Re: [PATCH] drm: panel: simple-panel: get the enable gpio as-is

[Thierry Reding <thierry.reding@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 3191 bytes --]

On Mon, Nov 07, 2016 at 03:26:56PM +0100, Philipp Zabel wrote:
> Am Montag, den 07.11.2016, 14:17 +0100 schrieb Thierry Reding:
> > On Mon, Nov 07, 2016 at 06:12:43PM +0800, Chen-Yu Tsai wrote:
> > > On Sun, Nov 6, 2016 at 7:09 PM, Icenowy Zheng <icenowy@aosc.xyz> wrote:
> > > > The enable gpio of simple-panel may be used by a simplefb or other
> > > > driver on the panel's display before the KMS driver get load.
> > > >
> > > > Get the GPIO as-is, so the panel won't be disabled, and the simplefb
> > > > can work.
> > > >
> > > > Signed-off-by: Icenowy Zheng <icenowy@aosc.xyz>
> > > > ---
> > > >  drivers/gpu/drm/panel/panel-simple.c | 2 +-
> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/gpu/drm/panel/panel-simple.c b/drivers/gpu/drm/panel/panel-simple.c
> > > > index 113db3c..ccee4c1 100644
> > > > --- a/drivers/gpu/drm/panel/panel-simple.c
> > > > +++ b/drivers/gpu/drm/panel/panel-simple.c
> > > > @@ -312,7 +312,7 @@ static int panel_simple_probe(struct device *dev, const struct panel_desc *desc)
> > > >                 return PTR_ERR(panel->supply);
> > > >
> > > >         panel->enable_gpio = devm_gpiod_get_optional(dev, "enable",
> > > > -                                                    GPIOD_OUT_LOW);
> > > > +                                                    GPIOD_ASIS);
> > > 
> > > The GPIO requested as-is might be in input mode. You should change the
> > > gpiod_set_value calls to gpiod_direction_output calls. The later also
> > > allows you to give an initial value. Not sure if it checks for cansleep
> > > like the set_value calls though.
> > 
> > I'd prefer not to add gpiod_direction_output() calls outside of
> > ->probe(). Instead, could we make this patch be smart about taking over
> > from an earlier user? Could it read the current direction and value of
> > the GPIO and not disable it if it had previously been enabled?
> 
> Seconded, the PWM backlight driver in drivers/video/backlight/pwm_bl.c
> already does a similar thing.
> 
> > And even if we go this extra mile there's a possibility that the GPIO
> > was just left dangling by earlier software (or hardware) and leaving it
> > on would actually be worse than turning the panel off.
> 
> Is this something the encoder driver should communicate to the panel?
> That one will know whether its atomic_reset state is enabled or
> disabled.

At that point it's already too late. Configuration of the GPIO here is
within the panel's ->probe() implementation, which is before it can ever
even get attached to an encoder/connector.

I can't think of a proper solution for this, but perhaps the best
heuristic would be to request with GPIOD_ASIS and then query the
direction. If it's configured as output we could assume that somebody's
configured it explicitly and it has the correct value.

That could still break existing use-cases, but it would at least allow
basic hand-over.

Also, if leaving the GPIO configured as-is causes glitches or other
issues, then these are observable with the current code as well, until
the panel driver takes over and disables everything.

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 801 bytes --]