Odd problem with beta2's qemu-server

[tog]

I just updated from Proxmox VE beta1 to beta2 using the apt-get dist-upgrade method. Everything was fine except my Win2k Server KVM. The FreeBSD 6.3/i386 KVM was perfectly fine and happy, it was just the Win2k server KVM that blew up:

So I tried and tried various things to get it running again (including yesterday's backups) to no avail. After a couple hours I finally found and then double-checked and double-tested the conditions needed to reverse the problem: downgrade qemu-server to beta1's. That's it, I'm now using beta2's pve-kvm and beta2's kernel, but I have to use beta1's older qemu-server package or Win2k Server explodes.

I can upgrade to beta2's qemu-server and put the updated /etc/qemu-server/XXX.conf in there in the newer format that it likes and Win2k will explode. I can downgrade to beta1's qemu-server and put the older appropriate /etc/qemu-server/XXX.conf back in and Win2k works fine again.

I took a look at the kvm command line differences between qemu-server versions, the only significant (to my eyes) thing I could find was that beta1 does -drive file=/var/whatever,media=disk and beta2 does not use ,media=disk on the primary IDE drive's -drive file line. That must not have been it, though, because I tried beta2's qemu-server and added ,media=disk to my ide0 line and it showed up in the kvm command line but Win2k still blew up.

So I'm not sure what differences in the older and newer qemu-server are causing the issue, here are the kvm command lines used by each:

beta1 (works fine, Win2k is happy):

/usr/bin/kvm
-monitor pty
-vnc unix:/tmp/qemu-server-vnc-101.socket,password
-usbdevice tablet
-name host.domain.com
-localtime
-k en-us
-drive file=/var/lib/vz/images/101/vm-101-default.qcow2,index=0,if=ide,media=disk
-drive index=2,media=cdrom
-m 1024
-net nic,macaddr=52:54:00:12:34:55,model=e1000
-net tap

beta2 (Win2k explodes):

/usr/bin/kvm
-monitor pty
-vnc unix:/tmp/qemu-server-vnc-101.socket,password
-usbdevice tablet
-name host.domain.com
-boot c
-localtime
-k en-us
-drive file=/var/lib/vz/images/101/vm-101-default.qcow2,if=ide,index=0,boot=on
-drive file=,if=ide,index=2,media=cdrom
-m 1024
-net tap,vlan=0,script=/var/lib/qemu-server/bridge-vlan0
-net nic,vlan=0,model=e1000,macaddr=52:54:00:12:34:55

 

[tog]

If the Proxmox guys take an interest in fixing this, perhaps a nice shortcut in figuring out and fixing this problem with beta2's qemu-server might be for me to provide my Win2k .qcow2 (and /etc/qemu-server/XXX.conf) for testing, though it is 6 gigs gzipped.

 

[tom]

If the Proxmox guys take an interest in fixing this, perhaps a nice shortcut in figuring out and fixing this problem with beta2's qemu-server might be for me to provide my Win2k .qcow2 (and /etc/qemu-server/XXX.conf) for testing, though it is 6 gigs gzipped.

we also experiences such behavior in win2003. but this happens only rarly and only on shutdown cycles. you can fix it booting into the windows recovery console and correct the boot loader (I use bootcfg /rebuild). As far as I know this option is not availabel in w2k, but you can also use a winxp to fix a w2k loader.

FYI, the issue is not related to qemu-server, its the KVM version. Also, in the actual testing version of Proxmox VE (not released yet) this issue is never seen yet but still in testing.

 

[tom]

If the Proxmox guys take an interest in fixing this, perhaps a nice shortcut in figuring out and fixing this problem with beta2's qemu-server might be for me to provide my Win2k .qcow2 (and /etc/qemu-server/XXX.conf) for testing, though it is 6 gigs gzipped.

yes, if possible I would like to test this against the current development versions - you can email me the download link, t.huber@proxmox.com

 

[tog]

we also experiences such behavior in win2003. but this happens only rarly and only on shutdown cycles. you can fix it booting into the windows recovery console and correct the boot loader (I use bootcfg /rebuild). As far as I know this option is not availabel in w2k, but you can also use a winxp to fix a w2k loader.

FYI, the issue is not related to qemu-server, its the KVM version. Also, in the actual testing version of Proxmox VE (not released yet) this issue is never seen yet but still in testing.

I hate to argue, but actually I think mine is a different issue than this one you've described.

1) I am running the latest kernel and kvm userland available for the Proxmox VE dist and it is running fine and happy. Only qemu-server changes to fix or break it.

1) I can take the exact same .qcow2 image without fixing it or doing anything to it, under kvm-71, with qemu-server from beta1 and works fine. Everything else remains the same, no changes to the .qcow2 file at all, just change beta1's qemu-server to beta2's qemu-server and immediately I get the above screenshotted problem.

I can switch back and forth between the two versions of qemu-server without doing anything to the .qcow2 file or having to repair anything with a winxp cd and reproduce that problem or fix that problem.

I will email you privately with the necessary files so you can stick it on your own test box and hopefully reproduce the issue.

Thanks!

 

[dietmar]

I also think it is a kvm real-mode emulation problem - The disk images seems to be ok.

- Dietmar

 

[tog]

It may be, but it is an issue that pops up only when beta2's qemu-server starts up kvm with its unique blend of command line options.

When beta1's qemu-server starts kvm with its style of kvm command line options, it works fine.

It's strange and I can't pinpoint which command line option could be doing it so I am just using all updated beta2 components with beta1's qemu-server for now. I know if I create a new KVM from the web interface that probably wont generate /etc/qemu-server/XXX.conf files that beta1's qemu-server understands, but I'll fix things up manually for now for the rare occasion that I might create a new KVM.

I've sent Tom the qcow2 image, hopefully he can reproduce and figure out what's doing it.

 

[dietmar]

It works with kernel/kvm from beta2 and qemu-server from beta 1?

 

[tog]

Yes, that is correct. That's how I am using it now.

 

[dietmar]

We found out that the issue is related to the 'boot=on' option which is used to specify the boot order.

Try to delete the "bootdisk" option from the VM configuration file - does it work then?

- Dietmar

 

[tog]

You are correct, removing "bootdisk: ide0" from the config file has allowed me to use the beta2 qemu-server with that Windows2000 KVM.