On Tue, Feb 28, 2017 at 12:06:21PM +0100, Paolo Bonzini wrote: > I think it's just a trivial ordering bug: > > diff --git a/arch/x86/kernel/kvm.c b/arch/x86/kernel/kvm.c > index 14f65a5f938e..5215629ad692 100644 > --- a/arch/x86/kernel/kvm.c > +++ b/arch/x86/kernel/kvm.c > @@ -161,8 +161,8 @@ void kvm_async_pf_task_wait(u32 token) > */ > rcu_irq_exit(); > native_safe_halt(); > - rcu_irq_enter(); > local_irq_disable(); > + rcu_irq_enter(); > } > } > if (!n.halted) > > > because the async page fault code checks for preempt_disable() already, > and kmap_atomic already does preempt_disable outside pagefault_disable. > > Let me know if the above is enough to fix it for you. Well, unfortunately I can't reproduce the issue with enough reliability to say whether or not the fix is making a difference. :( FWIW, I did re-run my tests with this fix applied, and no new errors cropped up. - Ross
