On Sat, Jan 24, 2026 at 05:52:53PM +0100, Johan Hovold wrote:
> On Fri, Jan 16, 2026 at 08:10:35AM +0000, Tzung-Bi Shih wrote:
> > There are independent lifecycle instances (e.g., other drivers) can save
> > a raw pointer to the struct gpio_device (e.g., via gpio_device_find())
> > or struct gpio_desc (e.g., via gpio_to_desc()).  In some operations,
> > they have to access the underlying struct gpio_chip.
> > 
> > Leverage revocable for them so that they don't need to handle the
> > synchronization by accessing the SRCU explicitly.
> > 
> > Signed-off-by: Tzung-Bi Shih <[email protected]>
> 
> >  static int gpiod_get_raw_value_commit(const struct gpio_desc *desc)
> >  {
> > -   struct gpio_device *gdev;
> >     struct gpio_chip *gc;
> >     int value;
> > +   DEFINE_REVOCABLE(rev, desc->gdev->chip_rp);
> 
> DEFINE_REVOCABLE() is racy and can lead to use-after-free since nothing
> prevents chip_rp from being revoked and freed while the
> revocable_alloc() hidden in DEFINE_REVOCABLE() is running.

This was supposed to say "revocable_init()" (i.e. revocable_alloc()
without the memory allocation).
 
> >  
> > -   /* FIXME Unable to use gpio_chip_guard due to const desc. */
> > -   gdev = desc->gdev;
> > -
> > -   guard(srcu)(&gdev->srcu);
> > -
> > -   gc = srcu_dereference(gdev->chip, &gdev->srcu);
> > +   REVOCABLE_TRY_ACCESS_WITH(rev, gc);
> >     if (!gc)
> >             return -ENODEV;
 
Johan

Reply via email to