[lug] IBM thinkpad X23

D. Stimits stimits at idcomm.com
Thu Jul 18 14:40:22 MDT 2002


Hugh Brown wrote:
> 
> On Thu, 2002-07-18 at 14:50, D. Stimits wrote:
> > Warren Sanders wrote:
> > >
> > > Try disabling DRI by commenting out 'Load "dri"' in the config file,
> > > then restart X.
> 
> Thanks to Warren and D.  This seems to have cured it.  I assume that I
> have turned off direct hardware access by doing this.
> 
> >
> > I will second this, and add to also make sure the set screen saver is
> > not OpenGL. Although it is pretty hard to believe some of the apps and
> > time periods involved would use DRI or allow screensavers to kick in. If
> > you have outside networking, you might want to find out if it can still
> > be pinged, and perhaps enable either sshd or telnet while debugging, so
> > you can see if it really locked up, or if just the keyboard and video is
> > locked up. If you can get in and su to root, try running init to console
> > login level. If you have a serial port, try enabling the kernel debugger
> > and magic-sysrq and get a null modem cable for serial port login (you
> > can almost always get to the kernel debugger and get a backtrace even
> > under a full lockup if you have a serial port connect).
> 
> It was dead.  No response on the network, no response on sshd.  How do
> you enable the kernel debugger/magic-sysrq?  I suppose I should go get a
> null modem cable to drag around with me.

This is the only way to debug when it is fully locked up. It sounds like
either AGP or the GLX modules are not correct for this server and/or
kernel version. With it back in the mode that locks up, you might view
which modules are loaded. Do an rmmod for the modules of GLX or AGP (go
to console init level if needed), then do a manual insmod -v, and see if
it says anything about unresolved symbols. Look in /var/log/messages,
and see if there are any messages about "modprobe: Can't locate module".

As for the actual use of magic-sysrq and kernel debugger, they first
must be compiled into the kernel (last options on kernel config). I
*think* most distros now include those options, but if they are not
there (KDB and magic-sysrq), you might need different kernel source.
Both have kernel source dir Documentation/ files, one being sysrq.txt
(useful to print the parts that say which key combos do what), the other
having its own subdirectory "kdb" with a number of files. You might want
to copy those man pages to /usr/local/man/man1/, and be sure permissions
are readable for all (if you see "kdb.mm", rename it "kdb.man" or add a
sym link to "kdb.man"). Of particular importance is kdb_bt man page, the
bt command in the debugger, which gives a backtrace.

Enabling magic-sysrq for a kernel that has the ability compiled, usually
means adding this line to /etc/sysctl.conf (at least on redhat):
kernel.sysrq =1
(or see the document file and echo "1" to the right proc file)

I think the kernel debugger is always available unless you set
"CONFIG_KDB_OFF", in which case you can echo "1" > /proc/sys/kernel/kdb
(see if it is already 1). The default key to trigger kdb is pause, you
can use "go" to start the system running again (except for the debugger
window). One warning is that if you use the reboot command in the
debugger, it is instant, it does not umount first. If you get a
backtrace (you can use btp <pid> to get a backtrace of just one
process), the function names are the most important part, most of the
arguments won't matter until someone has a specific debug question.
Also, you can change the activate key from pause to something else,
though I have not bothered. Because it is X11 and DRI you are debugging,
the serial console is the only way you can do this (I am told by Eric
Sandeen that X11 messes up kernel debugging, at least from the display
point of view, and have not tested otherwise; my own use of kernel
debugging also involves X11 kill scenarios, only console was available).

> 
> >
> > Btw, what graphics card does it have?
> 
> ATI Radeon Mobility M6 LY (per lspci)

I would double read the docs on the DRI stuff for it, and be sure both
the AGP and the X11 modules are compiled for whatever kernel and XFree86
server you have.

D. Stimits, stimits at idcomm.com

> 
> Hugh
> 
> >
> > D. Stimits, stimits @ idcomm.com
> >
> > >
> > > Hugh Brown wrote:
> > >
> > > >My boss got a new IBM thinkpad X23. and it is light as can be.  I just
> > > >installed rh 7.3 and have an interesting problem that I'm not quite sure
> > > >how to go about solving.
> > > >
> > > >Anytime I do anything that involves graphics (running X, authconfig,
> > > >etc), the system locks up within a minute or two.  I have to hard reboot
> > > >in order to bring the system back up.  It is a vanilla machine with all
> > > >the patches applied.
> > > >
> > > >I can't seem to reproduce the problem under win2k.
> 
> _______________________________________________
> Web Page:  http://lug.boulder.co.us
> Mailing List: http://lists.lug.boulder.co.us/mailman/listinfo/lug
> Join us on IRC: lug.boulder.co.us port=6667 channel=#colug



More information about the LUG mailing list