Red Hat Bugzilla – Bug 185235
xen0 frequently crashes during boot
Last modified: 2007-11-30 17:11:27 EST
Description of problem:
Booting xen0 kernel sometimes crashes for me since upgrading to latest rawhide,
hasn't happened with corresponding non-xen kernel
Version-Release number of selected component (if applicable):
approx 1 time in 3
Steps to Reproduce:
1. boot in runlevel 5
see attached log
Created attachment 126009 [details]
serial console log of boot and crash
I'm now not particularly convinced this is a xen issue, more a general kernel
issue. I just had a similar crash on the non-xen kernel, didn't occur during
bootup, but after the machine had been left idle for a few hours, typically I no
longer had the serial console in operation and no doubt the important bit of the
crash had whizzed off the top of the 80x25 screen :-(
I've moved this from xen to kernel, if anyone knows better feel free to move it
back to kernel-xen or xen ...
Though I've experienced this crash on bootup 3 times with xen and once without
xen (can't swear it was exactly the same crash without xen) once the system does
boot it is stable, it has remained up for 3 days and 20 hours, I'm just about to
update to today's rawhide ...
This occured again for me during my first boot of 2.6.15-1.2054_FC5xen0
This bug is blocking on BZ#179629, which seems to be a security bug, so I can't
see it, can anyone fill me in with the gist of it?
179629 is just an internal "tracking" bug which we use to keep track of some
outstanding Xen issues and tasks --- it's not an actual bug report, and so don't
worry, it doesn't contain anything relevant to this report. (That's why this
bug does not depend on 179629, but rather the other way around.)
Also, it would be *really* useful if you were ever able to capture this oops on
non-xen, so that we can compare the two and be sure it's really the same issue
in each case. The oops you logged is in a xen-specific function, and the test:
if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
looks like it could conceivably be related to Xen's use of writable page tables.
If we could confirm that it was definitely seen without Xen, that would narrow
things down greatly.
*** This bug has been marked as a duplicate of 184378 ***