On Mon, Sep 9, 2019 at 6:59 PM Agustín DallʼAlba
<agustin@xxxxxxxxxxxxxxx> wrote:
>
> On Mon, 2019-09-09 at 14:44 +0200, Vitaly Wool wrote:
> > On Mon, Sep 9, 2019 at 9:17 AM Vitaly Wool <vitalywool@xxxxxxxxx> wrote:
> > > On Mon, Sep 9, 2019 at 2:14 AM Agustín DallʼAlba <agustin@xxxxxxxxxxxxxxx> wrote:
> > > > However trace 2 (__zswap_pool_release blocked for more than xxxx
> > > > seconds) still happens.
> > >
> > > That one is pretty new and seems to have been caused by
> > > d776aaa9895eb6eb770908e899cb7f5bd5025b3c ("mm/z3fold.c: fix race
> > > between migration and destruction").
> > > I'm looking into this now and CC'ing Henry just in case.
> >
> > Agustin, could you please try reverting that commit? I don't think
> > it's working as it should.
> >
> > > ~Vitaly
>
> I reverted the commit and I haven't seen the deadlock happen again.
> Should I be experiencing some side effect of having reverted that?
>
> Thanks,
> Agustín
>
Hi Agustin,
Having reverted the commit you should only experience problems if you
are changing the zswap backend at runtime. There was a race condition
where a migrating page could avoid destruction and eventually crash
the kernel.
Best,
Henry
