Uhm crap. The last system update just pulled a new beta version of NVIDIAs binary driver. Usually not that much of a deal, I went for testing due to a bug in their stable driver some years ago and never ran into a problem. Well, at least until now. This time the testing-curse found its way into my system and I found myself no longer able to start my video editor kdenlive. Segmentation Fault. I suspected my setup first since I am running a rather unusual multihead configuration but could also with just one monitor I ran into the same problem. Well, looks like I have to dig deeper. Some gdb later at least a clue:
Program received signal SIGSEGV, Segmentation fault.
0x00007ffff0b640c5 in XScreenCount (dpy=0x0) at Macros.c:109
109 int XScreenCount(Display *dpy) { return (ScreenCount(dpy)); }
(gdb) bt
#0 0x00007ffff0b640c5 in XScreenCount (dpy=0x0) at Macros.c:109
#1 0x00007fffea7f50da in glXGetClientString () from /usr/lib/libGLX.so.0
#2 0x00007ffff7fed3b3 in ?? () from /usr/lib/qt/plugins/xcbglintegrations/libqxcb-glx-integration.so
#3 0x00007ffff7fed5b1 in ?? () from /usr/lib/qt/plugins/xcbglintegrations/libqxcb-glx-integration.so
#4 0x00007ffff686fc7b in QSGRenderLoop::instance() () from /usr/lib/libQt5Quick.so.5
#5 0x00007ffff68a19e5 in QQuickWindowPrivate::init(QQuickWindow*, QQuickRenderControl*) () from /usr/lib/libQt5Quick.so.5
#6 0x00007ffff694b68d in QQuickView::QQuickView(QWindow*) () from /usr/lib/libQt5Quick.so.5
#7 0x00000000006bd190 in ?? ()
#8 0x00000000006c3d7b in ?? ()
#9 0x00000000007af3cc in ?? ()
#10 0x000000000046346b in ?? ()
#11 0x00007ffff188e610 in __libc_start_main () from /usr/lib/libc.so.6
#12 0x0000000000463949 in _start ()
So XScreenCount or libGLX is the suspect. In my case GLX is provided by NVIDIAs binary driver (nouveau had limited multihead support last time I checked). Also some QT5-stuff is mentioned in the backtrace.
A bit of search engine voodoo later it turns out NVIDIA already acknowledged the problem. Aaron Plattner writes:
I reproduced the problem and tracked it down to this buggy code in Qt5’s qxcbglxintegration.cpp:
static bool vendorChecked = false;
static bool glxPbufferUsable = true;
if (!vendorChecked) {
vendorChecked = true;
const char *glxvendor = glXGetClientString(glXGetCurrentDisplay(), GLX_VENDOR);
if (glxvendor && !strcmp(glxvendor, "ATI"))
glxPbufferUsable = false;
}
When this code is called during sddm-greeter startup, there’s no current GLX context, so this gets called with a NULL argument.
While here SDDM is not the problem I think kdenlive, which uses similar libraries, runs into the same problem. He also pushed a corresponding patch to NVIDIAs GIT-repository. Sadly the current HEAD is not 100% ABI compatible with the official driver release. Also I really didn’t want to get into my distributions xorg-packaging-foo. Ultimately I reverted to the current NVIDIA stable version 358.16 to get back into business. So – lesson confirmed: Beta releases can fix problems or cause problems. But who doesn’t like a bit of stability-gambling, right?
