Re: 2.6.8-rc2 crash(s)?

From: Randy.Dunlap
Date: Fri Jul 30 2004 - 17:37:57 EST


On Thu, 29 Jul 2004 22:03:42 -0700 Randy.Dunlap wrote:

| On Fri, 30 Jul 2004 00:50:53 -0400 Gene Heskett wrote:
|
| | On Thursday 29 July 2004 23:36, Randy.Dunlap wrote:
| | >On Thu, 29 Jul 2004 21:47:21 -0400 Gene Heskett wrote:
| | >| On Thursday 29 July 2004 18:14, Randy.Dunlap wrote:
| | >| [...]
| | >| I've gone clear back to a 2.6.7 kernel because thats the newest
| | >| one that has a diff when cmp'ing fs/dcache.c files to whats in
| | >| 2.6.8-rc2.
| | >|
| | >| I've had one Oops, virtually the same one, but it didn't kill the
| | >| machine like it would have if I was running 2.6.8-rc2.
| | >
| | >Yeah, oopsen often don't kill the entire machine.
| |
| | Running 2.6.8-rc2 it sure did, deader than a doornail. X's clock
| | stopped, the whole maryann. Keyboard leds off. Had to use the reset
| | button, and once or twice I had to do a full powerdown before it
| | would enter post and reboot. Then sit thru 20+ minutes of e2fscking
| | all the drives of course.
| |
| | >| >make fs/dcache.s
| | >|
| | >| Aha! Voila!! It doesn't work in the "fs" subdir, but back out to
| | >| the top of the src tree and it works just fine. Duh...
| | >
| | >Right, it needs the top-level makfile and kbuild machinery to do
| | > that.
| | >
| | >| Now, I must confess that what I'm looking at in those two files is
| | >| the .s is the source assembly that would normally be fed to gas,
| | >| and the objdump'ed version is the dissed object translated back to
| | >| gas source. If no mistakes, they should be pretty close to the
| | >| same I'd think. Am I on the right track? Or full of it?
| | >
| | >Yes, right.
| |
| | Which?
| |
| | Right track, or full of it? :-)
|
| Right track. ;)
|
| | In any event, I could send those two files along if you'd like, I'm
| | not an assembly guru on "amd/intel" chips, not even in my wildest
| | dreams .
| |
| | >| Here's the theory thats gradually formed in whats left of my mind:
| | >| --------------
| | >| 5 things changed in the kernel soft when I changed the mobo.
| | >| 1. The ide driver, from via686a to the nforce2 version.
| | >| 2. The video driver, because the old card failed and took the mobo
| | >| with it.
| | >| 3. Ethernet driver is now forcedeth instead of rtl-8139too
| | >| 4. A different alsa driver, from via8233 to intel-8x0
| | >| 5. The 4Gb switch is turned on in the kernel now as theres a gig
| | >| of ram on this board.
| | >| --------------
| | >
| | >You can easily use a non-high-memory enabled kernel. It will still
| | >use 896 MB of RAM (IIRC). Enabling highmem gets you another 128 MB.
| | >
| | >IDE and video are somewhat important, no?
| | >But the ethernet and ALSA drivers should be optional, at least for
| | > some testing... Ha, you said that below!
| | >
| | >| I can't do anything about the first 2, but I can do without the
| | >| last 200 megs of ram long enough to test that, and I can switch
| | >| back to the rtl-8139too card for ethernet, and I can turn off
| | >| alsasound.
| | >|
| | >| In the meantime I turned a bunch of stuff the logs were
| | >| complaining about off, like sgi_fam (what the heck is that?), some
| | >| ups daemons
| | >
| | >FAM is File Alteration Monitor, from SGI:
| | > http://oss.sgi.com/projects/fam/
| |
| | So I should start it back up? Its a security tool?, or a system
| | watchdog (or both?)
|
| No, you probably don't need it. Depends on your apps/usage,
| but it's certainly not _required_.
|
| | >| for brands I don't have, that sort of thing, and have a tail
| | >| running on the log. So far, its clean since the restart of
| | >| xinetd. Another 16 hours will tell most of the tale for this
| | >| particular instant configuration.
| |
| | I just had mozilla try to display an rpm or tgz file, killed X, so the
| | log got filled up a bit. Odd, I haven't had to step on the shift key
| | before clicking a download link in ages. Trying to get a newer
| | version of ymessenger. I have an old beta version and yahoo says its
| | too old.
| |
| | >| One final question if I may: What do I turn off (or on) in the
| | >| video dept of the kernel so that my screen doesn't go black after
| | >| vmlinuz is unpacked, and not come back on till "init" is run, at
| | >| which point the screen comes back on in what looks to be exactly
| | >| the same mode?
| | >
| | >Hm, do you have a serial console enabled (in the kernel config and
| | > in your kernel command line)? If not, please send your .config
| | > file (your probably did, but I'm lost in the maze of emails).
| |
| | Not that I can find, and I'm restricted to a make menuconfig, xconfig
| | is broken because its not seeing the correct libqt-mt library. Its
| | complaining about the version in kde3.2.3, but all the env stuff now
| | points it to kde3.3-beta2. How can I cause that to get refreshed for
| | the newer libraries now installed?
|
| Dunno, I've seen some similar problems.
|
| | I'll attach the 2.6.7 .config.
|
| I'll check it, but I don't have a quick answer for you on that one.

Well, I managed to whack an ext3 partition (/) and have to reinstall it.

Just from looking at your .config, I suggest turning off some/all/most
of these to see if it will boot and run for awhile:

CONFIG_HPET_TIMER=y
CONFIG_HPET_EMULATE_RTC=y

CONFIG_PREEMPT=y
CONFIG_INPUT_EVBUG=y

CONFIG_IPMI_HANDLER=y
CONFIG_IPMI_DEVICE_INTERFACE=y

CONFIG_4KSTACKS=y

--
~Randy
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/