On 06/11/2014 02:54 AM, Steve Longerbeam wrote:
> On 06/10/2014 08:11 AM, Hans Verkuil wrote:
>> On 06/09/2014 06:42 PM, Steve Longerbeam wrote:
>>> On 06/08/2014 01:36 AM, Hans Verkuil wrote:
>>>> Hi Steve!
>>>>
>>>> On 06/07/2014 11:56 PM, Steve Longerbeam wrote:
>>>>> Hi all,
>>>>>
>>>>> This patch set adds video capture support for the Freescale i.MX6 SOC.
>>>>>
>>>>> It is a from-scratch standardized driver that works with community
>>>>> v4l2 utilities, such as v4l2-ctl, v4l2-cap, and the v4l2src gstreamer
>>>>> plugin. It uses the latest v4l2 interfaces (subdev, videobuf2).
>>>>> Please see Documentation/video4linux/mx6_camera.txt for it's full list
>>>>> of features!
>>>>>
>>>>> The first 38 patches:
>>>>>
>>>>> - prepare the ipu-v3 driver for video capture support. The current driver
>>>>> contains only video display functionality to support the imx DRM drivers.
>>>>> At some point ipu-v3 should be moved out from under staging/imx-drm since
>>>>> it will no longer only support DRM.
>>>>>
>>>>> - Adds the device tree nodes and OF graph bindings for video capture support
>>>>> on sabrelite, sabresd, and sabreauto reference platforms.
>>>>>
>>>>> The new i.MX6 capture host interface driver is at patch 39.
>>>>>
>>>>> To support the sensors found on the sabrelite, sabresd, and sabreauto,
>>>>> three patches add sensor subdev's for parallel OV5642, MIPI CSI-2 OV5640,
>>>>> and the ADV7180 decoder chip, beginning at patch 40.
>>>>>
>>>>> There is an existing adv7180 subdev driver under drivers/media/i2c, but
>>>>> it needs some extra functionality to work on the sabreauto. It will need
>>>>> OF graph bindings support and gpio for a power-on pin on the sabreauto.
>>>>> It would also need to send a new subdev notification to take advantage
>>>>> of decoder status change handling provided by the host driver. This
>>>>> feature makes it possible to correctly handle "hot" (while streaming)
>>>>> signal lock/unlock and autodetected video standard changes.
>>>> A new V4L2_EVENT_SOURCE_CHANGE event has just been added for that.
>>>
>>> Hello Hans!
>>>
>>> Ok, V4L2_EVENT_SOURCE_CHANGE looks promising.
>>>
>>> But v4l2-framework.txt states that v4l2 events are passed to userland. So
>>> I want to make sure this framework will also work internally; that is,
>>> the decoder subdev (adv7180) can generate this event and it can be
>>> subscribed by the capture host driver. That it can be passed to userland
>>> is fine and would be useful, it's not necessary in this case.
>>
>> A subdevice can notify its parent device through v4l2_subdev_notify (see
>> v4l2-device.h). It would be nice if the event API and this notify mechanism
>> were unified or at least be more similar.
>>
>> It's on my TODO list, but it is fairly far down that list.
>>
>> Let me know if you are interested to improve this situation. I should be able
>> to give some pointers.
>>
>>
>
> Hi Hans,
>
> It doesn't look possible for subdev's to queue events, since events require
> a v4l2 fh context, so it looks like subdev notifications should stick
> around.
>
> But perhaps subdev notification can be modified to send a struct v4l2_event,
> rather than a u32 notification value. Then at least notify and events can
> share event types instead of duplicating them (such as what I caused by
> introducing this similar decoder status change notification).
>
> Something like:
>
> /* Send an event to v4l2_device. */
> static inline void v4l2_subdev_notify(struct v4l2_subdev *sd,
> struct v4l2_event *ev,
> void *arg)
> {
> if (sd && sd->v4l2_dev && sd->v4l2_dev->notify)
> sd->v4l2_dev->notify(sd, ev, arg);
> }
>
>
> Then the parent is free to consume this event internally, and/or queue it
> off to userland via v4l2_event_queue().
>
> Notification values could then completely disappear, replaced with new (or
> existing) event types.
>
> Is that what you had in mind, or something completely different?
That is what I had in mind, although the 'arg' argument can be dropped, since
that will be part of the event payload.
There is another notifier function as well in v4l2-ctrls.h: v4l2_ctrl_notify().
This should also switch to using v4l2_subdev_notify(). Instead of setting a
notify callback it should set a v4l2_subdev pointer: that will allow it to
call v4l2_subdev_notify().
Where things get tricky is with the event payload: there is no problem if it
is one of the standard events, but for kernel-internal events it is not quite
clear how those should be defined. Ideally you would like to be able to add
new structs inside the v4l2_event union, but you don't want those to be part
of the public API either.
I'm thinking something like this:
// Range for events that are internal to the kernel
#define V4L2_EVENT_INTERNAL_START 0x04000000
And internal events and associated payload structs are defined in v4l2-event.h.
Add this macro to v4l2-event.h:
#define V4L2_EVENT_CAST(ev, type) ((type *)(&(ev)->u))
Now you should be able to do something like this:
In v4l2-event.h:
#define V4L2_EVENT_INT_FOO (V4L2_EVENT_INTERNAL_START + 0)
struct v4l2_event_int_foo {
u32 val;
};
In the driver:
struct v4l2_event ev;
struct v4l2_event_int_foo *f = V4L2_EVENT_CAST(&ev, struct v4l2_event_int_foo);
f->val = 10;
// Or alternatively:
*V4L2_EVENT_CAST(&ev, u32) = 10;
It's the best I can come up with.
It would be nice to use the same v4l2_event struct for all notifications, it's much more
unified and elegant.
Regards,
Hans
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
