On Fri, May 08, 2020 at 06:00:57PM +0300, Dmitry Osipenko wrote:
> 08.05.2020 13:31, Thierry Reding пишет:
> ...
> >> Is it legal to touch DPAUX registers while DPAUX is in a suspended state?
> > 
> > DPAUX is never runtime suspended and the dependency from the I2C
> > controller on DPAUX should ensure that they are suspended and resumed in
> > the right order during system sleep.
> 
> 1. Could you please explain why DPAUX is never suspended? Isn't it a
> problem?
> 
> It looks a bit odd that driver is doing this [1][2]. RPM is supposed to
> be used for the *dynamic* power management. Should we remove RPM usage
> from the DPAUX driver?
> 
> [1]
> https://elixir.bootlin.com/linux/v5.7-rc4/source/drivers/gpu/drm/tegra/dpaux.c#L524
> [2]
> https://elixir.bootlin.com/linux/v5.7-rc4/source/drivers/gpu/drm/tegra/dpaux.c#L591

Looks more like the intention had been to eventually enable dynamic
power management but I never got around to it. The runtime PM
implementations are what's necessary to runtime suspend the device,
so all that should be needed is hook up the pm_runtime_get() and
pm_runtime_put() calls correctly.

> 2. Could you please explain why I2C driver has to care about restoring
> the pinmux state? Why pinctrl driver isn't doing that for I2C and
> everything else?

We could probably do it either way. I did it this way because the
runtime suspend/resume will get called either way, so might as well
reuse the same code paths rather than add context save/restore.

Thierry