On Wed, May 19, 2010 at 11:06:42AM +0300, Avi Kivity wrote: > On 05/19/2010 10:39 AM, Rusty Russell wrote: >> >> I think we're talking about the last 2 entries of the avail ring. That means >> the worst case is 1 false bounce every time around the ring. > > It's low, but why introduce an inefficiency when you can avoid doing it > for the same effort? > >> I think that's >> why we're debating it instead of measuring it :) >> > > Measure before optimize is good for code but not for protocols. > Protocols have to be robust against future changes. Virtio is warty > enough already, we can't keep doing local optimizations. > >> Note that this is a exclusive->shared->exclusive bounce only, too. >> > > A bounce is a bounce. > > Virtio is already way too bouncy due to the indirection between the > avail/used rings and the descriptor pool. A device with out of order > completion (like virtio-blk) will quickly randomize the unused > descriptor indexes, so every descriptor fetch will require a bounce. > > In contrast, if the rings hold the descriptors themselves instead of > pointers, we bounce (sizeof(descriptor)/cache_line_size) cache lines for > every descriptor, amortized. On the other hand, consider that on fast path we are never using all of the ring. With a good allocator we might be able to keep reusing only small part of the ring, instead of wrapping around all of it all of the time. > -- > Do not meddle in the internals of kernels, for they are subtle and quick to panic. -- 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