Convert tx_qp from a single contiguous array allocation to per-queue
individual allocations. Each mana_tx_qp struct is approximately 35KB.
With many queues (e.g., 32/64), the flat array requires a single
contiguous allocation that can fail under memory fragmentation.
Change mana_tx_qp *tx_qp to mana_tx_qp **tx_qp (array of pointers),
allocating each queue's mana_tx_qp individually via kvzalloc. This
reduces each allocation to ~35KB and provides vmalloc fallback,
avoiding allocation failure due to fragmentation.
Signed-off-by: Aditya Garg <gargaditya@linux.microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Link: https://patch.msgid.link/20260502074552.23857-2-gargaditya@linux.microsoft.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Allison Henderson says:
====================
selftests: rds: Log collection, TAP compliance and cleanups
This series is a set of bug fixes and improvements for the rds
selftests.
Patch 1 bumps the kselftest timeout from 400s to 800s. The original
limit was developed against a lean config, but the kselftest harness
counts boot time and gcov log collection against the limit, so a
default config with gcov enabled needs more headroom.
Patch 2 corrects some typos in the run.sh USAGE string and removes an
unused "-g" flag.
Patch 3 silences a handful of pylint warnings in test.py: it adds a
module docstring, suppresses the warnings tied to the sys.path.append
import trick, marks the long lived tcpdump Popen with disable-next
consider-using-with, and drops unused exception variables from two
BlockingIOError except clauses.
Patch 4 adds a -t flag to run.sh so the timeout can be overridden
if needed.
Patch 5 adds a RDS_LOG_DIR environment variable that specifies where
logs should be stored, or skips log collection if left unset
Patch 6 adds a SUDO_USER environment variable that sets the user
for tcpdump --relinquish-privileges. This avoid the permissions
drop that would leave pcaps empty on 9pfs since 9p does not
support chown
Patch 7 removes the initial tmp tcpdumps and instead saves the pcaps
directly to the logdir if it is set.
Patch 8 hoists the tcpdump shutdown into a helper and calls it from the
timeout signal handler so that the processes are properly terminated
and dumps are flushed
Patch 9 fixes gcov collection by ensuring debugfs is mounted, and
specifying the --root folder so that gcov can still find the kernel
source when it is run from the ksft test directory.
Patch 10 makes the test output TAP compliant so the kselftest runner
parses results correctly.
====================
Link: https://patch.msgid.link/20260504054143.4027538-1-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch updates the rds selftests output to be TAP compliant.
Use ksft_pr() to mark debug output with a leading '# ' so that TAP
parsers treat it as commentary, and convert all informational print()
calls to use ksft_pr(). sys.exit(0) is changed to os._exit(0) to
avoid duplicate prints from the buffered TAP output. The console
output from the tcpdump subprocess is silenced, and the gcov console
output is redirected to a gcovr.log.
Finally adjust the exit path so that the hash check loop sets a
return code instead exiting directly. Then print the TAP results
and totals lines before exiting.
Signed-off-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260504054143.4027538-11-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
debugfs is not mounted automatically in a virtme-ng guest, so the
gcov data copy from /sys/kernel/debug/gcov/ silently finds nothing
depending on whether debugfs is mounted by default on the host OS.
Fix this by mounting debugfs in run.sh before copying the gcda
files.
Finally when invoked through the kselftest runner, the working
directory is the test directory rather than the kernel source root.
gcovr defaults --root to the current working directory, which causes
it to filter out all coverage data for files under net/rds/ since
they are not under the test directory. Fix this by passing --root
to gcovr explicitly.
Signed-off-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260504054143.4027538-10-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The timeout signal handler for the rds selftests currently just
exits when the time limit is exceeded, and forgets to stop the
network dumps. Fix this by hoisting the tcpdump terminate commands
into a helper function, and call it from the signal handler before
exiting
Bound proc.wait() with a timeout (and fall back to proc.kill())
so an unresponsive tcpdump cannot hang the timeout path itself.
We also pop() tcpdump_procs as we iterate, so stop_pcaps() is safe
to call from both the normal cleanup path and the signal handler,
since the second invocation simply has nothing to do
Signed-off-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260504054143.4027538-9-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch removes the initial tmp tcpdumps and instead saves
the pcaps directly to the logdir if it is set.
Signed-off-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260504054143.4027538-8-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch modifies rds selftests to use the environment variable
SUDO_USER for tcpdumps if it is set. This is needed to avoid chown
operations on the vng 9pfs which is not supported. Passing a user
listed in sudoers avoids the tcpdump privilege drop which may
otherwise create empty pcaps
Signed-off-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260504054143.4027538-7-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch modifies the rds selftest to look for an env variable
RDS_LOG_DIR, and log all traces, pcaps and gcov collections to
the folder specified in RDS_LOG_DIR. If RDS_LOG_DIR is unset,
logs are not collected.
Signed-off-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260504054143.4027538-6-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a -t flag to run.sh to optionally override the default
timeout. The --timeout flag is already supported in test.py,
so just add the shorthand -t flag
Signed-off-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260504054143.4027538-5-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This patch fixes a few pylint errors in test.py. Remove unused exception
variables from except blocks, and disable warnings for imports that cannot
appear at the start of the module. Also disable warnings for the
tcpdump processes. The suggestion to use a with block does not apply
here since the process needs to outlive the parent to collect the dumps.
Lastly add the module docstring at the top of the module.
Signed-off-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260504054143.4027538-4-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The run.sh script does not have a -g flag. Update USAGE string with
correct flags. Also fix typo packet_duplcate -> packet_duplicate
Signed-off-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260504054143.4027538-3-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The 400s time out was originally developed under a leaner
kernel config that booted much faster than a default config.
Boot up is included as part of the over all test runtime, as
well as any log collection done when the test is complete.
A slower config combined with the gcov enabled test means
we'll need more time to accommodate the boot up and log
collection. So, bump time out to 800s.
Signed-off-by: Allison Henderson <achender@kernel.org>
Link: https://patch.msgid.link/20260504054143.4027538-2-achender@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
cros_typec_register_thunderbolt() missed initializing the `adata->lock`
mutex. This leads to a NULL dereference when the mutex is later
acquired (e.g. in cros_typec_altmode_work()).
Initialize the mutex in cros_typec_register_thunderbolt() to fix the
issue.
Cc: stable@vger.kernel.org
Fixes: 3b00be26b1 ("platform/chrome: cros_ec_typec: Thunderbolt support")
Reviewed-by: Benson Leung <bleung@chromium.org>
Reviewed-by: Abhishek Pandit-Subedi <abhishekpandit@chromium.org>
Link: https://lore.kernel.org/r/20260505053403.3335740-1-tzungbi@kernel.org
Signed-off-by: Tzung-Bi Shih <tzungbi@kernel.org>
Tariq Toukan says:
====================
net/mlx5: Fixes for Socket-Direct
This series fixes several race conditions and bugs in the mlx5
Socket-Direct (SD) single netdev flow.
Patch 1 serializes mlx5_sd_init()/mlx5_sd_cleanup() with
mlx5_devcom_comp_lock() and tracks the SD group state on the primary
device, preventing concurrent or duplicate bring-up/tear-down.
Patch 2 fixes the debugfs "multi-pf" directory being stored on the
calling device's sd struct instead of the primary's, which caused
memory leaks and recreation errors when cleanup ran from a different PF.
Patch 3 fixes a race where a secondary PF could access the primary's
auxiliary device after it had been unbound, by holding the primary's
device lock while operating on its auxiliary device.
Patch 4 fixes missing cleanup on ETH probe errors. The analogous gap on
the resume path requires introducing sd_suspend/resume APIs that only
destroy FW resources and is left for a follow-up series.
====================
Link: https://patch.msgid.link/20260504180206.268568-1-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When utilizing Socket-Direct single netdev functionality the driver
resolves the actual auxiliary device using mlx5_sd_get_adev(). However,
the current implementation returns the primary ETH auxiliary device
without holding the device lock, leading to a potential race condition
where the ETH device could be unbound or removed concurrently during
probe, suspend, resume, or remove operations.[1]
Fix this by introducing mlx5_sd_put_adev() and updating
mlx5_sd_get_adev() so that secondaries devices would get a ref and
acquire the device lock of the returned auxiliary device. After the lock
is acquired, a second devcom check is needed[2].
In addition, update The callers to pair the get operation with the new
put operation, ensuring the lock is held while the auxiliary device is
being operated on and released afterwards.
The "primary" designation is determined once in sd_register(). It's set
before devcom is marked ready, and it never changes after that.
In Addition, The primary path never locks a secondary: When the primary
device invoke mlx5_sd_get_adev(), it sees dev == primary and returns.
no additional lock is taken.
Therefore lock ordering is always: secondary_lock -> primary_lock. The
reverse never happens, so ABBA deadlock is impossible.
[1]
for example:
BUG: kernel NULL pointer dereference, address: 0000000000000370
PGD 0 P4D 0
Oops: Oops: 0000 [#1] SMP
CPU: 4 UID: 0 PID: 3945 Comm: bash Not tainted 6.19.0-rc3+ #1 NONE
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
RIP: 0010:mlx5e_dcbnl_dscp_app+0x23/0x100 [mlx5_core]
Call Trace:
<TASK>
mlx5e_remove+0x82/0x12a [mlx5_core]
device_release_driver_internal+0x194/0x1f0
bus_remove_device+0xc6/0x140
device_del+0x159/0x3c0
? devl_param_driverinit_value_get+0x29/0x80
mlx5_rescan_drivers_locked+0x92/0x160 [mlx5_core]
mlx5_unregister_device+0x34/0x50 [mlx5_core]
mlx5_uninit_one+0x43/0xb0 [mlx5_core]
remove_one+0x4e/0xc0 [mlx5_core]
pci_device_remove+0x39/0xa0
device_release_driver_internal+0x194/0x1f0
unbind_store+0x99/0xa0
kernfs_fop_write_iter+0x12e/0x1e0
vfs_write+0x215/0x3d0
ksys_write+0x5f/0xd0
do_syscall_64+0x55/0xe90
entry_SYSCALL_64_after_hwframe+0x4b/0x53
[2]
CPU0 (primary) CPU1 (secondary)
==========================================================================
mlx5e_remove() (device_lock held)
mlx5e_remove() (2nd device_lock held)
mlx5_sd_get_adev()
mlx5_devcom_comp_is_ready() => true
device_lock(primary)
mlx5_sd_get_adev() ==> ret adev
_mlx5e_remove()
mlx5_sd_cleanup()
// mlx5e_remove finished
// releasing device_lock
//need another check here...
mlx5_devcom_comp_is_ready() => false
Fixes: 381978d283 ("net/mlx5e: Create single netdev per SD group")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260504180206.268568-5-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
When _mlx5e_probe() fails, the preceding successful mlx5_sd_init() is
not undone. Auxiliary bus probe failure skips binding, so mlx5e_remove()
is never called for that adev and the matching mlx5_sd_cleanup() never
runs - leaking the per-dev SD struct.
Call mlx5_sd_cleanup() on the probe error path to balance
mlx5_sd_init().
A similar gap exists on the resume path: mlx5_sd_init() and
mlx5_sd_cleanup() are currently bundled with both probe/remove and
suspend/resume, even though only the FW alias state actually needs to
follow the suspend/resume lifecycle - the sd struct allocation and
devcom membership are software state that should track the full bound
lifetime. As a result, a failed resume can leave a still-bound device
with sd == NULL, which mlx5_sd_get_adev() can't distinguish from a
non-SD device. Fixing this requires sd_suspend/resume APIs which will
only destroy FW resources and is left for a follow-up series.
Fixes: 381978d283 ("net/mlx5e: Create single netdev per SD group")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260504180206.268568-4-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mlx5_sd_init() creates the "multi-pf" debugfs directory under the
primary device debugfs root, but stored the dentry in the calling
device's sd struct. When sd_cleanup() run on a different PF,
this leads to using the wrong sd->dfs for removing entries, which
results in memory leak and an error in when re-creating the SD.[1]
Fix it by explicitly storing the debugfs dentry in the primary
device sd struct and use it for all per-group files.
[1]
debugfs: 'multi-pf' already exists in '0000:08:00.1'
Fixes: 4375130bf5 ("net/mlx5: SD, Add debugfs")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260504180206.268568-3-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mlx5_sd_init() / mlx5_sd_cleanup() may run from multiple PFs in the same
Socket-Direct group. This can cause the SD bring-up/tear-down sequence
to be executed more than once or interleaved across PFs.
Protect SD init/cleanup with mlx5_devcom_comp_lock() and track the SD
group state on the primary device. Skip init if the primary is already
UP, and skip cleanup unless the primary is UP.
The state check on cleanup is needed because sd_register() drops the
devcom comp lock between marking the comp ready and assigning
primary_dev on each peer. A concurrent cleanup that acquires the lock
in this window would observe devcom_is_ready==true while primary_dev
is still NULL (causing mlx5_sd_get_primary() to return NULL) or while
the FW alias setup performed by mlx5_sd_init()'s body has not yet run
(causing sd_cmd_unset_primary() to dereference a NULL tx_ft). Gate the
cleanup body on primary_sd->state == MLX5_SD_STATE_UP, which is set
only at the very end of mlx5_sd_init() under the same comp lock - so
observing UP guarantees primary_dev, secondaries[], tx_ft, and dfs are
all populated. Also bail explicitly if mlx5_sd_get_primary() returns
NULL, in case state is checked on a peer whose primary_dev hasn't been
assigned yet.
In addition, move mlx5_devcom_comp_set_ready(false) from sd_unregister()
into the cleanup's locked section, including the !primary and
state != UP early-exit paths, so the device cannot unregister and free
its struct mlx5_sd while devcom is still marked ready. A concurrent
init acquiring the devcom lock will now observe devcom is no longer
ready and bail out immediately.
Fixes: 381978d283 ("net/mlx5e: Create single netdev per SD group")
Signed-off-by: Shay Drory <shayd@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260504180206.268568-2-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
devlink reload while PSP connections are active does:
mlx5_unload_one_devl_locked() -> mlx5_detach_device()
-> _mlx5e_suspend()
-> mlx5e_detach_netdev()
-> profile->cleanup_rx
-> profile->cleanup_tx
-> mlx5e_destroy_mdev_resources() -> mlx5_core_dealloc_pd() fails:
...
mlx5_core 0000:08:00.0: mlx5_cmd_out_err:821:(pid 19722):
DEALLOC_PD(0x801) op_mod(0x0) failed, status bad resource state(0x9),
syndrome (0xef0c8a), err(-22)
...
The reason for failure is the existence of TX keys, which are removed by
the PSP dev unregistration happening in:
profile->cleanup() -> mlx5e_psp_unregister() -> mlx5e_psp_cleanup()
-> psp_dev_unregister()
...but this isn't invoked in the devlink reload flow, only when changing
the NIC profile (e.g. when transitioning to switchdev mode) or on dev
teardown.
Move PSP device registration into mlx5e_nic_enable(), and unregistration
into the corresponding mlx5e_nic_disable(). These functions are called
during netdev attach/detach after RX & TX are set up.
This ensures that the keys will be gone by the time the PD is destroyed.
Fixes: 89ee2d92f6 ("net/mlx5e: Support PSP offload functionality")
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260504181100.269334-4-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently, during PSP init, priv->psp is initialized to an incompletely
built psp struct. Additionally, on fs init failure priv->psp is reset to
NULL.
Change this so that only a fully initialized priv->psp is set, which
makes the code easier to reason about in failure scenarios.
Fixes: af2196f494 ("net/mlx5e: Implement PSP operations .assoc_add and .assoc_del")
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260504181100.269334-3-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
priv->psp->psp is initialized with the PSP device as returned by
psp_dev_create(). This could also return an error, in which case a
future psp_dev_unregister() will result in unpleasantness.
Avoid that by using a local variable and only saving the PSP device when
registration succeeds.
In case psp_dev_create() fails, priv->psp and steering structs are left
in place, but they will be inert. The unchecked access of priv->psp in
mlx5e_psp_offload_handle_rx_skb() won't happen because without a PSP
device, there can be no SAs added and therefore no packets will be
successfully decrypted and be handed off to the SW handler.
Fixes: 89ee2d92f6 ("net/mlx5e: Support PSP offload functionality")
Signed-off-by: Cosmin Ratiu <cratiu@nvidia.com>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260504181100.269334-2-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
t7xx_port_enum_msg_handler() uses the modem-supplied port_count field as
a loop bound over port_msg->data[] without checking that the message buffer
contains sufficient data. A modem sending port_count=65535 in a 12-byte
buffer triggers a slab-out-of-bounds read of up to 262140 bytes.
Add a sizeof(*port_msg) check before accessing the port message header
fields to guard against undersized messages.
Add a struct_size() check after extracting port_count and before the loop.
In t7xx_parse_host_rt_data(), guard the rt_feature header read with a
remaining-buffer check before accessing data_len, validate feat_data_len
against the actual remaining buffer to prevent OOB reads and signed
integer overflow on offset.
Pass msg_len from both call sites: skb->len at the DPMAIF path after
skb_pull(), and the validated feat_data_len at the handshake path.
Fixes: da45d2566a ("net: wwan: t7xx: Add control port")
Cc: stable@vger.kernel.org
Signed-off-by: Pavitra Jha <jhapavitra98@gmail.com>
Link: https://patch.msgid.link/20260501110713.145563-1-jhapavitra98@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit 23e8b470c7 ("net: dsa: mv88e6xxx: Add devlink param for ATU
hash algorithm.") introduced ATU hash algorithm access via devlink, but
did not enable it for the 6320 family. Do it now.
Signed-off-by: Marek Behún <kabel@kernel.org>
Link: https://patch.msgid.link/20260504153227.1390546-6-kabel@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit 9e907d739c ("net: dsa: mv88e6xxx: add POT operation") did not
add the .pot_clear() method to the 6321 switch operations structure.
Add them now.
Signed-off-by: Marek Behún <kabel@kernel.org>
Link: https://patch.msgid.link/20260504153227.1390546-4-kabel@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The 6320 family supports the ALT_SPEED bit on ports 2, 5 and 6. Allow
this speed by implementing 6320 family specific .port_set_speed_duplex()
method.
Signed-off-by: Marek Behún <kabel@kernel.org>
Link: https://patch.msgid.link/20260504153227.1390546-3-kabel@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Willem de Bruijn says:
====================
selftests: drv-net: convert so_txtime to drv-net
In preparation for extending to pacing hardware offload, convert the
so_txtime.sh test to a drv-net test that can be run against netdevsim
and real hardware.
Two preparatory patches
1. support negative tests, where tests are expected to fail
2. add a tc helper
See individual patches for details and detailed changelog
====================
Link: https://patch.msgid.link/20260504174056.565319-1-willemdebruijn.kernel@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In preparation for extending to pacing hardware offload, convert the
so_txtime.sh test to a drv-net test that can be run against netdevsim
and real hardware.
Also update so_txtime.c to not exit on first failure, but run to
completion and report exit code there. This helps with debugging
unexpected results, especially when processing multiple packets,
as happens in the "reverse_order" testcase.
Signed-off-by: Willem de Bruijn <willemb@google.com>
----
v6 -> v7
- update test to use new argument expect_fail
- v6 received Reviewed-by, but dropped due to above (minor) change
v5 -> v6
- fix order in tools/testing/selftests/drivers/net/config
v4 -> v5
- move qdisc setup/restore into each test
- add tc to utils.py (separate patch)
- test expected failure (separate patch)
- fix pylint
- convert fail to pass for timing errors if KSFT_MACHINE_SLOW
(cmd does not special case KSFT_SKIP process returncode yet)
Responses to sashiko review
- The test converts per packet failure to errors, to continue
testing other packets, but other error() cases are not in scope.
- The test starts sender and receiver at an absolute future time,
like the original test. This assumes ~msec scale sync'ed clocks.
- The tc qdisc replace command works fine with noqueue. Tested
manually.
v3 -> v4
- restore original qdisc after test
- drop unnecessary underscore in tap test names
v2 -> v3
- Makefile: so_txtime from YNL_GEN_FILES to TEST_GEN_FILES (Sashiko, NIPA)
v1 -> v2
- move so_txtime.c for net/lib to drivers/net (Jakub)
- fix drivers/net/config order (Jakub)
- detect passing when failure is expected (Jakub, Sashiko)
- pass pylint --disable=R (Jakub)
- only call ksft_run once (Jakub)
- do not sleep if waiting time is negative (Sashiko)
- add \n when converting error() to fprintf() (Sashiko)
- 4 space indentation, instead of 2 space
- increase sync delay from 100 to 200ms, to fix rare vng flakes
Link: https://patch.msgid.link/20260504174056.565319-4-willemdebruijn.kernel@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add a wrapper similar to existing ip, ethtool, ... commands.
Tc takes a slightly different syntax. Account for that.
The first user is the next patch in this series, converting so_txtime
to drv-net. Pacing offload is supported by selected qdiscs only.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260504174056.565319-3-willemdebruijn.kernel@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Support negative tests, where cmd raises an exception if the command
succeeded.
Add optional argument expect_fail to cmd and bkg. Where fail fails the
test on unexpected error, expect_fail fails it on unexpected success.
Both fail on negative return code. Python subprocess may set a
negative return code on process crash or timeout. Those are never
anticipated failures.
Signed-off-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260504174056.565319-2-willemdebruijn.kernel@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Use the convenience macro PCI_DEVICE to initialize .vendor, .device,
.subvendor and .subdevice. Drop explicit zeros that the compiler also
fills in.
Signed-off-by: Uwe Kleine-König (The Capable Hub) <u.kleine-koenig@baylibre.com>
Link: https://patch.msgid.link/20260504151202.2139919-2-u.kleine-koenig@baylibre.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The .port_max_speed_mode() method is not used anymore since commit
40da0c32c3 ("net: dsa: mv88e6xxx: remove handling for DSA and CPU ports").
Drop it.
Signed-off-by: Marek Behún <kabel@kernel.org>
Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Link: https://patch.msgid.link/20260504152653.1389394-1-kabel@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
fq_codel_dump_class_stats() acquires qdisc spinlock only when requested
to follow flow->head chain.
As we did in sch_cake recently, add the missing READ_ONCE()/WRITE_ONCE()
annotations.
Fixes: edb09eb17e ("net: sched: do not acquire qdisc spinlock in qdisc/class stats dump")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Jamal Hadi Salim <jhs@mojatatu.com>
Link: https://patch.msgid.link/20260504163842.1162001-1-edumazet@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
-----BEGIN PGP SIGNATURE-----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=KQQr
-----END PGP SIGNATURE-----
Merge tag 'nf-26-05-05' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf
Pablo Neira Ayuso says:
====================
IPVS fixes for net
The following batch contains IPVS fixes for net to address issues
from the latest net-next pull request.
Julian Anastasov made the following summary:
1-3) Fixes for the recently added resizable hash tables
4) dest from trash can be leaked if ip_vs_start_estimator() fails
5) fixed races and locking for the estimation kthreads
6) fix for wrong roundup_pow_of_two() usage in the resizable hash
tables
7-8) v2 of the changes from Waiman Long to properly guard against
the housekeeping_cpumask() updates:
https://lore.kernel.org/netfilter-devel/20260331165015.2777765-1-longman@redhat.com/
I added missing Fixes tag. The original description:
Since commit 041ee6f372 ("kthread: Rely on HK_TYPE_DOMAIN for preferred
affinity management"), the HK_TYPE_KTHREAD housekeeping cpumask may no
longer be correct in showing the actual CPU affinity of kthreads that
have no predefined CPU affinity. As the ipvs networking code is still
using HK_TYPE_KTHREAD, we need to make HK_TYPE_KTHREAD reflect the
reality.
This patch series makes HK_TYPE_KTHREAD an alias of HK_TYPE_DOMAIN
and uses RCU to protect access to the HK_TYPE_KTHREAD housekeeping
cpumask.
Julian plans to post a nf-next patch to limit the connections by using
"conn_max" sysctl. With Simon Horman, they agreed that this is an old
problem that we do not have a limit of connections and it is not a
stopper for this patchset.
* tag 'nf-26-05-05' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf:
sched/isolation: Make HK_TYPE_KTHREAD an alias of HK_TYPE_DOMAIN
ipvs: Guard access of HK_TYPE_KTHREAD cpumask with RCU
ipvs: fix shift-out-of-bounds in ip_vs_rht_desired_size
ipvs: fix races around est_mutex and est_cpulist
ipvs: do not leak dest after get from dest trash
ipvs: fix the spin_lock usage for RT build
ipvs: fix races around the conn_lfactor and svc_lfactor sysctl vars
ipvs: fixes for the new ip_vs_status info
====================
Link: https://patch.msgid.link/20260505001648.360569-1-pablo@netfilter.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Kuniyuki Iwashima says:
====================
udp_tunnel: Speed up UDP tunnel device destruction (Part I)
Most of the UDP tunnel devices call synchronize_rcu() twice
during destruction, for example, vxlan has
1) synchronize_rcu() in udp_tunnel_sock_release()
2) synchronize_net() in vxlan_sock_release()
The goal of this series is to remove the former, and another
followup series removes the latter.
synchronize_rcu() was added in udp_tunnel_sock_release() by
commit 3cf7203ca6 ("net/tunnel: wait until all sk_user_data
reader finish before releasing the sock").
This was intended to protect the fast path of a dying vxlan
from dereferencing vxlan_sock->sock->sk after sock_orphan()
has set sock->sk to NULL.
Most of the UDP tunnel devices store struct socket to its
private struct, but it is NOT needed in the fast paths;
struct sock is used there, but struct socket is only used
for tunnel setup / teardown.
This is probably because UDP tunnel functions accept struct
socket, but even such functions do not need it, except for
udp_tunnel_sock_release(), which can safely access sk->sk_socket.
The overview of the series:
Patch 1 - 5 : Convert UDP tunnel helper to take struct sock
Patch 6 : Small fix for 10-years-old bug
Patch 7 - 14 : Store struct sock in tunnel devices
Patch 15 : Remove synchronize_rcu() in udp_tunnel_sock_release()
With this change, a script creating/upping vxlan in 4000 netns
runs 10x faster.
====================
[See Link for benchmark results.]
Link: https://patch.msgid.link/20260502031401.3557229-1-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit 3cf7203ca6 ("net/tunnel: wait until all sk_user_data
reader finish before releasing the sock") added synchronize_rcu()
in udp_tunnel_sock_release().
This was intended to protect the fast path of a dying vxlan device
from dereferencing vxlan_sock->sock->sk after sock_orphan() has set
sock->sk to NULL.
However, vxlan does not need to access struct socket itself
in the fast path; it only reads struct sock, and struct socket
is only used for tunnel setup and teardown.
This applies to all other UDP tunnel users, and they have been
converted to access struct sock directly.
In addition, each device-specific struct used in their fast paths
is freed after one RCU grace period. Since this occurs after
udp_tunnel_sock_release(), the struct is guaranteed to be freed
after struct udp_sock.
Therefore, synchronize_rcu() in udp_tunnel_sock_release() is
now redundant.
Let's remove it.
Tested:
A script creating/upping vxlan devices in 4000 netns runs 10x
faster with this change. We can see the same improvement with
other UDP tunnel devices as well.
$ cat vxlan.sh
for i in `seq 1 40`
do
(for j in `seq 1 100` ; do
unshare -n bash -c "ip link add vxlan0 type vxlan id 100 local 127.0.0.1 dstport 4789 && ip link set vxlan0 up";
done) &
done
wait
With bpftrace, we can see vxlan_stop() is significantly faster.
bpftrace -e '
kprobe:vxlan_stop {
@start[tid] = nsecs;
}
kretprobe:vxlan_stop /@start[tid]/ {
@duration_us = hist((nsecs - @start[tid]) / 1000);
delete(@start[tid]);
}
END {
printf("\nExecution time of vxlan_stop (us):\n");
}'
Before:
# time ./vxlan.sh // without bpftrace
real 0m50.615s
user 0m8.171s
sys 1m45.101s
@duration_us:
[4K, 8K) 1266 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ |
[8K, 16K) 1957 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
[16K, 32K) 764 |@@@@@@@@@@@@@@@@@@@@ |
[32K, 64K) 6 | |
[64K, 128K) 4 | |
[128K, 256K) 3 | |
After:
# time ./vxlan.sh // without bpftrace
real 0m5.247s
user 0m7.956s
sys 1m47.404s
@duration_us:
[16, 32) 3411 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@|
[32, 64) 383 |@@@@@ |
[64, 128) 107 |@ |
[128, 256) 79 |@ |
[256, 512) 16 | |
[512, 1K) 2 | |
[1K, 2K) 2 | |
Next step is to remove another synchronize_net() in vxlan_stop()
and variants in other devices.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-16-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tipc udp_bearer does not need to access struct socket itself in
the fast path; it only reads struct sock, and struct socket is
only used for tunnel setup and teardown.
Let's store struct sock directly in struct udp_bearer.
Note that cleanup_bearer() calls synchronize_net() after
udp_tunnel_sock_release(), so udp_bearer is not freed until
inflight fast paths finish.
Note also that synchronize_rcu() is added in the error path
of tipc_udp_enable() since udp_bearer will be kfree()d
immediately once we remove synchronize_rcu() in
udp_tunnel_sock_release().
This can be later converted to kfree_rcu().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-15-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
pfcp does not need to access struct socket itself in the fast
path; it only reads struct sock, and struct socket is only used
for tunnel setup and teardown.
Let's store struct sock directly in struct pfcp_dev.
pfcp_del_sock() is called from dev->netdev_ops->ndo_uninit().
The 2nd synchronize_net() in unregister_netdevice_many_notify()
ensures that inflight pfcp RX fast paths finish before pfcp_dev
is freed.
Note that synchronize_rcu() is added in the error path of
pfcp_newlink() since free_netdev() will free pfcp_dev immediately
once we remove synchronize_rcu() in udp_tunnel_sock_release().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-14-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
amt does not need to access struct socket itself in the fast path;
it only reads struct sock, and struct socket is only used for tunnel
setup and teardown.
Let's store struct sock directly in struct amt.
amt_dev_stop() is called as dev->netdev_ops->ndo_stop().
synchronize_net() in unregister_netdevice_many_notify() ensures
that inflight amt RX fast paths finish before amt_dev is freed.
amt no longer needs synchronize_rcu() in udp_tunnel_sock_release().
Note that amt_dev_stop() looks buggy; cancel_delayed_work_sync()
should be called after udp_tunnel_sock_release().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-13-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
fou does not need to access struct socket itself in the fast
path; it only reads struct sock, and struct socket is only used
for tunnel setup and teardown.
Let's store struct sock directly in struct fou.
fou_release() frees struct fou with kfree_rcu(), so fou no
longer needs synchronize_rcu() in udp_tunnel_sock_release().
Note that the error path in fou_create() looks buggy; once the
tunnel is set up and fou_add_to_port_list() fails, struct fou
should be freed with kfree_rcu() _after_ udp_tunnel_sock_release().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-12-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
bareudp does not need to access struct socket itself in the fast
path; it only reads struct sock, and struct socket is only used
for tunnel setup and teardown.
Let's store struct sock directly in struct bareudp_dev.
bareudp_sock_release() is called from dev->netdev_ops->ndo_stop().
synchronize_net() in unregister_netdevice_many_notify() ensures that
inflight bareudp RX fast paths finish before bareudp_dev is freed.
bareudp no longer needs synchronize_rcu() in udp_tunnel_sock_release().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-11-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
geneve does not need to access struct socket itself in the fast
path; it only reads struct sock, and struct socket is only used for
tunnel setup and teardown.
Let's store struct sock directly in struct geneve_sock.
__geneve_sock_release() frees geneve_sock with kfree_rcu(), so
geneve no longer needs synchronize_rcu() in udp_tunnel_sock_release().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-10-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We will remove synchronize_rcu() in udp_tunnel_sock_release().
We must ensure that vxlan_sock is freed after inflight RX fast path.
Let's free vxlan_sock with kfree_rcu().
Note that vxlan_sock.vni_list[] is 8K and struct rcu_head must
be placed before it.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-9-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit 3cf7203ca6 ("net/tunnel: wait until all sk_user_data
reader finish before releasing the sock") added synchronize_rcu()
in udp_tunnel_sock_release().
This was intended to protect the fast path of a dying vxlan device
from dereferencing vxlan_sock->sock->sk after sock_orphan() has set
sock->sk to NULL.
However, vxlan does not need to access struct socket itself in the
fast path; it only reads struct sock, and struct socket is only
used for tunnel setup and teardown.
Let's store struct sock directly in struct vxlan_sock.
In the next patch, we will free vxlan_sock with kfree_rcu(), then
vxlan no longer needs synchronize_rcu() in udp_tunnel_sock_release().
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-8-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
udp_tunnel_sock_release() could set sk->sk_user_data to NULL
while vxlan_gro_prepare_receive() is running.
Let's check if rcu_dereference_sk_user_data() is NULL after
skb_gro_remcsum_init().
Fixes: 5602c48cf8 ("vxlan: change vxlan to use UDP socket GRO")
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-7-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
None of the udp_tunnel users need struct socket in their
fast paths; it is only used for tunnel setup / teardown.
Even udp_tunnel_notify_{add,del}_rx_port() do not need
struct socket.
Let's change udp_tunnel_notify_{add,del}_rx_port() to take
struct sock instead of struct socket.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-6-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
None of the udp_tunnel users need struct socket in their
fast paths; it is only used for tunnel setup / teardown.
Even udp_tunnel_{push,drop}_rx_port() do not need struct socket.
Let's change udp_tunnel_{push,drop}_rx_port() to take struct
sock instead of struct socket.
Signed-off-by: Kuniyuki Iwashima <kuniyu@google.com>
Link: https://patch.msgid.link/20260502031401.3557229-5-kuniyu@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
