On Tue, Nov 30, 2021 at 8:23 PM Sean Christopherson <seanjc@xxxxxxxxxx> wrote: > > On Tue, Nov 30, 2021, Ignat Korchagin wrote: > > I have managed to reliably reproduce the issue on a QEMU VM (on a host > > with nested virtualisation enabled). Here are the steps: > > > > 1. Install gvisor as per > > https://gvisor.dev/docs/user_guide/install/#install-latest > > 2. Run > > $ for i in $(seq 1 100); do sudo runsc --platform=kvm --network=none > > do echo ok; done > > > > I've tried to recompile the kernel with the above patch, but > > unfortunately it does fix the issue. I'm happy to try other > > patches/fixes queued for 5.16-rc4 > > My best guest would be https://lore.kernel.org/all/20211120045046.3940942-5-seanjc@xxxxxxxxxx/, > that bug results in KVM installing SPTEs into an invalid root. I think that could > lead to a use-after-free and/or double-free, which is usually what leads to the > "Bad page state" errors. Unfortunately, that patch (alone) does not fix it in my repro environment. Ignat > > In the meantime, I'll try to repro. > > > > > arch/x86/kvm/../../../virt/kvm/kvm_main.c:171 > > ... > > > > > After this the machine starts spitting some traces starting with: > > > > > > > > [177247.871683][T2343516] BUG: Bad page state in process <comm> pfn:fe680a
