Re: [PATCH v3] gpio: interrupt consistency check for OF GPIO IRQs

[Linus Walleij]

On Mon, Sep 23, 2013 at 10:21 PM, Stephen Warren <swarren@xxxxxxxxxxxxx> wrote:
> On 09/23/2013 02:01 PM, Linus Walleij wrote:

>> And how to you block the same line from being gpio_request()ed
>> and set as output?
>
> To be honest, I really don't think this problem is terribly likely to
> occur, so I'm really not convinced that it's worth putting a lot of
> effort into solving it.

I have a different opionion, I think this is important. As GPIO
subsystem maintainer I need to be convinced of the integrity
of the system.

> If the problem does occur, it's trivial to see that this has happened by
> looking at /proc/interrupts and /sys/kernel/debug/gpio,

I basically want /sys/kernel/debug/gpio to say if a certain line
is tied for IRQ.

> That driver needs to maintain some state that indicates which of its
> IRQs have been requested. Any time a GPIO request (I mean e.g.
> set_output() not request_gpio)) comes in, the request needs to be
> validated against that IRQ usage state. If there's a conflict, deny the
> GPIO request.

I think this should be done by gpiolib, and I think it is trivial
given the right APIs. Putting it in the drivers will just create
an array of similar-look Rube Goldberg-machines and code
duplication. There is no point.

> Now, it's quite possible that the code to maintain this state and
> perform the checks will be similar/common across multiple drivers. If
> so, by all means implement the code somewhere common, and have the
> various irq_chip/gpio_chip drivers call into it.

And this is what we should do in gpiolib.

> The main thing is that all of this has to be driven/controlled/enabled
> by the gpio_chip/irq_chip driver itself, not as some global/blanket
> feature of the GPIO or IRQ subsystems.

Sure.

> Perhaps rather than having the gpio_chip/irq_chip drivers physically
> implement a function which calls this common code, they could set some
> flags/data/... in the struct gpio_chip/irq_chip indicating that they
> desire the core code that implements the error-checking to be enabled.

I think it should more be like a function they can call to flag
a GPIO as used for IRQ.

Yours,
Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/