Re: [PATCH v5 12/15] KVM: s390: do not restore IAM immediately before SIE entry

From: Michael Mueller
Date: Mon Jan 07 2019 - 12:53:34 EST

On 03.01.19 16:00, Pierre Morel wrote:
On 19/12/2018 20:17, Michael Mueller wrote:
The IAM shall no be restored when deliverable interruptions are
delivered to vcpus by means of the PSW swap mechanism. That would
trigger the GIB alert millicode although we know that SIE will be
able to handle the pending interruption on entry.

Signed-off-by: Michael Mueller <mimu@xxxxxxxxxxxxx>
 arch/s390/kvm/interrupt.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/arch/s390/kvm/interrupt.c b/arch/s390/kvm/interrupt.c
index 8307717e3caf..48a93f5e5333 100644
--- a/arch/s390/kvm/interrupt.c
+++ b/arch/s390/kvm/interrupt.c
@@ -328,11 +328,11 @@ static unsigned long disable_iscs(struct kvm_vcpu *vcpu,
ÂÂÂÂÂ return active_mask;
-static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu)
+static unsigned long deliverable_irqs(struct kvm_vcpu *vcpu, u16 irq_flags)
ÂÂÂÂÂ unsigned long active_mask;
-ÂÂÂ active_mask = pending_irqs(vcpu, IRQ_MASK_ALL);
+ÂÂÂ active_mask = pending_irqs(vcpu, irq_flags);
ÂÂÂÂÂ if (!active_mask)
ÂÂÂÂÂÂÂÂÂ return 0;
@@ -1090,7 +1090,7 @@ int kvm_s390_ext_call_pending(struct kvm_vcpu *vcpu)
 int kvm_s390_vcpu_has_irq(struct kvm_vcpu *vcpu, int exclude_stop)
-ÂÂÂ if (deliverable_irqs(vcpu))
+ÂÂÂ if (deliverable_irqs(vcpu, IRQ_MASK_ALL | IRQ_FLAG_IAM))

Why do we need to restore IAM here?

please see kvm_s390_handle_wait()

It calls kvm_arch_vcpu_runnable() / kvm_s390_vcpu_has_irq()

That's the place where we want the IAM to be restored when
no ISC is pending in the IPM anymore.