* Johan Hovold <johan@xxxxxxxxxx> [220411 10:23]:
> On Mon, Apr 11, 2022 at 01:10:34PM +0300, Tony Lindgren wrote:
> > * Johan Hovold <johan@xxxxxxxxxx> [220411 09:54]:
> > > On Mon, Apr 11, 2022 at 12:48:04PM +0300, Tony Lindgren wrote:
> > > > diff --git a/drivers/tty/serial/8250/8250_port.c b/drivers/tty/serial/8250/8250_port.c
> > > > --- a/drivers/tty/serial/8250/8250_port.c
> > > > +++ b/drivers/tty/serial/8250/8250_port.c
> > > > @@ -1681,8 +1681,10 @@ static void serial8250_start_tx(struct uart_port *port)
> > > > return;
> > > >
> > > > if (em485 &&
> > > > - em485->active_timer == &em485->start_tx_timer)
> > > > + em485->active_timer == &em485->start_tx_timer) {
> > > > + serial8250_rpm_put_tx(up);
> > > > return;
> > > > + }
> > >
> > > I was just taking a quick look at your report about this and also
> > > noticed this return statement.
> > >
> > > The runtime PM implementation is a bit of mess as we've discussed
> > > elsewhere, but the change you propose here doesn't look right.
> >
> > Frankly "a bit of mess" applies "a bit more" than just the serial runtime
> > PM :)
>
> Heh. I'm afraid that's all too true. :)
>
> > > start_tx() can be deferred in the rs485 case, but that doesn't mean you
> > > should suspend the device here. In fact, that look like it would just
> > > break runtime PM (the parts that may work to some extent).
> >
> > AFAIK there's currently nothing paired with the serial8250_rpm_get_tx(up)
> > call at the beginning of serial8250_start_tx() for the early exit cases
> > if start_tx_rs485() or __start_tx() won't get called.
> >
> > Care to clarify a bit more what you have in mind?
>
> The problem is that that serial8250_rpm_put_tx() you're adding may
> suspend the device unconditionally (i.e. regardless of any previous
> calls to serial8250_rpm_get_tx()).
>
> If rs485 tx is just being deferred you mustn't suspend the device before
> it has had a chance to start transmitting.
Hmm I'm pretty sure rs485 has the runtime PM usage count is currently
unbalanced. To me it seems em485->start_tx_timer calls start_tx()
again from serial8250_em485_handle_start_tx().
Anyways, let's deal with the regression patch first, this can wait a bit.
Regards,
Tony
