Hi all,
This is the draft version of the agenda for the media workshop in Korea. I am assuming that all presenters are subscribed to this mailinglist.
Note: the MC topics are tentative. I've placed them at the end, since they might change. They also have a tendency to take over the discussion, so let's do the smaller topics first :-)
Mauro, do you know which room we'll be in?
I am assuming that we start at 9 am and continue until approx. 5:30 pm. Do you know when the lunch break is?
I guessed at the duration of the various topics: please check and let me know if you need more/less time!
-----------------------------------------------------------
Draft agenda for the media workshop October 26, 2015, Seoul, South Korea
Please be on time! We start at 9 am, we have a full agenda, so we want to make good use of the time.
9:00-9:15: Get laptops hooked up, the projector running, wireless configured, your coffee cup filled, etc. etc. :-) 9:15-9:30: Hans: Introduction
9:30-10:15: Pawel: Update on codec APIs
10:15-10:30: Mauro: kABI documentation improvements
10:30-11:00: Junghak: new uAPI for DVB to use streaming I/O
11:00-11:15: Break
11:15-11:30: Hans: Proposal for adding support for HW colorspace conversion.
11:30-12:30: Hans: Should we replace v4l2_buffer with a new struct that is y2038-safe and that can be cleaned up and extended? (Think proper support for HW timestamps, etc.). See: https://patchwork.linuxtv.org/patch/31413/
And perhaps use this redesign to add support for multiple streams over one video node? I saw an example where the HW combines two streams into one, basically an m2m device, but today you'd need three nodes: two out and one capture in order to implement this.
12:30-13:30: Lunch
13:30-14:00: Hans: Status update of the CEC framework patch series. Discuss if we should really integrate with the RC core (not convinced about this).
14:00-14:30: Hans: Should the v4l2 core just call try_format to validate the format passed in by VIDIOC_CREATE_BUFFERS? Few drivers validate the format and if they do they usually do only partial validation. It's unclear what should happen here. Always calling try_format would also help the vb2 core code.
MC discussions: very tentative, and I'm just listing the topics without scheduling them and in no particular order:
Mauro: Discuss dynamic subtree removal
Hans: How to implement VIDIOC_SUBDEV_QUERYCAP?
Shuah: ENT_F_IO appears to be too generic, requiring additional steps to determine what kind of IO function it is really is. I propose preserving and allowing enough uniqueness to enable drivers to tag the functions when they create them. For example, if AUDIO functions are all tagged ENT_F_IO, bridge driver has to determine what kind of IO it really is. This requires finding the Interface entity associated with the function entity or comparing entity name strings. Instead, if a function can be tagged as Audio Capture when it gets created, these additional steps aren't necessary.
Shuah: Discuss snd-usb-audio control interface mapping to media. snd-usb-audio creates a control device file which could be exposed in a media graph. In some cases, it is a simple mixer control. Expose ENT_F_MIXER entity with an associated interface node MEDIA_INTF_T_ALSA_CONTROL
17:15-17:30: Hans: Discuss future improvements to the workshop format. Right now (as is clear from this schedule) the discussions are very technical, which makes it hard for non-core developers to attend. Perhaps we should do a split format: a half or full day of status updates, discussions to improve application integration, followed by core-developer-only discussions.
Regards,
Hans
Mauro, Pawel, Junghak, Shuah, Laurent,
Please check the time I've guesstimated for your topics and let me know if you need more/less time!
Shuah/Mauro: which MC topics should we do in the workshop? We have already a lot of topics, so dropping one or more MC topics if they are premature would help.
Mauro: I do think it will be useful if you do an overview of the current status of the MC work, that will certainly be of interest of non-core developers that are attending, and probably even of the core developers.
Also one additional topic:
Hans/Laurent:
How should poll() work for output streams: for write() it should block if there are no internal buffers available, but for stream I/O it makes more sense to block if there are no buffers to dequeue. Currently for stream I/O it blocks if all buffers are in use (i.e. the write() case). Do we need a separate V4L2_EVENT to signal whether a buffer can be dequeued? Or change the poll semantics?
Time: 30 min
Regards,
Hans
On 10/12/15 16:25, Hans Verkuil wrote:
Hi all,
This is the draft version of the agenda for the media workshop in Korea. I am assuming that all presenters are subscribed to this mailinglist.
Note: the MC topics are tentative. I've placed them at the end, since they might change. They also have a tendency to take over the discussion, so let's do the smaller topics first :-)
Mauro, do you know which room we'll be in?
I am assuming that we start at 9 am and continue until approx. 5:30 pm. Do you know when the lunch break is?
I guessed at the duration of the various topics: please check and let me know if you need more/less time!
Draft agenda for the media workshop October 26, 2015, Seoul, South Korea
Please be on time! We start at 9 am, we have a full agenda, so we want to make good use of the time.
9:00-9:15: Get laptops hooked up, the projector running, wireless configured, your coffee cup filled, etc. etc. :-) 9:15-9:30: Hans: Introduction
9:30-10:15: Pawel: Update on codec APIs
10:15-10:30: Mauro: kABI documentation improvements
10:30-11:00: Junghak: new uAPI for DVB to use streaming I/O
11:00-11:15: Break
11:15-11:30: Hans: Proposal for adding support for HW colorspace conversion.
11:30-12:30: Hans: Should we replace v4l2_buffer with a new struct that is y2038-safe and that can be cleaned up and extended? (Think proper support for HW timestamps, etc.). See: https://patchwork.linuxtv.org/patch/31413/
And perhaps use this redesign to add support for multiple streams over one video node? I saw an example where the HW combines two streams into one, basically an m2m device, but today you'd need three nodes: two out and one capture in order to implement this.
12:30-13:30: Lunch
13:30-14:00: Hans: Status update of the CEC framework patch series. Discuss if we should really integrate with the RC core (not convinced about this).
14:00-14:30: Hans: Should the v4l2 core just call try_format to validate the format passed in by VIDIOC_CREATE_BUFFERS? Few drivers validate the format and if they do they usually do only partial validation. It's unclear what should happen here. Always calling try_format would also help the vb2 core code.
MC discussions: very tentative, and I'm just listing the topics without scheduling them and in no particular order:
Mauro: Discuss dynamic subtree removal
Hans: How to implement VIDIOC_SUBDEV_QUERYCAP?
Shuah: ENT_F_IO appears to be too generic, requiring additional steps to determine what kind of IO function it is really is. I propose preserving and allowing enough uniqueness to enable drivers to tag the functions when they create them. For example, if AUDIO functions are all tagged ENT_F_IO, bridge driver has to determine what kind of IO it really is. This requires finding the Interface entity associated with the function entity or comparing entity name strings. Instead, if a function can be tagged as Audio Capture when it gets created, these additional steps aren't necessary.
Shuah: Discuss snd-usb-audio control interface mapping to media. snd-usb-audio creates a control device file which could be exposed in a media graph. In some cases, it is a simple mixer control. Expose ENT_F_MIXER entity with an associated interface node MEDIA_INTF_T_ALSA_CONTROL
17:15-17:30: Hans: Discuss future improvements to the workshop format. Right now (as is clear from this schedule) the discussions are very technical, which makes it hard for non-core developers to attend. Perhaps we should do a split format: a half or full day of status updates, discussions to improve application integration, followed by core-developer-only discussions.
Regards,
Hans
media-workshop mailing list media-workshop@linuxtv.org http://www.linuxtv.org/cgi-bin/mailman/listinfo/media-workshop
On 10/15/2015 01:43 AM, Hans Verkuil wrote:
Mauro, Pawel, Junghak, Shuah, Laurent,
Thanks for pulling this together.
Please check the time I've guesstimated for your topics and let me know if you need more/less time!
Shuah/Mauro: which MC topics should we do in the workshop? We have already a lot of topics, so dropping one or more MC topics if they are premature would help.
There is a substantial amount of work that is already complete and could benefit from a focussed review at the work-shop. This includes the MC Next Gen work Mauro/Javier did as well as the ALSA MC Next Gen work I did.
Mauro: I do think it will be useful if you do an overview of the current status of the MC work, that will certainly be of interest of non-core developers that are attending, and probably even of the core developers.
Also one additional topic:
Hans/Laurent:
How should poll() work for output streams: for write() it should block if there are no internal buffers available, but for stream I/O it makes more sense to block if there are no buffers to dequeue. Currently for stream I/O it blocks if all buffers are in use (i.e. the write() case). Do we need a separate V4L2_EVENT to signal whether a buffer can be dequeued? Or change the poll semantics?
Time: 30 min
Regards,
Hans
On 10/12/15 16:25, Hans Verkuil wrote:
Hi all,
This is the draft version of the agenda for the media workshop in Korea. I am assuming that all presenters are subscribed to this mailinglist.
Note: the MC topics are tentative. I've placed them at the end, since they might change. They also have a tendency to take over the discussion, so let's do the smaller topics first :-)
Mauro, do you know which room we'll be in?
I am assuming that we start at 9 am and continue until approx. 5:30 pm. Do you know when the lunch break is?
I guessed at the duration of the various topics: please check and let me know if you need more/less time!
Draft agenda for the media workshop October 26, 2015, Seoul, South Korea
Please be on time! We start at 9 am, we have a full agenda, so we want to make good use of the time.
9:00-9:15: Get laptops hooked up, the projector running, wireless configured, your coffee cup filled, etc. etc. :-) 9:15-9:30: Hans: Introduction
9:30-10:15: Pawel: Update on codec APIs
10:15-10:30: Mauro: kABI documentation improvements
10:30-11:00: Junghak: new uAPI for DVB to use streaming I/O
11:00-11:15: Break
I will be away giving a talk on the ALSA MC Next Gen work from 11:15 - 12:05
http://korealinuxforum2015.sched.org/event/f59175cc49f866dbc9bebc36f86fe663#...
If Media work-shop attendees want to attend the above talk, you will have to adjust the time for that. Might not be a bad idea, this talk will go in depth into the ALSA MC Next Gen work and covers one of the MC Next Gen topics.
11:15-11:30: Hans: Proposal for adding support for HW colorspace conversion.
If you want to attend the above talk, you will have to adjust the time for that.
11:30-12:30: Hans: Should we replace v4l2_buffer with a new struct that is y2038-safe and that can be cleaned up and extended? (Think proper support for HW timestamps, etc.). See: https://patchwork.linuxtv.org/patch/31413/
And perhaps use this redesign to add support for multiple streams over one video node? I saw an example where the HW combines two streams into one, basically an m2m device, but today you'd need three nodes: two out and one capture in order to implement this.
12:30-13:30: Lunch
13:30-14:00: Hans: Status update of the CEC framework patch series. Discuss if we should really integrate with the RC core (not convinced about this).
14:00-14:30: Hans: Should the v4l2 core just call try_format to validate the format passed in by VIDIOC_CREATE_BUFFERS? Few drivers validate the format and if they do they usually do only partial validation. It's unclear what should happen here. Always calling try_format would also help the vb2 core code.
MC discussions: very tentative, and I'm just listing the topics without scheduling them and in no particular order:
Mauro: Discuss dynamic subtree removal
Hans: How to implement VIDIOC_SUBDEV_QUERYCAP?
Shuah: ENT_F_IO appears to be too generic, requiring additional steps to determine what kind of IO function it is really is. I propose preserving and allowing enough uniqueness to enable drivers to tag the functions when they create them. For example, if AUDIO functions are all tagged ENT_F_IO, bridge driver has to determine what kind of IO it really is. This requires finding the Interface entity associated with the function entity or comparing entity name strings. Instead, if a function can be tagged as Audio Capture when it gets created, these additional steps aren't necessary.
I think we could still discuss this even if we decide MC Next Gen topics are premature.
Shuah: Discuss snd-usb-audio control interface mapping to media. snd-usb-audio creates a control device file which could be exposed in a media graph. In some cases, it is a simple mixer control. Expose ENT_F_MIXER entity with an associated interface node MEDIA_INTF_T_ALSA_CONTROL
I think we could still discuss this even if we decide MC Next Gen topics are premature. I am about to two more patches to add mixer and control support.
17:15-17:30: Hans: Discuss future improvements to the workshop format. Right now (as is clear from this schedule) the discussions are very technical, which makes it hard for non-core developers to attend. Perhaps we should do a split format: a half or full day of status updates, discussions to improve application integration, followed by core-developer-only discussions.
Regards,
Hans
media-workshop mailing list media-workshop@linuxtv.org http://www.linuxtv.org/cgi-bin/mailman/listinfo/media-workshop
Em Thu, 15 Oct 2015 08:16:37 -0600 Shuah Khan shuahkh@osg.samsung.com escreveu:
On 10/15/2015 01:43 AM, Hans Verkuil wrote:
Mauro, Pawel, Junghak, Shuah, Laurent,
Thanks for pulling this together.
Please check the time I've guesstimated for your topics and let me know if you need more/less time!
Shuah/Mauro: which MC topics should we do in the workshop? We have already a lot of topics, so dropping one or more MC topics if they are premature would help.
There is a substantial amount of work that is already complete and could benefit from a focussed review at the work-shop. This includes the MC Next Gen work Mauro/Javier did as well as the ALSA MC Next Gen work I did.
The workshop is not the place for patch review. It should be done before its date, if we want to have further discussions about MC.
Mauro: I do think it will be useful if you do an overview of the current status of the MC work, that will certainly be of interest of non-core developers that are attending, and probably even of the core developers.
We could do that, but what's there is basically what it was discussed at the MC summit. There are a few implementation details that are been discussed by e-mail during patch review, but, except if I misread the review comments, none of them are relevant enough that would justify a F2F discussion.
So, at least from the feedback I had so far, we're all at the same page with regards to MC next gen.
Am I wrong? If so, people should rise their (virtual) hand and drop some comments to this thread explaining what are the remaining things with the already-sent patches that, on their opinion, are not ok, in order for for us to prepare for the discussions and try to properly address them.
Also one additional topic:
Hans/Laurent:
How should poll() work for output streams: for write() it should block if there are no internal buffers available, but for stream I/O it makes more sense to block if there are no buffers to dequeue. Currently for stream I/O it blocks if all buffers are in use (i.e. the write() case). Do we need a separate V4L2_EVENT to signal whether a buffer can be dequeued? Or change the poll semantics?
Time: 30 min
Regards,
Hans
On 10/12/15 16:25, Hans Verkuil wrote:
Hi all,
This is the draft version of the agenda for the media workshop in Korea. I am assuming that all presenters are subscribed to this mailinglist.
Note: the MC topics are tentative. I've placed them at the end, since they might change. They also have a tendency to take over the discussion, so let's do the smaller topics first :-)
Mauro, do you know which room we'll be in?
Not yet.
I am assuming that we start at 9 am and continue until approx. 5:30 pm. Do you know when the lunch break is?
No. I'll ping LF early next week to try to get more info about those. I'm out of the office today, on a 3 days trip.
I guessed at the duration of the various topics: please check and let me know if you need more/less time!
Draft agenda for the media workshop October 26, 2015, Seoul, South Korea
Please be on time! We start at 9 am, we have a full agenda, so we want to make good use of the time.
9:00-9:15: Get laptops hooked up, the projector running, wireless configured, your coffee cup filled, etc. etc. :-) 9:15-9:30: Hans: Introduction
9:30-10:15: Pawel: Update on codec APIs
10:15-10:30: Mauro: kABI documentation improvements
10:30-11:00: Junghak: new uAPI for DVB to use streaming I/O
11:00-11:15: Break
I will be away giving a talk on the ALSA MC Next Gen work from 11:15 - 12:05
http://korealinuxforum2015.sched.org/event/f59175cc49f866dbc9bebc36f86fe663#...
If Media work-shop attendees want to attend the above talk, you will have to adjust the time for that. Might not be a bad idea, this talk will go in depth into the ALSA MC Next Gen work and covers one of the MC Next Gen topics.
11:15-11:30: Hans: Proposal for adding support for HW colorspace conversion.
If you want to attend the above talk, you will have to adjust the time for that.
11:30-12:30: Hans: Should we replace v4l2_buffer with a new struct that is y2038-safe and that can be cleaned up and extended? (Think proper support for HW timestamps, etc.). See: https://patchwork.linuxtv.org/patch/31413/
And perhaps use this redesign to add support for multiple streams over one video node? I saw an example where the HW combines two streams into one, basically an m2m device, but today you'd need three nodes: two out and one capture in order to implement this.
12:30-13:30: Lunch
13:30-14:00: Hans: Status update of the CEC framework patch series. Discuss if we should really integrate with the RC core (not convinced about this).
14:00-14:30: Hans: Should the v4l2 core just call try_format to validate the format passed in by VIDIOC_CREATE_BUFFERS? Few drivers validate the format and if they do they usually do only partial validation. It's unclear what should happen here. Always calling try_format would also help the vb2 core code.
MC discussions: very tentative, and I'm just listing the topics without scheduling them and in no particular order:
Mauro: Discuss dynamic subtree removal
I guess we should at least discuss about SETUP_LINK_V2 and how this would work in the case of dynamic topology changes. Well, we could do that by email, instead, as I think this won't be hard to discuss.
The way I see, the client should pass the topology version to setup link. If the Kernel's topology version doesn't match, Kernel would return an error.
This way, there will be no risk of trying to setup a link that was already gone.
I don't see the need of any other change there. So, setup_link v2 would be:
struct media_setup_link_v2 { __u32 topology_version; __u32 link_id; __u32 flags; __u32 reserved[5]; };
#define MEDIA_IOC_SETUP_LINK_V2 _IOWR('|', 0x05, struct media_setup_link_v2)
If we can agree with that, then we could remove this talk from the contents of the workshop.
Hans: How to implement VIDIOC_SUBDEV_QUERYCAP?
Shuah: ENT_F_IO appears to be too generic, requiring additional steps to determine what kind of IO function it is really is. I propose preserving and allowing enough uniqueness to enable drivers to tag the functions when they create them. For example, if AUDIO functions are all tagged ENT_F_IO, bridge driver has to determine what kind of IO it really is. This requires finding the Interface entity associated with the function entity or comparing entity name strings. Instead, if a function can be tagged as Audio Capture when it gets created, these additional steps aren't necessary.
I think we could still discuss this even if we decide MC Next Gen topics are premature.
Shuah: Discuss snd-usb-audio control interface mapping to media. snd-usb-audio creates a control device file which could be exposed in a media graph. In some cases, it is a simple mixer control. Expose ENT_F_MIXER entity with an associated interface node MEDIA_INTF_T_ALSA_CONTROL
I think we could still discuss this even if we decide MC Next Gen topics are premature. I am about to two more patches to add mixer and control support.
17:15-17:30: Hans: Discuss future improvements to the workshop format. Right now (as is clear from this schedule) the discussions are very technical, which makes it hard for non-core developers to attend. Perhaps we should do a split format: a half or full day of status updates, discussions to improve application integration, followed by core-developer-only discussions.
Regards,
Hans
media-workshop mailing list media-workshop@linuxtv.org http://www.linuxtv.org/cgi-bin/mailman/listinfo/media-workshop
On 10/16/2015 05:26 AM, Mauro Carvalho Chehab wrote:
Em Thu, 15 Oct 2015 08:16:37 -0600 Shuah Khan shuahkh@osg.samsung.com escreveu:
On 10/15/2015 01:43 AM, Hans Verkuil wrote:
Mauro, Pawel, Junghak, Shuah, Laurent,
Thanks for pulling this together.
Please check the time I've guesstimated for your topics and let me know if you need more/less time!
Shuah/Mauro: which MC topics should we do in the workshop? We have already a lot of topics, so dropping one or more MC topics if they are premature would help.
There is a substantial amount of work that is already complete and could benefit from a focussed review at the work-shop. This includes the MC Next Gen work Mauro/Javier did as well as the ALSA MC Next Gen work I did.
The workshop is not the place for patch review. It should be done before its date, if we want to have further discussions about MC.
Definitely. I am not suggesting reviewing patches. It would be a waste of time. What I meant was to identify and include any issues/problems that come out of a prior review of the patches that could benefit from a face-face to discussion.
thanks, -- Shuah
On 10/16/2015 01:26 PM, Mauro Carvalho Chehab wrote:
Em Thu, 15 Oct 2015 08:16:37 -0600 Shuah Khan shuahkh@osg.samsung.com escreveu:
On 10/15/2015 01:43 AM, Hans Verkuil wrote:
Mauro, Pawel, Junghak, Shuah, Laurent,
Thanks for pulling this together.
Please check the time I've guesstimated for your topics and let me know if you need more/less time!
Shuah/Mauro: which MC topics should we do in the workshop? We have already a lot of topics, so dropping one or more MC topics if they are premature would help.
There is a substantial amount of work that is already complete and could benefit from a focussed review at the work-shop. This includes the MC Next Gen work Mauro/Javier did as well as the ALSA MC Next Gen work I did.
The workshop is not the place for patch review. It should be done before its date, if we want to have further discussions about MC.
Mauro: I do think it will be useful if you do an overview of the current status of the MC work, that will certainly be of interest of non-core developers that are attending, and probably even of the core developers.
We could do that, but what's there is basically what it was discussed at the MC summit. There are a few implementation details that are been discussed by e-mail during patch review, but, except if I misread the review comments, none of them are relevant enough that would justify a F2F discussion.
So, at least from the feedback I had so far, we're all at the same page with regards to MC next gen.
Am I wrong? If so, people should rise their (virtual) hand and drop some comments to this thread explaining what are the remaining things with the already-sent patches that, on their opinion, are not ok, in order for for us to prepare for the discussions and try to properly address them.
This isn't for the core devs that were in Espoo, it is for the other attendees of the workshop who are likely not aware of what is going on or who have not been following the work. At least one (Vinod Koul) joins specifically for the MC part.
A 15 minute update would be welcome, I feel.
Regards,
Hans
Em Mon, 19 Oct 2015 12:58:55 +0200 Hans Verkuil hverkuil@xs4all.nl escreveu:
On 10/16/2015 01:26 PM, Mauro Carvalho Chehab wrote:
Em Thu, 15 Oct 2015 08:16:37 -0600 Shuah Khan shuahkh@osg.samsung.com escreveu:
On 10/15/2015 01:43 AM, Hans Verkuil wrote:
Mauro, Pawel, Junghak, Shuah, Laurent,
Thanks for pulling this together.
Please check the time I've guesstimated for your topics and let me know if you need more/less time!
Shuah/Mauro: which MC topics should we do in the workshop? We have already a lot of topics, so dropping one or more MC topics if they are premature would help.
There is a substantial amount of work that is already complete and could benefit from a focussed review at the work-shop. This includes the MC Next Gen work Mauro/Javier did as well as the ALSA MC Next Gen work I did.
The workshop is not the place for patch review. It should be done before its date, if we want to have further discussions about MC.
Mauro: I do think it will be useful if you do an overview of the current status of the MC work, that will certainly be of interest of non-core developers that are attending, and probably even of the core developers.
We could do that, but what's there is basically what it was discussed at the MC summit. There are a few implementation details that are been discussed by e-mail during patch review, but, except if I misread the review comments, none of them are relevant enough that would justify a F2F discussion.
So, at least from the feedback I had so far, we're all at the same page with regards to MC next gen.
Am I wrong? If so, people should rise their (virtual) hand and drop some comments to this thread explaining what are the remaining things with the already-sent patches that, on their opinion, are not ok, in order for for us to prepare for the discussions and try to properly address them.
This isn't for the core devs that were in Espoo, it is for the other attendees of the workshop who are likely not aware of what is going on or who have not been following the work. At least one (Vinod Koul) joins specifically for the MC part.
A 15 minute update would be welcome, I feel.
Works for me. Another alternative would be to do it on the second day, as a presentation. I'll check with Ted if we have a spare slot there.
Regards,
Hans
media-workshop mailing list media-workshop@linuxtv.org http://www.linuxtv.org/cgi-bin/mailman/listinfo/media-workshop
On Monday 19 October 2015 10:03:06 Mauro Carvalho Chehab wrote:
Em Mon, 19 Oct 2015 12:58:55 +0200 Hans Verkuil escreveu:
On 10/16/2015 01:26 PM, Mauro Carvalho Chehab wrote:
Em Thu, 15 Oct 2015 08:16:37 -0600 Shuah Khan escreveu:
On 10/15/2015 01:43 AM, Hans Verkuil wrote:
Mauro, Pawel, Junghak, Shuah, Laurent,
Thanks for pulling this together.
Please check the time I've guesstimated for your topics and let me know if you need more/less time!
Shuah/Mauro: which MC topics should we do in the workshop? We have already a lot of topics, so dropping one or more MC topics if they are premature would help.
There is a substantial amount of work that is already complete and could benefit from a focussed review at the work-shop. This includes the MC Next Gen work Mauro/Javier did as well as the ALSA MC Next Gen work I did.
The workshop is not the place for patch review. It should be done before its date, if we want to have further discussions about MC.
Mauro: I do think it will be useful if you do an overview of the current status of the MC work, that will certainly be of interest of non-core developers that are attending, and probably even of the core developers.
We could do that, but what's there is basically what it was discussed at the MC summit. There are a few implementation details that are been discussed by e-mail during patch review, but, except if I misread the review comments, none of them are relevant enough that would justify a F2F discussion.
So, at least from the feedback I had so far, we're all at the same page with regards to MC next gen.
Am I wrong? If so, people should rise their (virtual) hand and drop some comments to this thread explaining what are the remaining things with the already-sent patches that, on their opinion, are not ok, in order for for us to prepare for the discussions and try to properly address them.
This isn't for the core devs that were in Espoo, it is for the other attendees of the workshop who are likely not aware of what is going on or who have not been following the work. At least one (Vinod Koul) joins specifically for the MC part.
A 15 minute update would be welcome, I feel.
Works for me. Another alternative would be to do it on the second day, as a presentation. I'll check with Ted if we have a spare slot there.
I'd prefer that if possible, time will be a very scarce resource during the workshop. Thank you for handling this Mauro.
Hi Hans,
On Thursday 15 October 2015 09:43:41 Hans Verkuil wrote:
Mauro, Pawel, Junghak, Shuah, Laurent,
Please check the time I've guesstimated for your topics and let me know if you need more/less time!
Shuah/Mauro: which MC topics should we do in the workshop? We have already a lot of topics, so dropping one or more MC topics if they are premature would help.
Mauro: I do think it will be useful if you do an overview of the current status of the MC work, that will certainly be of interest of non-core developers that are attending, and probably even of the core developers.
Also one additional topic:
Hans/Laurent:
How should poll() work for output streams: for write() it should block if there are no internal buffers available, but for stream I/O it makes more sense to block if there are no buffers to dequeue. Currently for stream I/O it blocks if all buffers are in use (i.e. the write() case). Do we need a separate V4L2_EVENT to signal whether a buffer can be dequeued? Or change the poll semantics?
Time: 30 min
I was about to propose that one, thanks :-)
Regards,
Hans
On 10/12/15 16:25, Hans Verkuil wrote:
Hi all,
This is the draft version of the agenda for the media workshop in Korea. I am assuming that all presenters are subscribed to this mailinglist.
Note: the MC topics are tentative. I've placed them at the end, since they might change. They also have a tendency to take over the discussion, so let's do the smaller topics first :-)
Mauro, do you know which room we'll be in?
I am assuming that we start at 9 am and continue until approx. 5:30 pm. Do you know when the lunch break is?
I guessed at the duration of the various topics: please check and let me know if you need more/less time!
Draft agenda for the media workshop October 26, 2015, Seoul, South Korea
Please be on time! We start at 9 am, we have a full agenda, so we want to make good use of the time.
9:00-9:15: Get laptops hooked up, the projector running, wireless configured,> your coffee cup filled, etc. etc. :-)
9:15-9:30: Hans: Introduction
9:30-10:15: Pawel: Update on codec APIs
10:15-10:30: Mauro: kABI documentation improvements
10:30-11:00: Junghak: new uAPI for DVB to use streaming I/O
11:00-11:15: Break
11:15-11:30: Hans: Proposal for adding support for HW colorspace conversion.
11:30-12:30: Hans: Should we replace v4l2_buffer with a new struct that is
y2038-safe and that can be cleaned up and extended? (Think proper support for HW timestamps, etc.). See: https://patchwork.linuxtv.org/patch/31413/
And perhaps use this redesign to add support for multiple streams over one video node? I saw an example where the HW combines two streams into one, basically an m2m device, but today you'd need three nodes: two out and one capture in order to implement this.
We could squeeze the poll() topic there. We'll need way more than an hour to finalize a solution, but discussing the design and agreeing on a feature set should hopefully not take the full hour.
12:30-13:30: Lunch
13:30-14:00: Hans: Status update of the CEC framework patch series. Discuss if we should really integrate with the RC core (not convinced about this).
14:00-14:30: Hans: Should the v4l2 core just call try_format to validate the format passed in by VIDIOC_CREATE_BUFFERS? Few drivers validate the format and if they do they usually do only partial validation. It's unclear what should happen here. Always calling try_format would also help the vb2 core code.
I'd start by trying to clarify the VIDIOC_CREATE_BUFFERS API, with a bit of luck whether the V4L2 core should call try_format will be a direct consequence of the discussion. If time permits I'd like to get opinions on a possible VIDIOC_DELETE_BUFFERS ioctl.
MC discussions: very tentative, and I'm just listing the topics without scheduling them and in no particular order:
Mauro: Discuss dynamic subtree removal
Hans: How to implement VIDIOC_SUBDEV_QUERYCAP?
Shuah: ENT_F_IO appears to be too generic, requiring additional steps to determine what kind of IO function it is really is. I propose preserving and allowing enough uniqueness to enable drivers to tag the functions when they create them. For example, if AUDIO functions are all tagged ENT_F_IO, bridge driver has to determine what kind of IO it really is. This requires finding the Interface entity associated with the function entity or comparing entity name strings. Instead, if a function can be tagged as Audio Capture when it gets created, these additional steps aren't necessary.
Shuah: Discuss snd-usb-audio control interface mapping to media. snd-usb-audio creates a control device file which could be exposed in a media graph. In some cases, it is a simple mixer control. Expose ENT_F_MIXER entity with an associated interface node MEDIA_INTF_T_ALSA_CONTROL
Are we getting feedback from ALSA developers on this ? Will any of them be present ?
17:15-17:30: Hans: Discuss future improvements to the workshop format. Right now (as is clear from this schedule) the discussions are very technical, which makes it hard for non-core developers to attend. Perhaps we should do a split format: a half or full day of status updates, discussions to improve application integration, followed by core- developer-only discussions.
As a consequence or prerequisite, do we need longer/more frequent media workshops ?
-
Hans Verkuil -
Laurent Pinchart -
Mauro Carvalho Chehab -
Shuah Khan