Re: [sched-devel/latest] WARNING: at mm/slub.c:2443

["Pekka Enberg" <penberg@cs.helsinki.fi>]

On Wed, Apr 23, 2008 at 8:11 PM, Pete Zaitcev <zaitcev@redhat.com> wrote:
>  > --- a/drivers/usb/mon/mon_text.c
>  > +++ b/drivers/usb/mon/mon_text.c
>  > @@ -449,6 +449,7 @@ static struct mon_event_text *mon_text_read_wait(struct mon_reader_text *rp,
>  >               if (file->f_flags & O_NONBLOCK) {
>  >                       set_current_state(TASK_RUNNING);
>  >                       remove_wait_queue(&rp->wait, &waita);
>  > +                     kmem_cache_free(rp->e_slab, ep);
>  >                       return ERR_PTR(-EWOULDBLOCK);
>
>  The code looks like this:
>
>         while ((ep = mon_text_fetch(rp, mbus)) == NULL) {
>
>                 if (file->f_flags & O_NONBLOCK) {
>                         set_current_state(TASK_RUNNING);
>                         remove_wait_queue(&rp->wait, &waita);
>                         return ERR_PTR(-EWOULDBLOCK);
>
>  It's impossible to get inside the while() with non-null ep,
>  so your patch doesn't fix anything (even if kmem_cache_free
>  can survive a NULL object).

Yeah, I need a new pair of glasses. Thanks.

On Wed, Apr 23, 2008 at 8:11 PM, Pete Zaitcev <zaitcev@redhat.com> wrote:
>  Any other ideas?

Well, a race in mon_text_release() after the kmem_cache_free() loop
and the call to kmem_cache_destroy() but looks like that can't happen
either. Frans, is this easily triggerable? If so, can you try with
CONFIG_SLAB and CONFIG_DEBUG_SLAB enabled instead?