Description of problem: When installing a 32bit DomU on a x86_64 Dom0 the DomU crashed repeatedly. Version-Release number of selected component (if applicable): Fresh RHEL5.1 install. How reproducible: Always Steps to Reproduce: 1. xm create staging kernel=/var/lib/xen/bootstrap/i386/vmlinuz ramdisk=/var/lib/xen/bootstrap/i386/initrd.img bootloader="" extra="ip= 10.32.4.66 netmask=255.255.255.0 dns=10.32.63.5 gateway=10.32.4.254 method=http://file/rhel5 ks=http://file/ks-small.cfg lang=en_US keymap=de-latin1-nodeadkeys text" on_crash=preserve on_reboot=destroy 2. Wait and watch via vncviewer localhost:1 3. Domain crashes during package installation Actual results: Domain crashes Expected results: Domain installed Additional info: Rebooting the Dom0 seems to make one install run possible. A second install fails. The log doesn't show very much: [2007-12-13 23:56:50 xend.XendDomainInfo 3234] WARNING (XendDomainInfo:923) Domain has crashed: name=staging id=5. [2007-12-13 23:56:50 xend.XendDomainInfo 3234] INFO (XendDomainInfo:1886) Preserving dead domain staging (5). Any suggestions on debugging this?
a) use virt-install/virt-manager instead of xm and attach the /root/.virtinst/*.log files or /root/.virt-manager/*log files if any are created b) give the domain enough memory - xen defaults to '128' which is not enough for anaconda - it'll crash & burn with this little c) attach the *full* /var/log/xen/xend.log d) Do *not* use a 32-bit paravirt DomU on a 64-bit Dom0 - this is known to randomly crash.
(In reply to comment #1) > a) use virt-install/virt-manager instead of xm and attach the > /root/.virtinst/*.log files or /root/.virt-manager/*log files if any are > created gonna try that > b) give the domain enough memory - xen defaults to '128' which is not enough > for anaconda - it'll crash & burn with this little The domain had 1024M assigned via the config file > c) attach the *full* /var/log/xen/xend.log Done > d) Do *not* use a 32-bit paravirt DomU on a 64-bit Dom0 - this is known to > randomly crash. Urgh. Fixes in sight?
Created attachment 288661 [details] xend.log
An additional note to b) Anaconda does usually just seem to freeze, no messages at all. Some times however, I can see some remnants of /dev/ram0 or something similar written on the console.
This is all symptomatic of running a 32-bit guest on a 64-bit HV. We know this doesn't work in 5.1 and we can't fix it - we don't even know why its broken - it simply doesn't work. It may work in 5.2 - we won't know until we try the full QA cycle on it though.
Hi Daniel, I've got a bit more info, reconnecting the vnc viewer gives some trace information.
Created attachment 289111 [details] kernel trace Hi Daniel, I've got a bit more info, reconnecting the vnc viewer gives some trace information.
Unless you can reproduce the problems with a 64-bit guest, we're not going to spend any time on this ticket. 32-on-64 is known to randomly, non-reproducably crash - it is simply not possible to debug it. We've wasted weeks attempting to do so in the past.
Ihis is a dpulicate of 250427. This has been fixed in the RHEL 5.2 stream. If there is a separate Anaconda issue please file a separate BZ. Also if this happens on non-PV guests, please reopen. *** This bug has been marked as a duplicate of 250427 ***
As discussed in private mail, installing 2.6.18-59.el5bbHPmgmtxen as the Dom0 kernel fixes this issue.