Re: [PATCH v13 08/13] gpio: regmap: remove gpiochip before tearing down irqchip

[Bartosz Golaszewski]

On Tue, Nov 03, 2026 at 07:50:02PM +0200, Andy Shevchenko wrote:
> On Mon, Nov 02, 2026 at 09:33:20AM +0100, Michael Walle wrote:
> > On Mon Nov 2, 2026 at 7:46 AM CET, Long Zhao via B4 Relay wrote:
> > >
> > > gpio_regmap_unregister() currently calls regmap_del_irq_chip() before
> > > gpiochip_remove(). That disposes the mapped virtual IRQs and frees the
> > > irq_domain while ACPI GPIO-signaled events may still be requested and
> > > while gpiod_free_irqs() still needs gc->irq.domain.
> > 
> > Maybe Andy have an opinion here. But that deviated from the usual
> > cleanup in the reverse order and at least deserves a comment in the
> > code.
> 
> Yeah, this is unfortunate. I think to make it work as proposed the IRQ domain
> handling in GPIOLIB should be revisited. So, at least a good comment is a must,
> indeed.
> 
> > > Remove the gpio_chip first so ACPI teardown and USED_AS_IRQ cleanup run
> > > against a live domain, then delete the regmap irqchip. Use the same
> > > order on the registration error path, which previously leaked the
> > > irqchip when gpiochip_irqchip_add_domain() failed after
> > > regmap_add_irq_chip_fwnode().
> > 
> > This should be split into a seperate patch with a Fixes: tag.
> 
> Not sure if we need a separate change here as two are coupled.
> I leave it to Bart to decide.
>

Rule of thumb: can we have a smaller fix that can go in first and be
backported? If so: Split. Otherwise let's leave it as is.

Bart