Re: [Linux-kernel-mentees] [PATCH v1] gpio: ml: ioh: Convert to dev_pm_ops

[Bjorn Helgaas <helgaas@kernel.org>]

[+cc linux-pci]

On Thu, Apr 02, 2020 at 11:23:27PM +0300, Andy Shevchenko wrote:
> On Thu, Apr 2, 2020 at 11:16 PM Bjorn Helgaas <helgaas@kernel.org> wrote:
> > On Thu, Apr 02, 2020 at 09:33:46PM +0300, Andy Shevchenko wrote:
> > > On Thu, Apr 2, 2020 at 6:52 PM Vaibhav Gupta <vaibhavgupta40@gmail.com> wrote:
> > > >
> > > > Convert the legacy callback .suspend() and .resume()
> > > > to the generic ones.
> > >
> > > Thank you for the patch.
> > >
> > > Rather then doing this I think the best approach is to unify gpio-pch
> > > and gpio-ml-ioh together.
> > > Under umbrella of the task, the clean ups like above are highly
> > > appreciated.
> >
> > I'd be all in favor of that, but what Vaibhav is working toward is
> > eliminating use of legacy PM in PCI drivers.  I think unifying drivers
> > is really out of scope for that project.
> >
> > If you'd rather leave gpio-ml-ioh.c alone for now, I suggest that
> > Vaibhav move on to other PCI drivers that use legacy PM.  If we
> > convert all the others away from legacy PM and gpio-ml-ioh.c is the
> > only one remaining, then I guess we can revisit this :)
> 
> Then skip this driver for good.
> 
> > Or, maybe converting gpio-ml-ioh.c now, along the lines of
> > 226e6b866d74 ("gpio: pch: Convert to dev_pm_ops"), would be one small
> > step towards the eventual unification, by making gpio-pch and
> > gpio-ml-ioh a little more similar.
> 
> I think it will delay the real work here (very old code motivates
> better to get rid of it then semi-fixed one).

With respect, I think it is unreasonable to use the fact that
gpio-ml-ioh and gpio-pch should be unified to hold up the conversion
of gpio-ml-ioh to generic power management.

I do not want to skip gpio-ml-ioh for good, because it is one of the
few remaining drivers that use the legacy PCI PM interfaces.  We are
very close to being able to remove a significant amount of ugly code
from the PCI core.

gpio-ml-ioh and gpio-pch do look quite similar, and no doubt it would
be great to unify them.  But without datasheets or hardware to test,
that's not a trivial task, and I don't think that burden should fall
on anyone who wants to make any improvements to these drivers.

Another alternative would be to remove legacy PCI PM usage
(ioh_gpio_suspend() and ioh_gpio_resume()) from gpio-ml-ioh.  That
would mean gpio-ml-ioh wouldn't support power management at all, which
isn't a good thing, but maybe it would be even more motivation to
unify it with gpio-pch (which has already been converted by
226e6b866d74 ("gpio: pch: Convert to dev_pm_ops"))?

Bjorn