Re: [PATCH] spi/atmel: add support for runtime PM

[Kevin Hilman <khilman@kernel.org>]

"Yang, Wenyou" <Wenyou.Yang@atmel.com> writes:

>> -----Original Message-----
>> From: Kevin Hilman [mailto:khilman@kernel.org]
>> Sent: Friday, October 17, 2014 10:22 PM
>> To: Mark Brown
>> Cc: Yang, Wenyou; linux-spi@vger.kernel.org; linux-kernel@vger.kernel.org; Ferre,
>> Nicolas; linux-arm-kernel@lists.infradead.org
>> Subject: Re: [PATCH] spi/atmel: add support for runtime PM
>> 
>> Mark Brown <broonie@kernel.org> writes:
>> 
>> > On Fri, Oct 17, 2014 at 06:02:35AM -0700, Kevin Hilman wrote:
>> >> Wenyou Yang <wenyou.yang@atmel.com> writes:
>> >
>> >> > +	if (!pm_runtime_suspended(dev)) {
>> >> > +		clk_disable_unprepare(as->clk);
>> >> > +		pinctrl_pm_select_sleep_state(dev);
>> >> > +	}
>> >
>> >> a.k.a. pm_runtime_put_sync() since the ->runtime_suspend() callback
>> >> does the same thing.
>> >
>> > Will that do the right thing when runtime PM is disabled in Kconfig?
>> 
>> Good point.
>> 
>> Then the way to make this cleaner, and obvious on inspection that system
>> suspend/resume are doing the same thing as runtime suspend/resume is to have -
>> >suspend call the runtime_suspend function.
>> 
>> The runtime suspend/resume functions then should be wrapped in CONFIG_PM
>> instead of CONFIG_PM_RUNTIME.
> But if the runtime PM is disabled, __pm_runtime_idle() return -ENOSYS,
> which invoked by pm_runtime_put_sync(), in spite of the runtime
> suspend/resume functions wrapper,

You won't be calling _put_sync(), instead you'll just directly call
atmel_spi_runtime_suspend().

The goal is to make it obvious upon reading that ->suspend and
->runtime_suspend are doing exactly the same thing.

Kevin