mirror of
git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
synced 2025-08-05 00:34:52 +00:00
394fcd8a81
Currently, tcp sendmsg(MSG_ZEROCOPY) is building skbs with order-0 fragments. Compared to standard sendmsg(), these skbs usually contain up to 16 fragments on arches with 4KB page sizes, instead of two. This adds considerable costs on various ndo_start_xmit() handlers, especially when IOMMU is in the picture. As high performance applications are often using huge pages, we can try to combine adjacent pages belonging to same compound page. Tested on AMD Rome platform, with IOMMU, nominal single TCP flow speed is roughly doubled (~55Gbit -> ~100Gbit), when user application is using hugepages. For reference, nominal single TCP flow speed on this platform without MSG_ZEROCOPY is ~65Gbit. Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Willem de Bruijn <willemb@google.com> Signed-off-by: David S. Miller <davem@davemloft.net> |
||
|---|---|---|
| .. | ||
| bpf_sk_storage.c | ||
| datagram.c | ||
| datagram.h | ||
| dev.c | ||
| dev_addr_lists.c | ||
| dev_ioctl.c | ||
| devlink.c | ||
| drop_monitor.c | ||
| dst.c | ||
| dst_cache.c | ||
| failover.c | ||
| fib_notifier.c | ||
| fib_rules.c | ||
| filter.c | ||
| flow_dissector.c | ||
| flow_offload.c | ||
| gen_estimator.c | ||
| gen_stats.c | ||
| gro_cells.c | ||
| hwbm.c | ||
| link_watch.c | ||
| lwt_bpf.c | ||
| lwtunnel.c | ||
| Makefile | ||
| neighbour.c | ||
| net-procfs.c | ||
| net-sysfs.c | ||
| net-sysfs.h | ||
| net-traces.c | ||
| net_namespace.c | ||
| netclassid_cgroup.c | ||
| netevent.c | ||
| netpoll.c | ||
| netprio_cgroup.c | ||
| page_pool.c | ||
| pktgen.c | ||
| ptp_classifier.c | ||
| request_sock.c | ||
| rtnetlink.c | ||
| scm.c | ||
| secure_seq.c | ||
| skbuff.c | ||
| skmsg.c | ||
| sock.c | ||
| sock_diag.c | ||
| sock_map.c | ||
| sock_reuseport.c | ||
| stream.c | ||
| sysctl_net_core.c | ||
| timestamping.c | ||
| tso.c | ||
| utils.c | ||
| xdp.c | ||