On Tue, Mar 3, 2020 at 12:33 PM Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
>
> On Tue, Mar 03, 2020 at 11:42:31AM -0800, Yang Shi wrote:
> > On Tue, Mar 3, 2020 at 10:15 AM Shakeel Butt <shakeelb@xxxxxxxxxx> wrote:
> > >
> > > On Tue, Mar 3, 2020 at 9:47 AM Yang Shi <shy828301@xxxxxxxxx> wrote:
> > > >
> > > > On Tue, Mar 3, 2020 at 2:53 AM Tetsuo Handa
> > > > <penguin-kernel@xxxxxxxxxxxxxxxxxxx> wrote:
> > > > >
> > > > > Hello, Naresh.
> > > > >
> > > > > > [ 98.003346] WARNING: CPU: 2 PID: 340 at
> > > > > > include/linux/sched/mm.h:323 alloc_page_buffers+0x210/0x288
> > > > >
> > > > > This is
> > > > >
> > > > > /**
> > > > > * memalloc_use_memcg - Starts the remote memcg charging scope.
> > > > > * @memcg: memcg to charge.
> > > > > *
> > > > > * This function marks the beginning of the remote memcg charging scope. All the
> > > > > * __GFP_ACCOUNT allocations till the end of the scope will be charged to the
> > > > > * given memcg.
> > > > > *
> > > > > * NOTE: This function is not nesting safe.
> > > > > */
> > > > > static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
> > > > > {
> > > > > WARN_ON_ONCE(current->active_memcg);
> > > > > current->active_memcg = memcg;
> > > > > }
> > > > >
> > > > > which is about memcg. Redirecting to linux-mm.
> > > >
> > > > Isn't this triggered by ("loop: use worker per cgroup instead of
> > > > kworker") in linux-next, which converted loop driver to use worker per
> > > > cgroup, so it may have multiple workers work at the mean time?
> > > >
> > > > So they may share the same "current", then it may cause kind of nested
> > > > call to memalloc_use_memcg().
> > > >
> > > > Could you please try the below debug patch? This is not the proper
> > > > fix, but it may help us narrow down the problem.
> > > >
> > > > diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h
> > > > index c49257a..1cc1cdc 100644
> > > > --- a/include/linux/sched/mm.h
> > > > +++ b/include/linux/sched/mm.h
> > > > @@ -320,6 +320,10 @@ static inline void
> > > > memalloc_nocma_restore(unsigned int flags)
> > > > */
> > > > static inline void memalloc_use_memcg(struct mem_cgroup *memcg)
> > > > {
> > > > + if ((current->flags & PF_KTHREAD) &&
> > > > + current->active_memcg)
> > > > + return;
> > > > +
> > > > WARN_ON_ONCE(current->active_memcg);
> > > > current->active_memcg = memcg;
> > > > }
> > > >
> > >
> > > Maybe it's time to make memalloc_use_memcg() nesting safe.
> >
> > Need handle the below case:
> >
> > CPU A CPU B
> > memalloc_use_memcg
> > memalloc_use_memcg
> > memalloc_unuse_memcg
> > memalloc_unuse_memcg
> >
> >
> > They may manipulate the same task->active_memcg, so CPU B may still
> > see wrong memcg, and the last call to memalloc_unuse_memcg() on CPU B
> > may not restore active_memcg to NULL. And, some code depends on
> > correct active_memcg.
>
> It's safe because it's only `current` updating a private pointer -
> nobody is changing active_memcg of a different task. And a task cannot
> run on more than one CPU simultaneously.
Yes, you are correct.
