| ↑back Search ←Prev date Next date→ Show only urls | (Click on time to select a line by its url) |
| Who | What | When |
|---|---|---|
| angelo_ts | kbingham, gm ! what could be the reason of v4l2_async_notifier_register failing ? My driver calls v4l2_async_register_subdev but upper driver (likely ISP) fails on v4l2_async_notifier_register | [09:50] |
| kbingham | angelo_ts, Lots of questions I'm afraid - what kernel version are you running?
angelo_ts, What is the 'return code' of the failure? What parameters are you passing :) angelo_ts, The ISP driver should use v4l2_async_notifier_register() to be notified when the subdev is available. (because that's a separate driver) - so they have to work out the link between themselves (through DT) and the ISP will then be able to register it's video nodes once the subdev is available. | [09:54] |
| angelo_ts | kernel is a 4.9.51 from nxp (imx8 branch), code is https://pastecode.xyz/view/92e8e655 btw, ok, i check params and return value | [09:57] |
| kbingham | angelo_ts, v4.9 is *very old* and we did a lot of chagnes to the async frameworks lately.
So it's difficult to know what state the NXP branch is | [09:58] |
| angelo_ts | ok. Will check return code
dmesg show https://pastecode.xyz/view/4325c2f7 | [09:58] |
| kbingham | angelo_ts, Because we don't know what version of the async code you are running, you will likely want to instrument the v4l2_async_notifier_register() call itself with debug at the failure return points to see what happens.
( drivers/media/v4l2-core/v4l2-async.c ) | [10:01] |
| angelo_ts | ok thanks | [10:02] |
| kbingham | angelo_ts, Your dmesg print suggests it was probably -EINVAL
(but I can't see the code that prints that message, so I can't be certain that the ret code is passed back to that print unmodified) | [10:03] |
| angelo_ts | retval is -22 | [10:03] |
| kbingham | Ok - so -EINVAL. | [10:03] |
| angelo_ts | i am tracing params | [10:04] |
| kbingham | v4l2_async_notifier_register() thinks you have given it some invalid parameters :D | [10:04] |
| angelo_ts | 6.108110] mxc_md->subdev_notifier.subdevs 7aa21c98
[ 6.113080] mxc_md->num_sensors 00000000 [ 6.117196] mxc_md->subdev_notifier.bound 08549968 [ 6.121993] mxc_md->subdev_notifier.complete 085499f8 maybe the num_sensors is the cuplrit | [10:04] |
| kbingham | depends where you put that print - your code paste sets it to 0 after it's used for some reason. | [10:05] |
| paulk-leonov | ezequielg, mhh we have little info about the encoder part | [10:05] |
| angelo_ts | kbingham, yes, not clear to me why it is set to 0 | [10:06] |
| kbingham | angelo_ts, I presume the function "register_sensor_entities" is handling the mapping of the async_subdevs array, so you'll want to check what that's doing too. | [10:10] |
| angelo_ts | ok thanks ! | [10:10] |
| ok i fails on v4l2_async_test_notify and v4l2_device_register_subdev, tha returns -22 | [10:19] | |
| .......... (idle for 48mn) | ||
| mm seems kernel is registering 2 subdevices (not clear why 2, i am supposing to register just 1), the second one is without any name, causing v4l2_device_register_subdev returning einval | [11:07] | |
| kbingham | How many entries does "register_sensor_entities()" create? | [11:10] |
| angelo_ts | 4 | [11:11] |
| kbingham | But you expect just one?
Are those sensor entities the cameras? | [11:11] |
| angelo_ts | my understanding is that there should be 1 sensors with 4 cameras (pads), one for each VC
https://pastebin.com/j1Q6bQC4 | [11:12] |
| kbingham | angelo_ts, Ok - it could be implemented that way yes.
angelo_ts, Which code creates the mxc_md->async_subdevs ? (before tehy are assigned to mxc_md->subdev_notifier.subdevs) | [11:16] |
| angelo_ts | kbingham, looks to be this https://pastebin.com/GMpXhAf3 | [11:21] |
| ............... (idle for 1h12mn) | ||
| *** | jmleo has quit IRC (Read error: Connection reset by peer)
neg has quit IRC (Read error: Connection reset by peer) | [12:33] |
| ..... (idle for 21mn) | ||
| angelo_ts | kbingham, ok the error should be related to the second port attached to csi0 (i am on cs1)
so i disabled CSI0 and now i see "v4l-subdev0 amd v4l-subdev1" into /dev | [12:57] |
| ............... (idle for 1h11mn) | ||
| pinchartl | mchehab: https://osseu18.sched.com/event/FwGc/complex-cameras-on-linux-mauro-carvalho-chehab-samsung and https://osseu18.sched.com/event/FwHe/why-embedded-cameras-are-difficult-and-how-to-make-them-easy-laurent-pinchart-ideas-on-board
slight issue ? | [14:12] |
| ezequielg | paulk-leonov: i wasn't talking about the encoder :-) | [14:13] |
| paulk-leonov | ezequielg, oh right, I misread it
ezequielg, we do have the info we need for the decoder part | [14:13] |
| ezequielg | the rockchip vpu expects a cabac table for h264, but i am not sure it should be exposed to userspace.
as I'm not sure how standard this is. | [14:15] |
| paulk-leonov | ezequielg, is it in the bitstream or a constant for all videos? | [14:15] |
| ezequielg | it is constant, but i am not sure how the device uses it, the spec is not super verbose ;) | [14:16] |
| paulk-leonov | ok, I'm investigating
ezequielg, by the looks of it, it should remain hardcoded in the driver code IMO ezequielg, it seems that the values are specific to the hw (not the same between rk3288 and rk3399) so asking userspace to set these coefficients would defeat the purpose ah well, apparently they are the same coefficients with different endianness or close anyway I don't think userspace should have to deal with that, at all | [14:18] |
| hverkuil | mchehab: posted v18. I added the diff between v17 and v18 at the end of the cover letter, that should help you to review the changes. | [14:21] |
| paulk-leonov | if it's not on the spec, it probably shouldn't be in the controls | [14:22] |
| ezequielg | right
paulk-leonov: ok, another thing. have you started working on encoding for cedrus? | [14:23] |
| paulk-leonov | ezequielg, no I haven't, only look at it from a distance
it seems that the stateless-stateful difference won't be very substantial for encoding | [14:24] |
| hverkuil | paulk-leonov: tfiga: posted v18. The only relevant change is that MEDIA_IOC_REQUEST_ALLOC returns the fd directly and no longer uses a struct. Mauro preferred that, and I think I do as well. | [14:27] |
| paulk-leonov | alright! | [14:28] |
| ezequielg | great - getting closer hverkuil!
paulk-leonov: not sure what you mean regarding stateless/stateful. | [14:28] |
| hverkuil | mchehab: sailus: pinchartl: please look at this rfc: https://www.mail-archive.com/linux-media@vger.kernel.org/msg134053.html It is the only thing left to do. Implementing it is easy, but I want to know what you guys think of it before I implement it. | [14:31] |
| paulk-leonov | I mean that it's likely not most required controls and interfaces are already there
that most* | [14:31] |
| ....... (idle for 31mn) | ||
| ezequielg | paulk-leonov: from the libva pov, i have been looking at it and wondered if you guys were already working on it.
if not, then perhaps i will send you some patches, depending on how much progress i can make. i mean, for the encoding entrypoint and support | [15:02] |
| .......... (idle for 49mn) | ||
| paulk-leonov | ezequielg, oh nice! I also planned on taking a look at it after my internship is over :) | [15:52] |
| ....... (idle for 32mn) | ||
| *** | ezequielg has quit IRC (Quit: leaving) | [16:24] |
| ↑back Search ←Prev date Next date→ Show only urls | (Click on time to select a line by its url) |