Linux try_cmpxchg mysterious inline assembly - linux

Question

I need some help understanding Linux's `try_cmpxchg semantics and implementation. In the kernel source, it is implemented as:
#define __raw_try_cmpxchg(_ptr, _pold, _new, size, lock) \
({ \
bool success; \
__typeof__(_ptr) _old = (_pold); \
__typeof__(*(_ptr)) __old = *_old; \
__typeof__(*(_ptr)) __new = (_new); \
switch (size) { \
case __X86_CASE_B: \
{ \
volatile u8 *__ptr = (volatile u8 *)(_ptr); \
asm volatile(lock "cmpxchgb %[new], %[ptr]" \
CC_SET(z) \
: CC_OUT(z) (success), \
[ptr] "+m" (*__ptr), \
[old] "+a" (__old) \
: [new] "q" (__new) \
: "memory"); \
break; \
} \
case __X86_CASE_W: \
{ \
volatile u16 *__ptr = (volatile u16 *)(_ptr); \
asm volatile(lock "cmpxchgw %[new], %[ptr]" \
CC_SET(z) \
: CC_OUT(z) (success), \
[ptr] "+m" (*__ptr), \
[old] "+a" (__old) \
: [new] "r" (__new) \
: "memory"); \
break; \
} \
case __X86_CASE_L: \
{ \
volatile u32 *__ptr = (volatile u32 *)(_ptr); \
asm volatile(lock "cmpxchgl %[new], %[ptr]" \
CC_SET(z) \
: CC_OUT(z) (success), \
[ptr] "+m" (*__ptr), \
[old] "+a" (__old) \
: [new] "r" (__new) \
: "memory"); \
break; \
} \
case __X86_CASE_Q: \
{ \
volatile u64 *__ptr = (volatile u64 *)(_ptr); \
asm volatile(lock "cmpxchgq %[new], %[ptr]" \
CC_SET(z) \
: CC_OUT(z) (success), \
[ptr] "+m" (*__ptr), \
[old] "+a" (__old) \
: [new] "r" (__new) \
: "memory"); \
break; \
} \
default: \
__cmpxchg_wrong_size(); \
} \
if (unlikely(!success)) \
*_old = __old; \
likely(success); \
})
#define __try_cmpxchg(ptr, pold, new, size) \
__raw_try_cmpxchg((ptr), (pold), (new), (size), LOCK_PREFIX)
#define try_cmpxchg(ptr, pold, new) \
__try_cmpxchg((ptr), (pold), (new), sizeof(*(ptr)))
I am curious what those CC_SET and CC_OUT means. They are defined as:
/*
* Macros to generate condition code outputs from inline assembly,
* The output operand must be type "bool".
*/
#ifdef __GCC_ASM_FLAG_OUTPUTS__
# define CC_SET(c) "\n\t/* output condition code " #c "*/\n"
# define CC_OUT(c) "=#cc" #c
#else
# define CC_SET(c) "\n\tset" #c " %[_cc_" #c "]\n"
# define CC_OUT(c) [_cc_ ## c] "=qm"
#endif
Also, it would be great if you can explain the exact semantics of try_cmpxchg (not quite understand how can a atomic cmpxchg fail...)

Answer 1

Newer versions of gcc (I believe from version 6) support specific flag outputs. The macros are there to use this support if available, else fall back to the old way by doing a setCC instruction and a temporary output.
As to how cmpxchg can "fail": it does a compare so it fails if that compare fails, in which case the destination is unchanged and the current value is fetched from memory. Consult an instruction set reference for the details.