[Bug 792341] New: Deadlock after upgrade from 1.12.2 to 1.12.4 on IMX platform

[GStreamer (GNOME Bugzilla)]

https://bugzilla.gnome.org/show_bug.cgi?id=792341

            Bug ID: 792341
           Summary: Deadlock after upgrade from 1.12.2 to 1.12.4 on IMX
                    platform
    Classification: Platform
           Product: GStreamer
           Version: 1.12.4
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: Normal
         Component: gstreamer (core)
          Assignee: gstreamer-bugs at lists.freedesktop.org
          Reporter: clopez at igalia.com
        QA Contact: gstreamer-bugs at lists.freedesktop.org
                CC: bilboed at bilboed.com, philn at igalia.com,
                    slomo at coaxion.net
     GNOME version: ---

Created attachment 366517
  --> https://bugzilla.gnome.org/attachment.cgi?id=366517&action=edit
thread apply all bt

I have WPE (WebKit) running on a Freescale (NXP) i.MX6 board (ARMv7).

The browser (WPE) is running on a Weston session and its configured to leverage
accelerated video decoding via gstreamer-imx and gstreamer-gl (glupload).

The application is a web page that plays videos in a loop and changes from one
video to the next by pre-loading the next video some seconds before the current
one stops playing and making it appear on the screen at the same time it hides
and destroys the old one (internally this uses a different gstreamer pipeline
per <video> element)

This was working fine with 1.12.2, but after an upgrade to 1.12.4 it started to
freeze after a few minutes of playing and a restart was needed.

I tracked this down to the commits 884e4b78105db30c21a0bea96cc666252140ee37 and
1a6c4be242d6d684b03bea6cba339207358faf79 ( from bug 790431 ) that were merged
in 1.12.4.

I tested to revert this two commits locally and the freeze is gone.

I attached GDB to the WPEWebProcess when it was in frozen state and I dumped
some backtraces (one with "thread apply all bt" and the other one with "thread
apply all bt full"). Find both the files attached here. 

I think the deadlock happens between thread 1 and thread 13, as both are inside
gst_pad_set_active()

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.