ADDC(sum, t1)
 
 1:	move	t1, zero
@@ -280,7 +293,7 @@ LEAF(csum_partial)
 1:
 	.set	reorder
 	/* Add the passed partial csum.  */
-	ADDC(sum, a2)
+	ADDC32(sum, a2)
 	jr	ra
 	.set	noreorder
 	END(csum_partial)
@@ -681,7 +694,7 @@ EXC(	sb	t0, NBYTES-2(dst), .Ls_exc)
 	.set	pop
 1:
 	.set reorder
-	ADDC(sum, psum)
+	ADDC32(sum, psum)
 	jr	ra
 	.set noreorder