Re: [PATCH] PM-runtime: fix deadlock with ktime

[Vincent Guittot]

On Wed, 30 Jan 2019 at 10:39, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
>
> On Wed, Jan 30, 2019 at 10:14 AM Vincent Guittot
> <vincent.guittot@xxxxxxxxxx> wrote:
> >
> > Hi Geert,
> >
> > On Wed, 30 Jan 2019 at 09:21, Geert Uytterhoeven <geert@xxxxxxxxxxxxxx> wrote:
> > >
> > > Hi Vincent,
> > >
> > > On Wed, Jan 30, 2019 at 9:16 AM Vincent Guittot
> > > <vincent.guittot@xxxxxxxxxx> wrote:
> > > > A deadlock has been seen when swicthing clocksources which use PM runtime.
> > > > The call path is:
> > > > change_clocksource
> > > >     ...
> > > >     write_seqcount_begin
> > > >     ...
> > > >     timekeeping_update
> > > >         ...
> > > >         sh_cmt_clocksource_enable
> > > >             ...
> > > >             rpm_resume
> > > >                 pm_runtime_mark_last_busy
> > > >                     ktime_get
> > > >                         do
> > > >                             read_seqcount_begin
> > > >                         while read_seqcount_retry
> > > >     ....
> > > >     write_seqcount_end
> > > >
> > > > Although we should be safe because we haven't yet changed the clocksource
> > > > at that time, we can't because of seqcount protection.
> > > >
> > > > Use ktime_get_mono_fast_ns instead which is lock safe for such case
> > > >
> > > > Fixes: 8234f6734c5d ("PM-runtime: Switch autosuspend over to using hrtimers")
> > > > Reported-by: Biju Das <biju.das@xxxxxxxxxxxxxx>
> > > > Signed-off-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
> > >
> > > Thanks for your patch!
> > >
> > > /**
> > >  * ktime_get_mono_fast_ns - Fast NMI safe access to clock monotonic
> > >  *
> > >  * This timestamp is not guaranteed to be monotonic across an update.
> > >  * The timestamp is calculated by:
> > >  *
> > >  *      now = base_mono + clock_delta * slope
> > >  *
> > >  * So if the update lowers the slope, readers who are forced to the
> > >  * not yet updated second array are still using the old steeper slope.
> > >  *
> > >  * tmono
> > >  * ^
> > >  * |    o  n
> > >  * |   o n
> > >  * |  u
> > >  * | o
> > >  * |o
> > >  * |12345678---> reader order
> > >  *
> > >  * o = old slope
> > >  * u = update
> > >  * n = new slope
> > >  *
> > >  * So reader 6 will observe time going backwards versus reader 5.
> > >  *
> > >  * While other CPUs are likely to be able observe that, the only way
> > >  * for a CPU local observation is when an NMI hits in the middle of
> > >  * the update. Timestamps taken from that NMI context might be ahead
> > >  * of the following timestamps. Callers need to be aware of that and
> > >  * deal with it.
> > >  */
> > >
> > > As this function is not guaranteed to be monotonic, have you checked how
> > > the Runtime PM code behaves if time goes backwards? Does it just make
> > > a suboptimal decision or does it crash?
> >
> > As a worst case this will generate a suboptimal decision around the update
>
> So that should be explained in the changelog of the patch.  In detail,
> if poss, please.

Ok, I'm going to update the commit message