KVM: nVMX: Fully support nested VMX preemption timer
authorArthur Chunqi Li <yzt356@gmail.com>
Mon, 16 Sep 2013 08:11:44 +0000 (16:11 +0800)
committerPaolo Bonzini <pbonzini@redhat.com>
Thu, 10 Oct 2013 16:22:54 +0000 (18:22 +0200)
This patch contains the following two changes:
1. Fix the bug in nested preemption timer support. If vmexit L2->L0
with some reasons not emulated by L1, preemption timer value should
be save in such exits.
2. Add support of "Save VMX-preemption timer value" VM-Exit controls
to nVMX.

With this patch, nested VMX preemption timer features are fully
supported.

Signed-off-by: Arthur Chunqi Li <yzt356@gmail.com>
Reviewed-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/include/uapi/asm/msr-index.h
arch/x86/kvm/vmx.c

index bb04650..b93e09a 100644 (file)
 
 /* MSR_IA32_VMX_MISC bits */
 #define MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS (1ULL << 29)
+#define MSR_IA32_VMX_MISC_PREEMPTION_TIMER_SCALE   0x1F
 /* AMD-V MSRs */
 
 #define MSR_VM_CR                       0xc0010114
index 2db9164..0156560 100644 (file)
@@ -2200,7 +2200,13 @@ static __init void nested_vmx_setup_ctls_msrs(void)
 #ifdef CONFIG_X86_64
                VM_EXIT_HOST_ADDR_SPACE_SIZE |
 #endif
-               VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
+               VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT |
+               VM_EXIT_SAVE_VMX_PREEMPTION_TIMER;
+       if (!(nested_vmx_pinbased_ctls_high & PIN_BASED_VMX_PREEMPTION_TIMER) ||
+           !(nested_vmx_exit_ctls_high & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)) {
+               nested_vmx_exit_ctls_high &= ~VM_EXIT_SAVE_VMX_PREEMPTION_TIMER;
+               nested_vmx_pinbased_ctls_high &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
+       }
        nested_vmx_exit_ctls_high |= (VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
                VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER);
 
@@ -6724,6 +6730,27 @@ static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
        *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
 }
 
+static void nested_adjust_preemption_timer(struct kvm_vcpu *vcpu)
+{
+       u64 delta_tsc_l1;
+       u32 preempt_val_l1, preempt_val_l2, preempt_scale;
+
+       if (!(get_vmcs12(vcpu)->pin_based_vm_exec_control &
+                       PIN_BASED_VMX_PREEMPTION_TIMER))
+               return;
+       preempt_scale = native_read_msr(MSR_IA32_VMX_MISC) &
+                       MSR_IA32_VMX_MISC_PREEMPTION_TIMER_SCALE;
+       preempt_val_l2 = vmcs_read32(VMX_PREEMPTION_TIMER_VALUE);
+       delta_tsc_l1 = vmx_read_l1_tsc(vcpu, native_read_tsc())
+               - vcpu->arch.last_guest_tsc;
+       preempt_val_l1 = delta_tsc_l1 >> preempt_scale;
+       if (preempt_val_l2 <= preempt_val_l1)
+               preempt_val_l2 = 0;
+       else
+               preempt_val_l2 -= preempt_val_l1;
+       vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, preempt_val_l2);
+}
+
 /*
  * The guest has exited.  See if we can fix it or if we need userspace
  * assistance.
@@ -7134,6 +7161,8 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
        atomic_switch_perf_msrs(vmx);
        debugctlmsr = get_debugctlmsr();
 
+       if (is_guest_mode(vcpu) && !vmx->nested.nested_run_pending)
+               nested_adjust_preemption_timer(vcpu);
        vmx->__launched = vmx->loaded_vmcs->launched;
        asm(
                /* Store host registers */
@@ -7543,6 +7572,7 @@ static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
 {
        struct vcpu_vmx *vmx = to_vmx(vcpu);
        u32 exec_control;
+       u32 exit_control;
 
        vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
        vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
@@ -7716,7 +7746,10 @@ static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
         * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
         * bits are further modified by vmx_set_efer() below.
         */
-       vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
+       exit_control = vmcs_config.vmexit_ctrl;
+       if (vmcs12->pin_based_vm_exec_control & PIN_BASED_VMX_PREEMPTION_TIMER)
+               exit_control |= VM_EXIT_SAVE_VMX_PREEMPTION_TIMER;
+       vmcs_write32(VM_EXIT_CONTROLS, exit_control);
 
        /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
         * emulated by vmx_set_efer(), below.
@@ -8124,6 +8157,11 @@ static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
        vmcs12->guest_pending_dbg_exceptions =
                vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
 
+       if ((vmcs12->pin_based_vm_exec_control & PIN_BASED_VMX_PREEMPTION_TIMER) &&
+           (vmcs12->vm_exit_controls & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER))
+               vmcs12->vmx_preemption_timer_value =
+                       vmcs_read32(VMX_PREEMPTION_TIMER_VALUE);
+
        /*
         * In some cases (usually, nested EPT), L2 is allowed to change its
         * own CR3 without exiting. If it has changed it, we must keep it.