Re: [PATCH] gpio: thunderx: fix irq_request_resources

[Tim Harvey <tharvey@xxxxxxxxxxxxx>]

On Fri, Mar 13, 2020 at 12:12 PM Robert Richter <rrichter@xxxxxxxxxxx> wrote:
>
> On 13.03.20 16:31:51, Robert Richter wrote:
> > On 11.03.20 08:43:53, Tim Harvey wrote:
> > > If there are no parent resources do not call irq_chip_request_resources_parent
> > > at all as this will return an error.
> > >
> > > This resolves a regression where devices using a thunderx gpio as an interrupt
> > > would fail probing.
> > >
> > > Fixes: 0d04d0c ("gpio: thunderx: Use the default parent apis for {request,release}_resources")
> > > Signed-off-by: Tim Harvey <tharvey@xxxxxxxxxxxxx>
> > > ---
> > >  drivers/gpio/gpio-thunderx.c | 9 ++++++---
> > >  1 file changed, 6 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/gpio/gpio-thunderx.c b/drivers/gpio/gpio-thunderx.c
> > > index 4627704..f84b9b1 100644
> > > --- a/drivers/gpio/gpio-thunderx.c
> > > +++ b/drivers/gpio/gpio-thunderx.c
> > > @@ -366,15 +366,18 @@ static int thunderx_gpio_irq_request_resources(struct irq_data *data)
> > >  {
> > >     struct thunderx_line *txline = irq_data_get_irq_chip_data(data);
> > >     struct thunderx_gpio *txgpio = txline->txgpio;
> > > +   struct irq_data *parent_data = data->parent_data;
> > >     int r;
> > >
> > >     r = gpiochip_lock_as_irq(&txgpio->chip, txline->line);
> > >     if (r)
> > >             return r;
> > >
> > > -   r = irq_chip_request_resources_parent(data);
> > > -   if (r)
> > > -           gpiochip_unlock_as_irq(&txgpio->chip, txline->line);
> > > +   if (parent_data && parent_data->chip->irq_request_resources) {
> > > +           r = irq_chip_request_resources_parent(data);
> > > +           if (r)
> > > +                   gpiochip_unlock_as_irq(&txgpio->chip, txline->line);
> > > +   }
> >
> > There is no unlocking for the else case. I would assume the
> > parent_data should be checked before grabbing the lock, or is this
> > intended?
>
> Looking at the original code, the parent resources are requested only
> if existing. So the change is ok.
>
> On the other hand, the overall change using irq_chip_{request,
> release}_resources_parent() became pointless now. It is unreadable and
> more complex now. Thus, commit 0d04d0c should just be reverted.
>
> The function interface is limited. Instead of letting the child device
> deal with the parent, parent requests should be handled directly in
> irq_request_resources(). Another aspect is that the code for this
> driver has been already removed upstream and ti_sci_inta_msi.c is the
> last remaining user of it. This speaks also for a removal by a revert.
>

Robert,

A revert does make the most sense to me and it works for 5.2, 5.3, and
5.5 but the revert fails for 5.4 and needs some manual intervention.

I'm not sure how to proceed with this.

Tim