When logging in and gnome-shell is supposed to be started, the system freezes as soon as LIBGL_ALWAYS_INDIRECT=1 is set. Sometimes messages in .xsession-errors suggest that gnome-shell was killed by a signal, sometimes there seems no indication of what is happening and X seems alltogether stuck. Switching to console works though. Reproducible: Always Steps to Reproduce: 1. log into gnome with gnome-shell and have LIBGL_ALWAYS_INDIRECT=1 Actual Results: crash/hangup Expected Results: Working shell I am using KMS and the OSS radeon drivers (not fglrx). It seems that something in the whole setup (KMS,X,OSS Radeon drivers) with LIBGL_ALWAYS_INDIRECT=1 makes gnome-shell behave undefined. This should be fixed by either doing a proper fallback, or killing the session and informing the user properly. Sometimes I got a screen saying that something went wrong and offered me to log out, but there was no indication anywhere I could find, what the problem was. Since there was nothing in any log etc. there's no further information I can provide, except, what fixed it for me.
What version of mesa are you using, and with what USE flags? What version of xorg-server, libdrm, and linux kernel? How are you starting gnome-shell (logging in from gdm, or another login manager, or manually launching it somehow - please explain)? How are you setting LIBGL_ALWAYS_INDIRECT? Are there gnome-shell, radeon, or gl-related error messages or warnings in dmesg, /var/log/messages, ~/.xsesession-errors(.old), or /var/log/Xorg.0.log(.old)? Please provide the output of "eselect mesa show" and "emerge --info".
uname -sr Linux 2.6.38.8 ---------------------------- [ebuild R ] x11-libs/libdrm-2.4.33 USE="-libkms -static-libs" VIDEO_CARDS="radeon -intel -nouveau (-omap) -vmware" 0 kB [ebuild R ] media-libs/mesa-8.0.2 USE="classic egl gallium llvm nptl shared-glapi -bindist -d3d -debug -g3dvl -gbm -gles1 -gles2 -openvg -osmesa -pax_kernel -pic (-selinux) -shared-dricore -vdpau -wayland -xa -xvmc" VIDEO_CARDS="radeon -i915 -i965 -intel -nouveau -r100 -r200 -r300 -r600 -vmware" 0 kB [ebuild R ] x11-base/xorg-server-1.12.0-r1 USE="ipv6 kdrive nptl udev xnest xorg -dmx -doc -minimal (-selinux) -static-libs -tslib -xvfb" 0 kB ---- I was/am using gdm3 and started the session from there ---- The env variable 'LIBGL_ALWAYS_INDIRECT' was set through /etc/profile (via env.d file) ---- eselect mesa show r300 gallium r600 gallium sw gallium ----- oncerning the logs, I'll need to add the env var again and check all of them...
Excerpt from dmesg: gnome-settings-[1903]: segfault at b3f24296 ip b3f24296 sp bfa2f2ac error 14 in libltdl.so.7.3.0[b3f57000+7000] gnome-shell[2680]: segfault at 0 ip b63bbd42 sp bffc6d30 error 4 in libcogl.so.5.0.1[b63a1000+5e000] gnome-shell[2768]: segfault at 2c ip b647a118 sp bff2b380 error 4 in libcogl.so.5.0.1[b643c000+5e000] (Added the gnome-settings stuff sine it happens right before the gshell crash and doesn'T seem to happen when env var is not set - Could be ralted) .xsession-errors: Failed to play sound: File or data not found ** Message: applet now removed from the notification area gnome-session[2613]: WARNING: Application 'gnome-shell.desktop' killed by signal gnome-session[2613]: WARNING: App 'gnome-shell.desktop' respawning too quickly gnome-session[2613]: WARNING: Application 'gnome-shell.desktop' killed by signal .Xorg.0.log: No EEs no WWs related to GL or graphics at all.
To find out in whose code the crash happens, try to get a backtrace. Do note that X.org upstream does not care much about indirect rendering, so it is unlikely that this problem will be fixed if X/mesa is at fault. If no other way works, start gnome-session from failsafe xterm, with libsigsegv in LD_PRELOAD.
Also check with gnome 3.8
