nVMX: Implement emulated Page Modification Logging
authorBandan Das <bsd@redhat.com>
Fri, 5 May 2017 19:25:14 +0000 (15:25 -0400)
committerPaolo Bonzini <pbonzini@redhat.com>
Tue, 9 May 2017 09:54:22 +0000 (11:54 +0200)
With EPT A/D enabled, processor access to L2 guest
paging structures will result in a write violation.
When this happens, write the GUEST_PHYSICAL_ADDRESS
to the pml buffer provided by L1 if the access is
write and the dirty bit is being set.

This patch also adds necessary checks during VMEntry if L1
has enabled PML. If the PML index overflows, we change the
exit reason and run L1 to simulate a PML full event.

Signed-off-by: Bandan Das <bsd@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kvm/vmx.c

index 44508522a1c7fc2dbea9bdce782c79bedc4b1ebb..7d056f5385e7adf8cb2c23fd8bfe76d636b153f8 100644 (file)
@@ -248,6 +248,7 @@ struct __packed vmcs12 {
        u64 xss_exit_bitmap;
        u64 guest_physical_address;
        u64 vmcs_link_pointer;
+       u64 pml_address;
        u64 guest_ia32_debugctl;
        u64 guest_ia32_pat;
        u64 guest_ia32_efer;
@@ -369,6 +370,7 @@ struct __packed vmcs12 {
        u16 guest_ldtr_selector;
        u16 guest_tr_selector;
        u16 guest_intr_status;
+       u16 guest_pml_index;
        u16 host_es_selector;
        u16 host_cs_selector;
        u16 host_ss_selector;
@@ -407,6 +409,7 @@ struct nested_vmx {
        /* Has the level1 guest done vmxon? */
        bool vmxon;
        gpa_t vmxon_ptr;
+       bool pml_full;
 
        /* The guest-physical address of the current VMCS L1 keeps for L2 */
        gpa_t current_vmptr;
@@ -742,6 +745,7 @@ static const unsigned short vmcs_field_to_offset_table[] = {
        FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
        FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
        FIELD(GUEST_INTR_STATUS, guest_intr_status),
+       FIELD(GUEST_PML_INDEX, guest_pml_index),
        FIELD(HOST_ES_SELECTOR, host_es_selector),
        FIELD(HOST_CS_SELECTOR, host_cs_selector),
        FIELD(HOST_SS_SELECTOR, host_ss_selector),
@@ -767,6 +771,7 @@ static const unsigned short vmcs_field_to_offset_table[] = {
        FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
        FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
        FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
+       FIELD64(PML_ADDRESS, pml_address),
        FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
        FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
        FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
@@ -1353,6 +1358,11 @@ static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
                vmx_xsaves_supported();
 }
 
+static inline bool nested_cpu_has_pml(struct vmcs12 *vmcs12)
+{
+       return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML);
+}
+
 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
 {
        return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
@@ -9369,13 +9379,20 @@ static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
                struct x86_exception *fault)
 {
        struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
+       struct vcpu_vmx *vmx = to_vmx(vcpu);
        u32 exit_reason;
+       unsigned long exit_qualification = vcpu->arch.exit_qualification;
 
-       if (fault->error_code & PFERR_RSVD_MASK)
+       if (vmx->nested.pml_full) {
+               exit_reason = EXIT_REASON_PML_FULL;
+               vmx->nested.pml_full = false;
+               exit_qualification &= INTR_INFO_UNBLOCK_NMI;
+       } else if (fault->error_code & PFERR_RSVD_MASK)
                exit_reason = EXIT_REASON_EPT_MISCONFIG;
        else
                exit_reason = EXIT_REASON_EPT_VIOLATION;
-       nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
+
+       nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
        vmcs12->guest_physical_address = fault->address;
 }
 
@@ -9718,6 +9735,22 @@ static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
        return 0;
 }
 
+static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
+                                        struct vmcs12 *vmcs12)
+{
+       u64 address = vmcs12->pml_address;
+       int maxphyaddr = cpuid_maxphyaddr(vcpu);
+
+       if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML)) {
+               if (!nested_cpu_has_ept(vmcs12) ||
+                   !IS_ALIGNED(address, 4096)  ||
+                   address >> maxphyaddr)
+                       return -EINVAL;
+       }
+
+       return 0;
+}
+
 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
                                       struct vmx_msr_entry *e)
 {
@@ -10253,6 +10286,9 @@ static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
        if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
                return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
 
+       if (nested_vmx_check_pml_controls(vcpu, vmcs12))
+               return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
+
        if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
                                vmx->nested.nested_vmx_procbased_ctls_low,
                                vmx->nested.nested_vmx_procbased_ctls_high) ||
@@ -11151,6 +11187,46 @@ static void vmx_flush_log_dirty(struct kvm *kvm)
        kvm_flush_pml_buffers(kvm);
 }
 
+static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
+{
+       struct vmcs12 *vmcs12;
+       struct vcpu_vmx *vmx = to_vmx(vcpu);
+       gpa_t gpa;
+       struct page *page = NULL;
+       u64 *pml_address;
+
+       if (is_guest_mode(vcpu)) {
+               WARN_ON_ONCE(vmx->nested.pml_full);
+
+               /*
+                * Check if PML is enabled for the nested guest.
+                * Whether eptp bit 6 is set is already checked
+                * as part of A/D emulation.
+                */
+               vmcs12 = get_vmcs12(vcpu);
+               if (!nested_cpu_has_pml(vmcs12))
+                       return 0;
+
+               if (vmcs12->guest_pml_index > PML_ENTITY_NUM) {
+                       vmx->nested.pml_full = true;
+                       return 1;
+               }
+
+               gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
+
+               page = nested_get_page(vcpu, vmcs12->pml_address);
+               if (!page)
+                       return 0;
+
+               pml_address = kmap(page);
+               pml_address[vmcs12->guest_pml_index--] = gpa;
+               kunmap(page);
+               nested_release_page_clean(page);
+       }
+
+       return 0;
+}
+
 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
                                           struct kvm_memory_slot *memslot,
                                           gfn_t offset, unsigned long mask)
@@ -11510,6 +11586,7 @@ static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
        .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
        .flush_log_dirty = vmx_flush_log_dirty,
        .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
+       .write_log_dirty = vmx_write_pml_buffer,
 
        .pre_block = vmx_pre_block,
        .post_block = vmx_post_block,