Re: list corruption in deferred_split_scan()

[Yang Shi]

On 7/18/19 5:54 PM, Qian Cai wrote:
> > On Jul 12, 2019, at 3:12 PM, Yang Shi <yang.shi@xxxxxxxxxxxxxxxxx> wrote:
> >
> >
> >
> > On 7/11/19 2:07 PM, Qian Cai wrote:
> > > On Wed, 2019-07-10 at 17:16 -0700, Yang Shi wrote:
> > > > Hi Qian,
> > > >
> > > >
> > > > Thanks for reporting the issue. But, I can't reproduce it on my machine.
> > > > Could you please share more details about your test? How often did you
> > > > run into this problem?
> > > I can almost reproduce it every time on a HPE ProLiant DL385 Gen10 server. Here
> > > is some more information.
> > >
> > > # cat .config
> > >
> > > https://raw.githubusercontent.com/cailca/linux-mm/master/x86.config
> > I tried your kernel config, but I still can't reproduce it. My compiler doesn't have retpoline support, so CONFIG_RETPOLINE is disabled in my test, but I don't think this would make any difference for this case.
> >
> > According to the bug call trace in the earlier email, it looks deferred _split_scan lost race with put_compound_page. The put_compound_page would call free_transhuge_page() which delete the page from the deferred split queue, but it may still appear on the deferred list due to some reason.
> >
> > Would you please try the below patch?
> >
> > diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> > index b7f709d..66bd9db 100644
> > --- a/mm/huge_memory.c
> > +++ b/mm/huge_memory.c
> > @@ -2765,7 +2765,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
> >          if (!mapcount && page_ref_freeze(head, 1 + extra_pins)) {
> >                  if (!list_empty(page_deferred_list(head))) {
> >                          ds_queue->split_queue_len--;
> > -                       list_del(page_deferred_list(head));
> > +                       list_del_init(page_deferred_list(head));
> >                  }
> >                  if (mapping)
> >                          __dec_node_page_state(page, NR_SHMEM_THPS);
> > @@ -2814,7 +2814,7 @@ void free_transhuge_page(struct page *page)
> >          spin_lock_irqsave(&ds_queue->split_queue_lock, flags);
> >          if (!list_empty(page_deferred_list(page))) {
> >                  ds_queue->split_queue_len--;
> > -               list_del(page_deferred_list(page));
> > +               list_del_init(page_deferred_list(page));
> >          }
> >          spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags);
> >          free_compound_page(page);
> Unfortunately, I am no longer be able to reproduce the original list corruption with todayâs linux-next.

It is because the patches have been dropped from -mm tree by Andrew due 
to this problem I guess. You have to use next-20190711, or apply the 
patches on today's linux-next.