On Thu, Feb 07, 2013 at 02:20:53PM +0100, Paolo Bonzini wrote: > Il 07/02/2013 14:23, Michael S. Tsirkin ha scritto: > > On Thu, Feb 07, 2013 at 02:14:24PM +0100, Paolo Bonzini wrote: > >> Il 07/02/2013 14:09, Michael S. Tsirkin ha scritto: > >>>> One major difference between virtqueue_add_buf and virtqueue_add_sg > >>>> is that the latter uses scatterlist iterators, which follow chained > >>>> scatterlist structs and stop at ending markers. In order to avoid code > >>>> duplication, and use the new API from virtqueue_add_buf (patch 8), we > >>>> need > >>>> to change all existing callers of virtqueue_add_buf to provide > >>>> well-formed > >>>> scatterlists. This is what patches 2-7 do. For virtio-blk it is easiest > >>>> to just switch to the new API, just like for virtio-scsi. For virtio-net > >>>> the ending marker must be reset after calling virtqueue_add_buf, in > >>>> preparation for the next usage of the scatterlist. Other drivers are > >>>> safe already. > >>> > >>> What are the changes as compared to the previous version? > >>> How about some comments made on the previous version? > >>> See e.g. > >>> https://patchwork.kernel.org/patch/1891541/ > >> > >> Two changes: 1) added virtqueue_add_sg_single; 2) reimplemented > >> virtqueue_add_buf in terms of the new API, which requires virtio-blk and > >> virtio-net changes. > >> > >> The virtio-blk and virtio-net changes are based on some ideas in the > >> patch Rusty posted, but virtio-net is a bit simpler and virtio-blk was > >> redone from scratch. > >> > >>> Generally we have code for direct and indirect which is already > >>> painful. We do not want 4 more variants of this code. > >> > >> Yes, indeed, the other main difference is that I'm now reimplementing > >> virtqueue_add_buf using the new functions. So: > >> > >> - we previously had 2 variants (direct/indirect) > >> > >> - v1 had 4 variants (direct/indirect x add_buf/add_sg) > >> > >> - v2 has 4 variants (direct/indirect x add_sg/add_sg_single) > > > > single is never indirect so should have a single variant. > > Single means *this piece* (for example a request header) is single. It > could still end up in an indirect buffer because QEMU does not support > mixed direct/indirect buffers. > > Paolo
Yes but why is the optimization worth it? It makes sense if all we want to do is add a single buffer in one go, this would give us virtqueue_add_buf_single. But if we are building up an s/g list anyway, speeding up one of the entries a tiny bit seems very unlikely to be measureable. No? > >>>> This is an RFC for two reasons. First, because I haven't done enough > >>>> testing yet (especially with all the variations on receiving that > >>>> virtio-net has). Second, because I still have two struct vring_desc * > >>>> fields in virtqueue API, which is a layering violation. I'm not really > >>>> sure how important that is and how to fix that---except by making the > >>>> fields void*. > >>> > >>> Hide the whole structure as part of vring struct, the problem will go > >>> away. > >> > >> Yes, that's the other possibility. Will do for the next submission. > >> > >> Paolo > >> > >>>> Paolo > >>>> Paolo Bonzini (8): > >>>> virtio: add functions for piecewise addition of buffers > >>>> virtio-blk: reorganize virtblk_add_req > >>>> virtio-blk: use virtqueue_start_buf on bio path > >>>> virtio-blk: use virtqueue_start_buf on req path > >>>> scatterlist: introduce sg_unmark_end > >>>> virtio-net: unmark scatterlist ending after virtqueue_add_buf > >>>> virtio-scsi: use virtqueue_start_buf > >>>> virtio: reimplement virtqueue_add_buf using new functions > >>>> > >>>> block/blk-integrity.c | 2 +- > >>>> block/blk-merge.c | 2 +- > >>>> drivers/block/virtio_blk.c | 165 +++++++++-------- > >>>> drivers/net/virtio_net.c | 21 ++- > >>>> drivers/scsi/virtio_scsi.c | 103 +++++------ > >>>> drivers/virtio/virtio_ring.c | 417 > >>>> +++++++++++++++++++++++++++--------------- > >>>> include/linux/scatterlist.h | 16 ++ > >>>> include/linux/virtio.h | 25 +++ > >>>> 8 files changed, 460 insertions(+), 291 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
