Re: [v2 2/3] media: ov8856: Add devicetree support

[Sakari Ailus]

Hi Robert,

On Fri, Mar 27, 2020 at 11:32:29AM +0100, Robert Foss wrote:
> On Thu, 26 Mar 2020 at 15:47, Sakari Ailus <sakari.ailus@xxxxxx> wrote:
> >
> > Hi Robert,
> >
> > On Thu, Mar 26, 2020 at 12:56:37PM +0100, Robert Foss wrote:
> > ...
> > > > > +static int __ov8856_power_on(struct ov8856 *ov8856)
> > > > > +{
> > > > > +     struct i2c_client *client = v4l2_get_subdevdata(&ov8856->sd);
> > > > > +     int ret;
> > > > > +
> > > > > +     ret = clk_prepare_enable(ov8856->xvclk);
> > > > > +     if (ret < 0) {
> > > > > +             dev_err(&client->dev, "failed to enable xvclk\n");
> > > > > +             return ret;
> > > > > +     }
> > > > > +
> > > > > +     gpiod_set_value_cansleep(ov8856->reset_gpio, GPIOD_OUT_HIGH);
> > > > > +
> > > > > +     ret = regulator_bulk_enable(ARRAY_SIZE(ov8856_supply_names),
> > > > > +                                 ov8856->supplies);
> > > > > +     if (ret < 0) {
> > > > > +             dev_err(&client->dev, "failed to enable regulators\n");
> > > > > +             goto disable_clk;
> > > > > +     }
> > > > > +
> > > > > +     gpiod_set_value_cansleep(ov8856->reset_gpio, GPIOD_OUT_LOW);
> > > > > +
> > > > > +     usleep_range(1500, 1800);
> > > >
> > > > I think you could omit the delay on ACPI based systems. Or just bail out
> > > > early in that case.
> > >
> > > I'll add a check for reset_gpio being NULL, and skip the sleep for that case.
> >
> > There could also be a regulator but no GPIO.
> >
> > I think if you don't have either, then certainly there's no need for a
> > delay.
> 
> Removing the delay if no action is taken makes sense, but I'm not sure
> how best to do it.
> If there are no regulators dummy ones are created automatically, which
> makes distinguishing between a little bit cumbersome. The regulator
> structs could of course all be inspected, and if all are dummy ones,
> the delay could be skipped. But is there a neater way of doing this?
> Manually inspecting the regs strikes me as a bit inelegant.

I guess the cleanest, easy way to make this right, albeit slightly
unoptimal in very rare cases where you have none of the above resources in
a DT system, is to bail out if you're running on an ACPI based system.

I.e. checking for e.g. is_acpi_node(dev->fwnode).

-- 
Sakari Ailus