On Fri, Oct 31, 2014 at 5:57 PM, S?ren Brinkmann
[off-list ref] wrote:

On Fri, 2014-10-31 at 09:17AM +0100, Linus Walleij wrote:

quoted

Again it seems to be a sequencing problem. And device tree is
not good at sequences, therefore all states should be self-contained.

I agree, but how would I define a pin with pull-up enabled and
tri-state disabled - assume the pin is currently in a random state that
can have those things set/not set arbitrarily.

I was more thinking as everything you don't enable explicitly
in a state is per definition disabled.

So if you are in state A and tri-state is enabled there and you
move to state B where pull-up is enabled, then tri-state should
be disabled, since it is not explicitly enabled.

I can't put bias-disable in DT since it would potentially disable both
and the pull-up enable would have only a transient effect.

Well look at the callback from the core:

        int (*pin_config_set) (struct pinctrl_dev *pctldev,
                               unsigned pin,
                               unsigned long *configs,
                               unsigned num_configs);

You get all configs in an array. The driver can walk over the list and
make informed decisions on what to do *BEFORE* poking any registers.

Avoiding transients as you describe is part of why the callback
looks as it does. This is why every driver has its own for-loop.

I can't do the sequencing in the driver either. If I see pull-up enable,
I can't imply an effect on tri-state since I can't know whether there
was/will be a tri-state property that sets it as well.

If you define that each state is self-contained you can.

Yours,
Linus Walleij