Re: [PATCH Part2 v6 37/49] KVM: SVM: Add support to handle MSR based Page State Change VMGEXIT

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon, Sep 19, 2022 at 5:47 PM Ashish Kalra <ashkalra@xxxxxxx> wrote:
>
>
> On 9/19/22 22:18, Tom Lendacky wrote:
> > On 9/19/22 17:02, Alper Gun wrote:
> >> On Mon, Sep 19, 2022 at 2:38 PM Tom Lendacky
> >> <thomas.lendacky@xxxxxxx> wrote:
> >>>
> >>> On 9/19/22 12:53, Alper Gun wrote:
> >>>> On Fri, Aug 19, 2022 at 9:54 AM Peter Gonda <pgonda@xxxxxxxxxx> wrote:
> >>>>>
> >>>>>> +
> >>>>>> +static int __snp_handle_page_state_change(struct kvm_vcpu *vcpu,
> >>>>>> enum psc_op op, gpa_t gpa,
> >>>>>> +                                         int level)
> >>>>>> +{
> >>>>>> +       struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
> >>>>>> +       struct kvm *kvm = vcpu->kvm;
> >>>>>> +       int rc, npt_level;
> >>>>>> +       kvm_pfn_t pfn;
> >>>>>> +       gpa_t gpa_end;
> >>>>>> +
> >>>>>> +       gpa_end = gpa + page_level_size(level);
> >>>>>> +
> >>>>>> +       while (gpa < gpa_end) {
> >>>>>> +               /*
> >>>>>> +                * If the gpa is not present in the NPT then
> >>>>>> build the NPT.
> >>>>>> +                */
> >>>>>> +               rc = snp_check_and_build_npt(vcpu, gpa, level);
> >>>>>> +               if (rc)
> >>>>>> +                       return -EINVAL;
> >>>>>> +
> >>>>>> +               if (op == SNP_PAGE_STATE_PRIVATE) {
> >>>>>> +                       hva_t hva;
> >>>>>> +
> >>>>>> +                       if (snp_gpa_to_hva(kvm, gpa, &hva))
> >>>>>> +                               return -EINVAL;
> >>>>>> +
> >>>>>> +                       /*
> >>>>>> +                        * Verify that the hva range is
> >>>>>> registered. This enforcement is
> >>>>>> +                        * required to avoid the cases where a
> >>>>>> page is marked private
> >>>>>> +                        * in the RMP table but never gets
> >>>>>> cleanup during the VM
> >>>>>> +                        * termination path.
> >>>>>> +                        */
> >>>>>> +                       mutex_lock(&kvm->lock);
> >>>>>> +                       rc = is_hva_registered(kvm, hva,
> >>>>>> page_level_size(level));
> >>>>>> +                       mutex_unlock(&kvm->lock);
> >>>>>> +                       if (!rc)
> >>>>>> +                               return -EINVAL;
> >>>>>> +
> >>>>>> +                       /*
> >>>>>> +                        * Mark the userspace range unmerable
> >>>>>> before adding the pages
> >>>>>> +                        * in the RMP table.
> >>>>>> +                        */
> >>>>>> +                       mmap_write_lock(kvm->mm);
> >>>>>> +                       rc = snp_mark_unmergable(kvm, hva,
> >>>>>> page_level_size(level));
> >>>>>> +                       mmap_write_unlock(kvm->mm);
> >>>>>> +                       if (rc)
> >>>>>> +                               return -EINVAL;
> >>>>>> +               }
> >>>>>> +
> >>>>>> +               write_lock(&kvm->mmu_lock);
> >>>>>> +
> >>>>>> +               rc = kvm_mmu_get_tdp_walk(vcpu, gpa, &pfn,
> >>>>>> &npt_level);
> >>>>>> +               if (!rc) {
> >>>>>> +                       /*
> >>>>>> +                        * This may happen if another vCPU
> >>>>>> unmapped the page
> >>>>>> +                        * before we acquire the lock. Retry the
> >>>>>> PSC.
> >>>>>> +                        */
> >>>>>> + write_unlock(&kvm->mmu_lock);
> >>>>>> +                       return 0;
> >>>>>> +               }
> >>>>>
> >>>>> I think we want to return -EAGAIN or similar if we want the caller to
> >>>>> retry, right? I think returning 0 here hides the error.
> >>>>>
> >>>>
> >>>> The problem here is that the caller(linux guest kernel) doesn't retry
> >>>> if PSC fails. The current implementation in the guest kernel is that
> >>>> if a page state change request fails, it terminates the VM with
> >>>> GHCB_TERM_PSC reason.
> >>>> Returning 0 here is not a good option because it will fail the PSC
> >>>> silently and will probably cause a nested RMP fault later. Returning
> >>>
> >>> Returning 0 here is ok because the PSC current index into the PSC
> >>> structure will not be updated and the guest will then retry (see the
> >>> loop
> >>> in vmgexit_psc() in arch/x86/kernel/sev.c).
> >>>
> >>> Thanks,
> >>> Tom
> >>
> >> But the host code updates the index. It doesn't leave the loop because
> >> rc is 0. The guest will think that it is successful.
> >> rc = __snp_handle_page_state_change(vcpu, op, gpa, level);
> >> if (rc)
> >> goto out;
> >>
> >> Also the page state change request with MSR is not retried. It
> >> terminates the VM if the MSR request fails.
> >
> > Ah, right. I see what you mean. It should probably return a -EAGAIN
> > instead of 0 and then the if (rc) check should be modified to
> > specifically look for -EAGAIN and goto out after setting rc to 0.
> >
> > But that does leave the MSR protocol open to the problem that you
> > mention, so, yes, retry logic in snp_handle_page_state_change() for a
> > -EAGAIN seems reasonable.
> >
> > Thanks,
> > Tom
>
> I believe it makes more sense to add the retry logic within
> __snp_handle_page_state_change() itself, as that will make it work for
> both the GHCB MSR protocol and the GHCB VMGEXIT requests.

You are suggesting we just retry 'kvm_mmu_get_tdp_walk' inside of
__snp_handle_page_state_change()? That should work but how many times
do we retry? If we return EAGAIN or error we can leave it up to the
caller

>
> Thanks, Ashish
>
> >
> >>
> >>>
> >>>> an error also terminates the guest immediately with current guest
> >>>> implementation. I think the best approach here is adding a retry logic
> >>>> to this function. Retrying without returning an error should help it
> >>>> work because snp_check_and_build_npt will be called again and in the
> >>>> second attempt this should work.
> >>>>
> >>>>>> +
> >>>>>> +               /*
> >>>>>> +                * Adjust the level so that we don't go higher
> >>>>>> than the backing
> >>>>>> +                * page level.
> >>>>>> +                */
> >>>>>> +               level = min_t(size_t, level, npt_level);
> >>>>>> +
> >>>>>> +               trace_kvm_snp_psc(vcpu->vcpu_id, pfn, gpa, op,
> >>>>>> level);
> >>>>>> +
> >>>>>> +               switch (op) {
> >>>>>> +               case SNP_PAGE_STATE_SHARED:
> >>>>>> +                       rc = snp_make_page_shared(kvm, gpa, pfn,
> >>>>>> level);
> >>>>>> +                       break;
> >>>>>> +               case SNP_PAGE_STATE_PRIVATE:
> >>>>>> +                       rc = rmp_make_private(pfn, gpa, level,
> >>>>>> sev->asid, false);
> >>>>>> +                       break;
> >>>>>> +               default:
> >>>>>> +                       rc = -EINVAL;
> >>>>>> +                       break;
> >>>>>> +               }
> >>>>>> +
> >>>>>> +               write_unlock(&kvm->mmu_lock);
> >>>>>> +
> >>>>>> +               if (rc) {
> >>>>>> +                       pr_err_ratelimited("Error op %d gpa %llx
> >>>>>> pfn %llx level %d rc %d\n",
> >>>>>> +                                          op, gpa, pfn, level, rc);
> >>>>>> +                       return rc;
> >>>>>> +               }
> >>>>>> +
> >>>>>> +               gpa = gpa + page_level_size(level);
> >>>>>> +       }
> >>>>>> +
> >>>>>> +       return 0;
> >>>>>> +}
> >>>>>> +



[Index of Archives]     [KVM ARM]     [KVM ia64]     [KVM ppc]     [Virtualization Tools]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Questions]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux