Re: [PATCH v4 11/13] pinctrl: meson: Replace custom code by gpiochip_node_count() call

[Andy Shevchenko]

On Thu, Apr 14, 2022 at 09:28:30PM +0300, Andy Shevchenko wrote:
> On Thu, Apr 14, 2022 at 07:06:21PM +0300, Andy Shevchenko wrote:
> > On Thu, Apr 14, 2022 at 6:32 PM Martin Blumenstingl
> > <martin.blumenstingl@xxxxxxxxxxxxxx> wrote:
> > > On Thu, Apr 14, 2022 at 3:51 PM Andy Shevchenko
> > > <andy.shevchenko@xxxxxxxxx> wrote:
> > > [...]
> > > > > This patch landed in linux next-20220413 as commit 88834c75cae5
> > > > > ("pinctrl: meson: Replace custom code by gpiochip_node_count() call").
> > > > > Unfortunately it breaks booting of all my Amlogic-based test boards
> > > > > (Odroid C4, N2, Khadas VIM3, VIM3l). MMC driver is no longer probed and
> > > > > boards are unable to mount rootfs. Reverting this patch on top of
> > > > > linux-next fixes the issue.
> > > >
> > > > Thank you for letting me know, I'll withdraw it and investigate.
> > > If needed I can investigate further later today/tomorrow. I think the
> > > problem is that our node name doesn't follow the .dts recommendation.
> > >
> > > For GXL (arch/arm64/boot/dts/amlogic/meson-gxl.dtsi) the GPIO
> > > controller nodes are for example:
> > >   gpio: bank@4b0 {
> > >       ...
> > >   }
> > > and
> > >   gpio_ao: bank@14 {
> > >       ...
> > >   }
> > >
> > > See also:
> > > $ git grep -C6 gpio-controller arch/arm64/boot/dts/amlogic/*.dtsi
> > >
> > > Marek did not state which error he's getting but I suspect it fails
> > > with "no gpio node found".
> > 
> > Would be interesting to know that, yeah.
> > 
> > The subtle difference between the patched and unpatched version is
> > that the former uses only available nodes, it means that node is not
> > available by some reason and then the error would be the one you
> > guessed.
> 
> Looking into the difference between iterating via available nodes I have found
> nothing suspicious. Your DTSes do not have status property, so it assumes the
> node is available.
> 
> I'm quite puzzled what's going on there. Because I can't see what the logical
> difference the patch brought in.
> 
> P.S. In any case it's withdrawn now and shouldn't appear in the next Linux
> Next. But I would really appreciate more input on this.

Okay, now I got it. The "name" of the node is not the same as containing the
property with a given name. So, the faulting line of the code is this one:

gpio_np = to_of_node(device_get_named_child_node(pc->dev, "gpio-controller"));


-- 
With Best Regards,
Andy Shevchenko