I don’t feel like this should even be possible, but it’s happening…
I’m running Linux as my host OS with a Windows VM for gaming. Standard KVM/Qemu setup. The VM has GPU passthrough and of course (it’s for gaming…). But there is one game I try to play on it that pretty consistently crashes. Well so what you say, games crash sometimes. But this game crashes so hard it not only locks up the Windows VM, it actually manages to lock up the Linux host OS too.
How would something like that even happen? The Windows VM has its own display of course but this is a problem even if I passthrough a separate keyboard and mouse for the Windows VM. So I still have a keyboard and mouse hooked up to the host and the host’s own display and yet it’s totally unresponsive to anything I do. I’m left with no choice but to do a hard reset.
I’m just generally puzzled by this whole experience. Incidentally other games work fine. But part of the whole reason I’m using a VM instead of just running Windows was to avoid crap like this…
I’m sorry if it sounded like I was looking for tech support. I really wasn’t. I was mainly just curious if anyone had experienced anything like this before, as it strikes me as quite bizarre. Yes ‘bugs happen’, but part of the whole point of virtualization is to containerize things in a way that really ought to be limiting things a bug like this could affect.
As far as letting the ‘appropriate people’ know goes, I don’t even know who that is because I don’t have any idea what system is failing. I tried running tail -f on my kernel and syslog log files on my host OS when it happens to see if they at least had time to post an error message or something, but they don’t. It’s a total system lockup and I don’t really have anything to go off of.
Absolutely. I’ve crashed KVM, Virtual Box, older version of VMware workstation and server (now defunct). Sure, I was trying to do thing like run OS/2, use real mode drivers in a VM, etc… so I was painting outside the lines a little. None of it was of any business need so not worth trying to get fixed.
Passthrough isn’t new, but it’s definitely a more advanced function that requires perfect cooperation between a bunch of hardware and software. I think that it’s easy to imagine how even a small bug related to the hypervisor could bring down the whole system.