If an entity has sink pads, at least one of them must be connected to another pad with an enabled link. If a driver with multiple sink pads has more strict requirements the check should be done in the driver itself.
Just requiring one sink pad is connected with an enabled link is enough API-wise: entities with sink pads with only disabled links should not be allowed to stream in the first place, but also in a different operation mode a device might require only one of its pads connected with an active link.
If an entity has an ability to function as a source entity another logical entity connected to the aforementioned one should be used for the purpose.
Signed-off-by: Sakari Ailus sakari.ailus@iki.fi --- drivers/media/media-entity.c | 16 +++++++++++++--- 1 files changed, 13 insertions(+), 3 deletions(-)
diff --git a/drivers/media/media-entity.c b/drivers/media/media-entity.c index e1cd132..8846ea7 100644 --- a/drivers/media/media-entity.c +++ b/drivers/media/media-entity.c @@ -227,6 +227,7 @@ __must_check int media_entity_pipeline_start(struct media_entity *entity, media_entity_graph_walk_start(&graph, entity);
while ((entity = media_entity_graph_walk_next(&graph))) { + bool has_sink = false, active_sink = false; unsigned int i;
entity->stream_count++; @@ -243,18 +244,27 @@ __must_check int media_entity_pipeline_start(struct media_entity *entity, for (i = 0; i < entity->num_links; i++) { struct media_link *link = &entity->links[i];
+ /* Are we the sink or not? */ + if (link->sink->entity != entity) + continue; + + has_sink = true; + /* Is this pad part of an enabled link? */ if (!(link->flags & MEDIA_LNK_FL_ENABLED)) continue;
- /* Are we the sink or not? */ - if (link->sink->entity != entity) - continue; + active_sink = true;
ret = entity->ops->link_validate(link); if (ret < 0 && ret != -ENOIOCTLCMD) goto error; } + + if (has_sink && !active_sink) { + ret = -EPIPE; + goto error; + } }
mutex_unlock(&mdev->graph_mutex);
Hi all,
Comments would be appreciated, either positive or negative. The omap3isp driver does the same check itself currently, but I think this is more generic than that.
Thanks.
On Fri, Oct 26, 2012 at 10:46:17PM +0300, Sakari Ailus wrote:
Hi Sakari,
On 11/13/2012 03:24 PM, Sakari Ailus wrote:
Why not leave it to individual drivers ? I'm not sure if it is a good idea not to allow an entity with sink pads to be used as a source only. It might be appropriate for most of the cases but likely not all. I'm inclined not to add this requirement in the API. Just my opinion though.
-- Thanks, Sylwester
Hi Sylwester,
On Wednesday 14 November 2012 10:23:19 Sylwester Nawrocki wrote:
I have mixed feelings about this patch too, which is why I've asked Sakari to cross-post it. It's pretty easy to add this check to the core now, but pushing it back to drivers late if we realize it's too restrictive would be difficult. I think my preference would go for a helper function that drivers can use, possibly first waiting until a second driver requires this kind of checks before implementing it.
On Wed, Nov 14, 2012 at 11:58:42AM +0100, Laurent Pinchart wrote:
I'd like to see a driver that doesn't. Quite likelye either it has no configurable links or it's broken. :-)
Hi Sylwester,
Thanks for the comments.
On Wed, Nov 14, 2012 at 10:23:19AM +0100, Sylwester Nawrocki wrote:
I'm just wondering what would be the use case for that.
What comes closest is generating a test pattern, but even that should be a separate subdev: the test pattern can be enabled by enabling the link from the test pattern generator subdev.
As it seems not everyone is outright happy about the idea of making this mandatory, then how about making it optional?
I'd hate having a link validate function for each subdev e.g. in the OMAP 3 ISP driver that just checks that its sink pad is actually connected with an enabled link. That'd be lots of mostly useless code. If this is done in the framework, the drivers will be spared from copying this code in a number of places. Which was why I originally wrote this patch. The alternative is to re-parse the whole graph in the driver which I'd also like to avoid.
One opion I can think of is to call link_validate op of struct media_entity_operations also for disabled links on entities that are connected through active links (on V4L2 to a video node right before streaming, for example).
That'd make it easy to perform the check in the drivers.
What do you think?
Hi Sakari,
On Wednesday 14 November 2012 23:13:45 Sakari Ailus wrote:
That would force creating a separate subdev just to support test pattern generation, I'm not sure if I want to push for that. There might also be other use cases not related to V4L (thus the cross-posting).
Agreed.
I'd also prefer to avoid that *if*possible*, as we already walk the graph during link validation.
I think that would be a bit too complex. Drivers (or the V4L core) would need to gather data from multiple links in some state object to find out if the complete pipeline is valid or not.
Another option would be to set a flag somewhere to indicate whether the check should be performed by the media core or left to drivers. As different types of drivers might need different types of checks, I think I would prefer for now to walk the graph one more time in the OMAP3 ISP driver, as currently done, and revisit this issue when we will have a couple of drivers implementing pipeline validity checks. I'm just a bit uncomfortable adding core code for a feature that has a single user at the moment without a clear view regarding how it would scale.
Hi Laurent,
Thanks for the comments.
On Thu, Nov 15, 2012 at 01:46:23AM +0100, Laurent Pinchart wrote:
The benefit of standard interfaces and uniform behaviour of those interfaces is that you can configure test pattern generation using e.g. media-ctl and yavta. Nothing else; not even private controls are needed.
use cases not related to V4L (thus the cross-posting).
That's certainly possible.
...
Honestly, I think that a driver that has configurable links but does not require this check to be done is quite probably broken. I'm not really convinced there is even a use case for a subdev operating its all sink pads disconnected.
Btw. this "somewhere" could be struct media_pad. It already has a "flags" field. We could call it MEDIA_PAD_FL_MUST_CONNECT.
dri-devel@lists.freedesktop.org
-
Laurent Pinchart -
Sakari Ailus -
Sylwester Nawrocki