The new stats calculations add several additional calls to
macb/gem_update_stats() and accesses to bp->hw_stats. These are
protected by a spinlock since commit fa52f15c74 ("net: cadence: macb:
Synchronize stats calculations"), which was applied in parallel. Add
some locking now that the net has been merged into net-next.
Fixes: f6af690a29 ("net: cadence: macb: Report standard stats")
Signed-off-by: Sean Anderson <sean.anderson@linux.dev>
Link: https://patch.msgid.link/20250303231832.1648274-1-sean.anderson@linux.dev
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Eric Dumazet says:
====================
tcp: scale connect() under pressure
Adoption of bhash2 in linux-6.1 made some operations almost twice
more expensive, because of additional locks.
This series adds RCU in __inet_hash_connect() to help the
case where many attempts need to be made before finding
an available 4-tuple.
This brings a ~200 % improvement in this experiment:
Server:
ulimit -n 40000; neper/tcp_crr -T 200 -F 30000 -6 --nolog
Client:
ulimit -n 40000; neper/tcp_crr -T 200 -F 30000 -6 --nolog -c -H server
Before series:
utime_start=0.288582
utime_end=1.548707
stime_start=20.637138
stime_end=2002.489845
num_transactions=484453
latency_min=0.156279245
latency_max=20.922042756
latency_mean=1.546521274
latency_stddev=3.936005194
num_samples=312537
throughput=47426.00
perf top on the client:
49.54% [kernel] [k] _raw_spin_lock
25.87% [kernel] [k] _raw_spin_lock_bh
5.97% [kernel] [k] queued_spin_lock_slowpath
5.67% [kernel] [k] __inet_hash_connect
3.53% [kernel] [k] __inet6_check_established
3.48% [kernel] [k] inet6_ehashfn
0.64% [kernel] [k] rcu_all_qs
After this series:
utime_start=0.271607
utime_end=3.847111
stime_start=18.407684
stime_end=1997.485557
num_transactions=1350742
latency_min=0.014131929
latency_max=17.895073144
latency_mean=0.505675853 # Nice reduction of latency metrics
latency_stddev=2.125164772
num_samples=307884
throughput=139866.80 # 194 % increase
perf top on client:
56.86% [kernel] [k] __inet6_check_established
17.96% [kernel] [k] __inet_hash_connect
13.88% [kernel] [k] inet6_ehashfn
2.52% [kernel] [k] rcu_all_qs
2.01% [kernel] [k] __cond_resched
0.41% [kernel] [k] _raw_spin_lock
====================
Link: https://patch.msgid.link/20250302124237.3913746-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When __inet_hash_connect() has to try many 4-tuples before
finding an available one, we see a high spinlock cost from
the many spin_lock_bh(&head->lock) performed in its loop.
This patch adds an RCU lookup to avoid the spinlock cost.
check_established() gets a new @rcu_lookup argument.
First reason is to not make any changes while head->lock
is not held.
Second reason is to not make this RCU lookup a second time
after the spinlock has been acquired.
Tested:
Server:
ulimit -n 40000; neper/tcp_crr -T 200 -F 30000 -6 --nolog
Client:
ulimit -n 40000; neper/tcp_crr -T 200 -F 30000 -6 --nolog -c -H server
Before series:
utime_start=0.288582
utime_end=1.548707
stime_start=20.637138
stime_end=2002.489845
num_transactions=484453
latency_min=0.156279245
latency_max=20.922042756
latency_mean=1.546521274
latency_stddev=3.936005194
num_samples=312537
throughput=47426.00
perf top on the client:
49.54% [kernel] [k] _raw_spin_lock
25.87% [kernel] [k] _raw_spin_lock_bh
5.97% [kernel] [k] queued_spin_lock_slowpath
5.67% [kernel] [k] __inet_hash_connect
3.53% [kernel] [k] __inet6_check_established
3.48% [kernel] [k] inet6_ehashfn
0.64% [kernel] [k] rcu_all_qs
After this series:
utime_start=0.271607
utime_end=3.847111
stime_start=18.407684
stime_end=1997.485557
num_transactions=1350742
latency_min=0.014131929
latency_max=17.895073144
latency_mean=0.505675853 # Nice reduction of latency metrics
latency_stddev=2.125164772
num_samples=307884
throughput=139866.80 # 190 % increase
perf top on client:
56.86% [kernel] [k] __inet6_check_established
17.96% [kernel] [k] __inet_hash_connect
13.88% [kernel] [k] inet6_ehashfn
2.52% [kernel] [k] rcu_all_qs
2.01% [kernel] [k] __cond_resched
0.41% [kernel] [k] _raw_spin_lock
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jason Xing <kerneljasonxing@gmail.com>
Tested-by: Jason Xing <kerneljasonxing@gmail.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://patch.msgid.link/20250302124237.3913746-5-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When __inet_hash_connect() has to try many 4-tuples before
finding an available one, we see a high spinlock cost from
__inet_check_established() and/or __inet6_check_established().
This patch adds an RCU lookup to avoid the spinlock
acquisition when the 4-tuple is found in the hash table.
Note that there are still spin_lock_bh() calls in
__inet_hash_connect() to protect inet_bind_hashbucket,
this will be fixed later in this series.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jason Xing <kerneljasonxing@gmail.com>
Tested-by: Jason Xing <kerneljasonxing@gmail.com>
Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>
Link: https://patch.msgid.link/20250302124237.3913746-2-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The address of a data structure member was determined before
a corresponding null pointer check in the implementation of
the function “tipc_link_tnl_prepare”.
Thus avoid the risk for undefined behaviour by moving the definition
for the local variable “fdefq” into an if branch at the end.
This issue was detected by using the Coccinelle software.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Link: https://patch.msgid.link/08fe8fc3-19c3-4324-8719-0ee74b0f32c9@web.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit 29b036be1b ("selftests: drv-net: test XDP, HDS auto and
the ioctl path") added a new test case in the net tree, now that
this code has made its way to net-next convert it to use the env.rpath()
helper instead of manually computing the relative path.
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Link: https://patch.msgid.link/20250228212956.25399-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
For PPPoE, PPTP, and PPPoL2TP, the start_xmit() function directly
forwards packets to the underlying network stack and never returns
anything other than 1. So these interfaces do not require a qdisc,
and the IFF_NO_QUEUE flag should be set.
Introduces a direct_xmit flag in struct ppp_channel to indicate when
IFF_NO_QUEUE should be applied. The flag is set in ppp_connect_channel()
for relevant protocols.
While at it, remove the usused latency member from struct ppp_channel.
Signed-off-by: Qingfang Deng <dqfext@gmail.com>
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Link: https://patch.msgid.link/20250301135517.695809-1-dqfext@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Lee Trager says:
====================
eth: fbnic: Cleanup macros and string function
We have received some feedback that the macros we use for reading FW mailbox
attributes are too large in scope and confusing to understanding. Additionally
the string function did not provide errors allowing it to silently succeed.
This patch set fixes theses issues.
====================
Link: https://patch.msgid.link/20250228191935.3953712-1-lee@trager.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Replace the firmware field macros with new macros which follow typical
kernel standards. No variables are required to be predefined for use and
results are now returned. These macros are prefixed with fta or fbnic
TLV attribute.
Signed-off-by: Lee Trager <lee@trager.us>
Link: https://patch.msgid.link/20250228191935.3953712-4-lee@trager.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Allow fbnic_tlv_attr_get_string() to return an error code. In the event the
source mailbox attribute is missing return -EINVAL. Like nla_strscpy() return
-E2BIG when the source string is larger than the destination string. In this
case the amount of data copied is equal to dstsize.
Signed-off-by: Lee Trager <lee@trager.us>
Link: https://patch.msgid.link/20250228191935.3953712-3-lee@trager.us
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When this binding was originally written, all known TSEC Ethernet
controllers had a Ten-Bit Interface (TBI). However, some datasheets such
as for the MPC8315E suggest that this is not universally true:
The eTSECs do not support TBI, GMII, and FIFO operating modes, so all
references to these interfaces and features should be ignored for this
device.
Acked-by: Rob Herring (Arm) <robh@kernel.org>
Signed-off-by: J. Neuschäfer <j.ne@posteo.net>
Link: https://patch.msgid.link/20250228-gianfar-yaml-v2-2-6beeefbd4818@posteo.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Andrei Botila says:
====================
net: phy: nxp-c45-tja11xx: add support for TJA1121
This patch series adds .match_phy_device for the existing TJAs
to differentiate between TJA1103/TJA1104 and TJA1120/TJA1121.
TJA1103 and TJA1104 share the same PHY_ID but TJA1104 has MACsec
capabilities while TJA1103 doesn't.
Also add support for TJA1121 which is based on TJA1120 hardware
with additional MACsec IP.
====================
Link: https://patch.msgid.link/20250228154320.2979000-1-andrei.botila@oss.nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Matthieu Baerts says:
====================
mptcp: improve code coverage and small optimisations
This small series have various unrelated patches:
- Patch 1 and 2: improve code coverage by validating mptcp_diag_dump_one
thanks to a new tool displaying MPTCP info for a specific token.
- Patch 3: a fix for a commit which is only in net-next.
- Patch 4: reduce parameters for one in-kernel PM helper.
- Patch 5: exit early when processing an ADD_ADDR echo to avoid unneeded
operations.
====================
Link: https://patch.msgid.link/20250228-net-next-mptcp-coverage-small-opti-v1-0-f933c4275676@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When the userspace PM is used, or when the in-kernel limits are reached,
there will be no need to schedule the PM worker to signal new addresses.
That corresponds to pm->work_pending set to 0.
In this case, an early exit can be done in mptcp_pm_add_addr_echoed()
not to hold the PM lock, and iterate over the announced addresses list,
not to schedule the worker anyway in this case. This is similar to what
is done when a connection or a subflow has been established.
Reviewed-by: Geliang Tang <geliang@kernel.org>
Signed-off-by: Matthieu Baerts (NGI0) <matttbe@kernel.org>
Link: https://patch.msgid.link/20250228-net-next-mptcp-coverage-small-opti-v1-5-f933c4275676@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Johannes Berg says:
====================
First 6.15 material:
* cfg80211/mac80211
- remove cooked monitor support
- strict mode for better AP testing
- basic EPCS support
- OMI RX bandwidth reduction support
* rtw88
- preparation for RTL8814AU support
* rtw89
- use wiphy_lock/wiphy_work
- preparations for MLO
- BT-Coex improvements
- regulatory support in firmware files
* iwlwifi
- preparations for the new iwlmld sub-driver
* tag 'wireless-next-2025-03-04-v2' of https://git.kernel.org/pub/scm/linux/kernel/git/wireless/wireless-next: (128 commits)
wifi: iwlwifi: remove mld/roc.c
wifi: mac80211: refactor populating mesh related fields in sinfo
wifi: cfg80211: reorg sinfo structure elements for mesh
wifi: iwlwifi: Fix spelling mistake "Increate" -> "Increase"
wifi: iwlwifi: add Debug Host Command APIs
wifi: iwlwifi: add IWL_MAX_NUM_IGTKS macro
wifi: iwlwifi: add OMI bandwidth reduction APIs
wifi: iwlwifi: remove mvm prefix from iwl_mvm_d3_end_notif
wifi: iwlwifi: remember if the UATS table was read successfully
wifi: iwlwifi: export iwl_get_lari_config_bitmap
wifi: iwlwifi: add support for external 32 KHz clock
wifi: iwlwifi: mld: add a debug level for EHT prints
wifi: iwlwifi: mld: add a debug level for PTP prints
wifi: iwlwifi: remove mvm prefix from iwl_mvm_esr_mode_notif
wifi: iwlwifi: use 0xff instead of 0xffffffff for invalid
wifi: iwlwifi: location api cleanup
wifi: cfg80211: expose update timestamp to drivers
wifi: mac80211: add ieee80211_iter_chan_contexts_mtx
wifi: mac80211: fix integer overflow in hwmp_route_info_get()
wifi: mac80211: Fix possible integer promotion issue
...
====================
Link: https://patch.msgid.link/20250304125605.127914-3-johannes@sipsolutions.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Breno Leitao says:
====================
netconsole: Add taskname sysdata support
This patchset introduces a new feature to the netconsole extradata
subsystem that enables the inclusion of the current task's name in the
sysdata output of netconsole messages.
This enhancement is particularly valuable for large-scale deployments,
such as Meta's, where netconsole collects messages from millions of
servers and stores them in a data warehouse for analysis. Engineers
often rely on these messages to investigate issues and assess kernel
health.
One common challenge we face is determining the context in which
a particular message was generated. By including the task name
(task->comm) with each message, this feature provides a direct answer to
the frequently asked question: "What was running when this message was
generated?"
This added context will significantly improve our ability to diagnose
and troubleshoot issues, making it easier to interpret output of
netconsole.
The patchset consists of seven patches that implement the following changes:
* Refactor CPU number formatting into a separate function
* Prefix CPU_NR sysdata feature with SYSDATA_
* Patch to covert a bitwise operation into boolean
* Add configfs controls for taskname sysdata feature
* Add taskname to extradata entry count
* Add support for including task name in netconsole's extra data output
* Document the task name feature in Documentation/networking/netconsole.rst
* Add test coverage for the task name feature to the existing sysdata selftest script
These changes allow users to enable or disable the task name feature via
configfs and provide additional context for kernel messages by showing
which task generated each console message.
I have tested these patches on some servers and they seem to work as
expected.
v1: https://lore.kernel.org/r/20250221-netcons_current-v1-0-21c86ae8fc0d@debian.org
Signed-off-by: Breno Leitao <leitao@debian.org>
====================
Link: https://patch.msgid.link/20250228-netcons_current-v2-0-f53ff79a0db2@debian.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Add test coverage for the netconsole task name feature to the existing
sysdata selftest script. This extends the test infrastructure to verify
that task names are correctly appended when enabled and absent when
disabled.
The test validates that:
- Task names appear in the expected format "taskname=<name>"
- Task names are included when the feature is enabled
- Task names are excluded when the feature is disabled
- The feature works correctly alongside other sysdata fields like CPU
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Add documentation for the netconsole task name feature in
Documentation/networking/netconsole.rst. This explains how to enable
task name via configfs and demonstrates the output format.
The documentation includes:
- How to enable/disable the feature via taskname_enabled
- The format of the task name in the output
- An example showing the task name appearing in messages
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This is the core patch for this whole patchset. Add support for
including the current task's name in netconsole's extra data output.
This adds a new append_taskname() function that writes the task name
(from current->comm) into the target's extradata buffer, similar to how
CPU numbers are handled.
The task name is included when the SYSDATA_TASKNAME field is set,
appearing in the format "taskname=<name>" in the output. This additional
context can help with debugging by showing which task generated each
console message.
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Add configfs interface to enable/disable the taskname sysdata feature.
This adds the following functionality:
The implementation follows the same pattern as the existing CPU number
feature, ensuring consistent behavior and error handling across sysdata
features.
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
New SYSDATA_TASKNAME feature flag to track when taskname append is enabled.
Additional check in count_extradata_entries() to include taskname in
total, counting it as an entry in extradata. This function is used to
check if we are not overflowing the number of extradata items.
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Extract CPU number formatting logic from prepare_extradata() into a new
append_cpu_nr() function.
This refactoring improves code organization by isolating CPU number
formatting into its own function while reducing the complexity of
prepare_extradata().
The change prepares the codebase for the upcoming taskname feature by
establishing a consistent pattern for handling sysdata features.
The CPU number formatting logic itself remains unchanged; only its
location has moved to improve maintainability.
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Convert the current state assignment to use explicit boolean conversion,
making the code more robust and easier to read. This change adds a
double-negation operator to ensure consistent boolean conversion as
suggested by Paolo[1].
This approach aligns with the existing pattern used in
sysdata_cpu_nr_enabled_show().
Link: https://lore.kernel.org/all/7309e760-63b0-4b58-ad33-2fb8db361141@redhat.com/ [1]
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Rename the CPU_NR enum value to SYSDATA_CPU_NR to establish a consistent
naming convention for sysdata features. This change prepares for
upcoming additions to the sysdata feature set by clearly grouping
related features under the SYSDATA prefix.
This change is purely cosmetic and does not modify any functionality.
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This file should never have been part of the commit, remove it.
Fixes: af3be90884 ("wifi: iwlwifi: support ROC version 6")
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
This patch implements the .ndo_eth_ioctl() to
read and write the PHY register.
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The MAC hardware is on the BMC side, and the driver is on the host side.
When the driver is abnormal, the BMC cannot directly detect the
exception cause.
Therefore, this patch implements the BMC diagnosis feature.
When users query driver diagnosis information on the BMC, the driver
detects the query request in the scheduled task and reports
driver statistics and link status to the BMC through the bar space.
The BMC collects logs to analyze exception causes.
Currently, the scheduled task is executed every 30 seconds
To quickly respond to user query requests,
this patch changes the scheduled task to once every second.
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
If the rate changed frequently, the PHY link ok,
but the MAC link maybe fails.
As a result, the network port is unavailable.
According to the documents of the chip,
core_reset needs to do to fix the fault.
In hw_adjus_link(), the core_reset is added to try to
ensure that MAC link status is normal.
In addition, MAC link failure detection is added.
If the MAC link fails after core_reset, driver invokes
the phy_stop() and phy_start() to re-link.
Due to phydev->lock, re-link cannot be triggered
in adjust_link(). Therefore, this operation
is invoked in a scheduled task.
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
the hardware error was reported by interrupt,
and need be fixed by doing function reset,
but the whole reset flow takes a long time,
should not do it in irq handler,
so do it in scheduled task.
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
This patch implements the rx checksum offload feature.
The tx checksum offload processing in .ndo_start_xmit()
has been accepted. This patch also adds the tx checksum
feature, including NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The driver supports many hw statistics. This patch supports
dump statistics through ethtool_ops and ndo.get_stats64().
The type of hw statistics register is u32,
To prevent the statistics register from overflowing,
the driver dump the statistics every 30 seconds.
in a scheduled task.
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Similar to PPE support for Mediatek devices, introduce PPE debugfs
in order to dump binded and unbinded flows.
Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
