Accelerate TX packet duplication with HSR rings.
This is only possible with the NPI-based "ocelot" tagging protocol, not
with "ocelot-8021q", because the latter does not use dsa_xmit_port_mask().
This has 2 implications:
- Depending on tagging protocol, we should set (or not set) the offload
feature flags. Switching tagging protocols is done with ports down, by
design. Additional calls to dsa_port_simple_hsr_join() can be put in
the ds->ops->change_tag_protocol() path, as I had originally tried,
but this would not work: dsa_user_setup_tagger() would later clear
the feature flag that we just set. So the additional call to
dsa_port_simple_hsr_join() should sit in the ds->ops->port_enable()
call.
- When joining a HSR ring and we are currently using "ocelot-8021q",
there are cases when we should return -EOPNOTSUPP (pessimistic) and
cases when we shouldn't (optimistic). In the pessimistic case, it is a
configuration that the port won't support even with the right tagging
protocol. Distinguishing between these 2 cases matters because if we
just return -EOPNOTSUPP regardless, we lose the dp->hsr_dev pointer
and can no longer replay the offload later for the optimistic case,
from felix_port_enable().
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251130131657.65080-8-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
It turns out that HSR offloads are so fine-grained that many DSA
switches can do a small part even though they weren't specifically
designed for the protocols supported by that driver (HSR and PRP).
Specifically NETIF_F_HW_HSR_DUP - it is simple packet duplication on
transmit, towards all (aka 2) ports members of the HSR device.
For many DSA switches, we know how to duplicate a packet, even though we
never typically use that feature. The transmit port mask from the
tagging protocol can have multiple bits set, and the switch should send
the packet once to every port with a bit set from that mask.
Nonetheless, not all tagging protocols are like this, and sometimes the
port is a single numeric value rather than a bit mask. For that reason,
and also because switches can sometimes change tagging protocols for
different ones, we need to make HSR offload helpers opt-in.
For devices that can do nothing else HSR-specific, we introduce
dsa_port_simple_hsr_join() and dsa_port_simple_hsr_leave(). These
functions monitor when two user ports of the same switch are part of the
same HSR device, and when that condition is true, they toggle the
NETIF_F_HW_HSR_DUP feature flag of both net devices.
Normally only dsa_port_simple_hsr_join() and dsa_port_simple_hsr_leave()
are needed. The dsa_port_simple_hsr_validate() helper is just to see
what kind of configuration could be offloadable using the generic
helpers. This is used by switch drivers which are not currently using
the right tagging protocol to offload this HSR ring, but could in
principle offload it after changing the tagger.
Suggested-by: David Yang <mmyangfl@gmail.com>
Cc: "Alvin Šipraga" <alsi@bang-olufsen.dk>
Cc: Chester A. Unal" <chester.a.unal@arinc9.com>
Cc: "Clément Léger" <clement.leger@bootlin.com>
Cc: Daniel Golle <daniel@makrotopia.org>
Cc: DENG Qingfang <dqfext@gmail.com>
Cc: Florian Fainelli <florian.fainelli@broadcom.com>
Cc: George McCollister <george.mccollister@gmail.com>
Cc: Hauke Mehrtens <hauke@hauke-m.de>
Cc: Jonas Gorski <jonas.gorski@gmail.com>
Cc: Kurt Kanzenbach <kurt@linutronix.de>
Cc: Linus Walleij <linus.walleij@linaro.org>
Cc: Sean Wang <sean.wang@mediatek.com>
Cc: UNGLinuxDriver@microchip.com
Cc: Woojung Huh <woojung.huh@microchip.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251130131657.65080-6-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
As discussed here:
https://lore.kernel.org/netdev/20240620090210.drop6jwh7e5qw556@skbuf/
the fact is that the xrs700x.c driver only supports offloading
HSR_PT_SLAVE_A and HSR_PT_SLAVE_B (which were the only port types at the
time the offload was written, _for this driver_).
Up until now, the API did not explicitly tell offloading drivers what
port has what role. So xrs700x can get confused and think that it can
support a configuration which it actually can't. There was a table in
the attached link which gave an example:
$ ip link add name hsr0 type hsr slave1 swp0 slave2 swp1 \
interlink swp2 supervision 45 version 1
HSR_PT_SLAVE_A HSR_PT_SLAVE_B HSR_PT_INTERLINK
----------------------------------------------------------------
user
space 0 1 2
requests
----------------------------------------------------------------
XRS700X
driver 1 2 -
understands
The switch would act as if the ring ports were swp1 and swp2.
Now that we have explicit hsr_get_port_type() API, let's use that to
work around the unintended semantical changes of the offloading API
brought by the introduction of interlink ports in HSR.
Fixes: 5055cccfc2 ("net: hsr: Provide RedBox support (HSR-SAN)")
Cc: Lukasz Majewski <lukma@denx.de>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: George McCollister <george.mccollister@gmail.com>
Link: https://patch.msgid.link/20251130131657.65080-5-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This mirrors what we do in dsa_port_lag_leave() and
dsa_port_bridge_leave(): when ds->ops->port_hsr_join() returns
-EOPNOTSUPP, we fall back to a software implementation where dp->hsr_dev
is NULL, and the unoffloaded port is no longer bothered with calls from
the HSR layer.
This helps, for example, with interlink ports which current DSA drivers
don't know how to offload. We have to check only in port_hsr_join() for
the port type, then in port_hsr_leave() we are sure we're dealing only
with known port types.
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251130131657.65080-4-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Since the introduction of HSR_PT_INTERLINK in commit 5055cccfc2 ("net:
hsr: Provide RedBox support (HSR-SAN)"), we see that different port
types require different settings for hardware offload, which was not the
case before when we only had HSR_PT_SLAVE_A and HSR_PT_SLAVE_B. But
there is currently no way to know which port is which type, so create
the hsr_get_port_type() API function and export it.
When hsr_get_port_type() is called from the device driver, the port can
must be found in the HSR port list. An important use case is for this
function to work from offloading drivers' NETDEV_CHANGEUPPER handler,
which is triggered by hsr_portdev_setup() -> netdev_master_upper_dev_link().
Therefore, we need to move the addition of the hsr_port to the HSR port
list prior to calling hsr_portdev_setup(). This makes the error
restoration path also more similar to hsr_del_port(), where
kfree_rcu(port) is already used.
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Lukasz Majewski <lukma@denx.de>
Signed-off-by: Xiaoliang Yang <xiaoliang.yang_1@nxp.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Łukasz Majewski <lukma@nabladev.com>
Link: https://patch.msgid.link/20251130131657.65080-3-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Parthiban Veerasooran says:
====================
Add SQI and SQI+ support for OATC14 10Base-T1S PHYs and Microchip T1S driver
This patch series adds Signal Quality Indicator (SQI) and enhanced SQI+
support for OATC14 10Base-T1S PHYs, along with integration into the
Microchip T1S PHY driver. This enables ethtool to report the SQI value for
OATC14 10Base-T1S PHYs.
====================
Link: https://patch.msgid.link/20251201032346.6699-1-parthiban.veerasooran@microchip.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add support for Signal Quality Index (SQI) reporting in the
Microchip T1S PHY driver for LAN867x Rev.D0 (OATC14-compliant) PHYs.
This patch registers the following callbacks in the microchip_t1s driver
structure:
- .get_sqi - returns the current SQI value
- .get_sqi_max - returns the maximum SQI value
This enables ethtool to report the SQI value for LAN867x Rev.D0 PHYs.
Signed-off-by: Parthiban Veerasooran <parthiban.veerasooran@microchip.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20251201032346.6699-3-parthiban.veerasooran@microchip.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add support for reading Signal Quality Indicator (SQI) and enhanced SQI+
from OATC14 10Base-T1S PHYs.
- Introduce MDIO register definitions for DCQ_SQI and DCQ_SQIPLUS.
- Add `genphy_c45_oatc14_get_sqi_max()` to return the maximum supported
SQI/SQI+ level.
- Add `genphy_c45_oatc14_get_sqi()` to return the current SQI or SQI+
value.
- Update `include/linux/phy.h` to expose the new APIs.
SQI+ capability is read from the Advanced Diagnostic Features Capability
register (ADFCAP). If SQI+ is supported, the driver calculates the value
from the MSBs of the DCQ_SQIPLUS register; otherwise, it falls back to
basic SQI (0-7 levels). This enables ethtool to report the SQI value for
OATC14 10Base-T1S PHYs.
Open Alliance TC14 10BASE-T1S Advanced Diagnostic PHY Features
Specification ref:
https://opensig.org/wp-content/uploads/2025/06/OPEN_Alliance_10BASE-T1S_Advanced_PHY_features_for-automotive_Ethernet_V2.1b.pdf
Signed-off-by: Parthiban Veerasooran <parthiban.veerasooran@microchip.com>
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Link: https://patch.msgid.link/20251201032346.6699-2-parthiban.veerasooran@microchip.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Tariq Toukan says:
====================
net/mlx5e: Enhance DCBNL get/set maxrate code
This series by Gal introduces multiple small code quality improvements
for the DCBNL operations mlx5e_dcbnl_ieee_[gs]etmaxrate().
No functional change.
====================
Link: https://patch.msgid.link/1764498334-1327918-1-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
MLX5E_100MB and MLX5E_1GB defines are confusing, MLX5E_100MB is not
equal to 100 * MEGA, and MLX5E_1GB is not equal to one GIGA, as they
hide the Kbps rate conversion required for ieee_maxrate.
Replace hardcoded bandwidth conversion values with standard unit
definitions from linux/units.h. Rename MLX5E_100MB/MLX5E_1GB to
MLX5E_100MB_TO_KB/MLX5E_1GB_TO_KB to clarify these are conversion
factors to Kbps, not absolute bandwidth values.
Signed-off-by: Gal Pressman <gal@nvidia.com>
Reviewed-by: Nimrod Oren <noren@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/1764498334-1327918-5-git-send-email-tariqt@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Jonas Gorski says:
====================
net: dsa: b53: fix ARL accesses for BCM5325/65 and allow VID 0
ARL entries on BCM5325 and BCM5365 were broken significantly in two
ways:
- Entries for the CPU port were using the wrong port id, pointing to a
non existing port.
- Setting the VLAN ID for entries was not done, adding them all to VLAN
0 instead.
While the former technically broke any communication to the CPU port,
with the latter they were added to the currently unused VID 0, so they
never became effective. Presumably the default PVID was set to 1 because
of these issues 0 was broken (and the root cause not found).
So fix writing and reading entries on BCM5325/65 by first fixing the CPU
port entries, then fixing setting the VLAN ID for entries.
Finally, re-allow VID 0 for BCM5325/65 to allow the whole 1-15 VLAN ID
range to be available to users, and align VLAN handling with all other
switch chips.
====================
Link: https://patch.msgid.link/20251128080625.27181-1-jonas.gorski@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
BCM5325/65's ARL entry registers do not contain the VID, only the search
result register does. ARL entries have a separate VID entry register for
the index into the VLAN table.
So make ARL entry accessors use the VID entry registers instead, and
move the VLAN ID field definition to the search register definition.
Fixes: c45655386e ("net: dsa: b53: add support for FDB operations on 5325/5365")
Signed-off-by: Jonas Gorski <jonas.gorski@gmail.com>
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Link: https://patch.msgid.link/20251128080625.27181-7-jonas.gorski@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
We currently use the mask 0xf for writing and reading b53_entry::port,
but this is only correct for unicast ARL entries. Multicast ARL entries
use a bitmask, and 0xf is not enough space for ports > 3, which includes
the CPU port.
So extend the mask accordingly to also fit port 4 (bit 4) and MII (bit
5). According to the datasheet the multicast port mask is [60:48],
making it 12 bit wide, but bits 60-55 are reserved anyway, and collide
with the priority field at [60:59], so I am not sure if this is valid.
Therefore leave it at the actual used range, [53:48].
The ARL search result register differs a bit, and there the mask is only
[52:48], so only spanning the user ports. The MII port bit is
contained in the Search Result Extension register. So create a separate
search result parse function that properly handles this.
Fixes: c45655386e ("net: dsa: b53: add support for FDB operations on 5325/5365")
Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
Signed-off-by: Jonas Gorski <jonas.gorski@gmail.com>
Link: https://patch.msgid.link/20251128080625.27181-6-jonas.gorski@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Raju Rangoju says:
====================
amd-xgbe: schedule NAPI on RBU event
During the RX overload the Rx buffers may not be refilled, trying to
schedule the NAPI when an Rx Buffer Unavailable is signaled may help in
improving the such situation, in case we missed an IRQ.
====================
Link: https://patch.msgid.link/20251129175016.3034185-1-Raju.Rangoju@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
To capture TX packets during a test, we are currently intercepting the
dst->output with an implementation that adds the transmitted packet to
a skb queue attached to the test-specific mock dst. The netdev itself is
not involved in the test TX path.
Instead, we can just use our test device to stash TXed packets for later
inspection by the test. This means we can include the actual
mctp_dst_output() implementation in the test (by setting dst.output in
the test case), and don't need to be creating fake dst objects, or their
corresponding skb queues.
We need to ensure that the netdev is up to allow delivery to
ndo_start_xmit, but the tests assume active devices at present anyway.
Signed-off-by: Jeremy Kerr <jk@codeconstruct.com.au>
Link: https://patch.msgid.link/20251126-dev-mctp-test-tx-queue-v2-1-4e5bbd1d6c57@codeconstruct.com.au
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Testing in two circumstances:
1. back to back optical SFP+ connection between two LS1028A-QDS ports
with the SCH-26908 riser card
2. T1042 with on-board AQR115 PHY using "OCSGMII", as per
https://lore.kernel.org/lkml/aIuEvaSCIQdJWcZx@FUE-ALEWI-WINX/
strongly suggests that enabling in-band auto-negotiation is actually
possible when the lane baud rate is 3.125 Gbps.
It was previously thought that this would not be the case, because it
was only tested on 2500base-x links with on-board Aquantia PHYs, where
it was noticed that MII_LPA is always reported as zero, and it was
thought that this is because of the PCS.
Test case #1 above shows it is not, and the configured MII_ADVERTISE on
system A ends up in the MII_LPA on system B, when in 2500base-x mode
(IF_MODE=0).
Test case #2, which uses "SGMII" auto-negotiation (IF_MODE=3) for the
3.125 Gbps lane, is actually a misconfiguration, but it is what led to
the discovery.
There is actually an old bug in the Lynx PCS driver - it expects all
register values to contain their default out-of-reset values, as if the
PCS were initialized by the Reset Configuration Word (RCW) settings.
There are 2 cases in which this is problematic:
- if the bootloader (or previous kexec-enabled Linux) wrote a different
IF_MODE value
- if dynamically changing the SerDes protocol from 1000base-x to
2500base-x, e.g. by replacing the optical SFP module.
Specifically in test case #2, an accidental alignment between the
bootloader configuring the PCS to expect SGMII in-band code words, and
the AQR115 PHY actually transmitting SGMII in-band code words when
operating in the "OCSGMII" system interface protocol, led to the PCS
transmitting replicated symbols at 3.125 Gbps baud rate. This could only
have happened if the PCS saw and reacted to the SGMII code words in the
first place.
Since test #2 is invalid from a protocol perspective (there seems to be
no standard way of negotiating the data rate of 2500 Mbps with SGMII,
and the lower data rates should remain 10/100/1000), in-band auto-negotiation
for 2500base-x effectively means Clause 37 (i.e. IF_MODE=0).
Make 2500base-x be treated like 1000base-x in this regard, by removing
all prior limitations and calling lynx_pcs_config_giga().
This adds a new feature: LINK_INBAND_ENABLE and at the same time fixes
the Lynx PCS's long standing problem that the registers (specifically
IF_MODE, but others could be misconfigured as well) are not written by
the driver to the known valid values for 2500base-x.
Co-developed-by: Alexander Wilhelm <alexander.wilhelm@westermo.com>
Signed-off-by: Alexander Wilhelm <alexander.wilhelm@westermo.com>
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Link: https://patch.msgid.link/20251125103507.749654-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This removes some noise that can be distracting while looking at
selftests by redirecting socat stderr to /dev/null.
Before this commit, netcons_basic would output:
Running with target mode: basic (ipv6)
2025/11/29 12:08:03 socat[259] W exiting on signal 15
2025/11/29 12:08:03 socat[271] W exiting on signal 15
basic : ipv6 : Test passed
Running with target mode: basic (ipv4)
2025/11/29 12:08:05 socat[329] W exiting on signal 15
2025/11/29 12:08:05 socat[322] W exiting on signal 15
basic : ipv4 : Test passed
Running with target mode: extended (ipv6)
2025/11/29 12:08:08 socat[386] W exiting on signal 15
2025/11/29 12:08:08 socat[386] W exiting on signal 15
2025/11/29 12:08:08 socat[380] W exiting on signal 15
extended : ipv6 : Test passed
Running with target mode: extended (ipv4)
2025/11/29 12:08:10 socat[440] W exiting on signal 15
2025/11/29 12:08:10 socat[435] W exiting on signal 15
2025/11/29 12:08:10 socat[435] W exiting on signal 15
extended : ipv4 : Test passed
After these changes, output looks like:
Running with target mode: basic (ipv6)
basic : ipv6 : Test passed
Running with target mode: basic (ipv4)
basic : ipv4 : Test passed
Running with target mode: extended (ipv6)
extended : ipv6 : Test passed
Running with target mode: extended (ipv4)
extended : ipv4 : Test passed
Signed-off-by: Andre Carvalho <asantostc@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20251129-netcons-socat-noise-v1-1-605a0cea8fca@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
New NIPA installation had been reporting a few flaky tests.
arp_ndisc_evict_nocarrier is most flaky of them all.
I suspect that the flakiness is due to udev swapping the MAC
addresses on the interfaces. Extend the message in
arp_ndisc_evict_nocarrier to hint at this potential issue.
Having the neigh get fail right after ping is rather unusual,
unless udev changes the MAC addr causing a flush in the meantime.
Link: https://patch.msgid.link/20251127194556.2409574-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Breno Leitao says:
====================
net: freescale: migrate to .get_rx_ring_count() ethtool callback
This series migrates Freescale network drivers to use the new .get_rx_ring_count()
ethtool callback introduced in commit 84eaf4359c ("net: ethtool: add
get_rx_ring_count callback to optimize RX ring queries").
The new callback simplifies the .get_rxnfc() implementation by removing
ETHTOOL_GRXRINGS handling and moving it to a dedicated callback. This provides
a cleaner separation of concerns and aligns these drivers with the modern
ethtool API.
The series updates the following Freescale drivers:
- enetc
- dppa2
- gianfar
====================
Link: https://patch.msgid.link/20251128-gxring_freescale-v1-0-22a978abf29e@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Convert the enetc driver to use the new .get_rx_ring_count
ethtool operation instead of implementing .get_rxnfc for handling
ETHTOOL_GRXRINGS command. This simplifies the code in two ways:
1. For enetc_get_rxnfc(): Remove the ETHTOOL_GRXRINGS case from the
switch statement while keeping other cases for classifier rules.
2. For enetc4_get_rxnfc(): Remove it completely and use
enetc_get_rxnfc() instead.
Now on, enetc_get_rx_ring_count() is the callback that returns the
number of RX rings for enetc driver.
Also, remove the documentation around enetc4_get_rxnfc(), which was not
matching what the function did(?!).
Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20251128-gxring_freescale-v1-3-22a978abf29e@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
