mjourdan: have you tried looking at V4L2 ioctl debug logs? (the ones you enable with /sys/class/video4linux/videoX/{dev_,}debug mjourdan: also, are you using the native v4l2 support or via a custom ffmpeg? tfiga: using the native v4l2 support. Upon seeking, the last log from v4l2_video_decode_accelerator.cc is ResetDoneTask(), and then nothing. Visually it's just an endless loading circle over the video. I'll try with V4L2 ioctl debugging! mjourdan: https://hverkuil.home.xs4all.nl/spec/kapi/v4l2-dev.html#video-device-debugging hverkuil: thanks :) tfiga: Ah.. looks like all but one capture buffers get queued back after streamoffs/streamons, but I need all of them.. I'm guessing the one not queued is the still frame displayed in the video while waiting for the seek to go through ? paulk-leonov: ping hverkuil, pong Hi Paul, I wonder if you could do a quick cedrus & v4l2-compliance test for me. I committed my Request API compliance tests yesterday, but I'd like to have it tested with the cedrus driver as well. I know the test will fail, but I'd like to know where it will fail. In order to make the test work correctly more work will be needed. hverkuil, sure thing, I will give it a spin later today or tomorrow but that will be developed first for vicodec by an Outreachy intern when she starts work on this in December. mjourdan: that's possible mjourdan: do you need all or you just need a fixed number? if the latter, maybe you need to return a higher value in MIN_BUFFERS_FOR_CAPTURE? tfiga: All of them sadly. The firmware wants the full list at init and is free to write to any buffer it wants to at any time. so I have min_buffers_for_capture set to the amount of buffers allocated. tfiga: good review of the rockchip jpeg encoder. Decided to drop the pull request and wait for your comments to be fixed. tfiga: I know all about backlog. After spending almost two weeks full time on this I am now (mostly) on top of things again. hverkuil: we have a no meetings week this week and so it's a chance for me to actually spend the time on reviewing mjourdan: why do you need to re-register them after streamoff? I suppose there is no reqbufs in the meantime? tfiga: because both streamoffs trigger a poweroff of the IP :P mjourdan: okay, something to fix then :) mjourdan: also, it's MMAP mode I assume? you know all the buffers even if they're not queued to oyu given the restrictions, DMABUF mode wouldn't work (which is actually not so rare for statefull decoders) tfiga: I guess I do but I can't risk the firmware writing to buffers that are not queued in yet. And yeah it's MMAP mjourdan: yeah, the buffers that are not queued shouldn't be written to tfiga: oh I mistook min_buffers_for_capture with queue->min_buffers_needed. Yeah I'm trying right now to raise MIN_BUFFERS_FOR_CAPTURE tfiga: alright, fixed it by increasing MIN_BUFFERS_FOR_CATPURE (+ a few fixes related to that), and always sending a src change event even if the resolution/dpbsize remain the same It's overall a bit hacky though, esp. with the part where I now set min_buffers_needed to (dst_bufs_num-1).. probably won't go into the PR hverkuil: I sent a revert for the bindings doc. That is why the bindings doc patch is on that series. ezequielg: Ah, and that revert probably hasn't been pulled in yet. Is it even in a pull request? Can you check patchwork? let me check no, mchehab hasn't picked the revert patch yet. hverkuil: the thing is, the vpu dt bindings doc will land in v4.20, unless it's reverted. hverkuil: so I thought it was better to have that reverted in v4.20, until the driver is accepted. Ah, it wasn't picked up by anyone. I'll make a pull request for 4.20 for it. thanks! done mjourdan: the resolution change is an interesting trick indeed, although you might need the patch to allow REQBUFS(0) with exported DMA-bufs in use tfiga: already patched that a few days ago :D . I wouldn't have been able to even play something without it. mjourdan: okay :)