[PATCH v2 2/4] gpu: host1x: add runtime pm support for gr2d

Mayuresh Kulkarni mkulkarni at nvidia.com
Fri Jun 14 07:45:42 PDT 2013


On Thursday 13 June 2013 08:44 PM, Stephen Warren wrote:
> On 06/13/2013 03:53 AM, Mayuresh Kulkarni wrote:
>
> Patch description?

I thought the patch subject is sufficient to tell what it is it doing. 
Description here would be repetition in my opinion.

Also, the cover letter for the patch-set series is verbose enough.

>
>
>> diff --git a/drivers/gpu/host1x/drm/gr2d.c b/drivers/gpu/host1x/drm/gr2d.c
>
>> +#ifdef CONFIG_PM_RUNTIME
>> +	pm_runtime_enable(&pdev->dev);
>> +	pm_runtime_get_sync(&pdev->dev);
>> +#else
>>   	err = clk_prepare_enable(gr2d->clk);
>>   	if (err) {
>>   		dev_err(dev, "cannot turn on clock\n");
>>   		return err;
>>   	}
>> +#endif
>
> The #else block here is a cut/paste of the body of
> gr2d_runtime_resume(). It'd be better to call that function instead. The
> following is what I ended up with in the Tegra ASoC driver in order to
> support runtime PM on or off:
>
>          pm_runtime_enable(&pdev->dev);
>          if (!pm_runtime_enabled(&pdev->dev)) {
>                  ret = tegra20_i2s_runtime_resume(&pdev->dev);
>                  if (ret)
>                          goto err_pm_disable;
>          }
>

Thanks for the tip. Runtime detection is better than compile time here.

>> @@ -328,10 +338,51 @@ static int __exit gr2d_remove(struct platform_device *pdev)
>>
>>   	host1x_channel_free(gr2d->channel);
>>   	clk_disable_unprepare(gr2d->clk);
>
> Don't you need to remove that clk disable, or make it conditional upon
> !PM_RUNTIME?

Yes you are correct.

>
>> +#ifdef CONFIG_PM_RUNTIME
>> +	pm_runtime_disable(&pdev->dev);
>> +#endif
>
> Similarly, perhaps something like the following here:
>
>          pm_runtime_disable(&pdev->dev);
>          if (!pm_runtime_status_suspended(&pdev->dev))
>                  tegra20_i2s_runtime_suspend(&pdev->dev);
>
>> @ -591,10 +592,18 @@ int host1x_job_submit(struct host1x_job *job)
>>   {
>>   	struct host1x *host = dev_get_drvdata(job->channel->dev->parent);
>>
>> +#ifdef CONFIG_PM_RUNTIME
>> +	pm_runtime_get_sync(job->channel->dev);
>> +#endif
>> +
>>   	return host1x_hw_channel_submit(host, job);
>>   }
>>
>>   int host1x_job_complete(struct host1x_job *job)
>>   {
>> +#ifdef CONFIG_PM_RUNTIME
>> +	return pm_runtime_put(job->channel->dev);
>> +#else
>>   	return 0;
>> +#endif
>>   }
>
> I don't think you need any of those ifdefs; simply call the
> pm_runtime_*() functions all the time, and they'll be successful no-ops
> if !PM_RUNTIME.
>
OK. But I thought it will be better to be verbose (which is not needed).



More information about the dri-devel mailing list