Re: [PATCH] ALSA: hda/tegra: enable clock during probe

From: Sameer Pujar
Date: Mon Jan 28 2019 - 01:07:12 EST



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@xxxxxxxxxx>
Reviewed-by: Ravindra Lokhande <rlokhande@xxxxxxxxxx>
Reviewed-by: Jon Hunter <jonathanh@xxxxxxxxxx>
(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.

Thanks,
Sameer.

Jon