linux/tools/include/uapi/linux
Bobby Eshleman b27a8560ee net: devmem: allow rx-page-size > PAGE_SIZE per dmabuf binding
Every devmem dmabuf binding today hands the page_pool PAGE_SIZE niovs.
This caps a single RX descriptor at PAGE_SIZE, burning CPU on buffer
churn for large flows.

Add a bind-time netlink attribute, NETDEV_A_DMABUF_RX_PAGE_SIZE, that
lets userspace request a larger niov size. The value must be a power of
two >= PAGE_SIZE.

The TX path is changed to always pass PAGE_SIZE.

Measurements:
Setup: kperf in devmem RX/TX cuda mode, 4 flows, 64 MB messages, 60s,
dctcp, num-rx-queues=4, dmabuf-rx/tx-size-mb=2048, 10 runs per niov
size, mlx5.

CPU Util:

   niov        net sirq %        net idle %         app sys %        app idle %
  -----  ----------------  ----------------  ----------------  ----------------
     4K   62.38 +/-  8.27   33.40 +/-  7.51   54.15 +/- 10.23   43.67 +/- 10.53
    16K   58.91 +/-  5.35   35.23 +/-  5.88   41.05 +/-  8.87   56.42 +/-  9.24
    32K   64.12 +/-  0.68   31.09 +/-  1.48   44.54 +/-  3.51   52.63 +/-  3.65
    64K   54.69 +/-  5.54   39.67 +/-  5.81   35.47 +/-  3.11   61.97 +/-  3.27

RX app sys % drops ~19% from 4K to 64K.

Throughput:

   niov       RX dev Gbps   RX flow avg Gbps
  -----  ----------------  -----------------
     4K  300.63 +/- 53.21    75.16 +/- 13.30
    16K  321.35 +/- 28.20    80.34 +/-  7.05
    32K  347.63 +/-  2.20    86.91 +/-  0.55
    64K  332.11 +/- 14.26    83.03 +/-  3.56

Throughput seems to increase, but the stdev is pretty wide so could just
be noise.

kperf support (not yet merged):
8837577f92

Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Reviewed-by: Mina Almasry <almasrymina@google.com>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Signed-off-by: Bobby Eshleman <bobbyeshleman@meta.com>
Link: https://patch.msgid.link/20260805-tcpdm-large-niovs-v8-1-3e0225e2808c@meta.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2026-08-07 18:32:39 -07:00
..
tc_act headers: Remove some left-over license text 2022-09-27 07:48:01 -07:00
acct.h tools headers UAPI: sync linux/taskstats.h for procacct.c 2026-06-03 16:25:50 -07:00
bits.h uapi: bitops: use UAPI-safe variant of BITS_PER_LONG again (2) 2025-07-08 10:23:13 -04:00
bpf_common.h
bpf_perf_event.h tools, headers: Sync struct bpf_perf_event_data 2021-01-26 00:15:03 +01:00
bpf.h bpf: Add support to specify uprobe_multi target via file descriptor 2026-06-14 17:24:25 -07:00
btf.h bpf: Extend BTF UAPI vlen, kinds to use unused bits 2026-04-20 17:52:48 -07:00
const.h tools headers UAPI: Sync linux/const.h with the kernel sources 2026-07-21 21:00:09 -03:00
coredump.h tools: add coredump.h header 2025-06-12 14:00:32 +02:00
elf.h tools/include: Add uapi/linux/elf.h 2025-03-03 20:00:12 +01:00
erspan.h
fanotify.h selftests/fs/mount-notify: build with tools include dir 2025-05-12 11:40:12 +02:00
filter.h tools: bpf: Use local copy of headers including uapi/linux/filter.h 2020-07-21 10:50:35 +01:00
fs.h tools headers UAPI: sync linux/fs.h with the kernel sources 2025-05-11 17:48:16 -07:00
fscrypt.h tools headers: Update the fs headers with the kernel sources 2025-06-16 14:05:10 -03:00
genetlink.h tools include: Add headers to make tools builds more hermetic 2025-10-02 15:13:19 -03:00
hw_breakpoint.h Move bp_type_idx to include/linux/hw_breakpoint.h 2023-03-10 21:05:16 +01:00
if_addr.h tools include: Add headers to make tools builds more hermetic 2025-10-02 15:13:19 -03:00
if_link.h bonding: 3ad: add lacp_strict configuration knob 2026-06-10 07:53:26 -07:00
if_tun.h tun/tap: add IFF_BACKPRESSURE flag 2026-08-07 17:29:15 -07:00
if_xdp.h net: xsk: introduce XDP_MAX_TX_SKB_BUDGET setsockopt 2025-07-10 14:48:29 +02:00
in.h tools headers: Update the socket headers with the kernel sources 2025-04-10 09:28:24 -07:00
io_uring.h tools headers: Grab copy of io_uring.h 2023-10-19 16:42:03 -06:00
kcmp.h
kvm.h tools headers: Sync KVM headers with the kernel sources 2026-07-21 21:00:10 -03:00
memfd.h selftests/mm: fix additional build errors for selftests 2024-04-25 20:56:42 -07:00
mman.h mm: add MAP_DROPPABLE for designating always lazily freeable mappings 2024-07-19 20:22:12 +02:00
module_signature.h tools uapi headers: add linux/module_signature.h 2026-03-24 21:42:37 +00:00
mount.h tools: update mount.h header 2026-03-12 13:33:54 +01:00
neighbour.h tools include: Add headers to make tools builds more hermetic 2025-10-02 15:13:19 -03:00
netdev.h net: devmem: allow rx-page-size > PAGE_SIZE per dmabuf binding 2026-08-07 18:32:39 -07:00
netfilter_arp.h tools include: Add headers to make tools builds more hermetic 2025-10-02 15:13:19 -03:00
netfilter.h tools include: Add headers to make tools builds more hermetic 2025-10-02 15:13:19 -03:00
netlink.h
nsfs.h nsfs: update tools header 2025-11-03 17:41:18 +01:00
openat2.h selftest: add tests for O_EMPTYPATH 2026-05-21 10:53:37 +02:00
perf_event.h perf beauty: Update the linux/perf_event.h copy with the kernel sources 2026-03-04 11:51:36 -03:00
pkt_cls.h net/sched: Remove uapi support for tcindex classifier 2024-01-02 14:25:51 +00:00
pkt_sched.h net/sched: Remove uapi support for CBQ qdisc 2024-01-02 14:25:51 +00:00
prctl.h Updates for the generic entry code: 2025-07-29 15:14:29 -07:00
rtnetlink.h tools headers UAPI: Sync linux/rtnetlink.h with the kernel sources 2026-07-21 21:00:09 -03:00
seccomp.h tools headers UAPI: Copy seccomp.h to be able to build 'perf bench' in older systems 2023-09-13 08:48:48 -03:00
seg6_local.h
seg6.h treewide: uapi: Replace zero-length arrays with flexible-array members 2022-06-28 21:26:05 +02:00
stat.h tools headers: Update the fs headers with the kernel sources 2025-06-16 14:05:10 -03:00
stddef.h tools/headers: Regenerate stddef.h to fix BPF selftests 2026-05-02 09:37:44 -07:00
taskstats.h tools headers UAPI: sync linux/taskstats.h 2026-03-27 21:19:43 -07:00
tcp.h bpf: Remove extra lock_sock for TCP_ZEROCOPY_RECEIVE 2021-01-20 14:23:00 -08:00
tls.h
types.h tools/include: make uapi/linux/types.h usable from assembly 2025-04-06 12:55:31 -07:00
userfaultfd.h selftests/mm: fix additional build errors for selftests 2024-04-25 20:56:42 -07:00