Re: [PATCH] lib/test_meminit: handle kmem_cache_alloc() failure

[Tuo Li]

On Sun, Oct 11, 2026 at 7:09 AM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Fri,  9 Oct 2026 15:30:40 +0800 Tuo Li <islituo@xxxxxxxxx> wrote:
>
> > In do_kmem_cache_size(), kmem_cache_alloc() may fail and return NULL.
> > The returned pointer is subsequently passed to check_buf() and
> > fill_with_garbage_skip() without being checked, which may result in a
> > NULL pointer dereference.
> >
> > Handle the allocation failure by marking the test as failed and
> > continuing with the next iteration.
> >
> > ...
> >
> > --- a/lib/test_meminit.c
> > +++ b/lib/test_meminit.c
> > @@ -241,6 +241,10 @@ static int __init do_kmem_cache_size(size_t size, bool want_ctor,
> >               }
> >
> >               buf = kmem_cache_alloc(c, alloc_mask);
> > +             if (!buf) {
> > +                     fail = true;
> > +                     continue;
> > +             }
> >               /* Check that buf is zeroed, if it must be. */
> >               fail |= check_buf(buf, size, want_ctor, want_rcu, want_zero);
> >               fill_with_garbage_skip(buf, size, want_ctor ? CTOR_BYTES : 0);
>
> Thanks, but we have an assumption that __init-time allocations cannot
> fail and so they don't need to be checked.  Because if they do fail,
> something has gone horribly wrong and this kernel cannot work and a
> later null-deref of "buf" is a suitable way of reporting this.
>
>

Thanks for your explanation!