Re: Oops for sale

[Avi Kivity <avi@qumranet.com>]

Gerd Hoffmann wrote:
> Log attached.  Scanning it and having a brief look at the source code
> makes me suspect it could be a flags update triggering this: Due to
> 64bit xen guest kernels running in ring3 xenner may just flip the user
> bit (off->on) in the page table handler, which is a unusual operation I
> guess ...
>   

The unshadow-on-userspace-access heuristic is broken.  Since it is also 
bad for Xenner, try removing the following lines:

        spte |= PT_WRITABLE_MASK;
        if (user_fault) {
            mmu_unshadow(vcpu->kvm, gfn);
            goto unshadowed;
        }


-- 
Do not meddle in the internals of kernels, for they are subtle and quick to panic.