Re: [PATCH v2 RESEND] PCI: imx6: Replace legacy gpio interface for gpiod interface

[Lucas Stach <l.stach@pengutronix.de>]

Am Montag, dem 01.11.2021 um 11:44 -0300 schrieb Maíra Canal:
> ?
> > >       /* Some boards don't have PCIe reset GPIO. */
> > > -     if (gpio_is_valid(imx6_pcie->reset_gpio)) {
> > > -             gpio_set_value_cansleep(imx6_pcie->reset_gpio,
> > > +     if (imx6_pcie->reset_gpio) {
> > > +             gpiod_set_value_cansleep(imx6_pcie->reset_gpio,
> > >                                       imx6_pcie->gpio_active_high);
> > >               msleep(100);
> > > -             gpio_set_value_cansleep(imx6_pcie->reset_gpio,
> > > +             gpiod_set_value_cansleep(imx6_pcie->reset_gpio,
> > >                                       !imx6_pcie->gpio_active_high);
> > 
> > I don't think this is correct. gpiod_set_value sets the logical line
> > state, so if the GPIO is specified as active-low in the DT, the real
> > line state will be negated. The only reason why the reset-gpio-active-
> > high property even exists is that old DTs might specify the wrong GPIO
> > polarity in the reset-gpio DT description. I think you need to use to
> > gpiod_set_raw_value API here to get the expected real line state even
> > with a broken DT description.
> > 
> > Regards,
> > Lucas
> > 
> 
> I'm a beginner in kernel development, so I'm sorry for the question.
> If I change gpiod_set_value_cansleep for gpiod_set_raw_value, wouldn't
> I change the behavior of the driver? I replaced
> gpio_set_value_cansleep for gpiod_set_value_cansleep because they have
> the same behavior and I didn't change the logic states. Thank you for
> the feedback!

Yes, you need to use the _cansleep variant of the API to keep the
context information. The point I was trying to make was that you
probably (please double check, that's just an assumption on my side)
need to use the _raw variant of the gpiod API to keep the current
behavior of the driver, as we are setting the physical line state
purely depending on the reset-gpio-active-high property presence, not
the logical line state, which would take into account the polarity
specified in the DT gpio descriptor.

I guess the right API call here would be
gpiod_set_raw_value_cansleep().

Regards,
Lucas