Re: [RFC PATCH v1 3/5] media: tegra-video: Move PM runtime handle to streaming

[Sowjanya Komatineni <skomatineni-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>]

On 4/28/20 8:15 AM, Dmitry Osipenko wrote:
> External email: Use caution opening links or attachments
>
>
> 28.04.2020 18:08, Sowjanya Komatineni пишет:
>> On 4/28/20 7:59 AM, Dmitry Osipenko wrote:
>>> External email: Use caution opening links or attachments
>>>
>>>
>>> 28.04.2020 17:51, Sowjanya Komatineni пишет:
>>>> On 4/28/20 6:59 AM, Dmitry Osipenko wrote:
>>>>> External email: Use caution opening links or attachments
>>>>>
>>>>>
>>>>> 28.04.2020 07:20, Sowjanya Komatineni пишет:
>>>>>> diff --git a/drivers/staging/media/tegra-video/csi.c
>>>>>> b/drivers/staging/media/tegra-video/csi.c
>>>>>> index b3dd0c3..29ccdae 100644
>>>>>> --- a/drivers/staging/media/tegra-video/csi.c
>>>>>> +++ b/drivers/staging/media/tegra-video/csi.c
>>>>>> @@ -272,8 +272,25 @@ static int tegra_csi_s_stream(struct v4l2_subdev
>>>>>> *subdev, int enable)
>>>>>>          struct tegra_vi_channel *chan =
>>>>>> v4l2_get_subdev_hostdata(subdev);
>>>>>>          struct tegra_csi_channel *csi_chan = to_csi_chan(subdev);
>>>>>>          struct tegra_csi *csi = csi_chan->csi;
>>>>>> +     int ret;
>>>>>> +
>>>>>> +     if (enable && atomic_add_return(1, &csi->clk_refcnt) == 1) {
>>>>>> +             ret = pm_runtime_get_sync(csi->dev);
>>>>>> +             if (ret < 0) {
>>>>>> +                     dev_err(csi->dev,
>>>>>> +                             "failed to get runtime PM: %d\n", ret);
>>>>>> +                     pm_runtime_put_noidle(csi->dev);
> Why this pm_runtime_put_noidle() is needed? This should be wrong, please
> remove it.

pm_runtime_get_sync() increments power.usage_count prior to rpm_resume

if rpm_resume fails it does not decrement usage_count.

So to balance count on failure, calling pm_runtime_put_noidle()

>
>>>>>> +                     atomic_dec(&csi->clk_refcnt);
>>>>>> +                     return ret;
>>>>>> +             }
>>>>>> +     }
>>>>>> +
>>>>>> +     ret = csi->ops->csi_streaming(csi_chan, chan->pg_mode, enable);
>>>>>>
>>>>>> -     return csi->ops->csi_streaming(csi_chan, chan->pg_mode, enable);
>>>>>> +     if ((ret < 0 || !enable) &&
>>>>>> atomic_dec_and_test(&csi->clk_refcnt))
>>>>>> +             pm_runtime_put_sync(csi->dev);
>>>>> Runtime PM maintains its own refcount, why these
>>>>> clk_refcnt/power_on_refcnt are needed?
>>>> Streaming is per channel and we can't turn power/clocks off while other
>>>> channels may still be running.
>>>>
>>> All channels use the same CSI device. You should remove the custom
>>> refcounting.
>>>
>>> BTW, next time you'll really need to do refcounting, use the generic
>>> kref.
>> Before channel stream we enable power/clocks and after streaming we stop.
>>
>> So without refcount, on stream stop of any of the channel RPM put turns
>> power/clock but other channels will still be streaming.
>>
> Runtime PM uses its own refcounting. Please consult the RPM code and
> documentation.
>
> https://elixir.bootlin.com/linux/v5.7-rc3/source/include/linux/pm_runtime.h#L78
> https://elixir.bootlin.com/linux/v5.7-rc3/source/drivers/base/power/runtime.c#L1079
Got it. Will remove refcount for RPM put/get