[PATCH v7 25/44] [media] replace all occurrences of MEDIA_ENT_T_DEVNODE_V4L

Mauro Carvalho Chehab mchehab at osg.samsung.com
Tue Aug 25 08:12:29 PDT 2015


Em Tue, 25 Aug 2015 15:54:09 +0200
Hans Verkuil <hverkuil at xs4all.nl> escreveu:

> On 08/25/15 13:32, Mauro Carvalho Chehab wrote:
> > Em Tue, 25 Aug 2015 11:23:24 +0200
> > Hans Verkuil <hverkuil at xs4all.nl> escreveu:
> > 
> >> On 08/23/15 22:17, Mauro Carvalho Chehab wrote:
> >>> Now that interfaces and entities are distinct, it makes no sense
> >>> of keeping something named as MEDIA_ENT_T_DEVNODE.
> >>>
> >>> This change was done with this script:
> >>>
> >>> 	for i in $(git grep -l MEDIA_ENT_T|grep -v uapi/linux/media.h); do sed s,MEDIA_ENT_T_DEVNODE_V4L,MEDIA_ENT_T_V4L2_VIDEO, <$i >a && mv a $i; done
> >>>
> >>> Signed-off-by: Mauro Carvalho Chehab <mchehab at osg.samsung.com>
> >>>
> >>> diff --git a/Documentation/DocBook/media/v4l/media-ioc-enum-entities.xml b/Documentation/DocBook/media/v4l/media-ioc-enum-entities.xml
> >>> index 5872f8bbf774..910243d4edb8 100644
> >>> --- a/Documentation/DocBook/media/v4l/media-ioc-enum-entities.xml
> >>> +++ b/Documentation/DocBook/media/v4l/media-ioc-enum-entities.xml
> >>> @@ -183,7 +183,7 @@
> >>>  	    <entry>Unknown device node</entry>
> >>>  	  </row>
> >>>  	  <row>
> >>> -	    <entry><constant>MEDIA_ENT_T_DEVNODE_V4L</constant></entry>
> >>> +	    <entry><constant>MEDIA_ENT_T_V4L2_VIDEO</constant></entry>
> >>>  	    <entry>V4L video, radio or vbi device node</entry>
> >>>  	  </row>
> >>
> >> OK, this makes no sense and that ties in with my confusion of the previous patch.
> >>
> >> These are not device nodes, in the new scheme these are DMA entities (I know,
> >> naming TDB) that have an associated interface.
> > 
> > Yes. Well, DMA is a bad name. It won't cover USB devices, where the DMA
> > engine is outside the V4L2 drivers, nor it would work for RDS radio data,
> > with may not need any DMA at all on no-USB devices, as the data flows via
> > the I2C bus.
> > 
> >> I think a much better approach would be to add entity type(s) for such DMA
> >> engines in patch 24, then use that new name in existing drivers and split
> >> up the existing DEVNODE_V4L media_entity into a media_entity and a
> >> media_intf_devnode:
> > 
> > Sorry, but I didn't get. That's precisely what I did ;)
> > 
> >> The current media_entity defined in struct video_device has to be replaced
> >> by media_intf_devnode, and the DMA entity has to be added as a new entity
> >> to these drivers.
> > 
> > If I do this way, it would break bisectability. I need first to replace
> > the names, but keep them as entities, and then add the interfaces.
> > 
> >>
> >> This reflects these two action items from our meeting:
> >>
> >> Migration: add v4l-subdev media_interface: Laurent
> >> Migration: add explicit DMA Engine entity: Laurent
> >>
> >> Unless Laurent says differently I think this is something you'll have to
> >> do given Laurent's workload.
> > 
> > Yes. The above action items are covered on this series.
> > 
> > What patch 24 does is to define the new namespace, moving the legacy
> > symbols kept due to backward compatibility on a separate part of the
> > header.
> > 
> > Then, patches 25-38 replace the occurrences of the deprecated names
> > by the new ones.
> > 
> > Nothing is touched at the interfaces yet, to avoid breaking bisectability.
> 
> I don't follow why that would break bisect.

It won't break compilation, but it will break runtime.

I mean: if we replace the current occurrences of the
"video output data entities" [1] any userspace app that would be used to test 
somethingwill stop working.

Ok, that means that it would break bisectability for us ;)
Still, better to avoid.

[1] I don't like the "DMA" entities term, as it is too broken.
I prefer to refer to them with some other name, like I/O entities. 
However, even this name is not perfect. Those are, in reality, a
"data interface", while what we call interface is actually a 
"control interface", but calling like that would be confusing, I think.
So, I'll simply call it as "video/vbi/... output data entities".

> 
> > Then, the next patches add interfaces support at the V4L side.
> 
> So this is not yet included in this patch series? That would explain
> my confusion. If it is, then I need to take another look on Friday.

It is on the 3 patches I sent yesterday, after this patch series:
	https://patchwork.linuxtv.org/patch/31081/
	https://patchwork.linuxtv.org/patch/31082/
	https://patchwork.linuxtv.org/patch/31083/

Please notice that the above patch series is not complete, as
there's something non-trivial to be addressed on non-subdev
V4L2 interfaces: how to create the indirect links.

By indirect links, I meant to refer to the interface links that
don't control an entity directly, but via the internal hardware
control/I2C bus(es).

So, a video interface, on a PC customer's hardware controls not only
the video output data entity, but it also indirectly controls the
tuner and the analog demod. Or, on a webcam hardware, it will also
controls the sensor.

However, on platform drivers, it controls just the
"video output data entity" that is directly associated with it via 
its device node.

We need to add some support to automatically create those links,
once available, but only if the device is a PC customer's hardware.

Btw, that's another reason to postpone it: creating the interfaces
offer this additional challenge, while creating the entities are
easy, as nothing changes there.

Regards,
Mauro



More information about the linux-arm-kernel mailing list