[PATCH] KVM: x86: Fix handling of pending signal on uninitialized AP

[PATCH] KVM: x86: Fix handling of pending signal on uninitialized AP

[Jan H. Schönherr]

KVM API says that KVM_RUN will return with -EINTR when a signal is
pending. However, if a vCPU is in KVM_MP_STATE_UNINITIALIZED, then
the return value is unconditionally -EAGAIN.

Copy over some code from vcpu_run(), so that the case of a pending
signal results in the expected return value.

Signed-off-by: Jan H. Schönherr <jschoenh@amazon.de>
---
 arch/x86/kvm/x86.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index 272320e..40039cd 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -7203,6 +7203,11 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 		kvm_apic_accept_events(vcpu);
 		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
 		r = -EAGAIN;
+		if (signal_pending(current)) {
+			r = -EINTR;
+			vcpu->run->exit_reason = KVM_EXIT_INTR;
+			++vcpu->stat.signal_exits;
+		}
 		goto out;
 	}
 
-- 
2.3.1.dirty

Re: [PATCH] KVM: x86: Fix handling of pending signal on uninitialized AP

[Jan H. Schönherr]

On 09/06/2017 02:33 PM, David Hildenbrand wrote:
> On 06.09.2017 00:27, Jan H. Schönherr wrote:
>> KVM API says that KVM_RUN will return with -EINTR when a signal is
>> pending. However, if a vCPU is in KVM_MP_STATE_UNINITIALIZED, then
>> the return value is unconditionally -EAGAIN.
>>
>> Copy over some code from vcpu_run(), so that the case of a pending
>> signal results in the expected return value.
>>
>> Signed-off-by: Jan H. Schönherr <jschoenh@amazon.de>
>> ---
>>   arch/x86/kvm/x86.c | 5 +++++
>>   1 file changed, 5 insertions(+)
>>
>> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
>> index 272320e..40039cd 100644
>> --- a/arch/x86/kvm/x86.c
>> +++ b/arch/x86/kvm/x86.c
>> @@ -7203,6 +7203,11 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
>>   		kvm_apic_accept_events(vcpu);
>>   		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
>>   		r = -EAGAIN;
>> +		if (signal_pending(current)) {
>> +			r = -EINTR;
>> +			vcpu->run->exit_reason = KVM_EXIT_INTR;
>> +			++vcpu->stat.signal_exits;
>> +		}
>>   		goto out;
>>   	}
>>   
>>
> 
> I am not sure if this is the right thing to do. E.g. also on s390x a
> -EINVAL is indicated if the VCPU is stopped.
> 
> If the documentation is unclear, maybe that one should be fixed. I don't
> see this to be relevant in practice, or is it?
> 
> -EINTR will only be returned if nothing else hinders the VCPU from running.
> 

In practice, in allows me to distinguish, whether I can reenter KVM_RUN immediately
(-EAGAIN), or whether I have to check for signals first (-EINTR), or whether I need
to have a look at the actual exit reason (0).

Right now, KVM_RUN returns *immediately* with -EAGAIN when there is a pending signal
for a vCPU thread that's in KVM_MP_STATE_UNINITIALIZED. And it does so again and
again until you consume the signal in user space.

Regards
Jan

PS: I just noticed, that the kvm_run->immediate_exit handling also does not work in this
case. Instead, the vCPU thread goes to sleep in the KVM_MP_STATE_UNINITIALIZED case.
(Let me prepare a patch for that...) So, I don't agree, that this is a case, where only
the documentation needs a fix.

Re: [PATCH] KVM: x86: Fix handling of pending signal on uninitialized AP

[Radim Krčmář]

2017-09-06 17:33+0200, Jan H. Schönherr:
> On 09/06/2017 02:33 PM, David Hildenbrand wrote:
> > On 06.09.2017 00:27, Jan H. Schönherr wrote:
> > > KVM API says that KVM_RUN will return with -EINTR when a signal is
> > > pending. However, if a vCPU is in KVM_MP_STATE_UNINITIALIZED, then
> > > the return value is unconditionally -EAGAIN.
> > > 
> > > Copy over some code from vcpu_run(), so that the case of a pending
> > > signal results in the expected return value.
> > > 
> > > Signed-off-by: Jan H. Schönherr <jschoenh@amazon.de>
> > > ---
> > >   arch/x86/kvm/x86.c | 5 +++++
> > >   1 file changed, 5 insertions(+)
> > > 
> > > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > > index 272320e..40039cd 100644
> > > --- a/arch/x86/kvm/x86.c
> > > +++ b/arch/x86/kvm/x86.c
> > > @@ -7203,6 +7203,11 @@ int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
> > >   		kvm_apic_accept_events(vcpu);
> > >   		kvm_clear_request(KVM_REQ_UNHALT, vcpu);
> > >   		r = -EAGAIN;
> > > +		if (signal_pending(current)) {
> > > +			r = -EINTR;
> > > +			vcpu->run->exit_reason = KVM_EXIT_INTR;
> > > +			++vcpu->stat.signal_exits;
> > > +		}
> > >   		goto out;
> > >   	}
> > > 
> > 
> > I am not sure if this is the right thing to do. E.g. also on s390x a
> > -EINVAL is indicated if the VCPU is stopped.
> > 
> > If the documentation is unclear, maybe that one should be fixed. I don't
> > see this to be relevant in practice, or is it?
> > 
> > -EINTR will only be returned if nothing else hinders the VCPU from running.
> > 
> 
> In practice, in allows me to distinguish, whether I can reenter KVM_RUN immediately
> (-EAGAIN), or whether I have to check for signals first (-EINTR), or whether I need
> to have a look at the actual exit reason (0).

Makes sense and the worst thing I found was ugliness of the code ...
Queued, thanks.