Skip to contentSkip to conversation

KVM: x86: Nested fixes (mostly #DF/#TF)

PATCH Applied Sean Christopherson 3 participants 7 messages Apr 7, 2022 – Apr 8, 2022

Subsystems (3)
lore
Share & tools
lore

Sean Christopherson

author maintainer
#0 lore raw

cover KVM: x86: Nested fixes (mostly #DF/#TF)

Patch 1 removes an assertion that is going to fail miserably once KVM allows save/restore of pending triple faults.

Patches 2 and 3 are fixes for edge cases for nVMX related to saving exit info on failed VM-Entry, #DF and triple fault, that in all likelihood no one cares about.

Sean Christopherson (3):
KVM: x86: Drop WARNs that assert a triple fault never "escapes" from

    L2

KVM: nVMX: Leave most VM-Exit info fields unmodified on failed
VM-Entry
KVM: nVMX: Clear IDT vectoring on nested VM-Exit for double/triple
fault

 arch/x86/kvm/svm/nested.c |  3 ---
 arch/x86/kvm/vmx/nested.c | 48 ++++++++++++++++++++++++++++++---------
 arch/x86/kvm/vmx/vmcs.h   |  5 ++++
 3 files changed, 42 insertions(+), 14 deletions(-)

base-commit: 7d4cda14f55ae6998220aaecd2ab00e20d0e8f57

Signature · 1 line
2.35.1.1094.g7c7d902a7c-goog
author maintainer
#1 lore raw

PATCH 1/3 KVM: x86: Drop WARNs that assert a triple fault never "escapes" from L2

Remove WARNs that sanity check that KVM never lets a triple fault for L2 escape and incorrectly end up in L1. In normal operation, the sanity check is perfectly valid, but it incorrectly assumes that it's impossible for userspace to induce KVM_REQ_TRIPLE_FAULT without bouncing through KVM_RUN (which guarantees kvm_check_nested_state() will see and handle the triple fault).

The WARN can currently be triggered if userspace injects a machine check while L2 is active and CR4.MCE=0. And a future fix to allow save/restore of KVM_REQ_TRIPLE_FAULT, e.g. so that a synthesized triple fault isn't lost on migration, will make it trivially easy for userspace to trigger the WARN.

Clearing KVM_REQ_TRIPLE_FAULT when forcibly leaving guest mode is tempting, but wrong, especially if/when the request is saved/restored, e.g. if userspace restores events (including a triple fault) and then restores nested state (which may forcibly leave guest mode). Ignoring the fact that KVM doesn't currently provide the necessary APIs, it's userspace's responsibility to manage pending events during save/restore.

  ------------[ cut here ]------------
  WARNING: CPU: 7 PID: 1399 at arch/x86/kvm/vmx/nested.c:4522 nested_vmx_vmexit+0x7fe/0xd90 [kvm_intel]
  Modules linked in: kvm_intel kvm irqbypass
  CPU: 7 PID: 1399 Comm: state_test Not tainted 5.17.0-rc3+ #808
  Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015
  RIP: 0010:nested_vmx_vmexit+0x7fe/0xd90 [kvm_intel]
  Call Trace:
   <TASK>
   vmx_leave_nested+0x30/0x40 [kvm_intel]
   vmx_set_nested_state+0xca/0x3e0 [kvm_intel]
   kvm_arch_vcpu_ioctl+0xf49/0x13e0 [kvm]
   kvm_vcpu_ioctl+0x4b9/0x660 [kvm]
   __x64_sys_ioctl+0x83/0xb0
   do_syscall_64+0x3b/0xc0
   entry_SYSCALL_64_after_hwframe+0x44/0xae
   </TASK>
  ---[ end trace 0000000000000000 ]---
Fixes cb6a32c2b877 ("KVM: x86: Handle triple fault in L2 without killing L1")
Cc Chenyi Qiang <[email protected]>
---
 arch/x86/kvm/svm/nested.c | 3 ---
 arch/x86/kvm/vmx/nested.c | 3 ---
 2 files changed, 6 deletions(-)
arch/x86/kvm/svm/nested.c +0 -3
index c1ef7867797a..bed5e1692cef 100644
@@ -894,9 +894,6 @@ int nested_svm_vmexit(struct vcpu_svm *svm)
 	struct kvm_host_map map;
 	int rc;
 
-	/* Triple faults in L2 should never escape. */
-	WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
-
 	rc = kvm_vcpu_map(vcpu, gpa_to_gfn(svm->nested.vmcb12_gpa), &map);
 	if (rc) {
 		if (rc == -EINVAL)
arch/x86/kvm/vmx/nested.c +0 -3
index 50d83b9f8067..ec4cbf583921 100644
@@ -4514,9 +4514,6 @@ void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 vm_exit_reason,
 	/* trying to cancel vmlaunch/vmresume is a bug */
 	WARN_ON_ONCE(vmx->nested.nested_run_pending);
 
-	/* Similarly, triple faults in L2 should never escape. */
-	WARN_ON_ONCE(kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu));
-
 	if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
 		/*
 		 * KVM_REQ_GET_NESTED_STATE_PAGES is also used to map
Signature · 1 line
2.35.1.1094.g7c7d902a7c-goog
author maintainer
#2 lore raw

PATCH 2/3 KVM: nVMX: Leave most VM-Exit info fields unmodified on failed VM-Entry

In reply to Sean Christopherson

Don't modify vmcs12 exit fields except EXIT_REASON and EXIT_QUALIFICATION when performing a nested VM-Exit due to failed VM-Entry. Per the SDM, only the two aformentioned fields are filled and "All other VM-exit information fields are unmodified".

Fixes 4704d0befb07 ("KVM: nVMX: Exiting from L2 to L1")
---
 arch/x86/kvm/vmx/nested.c | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)
arch/x86/kvm/vmx/nested.c +10 -5
index ec4cbf583921..9a4938955bad 100644
@@ -4198,12 +4198,12 @@ static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
 	if (to_vmx(vcpu)->exit_reason.enclave_mode)
 		vmcs12->vm_exit_reason |= VMX_EXIT_REASONS_SGX_ENCLAVE_MODE;
 	vmcs12->exit_qualification = exit_qualification;
-	vmcs12->vm_exit_intr_info = exit_intr_info;
-
-	vmcs12->idt_vectoring_info_field = 0;
-	vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
-	vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
 
+	/*
+	 * On VM-Exit due to a failed VM-Entry, the VMCS isn't marked launched
+	 * and only EXIT_REASON and EXIT_QUALIFICATION are updated, all other
+	 * exit info fields are unmodified.
+	 */
 	if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
 		vmcs12->launch_state = 1;
 
@@ -4215,8 +4215,13 @@ static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
 		 * Transfer the event that L0 or L1 may wanted to inject into
 		 * L2 to IDT_VECTORING_INFO_FIELD.
 		 */
+		vmcs12->idt_vectoring_info_field = 0;
 		vmcs12_save_pending_event(vcpu, vmcs12);
 
+		vmcs12->vm_exit_intr_info = exit_intr_info;
+		vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
+		vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
+
 		/*
 		 * According to spec, there's no need to store the guest's
 		 * MSRs if the exit is due to a VM-entry failure that occurs
Signature · 1 line
2.35.1.1094.g7c7d902a7c-goog
author maintainer
#3 lore raw

PATCH 3/3 KVM: nVMX: Clear IDT vectoring on nested VM-Exit for double/triple fault

In reply to Sean Christopherson

Clear the IDT vectoring field in vmcs12 on next VM-Exit due to a double or triple fault. Per the SDM, a VM-Exit isn't considered to occur during event delivery if the exit is due to an intercepted double fault or a triple fault. Opportunistically move the default clearing (no event "pending") into the helper so that it's more obvious that KVM does indeed handle this case.

Note, the double fault case is worded rather wierdly in the SDM:

The original event results in a double-fault exception that causes the VM exit directly.

Temporarily ignoring injected events, double faults can _only_ occur if an exception occurs while attempting to deliver a different exception, i.e. there's _always_ an original event. And for injected double fault, while there's no original event, injected events are never subject to interception.

Presumably the SDM is calling out that a the vectoring info will be valid if a different exit occurs after a double fault, e.g. if a #PF occurs and is intercepted while vectoring #DF, then the vectoring info will show the double fault. In other words, the clause can simply be read as:

The VM exit is caused by a double-fault exception.

Fixes 4704d0befb07 ("KVM: nVMX: Exiting from L2 to L1")
Cc Chenyi Qiang <[email protected]>
---
 arch/x86/kvm/vmx/nested.c | 32 ++++++++++++++++++++++++++++----
 arch/x86/kvm/vmx/vmcs.h   |  5 +++++
 2 files changed, 33 insertions(+), 4 deletions(-)
arch/x86/kvm/vmx/nested.c +28 -4
index 9a4938955bad..a6688663da4d 100644
@@ -3691,12 +3691,34 @@ vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
 }
 
 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
-				      struct vmcs12 *vmcs12)
+				      struct vmcs12 *vmcs12,
+				      u32 vm_exit_reason, u32 exit_intr_info)
 {
 	u32 idt_vectoring;
 	unsigned int nr;
 
-	if (vcpu->arch.exception.injected) {
+	/*
+	 * Per the SDM, VM-Exits due to double and triple faults are never
+	 * considered to occur during event delivery, even if the double/triple
+	 * fault is the result of an escalating vectoring issue.
+	 *
+	 * Note, the SDM qualifies the double fault behavior with "The original
+	 * event results in a double-fault exception".  It's unclear why the
+	 * qualification exists since exits due to double fault can occur only
+	 * while vectoring a different exception (injected events are never
+	 * subject to interception), i.e. there's _always_ an original event.
+	 *
+	 * The SDM also uses NMI as a confusing example for the "original event
+	 * causes the VM exit directly" clause.  NMI isn't special in any way,
+	 * the same rule applies to all events that cause an exit directly.
+	 * NMI is an odd choice for the example because NMIs can only occur on
+	 * instruction boundaries, i.e. they _can't_ occur during vectoring.
+	 */
+	if ((u16)vm_exit_reason == EXIT_REASON_TRIPLE_FAULT ||
+	    ((u16)vm_exit_reason == EXIT_REASON_EXCEPTION_NMI &&
+	     is_double_fault(exit_intr_info))) {
+		vmcs12->idt_vectoring_info_field = 0;
+	} else if (vcpu->arch.exception.injected) {
 		nr = vcpu->arch.exception.nr;
 		idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
 
@@ -3729,6 +3751,8 @@ static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
 			idt_vectoring |= INTR_TYPE_EXT_INTR;
 
 		vmcs12->idt_vectoring_info_field = idt_vectoring;
+	} else {
+		vmcs12->idt_vectoring_info_field = 0;
 	}
 }
 
@@ -4215,8 +4239,8 @@ static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
 		 * Transfer the event that L0 or L1 may wanted to inject into
 		 * L2 to IDT_VECTORING_INFO_FIELD.
 		 */
-		vmcs12->idt_vectoring_info_field = 0;
-		vmcs12_save_pending_event(vcpu, vmcs12);
+		vmcs12_save_pending_event(vcpu, vmcs12,
+					  vm_exit_reason, exit_intr_info);
 
 		vmcs12->vm_exit_intr_info = exit_intr_info;
 		vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
arch/x86/kvm/vmx/vmcs.h +5 -0
index e325c290a816..2b9d7a7e83f7 100644
@@ -104,6 +104,11 @@ static inline bool is_breakpoint(u32 intr_info)
 	return is_exception_n(intr_info, BP_VECTOR);
 }
 
+static inline bool is_double_fault(u32 intr_info)
+{
+	return is_exception_n(intr_info, DF_VECTOR);
+}
+
 static inline bool is_page_fault(u32 intr_info)
 {
 	return is_exception_n(intr_info, PF_VECTOR);
Signature · 1 line
2.35.1.1094.g7c7d902a7c-goog
#4 lore raw

Re: [PATCH 3/3] KVM: nVMX: Clear IDT vectoring on nested VM-Exit for double/triple fault

Quoted 18 lines · Sean Christopherson

Clear the IDT vectoring field in vmcs12 on next VM-Exit due to a double or triple fault. Per the SDM, a VM-Exit isn't considered to occur during event delivery if the exit is due to an intercepted double fault or a triple fault. Opportunistically move the default clearing (no event "pending") into the helper so that it's more obvious that KVM does indeed handle this case.

Note, the double fault case is worded rather wierdly in the SDM:

The original event results in a double-fault exception that causes the VM exit directly.

Temporarily ignoring injected events, double faults can _only_ occur if an exception occurs while attempting to deliver a different exception, i.e. there's _always_ an original event. And for injected double fault, while there's no original event, injected events are never subject to interception.

Quoted 4 lines

Presumably the SDM is calling out that a the vectoring info will be valid if a different exit occurs after a double fault, e.g. if a #PF occurs and is intercepted while vectoring #DF, then the vectoring info will show the double fault.

Wouldn't it be a tripe fault exit in this case?

Quoted 3 lines

In other words, the clause can simply be read as:

The VM exit is caused by a double-fault exception.

#5 lore raw

Re: [PATCH 3/3] KVM: nVMX: Clear IDT vectoring on nested VM-Exit for double/triple fault

Quoted 19 lines · Xiaoyao Li
Quoted 18 lines · Sean Christopherson

Clear the IDT vectoring field in vmcs12 on next VM-Exit due to a double or triple fault.  Per the SDM, a VM-Exit isn't considered to occur during event delivery if the exit is due to an intercepted double fault or a triple fault.  Opportunistically move the default clearing (no event "pending") into the helper so that it's more obvious that KVM does indeed handle this case.

Note, the double fault case is worded rather wierdly in the SDM:

   The original event results in a double-fault exception that causes the VM exit directly.

Temporarily ignoring injected events, double faults can _only_ occur if an exception occurs while attempting to deliver a different exception, i.e. there's _always_ an original event.  And for injected double fault, while there's no original event, injected events are never subject to interception.

Quoted 6 lines
Quoted 4 lines

Presumably the SDM is calling out that a the vectoring info will be valid if a different exit occurs after a double fault, e.g. if a #PF occurs and is intercepted while vectoring #DF, then the vectoring info will show the double fault.

Wouldn't it be a tripe fault exit in this case?

It won't since #PF is intercepted by exception bitmap in your case.

Quoted 4 lines
Quoted 3 lines

In other words, the clause can simply be read as:

   The VM exit is caused by a double-fault exception.