Skip to contentSkip to conversation

KVM: PPC: booke: Fix int_pending calculation for MSR[EE]

PATCH In review Scott Wood 2 participants 6 messages Aug 26, 2011 – Sep 2, 2011

Subsystems (2)
lore
Share & tools
lore

Scott Wood

author
#0 lore raw

PATCH 1/5 KVM: PPC: booke: Fix int_pending calculation for MSR[EE]

int_pending was only being lowered if a bit in pending_exceptions was cleared during exception delivery -- but for interrupts, we clear it during IACK/TSR emulation. This caused paravirt for enabling MSR[EE] to be ineffective.

---
 arch/powerpc/kvm/booke.c |    6 +-----
 1 files changed, 1 insertions(+), 5 deletions(-)
arch/powerpc/kvm/booke.c +1 -5
index d967faf..aeb69b2 100644
@@ -292,7 +292,6 @@ static int kvmppc_booke_irqprio_deliver(struct kvm_vcpu *vcpu,
 void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 {
 	unsigned long *pending = &vcpu->arch.pending_exceptions;
-	unsigned long old_pending = vcpu->arch.pending_exceptions;
 	unsigned int priority;
 
 	priority = __ffs(*pending);
@@ -306,10 +305,7 @@ void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
 	}
 
 	/* Tell the guest about our interrupt status */
-	if (*pending)
-		vcpu->arch.shared->int_pending = 1;
-	else if (old_pending)
-		vcpu->arch.shared->int_pending = 0;
+	vcpu->arch.shared->int_pending = !!*pending;
 }
 
 int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
Signature · 1 line
1.7.6
#1 lore raw

Re: [PATCH 1/5] KVM: PPC: booke: Fix int_pending calculation for

Quoted 4 lines · Scott Wood

int_pending was only being lowered if a bit in pending_exceptions was cleared during exception delivery -- but for interrupts, we clear it during IACK/TSR emulation. This caused paravirt for enabling MSR[EE] to be ineffective.

But that means that int_pending can still be 1 even though there is none pending as we don't get the call to deliver_interrupts when it gets lowered. Please create a common function to remove a bit from pending_exceptions and do the check there.

Alex

Quoted 29 lines
---
  arch/powerpc/kvm/booke.c |    6 +-----
  1 files changed, 1 insertions(+), 5 deletions(-)
arch/powerpc/kvm/booke.c +1 -5
index d967faf..aeb69b2 100644
@@ -292,7 +292,6 @@ static int kvmppc_booke_irqprio_deliver(struct kvm_vcpu *vcpu,
  void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
  {
  	unsigned long *pending =&vcpu->arch.pending_exceptions;
-	unsigned long old_pending = vcpu->arch.pending_exceptions;
  	unsigned int priority;
 
  	priority = __ffs(*pending);
@@ -306,10 +305,7 @@ void kvmppc_core_deliver_interrupts(struct kvm_vcpu *vcpu)
  	}
 
  	/* Tell the guest about our interrupt status */
-	if (*pending)
-		vcpu->arch.shared->int_pending = 1;
-	else if (old_pending)
-		vcpu->arch.shared->int_pending = 0;
+	vcpu->arch.shared->int_pending = !!*pending;
  }
 
  int kvmppc_vcpu_run(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
author
#2 lore raw

Re: [PATCH 1/5] KVM: PPC: booke: Fix int_pending calculation for

In reply to Scott Wood
Quoted 10 lines · Alexander Graf
Quoted 4 lines · Scott Wood

int_pending was only being lowered if a bit in pending_exceptions was cleared during exception delivery -- but for interrupts, we clear it during IACK/TSR emulation. This caused paravirt for enabling MSR[EE] to be ineffective.

But that means that int_pending can still be 1 even though there is none pending as we don't get the call to deliver_interrupts when it gets lowered. Please create a common function to remove a bit from pending_exceptions and do the check there.

I can do that if you want, but kvmppc_core_deliver_interrupts() should always get called before we return to the guest. Dequeues that are asynchronous to a guest exit should be very rare, and would be cured on the first subsequent guest exit.

-Scott

#3 lore raw

Re: [PATCH 1/5] KVM: PPC: booke: Fix int_pending calculation for MSR[EE] paravirt

In reply to Scott Wood
Quoted 12 lines · Scott Wood
Quoted 10 lines · Alexander Graf
Quoted 4 lines · Scott Wood

int_pending was only being lowered if a bit in pending_exceptions was cleared during exception delivery -- but for interrupts, we clear it during IACK/TSR emulation. This caused paravirt for enabling MSR[EE] to be ineffective.

But that means that int_pending can still be 1 even though there is none pending as we don't get the call to deliver_interrupts when it gets lowered. Please create a common function to remove a bit from pending_exceptions and do the check there.

Quoted 4 lines

I can do that if you want, but kvmppc_core_deliver_interrupts() should always get called before we return to the guest. Dequeues that are asynchronous to a guest exit should be very rare, and would be cured on the first subsequent guest exit.

Yes, but this means we have yet another subtile indirect assumption. The more we have those, the more people who work on the code need to know about the code to actually work on it. So the easier we can keep the scheme, the better IMHO.

So yes, please replace clear_bit(pending, ...) with clear_bit_pending(...) or so which then would also do the required magic to set the shared page values.

Alex

author
#4 lore raw

Re: [PATCH 1/5] KVM: PPC: booke: Fix int_pending calculation for

In reply to Scott Wood
Quoted 15 lines · Alexander Graf
Quoted 12 lines · Scott Wood
Quoted 10 lines · Alexander Graf
Quoted 4 lines · Scott Wood

int_pending was only being lowered if a bit in pending_exceptions was cleared during exception delivery -- but for interrupts, we clear it during IACK/TSR emulation. This caused paravirt for enabling MSR[EE] to be ineffective.

But that means that int_pending can still be 1 even though there is none pending as we don't get the call to deliver_interrupts when it gets lowered. Please create a common function to remove a bit from pending_exceptions and do the check there.

Quoted 8 lines
Quoted 4 lines

I can do that if you want, but kvmppc_core_deliver_interrupts() should always get called before we return to the guest. Dequeues that are asynchronous to a guest exit should be very rare, and would be cured on the first subsequent guest exit.

Yes, but this means we have yet another subtile indirect assumption. The more we have those, the more people who work on the code need to know about the code to actually work on it. So the easier we can keep the scheme, the better IMHO.

So yes, please replace clear_bit(pending, ...) with clear_bit_pending(...) or so which then would also do the required magic to set the shared page values.

OK.

-Scott

author
#5 lore raw

Re: [PATCH 1/5] KVM: PPC: booke: Fix int_pending calculation for

In reply to Scott Wood
Quoted 20 lines · Alexander Graf
Quoted 12 lines · Scott Wood
Quoted 10 lines · Alexander Graf
Quoted 4 lines · Scott Wood

int_pending was only being lowered if a bit in pending_exceptions was cleared during exception delivery -- but for interrupts, we clear it during IACK/TSR emulation. This caused paravirt for enabling MSR[EE] to be ineffective.

But that means that int_pending can still be 1 even though there is none pending as we don't get the call to deliver_interrupts when it gets lowered. Please create a common function to remove a bit from pending_exceptions and do the check there.

Quoted 4 lines

I can do that if you want, but kvmppc_core_deliver_interrupts() should always get called before we return to the guest. Dequeues that are asynchronous to a guest exit should be very rare, and would be cured on the first subsequent guest exit.

Quoted 8 lines

Yes, but this means we have yet another subtile indirect assumption. The more we have those, the more people who work on the code need to know about the code to actually work on it. So the easier we can keep the scheme, the better IMHO.

So yes, please replace clear_bit(pending, ...) with clear_bit_pending(...) or so which then would also do the required magic to set the shared page values.

Then we need to deal with races, if another exception is queued after we check pending_exceptions but before we clear int_pending, whereas the guest exit path will always be invoked if we queue an extirq or timer.

What if we rename it from deliver_interrupts() to prepare_to_enter(), to make it more obvious?

We also will want to check for MSR[WE] before returning to the guest, rather than immediately when MSR is set, and this would be a place to put that.

-Scott