Mailing List Archive

On 18/09/21 13:30, Hao Xiang wrote:
> exit_reason.bus_lock_detected is not only set when bus lock VM exit
> was preempted, in fact, this bit is always set if bus locks are
> detected no matter what the exit_reason.basic is.
>
> So the bus_lock_vmexit handling in vmx_handle_exit should be duplicated
> when exit_reason.basic is EXIT_REASON_BUS_LOCK(74). We can avoid it by
> checking if bus lock vmexit was preempted in vmx_handle_exit.

I don't understand, does this mean that bus_lock_detected=1 if
basic=EXIT_REASON_BUS_LOCK? If so, can we instead replace the contents
of handle_bus_lock_vmexit with

/* Do nothing and let vmx_handle_exit exit to userspace. */
WARN_ON(!to_vmx(vcpu)->exit_reason.bus_lock_detected);
return 0;

?

That would be doable only if this is architectural behavior and not a
processor erratum, of course.

Thanks,

Paolo

> Signed-off-by: Hao Xiang <hao.xiang@linux.alibaba.com>
> ---
> arch/x86/kvm/vmx/vmx.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index 0c2c0d5..5ddf1df 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -6054,7 +6054,8 @@ static int vmx_handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
> * still need to 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 (to_vmx(vcpu)->exit_reason.bus_lock_detected &&
> + to_vmx(vcpu)->exit_reason.basic != EXIT_REASON_BUS_LOCK) {
> if (ret > 0)
> vcpu->run->exit_reason = KVM_EXIT_X86_BUS_LOCK;
>
>

On 9/22/2021 6:02 PM, Paolo Bonzini wrote:
> On 18/09/21 13:30, Hao Xiang wrote:
>> exit_reason.bus_lock_detected is not only set when bus lock VM exit
>> was preempted, in fact, this bit is always set if bus locks are
>> detected no matter what the exit_reason.basic is.
>>
>> So the bus_lock_vmexit handling in vmx_handle_exit should be duplicated
>> when exit_reason.basic is EXIT_REASON_BUS_LOCK(74). We can avoid it by
>> checking if bus lock vmexit was preempted in vmx_handle_exit.
>
> I don't understand, does this mean that bus_lock_detected=1 if
> basic=EXIT_REASON_BUS_LOCK?  If so, can we instead replace the contents
> of handle_bus_lock_vmexit with
>
>     /* Do nothing and let vmx_handle_exit exit to userspace.  */
>     WARN_ON(!to_vmx(vcpu)->exit_reason.bus_lock_detected);
>     return 0;
>
> ?
>
> That would be doable only if this is architectural behavior and not a
> processor erratum, of course.

EXIT_REASON.bus_lock_detected may or may not be set when exit reason ==
EXIT_REASON_BUS_LOCK. Intel will update ISE or SDM to state it.

Maybe we can do below in handle_bus_lock_vmexit handler:

if (!to_vmx(vcpu)->exit_reason.bus_lock_detected)
to_vmx(vcpu)->exit_reason.bus_lock_detected = 1;

But is manually changing the hardware reported value for software
purpose a good thing?


> Thanks,
>
> Paolo
>
>> Signed-off-by: Hao Xiang <hao.xiang@linux.alibaba.com>
>> ---
>>   arch/x86/kvm/vmx/vmx.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
>> index 0c2c0d5..5ddf1df 100644
>> --- a/arch/x86/kvm/vmx/vmx.c
>> +++ b/arch/x86/kvm/vmx/vmx.c
>> @@ -6054,7 +6054,8 @@ static int vmx_handle_exit(struct kvm_vcpu
>> *vcpu, fastpath_t exit_fastpath)
>>        * still need to 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 (to_vmx(vcpu)->exit_reason.bus_lock_detected &&
>> +            to_vmx(vcpu)->exit_reason.basic != EXIT_REASON_BUS_LOCK) {
>>           if (ret > 0)
>>               vcpu->run->exit_reason = KVM_EXIT_X86_BUS_LOCK;
>>
>

On 22/09/21 12:32, Xiaoyao Li wrote:
>>
>
> EXIT_REASON.bus_lock_detected may or may not be set when exit reason ==
> EXIT_REASON_BUS_LOCK. Intel will update ISE or SDM to state it.
>
> Maybe we can do below in handle_bus_lock_vmexit handler:
>
>     if (!to_vmx(vcpu)->exit_reason.bus_lock_detected)
>         to_vmx(vcpu)->exit_reason.bus_lock_detected = 1;
>
> But is manually changing the hardware reported value for software
> purpose a good thing?

No. That said, Hao's patch is just making the code clearer; there's no
behavioral change since the "if" will just redo the same assignments as
handle_bus_lock_vmexit.

Paolo

On Wed, Sep 22, 2021, Xiaoyao Li wrote:
> On 9/22/2021 6:02 PM, Paolo Bonzini wrote:
> > On 18/09/21 13:30, Hao Xiang wrote:
> > > exit_reason.bus_lock_detected is not only set when bus lock VM exit
> > > was preempted, in fact, this bit is always set if bus locks are
> > > detected no matter what the exit_reason.basic is.
> > >
> > > So the bus_lock_vmexit handling in vmx_handle_exit should be duplicated
> > > when exit_reason.basic is EXIT_REASON_BUS_LOCK(74). We can avoid it by
> > > checking if bus lock vmexit was preempted in vmx_handle_exit.
> >
> > I don't understand, does this mean that bus_lock_detected=1 if
> > basic=EXIT_REASON_BUS_LOCK?? If so, can we instead replace the contents
> > of handle_bus_lock_vmexit with
> >
> > ????/* Do nothing and let vmx_handle_exit exit to userspace.? */
> > ????WARN_ON(!to_vmx(vcpu)->exit_reason.bus_lock_detected);
> > ????return 0;
> >
> > ?
> >
> > That would be doable only if this is architectural behavior and not a
> > processor erratum, of course.
>
> EXIT_REASON.bus_lock_detected may or may not be set when exit reason ==
> EXIT_REASON_BUS_LOCK. Intel will update ISE or SDM to state it.
>
> Maybe we can do below in handle_bus_lock_vmexit handler:
>
> if (!to_vmx(vcpu)->exit_reason.bus_lock_detected)
> to_vmx(vcpu)->exit_reason.bus_lock_detected = 1;
>
> But is manually changing the hardware reported value for software purpose a
> good thing?

In this case, I'd say yes. Hardware having non-deterministic behavior is the not
good thing, KVM would simply be correctly the not-technically-an-erratum erratum.

Set it unconditionally and then handle everything in common path. This has the
added advantage of having only one site that deals with KVM_RUN_X86_BUS_LOCK.

diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
index 33f92febe3ce..aa9372452e49 100644
--- a/arch/x86/kvm/vmx/vmx.c
+++ b/arch/x86/kvm/vmx/vmx.c
@@ -5561,9 +5561,9 @@ 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;
+ /* The dedicated flag may or may not be set by hardware. /facepalm. */
+ vcpu->exit_reason.bus_lock_detected = true;
+ return 1;
}

/*
@@ -6050,9 +6050,8 @@ 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)

On 2021/9/22 20:40, Paolo Bonzini wrote:
> On 22/09/21 12:32, Xiaoyao Li wrote:
>>>
>>
>> EXIT_REASON.bus_lock_detected may or may not be set when exit reason
>> == EXIT_REASON_BUS_LOCK. Intel will update ISE or SDM to state it.
>>
>> Maybe we can do below in handle_bus_lock_vmexit handler:
>>
>>      if (!to_vmx(vcpu)->exit_reason.bus_lock_detected)
>>          to_vmx(vcpu)->exit_reason.bus_lock_detected = 1;
>>
>> But is manually changing the hardware reported value for software
>> purpose a good thing?
>
> No.  That said, Hao's patch is just making the code clearer; there's
> no behavioral change since the "if" will just redo the same
> assignments as handle_bus_lock_vmexit.
>
> Paolo
I agree Paolo. EXIT_REASON.bus_lock_detected may or may not be set when
exit_reason=EXIT_REASON_BUS_LOCK,  It clould depend on hardware
implementaion. No matter when intel states it clearly, I think it is
better that we avoid repeated assignment by adding additional check
condition in vmx_handle_exit.  Of course , it is also ok that
hand_bus_lock_vmexit do nothing , but the code is not clear, and the
code logic will be inconsistent with spec description.

On 23/09/21 02:59, Hao Xiang wrote:
> EXIT_REASON.bus_lock_detected may or may not be set when
> exit_reason=EXIT_REASON_BUS_LOCK,  It clould depend on hardware
> implementaion. No matter when intel states it clearly, I think it is
> better that we avoid repeated assignment by adding additional check
> condition in vmx_handle_exit.  Of course , it is also ok that
> hand_bus_lock_vmexit do nothing , but the code is not clear, and the
> code logic will be inconsistent with spec description.

For 5.16 we'll go with something like Sean's sketch, that sets the
bus_lock_detected bit on EXIT_REASON_BUS_LOCK.

Paolo

On 2021/9/22 22:58, Sean Christopherson wrote:
> On Wed, Sep 22, 2021, Xiaoyao Li wrote:
>> On 9/22/2021 6:02 PM, Paolo Bonzini wrote:
>>> On 18/09/21 13:30, Hao Xiang wrote:
>>>> exit_reason.bus_lock_detected is not only set when bus lock VM exit
>>>> was preempted, in fact, this bit is always set if bus locks are
>>>> detected no matter what the exit_reason.basic is.
>>>>
>>>> So the bus_lock_vmexit handling in vmx_handle_exit should be duplicated
>>>> when exit_reason.basic is EXIT_REASON_BUS_LOCK(74). We can avoid it by
>>>> checking if bus lock vmexit was preempted in vmx_handle_exit.
>>> I don't understand, does this mean that bus_lock_detected=1 if
>>> basic=EXIT_REASON_BUS_LOCK?  If so, can we instead replace the contents
>>> of handle_bus_lock_vmexit with
>>>
>>>     /* Do nothing and let vmx_handle_exit exit to userspace.  */
>>>     WARN_ON(!to_vmx(vcpu)->exit_reason.bus_lock_detected);
>>>     return 0;
>>>
>>> ?
>>>
>>> That would be doable only if this is architectural behavior and not a
>>> processor erratum, of course.
>> EXIT_REASON.bus_lock_detected may or may not be set when exit reason ==
>> EXIT_REASON_BUS_LOCK. Intel will update ISE or SDM to state it.
>>
>> Maybe we can do below in handle_bus_lock_vmexit handler:
>>
>> if (!to_vmx(vcpu)->exit_reason.bus_lock_detected)
>> to_vmx(vcpu)->exit_reason.bus_lock_detected = 1;
>>
>> But is manually changing the hardware reported value for software purpose a
>> good thing?
> In this case, I'd say yes. Hardware having non-deterministic behavior is the not
> good thing, KVM would simply be correctly the not-technically-an-erratum erratum.
>
> Set it unconditionally and then handle everything in common path. This has the
> added advantage of having only one site that deals with KVM_RUN_X86_BUS_LOCK.
>
> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
> index 33f92febe3ce..aa9372452e49 100644
> --- a/arch/x86/kvm/vmx/vmx.c
> +++ b/arch/x86/kvm/vmx/vmx.c
> @@ -5561,9 +5561,9 @@ 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;
> + /* The dedicated flag may or may not be set by hardware. /facepalm. */
> + vcpu->exit_reason.bus_lock_detected = true;
> + return 1;
> }
>
> /*
> @@ -6050,9 +6050,8 @@ 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)
I agree with your modifications. And I will  re-submit the patch. Thanks.