Re: [PATCH 9/9] media: uvcvideo: Support granular power saving for compat syscalls

[Hans de Goede <hansg@kernel.org>]

Hi,

On 2-Jun-25 13:40, Ricardo Ribalda wrote:
> Hi Hans
> 
> On Mon, 2 Jun 2025 at 13:24, Hans de Goede <hansg@kernel.org> wrote:
>>
>> On 2-Jun-25 13:11, Ricardo Ribalda wrote:
>>> On Mon, 2 Jun 2025 at 13:07, Hans de Goede <hansg@kernel.org> wrote:
>>>>
>>>> Hi Ricardo,
>>>>
>>>> On 2-Jun-25 12:27, Ricardo Ribalda wrote:
>>>>> Hi Hans
>>>>>
>>>>> On Mon, 2 Jun 2025 at 12:11, Hans de Goede <hansg@kernel.org> wrote:
>>>>>>
>>>>>> Hi Ricardo,
>>>>>>
>>>>>> On 28-May-25 19:58, Ricardo Ribalda wrote:
>>>>>>> Right now we cannot support granular power saving on compat syscalls
>>>>>>> because the VIDIOC_*32 NRs defines are not accessible to drivers.
>>>>>>>
>>>>>>> Use the video_translate_cmd() helper to convert the compat syscall NRs
>>>>>>> into syscall NRs.
>>>>>>>
>>>>>>> Signed-off-by: Ricardo Ribalda <ribalda@chromium.org>
>>>>>>> ---
>>>>>>>  drivers/media/usb/uvc/uvc_v4l2.c     | 9 ++-------
>>>>>>>  drivers/media/v4l2-core/v4l2-ioctl.c | 3 ++-
>>>>>>>  include/media/v4l2-ioctl.h           | 1 +
>>>>>>>  3 files changed, 5 insertions(+), 8 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/media/usb/uvc/uvc_v4l2.c b/drivers/media/usb/uvc/uvc_v4l2.c
>>>>>>> index fcb1b79c214849ce4da96a86a688d777b32cc688..048ee7e01808c8944f9bd46e5df2931b9c146ad5 100644
>>>>>>> --- a/drivers/media/usb/uvc/uvc_v4l2.c
>>>>>>> +++ b/drivers/media/usb/uvc/uvc_v4l2.c
>>>>>>> @@ -1282,15 +1282,10 @@ static long uvc_v4l2_pm_ioctl(struct file *file,
>>>>>>>  static long uvc_v4l2_unlocked_ioctl(struct file *file,
>>>>>>>                                   unsigned int cmd, unsigned long arg)
>>>>>>>  {
>>>>>>> -     /*
>>>>>>> -      * For now, we do not support granular power saving for compat
>>>>>>> -      * syscalls.
>>>>>>> -      */
>>>>>>> -     if (in_compat_syscall())
>>>>>>> -             return uvc_v4l2_pm_ioctl(file, cmd, arg);
>>>>>>> +     unsigned int converted_cmd = video_translate_cmd(cmd);
>>>>>>
>>>>>> It looks like something went wrong here and you did not test-compile this?
>>>>>> video_translate_cmd() is private to drivers/media/v4l2-core/v4l2-ioctl.c
>>>>>> so this should not compile.
>>>>>
>>>>> Hmm... Actually I am pretty sure that I tested it on real hardware.
>>>>>
>>>>> Did you miss the EXPORT_SYMBOL() on the patch?
>>>>
>>>> Ah yes I did miss that, sorry.
>>>
>>> My bad, I doubt it till the last second if I should split it or not :)
>>>
>>>>
>>>> For the next time please split core changes out into their own
>>>> separate patches.
>>>>
>>>> In this case I think the core changes are not necessary instead
>>>> you can just do:
>>>>
>>>>         unsigned int converted_cmd = cmd;
>>>>
>>>> #ifdef CONFIG_COMPAT
>>>>         converted_cmd = v4l2_compat_translate_cmd(cmd);
>>>> #endif
>>>
>>> I believe this should work as well:
>>>
>>> unsigned int converted_cmd = cmd;
>>> if (in_compat_syscall())
>>>   converted_cmd = v4l2_compat_translate_cmd(cmd);
>>>
>>> the compiler knows that CONFIG_COMPAT=n means in_compat_syscall() will
>>> be always fails.
>>>
>>> If it is ok with you (and it actually works :) ) I will use this version.
>>
>> I agree that that is cleaner/better and I also think it should work,
>> so lets go with that.
> 
> Actually, v4l2_compat_translate_cmd() does not seem to be EXPORT_SYMBOL()ed
> 
> So I still need to do some changes in the core.
> (It also does not handle COMPAT_32BIT_TIME... but in this case it
> seems to be the same).
> 
> 
> Any preference between what to use: v4l2_compat_translate_cmd() vs
> video_translate_cmd()?

v4l2_compat_translate_cmd() is already exposed in include/media/v4l2-ioctl.h
so I think it is best to go with that function.

Regards,

Hans