KVM: nVMX: Make direct IRQ/NMI injection work

From: Jan Kiszka <[email protected]>

Check if the interrupt or NMI window exit is for L1 by testing if it has the corresponding controls enabled. This is required when we allow direct injection from L0 to L2

Signed-off-by Jan Kiszka <[email protected]>

---
 arch/x86/kvm/vmx.c |    9 ++-------
 1 files changed, 2 insertions(+), 7 deletions(-)

arch/x86/kvm/vmx.c +2 -7

index ad978a6..126d047 100644
@@ -6111,14 +6111,9 @@ static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
 	case EXIT_REASON_TRIPLE_FAULT:
 		return 1;
 	case EXIT_REASON_PENDING_INTERRUPT:
+		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
 	case EXIT_REASON_NMI_WINDOW:
-		/*
-		 * prepare_vmcs02() set the CPU_BASED_VIRTUAL_INTR_PENDING bit
-		 * (aka Interrupt Window Exiting) only when L1 turned it on,
-		 * so if we got a PENDING_INTERRUPT exit, this must be for L1.
-		 * Same for NMI Window Exiting.
-		 */
-		return 1;
+		return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
 	case EXIT_REASON_TASK_SWITCH:
 		return 1;
 	case EXIT_REASON_CPUID:

Signature · 1 line

1.7.3.4

From: Jan Kiszka <[email protected]>

The basic idea is to always transfer the pending event injection on vmexit into the architectural state of the VCPU and then drop it from there if it turns out that we left L2 to enter L1.

VMX and SVM are now identical in how they recover event injections from unperformed vmlaunch/vmresume: We detect that VM_ENTRY_INTR_INFO_FIELD still contains a valid event and, if yes, transfer the content into L1's idt_vectoring_info_field.

However, we differ on how to deal with events that L0 wanted to inject into L2. Likely, this case is still broken in SVM. For VMX, the function vmcs12_save_pending_events deals with transferring pending L0 events into the queue of L1. That is mandatory as L1 may decide to switch the guest state completely, invalidating or preserving the pending events for later injection (including on a different node, once we support migration).

Note that we treat directly injected NMIs differently as they can hit both L1 and L2. In this case, we let L0 try to injection again also over L1 after leaving L2.

To avoid that we incorrectly leak an event into the architectural VCPU state that L1 wants to inject, we skip cancellation on nested run.

Signed-off-by Jan Kiszka <[email protected]>

---
 arch/x86/kvm/vmx.c |  118 ++++++++++++++++++++++++++++++++++++++--------------
 1 files changed, 87 insertions(+), 31 deletions(-)

arch/x86/kvm/vmx.c +47 -29

index 126d047..ca74358 100644
@@ -6492,8 +6492,6 @@ static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
 
 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
 {
-	if (is_guest_mode(&vmx->vcpu))
-		return;
 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
 				  VM_EXIT_INSTRUCTION_LEN,
 				  IDT_VECTORING_ERROR_CODE);

@@ -6501,7 +6499,7 @@ static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
 
 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
 {
-	if (is_guest_mode(vcpu))
+	if (to_vmx(vcpu)->nested.nested_run_pending)
 		return;
 	__vmx_complete_interrupts(vcpu,
 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),

@@ -6534,21 +6532,6 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
 	struct vcpu_vmx *vmx = to_vmx(vcpu);
 	unsigned long debugctlmsr;
 
-	if (is_guest_mode(vcpu) && !vmx->nested.nested_run_pending) {
-		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
-		if (vmcs12->idt_vectoring_info_field &
-				VECTORING_INFO_VALID_MASK) {
-			vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
-				vmcs12->idt_vectoring_info_field);
-			vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
-				vmcs12->vm_exit_instruction_len);
-			if (vmcs12->idt_vectoring_info_field &
-					VECTORING_INFO_DELIVER_CODE_MASK)
-				vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
-					vmcs12->idt_vectoring_error_code);
-		}
-	}
-
 	/* Record the guest's net vcpu time for enforced NMI injections. */
 	if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
 		vmx->entry_time = ktime_get();

@@ -6707,17 +6690,6 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
 
 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
 
-	if (is_guest_mode(vcpu)) {
-		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
-		vmcs12->idt_vectoring_info_field = vmx->idt_vectoring_info;
-		if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
-			vmcs12->idt_vectoring_error_code =
-				vmcs_read32(IDT_VECTORING_ERROR_CODE);
-			vmcs12->vm_exit_instruction_len =
-				vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
-		}
-	}
-
 	vmx->loaded_vmcs->launched = 1;
 
 	vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);

@@ -7324,6 +7296,52 @@ vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
 			vcpu->arch.cr4_guest_owned_bits));
 }
 
+static void vmcs12_save_pending_events(struct kvm_vcpu *vcpu,
+				       struct vmcs12 *vmcs12)
+{
+	u32 idt_vectoring;
+	unsigned int nr;
+
+	/*
+	 * We only transfer exceptions and maskable interrupts. It is fine if
+	 * L0 retries to inject a pending NMI over L1.
+	 */
+	if (vcpu->arch.exception.pending) {
+		nr = vcpu->arch.exception.nr;
+		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
+
+		if (kvm_exception_is_soft(nr)) {
+			vmcs12->vm_exit_instruction_len =
+				vcpu->arch.event_exit_inst_len;
+			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
+		} else
+			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
+
+		if (vcpu->arch.exception.has_error_code) {
+			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
+			vmcs12->idt_vectoring_error_code =
+				vcpu->arch.exception.error_code;
+		}
+
+		vmcs12->idt_vectoring_info_field = idt_vectoring;
+	} else if (vcpu->arch.interrupt.pending) {
+		nr = vcpu->arch.interrupt.nr;
+		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
+
+		if (vcpu->arch.interrupt.soft) {
+			idt_vectoring |= INTR_TYPE_SOFT_INTR;
+			vmcs12->vm_entry_instruction_len =
+				vcpu->arch.event_exit_inst_len;
+		} else
+			idt_vectoring |= INTR_TYPE_EXT_INTR;
+
+		vmcs12->idt_vectoring_info_field = idt_vectoring;
+	}
+
+	kvm_clear_exception_queue(vcpu);
+	kvm_clear_interrupt_queue(vcpu);
+}
+
 /*
  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),

arch/x86/kvm/vmx.c +14 -0

+			/*
+			 * Preserve the event that was supposed to be injected
+			 * by L1 via emulating it would have been returned in
+			 * IDT_VECTORING_INFO_FIELD.
+			 */
+			vmcs12->idt_vectoring_info_field =
+				vmcs12->vm_entry_intr_info_field;
+			vmcs12->idt_vectoring_error_code =
+				vmcs12->vm_entry_exception_error_code;
+			vmcs12->vm_exit_instruction_len =
+				vmcs12->vm_entry_instruction_len;
+			vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
+
+			/*

arch/x86/kvm/vmx.c +11 -0

+			vmcs12_save_pending_events(vcpu, vmcs12);
+
+		/* clear vm-entry fields which are to be cleared on exit */
 		vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
+	}
+
+	/*
+	 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
+	 * preserved above and would only end up incorrectly in L1.
+	 */
+	kvm_clear_exception_queue(vcpu);
+	kvm_clear_interrupt_queue(vcpu);
 }
 
 /*

Signature · 1 line

1.7.3.4

Quoted 11 lines · Jan Kiszka

From: Jan Kiszka <[email protected]>

The basic idea is to always transfer the pending event injection on vmexit into the architectural state of the VCPU and then drop it from there if it turns out that we left L2 to enter L1.

VMX and SVM are now identical in how they recover event injections from unperformed vmlaunch/vmresume: We detect that VM_ENTRY_INTR_INFO_FIELD still contains a valid event and, if yes, transfer the content into L1's idt_vectoring_info_field.

But how this can happens with VMX code? VMX has this nested_run_pending thing that prevents #vmexit emulation from happening without vmlaunch. This means that VM_ENTRY_INTR_INFO_FIELD should never be valid during #vmexit emulation since it is marked invalid during vmlaunch.

Quoted 9 lines

However, we differ on how to deal with events that L0 wanted to inject into L2. Likely, this case is still broken in SVM. For VMX, the function vmcs12_save_pending_events deals with transferring pending L0 events into the queue of L1. That is mandatory as L1 may decide to switch the guest state completely, invalidating or preserving the pending events for later injection (including on a different node, once we support migration).

Quoted 4 lines

Note that we treat directly injected NMIs differently as they can hit both L1 and L2. In this case, we let L0 try to injection again also over L1 after leaving L2.

Hmm, where SDM says NMI behaves this way?

Quoted 4 lines

To avoid that we incorrectly leak an event into the architectural VCPU state that L1 wants to inject, we skip cancellation on nested run.

How the leak can happen?

Quoted 122 lines

Signed-off-by Jan Kiszka <[email protected]>

---
 arch/x86/kvm/vmx.c |  118 ++++++++++++++++++++++++++++++++++++++--------------
 1 files changed, 87 insertions(+), 31 deletions(-)

arch/x86/kvm/vmx.c +47 -29

index 126d047..ca74358 100644
@@ -6492,8 +6492,6 @@ static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
 
 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
 {
-	if (is_guest_mode(&vmx->vcpu))
-		return;
 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
 				  VM_EXIT_INSTRUCTION_LEN,
 				  IDT_VECTORING_ERROR_CODE);

@@ -6501,7 +6499,7 @@ static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
 
 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
 {
-	if (is_guest_mode(vcpu))
+	if (to_vmx(vcpu)->nested.nested_run_pending)
 		return;
 	__vmx_complete_interrupts(vcpu,
 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),

@@ -6534,21 +6532,6 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
 	struct vcpu_vmx *vmx = to_vmx(vcpu);
 	unsigned long debugctlmsr;
 
-	if (is_guest_mode(vcpu) && !vmx->nested.nested_run_pending) {
-		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
-		if (vmcs12->idt_vectoring_info_field &
-				VECTORING_INFO_VALID_MASK) {
-			vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
-				vmcs12->idt_vectoring_info_field);
-			vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
-				vmcs12->vm_exit_instruction_len);
-			if (vmcs12->idt_vectoring_info_field &
-					VECTORING_INFO_DELIVER_CODE_MASK)
-				vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
-					vmcs12->idt_vectoring_error_code);
-		}
-	}
-
 	/* Record the guest's net vcpu time for enforced NMI injections. */
 	if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
 		vmx->entry_time = ktime_get();

@@ -6707,17 +6690,6 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
 
 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
 
-	if (is_guest_mode(vcpu)) {
-		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
-		vmcs12->idt_vectoring_info_field = vmx->idt_vectoring_info;
-		if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
-			vmcs12->idt_vectoring_error_code =
-				vmcs_read32(IDT_VECTORING_ERROR_CODE);
-			vmcs12->vm_exit_instruction_len =
-				vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
-		}
-	}
-
 	vmx->loaded_vmcs->launched = 1;
 
 	vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);

@@ -7324,6 +7296,52 @@ vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
 			vcpu->arch.cr4_guest_owned_bits));
 }
 
+static void vmcs12_save_pending_events(struct kvm_vcpu *vcpu,
+				       struct vmcs12 *vmcs12)
+{
+	u32 idt_vectoring;
+	unsigned int nr;
+
+	/*
+	 * We only transfer exceptions and maskable interrupts. It is fine if
+	 * L0 retries to inject a pending NMI over L1.
+	 */
+	if (vcpu->arch.exception.pending) {
+		nr = vcpu->arch.exception.nr;
+		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
+
+		if (kvm_exception_is_soft(nr)) {
+			vmcs12->vm_exit_instruction_len =
+				vcpu->arch.event_exit_inst_len;
+			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
+		} else
+			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
+
+		if (vcpu->arch.exception.has_error_code) {
+			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
+			vmcs12->idt_vectoring_error_code =
+				vcpu->arch.exception.error_code;
+		}
+
+		vmcs12->idt_vectoring_info_field = idt_vectoring;
+	} else if (vcpu->arch.interrupt.pending) {
+		nr = vcpu->arch.interrupt.nr;
+		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
+
+		if (vcpu->arch.interrupt.soft) {
+			idt_vectoring |= INTR_TYPE_SOFT_INTR;
+			vmcs12->vm_entry_instruction_len =
+				vcpu->arch.event_exit_inst_len;
+		} else
+			idt_vectoring |= INTR_TYPE_EXT_INTR;
+
+		vmcs12->idt_vectoring_info_field = idt_vectoring;
+	}
+
+	kvm_clear_exception_queue(vcpu);
+	kvm_clear_interrupt_queue(vcpu);
+}
+
 /*
  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),

arch/x86/kvm/vmx.c static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12) · line 7440

Quoted 10 lines

arch/x86/kvm/vmx.c +5 -2

@@ -7415,9 +7433,47 @@ static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
 
-	/* clear vm-entry fields which are to be cleared on exit */
-	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
+	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
+		if ((vmcs12->vm_entry_intr_info_field &
+		     INTR_INFO_VALID_MASK) &&
+		    (vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) &
+		     INTR_INFO_VALID_MASK)) {

Again I do not see how this condition can be true.

Quoted 20 lines

arch/x86/kvm/vmx.c +20 -0

+			/*
+			 * Preserve the event that was supposed to be injected
+			 * by L1 via emulating it would have been returned in
+			 * IDT_VECTORING_INFO_FIELD.
+			 */
+			vmcs12->idt_vectoring_info_field =
+				vmcs12->vm_entry_intr_info_field;
+			vmcs12->idt_vectoring_error_code =
+				vmcs12->vm_entry_exception_error_code;
+			vmcs12->vm_exit_instruction_len =
+				vmcs12->vm_entry_instruction_len;
+			vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
+
+			/*
+			 * We do not drop NMIs that targeted L2 below as they
+			 * can also be reinjected over L1. But if this event
+			 * was an NMI, it was synthetic and came from L1.
+			 */
+			vcpu->arch.nmi_injected = false;
+		} else

arch/x86/kvm/vmx.c static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12) · line 7464

Quoted 4 lines

arch/x86/kvm/vmx.c +4 -0

+			/*
+			 * Transfer the event L0 may wanted to inject into L2
+			 * to IDT_VECTORING_INFO_FIELD.
+			 */

I do not understand the comment. This transfers an event from event queue into vmcs12. Since vmx_complete_interrupts() transfers event that L1 tried to inject into event queue too he we handle not only L0->L2, but also L1->L2 events too.

In fast I think only "else" part of this if() is needed.

Quoted 12 lines · Gleb Natapov

Quoted 11 lines · Jan Kiszka

From: Jan Kiszka <[email protected]>

The basic idea is to always transfer the pending event injection on vmexit into the architectural state of the VCPU and then drop it from there if it turns out that we left L2 to enter L1.

VMX and SVM are now identical in how they recover event injections from unperformed vmlaunch/vmresume: We detect that VM_ENTRY_INTR_INFO_FIELD still contains a valid event and, if yes, transfer the content into L1's idt_vectoring_info_field.

Quoted 4 lines

But how this can happens with VMX code? VMX has this nested_run_pending thing that prevents #vmexit emulation from happening without vmlaunch. This means that VM_ENTRY_INTR_INFO_FIELD should never be valid during #vmexit emulation since it is marked invalid during vmlaunch.

Now that nmi/interrupt_allowed is strict /wrt nested_run_pending again, it may indeed no longer happen. It was definitely a problem before, also with direct vmexit on pending INIT. Requires a second thought, maybe also a WARN_ON(vmx->nested.nested_run_pending) in nested_vmx_vmexit.

Quoted 9 lines

Quoted 9 lines

However, we differ on how to deal with events that L0 wanted to inject into L2. Likely, this case is still broken in SVM. For VMX, the function vmcs12_save_pending_events deals with transferring pending L0 events into the queue of L1. That is mandatory as L1 may decide to switch the guest state completely, invalidating or preserving the pending events for later injection (including on a different node, once we support migration).

Quoted 5 lines

Quoted 4 lines

Note that we treat directly injected NMIs differently as they can hit both L1 and L2. In this case, we let L0 try to injection again also over L1 after leaving L2.

Hmm, where SDM says NMI behaves this way?

NMIs are only blocked in root mode if we took an NMI-related vmexit (or, of course, an NMI is being processed). Thus, every arriving NMI can either hit the guest or the host - pure luck.

However, I have missed the fact that an NMI may have been injected from L1 as well. If injection triggers a vmexit, that NMI could now leak into L1. So we have to process them as well in vmcs12_save_pending_events.

Quoted 5 lines

Quoted 4 lines

To avoid that we incorrectly leak an event into the architectural VCPU state that L1 wants to inject, we skip cancellation on nested run.

How the leak can happen?

See above, this likely no longer applies.

Quoted 154 lines

Quoted 122 lines

Signed-off-by Jan Kiszka <[email protected]>

---
 arch/x86/kvm/vmx.c |  118 ++++++++++++++++++++++++++++++++++++++--------------
 1 files changed, 87 insertions(+), 31 deletions(-)

arch/x86/kvm/vmx.c +47 -29

index 126d047..ca74358 100644
@@ -6492,8 +6492,6 @@ static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
 
 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
 {
-	if (is_guest_mode(&vmx->vcpu))
-		return;
 	__vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
 				  VM_EXIT_INSTRUCTION_LEN,
 				  IDT_VECTORING_ERROR_CODE);

@@ -6501,7 +6499,7 @@ static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
 
 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
 {
-	if (is_guest_mode(vcpu))
+	if (to_vmx(vcpu)->nested.nested_run_pending)
 		return;
 	__vmx_complete_interrupts(vcpu,
 				  vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),

@@ -6534,21 +6532,6 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
 	struct vcpu_vmx *vmx = to_vmx(vcpu);
 	unsigned long debugctlmsr;
 
-	if (is_guest_mode(vcpu) && !vmx->nested.nested_run_pending) {
-		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
-		if (vmcs12->idt_vectoring_info_field &
-				VECTORING_INFO_VALID_MASK) {
-			vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
-				vmcs12->idt_vectoring_info_field);
-			vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
-				vmcs12->vm_exit_instruction_len);
-			if (vmcs12->idt_vectoring_info_field &
-					VECTORING_INFO_DELIVER_CODE_MASK)
-				vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
-					vmcs12->idt_vectoring_error_code);
-		}
-	}
-
 	/* Record the guest's net vcpu time for enforced NMI injections. */
 	if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
 		vmx->entry_time = ktime_get();

@@ -6707,17 +6690,6 @@ static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
 
 	vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
 
-	if (is_guest_mode(vcpu)) {
-		struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
-		vmcs12->idt_vectoring_info_field = vmx->idt_vectoring_info;
-		if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
-			vmcs12->idt_vectoring_error_code =
-				vmcs_read32(IDT_VECTORING_ERROR_CODE);
-			vmcs12->vm_exit_instruction_len =
-				vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
-		}
-	}
-
 	vmx->loaded_vmcs->launched = 1;
 
 	vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);

@@ -7324,6 +7296,52 @@ vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
 			vcpu->arch.cr4_guest_owned_bits));
 }
 
+static void vmcs12_save_pending_events(struct kvm_vcpu *vcpu,
+				       struct vmcs12 *vmcs12)
+{
+	u32 idt_vectoring;
+	unsigned int nr;
+
+	/*
+	 * We only transfer exceptions and maskable interrupts. It is fine if
+	 * L0 retries to inject a pending NMI over L1.
+	 */
+	if (vcpu->arch.exception.pending) {
+		nr = vcpu->arch.exception.nr;
+		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
+
+		if (kvm_exception_is_soft(nr)) {
+			vmcs12->vm_exit_instruction_len =
+				vcpu->arch.event_exit_inst_len;
+			idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
+		} else
+			idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
+
+		if (vcpu->arch.exception.has_error_code) {
+			idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
+			vmcs12->idt_vectoring_error_code =
+				vcpu->arch.exception.error_code;
+		}
+
+		vmcs12->idt_vectoring_info_field = idt_vectoring;
+	} else if (vcpu->arch.interrupt.pending) {
+		nr = vcpu->arch.interrupt.nr;
+		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
+
+		if (vcpu->arch.interrupt.soft) {
+			idt_vectoring |= INTR_TYPE_SOFT_INTR;
+			vmcs12->vm_entry_instruction_len =
+				vcpu->arch.event_exit_inst_len;
+		} else
+			idt_vectoring |= INTR_TYPE_EXT_INTR;
+
+		vmcs12->idt_vectoring_info_field = idt_vectoring;
+	}
+
+	kvm_clear_exception_queue(vcpu);
+	kvm_clear_interrupt_queue(vcpu);
+}
+
 /*
  * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
  * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),

arch/x86/kvm/vmx.c static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12) · line 7440

Quoted 10 lines

arch/x86/kvm/vmx.c +5 -2

@@ -7415,9 +7433,47 @@ static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
 	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
 	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
 
-	/* clear vm-entry fields which are to be cleared on exit */
-	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
+	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
+		if ((vmcs12->vm_entry_intr_info_field &
+		     INTR_INFO_VALID_MASK) &&
+		    (vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) &
+		     INTR_INFO_VALID_MASK)) {

Again I do not see how this condition can be true.

Quoted 20 lines

arch/x86/kvm/vmx.c +20 -0

+			/*
+			 * Preserve the event that was supposed to be injected
+			 * by L1 via emulating it would have been returned in
+			 * IDT_VECTORING_INFO_FIELD.
+			 */
+			vmcs12->idt_vectoring_info_field =
+				vmcs12->vm_entry_intr_info_field;
+			vmcs12->idt_vectoring_error_code =
+				vmcs12->vm_entry_exception_error_code;
+			vmcs12->vm_exit_instruction_len =
+				vmcs12->vm_entry_instruction_len;
+			vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
+
+			/*
+			 * We do not drop NMIs that targeted L2 below as they
+			 * can also be reinjected over L1. But if this event
+			 * was an NMI, it was synthetic and came from L1.
+			 */
+			vcpu->arch.nmi_injected = false;
+		} else

Quoted 8 lines

arch/x86/kvm/vmx.c static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12) · line 7464

Quoted 4 lines

arch/x86/kvm/vmx.c +4 -0

+			/*
+			 * Transfer the event L0 may wanted to inject into L2
+			 * to IDT_VECTORING_INFO_FIELD.
+			 */

I do not understand the comment. This transfers an event from event queue into vmcs12. Since vmx_complete_interrupts() transfers event that L1 tried to inject into event queue too he we handle not only L0->L2, but also L1->L2 events too.

I'm not sure if I fully understand your remark. Is it that the comment is only talking about L0 events? That is indeed not fully true, L1 events should make it to the architectural queue as well. Will adjust this.

Quoted 1 line

In fast I think only "else" part of this if() is needed.

Yes, probably.

Thanks,
Jan

Quoted 42 lines · Jan Kiszka

Quoted 12 lines · Gleb Natapov

Quoted 11 lines · Jan Kiszka

From: Jan Kiszka <[email protected]>

The basic idea is to always transfer the pending event injection on vmexit into the architectural state of the VCPU and then drop it from there if it turns out that we left L2 to enter L1.

VMX and SVM are now identical in how they recover event injections from unperformed vmlaunch/vmresume: We detect that VM_ENTRY_INTR_INFO_FIELD still contains a valid event and, if yes, transfer the content into L1's idt_vectoring_info_field.

Quoted 4 lines

But how this can happens with VMX code? VMX has this nested_run_pending thing that prevents #vmexit emulation from happening without vmlaunch. This means that VM_ENTRY_INTR_INFO_FIELD should never be valid during #vmexit emulation since it is marked invalid during vmlaunch.

Now that nmi/interrupt_allowed is strict /wrt nested_run_pending again, it may indeed no longer happen. It was definitely a problem before, also with direct vmexit on pending INIT. Requires a second thought, maybe also a WARN_ON(vmx->nested.nested_run_pending) in nested_vmx_vmexit.

Quoted 9 lines

Quoted 9 lines

However, we differ on how to deal with events that L0 wanted to inject into L2. Likely, this case is still broken in SVM. For VMX, the function vmcs12_save_pending_events deals with transferring pending L0 events into the queue of L1. That is mandatory as L1 may decide to switch the guest state completely, invalidating or preserving the pending events for later injection (including on a different node, once we support migration).

Quoted 5 lines

Quoted 4 lines

Note that we treat directly injected NMIs differently as they can hit both L1 and L2. In this case, we let L0 try to injection again also over L1 after leaving L2.

Hmm, where SDM says NMI behaves this way?

NMIs are only blocked in root mode if we took an NMI-related vmexit (or, of course, an NMI is being processed). Thus, every arriving NMI can either hit the guest or the host - pure luck.

Quoted 4 lines

However, I have missed the fact that an NMI may have been injected from L1 as well. If injection triggers a vmexit, that NMI could now leak into L1. So we have to process them as well in vmcs12_save_pending_events.

You mean "should not leak into L0" not L1?

Quoted 43 lines · Gleb Natapov

Quoted 42 lines · Jan Kiszka

Quoted 12 lines · Gleb Natapov

Quoted 11 lines · Jan Kiszka

From: Jan Kiszka <[email protected]>

The basic idea is to always transfer the pending event injection on vmexit into the architectural state of the VCPU and then drop it from there if it turns out that we left L2 to enter L1.

VMX and SVM are now identical in how they recover event injections from unperformed vmlaunch/vmresume: We detect that VM_ENTRY_INTR_INFO_FIELD still contains a valid event and, if yes, transfer the content into L1's idt_vectoring_info_field.

Quoted 4 lines

But how this can happens with VMX code? VMX has this nested_run_pending thing that prevents #vmexit emulation from happening without vmlaunch. This means that VM_ENTRY_INTR_INFO_FIELD should never be valid during #vmexit emulation since it is marked invalid during vmlaunch.

Now that nmi/interrupt_allowed is strict /wrt nested_run_pending again, it may indeed no longer happen. It was definitely a problem before, also with direct vmexit on pending INIT. Requires a second thought, maybe also a WARN_ON(vmx->nested.nested_run_pending) in nested_vmx_vmexit.

Quoted 9 lines

Quoted 9 lines

However, we differ on how to deal with events that L0 wanted to inject into L2. Likely, this case is still broken in SVM. For VMX, the function vmcs12_save_pending_events deals with transferring pending L0 events into the queue of L1. That is mandatory as L1 may decide to switch the guest state completely, invalidating or preserving the pending events for later injection (including on a different node, once we support migration).

Quoted 5 lines

Quoted 4 lines

Note that we treat directly injected NMIs differently as they can hit both L1 and L2. In this case, we let L0 try to injection again also over L1 after leaving L2.

Hmm, where SDM says NMI behaves this way?

NMIs are only blocked in root mode if we took an NMI-related vmexit (or, of course, an NMI is being processed). Thus, every arriving NMI can either hit the guest or the host - pure luck.

Quoted 5 lines

Quoted 4 lines

However, I have missed the fact that an NMI may have been injected from L1 as well. If injection triggers a vmexit, that NMI could now leak into L1. So we have to process them as well in vmcs12_save_pending_events.

You mean "should not leak into L0" not L1?

No, L1. If we keep the NMI in the architectural queue, L0 will try to reinject it over L1 after the vmexit to it.

Version 2 both takes review comments into account, reorders some patches that have dependencies and also addresses new findings regarding NMI injections. Fixes for vmx_interrupt_allowed and vmx_nmi_allowed were split up as they turn out to be more different to each other and are independent anyway. Finally there was still a bug in the handling of EXIT_REASON_NMI_WINDOW - the wrong condition was checked.

Applies on top of 'queue'.

Jan Kiszka (5):
KVM: nVMX: Fix injection of PENDING_INTERRUPT and NMI_WINDOW exits to

    L1

KVM: nVMX: Rework event injection and recovery
KVM: VMX: Move vmx_nmi_allowed after vmx_set_nmi_mask
KVM: nVMX: Fix conditions for interrupt injection
KVM: nVMX: Fix conditions for NMI injection

 arch/x86/kvm/vmx.c |  210 ++++++++++++++++++++++++++++++++++++++--------------
 1 files changed, 154 insertions(+), 56 deletions(-)