Re: [PATCH] KVM:x86: avoid VMCS access from non-vCPU context

[Paolo Bonzini <pbonzini@redhat.com>]

> In commit 108b249c453dd7132599ab6dc7e435a7036c193f, "KVM: x86: introduce
> get_kvmclock_ns", a function to obtain the time as would be done by the
> guest via kvmclock was introduced and used throughout the code.
> 
> The problem is that it reads the guest TSC value via kvm_read_l1_tsc
> which can only be called in vCPU context as it reads VMCS(TSC_OFFSET)
> directly (on Intel CPUs).  Therefore, when called in kvm_arch_vm_ioctl
> for KVM_[GS]ET_CLOCK, it returns bogus values, breaking save/restore.

The patch isn't good unfortunately, because only __get_kvmclock_ns
can return the pvclock value.  However, after Luiz's commit a545ab6a0085
("kvm: x86: add tsc_offset field to struct kvm_vcpu_arch", 2016-09-07)
the TSC offset is cached outside the VMCS/VMCB and all the complexity in
vmx.c and (to a lesser extent) svm.c can be dropped.  See the following
untested patch...

Paolo

diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index 4b20f7304b9c..bdde80731f49 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -948,7 +948,6 @@ struct kvm_x86_ops {
 	int (*get_lpage_level)(void);
 	bool (*rdtscp_supported)(void);
 	bool (*invpcid_supported)(void);
-	void (*adjust_tsc_offset_guest)(struct kvm_vcpu *vcpu, s64 adjustment);
 
 	void (*set_tdp_cr3)(struct kvm_vcpu *vcpu, unsigned long cr3);
 
@@ -958,8 +957,6 @@ struct kvm_x86_ops {
 
 	void (*write_tsc_offset)(struct kvm_vcpu *vcpu, u64 offset);
 
-	u64 (*read_l1_tsc)(struct kvm_vcpu *vcpu, u64 host_tsc);
-
 	void (*get_exit_info)(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2);
 
 	int (*check_intercept)(struct kvm_vcpu *vcpu,
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index f8157a36ab09..8ca1eca5038d 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -1138,21 +1138,6 @@ static void svm_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
 	mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
 }
 
-static void svm_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment)
-{
-	struct vcpu_svm *svm = to_svm(vcpu);
-
-	svm->vmcb->control.tsc_offset += adjustment;
-	if (is_guest_mode(vcpu))
-		svm->nested.hsave->control.tsc_offset += adjustment;
-	else
-		trace_kvm_write_tsc_offset(vcpu->vcpu_id,
-				     svm->vmcb->control.tsc_offset - adjustment,
-				     svm->vmcb->control.tsc_offset);
-
-	mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
-}
-
 static void avic_init_vmcb(struct vcpu_svm *svm)
 {
 	struct vmcb *vmcb = svm->vmcb;
@@ -3449,12 +3434,6 @@ static int cr8_write_interception(struct vcpu_svm *svm)
 	return 0;
 }
 
-static u64 svm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
-{
-	struct vmcb *vmcb = get_host_vmcb(to_svm(vcpu));
-	return vmcb->control.tsc_offset + host_tsc;
-}
-
 static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
 {
 	struct vcpu_svm *svm = to_svm(vcpu);
@@ -5422,8 +5401,6 @@ static inline void avic_post_state_restore(struct kvm_vcpu *vcpu)
 	.has_wbinvd_exit = svm_has_wbinvd_exit,
 
 	.write_tsc_offset = svm_write_tsc_offset,
-	.adjust_tsc_offset_guest = svm_adjust_tsc_offset_guest,
-	.read_l1_tsc = svm_read_l1_tsc,
 
 	.set_tdp_cr3 = set_tdp_cr3,
 
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
index 74a4df993a51..754c3a7f444a 100644
--- a/arch/x86/kvm/vmx.c
+++ b/arch/x86/kvm/vmx.c
@@ -421,7 +421,6 @@ struct nested_vmx {
 	/* vmcs02_list cache of VMCSs recently used to run L2 guests */
 	struct list_head vmcs02_pool;
 	int vmcs02_num;
-	u64 vmcs01_tsc_offset;
 	bool change_vmcs01_virtual_x2apic_mode;
 	/* L2 must run next, and mustn't decide to exit to L1. */
 	bool nested_run_pending;
@@ -2605,20 +2604,6 @@ static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
 }
 
 /*
- * Like guest_read_tsc, but always returns L1's notion of the timestamp
- * counter, even if a nested guest (L2) is currently running.
- */
-static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
-{
-	u64 tsc_offset;
-
-	tsc_offset = is_guest_mode(vcpu) ?
-		to_vmx(vcpu)->nested.vmcs01_tsc_offset :
-		vmcs_read64(TSC_OFFSET);
-	return host_tsc + tsc_offset;
-}
-
-/*
  * writes 'offset' into guest's timestamp counter offset register
  */
 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
@@ -2631,7 +2616,6 @@ static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
 		 * to the newly set TSC to get L2's TSC.
 		 */
 		struct vmcs12 *vmcs12;
-		to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
 		/* recalculate vmcs02.TSC_OFFSET: */
 		vmcs12 = get_vmcs12(vcpu);
 		vmcs_write64(TSC_OFFSET, offset +
@@ -2644,19 +2628,6 @@ static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
 	}
 }
 
-static void vmx_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment)
-{
-	u64 offset = vmcs_read64(TSC_OFFSET);
-
-	vmcs_write64(TSC_OFFSET, offset + adjustment);
-	if (is_guest_mode(vcpu)) {
-		/* Even when running L2, the adjustment needs to apply to L1 */
-		to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
-	} else
-		trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
-					   offset + adjustment);
-}
-
 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
 {
 	struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
@@ -10061,9 +10032,9 @@ static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
 
 	if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
 		vmcs_write64(TSC_OFFSET,
-			vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
+			vcpu->arch.tsc_offset + vmcs12->tsc_offset);
 	else
-		vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
+		vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
 	if (kvm_has_tsc_control)
 		decache_tsc_multiplier(vmx);
 
@@ -10293,8 +10264,6 @@ static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
 
 	enter_guest_mode(vcpu);
 
-	vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
-
 	if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
 		vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
 
@@ -10818,7 +10787,7 @@ static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
 	load_vmcs12_host_state(vcpu, vmcs12);
 
 	/* Update any VMCS fields that might have changed while L2 ran */
-	vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
+	vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
 	if (vmx->hv_deadline_tsc == -1)
 		vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
 				PIN_BASED_VMX_PREEMPTION_TIMER);
@@ -11339,8 +11308,6 @@ static void vmx_setup_mce(struct kvm_vcpu *vcpu)
 	.has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
 
 	.write_tsc_offset = vmx_write_tsc_offset,
-	.adjust_tsc_offset_guest = vmx_adjust_tsc_offset_guest,
-	.read_l1_tsc = vmx_read_l1_tsc,
 
 	.set_tdp_cr3 = vmx_set_cr3,
 
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index e954be8a3185..3017de0431bd 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -1409,7 +1409,7 @@ static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
 
 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
 {
-	return kvm_x86_ops->read_l1_tsc(vcpu, kvm_scale_tsc(vcpu, host_tsc));
+	return vcpu->arch.tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
 }
 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
 
@@ -1547,7 +1547,7 @@ void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
 					   s64 adjustment)
 {
-	kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
+	kvm_vcpu_write_tsc_offset(vcpu, vcpu->arch.tsc_offset + adjustment);
 }
 
 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
@@ -1555,7 +1555,7 @@ static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
 	if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
 		WARN_ON(adjustment < 0);
 	adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
-	kvm_x86_ops->adjust_tsc_offset_guest(vcpu, adjustment);
+	adjust_tsc_offset_guest(vcpu, adjustment);
 }
 
 #ifdef CONFIG_X86_64