22.06.2021 01:33, Jim Mattson пишет:
Maybe what you want is run->ready_for_interrupt_injection? And, if
that's not set, try KVM_RUN with run->request_interrupt_window set?
static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
{
return kvm_arch_interrupt_allowed(vcpu) &&
!kvm_cpu_has_interrupt(vcpu) &&
!kvm_event_needs_reinjection(vcpu) &&
kvm_cpu_accept_dm_intr(vcpu);
}
So judging from this snippet,
I wouldn't bet on the right indication
from run->ready_for_interrupt_injection
in our situation.
It doesn't check for vcpu->arch.exception.pending
or anything like that.
I believe, the exit to user-space
with pending synchronous exception
was not supposed to happen (but it does).
Also x86_emulate_instruction() seems
to be doing kvm_clear_exception_queue(vcpu)
before anything else, so obviously
such scenario is not trivial...
Possibly the non-emulate path
forgets to clear the queue on entry?
