Hi,
On Mon, Oct 11, 2021 at 09:17:54AM +0000, Xu Yang wrote:
> Hi heikki,
>
> > -----Original Message-----
> > From: Heikki Krogerus <heikki.krogerus@xxxxxxxxxxxxxxx>
> > Sent: Monday, October 11, 2021 2:31 PM
> > To: Xu Yang <xu.yang_2@xxxxxxx>; linux@xxxxxxxxxxxx
> > Cc: Jun Li <jun.li@xxxxxxx>; gregkh@xxxxxxxxxxxxxxxxxxx; linux-
> > usb@xxxxxxxxxxxxxxx; dl-linux-imx <linux-imx@xxxxxxx>
> > Subject: [EXT] Re: [PATCH v2] usb: typec: tcpm: fix issue of multiple
> > tcpm_set_state
> >
> > Caution: EXT Email
> >
> > Guenter, can you check this?
> >
> > On Fri, Aug 27, 2021 at 07:48:09PM +0800, Xu Yang wrote:
> > > There are potential problems when states are set as following:
> > >
> > > tcpm_set_state(A, 0)
> > > tcpm_set_state(B, X)
> > >
> > > As long as the state A is set and the state_machine work is queued
> > > successfully, state_machine work will be scheduled soon after. Before
> > > running into tcpm_state_machine_work(), there is a chance to set state
> > > B again. If it does occur:
> > >
> > > either (X = 0)
> > > port->state = B and state_machine work is queued again, then work
> > > will be executed twice.
> > > or (X != 0)
> > > port->state = A and port->delayed_state = B, then work will be
> > > executed once but timer is still running.
> > >
> > > For this situation, tcpm should only handle the most recent state
> > > change as if only one state is set just now. Therefore, if the
> > > state_machine work has already been queued, it can't be queued again
> > > before running into tcpm_state_machine_work().
> > >
> > > The state_machine_running flag already prevents from queuing the work,
> > > so we can make it contain the pending stage (after work be queued and
> > > before running into tcpm_state_machine_work). The
> > > state_machine_pending_or_running flag can be used to indicate that a
> > > state can be handled without queuing the work again.
> > >
> > > Because the state_machine work has been queued for state A, there is
> > > no way to cancel as it may be already dequeued later, and then will
> > > run into
> > > tcpm_state_machine_work() certainly. To handle the delayed state B,
> > > such an abnormal work should be skiped. If port->delayed_state !=
> > > INVALID_STATE and timer is still running, it's time to skip.
> > >
> > > Fixes: 4b4e02c83167 ("typec: tcpm: Move out of staging")
> > > cc: <stable@xxxxxxxxxxxxxxx>
> > > Signed-off-by: Xu Yang <xu.yang_2@xxxxxxx>
> >
> > What changed since v1?
> >
> > thanks,
>
> In patch v1, I committed two patches to fix two problems. However, the two
> problems are resulted by the same reason. And I try to address the issues
> after it occurs.
>
> In patch v2, according to Guenter's advice, I try to avoid such problems
> occurring from the source. So I set a state_machine_pending_or_running flag
> to indicate that a state can be handled without queuing the work again.
> Meanwhile, one patch is enough to address the problems.
This did not apply on top of Greg's latest usb-next branch anymore,
so you need to rebase and resend. Don't forget the changelog :-)
thanks,
--
heikki
