Re: [PATCH v5] mm/oom_kill.c: futex: Close a race between do_exit and the oom_reaper

[Nico Pache]

On 3/21/22 18:42, Davidlohr Bueso wrote:
> On Mon, 21 Mar 2022, Michal Hocko wrote:
> 
>> The more I am thinking about this the more I am getting convinced that
>> we should rather approach this differently and skip over vmas which can
>> be holding the list. Have you considered this option?
> 
> While I tend to agree with this over a hacky trylock approach, I cannot
> help but think that v3 was the right thing to do, at least conceptually.
Yeah conceptually the V3 was the first correct patch. It could use some slight
cleanup with a wrapper like in this v5 (has_robust_list), and instead of
returning it should set MMF_OOM_SKIP.
> Robust futex users here care enough about dealing with crashes while holding
> a lock that they sacrifice the performance of regular futexes. So the OOM
> killer should not cause this very thing. I went through previous threads
> but other than the user base (which I don't think would be very large
> just because of the performance implications), was there any other reason
> to no just set MMF_OOM_SKIP upon a robust list?
We could proceed with the V3 approach; however if we are able to find a complete
solution that keeps both functionalities (Concurrent OOM Reaping & Robust Futex)
working, I dont see why we wouldnt go for it.

If we can't find a good/reliable way to check if the vma contains the robust
list then I think we should just skip the OOM like in the v3.

Cheers,
-- Nico

> 
> Thanks,
> Davidlohr
>