On 28/02/20 15:58, Peter Xu wrote: > Yes I was thinking it the same way, however... I noticed that VFIO > will even mask edge INTx after getting IRQ (vfio_intx_handler).
INTx is always level-triggered so that's okay. > And if look into current KVM kernel implementation, it has done the > same logic to ack edge triggerred irqs (kvm_ioapic_update_eoi_one), > we should logically ack it here too even for edge-triggered? Right, that's because it's using ack notifiers also for in-kernel PIT and for RTC. But yeah, it makes sense to have userspace IOAPIC resamplefd match the functionality of the kernel ones. Thanks! Paolo
