plasma screen corruption/nothing to click on except crash notifier -> black screen

[Felix Miata]

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. Ctrl-Alt-BS exited X session
"normally", returning vtty normally.

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.

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.


Most of the above I sent originally to the fedora-devel mailing list,
whereupon I was directed first to:

> (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.)

Restarting plasma after the above suggested edit avoids the corruption and
allows the panel to start and function.

After more thread interaction I reported about having 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?

That comment directed me to post here. Full fedora list thread begins here:
https://lists.fedoraproject.org/pipermail/kde/2015-August/015630.html
-- 
"The wise are known for their understanding, and pleasant
words are persuasive." Proverbs 16:21 (New Living Translation)

 Team OS/2 ** Reg. Linux User #211409 ** a11y rocks!

Felix Miata  ***  http://fm.no-ip.com/