GNOME Bugzilla – Bug 796926
segfault when trying to postproc or decode on remote systems
Last modified: 2018-09-19 16:09:02 UTC
$ gst-launch-1.0 videotestsrc ! vaapipostproc ! fakesink Setting pipeline to PAUSED ... 0:00:00.038091895 28538 0x55f0f14a52f0 ERROR gldisplay gstgldisplay_gbm.c:181:gst_gl_display_gbm_setup_drm: No connected DRM connector found i965_dri.so does not support the 0xffffffff PCI ID. Caught SIGSEGV Spinning. Please run 'gdb gst-launch-1.0 28538' to continue debugging, Ctrl-C to quit, or Ctrl-\ to dump core. Note that it also happens on a remote, non-headless system (i.e. over ssh but with Xorg running): $ gst-launch-1.0 filesrc location=/tmp/test ! h264parse ! vaapih264dec ! fakesink Setting pipeline to PAUSED ... i965_dri.so does not support the 0xffffffff PCI ID. Caught SIGSEGV Spinning. Please run 'gdb gst-launch-1.0 1785' to continue debugging, Ctrl-C to quit, or Ctrl-\ to dump core. It does not happen when running locally. Discovered by trying to decode some samples (it does not matter if the file is here or not): $ gst-launch-1.0 filesrc location=/tmp/sample.h264 ! h264parse ! vaapih264dec ! fakesink Setting pipeline to PAUSED ... 0:00:00.037243094 27959 0x5561e37b7cd0 ERROR gldisplay gstgldisplay_gbm.c:181:gst_gl_display_gbm_setup_drm: No connected DRM connector found i965_dri.so does not support the 0xffffffff PCI ID. Caught SIGSEGV Spinning. Please run 'gdb gst-launch-1.0 27959' to continue debugging, Ctrl-C to quit, or Ctrl-\ to dump core. It crashes with all other decoders too. Interestingly, it does not if an encoder is present: $ gst-launch-1.0 videotestsrc ! vaapih264enc ! vaapih264dec ! fakesink NB: this is a regression, problem does not appear with 1.14.2 Trace:
+ Trace 238673
This looks like consequence of using the GBM's GL backend. Can you compile GstGL with out GBM and see if the the issue persist? Perhaps gstreamer-vaapi should ban the GBM display.
Rebuilt master with: gst-build (master)$ meson configure -Dgst-plugins-base:gl_winsys=x11 build gst-build (master)$ cd build && ninja Indeed, the issue is not persisting.
Florent, Could your reproduce the issue (recompiling) and get a full back-trace, because the one you posted doesn't show the crash, it might happen in other thread. Those systems have two video cards?
> Could your reproduce the issue (recompiling) and get a full back-trace, > because the one you posted doesn't show the crash, it might happen in other > thread. Actually, i now upgraded to 1.14.3 and i can't seem to reproduce this yet. If i reproduce it, i will post "bt full" output. I still see the following message on my system running xorg (ran over ssh), but it does not crash anymore: i965_dri.so does not support the 0xffffffff PCI ID. $ GST_DEBUG=vaapi:5 gst-launch-1.0 videotestsrc num-buffers=1 ! vaapipostproc ! fakesink 0:00:00.006320476 23509 0x559bf230e520 INFO vaapi gstvaapiutils.c:121:vaapi_initialize: VA-API version 1.2 0:00:00.006675600 23509 0x559bf230e520 INFO vaapi gstvaapiutils.c:121:vaapi_initialize: VA-API version 1.2 0:00:00.104822814 23508 0x55b3c6c84a00 INFO vaapi gstvaapiutils.c:121:vaapi_initialize: VA-API version 1.2 0:00:00.105309479 23508 0x55b3c6c84a00 INFO vaapi gstvaapiutils.c:121:vaapi_initialize: VA-API version 1.2 Définition du pipeline à PAUSED... i965_dri.so does not support the 0xffffffff PCI ID. 0:00:00.274348331 23508 0x55b3c6c84a00 INFO vaapi gstvaapiutils.c:121:vaapi_initialize: VA-API version 1.2 0:00:00.274874405 23508 0x55b3c6c84a00 INFO vaapi gstvaapiutils.c:121:vaapi_initialize: VA-API version 1.2 Le pipeline est en phase de PREROLL… Contexte de l’élément « vaapipostproc0 » : gst.gl.GLDisplay=context, gst.gl.GLDisplay=(GstGLDisplay)"\(GstGLDisplayGBM\)\ gldisplaygbm0"; Contexte de l’élément « vaapipostproc0 » : gst.vaapi.Display=context, gst.vaapi.Display=(GstVaapiDisplay)"\(GstVaapiDisplayDRM\)\ vaapidisplaydrm1"; > Those systems have two video cards? Nope, just the Intel GPU.
> I still see the following message on my system running xorg (ran over ssh), but > it does not crash anymore: > i965_dri.so does not support the 0xffffffff PCI ID. Yeah, I'm seeing that in my Haswell laptop. Don't know where it comes from. Can we close the bug?
Sure, go ahead