Re: [PATCH v2] kfence: skip DMA allocations

[Alexander Potapenko <glider@google.com>]

On Wed, Jun 30, 2021 at 9:02 AM Souptick Joarder <jrdr.linux@gmail.com> wrote:
>
> On Tue, Jun 29, 2021 at 9:47 PM Alexander Potapenko <glider@google.com> wrote:
> >
> > Allocation requests with __GFP_DMA/__GFP_DMA32 or
> > SLAB_CACHE_DMA/SLAB_CACHE_DMA32 cannot be fulfilled by KFENCE, because
> > they must reside in low memory, whereas KFENCE memory pool is located in
> > high memory.
> >
> > Skip such allocations to avoid crashes where DMAable memory is expected.
> >
> > Fixes: 0ce20dd84089 ("mm: add Kernel Electric-Fence infrastructure")
> > Cc: Andrew Morton <akpm@linux-foundation.org>
> > Cc: Dmitry Vyukov <dvyukov@google.com>
> > Cc: Marco Elver <elver@google.com>
> > Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > Cc: stable@vger.kernel.org # 5.12+
> > Signed-off-by: Alexander Potapenko <glider@google.com>
> >
> > ---
> >
> > v2:
> >  - added parentheses around the GFP clause, as requested by Marco
> > ---
> >  mm/kfence/core.c | 8 ++++++++
> >  1 file changed, 8 insertions(+)
> >
> > diff --git a/mm/kfence/core.c b/mm/kfence/core.c
> > index 4d21ac44d5d35..f7ce3d876bc9e 100644
> > --- a/mm/kfence/core.c
> > +++ b/mm/kfence/core.c
> > @@ -760,6 +760,14 @@ void *__kfence_alloc(struct kmem_cache *s, size_t size, gfp_t flags)
> >         if (size > PAGE_SIZE)
> >                 return NULL;
> >
> > +       /*
> > +        * Skip DMA allocations. These must reside in the low memory, which we
> > +        * cannot guarantee.
> > +        */
> > +       if ((flags & (__GFP_DMA | __GFP_DMA32)) ||
> > +           (s->flags & (SLAB_CACHE_DMA | SLAB_CACHE_DMA32)))
> > +               return NULL;
> > +
>
> I prefer to move this check at the top of the function.
> Although it won't make much difference except avoiding atomic operations
> in case this condition is true.

Agreed, we probably shouldn't be expecting a constant flow of
allocations from these zones that will be slowed down by this check.
On a related note, Marco suggested moving the PAGE_SIZE check to the
top of the function as well.

It will also make sense to check for GFP_ZONEMASK instead of just GFP DMA flags.
I couldn't see anyone passing e.g. __GFP_HIGHMEM or __GFP_MOVABLE to
kmem_cache_alloc(), but according to mm/slab.c it is possible, so just
to be on the safe side we'd better ignore them as well.

> >         return kfence_guarded_alloc(s, size, flags);
> >  }
> >
> > --
> > 2.32.0.93.g670b81a890-goog
> >
> >



-- 
Alexander Potapenko
Software Engineer

Google Germany GmbH
Erika-Mann-Straße, 33
80636 München

Geschäftsführer: Paul Manicle, Halimah DeLaine Prado
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg