Closed mighty-p closed 6 years ago
I tried finding out which commit causes this, but it seems that no recent commit did. So maybe this issue is in the code for longer (which puzzles me a bit since things seemed to run stable last week). In any case, the crash happens only when changing channels (not when the channel is already running for at least a few seconds).
For a few seconds, when I switch the channel and this happens, I see a broken image of the new channel (with a lot of green blocks/artefacts in it). So it seems something is here processed that is not ready yet (race condition?)
It looks like your driver runs an intentional assert() which generally means that something bad happened. However the driver is missing debug symbols so the backtrace does not print the key part as can be seen in stack frames 4,5,6 such as:
#4 0x00007fd40de53f44 in ?? ()
Another note: libva-1.7.0 is very old which causes the plugin to emit a build warning:
libva is too old - please, upgrade!"
We may not be able to help with such an old libva version. It is likely that you are hitting a bug in the library or driver that has been fixed in later releases.
The segfault went away when I went back from the pre 3.5 snapshot of ffmpeg to the "proven" 3.3.2 version (which is not ideal for HEVC). So probably this was neither a problem by my outdated libva nor by vaapidevice. I am closing this. Sorry for the false alarm.
VAAPIDEVICE SYSTEM INFORMATION REPORT
inxi
vainfo
ffmpeg
gcc
svdrpsend
On my Intel system, with latest GIT state, I get a segfault that I did not get before.