On 11/14/11 19:18, Stephen Warren wrote:
> Linus Walleij wrote at Monday, November 14, 2011 2:11 AM:
>> When requesting a single GPIO pin to be muxed in, some controllers
>> will need to poke a different value into the control register
>> depending on whether the pin will be used for GPIO output or GPIO
>> input. So pass this info along for the gpio_request_enable()
>> function, we assume this is not needed for the gpio_free_disable()
>> function for the time being.
> 
> I'm not sure this API change makes sense.
> 
> Functions gpio_direction_{input,output} already exist to configure the
> direction of a GPIO, and drivers should already be using them. These have
> to work to allow drivers to toggle the direction dynamically. Requiring
> them to additionally pass this same information to the pinmux driver when
> setting up the pinmux seems like extra redundant work.
> 
> Instead, shouldn't it work like this:
> 
> * If the pinmux driver implementation behind pinmux_request_gpio() needs
> to know the direction when configuring the HW, default to input for safety;
> that will prevent the SoC driving a signal on a GPIO that's driven by some
> other device.

If the GPIO has been configured for output by boot loader
and drives a value, and now you want Linux to take control over it,
then configuring it for input will not be safe at all.
I think this kind of flexibility is necessary (although it can be
implemented in different ways).

> 
> * Rely exclusively on gpio_direction_{input,output} to allow drivers to
> configure the direction.
> 
> * If the pinmux HW needs programming in response to the gpio_direction_*
> calls, have the GPIO and pinmux driver internally communicate to achieve
> this.
> 
> Does that seem reasonable?
> 

-- 
Regards,
Igor.