Re: [PATCH v2] gpiolib: notify user-space about line state changes triggered by kernel

[Kent Gibson]

On Thu, Aug 24, 2023 at 11:11:18AM +0200, Bartosz Golaszewski wrote:
> On Thu, Aug 24, 2023 at 11:08 AM Kent Gibson <warthog618@xxxxxxxxx> wrote:
> >
> > On Thu, Aug 24, 2023 at 10:55:44AM +0200, Bartosz Golaszewski wrote:
> > > From: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
> > >
> > > We currently only emit CHANGED_CONFIG events when the user-space changes
> > > GPIO config. We won't be notified if changes come from in-kernel. Let's
> > > call the notifier chain whenever kernel users change direction or any of
> > > the active-low, debounce or consumer name settings. We don't notify the
> > > user-space about the persistence as the uAPI has no notion of it.
> > >
> > > Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
> > > ---
> > > v1 -> v2:
> > > - use the gpiod_line_state_notify() helper
> > > - reorder the code in gpiod_set_debounce() for better readability
> > >
> > >  drivers/gpio/gpiolib.c | 16 ++++++++++++++--
> > >  1 file changed, 14 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> > > index 40a0022ea719..1cb7731550ca 100644
> > > --- a/drivers/gpio/gpiolib.c
> > > +++ b/drivers/gpio/gpiolib.c
> > > @@ -2439,6 +2439,7 @@ int gpiod_direction_input(struct gpio_desc *desc)
> > >       }
> > >       if (ret == 0) {
> > >               clear_bit(FLAG_IS_OUT, &desc->flags);
> > > +             gpiod_line_state_notify(desc, GPIO_V2_LINE_CHANGED_CONFIG);
> > >               ret = gpio_set_bias(desc);
> > >       }
> > >
> >
> > gpiod_direction_input() is called by cdev when a line is requested.
> > So requesting a line now creates two events - REQUESTED and
> > HANGED_CONFIG? Even worse - it calls gpiod_direction_input() first, so we
> > will get CHANGED_CONFIG then REQUESTED??
> >
> > And a config change from cdev can call this and then generate a
> > CHANGED_CONFIG event itself, so again double events.
> >
> > Same for output and probably debounce too (that one is a bit more
> > convoluted).
> 
> Ah, should have tested it with user-space too...
> 
> Back to the drawing board I guess. May be the reason why we're not
> doing it in the first place yet.
> 

I think we were only looking at it from the userspace PoV, so what
constitutes a "change" is only well defined when driven from userspace.
This might be difficult to generalise without redefining what
"changed" means, or providing cdev with a separate gpiolib API so you
can tell whether the change is actually part of a cdev request, or ...

So, yeah - more thought required.

Got a use case that requires this, or just a nice to have?

Cheers,
Kent.