[PATCH] malloc: Revert fastbins to old-style atomics

[Florian Weimer]

* Siddhesh Poyarekar:

> On 18/01/19 5:46 PM, Florian Weimer wrote:
>> Siddhesh, is this okay to commit before the release?
>
> This is OK.  +1 on putting the patch back in once master reopens and get
> architecture maintainers to do an RCA.

I've pushed the revert.

I looked at this some more (including the POWER disassembly), and I
think the old code approximates what would happen with
memory_order_consume (if we had a working implementation of that) and
exploits the data dependency detection in the CPU to reduce the need for
barriers.

Thanks,
Florian