1) Inline this small helper to reduce code size and decrease cpu costs.
2) Constify its argument.
3) Move it to include/net/netmem.h, as a prereq for the following patch.
$ scripts/bloat-o-meter -t vmlinux.2 vmlinux.3
add/remove: 0/2 grow/shrink: 0/4 up/down: 0/-158 (-158)
Function old new delta
validate_xmit_skb 866 857 -9
__pfx_net_is_devmem_iov 16 - -16
net_is_devmem_iov 22 - -22
get_netmem 152 130 -22
put_netmem 140 114 -26
tcp_recvmsg_locked 3860 3797 -63
Total: Before=22566015, After=22565857, chg -0.00%
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260122045720.1221017-4-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
I imagine (tm) that as the number of per-queue configuration
options grows some of them may conflict for certain drivers.
While the drivers can obviously do all the validation locally
doing so is fairly inconvenient as the config is fed to drivers
piecemeal via different ops (for different params and NIC-wide
vs per-queue).
Add a centralized callback for validating the queue config
in queue ops. The callback gets invoked before memory provider
is installed, and in the future should also be called when ring
params are modified.
The validation is done after each layer of configuration.
Since we can't fail MP un-binding we must make sure that
the config is valid both before and after MP overrides are
applied. This is moot for now since the set of MP and device
configs are disjoint. It will matter significantly in the future,
so adding it now so that we don't forget..
Link: https://patch.msgid.link/20260122005113.2476634-6-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We should follow the prepare/commit approach for queue configuration.
The qcfg struct should be added to dev->cfg rather than directly to
queue objects so that we can clone and discard the pending config
easily.
Remove the qcfg in struct netdev_rx_queue, and switch remaining callers
to netdev_queue_config(). netdev_queue_config() will construct the qcfg
on the fly based on device defaults and state of the queue.
ndo_default_qcfg becomes optional because having the callback itself
does not have any meaningful semantics to us.
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Link: https://patch.msgid.link/20260122005113.2476634-5-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We may choose to extend or reimplement the logic which renders
the per-queue config. The drivers should not poke directly into
the queue state. Add a helper for drivers to use when they want
to query the config for a specific queue.
Link: https://patch.msgid.link/20260122005113.2476634-3-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
GSO partial features for tunnels do not require any kind of support from
the underlying device: we can safely add them to the geneve UDP tunnel.
The only point of attention is the skb required features propagation in
the device xmit op: partial features must be stripped, except for
UDP_TUNNEL*.
Keep partial features disabled by default.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Link: https://patch.msgid.link/d851ca8e928cf05d68310bcbaeaa5e9e0b01e058.1769011015.git.pabeni@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_rate_check_app_limited() is used from tcp_sendmsg_locked()
fast path and from other callers.
Move it to tcp.c so that it can be inlined in tcp_sendmsg_locked().
Small increase of code, for better TCP performance.
$ scripts/bloat-o-meter -t vmlinux.old vmlinux.new
add/remove: 0/0 grow/shrink: 1/0 up/down: 87/0 (87)
Function old new delta
tcp_sendmsg_locked 4217 4304 +87
Total: Before=22566462, After=22566549, chg +0.00%
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Link: https://patch.msgid.link/20260121095923.3134639-3-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This function is called from one caller only, in TCP fast path.
Move it to tcp_input.c so that compiler can inline it.
$ scripts/bloat-o-meter -t vmlinux.old vmlinux.new
add/remove: 0/2 grow/shrink: 1/0 up/down: 226/-300 (-74)
Function old new delta
tcp_ack 5405 5631 +226
__pfx_tcp_rate_gen 16 - -16
tcp_rate_gen 284 - -284
Total: Before=22566536, After=22566462, chg -0.00%
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Link: https://patch.msgid.link/20260121095923.3134639-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The pipapo set backend is the only user of the .abort interface so far.
To speed up pipapo abort path, removals are skipped.
The follow up patch updates the rbtree to use to build an array of
ordered elements, then use binary search. This needs a new .abort
interface but, unlike pipapo, it also need to undo/remove elements.
Add a flag and use it from the pipapo set backend.
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Signed-off-by: Florian Westphal <fw@strlen.de>
Johannes Berg says:
====================
Another set of updates:
- various small fixes for ath10k/ath12k/mwifiex/rsi
- cfg80211 fix for HE bitrate overflow
- mac80211 fixes
- S1G beacon handling in scan
- skb tailroom handling for HW encryption
- CSA fix for multi-link
- handling of disabled links during association
* tag 'wireless-2026-11-22' of https://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless:
wifi: cfg80211: ignore link disabled flag from userspace
wifi: mac80211: apply advertised TTLM from association response
wifi: mac80211: parse all TTLM entries
wifi: mac80211: don't increment crypto_tx_tailroom_needed_cnt twice
wifi: mac80211: don't perform DA check on S1G beacon
wifi: ath12k: Fix wrong P2P device link id issue
wifi: ath12k: fix dead lock while flushing management frames
wifi: ath12k: Fix scan state stuck in ABORTING after cancel_remain_on_channel
wifi: ath12k: cancel scan only on active scan vdev
wifi: mwifiex: Fix a loop in mwifiex_update_ampdu_rxwinsize()
wifi: mac80211: correctly check if CSA is active
wifi: cfg80211: Fix bitrate calculation overflow for HE rates
wifi: rsi: Fix memory corruption due to not set vif driver data size
wifi: ath12k: don't force radio frequency check in freq_to_idx()
wifi: ath12k: fix dma_free_coherent() pointer
wifi: ath10k: fix dma_free_coherent() pointer
====================
Link: https://patch.msgid.link/20260122110248.15450-3-johannes@sipsolutions.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The rtnl lock might be locked, preventing ad_cond_set_peer_notif() from
acquiring the lock and updating send_peer_notif. This patch addresses
the issue by using a workqueue. Since updating send_peer_notif does
not require high real-time performance, such delayed updates are entirely
acceptable.
In fact, checking this value and using it in multiple places, all operations
are protected at the same time by rtnl lock, such as
- read send_peer_notif
- send_peer_notif--
- bond_should_notify_peers
By the way, rtnl lock is still required, when accessing bond.params.* for
updating send_peer_notif. In lacp mode, resetting send_peer_notif in
workqueue is safe, simple and effective way.
Additionally, this patch introduces bond_peer_notify_may_events(), which
is used to check whether an event should be sent. This function will be
used in both patch 1 and 2.
Cc: Jay Vosburgh <jv@jvosburgh.net>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: Simon Horman <horms@kernel.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Andrew Lunn <andrew+netdev@lunn.ch>
Cc: Nikolay Aleksandrov <razor@blackwall.org>
Cc: Hangbin Liu <liuhangbin@gmail.com>
Cc: Jason Xing <kerneljasonxing@gmail.com>
Suggested-by: Hangbin Liu <liuhangbin@gmail.com>
Signed-off-by: Tonghao Zhang <tonghao@bamaicloud.com>
Reviewed-by: Hangbin Liu <liuhangbin@gmail.com>
Link: https://patch.msgid.link/f95accb5db0b10ce3ed2f834fc70f716c9abbb9c.1768709239.git.tonghao@bamaicloud.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Florian Westphal says:
====================
Subject: netfilter: updates for net-next
1) Speed up nftables transactions after earlier transaction failed.
Due to a (harmeless) bug we remained in slow paranoia mode until
a successful transaction completes.
2) Allow generic tracker to resolve clashes, this avoids very rare
packet drops. From Yuto Hamaguchi.
3) Increase the cleanup budget to 64 entries in nf_conncount to reap
more entries in one go, from Fernando Fernandez Mancera.
4) Allow icmp trackers to resolve clashes, this avoids very rare
initial packet drop with test cases that have high-frequency pings.
After this all trackers except tcp and sctp allow clash resolution.
5) Disentangle netfilter headers, don't include nftables/xtables headers
in subsystems that are unrelated.
6) Don't rely on implicit includes coming from nf_conntrack_proto_gre.h.
7) Allow nfnetlink_queue nfq instance struct to get accounted via memcg,
from Scott Mitchell.
8) Reject bogus xt target/match data upfront via netlink policiy in
nft_compat interface rather than relying on x_tables API to do it.
9) Fix nf_conncount breakage when trying to limit loopback flows via
prerouting rule, from Fernando Fernandez Mancera.
This is a recent breakage but not seen as urgent enough to rush this
via net tree at this late stage in development cycle.
10) Fix a possible off-by-one when parsing tcp option in xtables tcpmss
match. Also handled via -next due to late stage in development
cycle.
* tag 'nf-next-26-01-20' of https://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf-next:
netfilter: xt_tcpmss: check remaining length before reading optlen
netfilter: nf_conncount: fix tracking of connections from localhost
netfilter: nft_compat: add more restrictions on netlink attributes
netfilter: nfnetlink_queue: nfqnl_instance GFP_ATOMIC -> GFP_KERNEL_ACCOUNT allocation
netfilter: nf_conntrack: don't rely on implicit includes
netfilter: don't include xt and nftables.h in unrelated subsystems
netfilter: nf_conntrack: enable icmp clash support
netfilter: nf_conncount: increase the connection clean up limit to 64
netfilter: nf_conntrack: Add allow_clash to generic protocol handler
netfilter: nf_tables: reset table validation state on abort
====================
Link: https://patch.msgid.link/20260120191803.22208-1-fw@strlen.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
FDO/LTO are unable to inline tcp6_gro_receive() from ipv6_gro_receive()
Make sure tcp6_check_fraglist_gro() is only called only when needed,
so that compiler can leave it out-of-line.
$ scripts/bloat-o-meter -t vmlinux.1 vmlinux.2
add/remove: 2/0 grow/shrink: 3/1 up/down: 1123/-253 (870)
Function old new delta
ipv6_gro_receive 1069 1846 +777
tcp6_check_fraglist_gro - 272 +272
ipv6_offload_init 218 274 +56
__pfx_tcp6_check_fraglist_gro - 16 +16
ipv6_gro_complete 433 435 +2
tcp6_gro_receive 959 706 -253
Total: Before=22592662, After=22593532, chg +0.00%
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20260120164903.1912995-3-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tcp_rate_skb_delivered() is only called from tcp_input.c.
Move it there and make it static.
Both gcc and clang are (auto)inlining it, TCP performance
is increased at a small space cost.
$ scripts/bloat-o-meter -t vmlinux.old vmlinux.new
add/remove: 0/2 grow/shrink: 3/0 up/down: 509/-187 (322)
Function old new delta
tcp_sacktag_walk 1682 1867 +185
tcp_ack 5230 5405 +175
tcp_shifted_skb 437 586 +149
__pfx_tcp_rate_skb_delivered 16 - -16
tcp_rate_skb_delivered 171 - -171
Total: Before=22566192, After=22566514, chg +0.00%
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Link: https://patch.msgid.link/20260118123204.2315993-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Pavel Begunkov says:
====================
Add support for providers with large rx buffer
Many modern NICs support configurable receive buffer lengths, and zcrx and
memory providers can use buffers larger than 4K to improve performance.
When paired with hw-gro larger rx buffer sizes can drastically reduce
the number of buffers traversing the stack and save a lot of processing
time. It also allows to give to users larger contiguous chunks of data.
Single stream benchmarks showed up to ~30% CPU util improvement.
E.g. comparison for 4K vs 32K buffers using a 200Gbit NIC:
packets=23987040 (MB=2745098), rps=199559 (MB/s=22837)
CPU %usr %nice %sys %iowait %irq %soft %idle
0 1.53 0.00 27.78 2.72 1.31 66.45 0.22
packets=24078368 (MB=2755550), rps=200319 (MB/s=22924)
CPU %usr %nice %sys %iowait %irq %soft %idle
0 0.69 0.00 8.26 31.65 1.83 57.00 0.57
This series adds net infrastructure for memory providers configuring
the size and implements it for bnxt. It's an opt-in feature for drivers,
they should advertise support for the parameter in the qops and must check
if the hardware supports the given size. It's limited to memory providers
as it drastically simplifies implementation. It doesn't affect the fast
path zcrx uAPI, and the user exposed parameter is defined in zcrx terms,
which allows it to be flexible and adjusted in the future.
A liburing example can be found at [2]
full branch:
[1] https://github.com/isilence/linux.git zcrx/large-buffers-v8
Liburing example:
[2] https://github.com/isilence/liburing.git zcrx/rx-buf-len
* tag 'net-queue-rx-buf-len-v9' of https://github.com/isilence/linux:
io_uring/zcrx: document area chunking parameter
selftests: iou-zcrx: test large chunk sizes
eth: bnxt: support qcfg provided rx page size
eth: bnxt: adjust the fill level of agg queues with larger buffers
eth: bnxt: store rx buffer size per queue
net: pass queue rx page size from memory provider
net: add bare bone queue configs
net: reduce indent of struct netdev_queue_mgmt_ops members
net: memzero mp params when closing a queue
====================
Link: https://patch.msgid.link/
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This reverts commit 77b9c4a438, reversing
changes made to 4515ec4ad5:
931420a2fc ("selftests/net: Add netkit container tests")
ab771c938d ("selftests/net: Make NetDrvContEnv support queue leasing")
6be87fbb27 ("selftests/net: Add env for container based tests")
61d99ce3df ("selftests/net: Add bpf skb forwarding program")
920da36341 ("netkit: Add xsk support for af_xdp applications")
eef51113f8 ("netkit: Add netkit notifier to check for unregistering devices")
b5ef109d22 ("netkit: Implement rtnl_link_ops->alloc and ndo_queue_create")
b5c3fa4a0b ("netkit: Add single device mode for netkit")
0073d2fd67 ("xsk: Proxy pool management for leased queues")
1ecea95dd3 ("xsk: Extend xsk_rcv_check validation")
804bf334d0 ("net: Proxy netdev_queue_get_dma_dev for leased queues")
0caa9a8dde ("net: Proxy net_mp_{open,close}_rxq for leased queues")
ff8889ff91 ("net, ethtool: Disallow leased real rxqs to be resized")
9e2103f361 ("net: Add lease info to queue-get response")
31127dedde ("net: Implement netdev_nl_queue_create_doit")
a5546e18f7 ("net: Add queue-create operation")
The series will conflict with io_uring work, and the code needs more
polish.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
several netfilter compilation units rely on implicit includes
coming from nf_conntrack_proto_gre.h.
Clean this up and add the required dependencies where needed.
nf_conntrack.h requires net_generic() helper.
Place various gre/ppp/vlan includes to where they are needed.
Signed-off-by: Florian Westphal <fw@strlen.de>
After the optimization to only perform one GC per jiffy, a new problem
was introduced. If more than 8 new connections are tracked per jiffy the
list won't be cleaned up fast enough possibly reaching the limit
wrongly.
In order to prevent this issue, only skip the GC if it was already
triggered during the same jiffy and the increment is lower than the
clean up limit. In addition, increase the clean up limit to 64
connections to avoid triggering GC too often and do more effective GCs.
This has been tested using a HTTP server and several
performance tools while having nft_connlimit/xt_connlimit or OVS limit
configured.
Output of slowhttptest + OVS limit at 52000 connections:
slow HTTP test status on 340th second:
initializing: 0
pending: 432
connected: 51998
error: 0
closed: 0
service available: YES
Fixes: d265929930 ("netfilter: nf_conncount: reduce unnecessary GC")
Reported-by: Aleksandra Rukomoinikova <ARukomoinikova@k2.cloud>
Closes: https://lore.kernel.org/netfilter/b2064e7b-0776-4e14-adb6-c68080987471@k2.cloud/
Signed-off-by: Fernando Fernandez Mancera <fmancera@suse.de>
Signed-off-by: Florian Westphal <fw@strlen.de>
When a process in a container wants to setup a memory provider, it will
use the virtual netdev and a leased rxq, and call net_mp_{open,close}_rxq
to try and restart the queue. At this point, proxy the queue restart on
the real rxq in the physical netdev.
For memory providers (io_uring zero-copy rx and devmem), it causes the
real rxq in the physical netdev to be filled from a memory provider that
has DMA mapped memory from a process within a container.
Signed-off-by: David Wei <dw@davidwei.uk>
Co-developed-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260115082603.219152-6-daniel@iogearbox.net
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Populate nested lease info to the queue-get response that returns the
ifindex, queue id with type and optionally netns id if the device
resides in a different netns.
Example with ynl client:
# ip a
[...]
4: enp10s0f0np0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 xdp/id:24 qdisc mq state UP group default qlen 1000
link/ether e8:eb:d3:a3:43:f6 brd ff:ff:ff:ff:ff:ff
inet 10.0.0.2/24 scope global enp10s0f0np0
valid_lft forever preferred_lft forever
inet6 fe80::eaeb:d3ff:fea3:43f6/64 scope link proto kernel_ll
valid_lft forever preferred_lft forever
[...]
# ethtool -i enp10s0f0np0
driver: mlx5_core
[...]
# ./pyynl/cli.py \
--spec ~/netlink/specs/netdev.yaml \
--do queue-get \
--json '{"ifindex": 4, "id": 15, "type": "rx"}'
{'id': 15,
'ifindex': 4,
'lease': {'ifindex': 8, 'netns-id': 0, 'queue': {'id': 1, 'type': 'rx'}},
'napi-id': 8227,
'type': 'rx',
'xsk': {}}
# ip netns list
foo (id: 0)
# ip netns exec foo ip a
[...]
8: nk@NONE: <BROADCAST,MULTICAST,NOARP,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000
link/ether 00:00:00:00:00:00 brd ff:ff:ff:ff:ff:ff
inet6 fe80::200:ff:fe00:0/64 scope link proto kernel_ll
valid_lft forever preferred_lft forever
[...]
# ip netns exec foo ethtool -i nk
driver: netkit
[...]
# ip netns exec foo ls /sys/class/net/nk/queues/
rx-0 rx-1 tx-0
# ip netns exec foo ./pyynl/cli.py \
--spec ~/netlink/specs/netdev.yaml \
--do queue-get \
--json '{"ifindex": 8, "id": 1, "type": "rx"}'
{'id': 1, 'ifindex': 8, 'type': 'rx'}
Note that the caller of netdev_nl_queue_fill_one() holds the netdevice
lock. For the queue-get we do not lock both devices. When queues get
{un,}leased, both devices are locked, thus if __netif_get_rx_queue_peer()
returns true, the peer pointer points to a valid device. The netns-id
is fetched via peernet2id_alloc() similarly as done in OVS.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Co-developed-by: David Wei <dw@davidwei.uk>
Signed-off-by: David Wei <dw@davidwei.uk>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20260115082603.219152-4-daniel@iogearbox.net
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Implement netdev_nl_queue_create_doit which creates a new rx queue in a
virtual netdev and then leases it to a rx queue in a physical netdev.
Example with ynl client:
# ./pyynl/cli.py \
--spec ~/netlink/specs/netdev.yaml \
--do queue-create \
--json '{"ifindex": 8, "type": "rx", "lease": {"ifindex": 4, "queue": {"type": "rx", "id": 15}}}'
{'id': 1}
Note that the netdevice locking order is always from the virtual to
the physical device.
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Co-developed-by: David Wei <dw@davidwei.uk>
Signed-off-by: David Wei <dw@davidwei.uk>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Link: https://patch.msgid.link/20260115082603.219152-3-daniel@iogearbox.net
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When the AP has an advertised TID to Link Mapping (TTLM) it shall
include the element in the association response. As such, when this
element is present it needs to be used for the currently dormant links.
See Draft P802.11REVmf_D1.0 section 35.3.7.2.3 ("Negotiation of TTLM")
for the details. The flag is also not usable in case userspace wants to
specify a negotiated TTLM during association.
Note that for the link reconfiguration case, mac80211 did not use the
information. Draft P802.11REVmf_D1.0 states in section 35.3.6.4 ("Link
reconfiguration to the setup links) that we "shall operate with all the
TIDs mapped to the newly added links ..."
All this means that the flag is not needed. The implementation should
parse the information from the association response.
Signed-off-by: Benjamin Berg <benjamin.berg@intel.com>
Reviewed-by: Johannes Berg <johannes.berg@intel.com>
Reviewed-by: Ilan Peer <ilan.peer@intel.com>
Signed-off-by: Miri Korenblit <miriam.rachel.korenblit@intel.com>
Link: https://patch.msgid.link/20260118093904.754e057896a5.Ifd06f5ef839a93bfd54d0593dc932870f95f3242@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Group together following struct netns_sysctl_ipv6 fields:
- flowlabel_consistency
- auto_flowlabels
- flowlabel_state_ranges
After this patch, ip6_make_flowlabel() uses a single cache line to fetch
auto_flowlabels and flowlabel_state_ranges (instead of two before the patch).
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260115094141.3124990-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We (Paolo and I) noticed that in the sending path touching an extra
cacheline due to cq_cached_prod_lock will impact the performance. After
moving the lock from struct xsk_buff_pool to struct xsk_queue, the
performance is increased by ~5% which can be observed by xdpsock.
An alternative approach [1] can be using atomic_try_cmpxchg() to have the
same effect. But unfortunately I don't have evident performance numbers to
prove the atomic approach is better than the current patch. The advantage
is to save the contention time among multiple xsks sharing the same pool
while the disadvantage is losing good maintenance. The full discussion can
be found at the following link.
[1]: https://lore.kernel.org/all/20251128134601.54678-1-kerneljasonxing@gmail.com/
Suggested-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: Jason Xing <kernelxing@tencent.com>
Link: https://patch.msgid.link/20260104012125.44003-3-kerneljasonxing@gmail.com
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Allow memory providers to configure rx queues with a custom receive
page size. It's passed in struct pp_memory_provider_params, which is
copied into the queue, so it's preserved across queue restarts. Then,
it's propagated to the driver in a new queue config parameter.
Drivers should explicitly opt into using it by setting
QCFG_RX_PAGE_SIZE, in which case they should implement ndo_default_qcfg,
validate the size on queue restart and honour the current config in case
of a reset.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
We'll need to pass extra parameters when allocating a queue for memory
providers. Define a new structure for queue configurations, and pass it
to qapi callbacks. It's empty for now, actual parameters will be added
in following patches.
Configurations should persist across resets, and for that they're
default-initialised on device registration and stored in struct
netdev_rx_queue. We also add a new qapi callback for defaulting a given
config. It must be implemented if a driver wants to use queue configs
and is optional otherwise.
Suggested-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Trivial change, reduce the indent. I think the original is copied
from real NDOs. It's unnecessarily deep, makes passing struct args
problematic.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Mina Almasry <almasrymina@google.com>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
To enable the cake_mq qdisc to reuse code from the mq qdisc, export a
bunch of functions from sch_mq. Split common functionality out from some
functions so it can be composed with other code, and export other
functions wholesale. To discourage wanton reuse, put the symbols into a
new NET_SCHED_INTERNAL namespace, and a sch_priv.h header file.
No functional change intended.
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://patch.msgid.link/20260109-mq-cake-sub-qdisc-v8-1-8d613fece5d8@redhat.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
In blamed commit, I added a check against the temporary queue
built in __dev_xmit_skb(). Idea was to drop packets early,
before any spinlock was acquired.
if (unlikely(defer_count > READ_ONCE(q->limit))) {
kfree_skb_reason(skb, SKB_DROP_REASON_QDISC_DROP);
return NET_XMIT_DROP;
}
It turned out that HTB Qdisc has a zero q->limit.
HTB limits packets on a per-class basis.
Some of our tests became flaky.
Add a new sysctl : net.core.qdisc_max_burst to control
how many packets can be stored in the temporary lockless queue.
Also add a new QDISC_BURST_DROP drop reason to better diagnose
future issues.
Thanks Neal !
Fixes: 100dfa74ca ("net: dev_queue_xmit() llist adoption")
Reported-and-bisected-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Neal Cardwell <ncardwell@google.com>
Link: https://patch.msgid.link/20260107104159.3669285-1-edumazet@google.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
RTL8127ATF supports a SFP+ port for fiber modules (10GBASE-SR/LR/ER/ZR and
DAC). The list of supported modes was provided by Realtek. According to the
r8127 vendor driver also 1G modules are supported, but this needs some more
complexity in the driver, and only 10G mode has been tested so far.
Therefore mainline support will be limited to 10G for now.
The SFP port signals are hidden in the chip IP and driven by firmware.
Therefore mainline SFP support can't be used here.
This PHY driver is used by the RTL8127ATF support in r8169.
RTL8127ATF reports the same PHY ID as the TP version. Therefore use a dummy
PHY ID. This PHY driver is used by the RTL8127ATF support in r8169.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Link: https://patch.msgid.link/e3d55162-210a-4fab-9abf-99c6954eee10@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Johannes Berg says:
====================
First set of changes for the current -next cycle, of note:
- ath12k gets an overhaul to support multi-wiphy device
wiphy and pave the way for future device support in
the same driver (rather than splitting to ath13k)
- mac80211 gets some better iteration macros
* tag 'wireless-next-2026-01-12' of https://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next: (120 commits)
wifi: mac80211: remove width argument from ieee80211_parse_bitrates
wifi: mac80211_hwsim: remove NAN by default
wifi: mac80211: improve station iteration ergonomics
wifi: mac80211: improve interface iteration ergonomics
wifi: cfg80211: include S1G_NO_PRIMARY flag when sending channel
wifi: mac80211: unexport ieee80211_get_bssid()
wl1251: Replace strncpy with strscpy in wl1251_acx_fw_version
wifi: iwlegacy: 3945-rs: remove redundant pointer check in il3945_rs_tx_status() and il3945_rs_get_rate()
wifi: mac80211: don't send an unused argument to ieee80211_check_combinations
wifi: libertas: fix WARNING in usb_tx_block
wifi: mwifiex: Allocate dev name earlier for interface workqueue name
wifi: wlcore: sdio: Use pm_ptr instead of #ifdef CONFIG_PM
wifi: cfg80211: Fix use_for flag update on BSS refresh
wifi: brcmfmac: rename function that frees vif
wifi: brcmfmac: fix/add kernel-doc comments
wifi: mac80211: Update csa_finalize to use link_id
wifi: cfg80211: add cfg80211_stop_link() for per-link teardown
wifi: ath12k: Skip DP peer creation for scan vdev
wifi: ath12k: move firmware stats request outside of atomic context
wifi: ath12k: add the missing RCU lock in ath12k_dp_tx_free_txbuf()
...
====================
Link: https://patch.msgid.link/20260112185836.378736-3-johannes@sipsolutions.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Right now, the only way to iterate stations is to declare an
iterator function, possibly data structure to use, and pass all
that to the iteration helper function. This is annoying, and
there's really no inherent need for it.
Add a new for_each_station() macro that does the iteration in
a more ergonomic way. To avoid even more exported functions, do
the old ieee80211_iterate_stations_mtx() as an inline using the
new way, which may also let the compiler optimise it a bit more,
e.g. via inlining the iterator function.
Link: https://patch.msgid.link/20260108143431.d2b641f6f6af.I4470024f7404446052564b15bcf8b3f1ada33655@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Right now, the only way to iterate interfaces is to declare an
iterator function, possibly data structure to use, and pass all
that to the iteration helper function. This is annoying, and
there's really no inherent need for it, except it was easier to
implement with the iflist mutex, but that's not used much now.
Add a new for_each_interface() macro that does the iteration in
a more ergonomic way. To avoid even more exported functions, do
the old ieee80211_iterate_active_interfaces_mtx() as an inline
using the new way, which may also let the compiler optimise it
a bit more, e.g. via inlining the iterator function.
Also provide for_each_active_interface() for the common case of
just iterating active interfaces.
Link: https://patch.msgid.link/20260108143431.f2581e0c381a.Ie387227504c975c109c125b3c57f0bb3fdab2835@changeid
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Currently, whenever cfg80211_stop_iface() is called, the entire iface
is stopped. However, there could be a need in AP/P2P_GO mode, where
one would like to stop a single link in MLO operation instead of the
whole MLD interface.
Hence, introduce cfg80211_stop_link() to allow drivers to tear down
only a specified AP/P2P_GO link during MLO operation. Passing -1
preserves the existing behavior of stopping the whole interface. Make
cfg80211_stop_iface() call this function by passing -1 to keep the
default behavior the same, that is, to stop all links and use
cfg80211_stop_link() with the desired link_id for AP/P2P_GO mode, to
stop only that link.
This brings no behavioral change for single-link/non-MLO interfaces,
and enables drivers to stop an AP/P2P_GO link without disrupting other
links on the same interface.
Signed-off-by: Manish Dharanenthiran <manish.dharanenthiran@oss.qualcomm.com>
Link: https://patch.msgid.link/20251127-stop_link-v2-1-43745846c5fd@qti.qualcomm.com
[make cfg80211_stop_iface() inline]
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
The optional and required hints in the tcp_congestion_ops are information
for the user of this interface to signalize its importance when
implementing these functions.
However, cong_avoid comment incorrectly tells that it is required,
in reality congestion control must provide one of either cong_avoid or
cong_control.
In addition, min_tso_segs has not had any comment optional/required
hints. So mark it as optional since it is used only in BBR.
Co-developed-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Daniel Sedlak <daniel.sedlak@cdn77.com>
Link: https://patch.msgid.link/20260105115533.1151442-1-daniel.sedlak@cdn77.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Use DEFINE_RAW_FLEX() to avoid thousands of -Wflex-array-member-not-at-end
warnings.
Remove struct ip_options_data, and adjust the rest of the code so that
flexible-array member struct ip_options_rcu::opt.__data[] ends last
in struct icmp_bxm.
Compensate for this by using the DEFINE_RAW_FLEX() helper to define each
on-stack struct instance that contained struct ip_options_data as a member,
and to define struct ip_options_rcu with a fixed on-stack size for its
nested flexible-array member opt.__data[].
Also, add a couple of code comments to prevent people from adding members
to a struct after another member that contains a flexible array.
With these changes, fix 2600 warnings of the following type:
include/net/inet_sock.h:65:33: warning: structure containing a flexible array member is not at the end of another structure [-Wflex-array-member-not-at-end]
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Link: https://patch.msgid.link/aVteBadWA6AbTp7X@kspp
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Problem description
-------------------
DSA has a mumbo-jumbo of reference handling of the conduit net device
and its kobject which, sadly, is just wrong and doesn't make sense.
There are two distinct problems.
1. The OF path, which uses of_find_net_device_by_node(), never releases
the elevated refcount on the conduit's kobject. Nominally, the OF and
non-OF paths should result in objects having identical reference
counts taken, and it is already suspicious that
dsa_dev_to_net_device() has a put_device() call which is missing in
dsa_port_parse_of(), but we can actually even verify that an issue
exists. With CONFIG_DEBUG_KOBJECT_RELEASE=y, if we run this command
"before" and "after" applying this patch:
(unbind the conduit driver for net device eno2)
echo 0000:00:00.2 > /sys/bus/pci/drivers/fsl_enetc/unbind
we see these lines in the output diff which appear only with the patch
applied:
kobject: 'eno2' (ffff002009a3a6b8): kobject_release, parent 0000000000000000 (delayed 1000)
kobject: '109' (ffff0020099d59a0): kobject_release, parent 0000000000000000 (delayed 1000)
2. After we find the conduit interface one way (OF) or another (non-OF),
it can get unregistered at any time, and DSA remains with a long-lived,
but in this case stale, cpu_dp->conduit pointer. Holding the net
device's underlying kobject isn't actually of much help, it just
prevents it from being freed (but we never need that kobject
directly). What helps us to prevent the net device from being
unregistered is the parallel netdev reference mechanism (dev_hold()
and dev_put()).
Actually we actually use that netdev tracker mechanism implicitly on
user ports since commit 2f1e8ea726 ("net: dsa: link interfaces with
the DSA master to get rid of lockdep warnings"), via netdev_upper_dev_link().
But time still passes at DSA switch probe time between the initial
of_find_net_device_by_node() code and the user port creation time, time
during which the conduit could unregister itself and DSA wouldn't know
about it.
So we have to run of_find_net_device_by_node() under rtnl_lock() to
prevent that from happening, and release the lock only with the netdev
tracker having acquired the reference.
Do we need to keep the reference until dsa_unregister_switch() /
dsa_switch_shutdown()?
1: Maybe yes. A switch device will still be registered even if all user
ports failed to probe, see commit 86f8b1c01a ("net: dsa: Do not
make user port errors fatal"), and the cpu_dp->conduit pointers
remain valid. I haven't audited all call paths to see whether they
will actually use the conduit in lack of any user port, but if they
do, it seems safer to not rely on user ports for that reference.
2. Definitely yes. We support changing the conduit which a user port is
associated to, and we can get into a situation where we've moved all
user ports away from a conduit, thus no longer hold any reference to
it via the net device tracker. But we shouldn't let it go nonetheless
- see the next change in relation to dsa_tree_find_first_conduit()
and LAG conduits which disappear.
We have to be prepared to return to the physical conduit, so the CPU
port must explicitly keep another reference to it. This is also to
say: the user ports and their CPU ports may not always keep a
reference to the same conduit net device, and both are needed.
As for the conduit's kobject for the /sys/class/net/ entry, we don't
care about it, we can release it as soon as we hold the net device
object itself.
History and blame attribution
-----------------------------
The code has been refactored so many times, it is very difficult to
follow and properly attribute a blame, but I'll try to make a short
history which I hope to be correct.
We have two distinct probing paths:
- one for OF, introduced in 2016 in commit 83c0afaec7 ("net: dsa: Add
new binding implementation")
- one for non-OF, introduced in 2017 in commit 71e0bbde0d ("net: dsa:
Add support for platform data")
These are both complete rewrites of the original probing paths (which
used struct dsa_switch_driver and other weird stuff, instead of regular
devices on their respective buses for register access, like MDIO, SPI,
I2C etc):
- one for OF, introduced in 2013 in commit 5e95329b70 ("dsa: add
device tree bindings to register DSA switches")
- one for non-OF, introduced in 2008 in commit 91da11f870 ("net:
Distributed Switch Architecture protocol support")
except for tiny bits and pieces like dsa_dev_to_net_device() which were
seemingly carried over since the original commit, and used to this day.
The point is that the original probing paths received a fix in 2015 in
the form of commit 679fb46c57 ("net: dsa: Add missing master netdev
dev_put() calls"), but the fix never made it into the "new" (dsa2)
probing paths that can still be traced to today, and the fixed probing
path was later deleted in 2019 in commit 93e86b3bc8 ("net: dsa: Remove
legacy probing support").
That is to say, the new probing paths were never quite correct in this
area.
The existence of the legacy probing support which was deleted in 2019
explains why dsa_dev_to_net_device() returns a conduit with elevated
refcount (because it was supposed to be released during
dsa_remove_dst()). After the removal of the legacy code, the only user
of dsa_dev_to_net_device() calls dev_put(conduit) immediately after this
function returns. This pattern makes no sense today, and can only be
interpreted historically to understand why dev_hold() was there in the
first place.
Change details
--------------
Today we have a better netdev tracking infrastructure which we should
use. Logically netdev_hold() belongs in common code
(dsa_port_parse_cpu(), where dp->conduit is assigned), but there is a
tradeoff to be made with the rtnl_lock() section which would become a
bit too long if we did that - dsa_port_parse_cpu() also calls
request_module(). So we duplicate a bit of logic in order for the
callers of dsa_port_parse_cpu() to be the ones responsible of holding
the conduit reference and releasing it on error. This shortens the
rtnl_lock() section significantly.
In the dsa_switch_probe() error path, dsa_switch_release_ports() will be
called in a number of situations, one being where dsa_port_parse_cpu()
maybe didn't get the chance to run at all (a different port failed
earlier, etc). So we have to test for the conduit being NULL prior to
calling netdev_put().
There have still been so many transformations to the code since the
blamed commits (rename master -> conduit, commit 0650bf52b3 ("net:
dsa: be compatible with masters which unregister on shutdown")), that it
only makes sense to fix the code using the best methods available today
and see how it can be backported to stable later. I suspect the fix
cannot even be backported to kernels which lack dsa_switch_shutdown(),
and I suspect this is also maybe why the long-lived conduit reference
didn't make it into the new DSA probing paths at the time (problems
during shutdown).
Because dsa_dev_to_net_device() has a single call site and has to be
changed anyway, the logic was just absorbed into the non-OF
dsa_port_parse().
Tested on the ocelot/felix switch and on dsa_loop, both on the NXP
LS1028A with CONFIG_DEBUG_KOBJECT_RELEASE=y.
Reported-by: Ma Ke <make24@iscas.ac.cn>
Closes: https://lore.kernel.org/netdev/20251214131204.4684-1-make24@iscas.ac.cn/
Fixes: 83c0afaec7 ("net: dsa: Add new binding implementation")
Fixes: 71e0bbde0d ("net: dsa: Add support for platform data")
Reviewed-by: Jonas Gorski <jonas.gorski@gmail.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251215150236.3931670-1-vladimir.oltean@nxp.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
