Re: [PATCH v2 3/9] irq/irq_sim: provide irq_sim_fire_type()

[Bartosz Golaszewski]

wt., 12 lut 2019 o 12:05 Uwe Kleine-KÃnig
<u.kleine-koenig@xxxxxxxxxxxxxx> napisaÅ(a):
>
> On Tue, Feb 12, 2019 at 10:27:54AM +0000, Marc Zyngier wrote:
> > On 12/02/2019 09:19, Bartosz Golaszewski wrote:
> > > When userspace wants to monitor GPIO line interrupts, the GPIO
> > > framework requests a threaded interrupt with IRQF_TRIGGER_FALLING,
> > > IRQF_TRIGGER_RISING or both. The testing module tries to act like real
> > > hardware and so if we pass only one of the *_TRIGGER_* flags, we want
> > > the simulated interrupt of corresponding type to be fired.
> >
> > Well, that's not how HW works.
>
> I cannot follow. I agree with Bartosz here. If you configure your SoC's
> irq-controller to only fire on a raising edge, you don't get an event
> when the line falls.
>
> > > Another solution - if you don't like this one -  would be to have more
> > > specialized functions: irq_sim_fire_rising() and
> > > irq_sim_fire_falling(). How about that?
> >
> > I think you're missing the point. So far, your API has been "an
> > interrupt has fired", no matter what the trigger is, and that's fine.
> > That's just modeling the output of an abstract interrupt controller into
> > whatever the irqsim is simulating.
> >
> > Now, what you're exposing is "this is how the line changed". Which is an
> > entirely different business, as you're now exposing the device output
> > line. Yes, you can model it with raising/falling, but you need at least
> > resampling for level interrupts, and actual edge detection (raising
> > followed by raising only generates a single interrupt, while
> > raising-falling-raising generates two).
>
> This matches my concern and that's why I suggested somewhere else in
> this thread to put the configuration of the sensitiveness and the actual
> tracking of the line in the same component (either irqsim or
> gpio-mockup). Given that there are only two irqsim users and the other
> one (something in iio) doesn't need that sensitiveness stuff (and I
> cannot imagine another user of irqsim with the sensitiveness support) I
> think it is best to move this to the mockup driver. That's how "normal"
> hardware drivers have to do it, too.
>

This is what v1 of this series did - it provided a function to
retrieve the type and the logic lived in gpio-mockup. Maybe we need to
get back to that solution.

Bart