[alsa-devel] [PATCH] ALSA: hda/tegra: enable clock during probe

[Takashi Iwai]

On Wed, 30 Jan 2019 10:35:35 +0100,
Jon Hunter wrote:
> 
> 
> On 28/01/2019 06:06, Sameer Pujar wrote:
> > 
> > On 1/25/2019 7:34 PM, Jon Hunter wrote:
> >> On 25/01/2019 13:58, Takashi Iwai wrote:
> >>> On Fri, 25 Jan 2019 14:26:27 +0100,
> >>> Jon Hunter wrote:
> >>>>
> >>>> On 25/01/2019 12:40, Takashi Iwai wrote:
> >>>>> On Fri, 25 Jan 2019 12:36:00 +0100,
> >>>>> Jon Hunter wrote:
> >>>>>>
> >>>>>> On 24/01/2019 19:08, Takashi Iwai wrote:
> >>>>>>> On Thu, 24 Jan 2019 18:36:43 +0100,
> >>>>>>> Sameer Pujar wrote:
> >>>>>>>> If CONFIG_PM is disabled or runtime PM calls are forbidden, the
> >>>>>>>> clocks
> >>>>>>>> will not be ON. This could cause issue during probe, where hda init
> >>>>>>>> setup is done. This patch checks whether runtime PM is enabled
> >>>>>>>> or not.
> >>>>>>>> If disabled, clocks are enabled in probe() and disabled in remove()
> >>>>>>>>
> >>>>>>>> This patch does following minor changes as cleanup,
> >>>>>>>>    * return code check for pm_runtime_get_sync() to take care of
> >>>>>>>> failure
> >>>>>>>>      and exit gracefully.
> >>>>>>>>    * In remove path runtime PM is disabled before calling
> >>>>>>>> snd_card_free().
> >>>>>>>>    * hda_tegra_disable_clocks() is moved out of CONFIG_PM_SLEEP
> >>>>>>>> check.
> >>>>>>>>    * runtime PM callbacks moved out of CONFIG_PM check
> >>>>>>>>
> >>>>>>>> Signed-off-by: Sameer Pujar <spujar at nvidia.com>
> >>>>>>>> Reviewed-by: Ravindra Lokhande <rlokhande at nvidia.com>
> >>>>>>>> Reviewed-by: Jon Hunter <jonathanh at nvidia.com>
> >>>>>>> (snip)
> >>>>>>>> @@ -555,6 +553,13 @@ static int hda_tegra_probe(struct
> >>>>>>>> platform_device *pdev)
> >>>>>>>>       if (!azx_has_pm_runtime(chip))
> >>>>>>>>           pm_runtime_forbid(hda->dev);
> >>>>>>>>   +    /* explicit resume if runtime PM is disabled */
> >>>>>>>> +    if (!pm_runtime_enabled(hda->dev)) {
> >>>>>>>> +        err = hda_tegra_runtime_resume(hda->dev);
> >>>>>>>> +        if (err)
> >>>>>>>> +            goto out_free;
> >>>>>>>> +    }
> >>>>>>>> +
> >>>>>>>>       schedule_work(&hda->probe_work);
> >>>>>>> Calling runtime_resume here is really confusing...
> >>>>>> Why? IMO it is better to have a single handler for resuming the
> >>>>>> device
> >>>>>> and so if RPM is not enabled we call the handler directly. This is
> >>>>>> what
> >>>>>> we have been advised to do in the past and do in other drivers.
> >>>>>> See ...
> >>>>> The point is that we're not "resuming" anything there.  It's in the
> >>>>> early probe stage, and the device state is uninitialized, not really
> >>>>> suspended.  It'd end up with just calling the same helper
> >>>>> (hda_tegra_enable_clocks()), though.
> >>>> Yes and you can make the same argument for every driver that calls
> >>>> pm_runtime_get_sync() during probe to turn on clocks, handle resets,
> >>>> etc, because at the end of the day the very first call to
> >>>> pm_runtime_get_sync() invokes the runtime_resume callback, when we have
> >>>> never been suspended.
> >>> Although there are some magical pm_runtime_*() in some places, most of
> >>> such pm_runtime_get_sync() is for the actual runtime PM management (to
> >>> prevent the runtime suspend), while the code above is for explicitly
> >>> setting up something for non-PM cases.
> >>>
> >>> And if pm_runtime_get_sync() is obviously superfluous, we should
> >>> remove such calls.  Really.
> >> Yes agree.
> >>
> >>>> Yes at the end of the day it is the same and given that we have done
> >>>> this elsewhere I think it is good to be consistent if/where we can.
> >>> The code becomes less readable, and that's a good reason against it :)
> >> I don't its less readable. However, I do think it is less error prone :-)
> > 
> > Do we have a consensus here? Request others to provide opinions to help
> > close on this.
> 
> I am not going to block this and ultimately it is Iwai-san call.
> 
> However, I wonder if it would be appropriate to move the whole ...
> 
>  if (pm_runtime_enabled())
>          ret = pm_runtime_get_sync();
>  else
>          ret = hda_tegra_runtime_resume();
> 
> ... into the probe_work function? In other words, we are just resuming
> when we really need to. Unless I am still misunderstanding Iwai-san
> comment. Otherwise if Iwai-san is happy with V2 then go with that.

Only from my personal taste, I find the v2 patch is better.
It like simpler, after all.  That is, the code in v1 patch

	probe() {
		....
		pm_runtime_enable();
		....
		if (!pm_runtime_enabled())
			hda_tegra_runtime_resume();
		schedule_work();
	}

	work() {
		pm_runtime_get_sync();
		....
		pm_runtime_put();
	}

becomes shorter in v2:

	probe() {
		....
		hda_tegra_enable_clocks();
		schedule_work();
	}

	work() {
		....
		pm_runtime_enable();
	}


However, the point about hda_tegra_remove() you raised in the v2 patch
is still valid.  (BTW, I guess the discussion followed in that thread
was somehow misunderstood; your argument was about hda_tegra_remove()
while Sameer discussed about the probe.)  It can be with
hda_tegra_disable_clocks() if we want more consistency.

Though, I don't mind too much about that as long as the proper comment
is given.


thanks,

Takashi