Re: [PATCH] mm: change memcg->oom_group access with atomic operations

From: Matthew Wilcox
Date: Tue Feb 21 2023 - 08:51:44 EST


On Mon, Feb 20, 2023 at 10:52:10PM -0800, Shakeel Butt wrote:
> On Mon, Feb 20, 2023 at 9:17 PM Roman Gushchin <roman.gushchin@xxxxxxxxx> wrote:
> > > On Feb 20, 2023, at 3:06 PM, Shakeel Butt <shakeelb@xxxxxxxxxx> wrote:
> > >
> > > On Mon, Feb 20, 2023 at 01:09:44PM -0800, Roman Gushchin wrote:
> > >>> On Mon, Feb 20, 2023 at 11:16:38PM +0800, Yue Zhao wrote:
> > >>> The knob for cgroup v2 memory controller: memory.oom.group
> > >>> will be read and written simultaneously by user space
> > >>> programs, thus we'd better change memcg->oom_group access
> > >>> with atomic operations to avoid concurrency problems.
> > >>>
> > >>> Signed-off-by: Yue Zhao <findns94@xxxxxxxxx>
> > >>
> > >> Hi Yue!
> > >>
> > >> I'm curious, have any seen any real issues which your patch is solving?
> > >> Can you, please, provide a bit more details.
> > >>
> > >
> > > IMHO such details are not needed. oom_group is being accessed
> > > concurrently and one of them can be a write access. At least
> > > READ_ONCE/WRITE_ONCE is needed here.
> >
> > Needed for what?
>
> For this particular case, documenting such an access. Though I don't
> think there are any architectures which may tear a one byte read/write
> and merging/refetching is not an issue for this.

Wouldn't a compiler be within its rights to implement a one byte store as:

load-word
modify-byte-in-word
store-word

and if this is a lockless store to a word which has an adjacent byte also
being modified by another CPU, one of those CPUs can lose its store?
And WRITE_ONCE would prevent the compiler from implementing the store
in that way.