Re: [PATCH v2 07/11] gpio: Add revocable provider handle for struct gpio_chip

[Tzung-Bi Shih <tzungbi@kernel.org>]

On Thu, Feb 05, 2026 at 05:57:15PM +0100, Bartosz Golaszewski wrote:
> On Thu, Feb 5, 2026 at 9:52 AM Tzung-Bi Shih <tzungbi@kernel.org> wrote:
> >
> > On Wed, Feb 04, 2026 at 07:58:44AM -0500, Bartosz Golaszewski wrote:
> > > On Tue, 3 Feb 2026 07:10:54 +0100, Tzung-Bi Shih <tzungbi@kernel.org> said:
> > > > diff --git a/drivers/gpio/gpiolib.h b/drivers/gpio/gpiolib.h
> > > > index 3abb90385829..cd136d5b52e9 100644
> > > > --- a/drivers/gpio/gpiolib.h
> > > > +++ b/drivers/gpio/gpiolib.h
> > > > @@ -52,6 +52,7 @@
> > > >   * @device_notifier: used to notify character device wait queues about the GPIO
> > > >   *                   device being unregistered
> > > >   * @srcu: protects the pointer to the underlying GPIO chip
> > > > + * @chip_rp: revocable provider handle for the corresponding struct gpio_chip.
> > > >   * @pin_ranges: range of pins served by the GPIO driver
> > > >   *
> > > >   * This state container holds most of the runtime variable data
> > > > @@ -79,6 +80,7 @@ struct gpio_device {
> > > >     struct workqueue_struct *line_state_wq;
> > > >     struct blocking_notifier_head device_notifier;
> > > >     struct srcu_struct      srcu;
> > > > +   struct revocable_provider __rcu *chip_rp;
> > > >
> > >
> > > Why __rcu? This doesn't live in a different address space, only the internal
> > > resource it protects does. If anything - this could be __attribute__((noderef))
> > > but even that is questionable as this is an opaque structure.
> >
> > For fixing a race on the pointer itself.  See also [1].
> >
> > [1] https://lore.kernel.org/all/20260129143733.45618-2-tzungbi@kernel.org
> 
> So we're just using a double RCU here? One to protect the resource and
> another to protect the protector of the resource? I can't say I'm a
> fan of this. I really want to like this interface but is there really
> no way to hide the implementation details from the caller? Isn't this
> the whole point? As it is: the user still has to care about an
> RCU-protected pointer.

Will think about it but I have no better idea for now.

Ideally, I think the user doesn't need to interact with the RCU (even if it's
annotated with __rcu) but revocable APIs should handle it correctly.