s390/cmpxchg: make loop condition for 1,2 byte cases precise

The cmpxchg implementation for 1 and 2 bytes consists of a 4 byte
cmpxchg loop. Currently, the decision to retry is imprecise, looping if
bits outside the target byte(s) change instead of retrying until the
target byte(s) differ from the old value.
E.g. if an attempt to exchange (prev_left_0 old_bytes prev_right_0) is
made and it fails because the word at the address is
(prev_left_1 x prev_right_1) where both x != old_bytes and one of the
prev_*_1 values differs from the respective prev_*_0 value, the cmpxchg
is retried, even if by a semantic equivalent to a normal cmpxchg, the
exchange would fail.
Instead exit the loop if x != old_bytes and retry otherwise.

Signed-off-by: Janis Schoetterl-Glausch <scgl@linux.ibm.com>
Link: https://lore.kernel.org/r/20221116144711.3811011-1-scgl@linux.ibm.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>

1 files changed, 34 insertions, 26 deletions

diff --git a/arch/s390/include/asm/cmpxchg.h b/arch/s390/include/asm/cmpxchg.h
index 1c5785b851ec..3f26416c2ad8 100644
--- a/arch/s390/include/asm/cmpxchg.h
+++ b/arch/s390/include/asm/cmpxchg.h
@@ -90,55 +90,63 @@ static __always_inline unsigned long __cmpxchg(unsigned long address,
 {
 	switch (size) {
 	case 1: {
-		unsigned int prev, tmp, shift;
+		unsigned int prev, shift, mask;
 
 		shift = (3 ^ (address & 3)) << 3;
 		address ^= address & 3;
+		old = (old & 0xff) << shift;
+		new = (new & 0xff) << shift;
+		mask = ~(0xff << shift);
 		asm volatile(
 			"	l	%[prev],%[address]\n"
-			"0:	nr	%[prev],%[mask]\n"
-			"	lr	%[tmp],%[prev]\n"
-			"	or	%[prev],%[old]\n"
-			"	or	%[tmp],%[new]\n"
-			"	cs	%[prev],%[tmp],%[address]\n"
+			"	nr	%[prev],%[mask]\n"
+			"	xilf	%[mask],0xffffffff\n"
+			"	or	%[new],%[prev]\n"
+			"	or	%[prev],%[tmp]\n"
+			"0:	lr	%[tmp],%[prev]\n"
+			"	cs	%[prev],%[new],%[address]\n"
 			"	jnl	1f\n"
 			"	xr	%[tmp],%[prev]\n"
+			"	xr	%[new],%[tmp]\n"
 			"	nr	%[tmp],%[mask]\n"
-			"	jnz	0b\n"
+			"	jz	0b\n"
 			"1:"
 			: [prev] "=&d" (prev),
-			  [tmp] "=&d" (tmp),
-			  [address] "+Q" (*(int *)address)
-			: [old] "d" ((old & 0xff) << shift),
-			  [new] "d" ((new & 0xff) << shift),
-			  [mask] "d" (~(0xff << shift))
-			: "memory", "cc");
+			  [address] "+Q" (*(int *)address),
+			  [tmp] "+&d" (old),
+			  [new] "+&d" (new),
+			  [mask] "+&d" (mask)
+			:: "memory", "cc");
 		return prev >> shift;
 	}
 	case 2: {
-		unsigned int prev, tmp, shift;
+		unsigned int prev, shift, mask;
 
 		shift = (2 ^ (address & 2)) << 3;
 		address ^= address & 2;
+		old = (old & 0xffff) << shift;
+		new = (new & 0xffff) << shift;
+		mask = ~(0xffff << shift);
 		asm volatile(
 			"	l	%[prev],%[address]\n"
-			"0:	nr	%[prev],%[mask]\n"
-			"	lr	%[tmp],%[prev]\n"
-			"	or	%[prev],%[old]\n"
-			"	or	%[tmp],%[new]\n"
-			"	cs	%[prev],%[tmp],%[address]\n"
+			"	nr	%[prev],%[mask]\n"
+			"	xilf	%[mask],0xffffffff\n"
+			"	or	%[new],%[prev]\n"
+			"	or	%[prev],%[tmp]\n"
+			"0:	lr	%[tmp],%[prev]\n"
+			"	cs	%[prev],%[new],%[address]\n"
 			"	jnl	1f\n"
 			"	xr	%[tmp],%[prev]\n"
+			"	xr	%[new],%[tmp]\n"
 			"	nr	%[tmp],%[mask]\n"
-			"	jnz	0b\n"
+			"	jz	0b\n"
 			"1:"
 			: [prev] "=&d" (prev),
-			  [tmp] "=&d" (tmp),
-			  [address] "+Q" (*(int *)address)
-			: [old] "d" ((old & 0xffff) << shift),
-			  [new] "d" ((new & 0xffff) << shift),
-			  [mask] "d" (~(0xffff << shift))
-			: "memory", "cc");
+			  [address] "+Q" (*(int *)address),
+			  [tmp] "+&d" (old),
+			  [new] "+&d" (new),
+			  [mask] "+&d" (mask)
+			:: "memory", "cc");
 		return prev >> shift;
 	}
 	case 4: {