On Tue, Dec 9, 2014 at 2:35 PM, Masao Uebayashi <[email protected]> wrote: > I see that vioif(4) schedules another softint from within > vioif_rx_vq_done(), that would latency a bit longer.
Good point. If I directly call vioif_rx_softint, the performance improves regardless of softint-ed interrupts on/off. The two performance are almost same and better than the vanilla. In the case of softint-ed interrupts on, I think we can safely remove the indirection. OTOH, in the case of off, I'm not sure we can do the same thing. > >> +#define VIRTIO_F_PCI_INTR_SOFTINT (2 << 0) > > Usually (1 << 1). Oops. You're right. Fixed. ozaki-r
