Re: [PATCH] media: i2c: ov5645: add HAS_EVENTS supporty

[Sakari Ailus <sakari.ailus@linux.intel.com>]

Hi Tommaso,

On Fri, Oct 18, 2024 at 12:48:21PM +0200, Tommaso Merciai wrote:
> Hi Sakari,
> Thanks for your review.
> 
> On Fri, Oct 18, 2024 at 10:33:56AM +0000, Sakari Ailus wrote:
> > Hi folks,
> > 
> > On Fri, Oct 18, 2024 at 12:18:15PM +0200, Tommaso Merciai wrote:
> > > Hi Prabhakar,
> > > Thanks for your comments.
> > > 
> > > On Fri, Oct 18, 2024 at 10:57:49AM +0100, Lad, Prabhakar wrote:
> > > > Hi Tommaso,
> > > > 
> > > > On Fri, Oct 18, 2024 at 10:28 AM Tommaso Merciai <tomm.merciai@gmail.com> wrote:
> > > > >
> > > > > Hi Laurent,
> > > > >
> > > > > On Wed, Oct 16, 2024 at 11:08:36PM +0300, Laurent Pinchart wrote:
> > > > > > Hi Tommaso,
> > > > > >
> > > > > > On Tue, Oct 15, 2024 at 06:37:58PM +0200, Tommaso Merciai wrote:
> > > > > > > On Tue, Oct 15, 2024 at 08:16:13AM +0200, Tommaso Merciai wrote:
> > > > > > > > On Mon, Oct 14, 2024 at 08:54:52PM +0300, Laurent Pinchart wrote:
> > > > > > > > > On Mon, Oct 14, 2024 at 07:38:40PM +0200, Tommaso Merciai wrote:
> > > > > > > > > > Controls can be exposed to userspace via a v4l-subdevX device, and
> > > > > > > > > > userspace has to be able to subscribe to control events so that it is
> > > > > > > > > > notified when the control changes value.
> > > > > > > > > > Add missing HAS_EVENTS support: flag and .(un)subscribe_event().
> > > > > > > > > >
> > > > > > > > > > Signed-off-by: Tommaso Merciai <tomm.merciai@gmail.com>
> > > > > > > > > > ---
> > > > > > > > > >  drivers/media/i2c/ov5645.c | 10 +++++++++-
> > > > > > > > > >  1 file changed, 9 insertions(+), 1 deletion(-)
> > > > > > > > > >
> > > > > > > > > > diff --git a/drivers/media/i2c/ov5645.c b/drivers/media/i2c/ov5645.c
> > > > > > > > > > index 0c32bd2940ec..2c5145d5c616 100644
> > > > > > > > > > --- a/drivers/media/i2c/ov5645.c
> > > > > > > > > > +++ b/drivers/media/i2c/ov5645.c
> > > > > > > > > > @@ -29,6 +29,7 @@
> > > > > > > > > >  #include <linux/slab.h>
> > > > > > > > > >  #include <linux/types.h>
> > > > > > > > > >  #include <media/v4l2-ctrls.h>
> > > > > > > > > > +#include <media/v4l2-event.h>
> > > > > > > > > >  #include <media/v4l2-fwnode.h>
> > > > > > > > > >  #include <media/v4l2-subdev.h>
> > > > > > > > > >
> > > > > > > > > > @@ -1034,6 +1035,11 @@ static const struct v4l2_subdev_video_ops ov5645_video_ops = {
> > > > > > > > > >       .s_stream = ov5645_s_stream,
> > > > > > > > > >  };
> > > > > > > > > >
> > > > > > > > > > +static const struct v4l2_subdev_core_ops ov5645_subdev_core_ops = {
> > > > > > > > > > +     .subscribe_event = v4l2_ctrl_subdev_subscribe_event,
> > > > > > > > > > +     .unsubscribe_event = v4l2_event_subdev_unsubscribe,
> > > > > > > > > > +};
> > > > > > > > > > +
> > > > > > > > > >  static const struct v4l2_subdev_pad_ops ov5645_subdev_pad_ops = {
> > > > > > > > > >       .enum_mbus_code = ov5645_enum_mbus_code,
> > > > > > > > > >       .enum_frame_size = ov5645_enum_frame_size,
> > > > > > > > > > @@ -1043,6 +1049,7 @@ static const struct v4l2_subdev_pad_ops ov5645_subdev_pad_ops = {
> > > > > > > > > >  };
> > > > > > > > > >
> > > > > > > > > >  static const struct v4l2_subdev_ops ov5645_subdev_ops = {
> > > > > > > > > > +     .core = &ov5645_subdev_core_ops,
> > > > > > > > > >       .video = &ov5645_video_ops,
> > > > > > > > > >       .pad = &ov5645_subdev_pad_ops,
> > > > > > > > > >  };
> > > > > > > > > > @@ -1178,7 +1185,8 @@ static int ov5645_probe(struct i2c_client *client)
> > > > > > > > > >
> > > > > > > > > >       v4l2_i2c_subdev_init(&ov5645->sd, client, &ov5645_subdev_ops);
> > > > > > > > > >       ov5645->sd.internal_ops = &ov5645_internal_ops;
> > > > > > > > > > -     ov5645->sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE;
> > > > > > > > > > +     ov5645->sd.flags |= V4L2_SUBDEV_FL_HAS_DEVNODE |
> > > > > > > > > > +                         V4L2_SUBDEV_FL_HAS_EVENTS;
> > > > > > > > >
> > > > > > > > > Instead of patching every subdev driver, should we handle all of this in
> > > > > > > > > the subdev core ? If a control handler is set for the subdev, we could
> > > > > > > > > set the HAS_EVENTS flag automatically, and default to
> > > > > > > > > v4l2_ctrl_subdev_subscribe_event() and v4l2_event_subdev_unsubscribe()
> > > > > > > > > if there are no control operations.
> > > > > > >
> > > > > > > Premit:
> > > > > > >  - Don't know if I'm wrong eh.
> > > > > >
> > > > > > Nobody knows :-)
> > > > > >
> > > > > > > This can be done into:
> > > > > > >
> > > > > > > __v4l2_subdev_init_finalize()
> > > > > > >
> > > > > > > Adding:
> > > > > > >
> > > > > > >     if (sd->ctrl_handler)
> > > > > > >             sd->flags |= V4L2_SUBDEV_FL_HAS_EVENTS;
> > > > > > >
> > > > > > > And check if there are no control operations using:
> > > > > > >
> > > > > > > bool has_subscribe_event;
> > > > > > > bool has_unsubscribe_event;
> > > > > > >
> > > > > > >
> > > > > > > has_subscribe_event = v4l2_subdev_has_op(sd, core, subscribe_event);
> > > > > > > has_unsubscribe_event = v4l2_subdev_has_op(sd, core, unsubscribe_event);
> > > > > > >
> > > > > > > if (!has_subscribe_event)
> > > > > > >     assign v4l2_ctrl_subdev_subscribe_event as default .subscribe ops(somehow)
> > > > > >
> > > > > > We can't change the ops structure as it's constant. Something like this
> > > > > > could do:
> > > > > >
> > > > > > diff --git a/drivers/media/v4l2-core/v4l2-subdev.c b/drivers/media/v4l2-core/v4l2-subdev.c
> > > > > > index 3a4ba08810d2..41ae18a0d41e 100644
> > > > > > --- a/drivers/media/v4l2-core/v4l2-subdev.c
> > > > > > +++ b/drivers/media/v4l2-core/v4l2-subdev.c
> > > > > > @@ -691,10 +691,24 @@ static long subdev_do_ioctl(struct file *file, unsigned int cmd, void *arg,
> > > > > >               return v4l2_event_dequeue(vfh, arg, file->f_flags & O_NONBLOCK);
> > > > > >
> > > > > >       case VIDIOC_SUBSCRIBE_EVENT:
> > > > > > -             return v4l2_subdev_call(sd, core, subscribe_event, vfh, arg);
> > > > > > +             if (v4l2_subdev_has_op(sd, core, subscribe_event))
> > > > > > +                     return v4l2_subdev_call(sd, core, subscribe_event, vfh,
> > > > > > +                                             arg);
> > > > > > +             else if ((sd->flags & V4L2_SUBDEV_FL_HAS_EVENTS) &&
> > > > > > +                      vfh->ctrl_handler)
> > > > > > +                     return v4l2_ctrl_subdev_subscribe_event(sd, vfh, arg);
> > > > > > +             else
> > > > > > +                     return -ENOIOCTLCMD;
> > > > > >
> > > > > >       case VIDIOC_UNSUBSCRIBE_EVENT:
> > > > > > -             return v4l2_subdev_call(sd, core, unsubscribe_event, vfh, arg);
> > > > > > +             if (!(sd->flags & V4L2_SUBDEV_FL_HAS_EVENTS))
> > > > > > +                     return -ENOIOCTLCMD;
> > > > > > +
> > > > > > +             if (v4l2_subdev_has_op(sd, core, unsubscribe_event))
> > > > > > +                     return v4l2_subdev_call(sd, core, unsubscribe_event,
> > > > > > +                                             vfh, arg);
> > > > > > +             else
> > > > > +                       return v4l2_event_subdev_unsubscribe(sd, vfh, arg);
> > > > >
> > > > > Thanks for your "more than an hint :)"
> > > > > I'm able to test this on ov5645:
> > > > >
> > > > > Adding:
> > > > >
> > > > > +++ b/drivers/media/i2c/ov5645.c
> > > > > @@ -1238,6 +1238,12 @@ static int ov5645_probe(struct i2c_client *client)
> > > > >
> > > > >         ov5645_init_state(&ov5645->sd, NULL);
> > > > >
> > > > > +       ret = v4l2_subdev_init_finalize(&ov5645->sd);
> > > > > +       if (ret < 0) {
> > > > > +               dev_err(dev, "subdev initialization error %d\n", ret);
> > > > > +               goto err_free_state;
> > > > > +       }
> > > > > +
> > > > >         ret = v4l2_async_register_subdev(&ov5645->sd);
> > > > >         if (ret < 0) {
> > > > >                 dev_err(dev, "could not register v4l2 device\n");
> > > > > @@ -1251,6 +1257,8 @@ static int ov5645_probe(struct i2c_client *client)
> > > > >
> > > > >         return 0;
> > > > >
> > > > > +err_free_state:
> > > > > +       v4l2_subdev_cleanup(&ov5645->sd);
> > > > >  err_pm_runtime:
> > > > >         pm_runtime_disable(dev);
> > > > >         pm_runtime_put_noidle(dev);
> > > > > @@ -1272,6 +1280,7 @@ static void ov5645_remove(struct i2c_client *client)
> > > > >
> > > > >         v4l2_async_unregister_subdev(&ov5645->sd);
> > > > >         media_entity_cleanup(&ov5645->sd.entity);
> > > > > +       v4l2_subdev_cleanup(&ov5645->sd);
> > > > >         v4l2_ctrl_handler_free(&ov5645->ctrls);
> > > > >         pm_runtime_disable(ov5645->dev);
> > > > >         if (!pm_runtime_status_suspended(ov5645->dev))
> > > > >
> > > > > Then from the compliance tool I'm getting now good results:
> > > > >
> > > > > Total for device /dev/v4l-subdev1: 44, Succeeded: 44, Failed: 0, Warnings: 0
> > > > >
> > > > > I will send these 2 patches later if you agree (1 v4l2-subdev 1 ov5645.c)
> > > > > Thanks again.
> > > > >
> > > > Thank you for the patch.
> > > > 
> > > > I am currently working on adding support for V4L2_SUBDEV_FL_HAS_EVENTS
> > > > and subscribe hooks[1] (and some more features [0] for ov5645 driver),
> > > > since the patch series adds internal pad which needs rework based on
> > > > patch series from Sakari which I will do soon and send v3 patches for
> > > > ov5645 driver.
> > > > 
> > > > [0] https://lore.kernel.org/all/20240910170610.226189-1-prabhakar.mahadev-lad.rj@bp.renesas.com/
> > > > [1] https://lore.kernel.org/all/20240910170610.226189-2-prabhakar.mahadev-lad.rj@bp.renesas.com/
> > > 
> > > Thanks for sharing this.
> > > Nice! then I can drop ov5645 patch and send only v4l2-subdev patch :)
> > > 
> > > Plan as suggested by Laurent is:
> > > 
> > > "Instead of patching every subdev driver, should we handle all of this in
> > > the subdev core ? If a control handler is set for the subdev, we could
> > > set the HAS_EVENTS flag automatically, and default to
> > > v4l2_ctrl_subdev_subscribe_event() and v4l2_event_subdev_unsubscribe()
> > > if there are no control operations"
> > > 
> > > Using:
> > > 
> > > diff --git a/drivers/media/v4l2-core/v4l2-subdev.c b/drivers/media/v4l2-core/v4l2-subdev.c
> > > index 3a4ba08810d2..fe584998f5e6 100644
> > > --- a/drivers/media/v4l2-core/v4l2-subdev.c
> > > +++ b/drivers/media/v4l2-core/v4l2-subdev.c
> > > @@ -691,10 +691,24 @@ static long subdev_do_ioctl(struct file *file, unsigned int cmd, void *arg,
> > >                 return v4l2_event_dequeue(vfh, arg, file->f_flags & O_NONBLOCK);
> > > 
> > >         case VIDIOC_SUBSCRIBE_EVENT:
> > > -               return v4l2_subdev_call(sd, core, subscribe_event, vfh, arg);
> > > +               if (v4l2_subdev_has_op(sd, core, subscribe_event))
> > > +                       return v4l2_subdev_call(sd, core, subscribe_event,
> > > +                                               vfh, arg);
> > > +               else if ((sd->flags & V4L2_SUBDEV_FL_HAS_EVENTS) &&
> > 
> > Useless use of else. Same below.
> 
> Better:
> 
> 	case VIDIOC_SUBSCRIBE_EVENT:
> 		if (v4l2_subdev_has_op(sd, core, subscribe_event))
> 			return v4l2_subdev_call(sd, core, subscribe_event,
> 						vfh, arg);
> 
> 		return v4l2_ctrl_subdev_subscribe_event(sd, vfh, arg);
> 
> 	case VIDIOC_UNSUBSCRIBE_EVENT:
> 		if (v4l2_subdev_has_op(sd, core, unsubscribe_event))
> 			return v4l2_subdev_call(sd, core, unsubscribe_event,
> 						vfh, arg);
> 
> 		return v4l2_event_subdev_unsubscribe(sd, vfh, arg);
> 
> Because V4L2_SUBDEV_FL_HAS_EVENTS and vfh->ctrl_handler are already
> check right?

v4l2_ctrl_subdev_{,un}subscribe_event() does check for the presence of the
control handler, but return -EINVAL if one doesn't exist. We should however
return -ENOIOCTLCMD, to indicate there's no such IOCTL if the HAS_EVENTS
flag is missing.

> > 
> > > +                       vfh->ctrl_handler)
> > > +                       return v4l2_ctrl_subdev_subscribe_event(sd, vfh, arg);
> > > +               else
> > > +                       return -ENOIOCTLCMD;
> > > 
> > >         case VIDIOC_UNSUBSCRIBE_EVENT:
> > > -               return v4l2_subdev_call(sd, core, unsubscribe_event, vfh, arg);
> > > +               if (!(sd->flags & V4L2_SUBDEV_FL_HAS_EVENTS))
> > > +                       return -ENOIOCTLCMD;
> > > +
> > > +               if (v4l2_subdev_has_op(sd, core, unsubscribe_event))
> > > +                       return v4l2_subdev_call(sd, core, unsubscribe_event,
> > > +                                               vfh, arg);
> > > +               else
> > > +                       return v4l2_event_subdev_unsubscribe(sd, vfh, arg);
> > > 
> > >  #ifdef CONFIG_VIDEO_ADV_DEBUG
> > >         case VIDIOC_DBG_G_REGISTER:
> > > @@ -1641,6 +1655,9 @@ int __v4l2_subdev_init_finalize(struct v4l2_subdev *sd, const char *name,
> > >                 }
> > >         }
> > > 
> > > +       if (sd->ctrl_handler)
> > > +               sd->flags |= V4L2_SUBDEV_FL_HAS_EVENTS;
> > > +
> > >         state = __v4l2_subdev_state_alloc(sd, name, key);
> > >         if (IS_ERR(state))
> > >                 return PTR_ERR(state);
> > > 
> > > In theory v4l2_subdev_init_finalize() and subdev_do_ioctl() will handle
> > > this.
> > > 
> > > Please Laurent correct me if I'm wrong.
> > > Prabhakar what do you think?
> > > Thanks in advance.
> > 
> > Looks good to me, with the above comment.

-- 
Kind regards,

Sakari Ailus