On 28/02/20 16:25, Peter Xu wrote: > My current thought is that we keep this hackery in userspace only so > we keep split+resamplefd forbidden in the kernel and be clean there.
It is better, yes. The kernel solution makes some sense because split irqchip _does_ have a concept of ioapic GSIs and routes. But the kernel patch would not fix the fact that remote-irr remains zero, which would need a userspace change anyway. Paolo
