Re: [PATCH v3 6/6] gpiolib: notify user-space about in-kernel line state changes
From: Bartosz Golaszewski
Date: Wed Oct 16 2024 - 06:12:43 EST
On Wed, Oct 16, 2024 at 11:43 AM Kent Gibson <warthog618@xxxxxxxxx> wrote:
>
> On Wed, Oct 16, 2024 at 11:22:07AM +0200, Bartosz Golaszewski wrote:
> > On Wed, Oct 16, 2024 at 11:17 AM Kent Gibson <warthog618@xxxxxxxxx> wrote:
> > >
> > > > >
> > > >
> > > > You mean, you get a CHANGED_CONFIG event but the debounce value is not
> > > > in the associated line info?
> > > >
> > >
> > > Correct.
> > >
> >
> > Ok, let me see.
> >
>
> When setting from userspace the issue is that linereq_set_config() setting the
> direction will emit, quite possibly before the debounce has been set. The
> edge_detector_setup() that does set it can also emit, though only if the
> hardware supports debounce. And then there could be a race between the
> notifier being called and the period being set in the supinfo.
> (the set will probably win that one)
>
> Debounce set from the kernel side is going to be an issue as cdev
> catches and stores the value from userspace to report in the supinfo - that
> isn't the case for kernel calls to gpiod_set_config().
>
> Seems moving the debounce value out of the desc and into cdev, which seemed a
> good idea at the time, might come back and bite now if it is no longer
> restricted to being cdev specific. Now there is an actual reason to
> store it in the desc :(.
>
I'm seeing commit:
commit 9344e34e7992fec95ce6210d95ac01437dd327ab
Author: Kent Gibson <warthog618@xxxxxxxxx>
Date: Tue Dec 19 08:41:54 2023 +0800
gpiolib: cdev: relocate debounce_period_us from struct gpio_desc
Store the debounce period for a requested line locally, rather than in
the debounce_period_us field in the gpiolib struct gpio_desc.
Add a global tree of lines containing supplemental line information
to make the debounce period available to be reported by the
GPIO_V2_GET_LINEINFO_IOCTL and the line change notifier.
Signed-off-by: Kent Gibson <warthog618@xxxxxxxxx>
Reviewed-by: Andy Shevchenko <andy@xxxxxxxxxx>
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>
But it doesn't explain *why* we did this and I don't remember the
story behind this change.
How bad would it be to go back to storing the debounce setting in the
descriptor?
Bart