[v2] kvm: x86: Add logical CPU to KVM_EXIT_FAIL_ENTRY info
diff mbox series

Message ID 20191214002014.144430-1-jmattson@google.com
State New
Headers show
Series
  • [v2] kvm: x86: Add logical CPU to KVM_EXIT_FAIL_ENTRY info
Related show

Commit Message

Jim Mattson Dec. 14, 2019, 12:20 a.m. UTC
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 <jmattson@google.com>
Reviewed-by: Oliver Upton <oupton@google.com>
Cc: Liran Alon <liran.alon@oracle.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
---
 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(+)

Comments

Liran Alon Dec. 14, 2019, 1:10 a.m. UTC | #1
> On 14 Dec 2019, at 2:20, Jim Mattson <jmattson@google.com> wrote:
> 
> 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 <jmattson@google.com>
> Reviewed-by: Oliver Upton <oupton@google.com>
> Cc: Liran Alon <liran.alon@oracle.com>
> Cc: Paolo Bonzini <pbonzini@redhat.com>

Reviewed-by: Liran Alon <liran.alon@oracle.com>

BTW, one could argue that receiving an unexpected exit-reason (i.e. KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON)
could also only occur in production either from a KVM bug or from a defective CPU. Similar to failed VM-entry.
Should we add similar behaviour to that as-well?

-Liran

> ---
> 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(+)
> 
> diff --git a/Documentation/virt/kvm/api.txt b/Documentation/virt/kvm/api.txt
> index ebb37b34dcfc..6e5d92406b65 100644
> --- a/Documentation/virt/kvm/api.txt
> +++ b/Documentation/virt/kvm/api.txt
> @@ -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
> diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
> index 122d4ce3b1ab..e07c5ce3ac93 100644
> --- a/arch/x86/kvm/svm.c
> +++ b/arch/x86/kvm/svm.c
> @@ -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 = vcpu->cpu;
> 		dump_vmcb(vcpu);
> 		return 0;
> 	}
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index e3394c839dea..17d1a1676fc0 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -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 = vcpu->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 = vcpu->cpu;
> 		return 0;
> 	}
> 
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index cf917139de6b..9e89a32056d1 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -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:
> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> index f0a16b4adbbd..09ba7174456d 100644
> --- a/include/uapi/linux/kvm.h
> +++ b/include/uapi/linux/kvm.h
> @@ -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
>
Jim Mattson Dec. 16, 2019, 5:43 p.m. UTC | #2
On Fri, Dec 13, 2019 at 5:10 PM Liran Alon <liran.alon@oracle.com> wrote:
>
>
>
> > On 14 Dec 2019, at 2:20, Jim Mattson <jmattson@google.com> wrote:
> >
> > 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 <jmattson@google.com>
> > Reviewed-by: Oliver Upton <oupton@google.com>
> > Cc: Liran Alon <liran.alon@oracle.com>
> > Cc: Paolo Bonzini <pbonzini@redhat.com>
>
> Reviewed-by: Liran Alon <liran.alon@oracle.com>
>
> BTW, one could argue that receiving an unexpected exit-reason (i.e. KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON)
> could also only occur in production either from a KVM bug or from a defective CPU. Similar to failed VM-entry.
> Should we add similar behaviour to that as-well?
>
> -Liran

That's a good point. We had one case of numerous VM-exits for INIT,
and I'm pretty sure that was a defective CPU too.
Paolo Bonzini Dec. 16, 2019, 5:51 p.m. UTC | #3
On 16/12/19 18:43, Jim Mattson wrote:
> That's a good point. We had one case of numerous VM-exits for INIT,
> and I'm pretty sure that was a defective CPU too.

We too, and that's the only conclusion we could reach.  And the other
one that I can remember was KVM_INTERNAL_ERROR_DELIVERY_EV with
EPT_VIOLATION exits, so I would add it to all KVM_EXIT_INTERNAL_ERROR.

Nowadays KVM_EXIT_FAIL_ENTRY would probably also be an internal error,
however it was somewhat more frequent back before Intel CPUs had
unrestricted guest support.

Paolo

Patch
diff mbox series

diff --git a/Documentation/virt/kvm/api.txt b/Documentation/virt/kvm/api.txt
index ebb37b34dcfc..6e5d92406b65 100644
--- a/Documentation/virt/kvm/api.txt
+++ b/Documentation/virt/kvm/api.txt
@@ -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
diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c
index 122d4ce3b1ab..e07c5ce3ac93 100644
--- a/arch/x86/kvm/svm.c
+++ b/arch/x86/kvm/svm.c
@@ -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 = vcpu->cpu;
 		dump_vmcb(vcpu);
 		return 0;
 	}
diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index e3394c839dea..17d1a1676fc0 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -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 = vcpu->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 = vcpu->cpu;
 		return 0;
 	}
 
diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index cf917139de6b..9e89a32056d1 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -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:
diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
index f0a16b4adbbd..09ba7174456d 100644
--- a/include/uapi/linux/kvm.h
+++ b/include/uapi/linux/kvm.h
@@ -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