page_pool_init() returns E2BIG when the page_pool size goes above 32K
pages. As some drivers are configuring the page_pool size according to
the MTU and ring size, there are cases where this limit is exceeded and
the queue creation fails.
The page_pool size doesn't have to cover a full queue, especially for
larger ring size. So clamp the size instead of returning an error. Do
this in the core to avoid having each driver do the clamping.
The current limit was deemed to high [1] so it was reduced to 16K to avoid
page waste.
[1] https://lore.kernel.org/all/1758532715-820422-3-git-send-email-tariqt@nvidia.com/
Signed-off-by: Dragos Tatulea <dtatulea@nvidia.com>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20250926131605.2276734-2-dtatulea@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Introduce a new document, flow_control.rst, to provide a comprehensive
guide on Ethernet Flow Control in Linux. The guide explains how flow
control works, how autonegotiation resolves pause capabilities, and how
to configure it using ethtool and Netlink.
In parallel, document the pause and pause-stat attributes in the
ethtool.yaml netlink spec. This enables the ynl tool to generate
kernel-doc comments for the corresponding enums in the UAPI header,
making the C interface self-documenting.
Finally, replace the legacy flow control section in phy.rst with a
reference to the new document and add pointers in the relevant C source
files.
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
Link: https://patch.msgid.link/20250924120241.724850-1-o.rempel@pengutronix.de
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Ivan Vecera says:
====================
dpll: add phase offset averaging factor
For some hardware, the phase shift may result from averaging previous values
and the newly measured value. In this case, the averaging is controlled by
a configurable averaging factor.
Add new device level attribute phase-offset-avg-factor, appropriate
callbacks and implement them in zl3073x driver.
====================
Link: https://patch.msgid.link/20250927084912.2343597-1-ivecera@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The DPLL phase measurement block uses an exponential moving average with
a configurable averaging factor. Measurements are taken at approximately
40 Hz or at the reference frequency, whichever is lower.
Currently, factor=2 is used to prioritize fast response for dynamic
phase changes. For applications needing a stable, precise average phase
offset where rapid changes are unlikely, a higher factor is recommended.
Implement the .phase_offset_avg_factor_get/set callbacks to allow a user
to adjust this factor.
Signed-off-by: Ivan Vecera <ivecera@redhat.com>
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://patch.msgid.link/20250927084912.2343597-4-ivecera@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add new callback operations for a dpll device:
- phase_offset_avg_factor_get(...) - to obtain current phase offset
averaging factor from dpll device,
- phase_offset_avg_factor_set(...) - to set phase offset averaging factor
Obtain the factor value using the get callback and provide it to the user
if the device driver implement this callback. Execute the set callback upon
user requests, if the driver implement it.
Signed-off-by: Ivan Vecera <ivecera@redhat.com>
v2:
* do not require 'set' callback to retrieve current value
* always call 'set' callback regardless of current value
Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Link: https://patch.msgid.link/20250927084912.2343597-3-ivecera@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tariq Toukan says:
====================
mlx5-next updates 2025-09-28
* tag 'mlx5-next-lag' of git://git.kernel.org/pub/scm/linux/kernel/git/mellanox/linux:
net/mlx5: IFC add balance ID and LAG per MP group bits
net/mlx5: Add IFC bit for TIR/SQ order capability
====================
Link: https://patch.msgid.link/1759093989-841873-1-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Kuniyuki Iwashima says:
====================
selftest: packetdrill: Import TFO server tests.
The series imports 15 TFO server tests from google/packetdrill and
adds 2 more tests.
The repository has two versions of tests for most scenarios; one uses
the non-experimental option (34), and the other uses the experimental
option (255) with 0xF989.
Basically, we only import the non-experimental version of tests, and
for the experimental option, tcp_fastopen_server_experimental_option.pkt
is added.
The following tests are not (yet) imported:
* icmp-baseline.pkt
* simple1.pkt / simple2.pkt / simple3.pkt
The former is completely covered by icmp-before-accept.pkt.
The later's delta is the src/dst IP pair to generate a different
cookie, but supporting dualstack requires churn in ksft_runner.sh,
so defered to future series. Also, sockopt-fastopen-key.pkt covers
the same function.
The following tests have the experimental version only, so converted
to the non-experimental option:
* client-ack-dropped-then-recovery-ms-timestamps.pkt
* sockopt-fastopen-key.pkt
For the imported tests, these common changes are applied.
* Add SPDX header
* Adjust path to default.sh
* Adjust sysctl w/ set_sysctls.py
* Use TFO_COOKIE instead of a raw hex value
* Use SOCK_NONBLOCK for socket() not to block accept()
* Add assertions for TCP state if commented
* Remove unnecessary delay (e.g. +0.1 setsockopt(SO_REUSEADDR), etc)
With this series, except for simple{1,2,3}.pkt, we can remove TFO server
tests in google/packetdrill.
====================
Link: https://patch.msgid.link/20250927213022.1850048-1-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
sockopt-fastopen-key.pkt does not have the non-experimental
version, so the Experimental version is converted, FOEXP -> FO.
The test sets net.ipv4.tcp_fastopen_key=0-0-0-0 and instead
sets another key via setsockopt(TCP_FASTOPEN_KEY).
The first listener generates a valid cookie in response to TFO
option without cookie, and the second listner creates a TFO socket
using the valid cookie.
TCP_FASTOPEN_KEY is adjusted to use the common key in default.sh
so that we can use TFO_COOKIE and support dualstack. Similarly,
TFO_COOKIE_ZERO for the 0-0-0-0 key is defined.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250927213022.1850048-13-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This imports the non-experimental version of opt34/*-trigger-rst.pkt.
| accept() | SYN data |
-----------------------------------+----------+----------+
listener-closed-trigger-rst.pkt | no | unread |
unread-data-closed-trigger-rst.pkt | yes | unread |
Both files test that close()ing a SYN_RECV socket with unread SYN data
triggers RST.
The files are renamed to have the common prefix, trigger-rst.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250927213022.1850048-11-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This imports the non-experimental version of opt34/reset-*.pkt.
| Child | RST | sk_err |
---------------------------------+---------+-------------------------------+---------+
reset-after-accept.pkt | TFO | after accept(), SYN_RECV | read() |
reset-close-with-unread-data.pkt | TFO | after accept(), SYN_RECV | write() |
reset-before-accept.pkt | TFO | before accept(), SYN_RECV | read() |
reset-non-tfo-socket.pkt | non-TFO | before accept(), ESTABLISHED | write() |
The first 3 files test scenarios where a SYN_RECV socket receives RST
before/after accept() and data in SYN must be read() without error,
but the following read() or fist write() will return ECONNRESET.
The last test is similar but with non-TFO socket.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250927213022.1850048-10-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This imports the non-experimental version of icmp-before-accept.pkt.
This file tests the scenario where an ICMP unreachable packet for a
not-yet-accept()ed socket changes its state to TCP_CLOSE, but the
SYN data must be read without error, and the following read() returns
EHOSTUNREACH.
Note that this test support only IPv4 as icmp is used.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250927213022.1850048-9-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The only difference between non-experimental vs experimental TFO
option handling is SYN+ACK generation.
When tcp_parse_fastopen_option() parses a TFO option, it sets
tcp_fastopen_cookie.exp to false if the option number is 34,
and true if 255.
The value is carried to tcp_options_write() to generate a TFO option
with the same option number.
Other than that, all the TFO handling is the same and the kernel must
generate the same cookie regardless of the option number.
Let's add a test for the handling so that we can consolidate
fastopen/server/ tests and fastopen/server/opt34 tests.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250927213022.1850048-7-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This imports basic TFO server tests from google/packetdrill.
The repository has two versions of tests for most scenarios; one uses
the non-experimental option (34), and the other uses the experimental
option (255) with 0xF989.
This only imports the following tests of the non-experimental version
placed in [0]. I will add a specific test for the experimental option
handling later.
| TFO | Cookie | Payload |
---------------------------+-----+--------+---------+
basic-rw.pkt | yes | yes | yes |
basic-zero-payload.pkt | yes | yes | no |
basic-cookie-not-reqd.pkt | yes | no | yes |
basic-non-tfo-listener.pkt | no | yes | yes |
pure-syn-data.pkt | yes | no | yes |
The original pure-syn-data.pkt missed setsockopt(TCP_FASTOPEN) and did
not test TFO server in some scenarios unintentionally, so setsockopt()
is added where needed. In addition, non-TFO scenario is stripped as
it is covered by basic-non-tfo-listener.pkt. Also, I added basic- prefix.
Link: https://github.com/google/packetdrill/tree/bfc96251310f/gtests/net/tcp/fastopen/server/opt34 #[0]
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250927213022.1850048-5-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
TCP Fast Open cookie is generated in __tcp_fastopen_cookie_gen_cipher().
The cookie value is generated from src/dst IPs and a key configured by
setsockopt(TCP_FASTOPEN_KEY) or net.ipv4.tcp_fastopen_key.
The default.sh sets net.ipv4.tcp_fastopen_key, and the original packetdrill
defines the corresponding cookie as TFO_COOKIE in run_all.py. [0]
Then, each test does not need to care about the value, and we can easily
update TFO_COOKIE in case __tcp_fastopen_cookie_gen_cipher() changes the
algorithm.
However, some tests use the bare hex value for specific IPv4 addresses
and do not support IPv6.
Let's define the same TFO_COOKIE in ksft_runner.sh.
We will replace such bare hex values with TFO_COOKIE except for a single
test for setsockopt(TCP_FASTOPEN_KEY).
Link: https://github.com/google/packetdrill/blob/7230b3990f94/gtests/net/packetdrill/run_all.py#L65 #[0]
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250927213022.1850048-4-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
To enable TCP Fast Open on a server, net.ipv4.tcp_fastopen must
have 0x2 (TFO_SERVER_ENABLE), and we need to do either
1. Call setsockopt(TCP_FASTOPEN) for the socket
2. Set 0x400 (TFO_SERVER_WO_SOCKOPT1) additionally to net.ipv4.tcp_fastopen
The default.sh sets 0x70403 so that each test does not need setsockopt().
(0x1 is TFO_CLIENT_ENABLE, and 0x70000 is ...???)
However, some tests overwrite net.ipv4.tcp_fastopen without
TFO_SERVER_WO_SOCKOPT1 and forgot setsockopt(TCP_FASTOPEN).
For example, pure-syn-data.pkt [0] tests non-TFO servers unintentionally,
except in the first scenario.
To prevent such an accident, let's require explicit setsockopt().
TFO_CLIENT_ENABLE is necessary for
tcp_syscall_bad_arg_fastopen-invalid-buf-ptr.pkt.
Link: https://github.com/google/packetdrill/blob/bfc96251310f/gtests/net/tcp/fastopen/server/opt34/pure-syn-data.pkt #[0]
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250927213022.1850048-3-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The cited commit forgot to update the ktap_set_plan call.
ktap_set_plan sets the number of tests (KSFT_NUM_TESTS), which must
match the number of executed tests (KTAP_CNT_PASS + KTAP_CNT_SKIP +
KTAP_CNT_XFAIL) in ktap_finished.
Otherwise, the selftest exit()s with 1.
Let's adjust KSFT_NUM_TESTS based on supported protocols.
While at it, misalignment is fixed up.
Fixes: a5c10aa3d1 ("selftests/net: packetdrill: Support single protocol test.")
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20250927213022.1850048-2-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Drop the open-coded register polling routines.
Use readl_poll_timeout_atomic() in atomic state.
Also adjust the delay time to 10us which seems more reasonable.
Tested on NXP i.MX8MP and ROCKCHIP RK3588 boards,
the break condition was met right after the first polling,
no delay involved at all.
So the 10us delay should be long enough for most cases.
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Furong Xu <0x1207@gmail.com>
Link: https://patch.msgid.link/20250927081036.10611-1-0x1207@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Matthieu Baerts says:
====================
mptcp: receive path improvement
This series includes several changes to the MPTCP RX path. The main
goals are improving the RX performances, and increase the long term
maintainability.
Some changes reflects recent(ish) improvements introduced in the TCP
stack: patch 1, 2 and 3 are the MPTCP counter part of SKB deferral free
and auto-tuning improvements. Note that patch 3 could possibly fix
additional issues, and overall such patch should protect from similar
issues to arise in the future.
Patches 4-7 are aimed at introducing the socket backlog usage which will
be done in a later series to process the packets received by the
different subflows while the msk socket is owned.
Patch 8 is not related to the RX path, but it contains additional tests
for new features recently introduced in net-next.
====================
Link: https://patch.msgid.link/20250927-net-next-mptcp-rcv-path-imp-v1-0-5da266aa9c1a@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Here are a few sub-tests for mptcp_join.sh, validating the new 'laminar'
endpoint type.
In a setup where subflows created using the routing rules would be
rejected by the listener, and where the latter announces one IP address,
some cases are verified:
- Without any 'laminar' endpoints: no new subflows are created.
- With one 'laminar' endpoint: a second subflow is created.
- With multiple 'laminar' endpoints: 2 IPv4 subflows are created.
- With one 'laminar' endpoint, but the server announcing a second IP
address, only one subflow is created.
- With one 'laminar' + 'subflow' endpoint, the same endpoint is only
used once.
Reviewed-by: Mat Martineau <martineau@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://patch.msgid.link/20250927-net-next-mptcp-rcv-path-imp-v1-8-5da266aa9c1a@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Some applications are stuck to the 20th century and still use
small SO_RCVBUF values.
After the blamed commit, we can drop packets especially
when using LRO/hw-gro enabled NIC and small MSS (1500) values.
LRO/hw-gro NIC pack multiple segments into pages, allowing
tp->scaling_ratio to be set to a high value.
Whenever the receive queue gets full, we can receive a small packet
filling RWIN, but with a high skb->truesize, because most NIC use 4K page
plus sk_buff metadata even when receiving less than 1500 bytes of payload.
Even if we refine how tp->scaling_ratio is estimated,
we could have an issue at the start of the flow, because
the first round of packets (IW10) will be sent based on
the initial tp->scaling_ratio (1/2)
Relax tcp_can_ingest() to use skb->len instead of skb->truesize,
allowing the peer to use final RWIN, assuming a 'perfect'
scaling_ratio of 1.
Fixes: 1d2fbaad7c ("tcp: stronger sk_rcvbuf checks")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250927092827.2707901-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Luiz Augusto von Dentz says:
====================
bluetooth-next pull request for net-next:
core:
- MAINTAINERS: add a sub-entry for the Qualcomm bluetooth driver
- Avoid a couple dozen -Wflex-array-member-not-at-end warnings
- bcsp: receive data only if registered
- HCI: Fix using LE/ACL buffers for ISO packets
- hci_core: Detect if an ISO link has stalled
- ISO: Don't initiate CIS connections if there are no buffers
- ISO: Use sk_sndtimeo as conn_timeout
drivers:
- btusb: Check for unexpected bytes when defragmenting HCI frames
- btusb: Add new VID/PID 13d3/3627 for MT7925
- btusb: Add new VID/PID 13d3/3633 for MT7922
- btusb: Add USB ID 2001:332a for D-Link AX9U rev. A1
- btintel: Add support for BlazarIW core
- btintel_pcie: Add support for _suspend() / _resume()
- btintel_pcie: Define hdev->wakeup() callback
- btintel_pcie: Add Bluetooth core/platform as comments
- btintel_pcie: Add id of Scorpious, Panther Lake-H484
- btintel_pcie: Refactor Device Coredump
* tag 'for-net-next-2025-09-27' of git://git.kernel.org/pub/scm/linux/kernel/git/bluetooth/bluetooth-next: (30 commits)
Bluetooth: Avoid a couple dozen -Wflex-array-member-not-at-end warnings
Bluetooth: hci_sync: Fix using random address for BIG/PA advertisements
Bluetooth: ISO: don't leak skb in ISO_CONT RX
Bluetooth: ISO: free rx_skb if not consumed
Bluetooth: ISO: Fix possible UAF on iso_conn_free
Bluetooth: SCO: Fix UAF on sco_conn_free
Bluetooth: bcsp: receive data only if registered
Bluetooth: btusb: Add new VID/PID 13d3/3633 for MT7922
Bluetooth: btusb: Add new VID/PID 13d3/3627 for MT7925
Bluetooth: remove duplicate h4_recv_buf() in header
Bluetooth: btusb: Check for unexpected bytes when defragmenting HCI frames
Bluetooth: hci_core: Print information of hcon on hci_low_sent
Bluetooth: hci_core: Print number of packets in conn->data_q
Bluetooth: Add function and line information to bt_dbg
Bluetooth: MGMT: Fix not exposing debug UUID on MGMT_OP_READ_EXP_FEATURES_INFO
Bluetooth: hci_core: Detect if an ISO link has stalled
Bluetooth: ISO: Use sk_sndtimeo as conn_timeout
Bluetooth: HCI: Fix using LE/ACL buffers for ISO packets
Bluetooth: ISO: Don't initiate CIS connections if there are no buffers
MAINTAINERS: add a sub-entry for the Qualcomm bluetooth driver
...
====================
Link: https://patch.msgid.link/20250927154616.1032839-1-luiz.dentz@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
__ptr_ring_zero_tail currently does the - 1 operation twice:
- during initialization of head
- at each loop iteration
Let's just do it in one place, all we need to do
is adjust the loop condition. this is better:
- a slightly clearer logic with less duplication
- uses prefix -- we don't need to save the old value
- one less - 1 operation - for example, when ring is empty
we now don't do - 1 at all, existing code does it once
Text size shrinks from 15081 to 15050 bytes.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/bcd630c7edc628e20d4f8e037341f26c90ab4365.1758976026.git.mst@redhat.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
For global RSS and multiple RSS scheme, the RSS type fields are defined
identically in the registers. So they can be defined as the macros
WX_RSS_FIELD_* to cleanup the codes. And to prepare for the RXFH support
in the next patch, move the rss_field to struct wx.
Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com>
Link: https://patch.msgid.link/20250926023843.34340-3-jiawenwu@trustnetic.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Convert the get_user() and __put_user() code to the
fast masked_user_access_begin()/unsafe_{get|put}_user()
variant.
This patch increases the performance of an UDP recvfrom()
receiver (netserver) on 120 bytes messages by 7 %
on an AMD EPYC 7B12 64-Core Processor platform.
Presence of audit_sockaddr() makes difficult
to avoid the stac/clac pair in the copy_to_user() call,
this is left for a future patch.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Link: https://patch.msgid.link/20250925230929.3727873-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Drop those frames causing Head-of-Line Blocking due to Scheduling
(HLBS) error to avoid HLBS interrupt flooding and netdev watchdog
timeouts due to blocked packets. Tx queues can be configured to drop
those blocked packets by setting Drop Frames causing Scheduling Error
(DFBS) bit of EST_CONTROL register.
Also, add per queue HLBS drop count.
Signed-off-by: Rohan G Thomas <rohan.g.thomas@altera.com>
Reviewed-by: Matthew Gerlach <matthew.gerlach@altera.com>
Reviewed-by: Furong Xu <0x1207@gmail.com>
Link: https://patch.msgid.link/20250925-hlbs_2-v3-1-3b39472776c2@altera.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Heiner Kallweit
Dragos Tatulea
Dmitry Antipov
Qingfang Deng
Oleksij Rempel
Jakub Kicinski
Ivan Vecera
Russell King (Oracle)
Kuniyuki Iwashima
Alok Tiwari
Eric Dumazet
Furong Xu
Matthieu Baerts (NGI0)
Paolo Abeni
Michael S. Tsirkin
Jiawen Wu
Rohan G Thomas