Re: [PATCH] memcg: page_alloc: skip bulk allocator for __GFP_ACCOUNT

[Michal Hocko <mhocko@suse.com>]

On Thu 14-10-21 08:01:16, Shakeel Butt wrote:
> On Thu, Oct 14, 2021 at 12:16 AM Michal Hocko <mhocko@suse.com> wrote:
> >
> > On Wed 13-10-21 12:43:38, Shakeel Butt wrote:
> > [...]
> > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > > index 668edb16446a..b3acad4615d3 100644
> > > --- a/mm/page_alloc.c
> > > +++ b/mm/page_alloc.c
> > > @@ -5215,6 +5215,10 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
> > >       unsigned int alloc_flags = ALLOC_WMARK_LOW;
> > >       int nr_populated = 0, nr_account = 0;
> > >
> > > +     /* Bulk allocator does not support memcg accounting. */
> > > +     if (unlikely(gfp & __GFP_ACCOUNT))
> > > +             goto out;
> >
> > Did you mean goto failed here? This would break some which do not
> > have any fallback. E.g. xfs_buf_alloc_pages but likely more.
> >
> > Sorry I could have been more specific when talking about bypassing the
> > bulk allocator. It is quite confusing because the bulk allocator
> > interface consists of the bulk allocator and the fallback to the normal
> > page allocator.
> >
> 
> I did consider 'goto failed' here but for that I have to move
> __GFP_ACCOUNT check after the "Already populated array" check in the
> function. Basically what's the point of doing other operations
> (incrementing nr_populated) if we are gonna skip bulk anyways.

I have to say I do not follow why that is a problem.

> Regarding xfs_buf_alloc_pages(), it is not using __GFP_ACCOUNT and
> vmalloc() is the only __GFP_ACCOUNT user at this point. So, not an
> issue for now but I suppose it is better to be future-proof and do the
> 'goto failed'.

Why do we want to have that silent trap?
-- 
Michal Hocko
SUSE Labs