gpiolib and sleeping gpios

[Ryan Mallon]

On 06/19/2010 06:21 PM, David Brownell wrote:
> 
>>
>> The runtime warnings will only show instances where the
>> non-sleeping
>> versions where called instead of the sleeping versions.
> 
> ... *AND* the GPIO requires the cansleep() version...
> 
> Right; such calls are errors.  We issue
> warnings since fault returns are inapplicable.

A driver which only uses the non-sleeping versions, but _could_ use the
cansleep variants (ie all calls to gpio_(set/get)_value are made from
contexts where it is possible to sleep) is not so easy to spot. Passing
a sleeping to gpio to such a driver will result in spurious warnings.

>> There is no
>> warning to say that we are calling the spinlock safe
>> version, where it is possible to sleep.
> 
> The call context isn't what controls whether
> gpio_get_value() or gpio_get_value_cansleep()
> is appropriate ... it's the GPIO itself, and
> how its implementation works.

No, a driver should not know anything about a gpio which is passed to
it. If a driver is able to call the cansleep variants, then it should,
and it will allow any gpio, sleeping or non-sleeping, to be used with
that driver.

If a driver uses a gpio in such a way that it cannot sleep, ie the
gpio_(get/set)_value calls are made from spinlock context, then only
gpios which do not sleep may be used with that driver.

Thats why I think specifying whether the gpio is able to sleep when it
is requested is a good idea. A driver which cannot use a sleeping gpio


> "possible to sleep" is a GPIO attribute,
> exposed by a predicate.  If spinlock-safe
> calls are used on GPIOs with that attribute,
>  a warning *IS* issued.

Possible to sleep is also an attribute of how a driver _uses_ a gpio.

>>
>> The point I was trying to make is that there are lots of
>> drivers which
>> will not work with gpios on sleeping io expanders because
>> they call the
>> spinlock safe gpio calls.
> 
> And they will trigger runtime warnings, and
> thus eventually get fixed.  The way to do that
> is to check if the GPIO needs the cansleep()
> call

Hmm, maybe this then for drivers which cannot accept sleeping gpios:

  if (gpio_cansleep(some_gpio)) {
  	  dev_err(&dev, "This driver only supports non-sleeping gpios");
	  return -EINVAL;
  }

  err = gpio_request(some_gpio, "some_gpio");

I think ideally, gpio_request should specify this via a flags argument, ie:

  #define GPIOF_NO_SLEEP	0x0
  #define GPIOF_CANSLEEP	0x1

  err = gpio_request(some_gpio, "some_gpio", GPIOF_NO_SLEEP);

~Ryan

-- 
Bluewater Systems Ltd - ARM Technology Solution Centre

Ryan Mallon         		5 Amuri Park, 404 Barbadoes St
ryan at bluewatersys.com         	PO Box 13 889, Christchurch 8013
http://www.bluewatersys.com	New Zealand
Phone: +64 3 3779127		Freecall: Australia 1800 148 751
Fax:   +64 3 3779135			  USA 1800 261 2934