s390/atomic: get rid of gcc atomic builtins

s390 is the only architecture in the kernel which makes use of gcc's
atomic builtin functions. Even though I don't see any technical
problem with that right now, remove this code and open-code
compare-and-swap loops again, like every other architecture is doing
it also.
We can switch to a generic implementation when other architectures are
doing that also.

See also https://lwn.net/Articles/586838/ for forther details.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>

arch/s390/include/asm/atomic_ops.h

@@ -156,22 +156,46 @@ __ATOMIC64_OPS(__atomic64_xor, "xgr")
 
 static inline int __atomic_cmpxchg(int *ptr, int old, int new)
 {
-	return __sync_val_compare_and_swap(ptr, old, new);
+	asm volatile(
+		"	cs	%[old],%[new],%[ptr]"
+		: [old] "+d" (old), [ptr] "+Q" (*ptr)
+		: [new] "d" (new)
+		: "cc", "memory");
+	return old;
 }
 
-static inline int __atomic_cmpxchg_bool(int *ptr, int old, int new)
+static inline bool __atomic_cmpxchg_bool(int *ptr, int old, int new)
 {
-	return __sync_bool_compare_and_swap(ptr, old, new);
+	int old_expected = old;
+
+	asm volatile(
+		"	cs	%[old],%[new],%[ptr]"
+		: [old] "+d" (old), [ptr] "+Q" (*ptr)
+		: [new] "d" (new)
+		: "cc", "memory");
+	return old == old_expected;
 }
 
 static inline long __atomic64_cmpxchg(long *ptr, long old, long new)
 {
-	return __sync_val_compare_and_swap(ptr, old, new);
+	asm volatile(
+		"	csg	%[old],%[new],%[ptr]"
+		: [old] "+d" (old), [ptr] "+S" (*ptr)
+		: [new] "d" (new)
+		: "cc", "memory");
+	return old;
 }
 
-static inline long __atomic64_cmpxchg_bool(long *ptr, long old, long new)
+static inline bool __atomic64_cmpxchg_bool(long *ptr, long old, long new)
 {
-	return __sync_bool_compare_and_swap(ptr, old, new);
+	long old_expected = old;
+
+	asm volatile(
+		"	csg	%[old],%[new],%[ptr]"
+		: [old] "+d" (old), [ptr] "+S" (*ptr)
+		: [new] "d" (new)
+		: "cc", "memory");
+	return old == old_expected;
 }
 
 #endif /* __ARCH_S390_ATOMIC_OPS__  */