Skip to contentSkip to conversation

KVM: nVMX: Consult only the "basic" exit reason when routing nested exit

PATCH Superseded Sean Christopherson 4 participants 5 messages Feb 27, 2020 – Feb 28, 2020

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 KVM: nVMX: Consult only the "basic" exit reason when routing nested exit

Consult only the basic exit reason, i.e. bits 15:0 of vmcs.EXIT_REASON, when determining whether a nested VM-Exit should be reflected into L1 or handled by KVM in L0.

For better or worse, the switch statement in nested_vmx_exit_reflected() currently defaults to "true", i.e. reflects any nested VM-Exit without dedicated logic. Because the case statements only contain the basic exit reason, any VM-Exit with modifier bits set will be reflected to L1, even if KVM intended to handle it in L0.

Practically speaking, this only affects EXIT_REASON_MCE_DURING_VMENTRY, i.e. a #MC that occurs on nested VM-Enter would be incorrectly routed to L1, as "failed VM-Entry" is the only modifier that KVM can currently encounter. The SMM modifiers will never be generated as KVM doesn't support/employ a SMI Transfer Monitor. Ditto for "exit from enclave", as KVM doesn't yet support virtualizing SGX, i.e. it's impossible to enter an enclave in a KVM guest (L1 or L2).

Fixes 644d711aa0e1 ("KVM: nVMX: Deciding if L0 or L1 should handle an L2 exit")
Cc Jim Mattson <[email protected]>
Cc Xiaoyao Li <[email protected]>
---
 arch/x86/kvm/vmx/nested.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/vmx/nested.c +1 -1
index 0946122a8d3b..127065bbde2c 100644
@@ -5554,7 +5554,7 @@ bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
 				vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
 				KVM_ISA_VMX);
 
-	switch (exit_reason) {
+	switch ((u16)exit_reason) {
 	case EXIT_REASON_EXCEPTION_NMI:
 		if (is_nmi(intr_info))
 			return false;
Signature · 1 line
2.24.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 42 lines · Sean Christopherson

Consult only the basic exit reason, i.e. bits 15:0 of vmcs.EXIT_REASON, when determining whether a nested VM-Exit should be reflected into L1 or handled by KVM in L0.

For better or worse, the switch statement in nested_vmx_exit_reflected() currently defaults to "true", i.e. reflects any nested VM-Exit without dedicated logic. Because the case statements only contain the basic exit reason, any VM-Exit with modifier bits set will be reflected to L1, even if KVM intended to handle it in L0.

Practically speaking, this only affects EXIT_REASON_MCE_DURING_VMENTRY, i.e. a #MC that occurs on nested VM-Enter would be incorrectly routed to L1, as "failed VM-Entry" is the only modifier that KVM can currently encounter. The SMM modifiers will never be generated as KVM doesn't support/employ a SMI Transfer Monitor. Ditto for "exit from enclave", as KVM doesn't yet support virtualizing SGX, i.e. it's impossible to enter an enclave in a KVM guest (L1 or L2).

Fixes 644d711aa0e1 ("KVM: nVMX: Deciding if L0 or L1 should handle an L2 exit")
Cc Jim Mattson <[email protected]>
Cc Xiaoyao Li <[email protected]>
---
 arch/x86/kvm/vmx/nested.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/vmx/nested.c +1 -2
index 0946122a8d3b..127065bbde2c 100644
@@ -5554,7 +5554,7 @@ bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
                                vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
                                KVM_ISA_VMX);
 
-       switch (exit_reason) {
+       switch ((u16)exit_reason) {
        case EXIT_REASON_EXCEPTION_NMI:
                if (is_nmi(intr_info))
                        return false;
--
Quoted 2 lines

2.24.1

Reviewed-by on v1 1/1
#2 lore raw
In reply to Sean Christopherson
Quoted 10 lines · Sean Christopherson

Consult only the basic exit reason, i.e. bits 15:0 of vmcs.EXIT_REASON, when determining whether a nested VM-Exit should be reflected into L1 or handled by KVM in L0.

For better or worse, the switch statement in nested_vmx_exit_reflected() currently defaults to "true", i.e. reflects any nested VM-Exit without dedicated logic. Because the case statements only contain the basic exit reason, any VM-Exit with modifier bits set will be reflected to L1, even if KVM intended to handle it in L0.

Quoted 7 lines

Practically speaking, this only affects EXIT_REASON_MCE_DURING_VMENTRY, i.e. a #MC that occurs on nested VM-Enter would be incorrectly routed to L1, as "failed VM-Entry" is the only modifier that KVM can currently encounter. The SMM modifiers will never be generated as KVM doesn't support/employ a SMI Transfer Monitor. Ditto for "exit from enclave", as KVM doesn't yet support virtualizing SGX, i.e. it's impossible to enter an enclave in a KVM guest (L1 or L2).

It seems nested_vmx_exit_reflected() deals only with the basic exit reason. If it doesn't need anything beyond bits 15:0, may be vmx_handle_exit() can pass just the base exit reason ?

Quoted 15 lines
Fixes 644d711aa0e1 ("KVM: nVMX: Deciding if L0 or L1 should handle an L2 exit")
Cc Jim Mattson <[email protected]>
Cc Xiaoyao Li <[email protected]>
---
  arch/x86/kvm/vmx/nested.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
arch/x86/kvm/vmx/nested.c +0 -0
index 0946122a8d3b..127065bbde2c 100644

arch/x86/kvm/vmx/nested.c bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason) · line 5560

Quoted 8 lines
arch/x86/kvm/vmx/nested.c +1 -1
@@ -5554,7 +5554,7 @@ bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
  				vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
  				KVM_ISA_VMX);
 
-	switch (exit_reason) {
+	switch ((u16)exit_reason) {
  	case EXIT_REASON_EXCEPTION_NMI:
  		if (is_nmi(intr_info))
  			return false;
#3 lore raw
Quoted 21 lines · Krish Sadhukhan
Quoted 10 lines · Sean Christopherson

Consult only the basic exit reason, i.e. bits 15:0 of vmcs.EXIT_REASON, when determining whether a nested VM-Exit should be reflected into L1 or handled by KVM in L0.

For better or worse, the switch statement in nested_vmx_exit_reflected() currently defaults to "true", i.e. reflects any nested VM-Exit without dedicated logic. Because the case statements only contain the basic exit reason, any VM-Exit with modifier bits set will be reflected to L1, even if KVM intended to handle it in L0.

Quoted 7 lines

Practically speaking, this only affects EXIT_REASON_MCE_DURING_VMENTRY, i.e. a #MC that occurs on nested VM-Enter would be incorrectly routed to L1, as "failed VM-Entry" is the only modifier that KVM can currently encounter. The SMM modifiers will never be generated as KVM doesn't support/employ a SMI Transfer Monitor. Ditto for "exit from enclave", as KVM doesn't yet support virtualizing SGX, i.e. it's impossible to enter an enclave in a KVM guest (L1 or L2).

Quoted 3 lines

It seems nested_vmx_exit_reflected() deals only with the basic exit reason. If it doesn't need anything beyond bits 15:0, may be vmx_handle_exit() can pass just the base exit reason ?

Argh. I was going to simply respond with "It traces exit_reason via trace_kvm_nested_vmexit().", but then I looked at the tracing code :-(

The tracepoints that print the names of the VM-Exit are flawed in the sense that they'll always print the raw value for VM-Exits with modifiers. E.g. a consistency check VM-Exit on invalid guest state will print 0x80000021 instead of INVALID_STATE.

Stripping bits 31:16 when invoking the tracepoint would fix the immediate issue, but I'm not sure I like that approach because doing so drops information that could potentially be quite helpful, e.g. if nested VM-Exit injection injected EXIT_REASON_MSR_LOAD_FAIL without also setting VMX_EXIT_REASONS_FAILED_VMENTRY, which could break/confuse the L1 VMM. I'm also not remotely confident that we won't screw this up again in the future :-)

So part of me thinks the best way to resolve the printing would be to modify VMX_EXIT_REASONS to do "| VMX_EXIT_REASONS_FAILED_VMENTRY" where appropriate, i.e. on INVALID_STATE, MSR_LOAD_FAIL and MCE_DURING_VMENTRY. The downside of that approach is it breaks again when new modifiers come along, e.g. SGX's ENCLAVE_EXIT. But again, the modifier is likely useful information.

I think the most foolproof and informative way to handle this would be to add a macro and/or helper function, e.g. kvm_print_vmx_exit_reason(), to wrap __print_symbolic(__entry->exit_code, VMX_EXIT_REASONS) so that it prints both the name of the basic exit reason as well as the names for any modifiers.

TL;DR: I still like this patch as is, especially since it'll be easy to backport. I'll send a separate patch for the tracepoint issue.

maintainer
#4 lore raw
Quoted 7 lines · Sean Christopherson

So part of me thinks the best way to resolve the printing would be to modify VMX_EXIT_REASONS to do "| VMX_EXIT_REASONS_FAILED_VMENTRY" where appropriate, i.e. on INVALID_STATE, MSR_LOAD_FAIL and MCE_DURING_VMENTRY. The downside of that approach is it breaks again when new modifiers come along, e.g. SGX's ENCLAVE_EXIT. But again, the modifier is likely useful information.

Quoted 8 lines

I think the most foolproof and informative way to handle this would be to add a macro and/or helper function, e.g. kvm_print_vmx_exit_reason(), to wrap __print_symbolic(__entry->exit_code, VMX_EXIT_REASONS) so that it prints both the name of the basic exit reason as well as the names for any modifiers.

TL;DR: I still like this patch as is, especially since it'll be easy to backport. I'll send a separate patch for the tracepoint issue.

Agreed, thanks. I queued this one.

Paolo