Skip to contentSkip to conversation

kvm: x86: Add logical CPU to KVM_EXIT_FAIL_ENTRY info

PATCH Superseded Jim Mattson 3 participants 4 messages Dec 13, 2019 – Dec 14, 2019

Subsystems (4)
Documentation x86 KVM KVM x86

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Jim Mattson

author
#0 lore raw

PATCH 1/1 kvm: x86: Add logical CPU to KVM_EXIT_FAIL_ENTRY info

More often than not, a failed VM-entry in a production environment is the result of a defective CPU (at least, insofar as Intel x86 is concerned). To aid in identifying the bad hardware, add the logical CPU to the information provided to userspace on a KVM exit with reason KVM_EXIT_FAIL_ENTRY. The presence of this additional information is indicated by a new capability, KVM_CAP_FAILED_ENTRY_CPU.

Signed-off-by Jim Mattson <[email protected]>
Reviewed-by Peter Shier <[email protected]>
Reviewed-by Oliver Upton <[email protected]>
---
 Documentation/virt/kvm/api.txt | 1 +
 arch/x86/kvm/svm.c             | 1 +
 arch/x86/kvm/vmx/vmx.c         | 2 ++
 arch/x86/kvm/x86.c             | 1 +
 include/uapi/linux/kvm.h       | 2 ++
 5 files changed, 7 insertions(+)
Documentation/virt/kvm/api.txt +1 -0
index ebb37b34dcfc..6e5d92406b65 100644
@@ -4245,6 +4245,7 @@ hardware_exit_reason.
 		/* KVM_EXIT_FAIL_ENTRY */
 		struct {
 			__u64 hardware_entry_failure_reason;
+			__u32 cpu; /* if KVM_CAP_FAILED_ENTRY_CPU */
 		} fail_entry;
 
 If exit_reason is KVM_EXIT_FAIL_ENTRY, the vcpu could not be run due
arch/x86/kvm/svm.c +1 -0
index 122d4ce3b1ab..4d06b2413c63 100644
@@ -4980,6 +4980,7 @@ static int handle_exit(struct kvm_vcpu *vcpu)
 		kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
 		kvm_run->fail_entry.hardware_entry_failure_reason
 			= svm->vmcb->control.exit_code;
+		kvm_run->fail_entry.cpu = raw_smp_processor_id();
 		dump_vmcb(vcpu);
 		return 0;
 	}
arch/x86/kvm/vmx/vmx.c +2 -0
index e3394c839dea..4d540b1c08e0 100644
@@ -5846,6 +5846,7 @@ static int vmx_handle_exit(struct kvm_vcpu *vcpu)
 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
 		vcpu->run->fail_entry.hardware_entry_failure_reason
 			= exit_reason;
+		vcpu->run->fail_entry.cpu = vmx->loaded_vmcs->cpu;
 		return 0;
 	}
 

@@ -5854,6 +5855,7 @@ static int vmx_handle_exit(struct kvm_vcpu *vcpu)
 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
 		vcpu->run->fail_entry.hardware_entry_failure_reason
 			= vmcs_read32(VM_INSTRUCTION_ERROR);
+		vcpu->run->fail_entry.cpu = vmx->loaded_vmcs->cpu;
 		return 0;
 	}
arch/x86/kvm/x86.c +1 -0
index cf917139de6b..9e89a32056d1 100644
@@ -3273,6 +3273,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	case KVM_CAP_GET_MSR_FEATURES:
 	case KVM_CAP_MSR_PLATFORM_INFO:
 	case KVM_CAP_EXCEPTION_PAYLOAD:
+	case KVM_CAP_FAILED_ENTRY_CPU:
 		r = 1;
 		break;
 	case KVM_CAP_SYNC_REGS:
include/uapi/linux/kvm.h +2 -0
index f0a16b4adbbd..09ba7174456d 100644
@@ -277,6 +277,7 @@ struct kvm_run {
 		/* KVM_EXIT_FAIL_ENTRY */
 		struct {
 			__u64 hardware_entry_failure_reason;
+			__u32 cpu;
 		} fail_entry;
 		/* KVM_EXIT_EXCEPTION */
 		struct {

@@ -1009,6 +1010,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_PPC_GUEST_DEBUG_SSTEP 176
 #define KVM_CAP_ARM_NISV_TO_USER 177
 #define KVM_CAP_ARM_INJECT_EXT_DABT 178
+#define KVM_CAP_FAILED_ENTRY_CPU 179
 
 #ifdef KVM_CAP_IRQ_ROUTING
Signature · 1 line
2.24.1.735.g03f4e72817-goog
#1 lore raw
Quoted 39 lines · Jim Mattson

More often than not, a failed VM-entry in a production environment is the result of a defective CPU (at least, insofar as Intel x86 is concerned). To aid in identifying the bad hardware, add the logical CPU to the information provided to userspace on a KVM exit with reason KVM_EXIT_FAIL_ENTRY. The presence of this additional information is indicated by a new capability, KVM_CAP_FAILED_ENTRY_CPU.

Signed-off-by Jim Mattson <[email protected]>
Reviewed-by Peter Shier <[email protected]>
Reviewed-by Oliver Upton <[email protected]>

---

Documentation/virt/kvm/api.txt | 1 +
arch/x86/kvm/svm.c             | 1 +
arch/x86/kvm/vmx/vmx.c         | 2 ++
arch/x86/kvm/x86.c             | 1 +
include/uapi/linux/kvm.h       | 2 ++
5 files changed, 7 insertions(+)
Documentation/virt/kvm/api.txt +1 -0
index ebb37b34dcfc..6e5d92406b65 100644
@@ -4245,6 +4245,7 @@ hardware_exit_reason.
 		/* KVM_EXIT_FAIL_ENTRY */
 		struct {
 			__u64 hardware_entry_failure_reason;
+			__u32 cpu; /* if KVM_CAP_FAILED_ENTRY_CPU */
 		} fail_entry;

If exit_reason is KVM_EXIT_FAIL_ENTRY, the vcpu could not be run due

arch/x86/kvm/svm.c +0 -0
index 122d4ce3b1ab..4d06b2413c63 100644

arch/x86/kvm/svm.c static int handle_exit(struct kvm_vcpu *vcpu) · line 4983

Quoted 4 lines
arch/x86/kvm/svm.c +1 -0
@@ -4980,6 +4980,7 @@ static int handle_exit(struct kvm_vcpu *vcpu)
 		kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
 		kvm_run->fail_entry.hardware_entry_failure_reason
 			= svm->vmcb->control.exit_code;
+		kvm_run->fail_entry.cpu = raw_smp_processor_id();

Why not just use vcpu->cpu?
Same for vmx_handle_exit() to be consistent.

Quoted 58 lines
arch/x86/kvm/svm.c +0 -0
 		dump_vmcb(vcpu);
 		return 0;
 	}
arch/x86/kvm/vmx/vmx.c +2 -0
index e3394c839dea..4d540b1c08e0 100644
@@ -5846,6 +5846,7 @@ static int vmx_handle_exit(struct kvm_vcpu *vcpu)
 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
 		vcpu->run->fail_entry.hardware_entry_failure_reason
 			= exit_reason;
+		vcpu->run->fail_entry.cpu = vmx->loaded_vmcs->cpu;
 		return 0;
 	}
 

@@ -5854,6 +5855,7 @@ static int vmx_handle_exit(struct kvm_vcpu *vcpu)
 		vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
 		vcpu->run->fail_entry.hardware_entry_failure_reason
 			= vmcs_read32(VM_INSTRUCTION_ERROR);
+		vcpu->run->fail_entry.cpu = vmx->loaded_vmcs->cpu;
 		return 0;
 	}
arch/x86/kvm/x86.c +1 -0
index cf917139de6b..9e89a32056d1 100644
@@ -3273,6 +3273,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	case KVM_CAP_GET_MSR_FEATURES:
 	case KVM_CAP_MSR_PLATFORM_INFO:
 	case KVM_CAP_EXCEPTION_PAYLOAD:
+	case KVM_CAP_FAILED_ENTRY_CPU:
 		r = 1;
 		break;
 	case KVM_CAP_SYNC_REGS:
include/uapi/linux/kvm.h +2 -0
index f0a16b4adbbd..09ba7174456d 100644
@@ -277,6 +277,7 @@ struct kvm_run {
 		/* KVM_EXIT_FAIL_ENTRY */
 		struct {
 			__u64 hardware_entry_failure_reason;
+			__u32 cpu;
 		} fail_entry;
 		/* KVM_EXIT_EXCEPTION */
 		struct {

@@ -1009,6 +1010,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_PPC_GUEST_DEBUG_SSTEP 176
 #define KVM_CAP_ARM_NISV_TO_USER 177
 #define KVM_CAP_ARM_INJECT_EXT_DABT 178
+#define KVM_CAP_FAILED_ENTRY_CPU 179
 
 #ifdef KVM_CAP_IRQ_ROUTING
Signature · 1 line
2.24.1.735.g03f4e72817-goog
author
#2 lore raw
Quoted 34 lines · Liran Alon
Quoted 34 lines · Jim Mattson

More often than not, a failed VM-entry in a production environment is the result of a defective CPU (at least, insofar as Intel x86 is concerned). To aid in identifying the bad hardware, add the logical CPU to the information provided to userspace on a KVM exit with reason KVM_EXIT_FAIL_ENTRY. The presence of this additional information is indicated by a new capability, KVM_CAP_FAILED_ENTRY_CPU.

Signed-off-by Jim Mattson <[email protected]>
Reviewed-by Peter Shier <[email protected]>
Reviewed-by Oliver Upton <[email protected]>

---

Documentation/virt/kvm/api.txt | 1 +
arch/x86/kvm/svm.c             | 1 +
arch/x86/kvm/vmx/vmx.c         | 2 ++
arch/x86/kvm/x86.c             | 1 +
include/uapi/linux/kvm.h       | 2 ++
5 files changed, 7 insertions(+)
Documentation/virt/kvm/api.txt +1 -0
index ebb37b34dcfc..6e5d92406b65 100644
@@ -4245,6 +4245,7 @@ hardware_exit_reason.
              /* KVM_EXIT_FAIL_ENTRY */
              struct {
                      __u64 hardware_entry_failure_reason;
+                     __u32 cpu; /* if KVM_CAP_FAILED_ENTRY_CPU */
              } fail_entry;
Quoted 13 lines

arch/x86/kvm/svm.c static int handle_exit(struct kvm_vcpu *vcpu) · line 4983

Quoted 10 lines

If exit_reason is KVM_EXIT_FAIL_ENTRY, the vcpu could not be run due

arch/x86/kvm/svm.c +1 -0
index 122d4ce3b1ab..4d06b2413c63 100644
@@ -4980,6 +4980,7 @@ static int handle_exit(struct kvm_vcpu *vcpu)
              kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
              kvm_run->fail_entry.hardware_entry_failure_reason
                      = svm->vmcb->control.exit_code;
+             kvm_run->fail_entry.cpu = raw_smp_processor_id();

Why not just use vcpu->cpu?
Same for vmx_handle_exit() to be consistent.

Ah. Perfect. Thanks.

Quoted 59 lines
Quoted 58 lines
arch/x86/kvm/svm.c +0 -0
              dump_vmcb(vcpu);
              return 0;
      }
arch/x86/kvm/vmx/vmx.c +2 -0
index e3394c839dea..4d540b1c08e0 100644
@@ -5846,6 +5846,7 @@ static int vmx_handle_exit(struct kvm_vcpu *vcpu)
              vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
              vcpu->run->fail_entry.hardware_entry_failure_reason
                      = exit_reason;
+             vcpu->run->fail_entry.cpu = vmx->loaded_vmcs->cpu;
              return 0;
      }
 

@@ -5854,6 +5855,7 @@ static int vmx_handle_exit(struct kvm_vcpu *vcpu)
              vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
              vcpu->run->fail_entry.hardware_entry_failure_reason
                      = vmcs_read32(VM_INSTRUCTION_ERROR);
+             vcpu->run->fail_entry.cpu = vmx->loaded_vmcs->cpu;
              return 0;
      }
arch/x86/kvm/x86.c +1 -0
index cf917139de6b..9e89a32056d1 100644
@@ -3273,6 +3273,7 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
      case KVM_CAP_GET_MSR_FEATURES:
      case KVM_CAP_MSR_PLATFORM_INFO:
      case KVM_CAP_EXCEPTION_PAYLOAD:
+     case KVM_CAP_FAILED_ENTRY_CPU:
              r = 1;
              break;
      case KVM_CAP_SYNC_REGS:
include/uapi/linux/kvm.h +2 -1
index f0a16b4adbbd..09ba7174456d 100644
@@ -277,6 +277,7 @@ struct kvm_run {
              /* KVM_EXIT_FAIL_ENTRY */
              struct {
                      __u64 hardware_entry_failure_reason;
+                     __u32 cpu;
              } fail_entry;
              /* KVM_EXIT_EXCEPTION */
              struct {

@@ -1009,6 +1010,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_PPC_GUEST_DEBUG_SSTEP 176
 #define KVM_CAP_ARM_NISV_TO_USER 177
 #define KVM_CAP_ARM_INJECT_EXT_DABT 178
+#define KVM_CAP_FAILED_ENTRY_CPU 179
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
--

2.24.1.735.g03f4e72817-goog

maintainer
#3 lore raw
In reply to Jim Mattson
Quoted 3 lines · Jim Mattson

More often than not, a failed VM-entry in a production environment is the result of a defective CPU (at least, insofar as Intel x86 is concerned).

It's conforting that someone else got to the same conclusion as we did...

Paolo

Quoted 8 lines

To aid in identifying the bad hardware, add the logical
CPU to the information provided to userspace on a KVM exit with reason KVM_EXIT_FAIL_ENTRY. The presence of this additional information is indicated by a new capability, KVM_CAP_FAILED_ENTRY_CPU.

Signed-off-by Jim Mattson <[email protected]>
Reviewed-by Peter Shier <[email protected]>
Reviewed-by Oliver Upton <[email protected]>