[PATCH] malloc: Use current (C11-style) atomics for fastbin access

[Florian Weimer]

* Anton Blanchard:

> Hi Florian,
>
>> This is another cleanup patch in preparation of the extended heap
>> protector (which will cover the fd/bk/fd_nextsize/bk_nextsize fields
>> in struct malloc_chunk, too).
>> 
>> By the way, there is an optimization opportunity for the tcache
>> backfill in _int_malloc: After the acquire MO load on the fastbin
>> list head, we can traverse the fastbin list as far as we need in
>> order to refill the tcache, and update the new list head with a
>> single CAS.  This does not have races (ABA races and the like)
>> because we have acquired the arena lock at this point.  Some backoff
>> is probably needed in case the fastbin list head is contended.  But
>> it is probably a good idea to do the first traversal at least once.
>
> I see a 16% regression on ppc64le with a simple threaded malloc test
> case. I guess the C11 atomics aren't as good as what we have in glibc.

Uh-oh.  Would you please check if replacing the two atomic_load_acquire
with atomic_load_relaxed restore the previous performance?

I believe both loads need to be acquire MO under the C11 memory model
(see the comments why), but the old code did not have them.

Thanks,
Florian