On Thu, Oct 11, 2018 at 9:00 PM Vitaly Kuznetsov <vkuznets@xxxxxxxxxx> wrote:
>
> Tianyu Lan <lantianyu1986@xxxxxxxxx> writes:
>
> > On Thu, Oct 11, 2018 at 8:18 PM Vitaly Kuznetsov <vkuznets@xxxxxxxxxx> wrote:
> >>
> >> Tianyu Lan <lantianyu1986@xxxxxxxxx> writes:
> >>
> >> > On Thu, Oct 11, 2018 at 6:32 PM Vitaly Kuznetsov <vkuznets@xxxxxxxxxx> wrote:
> >> >>
> >> >> I'm observing random crashes in multi-vCPU L2 guests running on KVM on
> >> >> Hyper-V. I bisected the issue to the commit 877ad952be3d ("KVM: vmx: Add
> >> >> tlb_remote_flush callback support"). Hyper-V TLFS states:
> >> >>
> >> >> "AddressSpace specifies an address space ID (an EPT PML4 table pointer)"
> >> >>
> >> >> So apparently, Hyper-V doesn't expect us to pass naked EPTP, only PML4
> >> >> pointer should be used. Strip off EPT configuration information before
> >> >> calling into vmx_hv_remote_flush_tlb().
> >> >
> >> > Hi Vitaly:
> >> > : Thanks to fix this. Sorry, I didn't meet the issue..
> >> > I think we may just store EPT PML4 table pointer without EPT
> >> > configuration information
> >> > in the ept_pointer field for this case.
> >> >
> >> > diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
> >> > index 619307b3e6bb..e316058b41a6 100644
> >> > --- a/arch/x86/kvm/vmx.c
> >> > +++ b/arch/x86/kvm/vmx.c
> >> > @@ -5379,7 +5379,7 @@ static void vmx_set_cr3(struct kvm_vcpu *vcpu,
> >> > unsigned long cr3)
> >> >
> >> > if (kvm_x86_ops->tlb_remote_flush) {
> >> > spin_lock(&to_kvm_vmx(kvm)->ept_pointer_lock);
> >> > - to_vmx(vcpu)->ept_pointer = eptp;
> >> > + to_vmx(vcpu)->ept_pointer = cr3;
> >>
> >> True, we can do that (and I even had a version of my patch doing so)
> >> but 'ept_pointer' will likely need to be renamed as it's not obvious
> >> why vcpu->ept_pointer != eptp.
> >>
> >
> > Yes. that need to rename ept_pointer.
> >
>
> Honestly, I would prefer to keep more information cached, e.g. if
> someone needs EPT configuration data later he can easily get it from
> ept_pointer and by putting raw cr3 there we'll just keep less.
>
Yes, that makes sense.
--
Best regards
Tianyu Lan
