[PATCH] mailbox: cancel timer before starting it

[Jassi Brar]

On Fri, Oct 16, 2020 at 1:54 PM Jerome Brunet <jbrunet at baylibre.com> wrote:
>
>
> On Fri 16 Oct 2020 at 19:33, Jassi Brar <jassisinghbrar at gmail.com> wrote:
>
> > On Fri, Oct 16, 2020 at 4:00 AM Jerome Brunet <jbrunet at baylibre.com> wrote:
> >>
> >>
> >> On Fri 16 Oct 2020 at 10:52, Ionela Voinescu <ionela.voinescu at arm.com> wrote:
> >>
> >> > On Thursday 15 Oct 2020 at 13:45:54 (-0500), Jassi Brar wrote:
> >> > [..]
> >> >> > >> --- a/drivers/mailbox/mailbox.c
> >> >> > >> +++ b/drivers/mailbox/mailbox.c
> >> >> > >> @@ -82,9 +82,13 @@ static void msg_submit(struct mbox_chan *chan)
> >> >> > >>  exit:
> >> >> > >>      spin_unlock_irqrestore(&chan->lock, flags);
> >> >> > >>
> >> >> > >> -    if (!err && (chan->txdone_method & TXDONE_BY_POLL))
> >> >> > >> -            /* kick start the timer immediately to avoid delays */
> >> >> > >> +    if (!err && (chan->txdone_method & TXDONE_BY_POLL)) {
> >> >> > >> +            /* Disable the timer if already active ... */
> >> >> > >> +            hrtimer_cancel(&chan->mbox->poll_hrt);
> >> >> > >> +
> >> >> > >> +            /* ... and kick start it immediately to avoid delays */
> >> >> > >>              hrtimer_start(&chan->mbox->poll_hrt, 0, HRTIMER_MODE_REL);
> >> >> > >> +    }
> >> >> > >>  }
> >> >> > >>
> >> >> > >>  static void tx_tick(struct mbox_chan *chan, int r)
> >> >> > >
> >> >> > > I've tracked a regression back to this commit. Details to reproduce:
> >> >> >
> >> >> > Hi Ionela,
> >> >> >
> >> >> > I don't have access to your platform and I don't get what is going on
> >> >> > from the log below.
> >> >> >
> >> >> > Could you please give us a bit more details about what is going on ?
> >> >> >
> >> >> > All this patch does is add hrtimer_cancel().
> >> >> > * It is needed if the timer had already been started, which is
> >> >> >   appropriate AFAIU
> >> >> > * It is a NO-OP is the timer is not active.
> >> >> >
> >> >> Can you please try using hrtimer_try_to_cancel() instead ?
> >> >>
> >> >
> >> > Yes, using hrtimer_try_to_cancel() instead works for me. But doesn't
> >> > this limit how effective this change is? AFAIU, this will possibly only
> >> > reduce the chances for the race condition, but not solve it.
> >> >
> >>
> >> It is also my understanding, hrtimer_try_to_cancel() would remove a
> >> timer which as not already started but would return withtout doing
> >> anything if the callback is already running ... which is the original
> >> problem
> >>
> > If we are running in the callback path, hrtimer_try_to_cancel will
> > return -1, in which case we could skip hrtimer_start.
> > Anyways, I think simply checking for hrtimer_active should effect the same.
> > I have submitted a patch, of course not tested.
>
> Yes it sloves this race but ...
>
Thanks for confirmation.

> If a race is possible between a timer callback rescheduling itself (which
> is not that uncommon) and another thread trying to cancel it
>
In our case, we should not be cancelling+restarting the timer in the
first place, because txdone_hrtimer will take care of it via
hrtimer_forward_now.

>, maybe
> there is something worth fixing in hrtimer ? Also, mailbox calls
> hrtimer_cancel() in unregister ... are we confident this would work ?
>
Yes. After unregister() every channel is supposed to die and so must
its resources.

-jassi