On 29/07/2014 at 10:00:17 +0200, Boris Brezillon wrote :

Hi Alexandre,

quoted

While this solves the particular issue Ji?? is seeing, this will not
solve the case where PA14 (CS0) is not used by the spi driver at all. It
will remained muxed as CS0 and toggle when the spi master needs to
access CS0 until another driver muxes it to something else. I still
believe we should explicitly ask pinctrl to mux them as gpios.

Do we really care about this case ?
After all, if a given pin needs a specific muxing during kernel boot
(i.e. a pin connected to a gpio-led that needs to stay in its previous
state or a pin connected to the reset line of a device that needs to
stay up and running during kernel boot) the bootloader/bootstrap should
have muxed this pin appropriately before booting the kernel.

Yeah, you are right.

What do you mean by "we should explicitly ask pinctrl to mux them as
gpios" ?
Do you mean configuring all the pins as GPIOs when the pin controller is
probed, or just adding a new pinctrl state configuring the pin as an
output GPIO and reference it in the pinctrl-0 property of the spi
controller.

If the former, you'll break devices that needs their pins to stay in
the state they were during the bootloader/boostrap phase.
The latter won't work if the pin you request as GPIO is later requested
by another device (which, if I'm correct, is exactly the case you're
trying to solve).

Again you are right, let's not care about that use case. I still feel
that the pinctrl-0 property has to be filled correctly.

-- 
Alexandre Belloni, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com