MIPS: fix truncation in __cmpxchg_small for short values

__cmpxchg_small erroneously uses u8 for load comparison which can
be either char or short. This patch changes the local variable to
u32 which is sufficiently sized, as the loaded value is already
masked and shifted appropriately. Using an integer size avoids
any unnecessary canonicalization from use of non native widths.

This patch is part of a series that adapts the MIPS small word
atomics code for xchg and cmpxchg on short and char to RISC-V.

Cc: RISC-V Patches <patches@groups.riscv.org>
Cc: Linux RISC-V <linux-riscv@lists.infradead.org>
Cc: Linux MIPS <linux-mips@linux-mips.org>
Signed-off-by: Michael Clark <michaeljclark@mac.com>
[paul.burton@mips.com:
  - Fix varialble typo per Jonas Gorski.
  - Consolidate load variable with other declarations.]
Signed-off-by: Paul Burton <paul.burton@mips.com>
Fixes: 3ba7f44d ("MIPS: cmpxchg: Implement 1 byte & 2 byte cmpxchg()")
Cc: stable@vger.kernel.org # v4.13+

arch/mips/kernel/cmpxchg.c

@@ -54,10 +54,9 @@ unsigned long __xchg_small(volatile void *ptr, unsigned long val, unsigned int s
 unsigned long __cmpxchg_small(volatile void *ptr, unsigned long old,
 			      unsigned long new, unsigned int size)
 {
-	u32 mask, old32, new32, load32;
+	u32 mask, old32, new32, load32, load;
 	volatile u32 *ptr32;
 	unsigned int shift;
-	u8 load;
 
 	/* Check that ptr is naturally aligned */
 	WARN_ON((unsigned long)ptr & (size - 1));