On Mon, Jun 6, 2016 at 4:35 PM, Paolo Bonzini <pbonzini@xxxxxxxxxx> wrote: > > > On 06/06/2016 15:55, Michael S. Tsirkin wrote: >> > According to the spec, if the VIRTIO_RING_F_EVENT_IDX feature bit is >> > negotiated the driver MUST set flags to 0. Not dirtying the available >> > ring in virtqueue_disable_cb may also have a positive performance impact. >> >> Question would be, is this a gain or a loss. We have an extra branch, >> and the write might serve to prefetch the cache line. >> >> > Writes to the used event field (vring_used_event) are still unconditional. >> > >> > Signed-off-by: Ladi Prosek <lprosek@xxxxxxxxxx> >> >> Wow you are right wrt the spec. Should we change the spec or the >> code though? > > I would change the spec and note that bit 0 of the flags is ignored if > event indices are in use. Changing the spec sounds good. I'll see if I can get any meaningful perf numbers with vring_bench, just in case. Would there be any interest in having the tool checked in the tree? There are several commits referencing vring_bench but it seems to exist only in a list archive - took me a while to figure it out. Also apologies for posting in a wrong list. Thanks, Ladi -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
