Host hang on boot after experimenting g with direct IO

[danja.vasiliev]

Hello,

I've been experimenting with directed IO for one of the guest VMs and now my host would hang on boot. I'm able to enter rescue mode via boot flags.

To enable directed IO on the host I followed Debian instructions //wiki.debian.org/VGAPassthrough


What would be the right way to debug this?

thanks!

 

[dcsapak]

please look at our documentation here: https://pve.proxmox.com/pve-docs/chapter-qm.html#qm_pci_passthrough
and post any errors and all relevant configs and options you set

 

[danja.vasiliev]

hi Dominik, thanks for your reply.

I went through the docs you linked and everything checks out. I also reverted all the changes related to directed IO, which did not help, the host still hangs after seconds of boot, without any error messages printed (boot without silent flag)

In emergency mode (boot flags: systemd.unit=emergency.target) i'm able to boot the host.
If then I run `pvedaemon start` the system hangs the same way as on boot, with no errors shown.

Attached is the recent /var/log/syslog (redacted)

 

[dcsapak]

log looks ok,
do you have vms with passthrough on 'start at boot' maybe there is some problem there?

If then I run `pvedaemon start` the system hangs the same way as on boot, with no errors shown.

you have to start pve-cluster first i guess

 

[danja.vasiliev]

log looks ok,
do you have vms with passthrough on 'start at boot' maybe there is some problem there?


you have to start pve-cluster first i guess

is pve-cluster a program or a service? if the latter i cannot run services in emergency mode..

yes, i might have VM with passthrough starting on boot - is there a way to disable that through the filesystem?

 

[dcsapak]

is pve-cluster a program or a service? if the latter i cannot run services in emergency mode..

ok that would explain why pvedaemon start hangs since it simply starts the service...

yes, i might have VM with passthrough starting on boot - is there a way to disable that through the filesystem?

you would have to mount the pmxcfs (with pmxcfs -l) and edit the vm config

 

[danja.vasiliev]

after moving PCIE card to another slot I was able to boot. for now I removed all directed IO configuration, and will experiment further at a later point.