Re: [PATCH][GPIO] Add IRQ edge setter to gpiolib

[Drasko DRASKOVIC <drasko.draskovic@gmail.com>]

On Tue, Oct 9, 2012 at 2:00 PM, Linus Walleij <linus.walleij@linaro.org> wrote:
> On Fri, Oct 5, 2012 at 3:16 PM, Drasko DRASKOVIC
> <drasko.draskovic@gmail.com> wrote:
>> [Me]
>>> If I understand correctly the below more or less exports
>>> struct irq_chip to userspace,
>>> trying to hide it by instead exposing a property of the
>>> containing struct gpio_chip and it worries me.
>>
>> No, it should not.
>
> You are exporting all of the defines from irq.h,
> IRQ_TYPE_NONE, IRQ_TYPE_EDGE_FALLING, etc
> to userspace. These are defined in <linux/irq.h> and that file
> has this comment on top:
>
> /*
>  * Please do not include this file in generic code.  There is currently
>  * no requirement for any architecture to implement anything held
>  * within this file.
>  *
>  * Thanks. --rmk
>  */
> And that comment is even only about generic *KERNEL* code,
> userspace is way, way more than that.

Yes, this seem to be true... I was looking for something similar to
gpio_edge_store(), but it is indeed static device attribute.

Would it then be more appropriate to pass char buffer as a param, and
"descriptively" precise the edge type ("rising", "falling", "both"),
as it is now done with GPIO device from command line ?
Or there can be some more elegant way to pass this information without
the need to include <linux/irq.h> or any other kernel's internal
structs?

One approach can be to add set_irq_type fnc pointer as a member to
gpio_chip structure... Gpiolib can then call directly this callback
upon chip export.

>
>> It operates only on already exported gpiochip
>> (similar to gpio_export_link()).
>> It just helps exported GPIO be configured in "interrupt" and not in
>> "normal" mode.
>
> So can you explain exactly why userspace want to configure
> GPIO pins in interrupt mode, when there is no way whatsoever
> for userspace to handle these IRQs?

Maybe I understand something wrong, but explicit configuration of GPIO
interrupt trigger type visible in sysfs is possible __only__ from the
userspace today, and that is exactly limitation I am addressing.

The only way known to me to set-up for example GPIO_X's interrupt
trigger edge to "rising" is something like this :
> echo "rising" > /sys/class/gpio/gpioX/edge
but kernel obviously can not (should not, at least) R/W a file.

To clarify, of course that kernel module can always call internal
.set_type callback of static-to-module irq_chip. However, this kind of
set-up will not at all be visible in userspace sysfs device attribute
"edge", which can even be not aligned to real HW set-up by the module.
I.e. we can imagine that kernel module sets up IRQ edge to "rising",
and sys (after creation) will say that edge is "none" - because it has
to be explicitly changed from userspace.

It is because sysfs' gpiolib holds edge information internally (within
gpio_desc.flags, static to gpiolib.c) , and can be discorelation
between what is really set-up by the driver in the background. Usually
they are aligned, as one will set-up edge type via command line (or
userspace program), and then gpiolib updated flags and calls driver's
set_type callback.

However, when driver module sets-up edge on it's own behalf, this
change is not updated in gpiolib, and upon boot we can end up with HW
adn IRQ that has "rising" edge type, but "cat"-ing
/sys/class/gpio/gpioX/edge would give "none".


So, finally - either we pass via gpiolib to set-up sysfs visible IRQ
edge type, or give kernel update gpiolib's internal flags (vey bad).

I hope this clarifies a little my motivation of defining IRQ edge type
via sysfs from kernel.

BR,
Drasko