[radeon] plasma screen corruption/nothing to click on except crash notifier -> black screen
host fi965 01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] RV370 [Radeon X600/X600 SE] using radeon driver F23 x86_64 freshly dnf upgraded from last same about 10 weeks ago (4.2.rc8 kernel) initial screen corruption eventually clears to black plasmashell PID 1621/1667/1687 segmentation fault (11) multi-user.target startup -> startx IIRC, theme selected is whatever alternative to Breeze was found, likely Oxygen If I un-disable compositing (set via xorg.conf*), crash notifier never appears, and screen corruption remains until Ctrl-Alt-BS .xsession-errors is not being written no BRC bug on point less than 5 months touched last found
After clicking off notifiers, screen goes black. No response to Ctrl-Alt-BS with keyboard configured via default /etc/X11/xorg.conf.d/00-keyboard.conf. Ctrl-Alt-Fn works. Ctrl-Alt-BS works after fixing 00-keyboard.conf and restarting X.
IceWM works normally on same installation. F22/KDE works normally on same host. Tumbleweed works normally on same host, though it's still using as much of KDE4 as remains available, Breeze is not there used, and Oxygen fonts are not there installed as a form of locking out Plasma5.
Same problem on same host with freshly upgraded Rawhide.
Prior kernels 4.1.rc2 and 4.0.rc4 didn't help. While in 4.0.rc4, something exhausted freespace on EXT4 F23 /. On reboot, it recovered to 37% free of 4.8G total, and dnf clean all brought free to 39%.
I replicated with variation on host big41 with rv380 radeon, but by incrementally dnf upgrading and starting after each increment. On the 5th group, kf5*, the trouble started. Excerpt from dnf.log containing the cuplrit:
http://fm.no-ip.com/Tmp/Linux/F/dnflog-badKF5.txt
The variation on host big41 (F23) is that the splash/desktop background and progress window paint instead of the screen corrupting. Post-crash the background is black.
I shut it down, moved the VGA cable to the onboard Intel 4000 port, removed the rv380, and the problem disappeared (1440x900 @108 DPI).
I shut it down again, installed a G84 (GT8600), put VGA cable on its #1 DVI port via adapter required for the LCD, and the problem remained gone (working 1440x900).
I shut down again, put in an ATI Cedar, connected the VGA cable to it, and again booted to a working 1440x900 Plasma desktop.
To follow up, I went to an ATI Radeon rv200 installation, dnf upgraded its (30 July) F23 to today's, and it works as expected, before and after today's upgrade.
Suggestions welcome, not sure whether this should have gone to xorg list, test list, devel list, somewhere upstream, or straight to BZ, instead of here in kde.
Felix Miata wrote:
host fi965 01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] RV370 [Radeon X600/X600 SE] using radeon driver F23 x86_64 freshly dnf upgraded from last same about 10 weeks ago (4.2.rc8 kernel) initial screen corruption eventually clears to black plasmashell PID 1621/1667/1687 segmentation fault (11) multi-user.target startup -> startx IIRC, theme selected is whatever alternative to Breeze was found, likely Oxygen If I un-disable compositing (set via xorg.conf*), crash notifier never appears, and screen corruption remains until Ctrl-Alt-BS .xsession-errors is not being written no BRC bug on point less than 5 months touched last found
That looks like broken OpenGL support in the driver.
Trying the QT_XCB_USE_SOFTWARE_OPENGL workaround that was implemented for the drivers (like rv200) that don't do OpenGL 2 at all can't hurt. Of course it will be slower, but at least it shouldn't crash…
(You can try editing /etc/X11/xinit/xinitrc.d/10-qt5-check-opengl2.sh to leave only these lines: QT_XCB_FORCE_SOFTWARE_OPENGL=1 export QT_XCB_FORCE_SOFTWARE_OPENGL to set the variable unconditionally.)
Kevin Kofler
Kevin Kofler composed on 2015-08-27 14:06 (UTC-0400):
Felix Miata wrote:
host fi965 01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] RV370 [Radeon X600/X600 SE] using radeon driver F23 x86_64 freshly dnf upgraded from last same about 10 weeks ago (4.2.rc8 kernel) initial screen corruption eventually clears to black plasmashell PID 1621/1667/1687 segmentation fault (11) multi-user.target startup -> startx IIRC, theme selected is whatever alternative to Breeze was found, likely Oxygen If I un-disable compositing (set via xorg.conf*), crash notifier never appears, and screen corruption remains until Ctrl-Alt-BS .xsession-errors is not being written no BRC bug on point less than 5 months touched last found
That looks like broken OpenGL support in the driver.
Trying the QT_XCB_USE_SOFTWARE_OPENGL workaround that was implemented for the drivers (like rv200) that don't do OpenGL 2 at all can't hurt. Of course it will be slower, but at least it shouldn't crashâ¦
(You can try editing /etc/X11/xinit/xinitrc.d/10-qt5-check-opengl2.sh to leave only these lines: QT_XCB_FORCE_SOFTWARE_OPENGL=1 export QT_XCB_FORCE_SOFTWARE_OPENGL to set the variable unconditionally.)
(Core2Duo 2.67GHz) host fi965 running 64 bit that resulted in me starting this thread won't POST any more, so instead of following up with host big41, I pulled the rv370 Radeon from host fi965, put it in (P4 2.8GHz) host gx27b running 32 bit, did a dnf update of F22, cloned it to another partition, dnf upgraded that to F23, and that did a tiny bit better. First boot to a plasma session got an automatic restart of Konsole (which I rarely close before logging off, expecting as in KDE4 an automatic reopen on login). Following the appearance of Konsole, the desktop behind it was all corrupted, then went black. Meanwhile, the panel never showed up, and 3 plasmashell abort popups appeared.
Restarting plasma after the above suggested edit avoids the corruption and allows the panel to start and function.
# rpm -qa | sort | egrep 'mesa|v-ati|r-Xorg' mesa-dri-drivers-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-filesystem-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libEGL-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libgbm-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libGL-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libglapi-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libGLES-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libGLU-9.0.0-9.fc23.i686 mesa-libwayland-egl-11.0.0-0.rc1.1.20150824.fc23.i686 xorg-x11-drv-ati-7.6.0-0.1.20150709git95f5d09.fc23.i686 xorg-x11-server-Xorg-1.17.2-2.fc23.i686
Felix Miata composed on 2015-08-27 22:23 (UTC-0400):
Kevin Kofler composed on 2015-08-27 14:06 (UTC-0400):
Felix Miata wrote:
host fi965 01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] RV370 [Radeon X600/X600 SE] using radeon driver F23 x86_64 freshly dnf upgraded from last same about 10 weeks ago (4.2.rc8 kernel) initial screen corruption eventually clears to black plasmashell PID 1621/1667/1687 segmentation fault (11) multi-user.target startup -> startx IIRC, theme selected is whatever alternative to Breeze was found, likely Oxygen If I un-disable compositing (set via xorg.conf*), crash notifier never appears, and screen corruption remains until Ctrl-Alt-BS .xsession-errors is not being written no BRC bug on point less than 5 months touched last found
That looks like broken OpenGL support in the driver.
Trying the QT_XCB_USE_SOFTWARE_OPENGL workaround that was implemented for the drivers (like rv200) that don't do OpenGL 2 at all can't hurt. Of course it will be slower, but at least it shouldn't crashâ¦
(You can try editing /etc/X11/xinit/xinitrc.d/10-qt5-check-opengl2.sh to leave only these lines: QT_XCB_FORCE_SOFTWARE_OPENGL=1 export QT_XCB_FORCE_SOFTWARE_OPENGL to set the variable unconditionally.)
(Core2Duo 2.67GHz) host fi965 running 64 bit that resulted in me starting this thread won't POST any more, so instead of following up with host big41, I pulled the rv370 Radeon from host fi965, put it in (P4 2.8GHz) host gx27b running 32 bit, did a dnf update of F22, cloned it to another partition, dnf upgraded that to F23, and that did a tiny bit better. First boot to a plasma session got an automatic restart of Konsole (which I rarely close before logging off, expecting as in KDE4 an automatic reopen on login). Following the appearance of Konsole, the desktop behind it was all corrupted, then went black. Meanwhile, the panel never showed up, and 3 plasmashell abort popups appeared.
Restarting plasma after the above suggested edit avoids the corruption and allows the panel to start and function.
Asked on ati driver list and got this wontfix bug response saying fix must come from KDE upstream if anywhere:
https://bugs.freedesktop.org/show_bug.cgi?id=91534
# rpm -qa | sort | egrep 'mesa|v-ati|r-Xorg' mesa-dri-drivers-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-filesystem-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libEGL-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libgbm-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libGL-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libglapi-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libGLES-11.0.0-0.rc1.1.20150824.fc23.i686 mesa-libGLU-9.0.0-9.fc23.i686 mesa-libwayland-egl-11.0.0-0.rc1.1.20150824.fc23.i686 xorg-x11-drv-ati-7.6.0-0.1.20150709git95f5d09.fc23.i686 xorg-x11-server-Xorg-1.17.2-2.fc23.i686
Felix Miata wrote:
Asked on ati driver list and got this wontfix bug response saying fix must come from KDE upstream if anywhere:
That is about KWin compositing though, which probably requires different OpenGL features than QML 2 (and thus Plasma 5).
Kevin Kofler
Felix Miata composed on 2015-08-25 23:13 (UTC-0400):
host fi965 01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] RV370 [Radeon X600/X600 SE] using radeon driver F23 x86_64 freshly dnf upgraded from last same about 10 weeks ago (4.2.rc8 kernel) initial screen corruption eventually clears to black plasmashell PID 1621/1667/1687 segmentation fault (11) multi-user.target startup -> startx IIRC, theme selected is whatever alternative to Breeze was found, likely Oxygen If I un-disable compositing (set via xorg.conf*), crash notifier never appears, and screen corruption remains until Ctrl-Alt-BS .xsession-errors is not being written no BRC bug on point less than 5 months touched last found
After clicking off notifiers, screen goes black. No response to Ctrl-Alt-BS with keyboard configured via default /etc/X11/xorg.conf.d/00-keyboard.conf. Ctrl-Alt-Fn works. Ctrl-Alt-BS works after fixing 00-keyboard.conf and restarting X.
IceWM works normally on same installation. F22/KDE works normally on same host. Tumbleweed works normally on same host, though it's still using as much of KDE4 as remains available, Breeze is not there used, and Oxygen fonts are not there installed as a form of locking out Plasma5.
Same problem on same host with freshly upgraded Rawhide.
Prior kernels 4.1.rc2 and 4.0.rc4 didn't help. While in 4.0.rc4, something exhausted freespace on EXT4 F23 /. On reboot, it recovered to 37% free of 4.8G total, and dnf clean all brought free to 39%.
I replicated with variation on host big41 with rv380 radeon, but by incrementally dnf upgrading and starting after each increment. On the 5th group, kf5*, the trouble started. Excerpt from dnf.log containing the cuplrit:
The variation on host big41 (F23) is that the splash/desktop background and progress window paint instead of the screen corrupting. Post-crash the background is black.
Having reported in a thread branch about host fi965 POST failure and moving the rv370 to host gx28b to replicate this problem, I came back to host big41 with rv380 expecting to replicate this in its openSUSE Tumbleweed installation after a fresh update, using FW 5.12.0 with Breeze and KDE's Oxygen fonts. No corruption or crashing on it though. It's mesa is 10.6.4, and repositories is only up to 10.6.5. So is this apparent regression due to F23's newer 11.0 mesa?
I shut it down, moved the VGA cable to the onboard Intel 4000 port, removed the rv380, and the problem disappeared (1440x900 @108 DPI).
I shut it down again, installed a G84 (GT8600), put VGA cable on its #1 DVI port via adapter required for the LCD, and the problem remained gone (working 1440x900).
I shut down again, put in an ATI Cedar, connected the VGA cable to it, and again booted to a working 1440x900 Plasma desktop.
To follow up, I went to an ATI Radeon rv200 installation, dnf upgraded its (30 July) F23 to today's, and it works as expected, before and after today's upgrade.
Suggestions welcome, not sure whether this should have gone to xorg list, test list, devel list, somewhere upstream, or straight to BZ, instead of here in kde.
Felix Miata wrote:
Having reported in a thread branch about host fi965 POST failure and moving the rv370 to host gx28b to replicate this problem, I came back to host big41 with rv380 expecting to replicate this in its openSUSE Tumbleweed installation after a fresh update, using FW 5.12.0 with Breeze and KDE's Oxygen fonts. No corruption or crashing on it though. It's mesa is 10.6.4, and repositories is only up to 10.6.5. So is this apparent regression due to F23's newer 11.0 mesa?
Please report those findings to the ati-driver developers. This is a pretty clear indication that the hardware limitations are not the problem here.
Kevin Kofler
-
Felix Miata -
Kevin Kofler