On Wed, Apr 07, 2021 at 09:23:14AM -0400, Dave Voutila wrote: > > Dave Voutila writes: > > > Mike Larkin writes: > > > >> On Wed, Apr 07, 2021 at 12:22:23AM +0200, Thomas L. wrote: > >>> On Tue, 6 Apr 2021 14:28:09 -0700 > >>> Mike Larkin <mlar...@nested.page> wrote: > >>> > >>> > On Tue, Apr 06, 2021 at 09:15:10PM +0200, Thomas L. wrote: > >>> > > On Tue, 6 Apr 2021 11:11:01 -0700 > >>> > > Mike Larkin <mlar...@nested.page> wrote: > >>> > > > Anything in the host's dmesg? > >>> > > > >>> > > >>> > *host* dmesg. I think you misread what I was after... > >>> > >>> The dmesg of the host was already attached to the first mail below the > >>> vm.conf (I mistakenly called the host hypervisor, which I realize now is > >>> not accurate). I figured since it was already attached, that > >>> you must mean the VM, compounding the confusion ... > >>> > >>> Kind regards, > >>> > >>> Thomas > >>> > >> > >> I see. > >> > >> You'll probably need to build a kernel with VMM_DEBUG and save that output > >> and > >> send it to me once a VM crashes. Note: it will generate a lot of output and > >> probably make things somewhat slower. > >> > >> -ml > > > > Thomas: I looked at your host dmesg and your provided vm.conf. It looks > > like 11 vm's with the default 512M memory and one (minecraft) with > > 8G. Your host seems to have only 16GB of memory, some of which is > > probably unavailable as it's used by the integrated gpu. I'm wondering > > if you are effectively oversusbcribing your memory here. > > > > I know we currently don't support swapping guest memory out, but not > > sure what happens if we don't have the physical memory to fault a page > > in and wire it. > > Looked a bit further and since your host is running 6.8 it doesn't have > wiring memory logic, but I'd still be cautious about oversubscribing > memory. >
Yep. Try -current and see if this can be reproduced. > > > > Even without a custom kernel with VMM_DEBUG, if it's a uvm_fault issue > > you should see a message in the kernel buffer. Something like: > > > > vmx_fault_page: uvm_fault returns N, GPA=0x...., rip=0x.... > > > > You can also run vmd(8) with debug logging (-v or -vv) and maybe capture > these events. Like with vmm(4) logging, it can be excessively verbose. > > > mlarkin: thoughts on my hypothesis? Am I wildly off course? > > > > -dv >
