KVM: nVMX: Fix nested VPID vmx exec control
authorWanpeng Li <wanpeng.li@hotmail.com>
Tue, 21 Mar 2017 04:18:53 +0000 (21:18 -0700)
committerPaolo Bonzini <pbonzini@redhat.com>
Thu, 23 Mar 2017 18:02:14 +0000 (19:02 +0100)
This can be reproduced by running kvm-unit-tests/vmx.flat on L0 w/ vpid disabled.

Test suite: VPID
Unhandled exception 6 #UD at ip 00000000004051a6
error_code=0000      rflags=00010047      cs=00000008
rax=0000000000000000 rcx=0000000000000001 rdx=0000000000000047 rbx=0000000000402f79
rbp=0000000000456240 rsi=0000000000000001 rdi=0000000000000000
r8=000000000000000a  r9=00000000000003f8 r10=0000000080010011 r11=0000000000000000
r12=0000000000000003 r13=0000000000000708 r14=0000000000000000 r15=0000000000000000
cr0=0000000080010031 cr2=0000000000000000 cr3=0000000007fff000 cr4=0000000000002020
cr8=0000000000000000
STACK: @4051a6 40523e 400f7f 402059 40028f

We should hide and forbid VPID in L1 if it is disabled on L0. However, nested VPID
enable bit is set unconditionally during setup nested vmx exec controls though VPID
is not exposed through nested VMX capablity. This patch fixes it by don't set nested
VPID enable bit if it is disabled on L0.

Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: stable@vger.kernel.org
Fixes: 5c614b3583e (KVM: nVMX: nested VPID emulation)
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/vmx.c

index e2f608283a5ab802bbffa93c46c7a07d401e97fe..5a82766e4b074aea3fa27c3e1d01e50c2c599ed8 100644 (file)
@@ -2753,7 +2753,6 @@ static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
                SECONDARY_EXEC_RDTSCP |
                SECONDARY_EXEC_DESC |
                SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
-               SECONDARY_EXEC_ENABLE_VPID |
                SECONDARY_EXEC_APIC_REGISTER_VIRT |
                SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
                SECONDARY_EXEC_WBINVD_EXITING |
@@ -2781,10 +2780,12 @@ static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
         * though it is treated as global context.  The alternative is
         * not failing the single-context invvpid, and it is worse.
         */
-       if (enable_vpid)
+       if (enable_vpid) {
+               vmx->nested.nested_vmx_secondary_ctls_high |=
+                       SECONDARY_EXEC_ENABLE_VPID;
                vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
                        VMX_VPID_EXTENT_SUPPORTED_MASK;
-       else
+       else
                vmx->nested.nested_vmx_vpid_caps = 0;
 
        if (enable_unrestricted_guest)