Re: ChipIdea USB regression

From: Rob Herring
Date: Tue Jan 18 2022 - 09:40:15 EST


On Mon, Jan 17, 2022 at 3:56 AM Charles Keepax
<ckeepax@xxxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Mon, Jan 17, 2022 at 09:26:56AM +0000, Charles Keepax wrote:
> > On Sat, Jan 15, 2022 at 09:55:23AM -0600, Rob Herring wrote:
> > > On Fri, Jan 14, 2022 at 5:18 AM Charles Keepax
> > > <ckeepax@xxxxxxxxxxxxxxxxxxxxx> wrote:
> > > > On Fri, Jan 14, 2022 at 10:56:20AM +0000, Charles Keepax wrote:
> > > > So when that patch copies the DT node to the new platform device
> > > > in ci_hdrc_add_device it copies the compatible stuff as well as
> > > > the IRQ stuff it was targeting, this presumably causes the kernel
> > > > to bind a new copy of the driver to that new device, which probes
> > > > and calls ci_hdrc_add_device again repeating the process until
> > > > it dies.
> > > >
> > > > Kinda looks to me like the best solution might just be to revert
> > > > the patch, I am not sure I see how that copy of the DT is supposed
> > > > to work?
> > >
> > > It's not copying the DT, but yes AFAICT it does match and bind the
> > > child device on the parent driver using the compatible match instead
> > > of matching on driver name. I think we can use the of_reuse_node flag
> > > to avoid this in the match, but that needs some more investigation.
> >
> > Assuming you mean the of_node_reused flag, looks like it already
> > being set, your code does this:
> >
> > @@ -864,6 +864,7 @@ struct platform_device *ci_hdrc_add_device(struct device *dev,
> > pdev->dev.parent = dev;
> > + device_set_of_node_from_dev(&pdev->dev, dev);
> >
> > And that function does this:
> >
> > void device_set_of_node_from_dev(struct device *dev, const struct device *dev2)
> > {
> > of_node_put(dev->of_node);
> > dev->of_node = of_node_get(dev2->of_node);
> > dev->of_node_reused = true;
> > }
> > EXPORT_SYMBOL_GPL(device_set_of_node_from_dev);
> >
> > I guess maybe that flag doesn't do what it is supposed to for
> > some reason?
> >
>
> Ah ok it seems that flag is only currently used by the pinctrl
> subsystem, didn't realise that was quite so new and not used
> anywhere. I guess we probably need to add something to the
> platform device code to use that flag too, if that is the way we
> want to run with this.

I pushed a patch[1] for kernel-ci to test if you want to give it a try, too.

Rob

[1] https://git.kernel.org/pub/scm/linux/kernel/git/robh/linux.git/log/?h=for-kernelci