Re: [PATCH 15/15] kasan: use stack_depot_evict for tag-based modes

[Andrey Konovalov]

On Wed, Aug 30, 2023 at 11:38 AM Marco Elver <elver@xxxxxxxxxx> wrote:
>
> > --- a/mm/kasan/tags.c
> > +++ b/mm/kasan/tags.c
> > @@ -96,7 +96,7 @@ static void save_stack_info(struct kmem_cache *cache, void *object,
> >                       gfp_t gfp_flags, bool is_free)
> >  {
> >       unsigned long flags;
> > -     depot_stack_handle_t stack;
> > +     depot_stack_handle_t stack, old_stack;
> >       u64 pos;
> >       struct kasan_stack_ring_entry *entry;
> >       void *old_ptr;
> > @@ -120,6 +120,8 @@ static void save_stack_info(struct kmem_cache *cache, void *object,
> >       if (!try_cmpxchg(&entry->ptr, &old_ptr, STACK_RING_BUSY_PTR))
> >               goto next; /* Busy slot. */
> >
> > +     old_stack = READ_ONCE(entry->stack);
>
> Why READ_ONCE? Is it possible that there is a concurrent writer once the
> slot has been "locked" with STACK_RING_BUSY_PTR?
>
> If there is no concurrency, it would be clearer to leave it unmarked and
> add a comment to that effect. (I also think a comment would be good to
> say what the WRITE_ONCE below pair with, because at this point I've
> forgotten.)

Hm, I actually suspect we don't need these READ/WRITE_ONCE to entry
fields at all. This seems to be a leftover from the initial series
when I didn't yet have the rwlock. The rwlock prevents the entries
from being read (in kasan_complete_mode_report_info) while being
written and the try_cmpxchg prevents the same entry from being
rewritten (in the unlikely case of wrapping during writing).

Marco, do you think we can drop these READ/WRITE_ONCE?

Thanks!