Re: [PATCH] gpiolib: don't support irq sharing for gpio edge detection

[Bartosz Golaszewski <brgl@xxxxxxxx>]

czw., 8 lis 2018 o 21:30 Uwe Kleine-König
<u.kleine-koenig@xxxxxxxxxxxxxx> napisał(a):
>
> Hello Bartosz,
>
> On Thu, Nov 08, 2018 at 06:07:43PM +0100, Bartosz Golaszewski wrote:
> > czw., 13 wrz 2018 o 11:13 Linus Walleij <linus.walleij@xxxxxxxxxx> napisał(a):
> > >
> > > On Mon, Aug 20, 2018 at 8:32 AM Uwe Kleine-König
> > > <u.kleine-koenig@xxxxxxxxxxxxxx> wrote:
> > >
> > > > Trying to work out the right event if it's not sure that the examined
> > > > gpio actually moved is impossible.
> > >
> > > I applied the patch with the following reasoning.
> > >
> > > 1) Uwe has a usecase and hardware that needs this so he can do
> > > what he needs to do. I.e. the patch corresponds to a practical need
> > > of a user.
> > >
> > > 2) Only theoretical counter-arguments.
> > >
> > > 3) Uwe clarified that it is fine to put IRQ sharing back if/when
> > > we support level-triggered IRQs from userspace. This makes a
> > > lot of sense to me.
> > >
> > > 4) Rough consensus and running code.
> > >
> > > I edited the subject and commit message to reflect that we are talking
> > > about the userspace events here, not in-kernel users.
> > >
> > > Edge triggers are tricky and I feel a bit uncertain about how we handle
> > > all kinds of electronics out there. I can think about a few design
> > > mistakes.
> > >
> > > At one point I had the feeling level interrupts on GPIO lines are
> > > completely bogus, but I guess they make some sense for an
> > > intelligent peripheral (like on I2C) with it own internal status
> > > register that you then go and handle in a thread.
> > >
> >
> > Hi Linus, Uwe,
> >
> > just FYI - this patch broke existing libgpiod tests in 4.20-rc1.
> >
> > With this change, if user space requests a line for events of specific
> > type, let's say rising edge, both falling and rising edge events will
> > be emitted as rising edge.
>
> I bet this is using the gpio-mockup driver. The problem is that the
> mockup driver fires an irq for each write independent of the sensitivity
> configured by the consumer.
>

Ah right, this normally resides in the interrupt controller. I'll see
if I can modify the driver.

Bart