]> asedeno.scripts.mit.edu Git - linux.git/commitdiff
KVM: nVMX: fix HOST_CR3/HOST_CR4 cache
authorLadi Prosek <lprosek@redhat.com>
Fri, 22 Sep 2017 05:53:15 +0000 (07:53 +0200)
committerPaolo Bonzini <pbonzini@redhat.com>
Fri, 22 Sep 2017 08:41:32 +0000 (10:41 +0200)
For nested virt we maintain multiple VMCS that can run on a vCPU. So it is
incorrect to keep vmcs_host_cr3 and vmcs_host_cr4, whose purpose is caching
the value of the rarely changing HOST_CR3 and HOST_CR4 VMCS fields, in
vCPU-wide data structures.

Hyper-V nested on KVM runs into this consistently for me with PCID enabled.
CR3 is updated with a new value, unlikely(cr3 != vmx->host_state.vmcs_host_cr3)
fires, and the currently loaded VMCS is updated. Then we switch from L2 to
L1 and the next exit reverts CR3 to its old value.

Fixes: d6e41f1151fe ("x86/mm, KVM: Teach KVM's VMX code that CR3 isn't a constant")
Signed-off-by: Ladi Prosek <lprosek@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/vmx.c

index 0726ca7a1b0291aedd168e90259817046b3aadda..c83d28b0ab056e2bb6f10f684f829cdc3f20310a 100644 (file)
@@ -200,6 +200,8 @@ struct loaded_vmcs {
        int cpu;
        bool launched;
        bool nmi_known_unmasked;
+       unsigned long vmcs_host_cr3;    /* May not match real cr3 */
+       unsigned long vmcs_host_cr4;    /* May not match real cr4 */
        struct list_head loaded_vmcss_on_cpu_link;
 };
 
@@ -600,8 +602,6 @@ struct vcpu_vmx {
                int           gs_ldt_reload_needed;
                int           fs_reload_needed;
                u64           msr_host_bndcfgs;
-               unsigned long vmcs_host_cr3;    /* May not match real cr3 */
-               unsigned long vmcs_host_cr4;    /* May not match real cr4 */
        } host_state;
        struct {
                int vm86_active;
@@ -5178,12 +5178,12 @@ static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
         */
        cr3 = __read_cr3();
        vmcs_writel(HOST_CR3, cr3);             /* 22.2.3  FIXME: shadow tables */
-       vmx->host_state.vmcs_host_cr3 = cr3;
+       vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
 
        /* Save the most likely value for this task's CR4 in the VMCS. */
        cr4 = cr4_read_shadow();
        vmcs_writel(HOST_CR4, cr4);                     /* 22.2.3, 22.2.5 */
-       vmx->host_state.vmcs_host_cr4 = cr4;
+       vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
 
        vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS);  /* 22.2.4 */
 #ifdef CONFIG_X86_64
@@ -9274,15 +9274,15 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
                vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
 
        cr3 = __get_current_cr3_fast();
-       if (unlikely(cr3 != vmx->host_state.vmcs_host_cr3)) {
+       if (unlikely(cr3 != vmx->loaded_vmcs->vmcs_host_cr3)) {
                vmcs_writel(HOST_CR3, cr3);
-               vmx->host_state.vmcs_host_cr3 = cr3;
+               vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
        }
 
        cr4 = cr4_read_shadow();
-       if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
+       if (unlikely(cr4 != vmx->loaded_vmcs->vmcs_host_cr4)) {
                vmcs_writel(HOST_CR4, cr4);
-               vmx->host_state.vmcs_host_cr4 = cr4;
+               vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
        }
 
        /* When single-stepping over STI and MOV SS, we must clear the