The implementation of cmpxchg64() for the ARM v6 and v7 architecture
casts parameter 2 and 3 (the old and new 64bit values) to an unsigned
long before calling the atomic_cmpxchg64() function. This clears
the top 32 bits of the old and new values, resulting in the wrong
values being compare-exchanged. Luckily, this only appears to be used
for 64-bit sched_clock, which we don't (yet) have on ARM.
This bug was introduced by commit
3e0f5a15f500 ("ARM: 7404/1: cmpxchg64:
use atomic64 and local64 routines for cmpxchg64").
Cc: <stable@vger.kernel.org>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Jaccon Bastiaansen <jaccon.bastiaansen@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
((__typeof__(*(ptr)))atomic64_cmpxchg(container_of((ptr), \
atomic64_t, \
counter), \
- (unsigned long)(o), \
- (unsigned long)(n)))
+ (unsigned long long)(o), \
+ (unsigned long long)(n)))
#define cmpxchg64_local(ptr, o, n) \
((__typeof__(*(ptr)))local64_cmpxchg(container_of((ptr), \
local64_t, \
a), \
- (unsigned long)(o), \
- (unsigned long)(n)))
+ (unsigned long long)(o), \
+ (unsigned long long)(n)))
#endif /* __LINUX_ARM_ARCH__ >= 6 */