netdev: expose page pool order via netlink
Subsystems (6)
v2 cover netdev: expose page pool order via netlink
This small series exposes io_uring's high order page configuration via the page_pool netlink interface and updates the appropriate selftest to check this value.
---
v2:
- Switched from exposing page_pool order to rx_buf_len via nl_fill of the io_uring memory provider.
- Updated selftest to check rx_buf_len.
- v1: https://lore.kernel.org/all/[email protected]/
---
Dragos Tatulea (2):
netdev: expose io_uring rx_page_order order via netlink io_uring/zcrx: selftests: verify rx_buf_len for large chunks
Documentation/netlink/specs/netdev.yaml | 9 ++++++- include/uapi/linux/netdev.h | 2 ++ io_uring/zcrx.c | 8 ++++++ tools/include/uapi/linux/netdev.h | 2 ++ .../selftests/drivers/net/hw/iou-zcrx.py | 26 ++++++++++++++++++- 5 files changed, 45 insertions(+), 2 deletions(-)
Signature · 1 line
2.54.0
PATCH v2 1/2 netdev: expose io_uring rx_page_order order via netlink
This adds observability for the io_uring zcrx rx-buf-len configuration.
--- Documentation/netlink/specs/netdev.yaml | 9 ++++++++- include/uapi/linux/netdev.h | 2 ++ io_uring/zcrx.c | 8 ++++++++ tools/include/uapi/linux/netdev.h | 2 ++ 4 files changed, 20 insertions(+), 1 deletion(-)
Documentation/netlink/specs/netdev.yaml +8 -1
@@ -127,7 +127,14 @@ attribute-sets: enum: xsk-flags - name: io-uring-provider-info - attributes: [] + attributes: + - + name: rx-buf-len + type: uint + doc: | + RX buffer length in bytes for this io_uring memory provider. + Reflects the rx_buf_len passed at io_uring zerocopy rx + registration time. - name: page-pool attributes:
include/uapi/linux/netdev.h +2 -0
@@ -97,6 +97,8 @@ enum { };enum { +NETDEV_A_IO_URING_PROVIDER_INFO_RX_BUF_LEN =1 , + __NETDEV_A_IO_URING_PROVIDER_INFO_MAX,NETDEV_A_IO_URING_PROVIDER_INFO_MAX = (__NETDEV_A_IO_URING_PROVIDER_INFO_MAX -1 ) };
io_uring/zcrx.c +8 -0
@@ -1156,6 +1156,7 @@ static void io_pp_zc_destroy(struct page_pool *pp)static int io_pp_nl_fill (void *mp_priv,struct sk_buff *rsp,struct netdev_rx_queue *rxq) { +struct io_zcrx_ifq *ifq = mp_priv;struct nlattr *nest;int type;
@@ -1163,6 +1164,13 @@ static int io_pp_nl_fill(void *mp_priv, struct sk_buff *rsp, nest =nla_nest_start (rsp, type);if (!nest)return -EMSGSIZE ; + +if (nla_put_uint (rsp,NETDEV_A_IO_URING_PROVIDER_INFO_RX_BUF_LEN , +1ULL << ifq->niov_shift)) { +nla_nest_cancel (rsp, nest); +return -EMSGSIZE ; + } +nla_nest_end (rsp, nest);return 0 ;
tools/include/uapi/linux/netdev.h +2 -0
@@ -97,6 +97,8 @@ enum { };enum { +NETDEV_A_IO_URING_PROVIDER_INFO_RX_BUF_LEN =1 , + __NETDEV_A_IO_URING_PROVIDER_INFO_MAX,NETDEV_A_IO_URING_PROVIDER_INFO_MAX = (__NETDEV_A_IO_URING_PROVIDER_INFO_MAX -1 ) };
Signature · 1 line
2.54.0
Re: [PATCH net-next v2 1/2] netdev: expose io_uring rx_page_order order via netlink
Quoted 1 line · Dragos Tatulea
This adds observability for the
io_uringzcrx rx-buf-len configuration.
It might be nicer to look it up in the queue, e.g. rxq->mp_params, and make it a queue attribute instead of zcrx specific one. In either case, no objections.
Signature · 1 line
Pavel Begunkov
Re: [PATCH net-next v2 1/2] netdev: expose io_uring rx_page_order order via netlink
Quoted 7 lines · Pavel Begunkov
Quoted 1 line · Dragos Tatulea
This adds observability for the
io_uringzcrx rx-buf-len configuration.It might be nicer to look it up in the queue, e.g.
rxq->mp_params, and make it a queue attribute instead of zcrx specific one. In either case, no objections.
In io_pp_nl_fill() or in page_pool_nl_fill() as it was done in v1 for order?
Thanks,
Dragos
Jakub Kicinski
maintainerRe: [PATCH net-next v2 1/2] netdev: expose io_uring rx_page_order order via netlink
Quoted 9 lines · On Sat, 13 Jun 2026 16:09:03 +0200 Dragos Tatulea wrote:
Quoted 6 lines · Pavel Begunkov
Quoted 1 line · Dragos Tatulea
This adds observability for the
io_uringzcrx rx-buf-len configuration.It might be nicer to look it up in the queue, e.g.
rxq->mp_params, and make it a queue attribute instead of zcrx specific one. In either case, no objections.In
io_pp_nl_fill()or inpage_pool_nl_fill()as it was done in v1 for order?
It's fine. We decided to make the "page size" a memory provider property, now we're going back to making it a queue level param? Like my RFC had that everyone hated so much? Sigh.
Pavel Begunkov
maintainerRe: [PATCH net-next v2 1/2] netdev: expose io_uring rx_page_order order via netlink
Quoted 9 lines · Jakub Kicinski
Quoted 8 lines · On Sat, 13 Jun 2026 16:09:03 +0200 Dragos Tatulea wrote:
Quoted 6 lines · Pavel Begunkov
Quoted 1 line · Dragos Tatulea
This adds observability for the
io_uringzcrx rx-buf-len configuration.It might be nicer to look it up in the queue, e.g.
rxq->mp_params, and make it a queue attribute instead of zcrx specific one. In either case, no objections.
Quoted 5 lines
Quoted 1 line
In
io_pp_nl_fill()or inpage_pool_nl_fill()as it was done in v1 for order?It's fine. We decided to make the "page size" a memory provider property, now we're going back to making it a queue level param? Like my RFC had that everyone hated so much? Sigh.
TBH, I never cared much how nl would show it, so not opposing either version. My idea is that even without plumbing in per-queue non-mp size configuration, it'd be nice to have a common way to check it b/w providers.
From the semantics and observability perspective, zcrx is probably not that interesting as the parameter is basically just a hint with no affect on uapi, and I'd assume people would rather see the page pool size or even the NIC's page size. But I guess it depends on what Dragos is really after with this patch.
Signature · 1 line
Pavel Begunkov
Jakub Kicinski
maintainerRe: [PATCH net-next v2 1/2] netdev: expose io_uring rx_page_order order via netlink
Quoted 13 lines · On Mon, 15 Jun 2026 12:16:47 +0100 Pavel Begunkov wrote:
Quoted 6 lines · Jakub Kicinski
Quoted 1 line · On Sat, 13 Jun 2026 16:09:03 +0200 Dragos Tatulea wrote:
In
io_pp_nl_fill()or inpage_pool_nl_fill()as it was done in v1 for order?It's fine. We decided to make the "page size" a memory provider property, now we're going back to making it a queue level param? Like my RFC had that everyone hated so much? Sigh.
TBH, I never cared much how nl would show it, so not opposing either version. My idea is that even without plumbing in per-queue non-mp size configuration, it'd be nice to have a common way to check it b/w providers.
Quoted 5 lines
From the semantics and observability perspective, zcrx is probably not
that interesting as the parameter is basically just a hint with no affect on uapi, and I'd assume people would rather see the page pool size or even the NIC's page size. But I guess it depends on what Dragos is really after with this patch.
Not sure what Dragos's use case is but IMO it's useful as system admin / vendor to be able to peek at the important params when user reports bad perf. The netlink MP info is meant for system observability.
Pavel Begunkov
maintainerRe: [PATCH net-next v2 1/2] netdev: expose io_uring rx_page_order order via netlink
In reply to Dragos TatuleaQuoted 9 lines · Dragos Tatulea
Quoted 7 lines · Pavel Begunkov
Quoted 1 line · Dragos Tatulea
This adds observability for the
io_uringzcrx rx-buf-len configuration.It might be nicer to look it up in the queue, e.g.
rxq->mp_params, and make it a queue attribute instead of zcrx specific one. In either case, no objections.In
io_pp_nl_fill()or inpage_pool_nl_fill()as it was done in v1 for order?
I didn't see v1, so yeah, I was thinking along the same lines as your previous version. Oh, well
Signature · 1 line
Pavel Begunkov
PATCH v2 2/2 io_uring/zcrx: selftests: verify rx_buf_len for large chunks
In reply to Dragos TatuleaCheck the newly added rx_buf_len page_pool field for io_uring in the existing large-chunks test after the receiver is up.
--- .../selftests/drivers/net/hw/iou-zcrx.py | 26 ++++++++++++++++++- 1 file changed, 25 insertions(+), 1 deletion(-)
tools/testing/selftests/drivers/net/hw/iou-zcrx.py +25 -1
@@ -2,6 +2,7 @@ # SPDX-License-Identifier: GPL-2.0 import re +import resource import time from os import path from lib.py import ksft_run, ksft_exit, KsftSkipEx, ksft_variants, KsftNamedVariant
@@ -41,6 +42,23 @@ def set_flow_rule_rss(cfg, rss_ctx_id): return int(values) +def check_iou_rx_buf_len(cfg, expected_rx_buf_len): + """Check the io-uring memory provider exposes the expected rx_buf_len.""" + q = cfg.netnl.queue_get({'ifindex': cfg.ifindex, 'type': 'rx', 'id': cfg.target}) + napi_id = q['napi-id'] + pools = cfg.netnl.page_pool_get({}, dump=True) + pools = [p for p in pools if p.get('napi-id') == napi_id + and 'io-uring' in p] + if len(pools) != 1: + raise Exception(f"Expected 1 io-uring page pool, found {len(pools)}") + rx_buf_len = pools[0]['io-uring'].get('rx-buf-len') + if rx_buf_len is None: + raise KsftSkipEx("io-uring 'rx-buf-len' attribute not supported") + if rx_buf_len != expected_rx_buf_len: + raise Exception(f'Expected io-uring rx-buf-len {expected_rx_buf_len}, ' + f'got {rx_buf_len}') + + def single(cfg): channels = cfg.ethnl.channels_get({'header': {'dev-index': cfg.ifindex}}) channels = channels['combined-count']
@@ -156,7 +174,10 @@ def test_zcrx_large_chunks(cfg) -> None: defer(lambda: open(hp_file, 'w', encoding='utf-8').write(str(nr_hugepages))) single(cfg) - rx_cmd = f"{cfg.bin_local} -s -p {cfg.port} -i {cfg.ifname} -q {cfg.target} -x 2" + page_size = resource.getpagesize() + nr_pages = 2 + rx_buf_len = nr_pages * page_size + rx_cmd = f"{cfg.bin_local} -s -p {cfg.port} -i {cfg.ifname} -q {cfg.target} -x {nr_pages}" tx_cmd = f"{cfg.bin_remote} -c -h {cfg.addr_v['6']} -p {cfg.port} -l 12840" probe = cmd(rx_cmd + " -d", fail=False)
@@ -166,6 +187,9 @@ def test_zcrx_large_chunks(cfg) -> None: mp_clear_wait(cfg) with bkg(rx_cmd, exit_wait=True): wait_port_listen(cfg.port, proto="tcp") + + check_iou_rx_buf_len(cfg, rx_buf_len) + cmd(tx_cmd, host=cfg.remote)
Signature · 1 line
2.54.0
This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This series was applied to netdev/net-next.git (main) by Jakub Kicinski <[email protected]>:
Quoted 4 lines · On Sat, 13 Jun 2026 00:17:02 +0300 you wrote:
This small series exposes io_uring's high order page configuration via the
page_poolnetlink interface and updates the appropriate selftest to check this value.
Quoted 8 lines
---
v2:
- Switched from exposingpage_poolorder torx_buf_lenvianl_fillof theio_uringmemory provider.
- Updated selftest to checkrx_buf_len.
- v1: https://lore.kernel.org/all/[email protected]/[...]
Here is the summary with links:
- [net-next,v2,1/2] netdev: expose io_uring rx_page_order order via netlink
https://git.kernel.org/netdev/net-next/c/5c4adb7fb46f - [net-next,v2,2/2] io_uring/zcrx: selftests: verify rx_buf_len for large chunks https://git.kernel.org/netdev/net-next/c/18f65355e112
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html