Re: [PATCH 16/23] gpio: nomadik: support shared GPIO IRQs

[Bartosz Golaszewski]

On Mon, Feb 19, 2024 at 4:54 PM Théo Lebrun <theo.lebrun@xxxxxxxxxxx> wrote:
>
> Hello,
>
> On Mon Feb 19, 2024 at 4:48 PM CET, Bartosz Golaszewski wrote:
> > On Wed, Feb 14, 2024 at 5:24 PM Théo Lebrun <theo.lebrun@xxxxxxxxxxx> wrote:
> > >
> > > Support a single IRQs used by multiple GPIO banks. Change the IRQ
> > > handler type from a chained handler (as used by gpiolib
> > > for ->parent_handler) to a threaded IRQ.
> > >
> > > Use a fake raw spinlock to ensure generic_handle_irq() is called in a
> > > no-irq context. See Documentation/driver-api/gpio/driver.rst, "CHAINED
> > > CASCADED GPIO IRQCHIPS" for additional information.
> > >
> >
> > Any reason for not using preempt_disable()?
>
> I did what the doc recommended:
>
> > The generic_handle_irq() is expected to be called with IRQ disabled,
> > so the IRQ core will complain if it is called from an IRQ handler which is
> > forced to a thread. The "fake?" raw lock can be used to work around this
> > problem::
> >
> >   raw_spinlock_t wa_lock;
> >   static irqreturn_t omap_gpio_irq_handler(int irq, void *gpiobank)
> >       unsigned long wa_lock_flags;
> >       raw_spin_lock_irqsave(&bank->wa_lock, wa_lock_flags);
> >       generic_handle_irq(irq_find_mapping(bank->chip.irq.domain, bit));
> >       raw_spin_unlock_irqrestore(&bank->wa_lock, wa_lock_flags);
>
> If you confirm I should be using preempt_disable() that's what I'll do
> in the next revision. I could even throw in a documentation patch if
> the advice is outdated.
>
> Thanks Bartosz,

This was added 9 years ago:

commit c307b002548590c5d8c32b964831de671ad4affe
Author: Grygorii Strashko <grygorii.strashko@xxxxxx>
Date:   Tue Oct 20 17:22:15 2015 +0300

    gpio: add a real time compliance notes

    Put in a compliance checklist.

    Signed-off-by: Grygorii Strashko <grygorii.strashko@xxxxxx>
    Signed-off-by: Linus Walleij <linus.walleij@xxxxxxxxxx>

I'm Cc'ing Grygorii - maybe he can remember if there was any reason
for using a spinlock over preempt_disable(). But for now I'd go with
the latter.

Bart

>
> --
> Théo Lebrun, Bootlin
> Embedded Linux and Kernel engineering
> https://bootlin.com