Re: [PATCH v2 3/3] memcg: no irq disable for memcg stock lock

[Alexei Starovoitov <alexei.starovoitov@gmail.com>]

On Fri, May 2, 2025 at 4:03 PM Shakeel Butt <shakeel.butt@gmail.com> wrote:
>
> On Fri, May 2, 2025 at 11:29 AM Alexei Starovoitov
> <alexei.starovoitov@gmail.com> wrote:
> >
> > On Thu, May 1, 2025 at 5:18 PM Shakeel Butt <shakeel.butt@linux.dev> wrote:
> > >
> > > There is no need to disable irqs to use memcg per-cpu stock, so let's
> > > just not do that. One consequence of this change is if the kernel while
> > > in task context has the memcg stock lock and that cpu got interrupted.
> > > The memcg charges on that cpu in the irq context will take the slow path
> > > of memcg charging. However that should be super rare and should be fine
> > > in general.
> > >
> > > Signed-off-by: Shakeel Butt <shakeel.butt@linux.dev>
> > > Acked-by: Vlastimil Babka <vbabka@suse.cz>
> > > ---
> > >  mm/memcontrol.c | 17 +++++++----------
> > >  1 file changed, 7 insertions(+), 10 deletions(-)
> > >
> > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c
> > > index cd81c70d144b..f8b9c7aa6771 100644
> > > --- a/mm/memcontrol.c
> > > +++ b/mm/memcontrol.c
> > > @@ -1858,7 +1858,6 @@ static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages,
> > >  {
> > >         struct memcg_stock_pcp *stock;
> > >         uint8_t stock_pages;
> > > -       unsigned long flags;
> > >         bool ret = false;
> > >         int i;
> > >
> > > @@ -1866,8 +1865,8 @@ static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages,
> > >                 return ret;
> > >
> > >         if (gfpflags_allow_spinning(gfp_mask))
> > > -               local_lock_irqsave(&memcg_stock.lock, flags);
> > > -       else if (!local_trylock_irqsave(&memcg_stock.lock, flags))
> > > +               local_lock(&memcg_stock.lock);
> > > +       else if (!local_trylock(&memcg_stock.lock))
> > >                 return ret;
> >
> > I don't think it works.
> > When there is a normal irq and something doing regular GFP_NOWAIT
> > allocation gfpflags_allow_spinning() will be true and
> > local_lock() will reenter and complain that lock->acquired is
> > already set... but only with lockdep on.
>
> Yes indeed. I dropped the first patch and didn't fix this one
> accordingly. I think the fix can be as simple as checking for
> in_task() here instead of gfp_mask. That should work for both RT and
> non-RT kernels.

Like:
if (in_task())
  local_lock(...);
else if (!local_trylock(...))

Most of the networking runs in bh, so it will be using
local_trylock() path which is probably ok in !PREEMPT_RT,
but will cause random performance issues in PREEMP_RT,
since rt_spin_trylock() will be randomly failing and taking
slow path of charging. It's not going to cause permanent
nginx 3x regression :), but unlucky slowdowns will be seen.
A task can grab that per-cpu rt_spin_lock and preempted
by network processing.