Resolved: Host kernel 2.6.18 and smp: regular crashes

Discussions related to using VirtualBox on Linux hosts.
Post Reply
urilabob
Posts: 67
Joined: 16. Aug 2009, 17:09
Primary OS: Linux other
VBox Version: OSE Fedora
Guest OSses: Fedora 11

Resolved: Host kernel 2.6.18 and smp: regular crashes

Post by urilabob »

Host: CentOS 5.3 (kernel 2.6.18) 64 bit
Hardware: intel X5472 2*4
Guest: fedora 11 or ubuntu (same problem with both) - tried with kernels 2.6.29 and 2.6.30
VBox: 3.0.6 (same problems, only worse, with 3.0.4)

I'm getting regular kernel oopses and crashes (in the latter case, the host cpu usage gets stuck at n*100%, where n is usually the number of processors I'm running, though sometimes (n-1)*100%) whenever I run more than one cpu. If I stay with one cpu, the guest runs absolutely rock-solid. So presumably it's an smp problem. For a long time, I thought it was a problem with fedora in particular. But I subsequently installed ubuntu as the guest, and got very similar hangs.

Running a similar installation on similar AMD hardware (i.e. 64-bit multi-cpu) and Fedora 10 host also runs rock solid under smp. So it seems that the variables are reduced to either intel vs AMD hardware, or CentOS 2.6.18 kernel vs the later one that will be in F10.

I notice that 2.6.18 is deprecated in the user manual as a guest, because of a race condition (but I think this was a general problem, not just smp). Does anyone know if there is a problem with 2.6.18 as a host (for example, is anyone successfully running smp under CentOS 5.3)? If so, I guess I could try to build a later kernel. Or are interactions with the specific hardware more likely to be the problem? (replacing the hardware isn't an option, unfortunately).

Thanks for any feedback.

Resolution: there are reports that when virtualbox is run on hosts with multiple cpus, each with multiple cores, that problems occur if the number of cpus allocated to a guest is greater than the number of physical cpus (not cores). This is certainly my experience. This host has now been upgraded to fedora 13 (i.e. 2.6.33 kernel) with a fedora 13 guest. It still crashes regularly if I run with more than 2 cpus allocated to the guest (though subjectively, probably less than it did when the host was CentOS with 2.6.18 kernel). It seems to run fairly well - a week or so uptime - when I run it with only 2 cpus allocated. It's a dual quad-core machine, so as I read the documentation, I should be able to allocate up to 8 cpus - or at least 7 - but it doesn't seem to work. The other machine mentioned above was a single cpu quad-core machine, and I think that's the difference - vbox seems to handle single-cpu multi-core OK, and multi-cpu OK, but maybe has some problems with multi-cpu multi-core (at least for the particular hardware we have). Unfortunate, because it means we can get more power into a guest on a single-cpu machine than on a multi-cpu.

So not exactly solved (I'd very much like to be able to give this guest more compute resources), but at least it seems clear why the problem is occurring, and probably not kernel-related.
Post Reply