On 03/02/2013, at 9:00 AM, Yunkai Zhang <yunkai...@gmail.com> wrote: > On Sun, Feb 3, 2013 at 8:15 PM, Igor Galić <i.ga...@brainsware.org> wrote: > >> On Sunday, February 03, 2013 10:38:37 AM z...@apache.org wrote: >>> TS-1006: Add a new wrapper: ink_atomic_decrement() >>> >>> This is a wrapper of __sync_fetch_and_sub() in GNU env. >> >> And what about non-GNU environments? >> > > I have writen a TODO list in the code:
We should investigate using C++ atomics where possible. It looks like GCC has changed the memory barrier semantics of the intrinsics over time; there's probably cases where this is hurting us ... http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51798 > > static inline void > ink_chunk_delete(InkFreeList *f, InkThreadCache *pCache, InkChunkInfo > *pChunk) > { > void *chunk_addr = pChunk->head; > > ... > > /* > * TODO: I had used ink_atomic_increment() here, but it would > * lead to incorrect value in linux OS, I don't know why: Comparing the generated code might be illuminating ... > * ink_atomic_decrement((int64_t *)&total_mem_in_byte, > -f->chunk_byte_size); > * > * So I create a new wrap, ink_atomic_decrement(), in ink_atomic.h, > * it works well. But we should create the same wrap for other OS. > */ > ink_atomic_decrement(&total_mem_in_byte, f->chunk_byte_size); > } > > I'll give new patch once I have other platforms to test and improve it. > > >> >>> I need this wrapper in reclaimable freelist, as ink_atomic_increment() >>> can't work correctly in some situation by unknown reason. >>> >>> Signed-off-by: Yunkai Zhang <qiushu....@taobao.com> >>> Signed-off-by: Zhao Yongming <ming....@gmail.com> >>> >>> >>> Project: http://git-wip-us.apache.org/repos/asf/trafficserver/repo >>> Commit: >> http://git-wip-us.apache.org/repos/asf/trafficserver/commit/4397abf7 >>> Tree: http://git-wip-us.apache.org/repos/asf/trafficserver/tree/4397abf7 >>> Diff: http://git-wip-us.apache.org/repos/asf/trafficserver/diff/4397abf7 >>> >>> Branch: refs/heads/master >>> Commit: 4397abf76c4fbb7babccc9cc11b0456f2a7a1516 >>> Parents: 5d8cc8a >>> Author: Yunkai Zhang <qiushu....@taobao.com> >>> Authored: Sat Feb 2 17:08:18 2013 +0800 >>> Committer: Zhao Yongming <ming....@gmail.com> >>> Committed: Sun Feb 3 10:52:58 2013 +0800 >>> >>> ---------------------------------------------------------------------- >>> lib/ts/ink_atomic.h | 9 ++++++++- >>> 1 files changed, 8 insertions(+), 1 deletions(-) >>> ---------------------------------------------------------------------- >>> >>> >>> >> http://git-wip-us.apache.org/repos/asf/trafficserver/blob/4397abf7/lib/ts/in >>> k_atomic.h >>> ---------------------------------------------------------------------- >> diff >>> --git a/lib/ts/ink_atomic.h b/lib/ts/ink_atomic.h >>> index b22cf24..9425eec 100644 >>> --- a/lib/ts/ink_atomic.h >>> +++ b/lib/ts/ink_atomic.h >>> @@ -160,6 +160,13 @@ ink_atomic_increment(volatile Type * mem, Amount >> count) >>> { return __sync_fetch_and_add(mem, (Type)count); >>> } >>> >>> +// ink_atomic_decrement(ptr, count) >>> +// Decrement @ptr by @count, returning the previous value. >>> +template <typename Type, typename Amount> static inline Type >>> +ink_atomic_decrement(volatile Type * mem, Amount count) { >>> + return __sync_fetch_and_sub(mem, (Type)count); >>> +} >>> + >>> // Special hacks for ARM 32-bit >>> #if defined(__arm__) && (SIZEOF_VOIDP == 4) >>> extern ProcessMutex __global_death; >>> @@ -213,6 +220,6 @@ ink_atomic_increment<int64_t>(pvint64 mem, int >> value) { >>> >>> #else /* not gcc > v4.1.2 */ >>> #error Need a compiler / libc that supports atomic operations, e.g. gcc >>> v4.1.2 or later -#endif >>> +#endif >>> >>> #endif /* _ink_atomic_h_ */ >> > > > > -- > Yunkai Zhang > Work at Taobao
