Re: [PATCH 2/2] of/platform: Disable generic device linking code for PowerPC

[Rob Herring <robh+dt@kernel.org>]

On Thu, Aug 15, 2019 at 9:04 PM Saravana Kannan <saravanak@google.com> wrote:
>
> On Wed, Aug 14, 2019 at 4:41 PM Rob Herring <robh+dt@kernel.org> wrote:
> >
> > On Tue, Aug 6, 2019 at 4:04 PM Saravana Kannan <saravanak@google.com> wrote:
> > >
> > > On Tue, Aug 6, 2019 at 2:27 PM Rob Herring <robh+dt@kernel.org> wrote:
> > > >
> > > > On Tue, Aug 6, 2019 at 1:27 PM Saravana Kannan <saravanak@google.com> wrote:
> > > > >
> > > > > PowerPC platforms don't use the generic of/platform code to populate the
> > > > > devices from DT.
> > > >
> > > > Yes, they do.
> > >
> > > No they don't. My wording could be better, but they don't use
> > > of_platform_default_populate_init()
> > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/of/platform.c#n511
> >
> > Right, but the rest of the of/platform code is used (guess where it
> > got moved here from?).
> >
> > > > > Therefore the generic device linking code is never used
> > > > > in PowerPC.  Compile it out to avoid warning about unused functions.
> > > >
> > > > I'd prefer this get disabled on PPC using 'if (IS_ENABLED(CONFIG_PPC))
> > > > return' rather than #ifdefs.
> > >
> > > I'm just moving the existing ifndef some lines above. I don't want to
> > > go change existing #ifndef in this patch. Maybe that should be a
> > > separate patch series that goes and fixes all such code in drivers/of/
> > > or driver/
> >
> > So the initcall was originally just supposed to call
> > of_platform_default_populate(), but it's grown beyond that. That could
> > make things fragile as it is possible for platforms to call
> > of_platform_populate() (directly or indirectly) before
> > of_platform_default_populate_init(). That was supposed to work, but
> > now I think it's getting more fragile.
>
> Can you clarify what's wrong with of_platfrom_populate() being called
> before of_platform_default_populate_init()? If that's what a platform
> wants to do, they can do it? I have some thoughts of my own, but I
> want to hear yours.

Really, I'd like to get rid of platforms doing their own calls. That's
mostly an arm32 issue. Most of what's left are either platforms using
auxdata which was supposed to be a transition thing or ones that set a
parent device (soc_device). The former takes work to finish converting
platforms to DT and I don't know what to do for the latter other than
always or never set a parent device. Also, I know there's an issue on
atmel where we can't remove their of_platform_populate call because it
changes the probe order and breaks their pinctrl and gpio driver (I
started a patch for that...).

Rob