Re: [PATCH] KVM: x86: add hint to skip hidden rdpkru under kvm_load_host_xsave_state

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

 



On 14/05/21 07:11, Andy Lutomirski wrote:
That's nice, but it fails to restore XINUSE[PKRU].  As far as I know,
that bit is live, and the only way to restore it to 0 is with
XRSTOR(S).

The manual says "It is possible for XINUSE[i] to be 1 even when state component i is in its initial configuration" so this is architecturally valid. Does the XINUSE optimization matter for PKRU which is a single word?

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index cebdaa1e3cf5..cd95adbd140c 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -912,10 +912,10 @@ void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
         }

         if (static_cpu_has(X86_FEATURE_PKU) &&
-           (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
-            (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
-           vcpu->arch.pkru != vcpu->arch.host_pkru)
-               __write_pkru(vcpu->arch.pkru);
+           vcpu->arch.pkru != vcpu->arch.host_pkru &&
+           ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
+            kvm_read_cr4_bits(vcpu, X86_CR4_PKE)))
+               __write_pkru(vcpu->arch.pkru, false);

Please tell me I'm missing something (e.g. KVM very cleverly managing
the PKRU register using intercepts) that makes this reliably load the
guest value.  An innocent or malicious guest could easily make that
condition evaluate to false, thus allowing the host PKRU value to be
live in guest mode.  (Or is something fancy going on here?)

RDPKRU/WRPKRU cannot be used unless CR4.PKE=1, but PKRU can still be accessed using XSAVE/XRSTOR. However both CR4 and XCR0 have their writes trapped, so the guest will not use the host PKRU value before the next vmexit if CR4.PKE=0 and XCR0.PKRU=0.

I don't even want to think about what happens if a perf NMI hits and
accesses host user memory while the guest PKRU is live (on VMX -- I
think this can't happen on SVM).

This is indeed a problem, which indeed cannot happen on SVM but is there on VMX. Note that the function above is not handling all of the xstate, it's handling the *XSAVE state*, that is XCR0, XSS and PKRU. Thus the window is small, but it's there.

Is it solvable at all, without having PKRU fields in the VMCS (and without masking NMIs in the LAPIC which would be too expensive)? Dave, Sean, what do you think?

  }
  EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);

@@ -925,11 +925,11 @@ void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
                 return;

         if (static_cpu_has(X86_FEATURE_PKU) &&
-           (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
-            (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
+           ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
+            kvm_read_cr4_bits(vcpu, X86_CR4_PKE))) {
                 vcpu->arch.pkru = rdpkru();
                 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
-                       __write_pkru(vcpu->arch.host_pkru);
+                       __write_pkru(vcpu->arch.host_pkru, true);
         }

Suppose the guest writes to PKRU and then, without exiting, sets PKE =
0 and XCR0[PKRU] = 0.  (Or are the intercepts such that this can't
happen except on SEV where maybe SEV magic makes the problem go away?)

Yes, see above. KVM needs to trap CR4 and XCR0 anyway (CR4 because you don't want the guest to clear e.g. MCE, XCR0 to forbid setting bits that the host kernel does not have in its own xstate).

I admit I'm fairly mystified as to why KVM doesn't handle PKRU like
the rest of guest XSTATE.
Because the rest of the guest XSTATE is live too early. The problem you mention above with respect to perf, where you access host memory with the guest PKRU, would be very much amplified.

It is basically the opposite problem of what you have in switch_fpu_finish, which loads PKRU eagerly while delaying the rest to the switch to userland.

Paolo




[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