Re: [PATCH] workqueue: Fix memory ordering race in queue_work*()

From: Hector Martin
Date: Tue Aug 16 2022 - 13:22:03 EST


On 17/08/2022 01.26, Tejun Heo wrote:
> Hello,
>
> On Tue, Aug 16, 2022 at 12:15:38PM +0800, Herbert Xu wrote:
>> Tejun Heo <tj@xxxxxxxxxx> wrote:
>>>
>>> Oh, tricky one and yeah you're absolutely right that it makes no sense to
>>> not guarantee barrier semantics when already pending. I didn't even know
>>> test_and_set_bit() wasn't a barrier when it failed. Thanks a lot for hunting
>>> down and fixing this. Applied to wq/for-6.0-fixes.
>>
>> Please revert this as test_and_set_bit was always supposed to be
>> a full memory barrier. This is an arch bug.
>
> Alright, reverting.

FYI, Linus applied the alternate fix directly to his tree already, so
this is fixed on M1 either way. However, you may want to pay attention
to the discussion about the subtlety of the workqueue code pairing
test_and_set_bit() not with clear_bit(), but rather atomic_long_set(),
since it *could* imply it is still broken or might be broken in the
future, on other architectures.

- Hector