Re: [PATCH 2/4] gpiolib: use const parameters when possible

[Alexandre Courbot]

 On Wed, Nov 19, 2014 at 6:02 PM, Uwe Kleine-KÃnig
<u.kleine-koenig@xxxxxxxxxxxxxx> wrote:
> Hello Alexandre,
>
> On Wed, Nov 19, 2014 at 05:57:46PM +0900, Alexandre Courbot wrote:
>> On Wed, Nov 19, 2014 at 5:44 PM, Uwe Kleine-KÃnig
>> <u.kleine-koenig@xxxxxxxxxxxxxx> wrote:
>> > Hello Alexandre,
>> >
>> > On Wed, Nov 19, 2014 at 05:33:42PM +0900, Alexandre Courbot wrote:
>> >> On Mon, Nov 17, 2014 at 6:09 PM, Uwe Kleine-KÃnig
>> >> <u.kleine-koenig@xxxxxxxxxxxxxx> wrote:
>> >> > Hello,
>> >> >
>> >> > On Wed, Feb 13, 2013 at 04:03:00PM +0900, Alexandre Courbot wrote:
>> >> >> From: Alexandre Courbot <acourbot@xxxxxxxxxx>
>> >> >>
>> >> >> Constify descriptor parameter of gpiod_* functions for those that
>> >> >> should obviously not modify it. This includes value or direction get,
>> >> >> cansleep, and IRQ number query.
>> >> >>
>> >> >> Signed-off-by: Alexandre Courbot <acourbot@xxxxxxxxxx>
>> >> > This patch is applied as def634338d3ffb32fbe9b0a2d70cc24ef909cd4f since
>> >> > v3.9-rc2.
>> >> >
>> >> >>  drivers/gpio/gpiolib.c | 29 ++++++++++++++++-------------
>> >> >>  1 file changed, 16 insertions(+), 13 deletions(-)
>> >> >>
>> >> >> diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
>> >> >> index 8a2cf9c..ad6df6b 100644
>> >> >> --- a/drivers/gpio/gpiolib.c
>> >> >> +++ b/drivers/gpio/gpiolib.c
>> >> >> @@ -207,7 +208,7 @@ static int gpiochip_find_base(int ngpio)
>> >> >>  }
>> >> >>
>> >> >>  /* caller ensures gpio is valid and requested, chip->get_direction may sleep  */
>> >> >> -static int gpiod_get_direction(struct gpio_desc *desc)
>> >> >> +static int gpiod_get_direction(const struct gpio_desc *desc)
>> >> >>  {
>> >> >>       struct gpio_chip        *chip;
>> >> >>       unsigned                offset;
>> >> >> @@ -223,11 +224,13 @@ static int gpiod_get_direction(struct gpio_desc *desc)
>> >> >>       if (status > 0) {
>> >> >>               /* GPIOF_DIR_IN, or other positive */
>> >> >>               status = 1;
>> >> >> -             clear_bit(FLAG_IS_OUT, &desc->flags);
>> >> >> +             /* FLAG_IS_OUT is just a cache of the result of get_direction(),
>> >> >> +              * so it does not affect constness per se */
>> >> >> +             clear_bit(FLAG_IS_OUT, &((struct gpio_desc *)desc)->flags);
>> >> > I think this is worse than not having gpiod_get_direction take a const
>> >> > descriptor.
>> >>
>> >> Yeah, this kind of sucks but ultimately gpiod_get_direction() is
>> >> without side-effect to the caller and the const parameter helps
>> >> highlighting that fact.
>> > Does that mean you want to keep the const and take the risk that it
>> > might introduce some hard to debug problem when the compiler changes its
>> > optimisation strategies? I hope you don't.
>>
>> Nope, I would like to keep that const but correctness is more
>> important than prototype aesthetics. I will fix that.
> Fine.
>
>>
>> >> > I don't know an example for this particular case where this could break,
>> >> > but usually casting away a const is bad and can break in various ways.
>> >> >
>> >> > Maybe even drop the public function gpiod_get_direction completely.
>> >> > As of next-20141114 there are only two users of this function (apart
>> >> > from drivers/gpio/gpiolib*) and both are wrong (as discussed in
>> >> > http://thread.gmane.org/gmane.linux.serial/16808/focus=4783).
>> >>
>> >> Or maybe gpiod_direction_*() should register the last set direction to
>> >> be used by gpiod_get_direction() if the controller does not implement
>> >> the get_direction op? On the other hand callers can also track the
>> >> direction themselves (since they presumably set it), so this function
>> >> is really only useful for users who want to read the *hardware* state
>> >> of a GPIO. Which might still be a useful feature?
>> > I think it's useful to be able to check how a given gpio is configured
>> > in hardware. So you really want to look into the register and not cache
>> > the direction in gpiod_direction_*.
>> > And I'd say having /sys/kernel/debug/gpio is good enough. So IMHO there
>> > is no reason to hide gpiod_get_direction from general misuse.
>>
>> Mmm so are you suggesting that we should keep gpiod_get_direction() in the end?
> I'd make gpiod_get_direction static and only use it to fill in
> /sys/kernel/debug/gpio.

That's very tempting. I see only atmel_serial.c using this function,
and there is no gpio_get_direction() declared anywhere so no user of
this either. I'm not sure what I was thinking when I decided to export
it?

So yeah, if you can fix the current users, I agree with your proposal.
--
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/