[PATCH v2] KVM: VMX: Remove redundant handling of bus lock vmexit

[PATCH v2] KVM: VMX: Remove redundant handling of bus lock vmexit

[Hao Xiang]

Hardware may or may not set exit_reason.bus_lock_detected on BUS_LOCK
VM-Exits. Dealing with KVM_RUN_X86_BUS_LOCK in handle_bus_lock_vmexit
could be redundant when exit_reason.basic is EXIT_REASON_BUS_LOCK.

We can remove redundant handling of bus lock vmexit. Force Setting
exit_reason.bus_lock_detected in handle_bus_lock_vmexit(), and deal with
KVM_RUN_X86_BUS_LOCK only in vmx_handle_exit().

Suggested-by: Sean Christopherson <seanjc@google.com>
Signed-off-by: Hao Xiang <hao.xiang@linux.alibaba.com>
---
v1 -> v2: a little modifications of comments

 arch/x86/kvm/vmx/vmx.c | 15 +++++++++------
 1 file changed, 9 insertions(+), 6 deletions(-)

diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index 116b089..22be02e 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -5562,9 +5562,13 @@ static int handle_encls(struct kvm_vcpu *vcpu)
 
 static int handle_bus_lock_vmexit(struct kvm_vcpu *vcpu)
 {
-	vcpu->run->exit_reason = KVM_EXIT_X86_BUS_LOCK;
-	vcpu->run->flags |= KVM_RUN_X86_BUS_LOCK;
-	return 0;
+	/*
+	 * Hardware may or may not set the BUS_LOCK_DETECTED flag on BUS_LOCK
+	 * VM-Exits, force setting the flag so that the logic in vmx_handle_exit()
+	 * doesn't have to handle the flag and the basic exit reason.
+	 */
+	to_vmx(vcpu)->exit_reason.bus_lock_detected = true;
+	return 1;
 }
 
 /*
@@ -6051,9 +6055,8 @@ static int vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
 	int ret = __vmx_handle_exit(vcpu, exit_fastpath);
 
 	/*
-	 * Even when current exit reason is handled by KVM internally, we
-	 * still need to exit to user space when bus lock detected to inform
-	 * that there is a bus lock in guest.
+	 * Exit to user space when bus lock detected to inform that there is
+	 * a bus lock in guest.
 	 */
 	if (to_vmx(vcpu)->exit_reason.bus_lock_detected) {
 		if (ret > 0)
-- 
1.8.3.1

Re: [PATCH v2] KVM: VMX: Remove redundant handling of bus lock vmexit

[Xiaoyao Li]

On 10/15/2021 11:57 AM, Hao Xiang wrote:
> Hardware may or may not set exit_reason.bus_lock_detected on BUS_LOCK
> VM-Exits. Dealing with KVM_RUN_X86_BUS_LOCK in handle_bus_lock_vmexit
> could be redundant when exit_reason.basic is EXIT_REASON_BUS_LOCK.
> 
> We can remove redundant handling of bus lock vmexit. Force Setting
> exit_reason.bus_lock_detected in handle_bus_lock_vmexit(), and deal with
> KVM_RUN_X86_BUS_LOCK only in vmx_handle_exit().
> 
> Suggested-by: Sean Christopherson <seanjc@google.com>
> Signed-off-by: Hao Xiang <hao.xiang@linux.alibaba.com>
> ---
> v1 -> v2: a little modifications of comments
> 
>   arch/x86/kvm/vmx/vmx.c | 15 +++++++++------
>   1 file changed, 9 insertions(+), 6 deletions(-)
> 
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index 116b089..22be02e 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -5562,9 +5562,13 @@ static int handle_encls(struct kvm_vcpu *vcpu)
>   
>   static int handle_bus_lock_vmexit(struct kvm_vcpu *vcpu)
>   {
> -	vcpu->run->exit_reason = KVM_EXIT_X86_BUS_LOCK;
> -	vcpu->run->flags |= KVM_RUN_X86_BUS_LOCK;
> -	return 0;
> +	/*
> +	 * Hardware may or may not set the BUS_LOCK_DETECTED flag on BUS_LOCK
> +	 * VM-Exits, force setting the flag 

>          so that the logic in vmx_handle_exit()
> +	 * doesn't have to handle the flag and the basic exit reason.

The code itself looks good.

But the comment is partly incorrect. Now we rely only on the 
bus_lock_detected flag so vmx_handle_exit() must check the flag.

> +	 */
> +	to_vmx(vcpu)->exit_reason.bus_lock_detected = true;
> +	return 1;
>   }
>   
>   /*
> @@ -6051,9 +6055,8 @@ static int vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
>   	int ret = __vmx_handle_exit(vcpu, exit_fastpath);
>   
>   	/*
> -	 * Even when current exit reason is handled by KVM internally, we
> -	 * still need to exit to user space when bus lock detected to inform
> -	 * that there is a bus lock in guest.
> +	 * Exit to user space when bus lock detected to inform that there is
> +	 * a bus lock in guest.
>   	 */
>   	if (to_vmx(vcpu)->exit_reason.bus_lock_detected) {
>   		if (ret > 0)
> 

Re: [PATCH v2] KVM: VMX: Remove redundant handling of bus lock vmexit

[Hao Xiang]

On 2021/10/15 14:46, Xiaoyao Li wrote:
> On 10/15/2021 11:57 AM, Hao Xiang wrote:
>> Hardware may or may not set exit_reason.bus_lock_detected on BUS_LOCK
>> VM-Exits. Dealing with KVM_RUN_X86_BUS_LOCK in handle_bus_lock_vmexit
>> could be redundant when exit_reason.basic is EXIT_REASON_BUS_LOCK.
>>
>> We can remove redundant handling of bus lock vmexit. Force Setting
>> exit_reason.bus_lock_detected in handle_bus_lock_vmexit(), and deal with
>> KVM_RUN_X86_BUS_LOCK only in vmx_handle_exit().
>>
>> Suggested-by: Sean Christopherson <seanjc@google.com>
>> Signed-off-by: Hao Xiang <hao.xiang@linux.alibaba.com>
>> ---
>> v1 -> v2: a little modifications of comments
>>
>>   arch/x86/kvm/vmx/vmx.c | 15 +++++++++------
>>   1 file changed, 9 insertions(+), 6 deletions(-)
>>
>> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
>> index 116b089..22be02e 100644
>> --- a/arch/x86/kvm/vmx/vmx.c
>> +++ b/arch/x86/kvm/vmx/vmx.c
>> @@ -5562,9 +5562,13 @@ static int handle_encls(struct kvm_vcpu *vcpu)
>>   static int handle_bus_lock_vmexit(struct kvm_vcpu *vcpu)
>>   {
>> -    vcpu->run->exit_reason = KVM_EXIT_X86_BUS_LOCK;
>> -    vcpu->run->flags |= KVM_RUN_X86_BUS_LOCK;
>> -    return 0;
>> +    /*
>> +     * Hardware may or may not set the BUS_LOCK_DETECTED flag on 
>> BUS_LOCK
>> +     * VM-Exits, force setting the flag 
> 
>>          so that the logic in vmx_handle_exit()
>> +     * doesn't have to handle the flag and the basic exit reason.
> 
> The code itself looks good.
> 
> But the comment is partly incorrect. Now we rely only on the 
> bus_lock_detected flag so vmx_handle_exit() must check the flag.
>Thank you for you advice. Is the following comment ok?
/*
  * Hardware may or may not set the BUS_LOCK_DETECTED flag on BUS_LOCK
  * VM-Exits. Forcely set the flag here, then check the flag in
  * vmx_handle_exit() and handle the bus lock vmexit if the flag is
  * set.
  */

>> +     */
>> +    to_vmx(vcpu)->exit_reason.bus_lock_detected = true;
>> +    return 1;
>>   }
>>   /*
>> @@ -6051,9 +6055,8 @@ static int vmx_handle_exit(struct kvm_vcpu 
>> *vcpu, fastpath_t exit_fastpath)
>>       int ret = __vmx_handle_exit(vcpu, exit_fastpath);
>>       /*
>> -     * Even when current exit reason is handled by KVM internally, we
>> -     * still need to exit to user space when bus lock detected to inform
>> -     * that there is a bus lock in guest.
>> +     * Exit to user space when bus lock detected to inform that there is
>> +     * a bus lock in guest.
>>        */
>>       if (to_vmx(vcpu)->exit_reason.bus_lock_detected) {
>>           if (ret > 0)
>>

Re: [PATCH v2] KVM: VMX: Remove redundant handling of bus lock vmexit

[Xiaoyao Li]

On 10/15/2021 3:28 PM, Hao Xiang wrote:
> 
> 
> On 2021/10/15 14:46, Xiaoyao Li wrote:
>> On 10/15/2021 11:57 AM, Hao Xiang wrote:
>>> Hardware may or may not set exit_reason.bus_lock_detected on BUS_LOCK
>>> VM-Exits. Dealing with KVM_RUN_X86_BUS_LOCK in handle_bus_lock_vmexit
>>> could be redundant when exit_reason.basic is EXIT_REASON_BUS_LOCK.
>>>
>>> We can remove redundant handling of bus lock vmexit. Force Setting
>>> exit_reason.bus_lock_detected in handle_bus_lock_vmexit(), and deal with
>>> KVM_RUN_X86_BUS_LOCK only in vmx_handle_exit().
>>>
>>> Suggested-by: Sean Christopherson <seanjc@google.com>
>>> Signed-off-by: Hao Xiang <hao.xiang@linux.alibaba.com>
>>> ---
>>> v1 -> v2: a little modifications of comments
>>>
>>>   arch/x86/kvm/vmx/vmx.c | 15 +++++++++------
>>>   1 file changed, 9 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
>>> index 116b089..22be02e 100644
>>> --- a/arch/x86/kvm/vmx/vmx.c
>>> +++ b/arch/x86/kvm/vmx/vmx.c
>>> @@ -5562,9 +5562,13 @@ static int handle_encls(struct kvm_vcpu *vcpu)
>>>   static int handle_bus_lock_vmexit(struct kvm_vcpu *vcpu)
>>>   {
>>> -    vcpu->run->exit_reason = KVM_EXIT_X86_BUS_LOCK;
>>> -    vcpu->run->flags |= KVM_RUN_X86_BUS_LOCK;
>>> -    return 0;
>>> +    /*
>>> +     * Hardware may or may not set the BUS_LOCK_DETECTED flag on 
>>> BUS_LOCK
>>> +     * VM-Exits, force setting the flag 
>>
>>>          so that the logic in vmx_handle_exit()
>>> +     * doesn't have to handle the flag and the basic exit reason.
>>
>> The code itself looks good.
>>
>> But the comment is partly incorrect. Now we rely only on the 
>> bus_lock_detected flag so vmx_handle_exit() must check the flag.
>> Thank you for you advice. Is the following comment ok?
> /*
>   * Hardware may or may not set the BUS_LOCK_DETECTED flag on BUS_LOCK
>   * VM-Exits. Forcely set the flag here, then check the flag in
>   * vmx_handle_exit() and handle the bus lock vmexit if the flag is
>   * set.
>   */

(I'm not sure if forcely is a valid word)

how about

/*
  * Hardware may or may not set the BUS_LOCK_DETECTED flag on BUS_LOCK
  * VM Exit. Unconditionally set the flag here and leave the handling to
  * vmx_handle_exit().
  */

>>> +     */
>>> +    to_vmx(vcpu)->exit_reason.bus_lock_detected = true;
>>> +    return 1;
>>>   }
>>>   /*
>>> @@ -6051,9 +6055,8 @@ static int vmx_handle_exit(struct kvm_vcpu 
>>> *vcpu, fastpath_t exit_fastpath)
>>>       int ret = __vmx_handle_exit(vcpu, exit_fastpath);
>>>       /*
>>> -     * Even when current exit reason is handled by KVM internally, we
>>> -     * still need to exit to user space when bus lock detected to 
>>> inform
>>> -     * that there is a bus lock in guest.
>>> +     * Exit to user space when bus lock detected to inform that 
>>> there is
>>> +     * a bus lock in guest.
>>>        */
>>>       if (to_vmx(vcpu)->exit_reason.bus_lock_detected) {
>>>           if (ret > 0)
>>>