Re: [RFC] gpio: expose pull-up/pull-down line flags to userspace

[Linus Walleij <linus.walleij@xxxxxxxxxx>]

On Mon, Sep 23, 2019 at 10:38 AM Bartosz Golaszewski <brgl@xxxxxxxx> wrote:

> I remember discussing it with Linus some time ago. This may not be as
> straightforward as simply adding new flags. Since PULL-UP/DOWN
> resistors can - among other parameters - also have configurable
> resistance, we'll probably need some kind of a structure for this
> ioctl() to pass any additional information to the kernel. Since we
> can't change ABI this may require adding a whole new ioctl() for
> extended configuration. This in turn has to be as future-proof as
> possible - if someone asks for user-space-configurable drive-strength,
> the new ioctl() should be ready for it.
>
> I should have some bandwidth in the coming days, so I'll try to give it a try.

What we did for the in-kernel API and the Device Tree ABI
was to simply say that if you need such elaborate control over
the line, it needs to be done with a proper pin control driver.

So for lines that just have the GPIO_PULL_UP or
GPIO_PULL_DOWN set as a (one-bit) flag, what you will
get is "typical" pull down/up (whatever the hardware default
is, or what the driver thinks is default, which should be safe
so the highest possible pull resistance I suppose).

So one option is to just go with these flags and explicitly
say that it will give a "system default (high resistance)
pull up/down".

That said, the pin controller back-end is fully capable of
accepting more elaborate configuration, so if we prefer then
we can make the more complex userspace ABI that can
set it to a desired-or-default resistance.

I lean toward simplicity here. I haven't seen that these
userspace consumers need very elaborate control of this
resistance, they are for one-off hacks and as such should
be fine with just default pull up/down I think?

I  think that specifying "this line will use pull up/down"
at request time and having the driver set a safe default
pull-up/down as response, (and pretty much what this
patch does) and then add another explicit
ioctl to refine it the day we need it is a viable way forward.

 in the future something like:
#define GPIOHANDLE_SET_LINE_CONFIG_IOCTL _IOWR(0xB4, 0x0a, struct
gpiohandle_config)

And then, when we need it, try to come up with some
really flexible ABI for the config, based on
include/linux/pinctrl/pinconf-generic.h

But no upfront code for that right now as it is not needed.
A practical usecase must come first.

Yours,
Linus Walleij