Memory corruption (QMatrix4x4::setToIdentity() has no crash potential).

- Do all backtraces look the same?
- Can you run kwin in valgrind?
- Please attach the output of "qdbus org.kde.kwin /KWin supportInformation"

The crash happens in the blurring effect, in particular in the cached blur path.

There's no guarantee because of the memory corruption, but *if* this induces the memory corruption you can either deactivate the blur effect or disable caching (the only checkbox in the effect config, run "kcmshell4 kwincompositing", effect is in list on 2nd tab)

Because of the flash invocation, what likely means some accelerated video playback, this is pot. related to the nouveau driver, but please elaborate on " nouveau w/ nvidia313 updates driver"

It is not possible to use the nvidia driver and the nouveau driver at the same time.
In doubt please attach (next to requested above output) the output of "glxinfo" as well as /var/log/Xorg.0.log

Created attachment 82504 [details]
attachment-9396-0.html

Thomas

Thanks for the detailed reply.  I have attached the files you requested.
 - I am fairly new to the backtraces and still learning how to read them, I
can only say that they look similar, but probably they are different.
-I have just now installed valgrind, but don't know how to use it yet
-Memory corruption apparently is the cause, I also sometimes get kernal
panics (flashing caps lock indicator) as well as other crash behavior that
does not create a bug report.  It does always correlate to accelerated
graphics and most often with flash video, but I have also had troubles with
pdf viewers and CAD or IDE applications.

I have disabled the blur effect and transparency as well.  I did not find
anything relating to caching.

I am pretty sure that the nvidia driver is the root cause.  I have tried
using different drivers, with similar behavior.  I tried to install the
nouveau to see if it fixed this, but apparently I did not get it configured
correctly.  Since nvidia software still was present, I assumed they were
still using the nvidia driver for low level calls for accelerated graphics.
 Would proper nouveau configuration fix this?

Cay Logan


On Sun, Sep 22, 2013 at 6:48 AM, Thomas Lübking
<thomas.luebking@gmail.com>wrote:

> https://bugs.kde.org/show_bug.cgi?id=325166
>
> --- Comment #1 from Thomas Lübking <thomas.luebking@gmail.com> ---
> Memory corruption (QMatrix4x4::setToIdentity() has no crash potential).
>
> - Do all backtraces look the same?
> - Can you run kwin in valgrind?
> - Please attach the output of "qdbus org.kde.kwin /KWin supportInformation"
>
> The crash happens in the blurring effect, in particular in the cached blur
> path.
>
> There's no guarantee because of the memory corruption, but *if* this
> induces
> the memory corruption you can either deactivate the blur effect or disable
> caching (the only checkbox in the effect config, run "kcmshell4
> kwincompositing", effect is in list on 2nd tab)
>
> Because of the flash invocation, what likely means some accelerated video
> playback, this is pot. related to the nouveau driver, but please elaborate
> on "
> nouveau w/ nvidia313 updates driver"
>
> It is not possible to use the nvidia driver and the nouveau driver at the
> same
> time.
> In doubt please attach (next to requested above output) the output of
> "glxinfo"
> as well as /var/log/Xorg.0.log
>
> --
> You are receiving this mail because:
> You reported the bug.

Created attachment 82505 [details]
attachment-9396-1.dat

Created attachment 82506 [details]
qdbus_out.txt

Created attachment 82507 [details]
glxinfo_out.txt

Created attachment 82508 [details]
Xorg.0.log

nvidia blob (and for the records: also synaptics)

(In reply to comment #2)
> Thanks for the detailed reply.  I have attached the files you requested.
>  - I am fairly new to the backtraces and still learning how to read them, I
> can only say that they look similar, but probably they are different.

The relevant part would be to end with lines containing

KWin::EffectsHandlerImpl::drawWindow [...] at ../../kwin/effects.cpp:451
KWin::BlurEffect::drawWindow [...] at ../../../kwin/effects/blur/blur.cpp:420
KWin::BlurEffect::doCachedBlur [...]

and eventually:
 
setToIdentity [...] at /usr/include/qt4/QtGui/qmatrix4x4.h

where [...] is a placeholder for random content (normal to be differen each and every time)

> -Memory corruption apparently is the cause
Please notice that this does NOT mean "your RAM is physically broken" but just that the actual bug wrote into memory portions that the code was not supposed to touch at that time; therefore every follow-up call is prone to expose this corruption and not necessarily the cause (and the QMatrix call is not the cause ;-)

> I also sometimes get kernel panics (flashing caps lock indicator) and indeed points a memory 
That's a severe issue and points a bug in the nvidia kernel module or it's compilation (compared to the used kernel)

> does not create a bug report.  It does always correlate to accelerated
> graphics and most often with flash video, but I have also had troubles with
> pdf viewers and CAD or IDE applications.

Does this always mean "fullscreen" applications? (They're all a bit prone to that)

In case, plase run "kcmshell4 kwincompositing" and in the advanced tab uncheck "suspend compositing for fullscreen windows"

> I have disabled the blur effect and transparency as well.  I did not find
> anything relating to caching.
Caching is a suboption to blurring. (There're two buttons on the right of the effect entry. One has a wrench or gear icon on it)

Did the crash occur afterwards?

> I am pretty sure that the nvidia driver is the root cause. 
Given the kernel panics, it's not unlikely, but neither certain: the kernel panics could be unrelated. Also it seems odd that various versions of the kernel module would carry the same issue.

> Since nvidia software still was present, I assumed they were
> still using the nvidia driver for low level calls for accelerated graphics.
No =)
Both talk to the HW and they prefer to do that exclusively.
Rule of thumb: install nouveau XOR nvidia - not both. It's possible, but you you've to properly configure the system to control which one's used (and they must not be mixed)

>  Would proper nouveau configuration fix this?
Impossible to say without knowing the actual problem.

> -I have just now installed valgrind, but don't know how to use it yet
If there's actually a problem in the kernel module: "not" - that'd be slightly more complex, but i'd rather bet on a "downstream" issue (broken/incompatible kernel module build) in that case.

-> "uname -r"

> Cay Logan
You're mailing to a bugtracker, there's no reason to quote messages but for inline reply.

Created attachment 82518 [details]
New crash information added by DrKonqi

kwin (4.11.1) on KDE Platform 4.11.1 using Qt 4.8.4

- What I was doing when the application crashed:   Watching this youtube video with Bill Gates http://www.youtube.com/watch?v=cBHJ-8Bch4E , the video continued playing.  The backtrace looks different, but the flash video player always leeds to some bizarre crash that doesn't show up otherwise.  This one was in reconq, but every browser has trouble.

-- Backtrace (Reduced):
#6  QEventLoop::processEvents (this=this@entry=0x7ffff552e200, flags=...) at kernel/qeventloop.cpp:145
#7  0x00007f1aeeeb4618 in QEventLoop::exec (this=this@entry=0x7ffff552e200, flags=...) at kernel/qeventloop.cpp:204
#8  0x00007f1aeeeb9cf6 in QCoreApplication::exec () at kernel/qcoreapplication.cpp:1218
#9  0x00007f1aee259ffc in QApplication::exec () at kernel/qapplication.cpp:3828
#10 0x00007f1af4ee99a3 in kdemain (argc=3, argv=0x7ffff552e638) at ../../kwin/main.cpp:589

I have had flash video problems since I bought this computer (in Windows
Vista).  I have had some variant of Ubuntu in dual boot mode from the
beginning as well.  In general I did not watch videos in the Ubuntu.
 Eventually, I started having other problems in Windows with PDF veiwers
and other things.  The more I tried to track it down, the more it eluded
me, but I would get Blue screens with page faults and various errors that
didn't make sense. It finally got to the point where windows would not boot
and I could not repair the boot sequence, but it booted fine in Ubuntu and
the disk is still working fine.  So I decided to just use Ubuntu and forget
windows.

Ubuntu was initially using nv and I did not experience any crashes, but I
was not able to watch videos effectively.  I switched to nvidia drivers and
started have similar problems with flash videos again.  Downloaded videos
seem to work fine and I don't recall any significant problems while in full
screen mode(The effects were already disabled in full screen mode).

Since disableing the blur effect I have had several crashes with flash
videos.  The first two did not generate reports.  The first one reported
that shockwave flash crashed in Chrome and then the kwin environment seemed
ok but every window had some probilem with closing or konsole windows would
not return a prompt after displaying 'ps' results, htop was only a blank
window, etc..

The second was a kernal panic, the screen still displaying but not
updating, no keyboard or mouse response.

The third a kwin segmentation fault, but the video kept on playing until
the forth crash in rekonq.

Its all very strange but nvidia and opengl seem to be the common point.

Created attachment 82519 [details]
New crash information added by DrKonqi

rekonq (2.3.2) on KDE Platform 4.11.1 using Qt 4.8.4

- What I was doing when the application crashed: The Bill Gates continued playing while I filed the first bug report, eventually reconq crashed on the same video.  Not that the particular video was the issue just that it is a long video.

-- Backtrace (Reduced):
#6  0x00007f8928682037 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
#7  0x00007f8928685698 in __GI_abort () at abort.c:90
#8  0x00007f892867ae03 in __assert_fail_base (fmt=0x7f89287d1578 "%s%s%s:%u: %s%sAssertion `%s' failed.\n%n", assertion=assertion@entry=0x7f891ef74ec7 "!xcb_xlib_unknown_seq_number", file=file@entry=0x7f891ef74e92 "../../src/xcb_io.c", line=line@entry=161, function=function@entry=0x7f891ef74fe0 <__PRETTY_FUNCTION__.14318> "append_pending_request") at assert.c:92
#9  0x00007f892867aeb2 in __GI___assert_fail (assertion=assertion@entry=0x7f891ef74ec7 "!xcb_xlib_unknown_seq_number", file=file@entry=0x7f891ef74e92 "../../src/xcb_io.c", line=line@entry=161, function=function@entry=0x7f891ef74fe0 <__PRETTY_FUNCTION__.14318> "append_pending_request") at assert.c:101
#10 0x00007f891ef0358f in append_pending_request (dpy=dpy@entry=0x9d978b0, sequence=11817445422221081073) at ../../src/xcb_io.c:159

Ok, I checked that nouveau was installed but the instructions said nvidia
drivers would be used if present.  I went into KDELauncher > Applications >
System > Additional Drivers  and removed the active nvidia driver.  I
rebooted and checked the Xorg.0.log file and verified that nouveau was the
current video driver.  I made it through the hour of the Bill Gates video
plus many many other flash videos without a single crash, segment fault or
any other adverse events.  The rekonq browser doesn't play youtube videos
now/yet, but I consider that to be a new/separate/insignificant problem.  I
am ready to call this one solved.  Thanks for all the help and kudos to the
nouveau team.



>

Nouveau won't use accelerated playback - esp. not for flash.

The other backtrace is right out of the eventloop - reg. the memory corruption state and the issues you encounter on windows as well, i don't think this is about the driver, but that the GPU is underpowered - could be due to powermizer - but as similar issues occur on windows, it's rather the Hardware (since it's a mobile GPU it'll likely be in a notebook and if that is underpowered, you've kinda lost)

You might be able to either enforce a performance level or just underclock the GPU (so instead of running Shaders @ 800MHz, clock them down to 700MHz)

Also ensure that your nouveau installation uses OpenGL and not XRender compositing.

OT and reg. youtube/flash: install minitube - far less annoying ;-)

Created attachment 84650 [details]
New crash information added by DrKonqi

kwin (4.11.4) on KDE Platform 4.11.4 using Qt 4.8.6

- What I was doing when the application crashed: Under System Settings -> Desktop Effects -> Advanced, i switched Compositing type from Opengl 1.2 to 3.1 (both Raster).
It crashed probably because of my nvidia drivers for my NVIDIA GeForce Go 7600  graphics card of my HP Pavilion dv9000. I'm running of course SolydK (built on debian) OS.

- Unusual behavior I noticed: The windows were partly transparent for a split second then everything flickered for a split second and system restored to OpenGL 1.2 so everything looked fine, and i got this warning for bug report and kwin crash.

- Custom settings of the application: OpenGL 1.2 set to Raster with nouveau drivers for my NVIDIA GeForce Go 7600 graphics card.

-- Backtrace (Reduced):
#5  0x00007fcd93cda1d5 in __GI_raise (sig=sig@entry=6) at ../nptl/sysdeps/unix/sysv/linux/raise.c:56
#6  0x00007fcd93cdd388 in __GI_abort () at abort.c:90
#7  0x00007fcd93cd3252 in __assert_fail_base (fmt=0x7fcd93e11080 "%s%s%s:%u: %s%sAssertion `%s' failed.\n%n", assertion=assertion@entry=0x7fcd91beac10 "!xcb_xlib_threads_sequence_lost", file=file@entry=0x7fcd91beaa5b "../../src/xcb_io.c", line=line@entry=274, function=function@entry=0x7fcd91beaf16 "poll_for_event") at assert.c:92
#8  0x00007fcd93cd3302 in __GI___assert_fail (assertion=0x7fcd91beac10 "!xcb_xlib_threads_sequence_lost", file=0x7fcd91beaa5b "../../src/xcb_io.c", line=274, function=0x7fcd91beaf16 "poll_for_event") at assert.c:101
[...]
#11 0x00007fcd91b7b74d in _XEventsQueued () from /usr/lib/x86_64-linux-gnu/libX11.so.6

The chip does not support OpenGL 3.1, only OpenGL 2.1
The driver should not crash, but this apparently happens.
-> Does OpenGL 2.1 work for you? (HINT: in doubt reboot after attempting to activate GL 3.1, the chip might end in an invalid state)

(In reply to comment #14)
> The chip does not support OpenGL 3.1, only OpenGL 2.1
> The driver should not crash, but this apparently happens.
> -> Does OpenGL 2.1 work for you? (HINT: in doubt reboot after attempting to
> activate GL 3.1, the chip might end in an invalid state)

Thanks for the reply, i suspected that was the problem, but uploaded the report just for the sake of integrity...
I do not have an OpenGL 2.1 option, but i have a 2.0 one, it does not crash, but instead loads an interface (even though i got a warning once, something about cannot enable..) which is mostly transparent (think corrupted image or something) which is makes  everything impossible to see except for areas that get PARTIALLY (very) updated whenever i move the mouse over a button or dynamic animation (like loading bar) or i get like a popup...and of course then i choose to revert back from changes. I also suspect this is exactly what OPENGL 3.1 does right before sending me a crash report and restoring openGL 1.2...it seems i need a new pc if i want to get the latest opengl support :(

OpenGL 2.0 (2.1 is technically supported by the GPU, but the config entry just means to use a glsl path) is supported by the chip and should sufficiently work with the nvidia blob (where you'll require the 304.xxx legacy package)
GL 3.1 has atm. no feature advantages in KWin, so using 2.0 would just be fine.

The transparent stuff sounds like a problem in nouveau shaders (i don't know how much the programmable engine works there, but nouveau 3D is still kinda "experimental")

In case you want to try the nvidia blob, ensure to not select a framebuffer console (ie. high resolution VT1) - nvidia does not support it and apparently most distros do not manage that for you :-(