Re: [PATCH v6 2/3] media: ov8856: Add devicetree support

From: Marco Felsch
Date: Mon May 04 2020 - 02:25:02 EST


Hi Robert,

On 20-04-30 18:11, Robert Foss wrote:
> Hey Marco,
>
> On Thu, 30 Apr 2020 at 14:07, Marco Felsch <m.felsch@xxxxxxxxxxxxxx> wrote:
> >
> > On 20-04-30 13:20, Sakari Ailus wrote:
> > > On Thu, Apr 30, 2020 at 12:11:57PM +0200, Marco Felsch wrote:
> > > > On 20-04-30 12:59, Sakari Ailus wrote:
> > > > > Hi Marco,
> > > > >
> > > > > On Thu, Apr 30, 2020 at 11:53:32AM +0200, Marco Felsch wrote:
> > > > > > Hi Sakari,
> > > > > >
> > > > > > On 20-04-30 12:45, Sakari Ailus wrote:
> > > > > > > Hi Marco,
> > > > > > >
> > > > > > > On Thu, Apr 30, 2020 at 11:35:24AM +0200, Marco Felsch wrote:
> > > >
> > > > ...
> > > >
> > > > > > > > > - if (mclk != OV8856_MCLK) {
> > > > > > > > > - dev_err(dev, "external clock %d is not supported", mclk);
> > > > > > > > > - return -EINVAL;
> > > > > > > > > + if (!is_acpi_node(fwnode)) {
> > > > > > > > > + ov8856->xvclk = devm_clk_get(dev, "xvclk");
> > > > > > > > > + if (IS_ERR(ov8856->xvclk)) {
> > > > > > > > > + dev_err(dev, "could not get xvclk clock (%pe)\n",
> > > > > > > > > + ov8856->xvclk);
> > > > > > > > > + return PTR_ERR(ov8856->xvclk);
> > > > > > > > > + }
> > > > > > > > > +
> > > > > > > > > + clk_set_rate(ov8856->xvclk, xvclk_rate);
> > > > > > > > > + xvclk_rate = clk_get_rate(ov8856->xvclk);
> > > > > > > > > }
> > > > > > > >
> > > > > > > > Why do we handle the clock only in DT case? Is there a problem with the
> > > > > > > > clock handling and ACPI?
> > > > > > >
> > > > > > > Not really, it's just that ACPI does not provide an interface to the clocks
> > > > > > > as such.
> > > > > >
> > > > > > But you will get a clk by devm_clk_get()?
> > > > >
> > > > > No, because ACPI does not expose one to drivers. Effectively the entire
> > > > > power sequences are implemented in ACPI, not in the driver.
> > > > >
> > > >
> > > > Ah okay, thanks for the explanation. I'm really not into the ACPI
> > > > stuff.. So this means the __power_off / power_on should only be done if
> > > > we are using DT's?
> > >
> > > Correct. That's why it bails out early. It could be yet earlier though,
> > > without doing anything.
> >
> > Yes I see. For easier and error less prone handling I would prefer:
> >
> > if (is_acpi_node())
> > return 0;
> >
> > as first instruction for __power_off/on().
>
> __ov8856_power_on() does make a check like that, albeit only after
> having run clk_prepare_enable() which won't do anything due to
> ov8856->xvclk==NULL. So this should be fixed and be moved to after the
> ACPI check.

Yep, I saw that. I didn't checked the clk_prepare_enable() function and
just saw that we don't request the clk for the acpi case and enable it
in both cases. This doesn't sound right to me.

> __ov8856_power_off() has no ACPI check, but all of the calls it makes
> are going to do nothing due to v8856->reset_gpio / v8856->reset_gpio /
> ov8856->xvclk all being NULL or dummies. For the sake of clarity an
> early ACPI check+return could be added, but if clarity is the goal a
> comment would work too.

Thanks god that most of the library functions taking NULL into account
=) But I think we have to take the regulator count into account. Again I
don't know how the ACPI part is working. What happens if we request
regulators which aren't listed within the ACPI table? In case of DT
there will be added dummy-regulator. If this is the case for ACPI too we
are ending in an unbalanced regulator enable/disable count since you
enable it for the DT case and disable it in both cases.

> >
> > Also I would refactor the ov8856_check_hwcfg() so the common part can be
> > used by this function and by a ov8856_parse_of() function. But thats
> > only my opinion.
>
> I'm trying to grok the above paragraph. You'd like to see something in
> the style of tc358743_probe_of()?

You don't have to if Saki is fine with the current patch. Just saying
that it would be a bit easier for the patch review.

Regards,
Marco

> >
> > Regards,
> > Marco
>