From Bugzilla Helper: User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.6) Gecko/20040510 Description of problem: I am not on this machine now. (Home) I could not give versions exactly. This system was fully updated before the crash from the ncsu mirror. Crash was on 7/21/04 The general specs are the crash occured with the smp kernel and after about an hour ad a half after running. Screensavers were set to cycle every minute. Better description later from email to test list. Version-Release number of selected component (if applicable): kernel-2.6*464smp How reproducible: Didn't try Steps to Reproduce: 1. run the smp kernel 2. login using runlevel 3 3. transfer 9 gigs of info via ftp to machine 4. notice screen locked up, other processes still functioning 5. try to ctl-alt-Fx to other terminals Actual Results: no response from keyboard entries. No screen changes on display. Expected Results: machine should have been cycling screensavers and a keypress or mouse movement should have displayed the programs running in the GUI. Additional info: On the dell computer that I got the video to work by advice by people on the list. I had X displaying a black screen w/ white dots scattered sparsely. I also could not switch to any screen terminal and the keyboard lights were out. I was running the ftp client to transfer files from a windows computer to the FC3T1 fresh install. This service was still transferring files and also the web server was still working. I eventually had the file transfer complete,then pressed ctl-alt-del and the computer eventually rebooted. I did not see any difference on the screen until I saw the grub menu. Then the computer booted up into windows. This indicates that the system should have entered runlevel 6 after the ctl-alt-del. A basic summary regarding the conditions the system was booted in follow: - Running the latest 494smp kernel - On bootup, a message regarding AGP and S3 conflicting disabling S3 - Some odd boot messages passing quickly by, no apparent errors displayed. - Screen savers set to cycle every minute randomly. Was logged into X as a regular user. Checked computer periodically and all seemed alright. After about 1.5 hrs, noticed screen displaying the speckled white dots and smears. Noted ftp transfer still proceeding. Also checked apache was functioning. processor is a P4 w/ hyperthreading, 512 MB memory. My questions are regarding what logs to provide? What config files to provide? What runlevel to boot into. (I use level 3 mostly, would level 1 be better?) Any other additions of useful testing practices appreciated. I don't want to wipe any info out that might aid in bug resolution. I added high priority because of keyboard non responsive and most people would consider machine crashed.
I booted up the computer in runlevel 1. i then checked and copied logs for messages, boot, x-session errors before telinit to runlevel 3 to submit report. Linux amrc-server 2.6.7-1.494smp #1 SMP Sat Jul 17 05:34:46 EDT 2004 i686 i686 i386 GNU/Linux The rest is just attached logs.
Created attachment 102140 [details] x session file before starting x - copied this does indicate errors reported. I hope something is decent there with the output
Created attachment 102142 [details] this is the latest log messages - checked in runlevel 1 This log shows that the system shut down and also the next attachmeng of xorg.log.0 will show points of lockup
Created attachment 102143 [details] before X started up -after shutdown There are several errors regarding system lockup. I think this will help. abandoning ship now!
Reviewing the log, I noticed that the process started to use the 8 meg of vesa memory from bios. It reported 8 meg of stolen memory later. It looks to me like it eventually used the agp, which is set to 128 MB. The bios version that the computer booted up using is version A04. The computer model is a Dell Optiplex - GX270 series. The processor is a Pentium 4 - 2.6 GHz. When I was uning the original BIOS that was on the computer. (bought with). The computer had BIOS version A03 and could only go into 640 x 480 video mode. The computer did not eventually switch to AGP or use the vesa 8 meg legacy from BIOS.
As information that the X server locks up and that the screensaver is independent. I tried this out for lockup again. Knowing that the X server was locked and the computer keyboard was usable. I blindly proceeded to ctl-alt-F5, then login. I repeated the steps for other mingetty consoles. Finally, I went to my root login and issued reboot in the terminal and pressed enter. The machine rebooted and the screen stayed the same with the screensaver that was active at the time. This was a screensaver that swirled a bluiesh color. Since the machine is not locked up and X just seems to be falling out. I was wondering if telinit to a different runlevel, after the X crash would reveal anything useful? There are some updates from development that I did not apply yet. They were basically libraries and a few misc. gcc++ and whatever the ncsu mirror has on 7/22/2004. Usually a day behind.
Created attachment 102174 [details] ps -A - what's up in procceses This seems to be DOS formatted. It looks like runn on lines. I hope this is useful.
Created attachment 102175 [details] With blocked up screen. This is dmesg saved to floppy I guess this will allow you to determine differences in the previous logs submitted> Is this locking up the video hardware? can it be refreshed/reset when errors encountered? ctl-alt-backspace does not change screen output. I did not test if ctl-alt-backspace would defaulted me to the terminal startx was initialized in. I assume the login terminal would have been reached.
Created attachment 102224 [details] process running before computer crashe output sent to file
Created attachment 102225 [details] This is what is left running processes Notice X is not there any longer. See next attachment for difference file
Created attachment 102226 [details] using diff from the two resulting files This is what was not running after the crash of X. I ran poweroff from the ssh console on the RHL8 computer to shut off the computer.
Created attachment 102294 [details] mtrr state before crash
Created attachment 102295 [details] /proc/mtrr after crash
Created attachment 102296 [details] this is the difference between the mtrr process value
Created attachment 102297 [details] these are processes running or deleted for repeatability ps -A I believe that tty2 gets wiped out and so does the X related processes. tty 2 is where X was launched.
Correction: tty3 on this test. tty2 was on the last trial! END
Trying every screensaver through the preview mode. Other than a few errors with yellow text that did not crash X, I found the following screensavers to crash X. noof (repeatable, tried twice) StarWars (Repeatable, tried three times) Another screensaver just gives a white screen. The screensaver is GLEidescope.
opened bug 128862 against StarWars and noof
*** Bug 128862 has been marked as a duplicate of this bug. ***
I made an observation in the two diff files that I tracked with before and after X crashed. In comment 11 there is rd-bomb designated as still around after the crash. I got lost with all the > and < designations but then spotted a similarity in comment 15 diff file. The screensaver does running when the crash happened is still in the running processes. Also, I noticed that GLBlur did not crash immediately when trying to preview it. However, it crashed shortly after it started running. Thanks, Jim
I tried running StarWars after upgrading both xscreensavers and xorg-x11 and booting with the latest kernel. The results were the same for crashing X w/ certain selected screensavers. I read the changelog for both xorg-x11 and for xscreensaver and did not see any applicable changes which would change the outcome of this bug. I noted some i810 changes listed, but none concerning memory corruption between the 1810 driver or cleaning up the limits for marked screensavers included. It seems to me that this problem seems to effect lower resolution graphics. glxgears seems to spew out high values.
Created attachment 102773 [details] Versions of xorg-x11 from CVS installed After installing the versions of xorg-x11 listed in attachment, the screensavers StarWars and noof did not cause the X server to crash during preview testing. The latest versions for xscreensaver was also installed before testing the screensavers. I figure X improvements reduced the crashing, since cosmetic changes were listed in the changelog. The improvements are great for the 845 card but did not work for another 815 card that used the same i810 driver. A bug report for the other card was submitted to xorg bugzilla. Thanks for this fix, intentional or accidental. :-) Jim
After a longer duration test with many screensavers selected and cycling, I determined that the stability of this particular version of X resolved the random disappearance of X. After reading posts on several mailing lists, It seems that the 830 card has problems logging out successfully with certain versions of X. I know that the newer version does not work properly with the 815 card. Anyway, the driver should be divided into three or whatever amount of driver versions needed, for customizing the reliability and performance characteristics for each card. This particular customizing of the driver works for maybe the 845 video card versions. Prior versions seem to work great for the 815 cards without killing X or having any instability issues. This bug is closed.