This patch set adds all the supported FLEXCOMs for the SAMA7D65 SoC.
This also adds the GMAC interfaces and enables GMAC0 interface for
the SAMA7D65 SoC.
With the FLEXCOMs added to the SoC the MCP16502 and the MAC address
EEPROM are both added to flexcom10.
The dt-binding for USART is here [1]. And the dt-binding for DMA has
been applied here [2].
The original thread for this is here [3]. The applied changes have been
removed for this resend
Jason Xing [Mon, 23 Jun 2025 12:01:59 +0000 (20:01 +0800)]
net: xsk: dpaa2: avoid repeatedly updating the global consumer
This patch avoids another update of the consumer at the end of
dpaa2_xsk_tx().
In the zero copy xmit path, two versions (batched and non-batched)
regarding how the consumer of tx ring changes are implemented in
xsk_tx_peek_release_desc_batch() that eventually updates the local
consumer to the global consumer in either of the following call trace:
1) batched mode:
xsk_tx_peek_release_desc_batch()
__xskq_cons_release()
2) non-batched mode:
xsk_tx_peek_release_desc_batch()
xsk_tx_peek_release_fallback()
xsk_tx_release()
As we can see, dpaa2_xsk_tx() doesn't need to call extra release function
to handle the sync of consumer itself.
Peiyang Wang [Mon, 23 Jun 2025 04:00:43 +0000 (12:00 +0800)]
net: hns3: clear hns alarm: comparison of integer expressions of different signedness
A static alarm exists in the hns and needs to be cleared.
The alarm is comparison of integer expressions of different
signedness including 's64' and 'long unsigned int',
'int' and 'long unsigned int', 'u32' and 'int',
'int' and 'unsigned int'.
Jijie Shao [Mon, 23 Jun 2025 04:00:39 +0000 (12:00 +0800)]
net: hns3: use hns3_get_ops() helper to reduce the unnecessary middle layer conversion
There are too many indirection layers in the HNS3 driver code,
This issue was previously discussed with the maintainer,
who suggested adding a helper function to fix the issue.
In fact, the hns3_get_ops() helper is already defined
and can fix this issue.
This patch uses hns3_get_ops() helper to reduce the unnecessary
middle layer conversion. Apply it to the whole HNS3 driver.
The former discusstion can be checked from the link.
Jijie Shao [Mon, 23 Jun 2025 04:00:38 +0000 (12:00 +0800)]
net: hns3: use hns3_get_ae_dev() helper to reduce the unnecessary middle layer conversion
There are too many indirection layers in the HNS3 driver code.
This issue was previously discussed with the maintainer,
who suggested adding a helper function to fix the issue.
In fact, the hns3_get_ae_dev() helper is already defined
and can fix this issue.
This patch uses hns3_get_ae_dev() helper to reduce the unnecessary
middle layer conversion. Apply it to the whole HNS3 driver.
The former discusstion can be checked from the link.
Qingfang Deng [Mon, 23 Jun 2025 03:34:31 +0000 (11:34 +0800)]
pppoe: drop PACKET_OTHERHOST before skb_share_check()
Align with ip_rcv() by dropping PACKET_OTHERHOST packets before
calling skb_share_check(). This avoids unnecessary skb processing
for packets that will be discarded anyway.
Paolo Abeni [Mon, 23 Jun 2025 10:53:55 +0000 (12:53 +0200)]
udp_tunnel: fix deadlock in udp_tunnel_nic_set_port_priv()
While configuring a vxlan tunnel in a system with a i40e NIC driver, I
observe the following deadlock:
WARNING: possible recursive locking detected
6.16.0-rc2.net-next-6.16_92d87230d899+ #13 Tainted: G E
--------------------------------------------
kworker/u256:4/1125 is trying to acquire lock: ffff88921ab9c8c8 (&utn->lock){+.+.}-{4:4}, at: i40e_udp_tunnel_set_port (/home/pabeni/net-next/include/net/udp_tunnel.h:343 /home/pabeni/net-next/drivers/net/ethernet/intel/i40e/i40e_main.c:13013) i40e
but task is already holding lock: ffff88921ab9c8c8 (&utn->lock){+.+.}-{4:4}, at: udp_tunnel_nic_device_sync_work (/home/pabeni/net-next/net/ipv4/udp_tunnel_nic.c:739) udp_tunnel
other info that might help us debug this:
Possible unsafe locking scenario:
Thanks to new PSE driver being proposed on the mailinglist figured out
there was some unclear documentation, which lead the developer on the
wrong path. Clarify these documentation.
The changes focus on clarifying the setup_pi_matrix callback behavior
and improving device tree binding descriptions, particularly around
channel-to-PI mapping relationships that are critical for proper PSE
controller integration.
Improve the setup_pi_matrix callback documentation to clarify its purpose
and usage. The enhanced description explains that PSE PI devicetree nodes
are pre-parsed before this callback is invoked, and drivers should utilize
pcdev->pi[x]->pairset[y].np to map PSE controller hardware ports to their
corresponding Power Interfaces.
This clarification helps driver implementers understand the callback's
role in establishing the hardware-to-PI relationship mapping.
Improve the channels property description to better explain the
relationship between physical delivery channels and PSE PI pairsets.
The previous description was unclear about how channels are referenced
and used in the port matrix mapping.
====================
Add support for 25G, 50G, and 100G to fbnic
The fbnic driver up till now had avoided actually reporting link as the
phylink setup only supported up to 40G configurations. This changeset is
meant to start addressing that by adding support for 50G and 100G interface
types.
With that basic support added fbnic can then set those types based on the
EEPROM configuration provided by the firmware and then report those speeds
out using the information provided via the phylink call for getting the
link ksettings. This provides the basic MAC support and enables supporting
the speeds as well as configuring flow control.
After this I plan to add support for a PHY that will represent the SerDes
PHY being used to manage the link as we need a way to indicate link
training into phylink to prevent link flaps on the PCS while the SerDes is
in training, and then after that I will look at rolling support for our
PCS/PMA into the XPCS driver.
====================
Alexander Duyck [Wed, 18 Jun 2025 22:08:09 +0000 (15:08 -0700)]
fbnic: Add support for setting/getting pause configuration
Phylink already handles most of the pieces necessary for configuring
autonegotation. With that being the case we can add support for
getting/setting pause now by just passing through the ethtool call to the
phylink code.
Alexander Duyck [Wed, 18 Jun 2025 22:08:02 +0000 (15:08 -0700)]
fbnic: Add support for reporting link config
This change adds some basic support for reporting the current link config
to the user via ethtool. Currently the main components reported are the
carrier status, link speed, and FEC.
For now we are handling the FEC directly as phylink doesn't have support
for it. The plan is to work on incorporating FEC support into phylink and
eventually adding the ability for us to set the FEC configuration through
phylink itself.
In addition as we don't yet have SFP or PHY support the listed modes
supported are including ones not supported by the media we are attached to.
That will hopefully be addressed once we can get the QSFP modules
supported.
Alexander Duyck [Wed, 18 Jun 2025 22:07:55 +0000 (15:07 -0700)]
fbnic: Set correct supported modes and speeds based on FW setting
The fbnic driver was using the XLGMII link mode to enable phylink, however
that mode wasn't the correct one to use as the NIC doesn't actually use
XLGMII, it is using a combinations of 25G, 50G, and 100G interface modes
and configuring those via pins exposed on the PCS, MAC, and PHY interfaces.
To more accurately reflect that we should drop the uxe of XGMII and XLGMII
and instead use the correct interface types.
Alexander Duyck [Wed, 18 Jun 2025 22:07:42 +0000 (15:07 -0700)]
fbnic: Replace link_mode with AUI
The way we were using "link_mode" was really more to describe the interface
between the attachment unit interface(s) we were using on the device.
Specifically the AUI is describing the modulation and the number of lanes
we are using. So we can simplify this by replacing link_mode with aui.
Alexander Duyck [Wed, 18 Jun 2025 22:07:35 +0000 (15:07 -0700)]
fbnic: Retire "AUTO" flags and cleanup handling of FW link settings
There were several issues in the way we were handling the link info coming
from firmware.
First is the fact that we were carrying around "AUTO" flags to indicate
that we needed to populate the values. We can just drop this and assume
that we will always be populating the settings from firmware. With this we
can also clean up the masking as the "AUTO" flags were just there to be
stripped anyway.
Second since we are getting rid of the "AUTO" setting we still need a way
to report that the link is not configured. We convert the link_mode "AUTO"
to "UNKNOWN" to do this. With this we can avoid reporting link up in the
phylink_pcs_get_state call as we will just set link to 0 and return without
updating the link speed. This is preferred versus the driver just forcing
50G which makes it harder to recover when the FW does start providing valid
settings.
With this the plan is to eventually replace the link_mode we use with the
interface_mode from phylink for all intents and purposes and have the two
be interchangeable. At that point we can convert the FW provided settings
over to something closer to link partner settings and give phylink greater
control of the interface allowing for user override of the settings and an
asynchronous setup of the link versus having to pull early settings from
firmware.
Alexander Duyck [Wed, 18 Jun 2025 22:07:28 +0000 (15:07 -0700)]
fbnic: Do not consider mailbox "initialized" until we have verified fw version
In order for us to make use of the information in the PHY we need to verify
that the FW capabilities message has been processed. To do so we should
poll until the FW version in the capabilities message is at least at the
minimum level needed to verify that the FW can support the basic PHY config
messages.
As such this change adds logic so that we will poll the mailbox until such
time as the FW version can be populated with an acceptable value. If it
doesn't reach a sufficicient value the mailbox will be considered to have
timed out.
Alexander Duyck [Wed, 18 Jun 2025 22:07:22 +0000 (15:07 -0700)]
net: phy: Add interface types for 50G and 100G
Add support for 802.3cd based interface types 50GBASE-R and 100GBASE-P.
This choice in naming is based on section 135 of the 802.3-2022 IEEE
Standard.
In addition it is adding support for what I am referring to as LAUI
which is based on annex 135C of the IEEE Standard, and shares many
similarities with the 25/50G consortium. The main difference between the
two is that IEEE spec refers to LAUI as the AUI before the RS(544/514) FEC,
whereas the 25/50G use this lane and frequency combination after going
through RS(528/514), Base-R or no FEC at all.
We frequently consult with Jesper's out-of-tree page_pool benchmark to
evaluate page_pool changes.
Import the benchmark into the upstream linux kernel tree so that (a)
we're all running the same version, (b) pave the way for shared
improvements, and (c) maybe one day integrate it with nipa, if possible.
Import bench_page_pool_simple from commit 35b1716d0c30 ("Add
page_bench06_walk_all"), from this repository:
https://github.com/netoptimizer/prototype-kernel.git
Changes done during upstreaming:
- Fix checkpatch issues.
- Remove the tasklet logic not needed.
- Move under tools/testing
- Create ksft for the benchmark.
- Changed slightly how the benchmark gets build. Out of tree, time_bench
is built as an independent .ko. Here it is included in
bench_page_pool.ko
Steps to run:
```
mkdir -p /tmp/run-pp-bench
make -C ./tools/testing/selftests/net/bench
make -C ./tools/testing/selftests/net/bench install INSTALL_PATH=/tmp/run-pp-bench
rsync --delete -avz --progress /tmp/run-pp-bench mina@$SERVER:~/
ssh mina@$SERVER << EOF
cd ~/run-pp-bench && sudo ./test_bench_page_pool.sh
EOF
```
Note that by default, the Makefile will build the benchmark for the
currently installed kernel in /lib/modules/$(shell uname -r)/build. To
build against the current tree, do:
make KDIR=$(pwd) -C ./tools/testing/selftests/net/bench
Jakub Kicinski [Tue, 24 Jun 2025 01:14:03 +0000 (18:14 -0700)]
Merge branch 'rework-irq-handling-in-mtk_eth_soc'
Frank Wunderlich says:
====================
rework IRQ handling in mtk_eth_soc
From: Frank Wunderlich <frank-w@public-files.de>
This series introduces named IRQs while keeping the index based way
for older dts.
Further it makes some cleanup like adding consts for index access and
avoids loading first IRQ which was not used on non SHARED_INT SoCs.
====================
Frank Wunderlich [Thu, 19 Jun 2025 13:21:24 +0000 (15:21 +0200)]
net: ethernet: mtk_eth_soc: only use legacy mode on missing IRQ name
If platform_get_irq_byname returns -ENXIO fall back to legacy (index
based) mode, but on other errors function should return this error.
Suggested-by: Daniel Golle <daniel@makrotopia.org> Signed-off-by: Frank Wunderlich <frank-w@public-files.de> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://patch.msgid.link/20250619132125.78368-5-linux@fw-web.de Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Frank Wunderlich [Thu, 19 Jun 2025 13:21:23 +0000 (15:21 +0200)]
net: ethernet: mtk_eth_soc: skip first IRQ if not used
On SoCs with dedicated RX and TX interrupts (all except MT7621 and
MT7628) platform_get_irq() is called for the first IRQ (eth->irq[0])
but it is never used.
Skip the first IRQ and reduce the IRQ-count to 2.
Signed-off-by: Frank Wunderlich <frank-w@public-files.de> Reviewed-by: Daniel Golle <daniel@makrotopia.org> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://patch.msgid.link/20250619132125.78368-4-linux@fw-web.de Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Frank Wunderlich [Thu, 19 Jun 2025 13:21:22 +0000 (15:21 +0200)]
net: ethernet: mtk_eth_soc: add consts for irq index
Use consts instead of fixed integers for accessing IRQ array.
Signed-off-by: Frank Wunderlich <frank-w@public-files.de> Reviewed-by: Simon Horman <horms@kernel.org> Reviewed-by: Daniel Golle <daniel@makrotopia.org> Link: https://patch.msgid.link/20250619132125.78368-3-linux@fw-web.de Signed-off-by: Jakub Kicinski <kuba@kernel.org>
====================
selftests: drv-net: stats: use skip instead of xfail
Alex posted support for configuring pause frames in fbnic. This flipped
the pause stats test from xfail to fail. Because CI considered xfail as
pass it now flags the test as failing. This shouldn't happen. Also we
currently report pause and FEC tests as passing on virtio which doesn't
make sense.
====================
Jakub Kicinski [Fri, 20 Jun 2025 16:11:09 +0000 (09:11 -0700)]
selftests: drv-net: stats: use skip instead of xfail for unsupported features
XFAIL is considered a form of a pass by our CI. For HW devices returning
XFAIL for unsupported features is counter-productive because our CI
knows not to expect any HW test to pass until it sees 10 passes in a row.
If we return xfail the test shows up as pass even if the device doesn't
support the feature. netdevsim supports all features necessary for
the stats test so there is no concern about running the test in SW mode.
Make the test skip rather than xfail if driver doesn't support FEC or pause.
Jakub Kicinski [Fri, 20 Jun 2025 16:11:08 +0000 (09:11 -0700)]
selftests: drv-net: stats: fix pylint issues
Small adjustments to make pylint happy.
One warning about unused argument remains because the test uses
global variables rather than attaching netlink sockets to cfg.
Fixing this would be too much of a change for a linter fix commit
like this one.
Eric Dumazet [Fri, 20 Jun 2025 13:30:01 +0000 (13:30 +0000)]
net: remove sock_i_uid()
Difference between sock_i_uid() and sk_uid() is that
after sock_orphan(), sock_i_uid() returns GLOBAL_ROOT_UID
while sk_uid() returns the last cached sk->sk_uid value.
None of sock_i_uid() callers care about this.
Use sk_uid() which is much faster and inlined.
Note that diag/dump users are calling sock_i_ino() and
can not see the full benefit yet.
Eric Dumazet [Fri, 20 Jun 2025 13:30:00 +0000 (13:30 +0000)]
net: annotate races around sk->sk_uid
sk->sk_uid can be read while another thread changes its
value in sockfs_setattr().
Add sk_uid(const struct sock *sk) helper to factorize the needed
READ_ONCE() annotations, and add corresponding WRITE_ONCE()
where needed.
Fixes: 86741ec25462 ("net: core: Add a UID field to struct sock.") Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: Lorenzo Colitti <lorenzo@google.com> Reviewed-by: Maciej Żenczykowski <maze@google.com> Link: https://patch.msgid.link/20250620133001.4090592-2-edumazet@google.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Arnd Bergmann [Fri, 20 Jun 2025 11:22:39 +0000 (13:22 +0200)]
caif: reduce stack size, again
I tried to fix the stack usage in this function a couple of years ago,
but there is still a problem with the latest gcc versions in some
configurations:
net/caif/cfctrl.c:553:1: error: the frame size of 1296 bytes is larger than 1280 bytes [-Werror=frame-larger-than=]
Reduce this once again, with a separate cfctrl_link_setup() function that
holds the bulk of all the local variables. It also turns out that the
param[] array that takes up a large portion of the stack is write-only
and can be left out here.
Arnd Bergmann [Fri, 20 Jun 2025 11:19:04 +0000 (13:19 +0200)]
lib: test_objagg: split test_hints_case() into two functions
With sanitizers enabled, this function uses a lot of stack, causing
a harmless warning:
lib/test_objagg.c: In function 'test_hints_case.constprop':
lib/test_objagg.c:994:1: error: the frame size of 1440 bytes is larger than 1408 bytes [-Werror=frame-larger-than=]
Most of this is from the two 'struct world' structures. Since most of
the work in this function is duplicated for the two, split it up into
separate functions that each use one of them.
The combined stack usage is still the same here, but there is no warning
any more, and the code is still safe because of the known call chain.
Frank Li [Wed, 18 Jun 2025 18:44:16 +0000 (14:44 -0400)]
dt-bindings: net: convert qca,qca7000.txt yaml format
Convert qca,qca7000.txt yaml format.
Additional changes:
- add refs: spi-peripheral-props.yaml, serial-peripheral-props.yaml and
ethernet-controller.yaml.
- simple spi and uart node name.
- use low case for mac address in examples.
- add check reg choose spi-peripheral-props.yaml or
spi-peripheral-props.yaml.
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> Signed-off-by: Frank Li <Frank.Li@nxp.com> Link: https://patch.msgid.link/20250618184417.2169745-1-Frank.Li@nxp.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Jakub Kicinski [Sat, 21 Jun 2025 17:19:43 +0000 (10:19 -0700)]
selftests: drv-net: import things in lib one by one
pylint doesn't understand our path hacks, and it generates a lot
of warnings for driver tests. Import what we use one by one, this
is hopefully not too tedious and it makes pylint happy.
lpc18xx uses plat_dat->mac_interface, despite wanting to validate the
PHY interface. Checking the DT files (arch/arm/boot/dts/nxp/lpc/), none
of them specify mac-mode which means mac_interface and phy_interface
will be identical.
mac_interface is only used when there is some kind of MII converter
between the DesignWare MAC and PHY, and describes the interface mode
that the DW MAC needs to use, whereas phy_interface describes the
interface mode that the PHY uses.
Noting that lpc18xx only supports MII and RMII interface modes, switch
this glue driver to use plat_dat->phy_interface, and to mark that the
mac_interface is not used, explicitly set it to PHY_INTERFACE_MODE_NA.
The latter is safe as the only user of mac_interface for this platform
would be in stmmac_check_pcs_mode(), which only checks for RGMII or
SGMII.
====================
Add support for the IPQ5018 Internal GE PHY
The IPQ5018 SoC contains an internal Gigabit Ethernet PHY with its
output pins that provide an MDI interface to either an external switch
in a PHY to PHY link architecture or directly to an attached RJ45
connector.
The PHY supports 10BASE-T/100BASE-TX/1000BASE-T link modes in SGMII
interface mode, CDT, auto-negotiation and 802.3az EEE.
The LDO controller found in the IPQ5018 SoC needs to be enabled to drive
power to the CMN Ethernet Block (CMN BLK) which the GE PHY depends on.
The LDO must be enabled in TCSR by writing to a specific register.
In a phy to phy architecture, DAC values need to be set to accommodate
for the short cable length.
====================
George Moussalem [Fri, 13 Jun 2025 01:55:08 +0000 (05:55 +0400)]
net: phy: qcom: at803x: Add Qualcomm IPQ5018 Internal PHY support
The IPQ5018 SoC contains a single internal Gigabit Ethernet PHY which
provides an MDI interface directly to an RJ45 connector or an external
switch over a PHY to PHY link.
The PHY supports 10BASE-T/100BASE-TX/1000BASE-T link modes in SGMII
interface mode, CDT, auto-negotiation and 802.3az EEE.
Let's add support for this PHY in the at803x driver as it falls within
the Qualcomm Atheros OUI.
George Moussalem [Fri, 13 Jun 2025 01:55:07 +0000 (05:55 +0400)]
dt-bindings: net: qca,ar803x: Add IPQ5018 Internal GE PHY support
Document the IPQ5018 Internal Gigabit Ethernet PHY found in the IPQ5018
SoC. Its output pins provide an MDI interface to either an external
switch in a PHY to PHY link scenario or is directly attached to an RJ45
connector.
The PHY supports 10/100/1000 mbps link modes, CDT, auto-negotiation and
802.3az EEE.
For operation, the LDO controller found in the IPQ5018 SoC for which
there is provision in the mdio-4019 driver.
Two common archictures across IPQ5018 boards are:
1. IPQ5018 PHY --> MDI --> RJ45 connector
2. IPQ5018 PHY --> MDI --> External PHY
In a phy to phy architecture, the DAC needs to be configured to
accommodate for the short cable length. As such, add an optional boolean
property so the driver sets preset DAC register values accordingly.
Miroslav Lichvar [Thu, 19 Jun 2025 13:53:42 +0000 (15:53 +0200)]
testptp: add option to enable external timestamping edges
Some drivers (e.g. ice) don't enable any edges by default when external
timestamping is requested by the PTP_EXTTS_REQUEST ioctl, which makes
testptp -e unusable for testing hardware supported by these drivers.
Add -E option to specify if the rising, falling, or both edges should
be enabled by the ioctl.
Signed-off-by: Miroslav Lichvar <mlichvar@redhat.com> Cc: Richard Cochran <richardcochran@gmail.com> Cc: Jacob Keller <jacob.e.keller@intel.com> Reviewed-by: Vadim Fedorenko <vadim.fedorenko@linux.dev> Signed-off-by: David S. Miller <davem@davemloft.net>
Jakub Kicinski [Fri, 20 Jun 2025 17:40:07 +0000 (10:40 -0700)]
netdevsim: fix UaF when counting Tx stats
skb may be freed as soon as we put it on the rx queue.
Use the len variable like the code did prior to the conversion.
Fixes: f9e2511d80c2 ("netdevsim: migrate to dstats stats collection") Signed-off-by: Jakub Kicinski <kuba@kernel.org> Acked-by: David S. Miller <davem@davemloft.net> Reviewed-by: Breno Leitao <leitao@debian.org> Signed-off-by: David S. Miller <davem@davemloft.net>
Yuyang Huang [Thu, 19 Jun 2025 03:51:16 +0000 (12:51 +0900)]
selftest: add selftest for anycast notifications
This commit adds a new kernel selftest to verify RTNLGRP_IPV6_ACADDR
notifications. The test works by adding/removing a dummy interface,
enabling packet forwarding, and then confirming that user space can
correctly receive anycast notifications.
The test relies on the iproute2 version to be 6.13+.
Tested by the following command:
$ vng -v --user root --cpus 16 -- \
make -C tools/testing/selftests TARGETS=net
TEST_PROGS=rtnetlink_notification.sh \
TEST_GEN_PROGS="" run_tests
Cc: Maciej Żenczykowski <maze@google.com> Cc: Lorenzo Colitti <lorenzo@google.com> Signed-off-by: Yuyang Huang <yuyanghuang@google.com> Reviewed-by: Simon Horman <horms@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
Wang Liang [Thu, 19 Jun 2025 03:08:54 +0000 (11:08 +0800)]
net/smc: remove unused function smc_lo_supports_v2
The smcd_ops->supports_v2 is only called in smcd_register_dev(), which
calls function smcd_supports_v2 for ism. For loopback-ism, function
smc_lo_supports_v2 is unused, remove it.
Kory Maincent [Fri, 20 Jun 2025 09:16:41 +0000 (11:16 +0200)]
net: pse-pd: Fix ethnl_pse_send_ntf() stub parameter type
The ethnl_pse_send_ntf() stub function has incorrect parameter type when
CONFIG_ETHTOOL_NETLINK is disabled. The function should take a net_device
pointer instead of phy_device pointer to match the actual implementation.
Reported-by: kernel test robot <lkp@intel.com> Closes: https://lore.kernel.org/oe-kbuild-all/202506200355.TqFiYUbN-lkp@intel.com/ Fixes: fc0e6db30941 ("net: pse-pd: Add support for reporting events") Signed-off-by: Kory Maincent <kory.maincent@bootlin.com> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://patch.msgid.link/20250620091641.2098028-1-kory.maincent@bootlin.com Signed-off-by: Jakub Kicinski <kuba@kernel.org>
====================
eth: finish migration to the new RXFH callbacks
Finish drivers conversions to callbacks added by
commit 9bb00786fc61 ("net: ethtool: add dedicated callbacks for
getting and setting rxfh fields"). Remove the conditional calling
in the core, rxnfc callbacks are no longer used for RXFH.
====================
Jakub Kicinski [Wed, 18 Jun 2025 20:38:23 +0000 (13:38 -0700)]
net: ethtool: don't mux RXFH via rxnfc callbacks
All drivers have been converted. Stop using the rxnfc fallbacks
for Rx Flow Hashing configuration.
Joe pointed out in earlier review that in ethtool_set_rxfh()
we need both .get_rxnfc and .get_rxfh_fields, because we need
both the ring count and flow hashing (because we call
ethtool_check_flow_types()). IOW the existing check added
for transitioning drivers was buggy.
Jakub Kicinski [Wed, 18 Jun 2025 20:38:21 +0000 (13:38 -0700)]
eth: hinic: migrate to new RXFH callbacks
Migrate to new callbacks added by commit 9bb00786fc61 ("net: ethtool:
add dedicated callbacks for getting and setting rxfh fields").
Zeroing data on SET is not necessary, the argument is not copied
back to user space. The driver has no other RXNFC functionality
so the SET callback can be now removed.
Jakub Kicinski [Wed, 18 Jun 2025 20:38:16 +0000 (13:38 -0700)]
eth: sfc: migrate to new RXFH callbacks
Migrate to new callbacks added by commit 9bb00786fc61 ("net: ethtool:
add dedicated callbacks for getting and setting rxfh fields").
This driver's RXFH config is read only / fixed so the conversion
is purely factoring out the handling into a helper. One thing of
note that this is one of the two drivers which pays attention to
rss_context.
Jakub Kicinski [Wed, 18 Jun 2025 20:38:15 +0000 (13:38 -0700)]
eth: sfc: siena: migrate to new RXFH callbacks
Migrate to new callbacks added by commit 9bb00786fc61 ("net: ethtool:
add dedicated callbacks for getting and setting rxfh fields").
This driver's RXFH config is read only / fixed so the conversion
is purely factoring out the handling into a helper.
Jakub Kicinski [Wed, 18 Jun 2025 20:38:14 +0000 (13:38 -0700)]
eth: sfc: falcon: migrate to new RXFH callbacks
Migrate to new callbacks added by commit 9bb00786fc61 ("net: ethtool:
add dedicated callbacks for getting and setting rxfh fields").
This driver's RXFH config is read only / fixed so the conversion
is purely factoring out the handling into a helper.
can: rcar_canfd: Describe channel-specific FD registers using C struct
The rcar_canfd_f_*() inline functions to obtain channel-specific CAN-FD
register offsets really describe a memory layout. Hence replace them by
a C structure, to simplify the code, and reduce kernel size.
This also gets rid of warnings about unused rcar_canfd_f_*() inline
functions, which are reported by recent versions of clang.
Suggested-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr> Reported-by: Jakub Kicinski <kuba@kernel.org> Closes: https://lore.kernel.org/20250618183827.5bebca8f@kernel.org Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> Acked-by: Vincent Mailhol <mailhol.vincent@wanadoo.fr> Link: https://patch.msgid.link/292b75b3bc8dd95f805f0223f606737071c8cf86.1750327217.git.geert+renesas@glider.be Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Simon Horman [Thu, 19 Jun 2025 13:58:32 +0000 (14:58 +0100)]
rds: Correct endian annotation of port and addr assignments
Correct the endianness annotation of port assignments:
A host byte order value (RDS_TCP_PORT) is correctly converted to
network byte order (big endian) using htons. But it is then cast back to
host byte order before assigning to a variable that expects a big endian
value. Address this by dropping the cast.
This is not a bug because, while the endian annotation is changed by
this patch, the assigned value is unchanged.
Also correct the endianness of address assignment.
A host byte order value (INADDR_ANY) is incorrectly assigned as-is to
a variable that expects a big endian value. Address this by converting
the value to network byte order (big endian).
This is not a bug because INADDR_ANY is 0, which is isomorphic
with regards to endian conversions. IOW, while the endian annotation
is changed by this patch, the assigned value is unchanged.
Incorrect endian annotations appear to date back to IPv4-only code added
by commit 70041088e3b9 ("RDS: Add TCP transport to RDS").
David S. Miller [Sat, 21 Jun 2025 13:26:25 +0000 (14:26 +0100)]
Merge branch 'gve-xdp-tx-redirect' into main
Joshua Washington says:
====================
gve: XDP TX and redirect support for DQ RDA
A previous patch series[1] introduced the ability to process XDP buffers
to the DQ RDA queue format. This is a follow-up patch series to
introduce XDP_TX and XDP_REDIRECT support and expose XDP support to the
kernel.
gve: add XDP_TX and XDP_REDIRECT support for DQ RDA
This patch adds support for XDP_TX and XDP_REDIRECT for the DQ RDA
queue format. To appropriately support transmission of XDP frames, a
new pending packet type GVE_TX_PENDING_PACKET_DQO_XDP_FRAME is
introduced for completion handling, as there was a previous assumption
that completed packets would be SKBs.
XDP_TX handling completes the basic XDP actions, so the feature is
recorded accordingly. This patch also enables the ndo_xdp_xmit callback
allowing DQ to handle XDP_REDIRECT packets originating from another
interface.
The XDP spinlock is moved to common TX ring fields so that it can be
used in both GQ and DQ. Originally, it was in a section which was
mutually exclusive for GQ and DQ.
In summary, 3 XDP features are exposed for the DQ RDA queue format:
1) NETDEV_XDP_ACT_BASIC
2) NETDEV_XDP_ACT_NDO_XMIT
3) NETDEV_XDP_ACT_REDIRECT
Note that XDP and header-data split are mutually exclusive for the time
being due to lack of multi-buffer XDP support.
This patch does not add support for the DQ QPL format. That is to come
in a future patch series.
Reviewed-by: Willem de Bruijn <willemb@google.com> Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com> Signed-off-by: Joshua Washington <joshwash@google.com> Signed-off-by: Harshitha Ramamurthy <hramamurthy@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
gve: refactor DQO TX methods to be more generic for XDP
This patch performs various minor DQO TX datapath refactors in
preparation for adding XDP_TX and XDP_REDIRECT support. The following
refactors are performed:
1) gve_tx_fill_pkt_desc_dqo() relies on a SKB pointer to
get whether checksum offloading should be enabled. This won't work
for the XDP case, which does not have a SKB. This patch updates the
method to use a boolean representing whether checksum offloading
should be enabled directly.
2) gve_maybe_stop_dqo() contains some synchronization between the true
TX head and the cached value, a synchronization which is common for
XDP queues and normal netdev queues. However, that method is reserved
for netdev TX queues. To avoid duplicate code, this logic is factored
out into a new method, gve_has_tx_slots_available().
3) gve_tx_update_tail() is added to update the TX tail, a functionality
that will be common between normal TX and XDP TX codepaths.
Reviewed-by: Willem de Bruijn <willemb@google.com> Signed-off-by: Joshua Washington <joshwash@google.com> Signed-off-by: Praveen Kaligineedi <pkaligineedi@google.com> Signed-off-by: Harshitha Ramamurthy <hramamurthy@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for XDP DQ support, the gve_xdp_xmit callback needs to
be generalized for all queue formats. This patch renames the GQ-specific
function to gve_xdp_xmit_gqi, and introduces a new gve_xdp_xmit callback
which branches on queue format.
Reviewed-by: Willem de Bruijn <willemb@google.com> Signed-off-by: Joshua Washington <joshwash@google.com> Signed-off-by: Harshitha Ramamurthy <hramamurthy@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
rvu_mbox_init function makes use of error path for
freeing memory which are local to the function in
both success and failure conditions. This is unusual hence
fix it by returning zero on success. With new cn20k code this
is freeing valid memory in success case also.
Fixes: e53ee4acb220 ("octeontx2-af: CN20k basic mbox operations and structures") Signed-off-by: Subbaraya Sundeep <sbhatta@marvell.com> Reviewed-by: Simon Horman <horms@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net>
Himanshu Mittal [Wed, 18 Jun 2025 17:55:36 +0000 (23:25 +0530)]
net: ti: icssg-prueth: Add prp offload support to ICSSG driver
Add support for ICSSG PRP mode which supports offloading of:
- Packet duplication and PRP trailer insertion
- Packet duplicate discard and PRP trailer removal
Signed-off-by: Himanshu Mittal <h-mittal1@ti.com> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://patch.msgid.link/20250618175536.430568-1-h-mittal1@ti.com Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Jakub Kicinski <kuba@kernel.org>
====================
ref_tracker: add ability to register a debugfs file for a ref_tracker_dir
For those just joining in, this series adds a new top-level
"ref_tracker" debugfs directory, and has each ref_tracker_dir register a
file in there as part of its initialization. It also adds the ability to
register a symlink with a more human-usable name that points to the
file, and does some general cleanup of how the ref_tracker object names
are handled.
Jeff Layton [Wed, 18 Jun 2025 14:24:22 +0000 (10:24 -0400)]
ref_tracker: eliminate the ref_tracker_dir name field
Now that we have dentries and the ability to create meaningful symlinks
to them, don't keep a name string in each tracker. Switch the output
format to print "class@address", and drop the name field.
Also, add a kerneldoc header for ref_tracker_dir_init().
Jeff Layton [Wed, 18 Jun 2025 14:24:21 +0000 (10:24 -0400)]
net: add symlinks to ref_tracker_dir for netns
After assigning the inode number to the namespace, use it to create a
unique name for each netns refcount tracker with the ns.inum and
net_cookie values in it, and register a symlink to the debugfs file for
it.
init_net is registered before the ref_tracker dir is created, so add a
late_initcall() to register its files and symlinks.
Jeff Layton [Wed, 18 Jun 2025 14:24:20 +0000 (10:24 -0400)]
ref_tracker: add a way to create a symlink to the ref_tracker_dir debugfs file
Add the ability for a subsystem to add a user-friendly symlink that
points to a ref_tracker_dir's debugfs file. Add a separate
debugfs_symlinks xarray and use that to track symlinks. The reaper
workqueue job will remove symlinks before their corresponding dentries.
Jeff Layton [Wed, 18 Jun 2025 14:24:19 +0000 (10:24 -0400)]
ref_tracker: automatically register a file in debugfs for a ref_tracker_dir
Currently, there is no convenient way to see the info that the
ref_tracking infrastructure collects. Attempt to create a file in
debugfs when called from ref_tracker_dir_init().
The file is given the name "class@%px", as having the unmodified address
is helpful for debugging. This should be safe since this directory is only
accessible by root
While ref_tracker_dir_init() is generally called from a context where
sleeping is OK, ref_tracker_dir_exit() can be called from anywhere.
Thus, dentry cleanup must be handled asynchronously.
Add a new global xarray that has entries with the ref_tracker_dir
pointer as the index and the corresponding debugfs dentry pointer as the
value. Instead of removing the debugfs dentry, have
ref_tracker_dir_exit() set a mark on the xarray entry and schedule a
workqueue job. The workqueue job then walks the xarray looking for
marked entries, and removes their xarray entries and the debugfs
dentries.
Because of this, the debugfs dentry can outlive the corresponding
ref_tracker_dir. Have ref_tracker_debugfs_show() take extra care to
ensure that it's safe to dereference the dir pointer before using it.
Jeff Layton [Wed, 18 Jun 2025 14:24:18 +0000 (10:24 -0400)]
ref_tracker: allow pr_ostream() to print directly to a seq_file
Allow pr_ostream to also output directly to a seq_file without an
intermediate buffer. The first caller of +ref_tracker_dir_seq_print()
will come in a later patch, so mark that __maybe_unused for now. That
designation will be removed once it is used.
Jeff Layton [Wed, 18 Jun 2025 14:24:17 +0000 (10:24 -0400)]
ref_tracker: add a static classname string to each ref_tracker_dir
A later patch in the series will be adding debugfs files for each
ref_tracker that get created in ref_tracker_dir_init(). The format will
be "class@%px". The current "name" string can vary between
ref_tracker_dir objects of the same type, so it's not suitable for this
purpose.
Add a new "class" string to the ref_tracker dir that describes the
the type of object (sans any individual info for that object).
Also, in the i915 driver, gate the creation of debugfs files on whether
the dentry pointer is still set to NULL. CI has shown that the
ref_tracker_dir can be initialized more than once.
Jeff Layton [Wed, 18 Jun 2025 14:24:16 +0000 (10:24 -0400)]
ref_tracker: have callers pass output function to pr_ostream()
In a later patch, we'll be adding a 3rd mechanism for outputting
ref_tracker info via seq_file. Instead of a conditional, have the caller
set a pointer to an output function in struct ostream. As part of this,
the log prefix must be explicitly passed in, as it's too late for the
pr_fmt macro.
Jeff Layton [Wed, 18 Jun 2025 14:24:15 +0000 (10:24 -0400)]
ref_tracker: add a top level debugfs directory for ref_tracker
Add a new "ref_tracker" directory in debugfs. Each individual refcount
tracker can register files under there to display info about
currently-held references.
Jeff Layton [Wed, 18 Jun 2025 14:24:14 +0000 (10:24 -0400)]
ref_tracker: don't use %pK in pr_ostream() output
As Thomas Weißschuh points out [1], it is now preferable to use %p
instead of hashed pointers with printk(), since raw pointers should no
longer be leaked into the kernel log. Change the ref_tracker
infrastructure to use %p instead of %pK in its formats.
The netpoll_setup() function has grown complex over time, mixing
different error handling and concerns like carrier waiting, IPv4 address
retrieval, and IPv6 address retrieval all within a single function,
which is huge (127 LoC).
This patch series refactors the netpoll_setup() function to improve code
organization and readability by extracting logical blocks into dedicated
helper functions. netpoll_setup() length is reduced to 72 LoC.
This series breaks down these responsibilities into focused helper
functions.
The changes are purely structural with no functional modifications.
This changes were tested with the netconsole tests and the netpoll
selftest (WIP)[1]
====================
netdevsim: implement RX statistics using NETDEV_PCPU_STAT_DSTATS
The netdevsim driver previously lacked RX statistics support, which
prevented its use with the GenerateTraffic() test framework, as this
framework verifies traffic flow by checking RX byte counts.
This patch migrates netdevsim from its custom statistics collection to
the NETDEV_PCPU_STAT_DSTATS framework, as suggested by Jakub. This
change not only standardizes the statistics handling but also adds the
necessary RX statistics support required by the test framework.
Breno Leitao [Wed, 18 Jun 2025 09:32:47 +0000 (02:32 -0700)]
netpoll: Extract IPv6 address retrieval function
Extract the IPv6 address retrieval logic from netpoll_setup() into
a dedicated helper function netpoll_take_ipv6() to improve code
organization and readability.
The function handles obtaining the local IPv6 address from the
network device, including proper address type matching between
local and remote addresses (link-local vs global), and includes
appropriate error handling when IPv6 is not supported or no
suitable address is available.
Breno Leitao [Wed, 18 Jun 2025 09:32:46 +0000 (02:32 -0700)]
netpoll: extract IPv4 address retrieval into helper function
Move the IPv4 address retrieval logic from netpoll_setup() into a
separate netpoll_take_ipv4() function to improve code organization
and readability. This change consolidates the IPv4-specific logic
and error handling into a dedicated function while maintaining
the same functionality.
Breno Leitao [Wed, 18 Jun 2025 09:32:45 +0000 (02:32 -0700)]
netpoll: Extract carrier wait function
Extract the carrier waiting logic into a dedicated helper function
netpoll_wait_carrier() to improve code readability and reduce
duplication in netpoll_setup().
Breno Leitao [Wed, 18 Jun 2025 08:32:45 +0000 (01:32 -0700)]
netdevsim: account dropped packet length in stats on queue free
Add a call to dev_dstats_rx_dropped_add() in nsim_queue_free() to
account for the number of packets dropped when purging the skb queue.
This improves the accuracy of RX drop statistics reported by netdevsim.
local_bh_{disable, enable}() protection is used to disable preemption,
which is necessary given that dev_dstats_rx_dropped_add() access
this_cpu_ptr(). See discussion in [1].
Breno Leitao [Wed, 18 Jun 2025 08:32:44 +0000 (01:32 -0700)]
net: add dev_dstats_rx_dropped_add() helper
Introduce the dev_dstats_rx_dropped_add() helper to allow incrementing
the rx_drops per-CPU statistic by an arbitrary value, rather than just
one. This is useful for drivers or code paths that need to account for
multiple dropped packets at once, such as when dropping entire queues.
Nicolas Escande [Tue, 17 Jun 2025 14:13:34 +0000 (16:13 +0200)]
neighbour: add support for NUD_PERMANENT proxy entries
As discussesd before in [0] proxy entries (which are more configuration
than runtime data) should stay when the link (carrier) goes does down.
This is what happens for regular neighbour entries.
So lets fix this by:
- storing in proxy entries the fact that it was added as NUD_PERMANENT
- not removing NUD_PERMANENT proxy entries when the carrier goes down
(same as how it's done in neigh_flush_dev() for regular neigh entries)
projects / thirdparty / kernel / linux.git / log
