1 files changed, 282 insertions, 2 deletions

diff --git a/target/i386/machine.c b/target/i386/machine.c
index 4aff1a763f..851b249d1a 100644
--- a/target/i386/machine.c
+++ b/target/i386/machine.c
@@ -231,6 +231,50 @@ static int cpu_pre_save(void *opaque)
         env->segs[R_SS].flags &= ~(env->segs[R_SS].flags & DESC_DPL_MASK);
     }
 
+#ifdef CONFIG_KVM
+    /* Verify we have nested virtualization state from kernel if required */
+    if (kvm_enabled() && cpu_has_vmx(env) && !env->nested_state) {
+        error_report("Guest enabled nested virtualization but kernel "
+                "does not support saving of nested state");
+        return -EINVAL;
+    }
+#endif
+
+    /*
+     * When vCPU is running L2 and exception is still pending,
+     * it can potentially be intercepted by L1 hypervisor.
+     * In contrast to an injected exception which cannot be
+     * intercepted anymore.
+     *
+     * Furthermore, when a L2 exception is intercepted by L1
+     * hypervisor, it's exception payload (CR2/DR6 on #PF/#DB)
+     * should not be set yet in the respective vCPU register.
+     * Thus, in case an exception is pending, it is
+     * important to save the exception payload seperately.
+     *
+     * Therefore, if an exception is not in a pending state
+     * or vCPU is not in guest-mode, it is not important to
+     * distinguish between a pending and injected exception
+     * and we don't need to store seperately the exception payload.
+     *
+     * In order to preserve better backwards-compatabile migration,
+     * convert a pending exception to an injected exception in
+     * case it is not important to distingiush between them
+     * as described above.
+     */
+    if (env->exception_pending && !(env->hflags & HF_GUEST_MASK)) {
+        env->exception_pending = 0;
+        env->exception_injected = 1;
+
+        if (env->exception_has_payload) {
+            if (env->exception_nr == EXCP01_DB) {
+                env->dr[6] = env->exception_payload;
+            } else if (env->exception_nr == EXCP0E_PAGE) {
+                env->cr[2] = env->exception_payload;
+            }
+        }
+    }
+
     return 0;
 }
 
@@ -278,6 +322,33 @@ static int cpu_post_load(void *opaque, int version_id)
     env->hflags &= ~HF_CPL_MASK;
     env->hflags |= (env->segs[R_SS].flags >> DESC_DPL_SHIFT) & HF_CPL_MASK;
 
+#ifdef CONFIG_KVM
+    if ((env->hflags & HF_GUEST_MASK) &&
+        (!env->nested_state ||
+        !(env->nested_state->flags & KVM_STATE_NESTED_GUEST_MODE))) {
+        error_report("vCPU set in guest-mode inconsistent with "
+                     "migrated kernel nested state");
+        return -EINVAL;
+    }
+#endif
+
+    /*
+     * There are cases that we can get valid exception_nr with both
+     * exception_pending and exception_injected being cleared.
+     * This can happen in one of the following scenarios:
+     * 1) Source is older QEMU without KVM_CAP_EXCEPTION_PAYLOAD support.
+     * 2) Source is running on kernel without KVM_CAP_EXCEPTION_PAYLOAD support.
+     * 3) "cpu/exception_info" subsection not sent because there is no exception
+     *    pending or guest wasn't running L2 (See comment in cpu_pre_save()).
+     *
+     * In those cases, we can just deduce that a valid exception_nr means
+     * we can treat the exception as already injected.
+     */
+    if ((env->exception_nr != -1) &&
+        !env->exception_pending && !env->exception_injected) {
+        env->exception_injected = 1;
+    }
+
     env->fpstt = (env->fpus_vmstate >> 11) & 7;
     env->fpus = env->fpus_vmstate & ~0x3800;
     env->fptag_vmstate ^= 0xff;
@@ -323,6 +394,35 @@ static bool steal_time_msr_needed(void *opaque)
     return cpu->env.steal_time_msr != 0;
 }
 
+static bool exception_info_needed(void *opaque)
+{
+    X86CPU *cpu = opaque;
+    CPUX86State *env = &cpu->env;
+
+    /*
+     * It is important to save exception-info only in case
+     * we need to distingiush between a pending and injected
+     * exception. Which is only required in case there is a
+     * pending exception and vCPU is running L2.
+     * For more info, refer to comment in cpu_pre_save().
+     */
+    return env->exception_pending && (env->hflags & HF_GUEST_MASK);
+}
+
+static const VMStateDescription vmstate_exception_info = {
+    .name = "cpu/exception_info",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .needed = exception_info_needed,
+    .fields = (VMStateField[]) {
+        VMSTATE_UINT8(env.exception_pending, X86CPU),
+        VMSTATE_UINT8(env.exception_injected, X86CPU),
+        VMSTATE_UINT8(env.exception_has_payload, X86CPU),
+        VMSTATE_UINT64(env.exception_payload, X86CPU),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
 static const VMStateDescription vmstate_steal_time_msr = {
     .name = "cpu/steal_time_msr",
     .version_id = 1,
@@ -634,7 +734,7 @@ static bool hyperv_runtime_enable_needed(void *opaque)
     X86CPU *cpu = opaque;
     CPUX86State *env = &cpu->env;
 
-    if (!cpu->hyperv_runtime) {
+    if (!hyperv_feat_enabled(cpu, HYPERV_FEAT_RUNTIME)) {
         return false;
     }
 
@@ -851,6 +951,182 @@ static const VMStateDescription vmstate_tsc_khz = {
     }
 };
 
+#ifdef CONFIG_KVM
+
+static bool vmx_vmcs12_needed(void *opaque)
+{
+    struct kvm_nested_state *nested_state = opaque;
+    return (nested_state->size >
+            offsetof(struct kvm_nested_state, data.vmx[0].vmcs12));
+}
+
+static const VMStateDescription vmstate_vmx_vmcs12 = {
+    .name = "cpu/kvm_nested_state/vmx/vmcs12",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .needed = vmx_vmcs12_needed,
+    .fields = (VMStateField[]) {
+        VMSTATE_UINT8_ARRAY(data.vmx[0].vmcs12,
+                            struct kvm_nested_state,
+                            KVM_STATE_NESTED_VMX_VMCS_SIZE),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static bool vmx_shadow_vmcs12_needed(void *opaque)
+{
+    struct kvm_nested_state *nested_state = opaque;
+    return (nested_state->size >
+            offsetof(struct kvm_nested_state, data.vmx[0].shadow_vmcs12));
+}
+
+static const VMStateDescription vmstate_vmx_shadow_vmcs12 = {
+    .name = "cpu/kvm_nested_state/vmx/shadow_vmcs12",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .needed = vmx_shadow_vmcs12_needed,
+    .fields = (VMStateField[]) {
+        VMSTATE_UINT8_ARRAY(data.vmx[0].shadow_vmcs12,
+                            struct kvm_nested_state,
+                            KVM_STATE_NESTED_VMX_VMCS_SIZE),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static bool vmx_nested_state_needed(void *opaque)
+{
+    struct kvm_nested_state *nested_state = opaque;
+
+    return ((nested_state->format == KVM_STATE_NESTED_FORMAT_VMX) &&
+            ((nested_state->hdr.vmx.vmxon_pa != -1ull) ||
+             (nested_state->hdr.vmx.smm.flags & KVM_STATE_NESTED_SMM_VMXON)));
+}
+
+static const VMStateDescription vmstate_vmx_nested_state = {
+    .name = "cpu/kvm_nested_state/vmx",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .needed = vmx_nested_state_needed,
+    .fields = (VMStateField[]) {
+        VMSTATE_U64(hdr.vmx.vmxon_pa, struct kvm_nested_state),
+        VMSTATE_U64(hdr.vmx.vmcs12_pa, struct kvm_nested_state),
+        VMSTATE_U16(hdr.vmx.smm.flags, struct kvm_nested_state),
+        VMSTATE_END_OF_LIST()
+    },
+    .subsections = (const VMStateDescription*[]) {
+        &vmstate_vmx_vmcs12,
+        &vmstate_vmx_shadow_vmcs12,
+        NULL,
+    }
+};
+
+static bool svm_nested_state_needed(void *opaque)
+{
+    struct kvm_nested_state *nested_state = opaque;
+
+    return (nested_state->format == KVM_STATE_NESTED_FORMAT_SVM);
+}
+
+static const VMStateDescription vmstate_svm_nested_state = {
+    .name = "cpu/kvm_nested_state/svm",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .needed = svm_nested_state_needed,
+    .fields = (VMStateField[]) {
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+static bool nested_state_needed(void *opaque)
+{
+    X86CPU *cpu = opaque;
+    CPUX86State *env = &cpu->env;
+
+    return (env->nested_state &&
+            (vmx_nested_state_needed(env->nested_state) ||
+             svm_nested_state_needed(env->nested_state)));
+}
+
+static int nested_state_post_load(void *opaque, int version_id)
+{
+    X86CPU *cpu = opaque;
+    CPUX86State *env = &cpu->env;
+    struct kvm_nested_state *nested_state = env->nested_state;
+    int min_nested_state_len = offsetof(struct kvm_nested_state, data);
+    int max_nested_state_len = kvm_max_nested_state_length();
+
+    /*
+     * If our kernel don't support setting nested state
+     * and we have received nested state from migration stream,
+     * we need to fail migration
+     */
+    if (max_nested_state_len <= 0) {
+        error_report("Received nested state when kernel cannot restore it");
+        return -EINVAL;
+    }
+
+    /*
+     * Verify that the size of received nested_state struct
+     * at least cover required header and is not larger
+     * than the max size that our kernel support
+     */
+    if (nested_state->size < min_nested_state_len) {
+        error_report("Received nested state size less than min: "
+                     "len=%d, min=%d",
+                     nested_state->size, min_nested_state_len);
+        return -EINVAL;
+    }
+    if (nested_state->size > max_nested_state_len) {
+        error_report("Recieved unsupported nested state size: "
+                     "nested_state->size=%d, max=%d",
+                     nested_state->size, max_nested_state_len);
+        return -EINVAL;
+    }
+
+    /* Verify format is valid */
+    if ((nested_state->format != KVM_STATE_NESTED_FORMAT_VMX) &&
+        (nested_state->format != KVM_STATE_NESTED_FORMAT_SVM)) {
+        error_report("Received invalid nested state format: %d",
+                     nested_state->format);
+        return -EINVAL;
+    }
+
+    return 0;
+}
+
+static const VMStateDescription vmstate_kvm_nested_state = {
+    .name = "cpu/kvm_nested_state",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .fields = (VMStateField[]) {
+        VMSTATE_U16(flags, struct kvm_nested_state),
+        VMSTATE_U16(format, struct kvm_nested_state),
+        VMSTATE_U32(size, struct kvm_nested_state),
+        VMSTATE_END_OF_LIST()
+    },
+    .subsections = (const VMStateDescription*[]) {
+        &vmstate_vmx_nested_state,
+        &vmstate_svm_nested_state,
+        NULL
+    }
+};
+
+static const VMStateDescription vmstate_nested_state = {
+    .name = "cpu/nested_state",
+    .version_id = 1,
+    .minimum_version_id = 1,
+    .needed = nested_state_needed,
+    .post_load = nested_state_post_load,
+    .fields = (VMStateField[]) {
+        VMSTATE_STRUCT_POINTER(env.nested_state, X86CPU,
+                vmstate_kvm_nested_state,
+                struct kvm_nested_state),
+        VMSTATE_END_OF_LIST()
+    }
+};
+
+#endif
+
 static bool mcg_ext_ctl_needed(void *opaque)
 {
     X86CPU *cpu = opaque;
@@ -1056,7 +1332,7 @@ VMStateDescription vmstate_x86_cpu = {
         VMSTATE_INT32(env.interrupt_injected, X86CPU),
         VMSTATE_UINT32(env.mp_state, X86CPU),
         VMSTATE_UINT64(env.tsc, X86CPU),
-        VMSTATE_INT32(env.exception_injected, X86CPU),
+        VMSTATE_INT32(env.exception_nr, X86CPU),
         VMSTATE_UINT8(env.soft_interrupt, X86CPU),
         VMSTATE_UINT8(env.nmi_injected, X86CPU),
         VMSTATE_UINT8(env.nmi_pending, X86CPU),
@@ -1080,6 +1356,7 @@ VMStateDescription vmstate_x86_cpu = {
         /* The above list is not sorted /wrt version numbers, watch out! */
     },
     .subsections = (const VMStateDescription*[]) {
+        &vmstate_exception_info,
         &vmstate_async_pf_msr,
         &vmstate_pv_eoi_msr,
         &vmstate_steal_time_msr,
@@ -1113,6 +1390,9 @@ VMStateDescription vmstate_x86_cpu = {
 #ifndef TARGET_X86_64
         &vmstate_efer32,
 #endif
+#ifdef CONFIG_KVM
+        &vmstate_nested_state,
+#endif
         NULL
     }
 };