On Wed, Jun 13, 2018 at 03:01:05AM +0200, Janusz Krzysztofik wrote:
> On Wednesday, June 13, 2018 12:17:24 AM CEST Dmitry Torokhov wrote:
> > On Sat, Jun 09, 2018 at 04:02:18PM +0200, Janusz Krzysztofik wrote:
> > > ...
> > > + priv->vcc = devm_regulator_get(&pdev->dev, "vcc");
> > > + if (IS_ERR(priv->vcc)) {
> > > + err = PTR_ERR(priv->vcc);
> > > + dev_err(&pdev->dev, "regulator request failed (%d)\n", err);
> > > + /* Fail softly if the regulator is not available yet */
> > > + if (err == -ENODEV)
> > > + err = -EPROBE_DEFER;
> >
> > Hmm, if regulator is not ready yet, devm_regulator_get() should be
> > returning -EPROBE_DEFER already, we should not have to convert -ENODEV
> > to -EPROBE_DEFER...
>
> Regulator is not ready because its initialization at subsys_initcall is
> deferred by not ready GPIO pin, that in turn is caused by gpio-mmio driver,
> unlike many other GPIO drivers, registered as late as at device_initcall.
>
> I agree devm_regulator_get() could return -EPROBE_DEFER in this case, but I
> can see it does that only when of_get_regulator() indicates the regulator
> should exist. In non-dt case there is apparently no way to justify if it
> should unless its consumer supply table was already in place. For that,
> registration of that table would have to be independent of successful
> registration of the regulator itself while it's not. Maybe it should, but
> that's a separate topic for a separate discussion, I think.
>
> > Is it because we have_full_constraints() returns false? You might need
> > to add call to regulator_has_full_constraints() to your board file.
>
> If have_full_constraints() returned true before the regulator or its consumer
> supply table is ready, devm_regulator_get() would happily return a dummy
> regulator and our keyboard would never get its power.
>
> I'm afraid we have to live with that return code conversion as long as the
> only user of this driver is not migrated to dt.
OK, fair enough. Can you please add a comment to that effect?
Thanks.
--
Dmitry
