Re: [PATCH 58/78] media: exynos-gsc: use pm_runtime_resume_and_get()

From: Mauro Carvalho Chehab
Date: Wed Apr 28 2021 - 03:17:16 EST


Em Wed, 28 Apr 2021 09:13:02 +0200
Mauro Carvalho Chehab <mchehab+huawei@xxxxxxxxxx> escreveu:

> Em Tue, 27 Apr 2021 13:50:44 +0200
> Sylwester Nawrocki <s.nawrocki@xxxxxxxxxxx> escreveu:
>
> > On 27.04.2021 11:42, Mauro Carvalho Chehab wrote:
> > > Em Tue, 27 Apr 2021 11:30:55 +0200
> > > Mauro Carvalho Chehab <mchehab+huawei@xxxxxxxxxx> escreveu:
> > >
> > >> Em Tue, 27 Apr 2021 10:18:12 +0200
> > >> Sylwester Nawrocki <s.nawrocki@xxxxxxxxxxx> escreveu:
> > >>
> > >>> On 24.04.2021 08:45, Mauro Carvalho Chehab wrote:
> > >>>> Commit dd8088d5a896 ("PM: runtime: Add pm_runtime_resume_and_get to deal with usage counter")
> > >>>> added pm_runtime_resume_and_get() in order to automatically handle
> > >>>> dev->power.usage_count decrement on errors.
> > >>>>
> > >>>> Use the new API, in order to cleanup the error check logic.
> > >>>>
> > >>>> Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@xxxxxxxxxx>
> > >>>> ---
> > >>>> drivers/media/platform/exynos-gsc/gsc-core.c | 3 +--
> > >>>> drivers/media/platform/exynos-gsc/gsc-m2m.c | 2 +-
> > >>>> 2 files changed, 2 insertions(+), 3 deletions(-)
> > >>>>
> > >>>> diff --git a/drivers/media/platform/exynos-gsc/gsc-core.c b/drivers/media/platform/exynos-gsc/gsc-core.c
> > >>>> index 9f41c2e7097a..9d5841194f6b 100644
> > >>>> --- a/drivers/media/platform/exynos-gsc/gsc-core.c
> > >>>> +++ b/drivers/media/platform/exynos-gsc/gsc-core.c
> > >>>> @@ -1210,7 +1210,7 @@ static int gsc_remove(struct platform_device *pdev)
> > >>>> struct gsc_dev *gsc = platform_get_drvdata(pdev);
> > >>>> int i;
> > >>>>
> > >>>> - pm_runtime_get_sync(&pdev->dev);
> > >>>> + pm_runtime_resume_and_get(&pdev->dev);
> > >>>>
> > >>>> gsc_unregister_m2m_device(gsc);
> > >>>> v4l2_device_unregister(&gsc->v4l2_dev);
> > >>>> @@ -1219,7 +1219,6 @@ static int gsc_remove(struct platform_device *pdev)
> > >>>> for (i = 0; i < gsc->num_clocks; i++)
> > >>>> clk_disable_unprepare(gsc->clock[i]);
> > >>>>
> > >>>> - pm_runtime_put_noidle(&pdev->dev);
> > >>>
> > >>> If we do this then the device usage count will not get decremented
> > >>> after the pm_runtime_resume_and_get() call above and after driver
> > >>> unload/load cycle it will not be possible to suspend the device.
> > >>> I wouldn't be changing anything in gsc_remove(), pm_runtime_get_sync()
> > >>> works better in that case.
> > >>
> > >> Good point.
> > >>
> > >> Actually, I don't see any reason why to call a PM resume
> > >> function - either being pm_runtime_get_sync() or
> > >> pm_runtime_resume_and_get().
> > >>
> > >> The code there could simply be:
> > >>
> > >> static int gsc_remove(struct platform_device *pdev)
> > >> {
> > >> struct gsc_dev *gsc = platform_get_drvdata(pdev);
> > >> int i;
> > >>
> > >> gsc_unregister_m2m_device(gsc);
> > >> v4l2_device_unregister(&gsc->v4l2_dev);
> > >>
> > >> vb2_dma_contig_clear_max_seg_size(&pdev->dev);
> > >> for (i = 0; i < gsc->num_clocks; i++)
> > >> clk_disable_unprepare(gsc->clock[i]);
> > >>
> > >> pm_runtime_disable(&pdev->dev);
> > >>
> > >> dev_dbg(&pdev->dev, "%s driver unloaded\n", pdev->name);
> > >> return 0;
> > >> }
> > >>
> > >> Eventually also adding:
> > >> pm_runtime_suspended(&pdev->dev);
> > >
> > > In time: I actually meant:
> > >
> > > pm_runtime_set_suspended(&pdev->dev);
> > >
> > > but after double-checking the PM runtime code, it sounds to me that
> > > just calling pm_runtime_disable() would be enough. Not 100% sure
> > > here. Btw, some media drivers call it after pm_runtime_disable(),
> > > while others don't do.
> >
> > I think if the device is brought into suspended state (e.g. by
> > disabling clocks as above) the pm_runtime_set_suspended() call
> > should be there. IOW a following sequence:
> >
> > pm_runtime_disable(dev);
> > if (!pm_runtime_status_suspended(dev))
> > /* put device into suspended state (disable clocks,
> > voltage regulators, assert GPIOs, etc. */
> > pm_runtime_set_suspended(dev);
>
> Not sure if this would work, as the clock framework would try
> to do things like calling clk_pm_runtime_put().
>
> Perhaps an alternative would be to just return an error if it
> can't resume PM runtime, e. g.:
>
> diff --git a/drivers/media/platform/exynos-gsc/gsc-core.c b/drivers/media/platform/exynos-gsc/gsc-core.c
> index 9f41c2e7097a..d47d02c75484 100644
> --- a/drivers/media/platform/exynos-gsc/gsc-core.c
> +++ b/drivers/media/platform/exynos-gsc/gsc-core.c
> @@ -1208,9 +1208,11 @@ static int gsc_probe(struct platform_device *pdev)
> static int gsc_remove(struct platform_device *pdev)
> {
> struct gsc_dev *gsc = platform_get_drvdata(pdev);
> - int i;
> + int ret, i;
>
> - pm_runtime_get_sync(&pdev->dev);
> + ret = pm_runtime_resume_and_get(&pdev->dev);
> + if (ret < 0)
> + return ret;

Nah, forget about that. Despite the platform driver having a return code,
support for it bogus:


static int platform_remove(struct device *_dev)
{
struct platform_driver *drv = to_platform_driver(_dev->driver);
struct platform_device *dev = to_platform_device(_dev);

if (drv->remove) {
int ret = drv->remove(dev);

if (ret)
dev_warn(_dev, "remove callback returned a non-zero value. This will be ignored.\n");
}
dev_pm_domain_detach(_dev, true);

return 0;
}

Thanks,
Mauro