Re: [PATCH v6 3/5] media: uvcvideo: Increase/decrease the PM counter per IOCTL

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

On Mon, Apr 28, 2025 at 01:54:38PM +0200, Hans de Goede wrote:
> On 23-Apr-25 00:58, Ricardo Ribalda wrote:
> > On Wed, 23 Apr 2025 at 04:37, Laurent Pinchart wrote:
> >> On Thu, Mar 27, 2025 at 09:05:29PM +0000, Ricardo Ribalda wrote:
> >>> Now we call uvc_pm_get/put from the device open/close. This low
> >>> level of granularity might leave the camera powered on in situations
> >>> where it is not needed.
> >>>
> >>> Increase the granularity by increasing and decreasing the Power
> >>> Management counter per ioctl. There are two special cases where the
> >>> power management outlives the ioctl: async controls and streamon. Handle
> >>> those cases as well.
> >>>
> >>> In a future patch, we will remove the uvc_pm_get/put from open/close.
> >>>
> >>> Reviewed-by: Hans de Goede <hdegoede@redhat.com>
> >>> Signed-off-by: Ricardo Ribalda <ribalda@chromium.org>
> >>> ---
> >>>  drivers/media/usb/uvc/uvc_ctrl.c | 37 +++++++++++++++++++++++++++----------
> >>>  drivers/media/usb/uvc/uvc_v4l2.c | 39 +++++++++++++++++++++++++++++++++++++--
> >>>  2 files changed, 64 insertions(+), 12 deletions(-)
> >>>
> >>> diff --git a/drivers/media/usb/uvc/uvc_ctrl.c b/drivers/media/usb/uvc/uvc_ctrl.c
> >>> index cbf19aa1d82374a08cf79b6a6787fa348b83523a..3fad289e41fd5a757f8dcf30a6238c694fc4250c 100644
> >>> --- a/drivers/media/usb/uvc/uvc_ctrl.c
> >>> +++ b/drivers/media/usb/uvc/uvc_ctrl.c
> >>> @@ -1812,38 +1812,49 @@ static void uvc_ctrl_send_slave_event(struct uvc_video_chain *chain,
> >>>       uvc_ctrl_send_event(chain, handle, ctrl, mapping, val, changes);
> >>>  }
> >>>
> >>> -static void uvc_ctrl_set_handle(struct uvc_fh *handle, struct uvc_control *ctrl,
> >>> -                             struct uvc_fh *new_handle)
> >>> +static int uvc_ctrl_set_handle(struct uvc_fh *handle, struct uvc_control *ctrl,
> >>> +                            struct uvc_fh *new_handle)
> >>>  {
> >>>       lockdep_assert_held(&handle->chain->ctrl_mutex);
> >>>
> >>>       if (new_handle) {
> >>> +             int ret;
> >>> +
> >>>               if (ctrl->handle)
> >>>                       dev_warn_ratelimited(&handle->stream->dev->udev->dev,
> >>>                                            "UVC non compliance: Setting an async control with a pending operation.");
> >>>
> >>>               if (new_handle == ctrl->handle)
> >>> -                     return;
> >>> +                     return 0;
> >>>
> >>>               if (ctrl->handle) {
> >>>                       WARN_ON(!ctrl->handle->pending_async_ctrls);
> >>>                       if (ctrl->handle->pending_async_ctrls)
> >>>                               ctrl->handle->pending_async_ctrls--;
> >>> +                     ctrl->handle = new_handle;
> >>> +                     handle->pending_async_ctrls++;
> >>
> >> As commented previously, your usage of the handle variable is confusing.
> >>
> >>                         ctrl->handle->pending_async_ctrls++;
> > 
> > I believe what makes it confusing is the function arguments.
> > 
> > Would you mind if I send a new patch introducing:
> > uvc_ctrl_set_handle() and uvc_ctrl_clear_handle().

It seems worth a try. I would still like the above change to be
addressed. Can you please send a patch for that too ?

> Ricardo, if you do end up making this change, please do so as a follow-up
> patch on top of current uvc/next so that we don't have to drop the whole
> series and then rebuild uvc/next from scratch.

Ack, that's my preference too.

> >>> +                     return 0;
> >>>               }
> >>>
> >>> +             ret = uvc_pm_get(handle->chain->dev);
> >>> +             if (ret)
> >>> +                     return ret;
> >>> +
> >>>               ctrl->handle = new_handle;
> >>>               handle->pending_async_ctrls++;
> >>
> >>                 ctrl->handle->pending_async_ctrls++;
> >>
> >>> -             return;
> >>> +             return 0;
> >>>       }
> >>>
> >>>       /* Cannot clear the handle for a control not owned by us.*/
> >>>       if (WARN_ON(ctrl->handle != handle))
> >>> -             return;
> >>> +             return -EINVAL;
> >>>
> >>>       ctrl->handle = NULL;
> >>>       if (WARN_ON(!handle->pending_async_ctrls))
> >>> -             return;
> >>> +             return -EINVAL;
> >>>       handle->pending_async_ctrls--;
> >>> +     uvc_pm_put(handle->chain->dev);
> >>> +     return 0;
> >>>  }
> >>>
> >>>  void uvc_ctrl_status_event(struct uvc_video_chain *chain,
> >>> @@ -2137,15 +2148,16 @@ static int uvc_ctrl_commit_entity(struct uvc_device *dev,
> >>>
> >>>               ctrl->dirty = 0;
> >>>
> >>> +             if (!rollback && handle && !ret &&
> >>> +                 ctrl->info.flags & UVC_CTRL_FLAG_ASYNCHRONOUS)
> >>> +                     ret = uvc_ctrl_set_handle(handle, ctrl, handle);
> >>> +
> >>>               if (ret < 0) {
> >>>                       if (err_ctrl)
> >>>                               *err_ctrl = ctrl;
> >>>                       return ret;
> >>>               }
> >>>
> >>> -             if (!rollback && handle &&
> >>> -                 ctrl->info.flags & UVC_CTRL_FLAG_ASYNCHRONOUS)
> >>> -                     uvc_ctrl_set_handle(handle, ctrl, handle);
> >>>       }
> >>>
> >>>       return 0;
> >>> @@ -3222,6 +3234,7 @@ int uvc_ctrl_init_device(struct uvc_device *dev)
> >>>  void uvc_ctrl_cleanup_fh(struct uvc_fh *handle)
> >>>  {
> >>>       struct uvc_entity *entity;
> >>> +     int i;
> >>>
> >>>       guard(mutex)(&handle->chain->ctrl_mutex);
> >>>
> >>> @@ -3236,7 +3249,11 @@ void uvc_ctrl_cleanup_fh(struct uvc_fh *handle)
> >>>               }
> >>>       }
> >>>
> >>> -     WARN_ON(handle->pending_async_ctrls);
> >>> +     if (!WARN_ON(handle->pending_async_ctrls))
> >>> +             return;
> >>> +
> >>> +     for (i = 0; i < handle->pending_async_ctrls; i++)
> >>> +             uvc_pm_put(handle->stream->dev);
> >>>  }
> >>>
> >>>  /*
> >>> diff --git a/drivers/media/usb/uvc/uvc_v4l2.c b/drivers/media/usb/uvc/uvc_v4l2.c
> >>> index 1d5be045d04ecbf17e65e14b390e494a294b735f..8bccf7e17528b62f2594c0dad99405034532973d 100644
> >>> --- a/drivers/media/usb/uvc/uvc_v4l2.c
> >>> +++ b/drivers/media/usb/uvc/uvc_v4l2.c
> >>> @@ -697,6 +697,9 @@ static int uvc_v4l2_release(struct file *file)
> >>>       if (uvc_has_privileges(handle))
> >>>               uvc_queue_release(&stream->queue);
> >>>
> >>> +     if (handle->is_streaming)
> >>> +             uvc_pm_put(stream->dev);
> >>> +
> >>>       /* Release the file handle. */
> >>>       uvc_dismiss_privileges(handle);
> >>>       v4l2_fh_del(&handle->vfh);
> >>> @@ -862,6 +865,11 @@ static int uvc_ioctl_streamon(struct file *file, void *fh,
> >>>       if (ret)
> >>>               return ret;
> >>>
> >>> +     ret = uvc_pm_get(stream->dev);
> >>
> >> Shouldn't this be done before calling uvc_queue_streamon() ? There's
> >> another PM reference being held by the ioctl handler, but if the code is
> >> refactored later, it would be good to make sure we resume the device
> >> before starting streaming.
> > 
> > I was trying to simplify the error handling and, as you say, the ioctl
> > handler already holds a reference. I do not mind sending a follow-up
> > patch changing the order.

One of the two uvc_queue_streamon() or uvc_pm_get() calls will need to
perform cleanup in case of error. As the cleanup should just be
uvc_pm_put(), it seems equally easy. I'd appreciate a follow-up patch.

> >>> +     if (ret) {
> >>> +             uvc_queue_streamoff(&stream->queue, type);
> >>> +             return ret;
> >>> +     }
> >>>       handle->is_streaming = true;
> >>>
> >>>       return 0;
> >>> @@ -879,7 +887,10 @@ static int uvc_ioctl_streamoff(struct file *file, void *fh,
> >>>       guard(mutex)(&stream->mutex);
> >>>
> >>>       uvc_queue_streamoff(&stream->queue, type);
> >>> -     handle->is_streaming = false;
> >>> +     if (handle->is_streaming) {
> >>> +             handle->is_streaming = false;
> >>> +             uvc_pm_put(stream->dev);
> >>> +     }
> >>>
> >>>       return 0;
> >>>  }
> >>> @@ -1378,9 +1389,11 @@ static int uvc_v4l2_put_xu_query(const struct uvc_xu_control_query *kp,
> >>>  #define UVCIOC_CTRL_MAP32    _IOWR('u', 0x20, struct uvc_xu_control_mapping32)
> >>>  #define UVCIOC_CTRL_QUERY32  _IOWR('u', 0x21, struct uvc_xu_control_query32)
> >>>
> >>> +DEFINE_FREE(uvc_pm_put, struct uvc_device *, if (_T) uvc_pm_put(_T))
> >>>  static long uvc_v4l2_compat_ioctl32(struct file *file,
> >>>                    unsigned int cmd, unsigned long arg)
> >>>  {
> >>> +     struct uvc_device *uvc_device __free(uvc_pm_put) = NULL;
> >>>       struct uvc_fh *handle = file->private_data;
> >>>       union {
> >>>               struct uvc_xu_control_mapping xmap;
> >>> @@ -1389,6 +1402,12 @@ static long uvc_v4l2_compat_ioctl32(struct file *file,
> >>>       void __user *up = compat_ptr(arg);
> >>>       long ret;
> >>>
> >>> +     ret = uvc_pm_get(handle->stream->dev);
> >>> +     if (ret)
> >>> +             return ret;
> >>> +
> >>> +     uvc_device = handle->stream->dev;
> >>
> >> Ouch... That's not nice very :-/
> > 
> > IIt is nicer than changing the returns with breaks, believe me I tried
> > :), and it is more prone to errors.
> > 
> > I thought about the CLASS, but it is not worth it with a single user.
> > I believe the current code is a good compromise, but I might be
> > biased.

Yes, a class may be a bit of an over-engineering. I'd prefer breaks
though.

> >> If you want to use the cleanup API, I think we could use guards with an
> >> init function such as
> >>
> >> struct uvc_device *__uvc_pm_get_init(struct uvc_device *dev, int *ret)
> >> {
> >>         *ret = uvc_pm_get(dev);
> >>         return *ret ? NULL : dev;
> >> }
> >>
> >> You can use DEFINE_CLASS() instead of DEFINE_GUARD() to control the
> >> arguments to the init function. Users of the guard could do
> >>
> >>         int ret;
> >>
> >>         guard(uvc_pm)(dev, &ret);
> >>         if (ret)
> >>                 return ret;
> >>
> >>         ...
> >>
> >> Another, simpler option would be to replace returns with breaks in
> >> uvc_v4l2_compat_ioctl32(). I'm tempted to do that in this patch, and
> >> switching to the cleanup API as a patch on top if desired.
> >>
> >>> +
> >>>       switch (cmd) {
> >>>       case UVCIOC_CTRL_MAP32:
> >>>               ret = uvc_v4l2_get_xu_mapping(&karg.xmap, up);
> >>> @@ -1423,6 +1442,22 @@ static long uvc_v4l2_compat_ioctl32(struct file *file,
> >>>  }
> >>>  #endif
> >>>
> >>> +static long uvc_v4l2_unlocked_ioctl(struct file *file,
> >>> +                                 unsigned int cmd, unsigned long arg)
> >>> +{
> >>> +     struct uvc_fh *handle = file->private_data;
> >>> +     int ret;
> >>> +
> >>> +     ret = uvc_pm_get(handle->stream->dev);
> >>> +     if (ret)
> >>> +             return ret;
> >>> +
> >>> +     ret = video_ioctl2(file, cmd, arg);
> >>> +
> >>> +     uvc_pm_put(handle->stream->dev);
> >>> +     return ret;
> >>> +}
> >>> +
> >>>  static ssize_t uvc_v4l2_read(struct file *file, char __user *data,
> >>>                   size_t count, loff_t *ppos)
> >>>  {
> >>> @@ -1507,7 +1542,7 @@ const struct v4l2_file_operations uvc_fops = {
> >>>       .owner          = THIS_MODULE,
> >>>       .open           = uvc_v4l2_open,
> >>>       .release        = uvc_v4l2_release,
> >>> -     .unlocked_ioctl = video_ioctl2,
> >>> +     .unlocked_ioctl = uvc_v4l2_unlocked_ioctl,
> >>>  #ifdef CONFIG_COMPAT
> >>>       .compat_ioctl32 = uvc_v4l2_compat_ioctl32,
> >>>  #endif

-- 
Regards,

Laurent Pinchart