Re: [alsa-devel] [PATCH 4/9] ASoC: tegra: add Tegra210 based I2S driver

[Dmitry Osipenko]

27.01.2020 08:22, Sameer Pujar ÐÐÑÐÑ:
> 
> 
> On 1/24/2020 7:34 PM, Dmitry Osipenko wrote:
>> External email: Use caution opening links or attachments
>>
>>
>> 24.01.2020 12:51, Jon Hunter ÐÐÑÐÑ:
>>> On 24/01/2020 09:07, Jon Hunter wrote:
>>>> On 23/01/2020 15:16, Dmitry Osipenko wrote:
>>>>> 23.01.2020 12:22, Sameer Pujar ÐÐÑÐÑ:
>>>>>>
>>>>>> On 1/22/2020 9:57 PM, Dmitry Osipenko wrote:
>>>>>>> External email: Use caution opening links or attachments
>>>>>>>
>>>>>>>
>>>>>>> 22.01.2020 14:52, Jon Hunter ÐÐÑÐÑ:
>>>>>>>> On 22/01/2020 07:16, Sameer Pujar wrote:
>>>>>>>>
>>>>>>>> ...
>>>>>>>>
>>>>>>>>>>>>>>> +static int tegra210_i2s_remove(struct platform_device
>>>>>>>>>>>>>>> *pdev)
>>>>>>>>>>>>>>> +{
>>>>>>>>>>>>>>> +ÂÂÂÂ pm_runtime_disable(&pdev->dev);
>>>>>>>>>>>>>>> +ÂÂÂÂ if (!pm_runtime_status_suspended(&pdev->dev))
>>>>>>>>>>>>>>> +ÂÂÂÂÂÂÂÂÂÂÂÂ tegra210_i2s_runtime_suspend(&pdev->dev);
>>>>>>>>>>>>>> This breaks device's RPM refcounting if it was disabled in
>>>>>>>>>>>>>> the
>>>>>>>>>>>>>> active
>>>>>>>>>>>>>> state. This code should be removed. At most you could warn
>>>>>>>>>>>>>> about the
>>>>>>>>>>>>>> unxpected RPM state here, but it shouldn't be necessary.
>>>>>>>>>>>>> I guess this was added for safety and explicit suspend
>>>>>>>>>>>>> keeps clock
>>>>>>>>>>>>> disabled.
>>>>>>>>>>>>> Not sure if ref-counting of the device matters when runtime
>>>>>>>>>>>>> PM is
>>>>>>>>>>>>> disabled and device is removed.
>>>>>>>>>>>>> I see few drivers using this way.
>>>>>>>>>>>> It should matter (if I'm not missing something) because RPM
>>>>>>>>>>>> should
>>>>>>>>>>>> be in
>>>>>>>>>>>> a wrecked state once you'll try to re-load the driver's module.
>>>>>>>>>>>> Likely
>>>>>>>>>>>> that those few other drivers are wrong.
>>>>>>>>>>>>
>>>>>>>>>>>> [snip]
>>>>>>>>>>> Once the driver is re-loaded and RPM is enabled, I don't
>>>>>>>>>>> think it
>>>>>>>>>>> would use
>>>>>>>>>>> the same 'dev' and the corresponding ref count. Doesn't it
>>>>>>>>>>> use the
>>>>>>>>>>> new
>>>>>>>>>>> counters?
>>>>>>>>>>> If RPM is not working for some reason, most likely it would
>>>>>>>>>>> be the
>>>>>>>>>>> case
>>>>>>>>>>> for other
>>>>>>>>>>> devices. What best driver can do is probably do a force suspend
>>>>>>>>>>> during
>>>>>>>>>>> removal if
>>>>>>>>>>> already not done. I would prefer to keep, since multiple drivers
>>>>>>>>>>> still
>>>>>>>>>>> have it,
>>>>>>>>>>> unless there is a real harm in doing so.
>>>>>>>>>> I took a closer look and looks like the counter actually
>>>>>>>>>> should be
>>>>>>>>>> reset. Still I don't think that it's a good practice to make
>>>>>>>>>> changes
>>>>>>>>>> underneath of RPM, it may strike back.
>>>>>>>>> If RPM is broken, it probably would have been caught during device
>>>>>>>>> usage.
>>>>>>>>> I will remove explicit suspend here if no any concerns from other
>>>>>>>>> folks.
>>>>>>>>> Thanks.
>>>>>>>> I recall that this was the preferred way of doing this from the RPM
>>>>>>>> folks. Tegra30 I2S driver does the same and Stephen had pointed
>>>>>>>> me to
>>>>>>>> this as a reference.
>>>>>>>> I believe that this is meant to ensure that the
>>>>>>>> device is always powered-off regardless of it RPM is enabled or
>>>>>>>> not and
>>>>>>>> what the current state is.
>>>>>>> Yes, it was kinda actual for the case of unavailable RPM.
>>>>>>> Anyways, /I think/ variant like this should have been more
>>>>>>> preferred:
>>>>>>>
>>>>>>> if (!pm_runtime_enabled(&pdev->dev))
>>>>>>> ÂÂÂÂÂÂÂÂÂ tegra210_i2s_runtime_suspend(&pdev->dev);
>>>>>>> else
>>>>>>> ÂÂÂÂÂÂÂÂÂ pm_runtime_disable(&pdev->dev);
>>>>>> I think it looks to be similar to what is there already.
>>>>>>
>>>>>> pm_runtime_disable(&pdev->dev); // it would turn out to be a dummy
>>>>>> call
>>>>>> if !RPM
>>>>>> if (!pm_runtime_status_suspended(&pdev->dev)) // it is true always
>>>>>> if !RPM
>>>>>> ÂÂÂÂÂÂÂÂ tegra210_i2s_runtime_suspend(&pdev->dev);
>>>>> Maybe this is fine for !RPM, but not really fine in a case of enabled
>>>>> RPM. Device could be in resumed state after pm_runtime_disable() if it
>>>>> wasn't suspended before the disabling.
>>>> I don't see any problem with this for the !RPM case.
>>> Sorry I meant the RPM case. In other words, I don't see a problem for
>>> neither the RPM case of the !RPM case.
>> 1. Device shall be in RPM-suspended state at the time of driver's
>> removal, unless there is a bug in the sound driver. Hence why do you
>> need the dead code which doesn't bring any practical value?
>>
>> 2. Making changes underneath of RPM is simply error-prone. It may hit
>> badly in the future once something will change in the RPM core.
> 
> I think we are stretching a bit more here when there is no any real harm.
> Right now it works well for both RPM and !RPM case and if we really need to
> fix something in future we can fix. Since my initial inclination was
> keeping
> the code as it is and Jon also has similar thoughts, I would retain this
> code.
> Sorry Dmitry, we can fix if something comes up and many other drivers would
> need this at that time.

The !RPM case isn't supported by Tegra anymore in upstream kernel. I'm
trying to help to make yours driver better and gave you reasons to
remove the unneeded code, while you're keep saying that "there is no
harm to retain it", which is not a reason to clutter up the code. I
don't feel that it's worthwhile to continue arguing here.