realtek: mdio: use generic bus command function for RTL838x
The Realtek mdio bus works similar for all devices with only
minor family specific differences. Basically command issuing
follows a generic style. Write command type and a trigger.
Afterwards wait until the trigger flag goes back to zero (aka
"command complete"). Unify this sequence in a central helper.
RTL838x read/write callers of this helper use a strange style of
error handling by issuing goto statements. Refactor this for better
readability. Additionally remove all debug prints. These are not
needed as the central read/write handlers provide a common logging
mechanism.
Thomas Richard [Mon, 5 Jan 2026 15:31:36 +0000 (16:31 +0100)]
kernel: add OSNOISE_TRACER kernel config option
This adds the OSNOISE_TRACER kernel configuration option. The osnoise
tracer allows to track interference experienced by an application due to
activities inside the operating system (like NMIs, IRQs ...).
Thomas Richard [Wed, 21 Jan 2026 12:49:32 +0000 (13:49 +0100)]
loongarch64: set PREEMPT_NONE as default preemption model
Loongarch64 currently uses the PREEMPT_VOLUNTARY model, which is not
aligned with OpenWrt's default. Switch to PREEMPT_NONE for consistency with
OpenWrt's default configuration. Preemption model selection is now
available via menuconfig for further customization.
Thomas Richard [Wed, 21 Jan 2026 12:34:33 +0000 (13:34 +0100)]
mediatek: mt7623: set PREEMPT_NONE as default model
Mediatek/mt7623 currently uses the PREEMPT model, which is not aligned with
OpenWrt's default. Switch to PREEMPT_NONE for consistency with OpenWrt's
default configuration. Preemption model selection is now available via
menuconfig for further customization.
Signed-off-by: Thomas Richard <thomas.richard@bootlin.com> Link: https://github.com/openwrt/openwrt/pull/21413
[Added some config options again] Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Thomas Richard [Wed, 21 Jan 2026 12:08:23 +0000 (13:08 +0100)]
gemini: set PREEMPT_NONE as default preemption model
Gemini currently uses the PREEMPT model, which is not aligned with
OpenWrt's default. Switch to PREEMPT_NONE for consistency with OpenWrt's
default configuration. Preemption model selection is now available via
menuconfig for further customization.
Thomas Richard [Wed, 21 Jan 2026 10:54:14 +0000 (11:54 +0100)]
layerscape: set PREEMPT_NONE as default preemption model
Layerscape currently uses the PREEMPT model, which is not aligned with
OpenWrt's default. Switch to PREEMPT_NONE for consistency with
OpenWrt's default configuration. Preemption model selection is now
available via menuconfig for further customization.
Thomas Richard [Mon, 5 Jan 2026 12:46:52 +0000 (13:46 +0100)]
kernel: add preemption models
Introduce preemption model selection with PREEMPT_NONE as the default.
PREEMPT_NONE is the traditional Linux preemption model and also the best
choice for servers.
Hauke Mehrtens [Thu, 22 Jan 2026 22:54:33 +0000 (23:54 +0100)]
firewall: change synflood_protect option name
The `syn_flood` option name is deprecated, `synflood_protect` should
be used instead. firewall3 and firewall4 both support this option since
a long time. LuCI already replaces the option name.
https://github.com/openwrt/luci/commit/0abcb39b623f0eff0fbcdfb99fca8f3224701e86
David Yang [Mon, 19 Jan 2026 17:51:02 +0000 (01:51 +0800)]
iproute2: add dcb utility
KERNEL_DCB was introduced in 40f1db9cb11d, however the dcb utility is not
enabled for iproute2. Although DCB is not generally available among
Ethernet cards, not having the dcb utility renders it completely
unchangeable.
John Audia [Tue, 15 Jul 2025 09:46:52 +0000 (05:46 -0400)]
base-files: remove ftp user and group
Any ftpd package such as vsftpd or atftpd (are there others?) should
create their own unprivileged user. Both of the aforementioned
packages do not even use this ftp user and run as root. That should be
addressed in separate commits.
Linus Walleij [Wed, 21 Jan 2026 21:24:46 +0000 (22:24 +0100)]
gemini: support upgrade on reference designs
The Gemini reference design-derived devices uses a partition
format which is predictable and we can exploit this to offer
some proper upgrade path.
The kernel for these contains a hack to use this partition
format unaltered by combining the partitions "Kern" and "Ramdisk"
to one image with all of the kernel+ramdisk in memory.
Then the "Application" which is used for the rootfs go into its
own partition.
Following the pattern of the factory image we create three
images named zImage, rd.gz and hddapp.tgz (these filenames
are misleading! They are just required by the old firmware.)
and flash each individually with "mtd" during upgrades.
Since the IB-4220-V has a different layout with a bigger kernel
space we parameterize this so we can handle this too. (More
fixes are needed for that device though.)
A way to upgrade older OpenWrt on these platforms to the latest
and greatest will be to copy the file
target/linux/gemini/base-files/lib/upgrade/platform.sh
to /lib/upgrade/platform.sh
on your running system and then run sysupgrade from the image
produced after this patch.
The script is picky to sanity check the partitions before
commencing upgrade.
This was tested with a full sysupgrade on the iTian SQ201.
- phy_probe() for setup of structures
- config_init() for device setup (after reset)
RTL8214FC is a combo phy and the currently active port can be switched
with the SFP helper functions that are triggered during SFP insertion
and removal. In case a fibre SFP is inserted while booting the SFP
trigger is run between the above mentioned stages. During the final
setup in config_init() the phy is reset to the copper port. Thus no
link is available on fibre and the SFP must be reinserted for normal
operation.
For a consistent behaviour the fibre/copper port setup must run before
the SFP probing and not afterwards. Move the setup code from config_init()
into phy_probe().
Fixes: 10ae743 ("realtek: phy: simplify RTL8214FC configuration") Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de> Link: https://github.com/openwrt/openwrt/pull/21582 Signed-off-by: Robert Marko <robimarko@gmail.com>
realtek: phy: optimize RTL8214FC register usage during media set
There are some shortcomings in rtl8214fc_media_set()
- It always uses the mdio raw page (4095) of RTL838x for writes.
That is wrong when the phy is attached to an RTL839x (raw = 8191)
- It uses the internal write only extended page companion (29)
- The extended page content is not preserved
realtek: phy: save RTL8214FC extended page during power set
Changing the fibre/copper power of a RTL8214FC changes the
extended page via register 29. This is the write only companion
of register 30. The register is afterwards overwritten to 0.
Use the proper extended page register 30 and preserve its content
during the operation.
realtek: phy: convert confusing BMCR_PDOWN usage for RTL8214FC
Bit 11 of registers 16, 19, 20, 21 in page 0x266 denotes if a port
of a RTL8214FC is set/favoured to fibre (0) or copper (1). For unknown
reasons the bit was mixed with BMCR_PDOWN. Convert this to a meaningful
define.
realtek: mdio: rtl838x: activate combo PHY media detection
There is a misunderstanding about BIT(7) aka EX_PHY_MAN_24_27 in
SMI_GLB_CTRL register. The SDK sets/clears it at different places and
it is not clear what it is for. Observation shows that it is essential
for a working MAC_LINK_MEDIA_STS register.
A RTL838x device has usally two configurations
- port 24/26 are 2 serdes driven fiber ports
- port 24-27 are 4 PHY driven combo ports
In the combo case the above bit must be set so that a switch between
copper and fiber can be detected. Cleanup the MDIO initialization
and remove the unneeded bit handling in the DSA driver.
George Sapkin [Sat, 24 Jan 2026 02:30:48 +0000 (04:30 +0200)]
build: purge DESCRIPTION
DESCRIPTION field has been deprecated since 2007, yet internal exporting
causes intermittent package build failures due to the deprecation check,
even with packages that don't have that fields set. Replace all
DESCRIPTION exports with alternative variables defined through shexport
and shvar helpers and remove the check.
Fixes: 547b127d ("make kernel module packaging code reusable and use it in madwifi") Fixes: 7558f028 ("DESCRIPTION:= is obselete, so complain if it is used and use TITLE if no description is set") Fixes: cc435322 ("build: optimize target metadata dump") Fixes: d081edf7 ("build: clean up and optimize ipkg control generator code") Signed-off-by: George Sapkin <george@sapk.in> Link: https://github.com/openwrt/openwrt/pull/21668 Signed-off-by: Robert Marko <robimarko@gmail.com>
George Sapkin [Fri, 23 Jan 2026 18:07:53 +0000 (20:07 +0200)]
imagebuilder: fix parsing ABI for apk packages
Fix parsing ABI when package has multiple tags and apk returns them in a
single line.
Fixes: 31cdd13d ("imagebuilder: add ABI suffix to packages when using apk") Signed-off-by: George Sapkin <george@sapk.in> Link: https://github.com/openwrt/openwrt/pull/21660 Signed-off-by: Robert Marko <robimarko@gmail.com>
Hauke Mehrtens [Thu, 22 Jan 2026 22:14:21 +0000 (23:14 +0100)]
lantiq: AVM Fritz 736x: fix PCIe reset GPIO
The vr9.dtsi configures the GPIO 38 as reset GPIO. Also the fon LED is
configured on GPIO 38. This conflicts and makes the probing of the PCIe
controller fail in OpenWrt 25.12.
The AVM GPL source code configured GPIO 21 as PCIe reset.
Some SPI-NAND driver macro definitions and function parameters have
been changed in the latest release[3]. Hence we also had to rework
the related local patches to follow the upstream changes.
Signed-off-by: Shiji Yang <yangshiji66@outlook.com>
[daniel@makrotopia.org: tested MT7622 SNAND and SPI-NOR] Signed-off-by: Daniel Golle <daniel@makrotopia.org>
George Sapkin [Fri, 23 Jan 2026 16:58:36 +0000 (18:58 +0200)]
imagebuilder: skip repository file when standalone
Standalone image builder doesn't have a repositories file as all
packages are included, which causes:
ERROR: failed to read repositories: PATH_TO_BUILDER/repositories: No such file or directory
The images are still built, so this is more of an informational error.
Pass related argument to apk only when CONFIG_IB_STANDALONE is not set.
Fixes: a8d17c21 ("imagebuilder: actually support IB from buildbot") Signed-off-by: George Sapkin <george@sapk.in> Link: https://github.com/openwrt/openwrt/pull/21658 Signed-off-by: Robert Marko <robimarko@gmail.com>
Gabriel Roper [Tue, 7 Oct 2025 04:21:40 +0000 (21:21 -0700)]
d1: Add device tree configuration for D1 RGB LED Controller
This adds the device tree patches written by Samuel Holland
to the kernel. The driver was merged into mainline Linux in 6.8,
but the device tree patches were never merged into mainline.
Signed-off-by: Gabriel Roper <lockheedmaniac@gmail.com>
The RTL8218D currently relies on proper U-Boot configuration. In
case that is not possible, provide a basic setup sequence that can
bring the PHY "alive". The SDK provides multiple configuration
sequences for two operation modes (XSGMII or QSGMII) and the different
SoC families. Due to limited testing resources only provide a setup
for RTL93xx devices and both modes at the moment.
John Crispin [Fri, 19 Sep 2025 15:05:30 +0000 (17:05 +0200)]
wifi-scripts: refactor iwinfo.uc to support dynamic data updates
Moved interface discovery and data population into an exported update()
function that can be called on-demand to refresh wireless interface
information. This allows using iwinfo.uc as a library inside daemons.
Felix Fietkau [Tue, 13 Jan 2026 12:37:50 +0000 (12:37 +0000)]
build: fix stale package install stamps for unselected variants
When a package variant changes from =y to =m, its .install stamp file
was not being cleaned up, causing both variants to appear in the
package install list.
Fix by making compile depend on prepare-package-install, ensuring
the stamp file is always cleared. Only =y packages recreate it.
Felix Fietkau [Tue, 13 Jan 2026 09:31:22 +0000 (09:31 +0000)]
build: auto-generate CONFLICTS for package variants sharing PROVIDES
Packages using PROVIDES to declare virtual package alternatives (like
tc-tiny, tc-bpf, tc-full all providing 'tc') could be simultaneously
selected as =y, causing installation conflicts. The PROVIDES mechanism
only handles dependency resolution, not mutual exclusion.
Add add_implicit_provides_conflicts() to automatically generate CONFLICTS
from default variants to non-default variants sharing the same PROVIDES.
This ensures only one variant can be built-in (=y) at a time.
Skip generating implicit conflicts when the non-default already has
explicit CONFLICTS with the default, to avoid Kconfig dependency cycles
with the select-based dependency resolution.
Shiji Yang [Sat, 3 Jan 2026 04:59:19 +0000 (12:59 +0800)]
mediatek: dts: mt7981: remove internal phy LED pinctrl
This default pinctrl doesn't work at all. We must add pinctrl phandle
to the ethernet-phy node or mac node. This patch removed the internal
phy LED pinctrl group definitions from the SOC dtsi because it is
currently only used by one device. Some incorrect LED pinctrls also
have been removed because these devices do not have hardware phy LED.
The PCIe NIC WED offload features depend on it. Also add the missing
"syscon" compatible for the wed_pcie node as it's accessed by
syscon_regmap_lookup_by_phandle().
Shiji Yang [Fri, 19 Dec 2025 11:39:55 +0000 (19:39 +0800)]
mediatek: dts: mt7981: remove useless nodes
These stuffs were copied from the vendor SDK. There are currently
no drivers compatible with them. The pending upstream patches did
not include them either.
This is in preparation for the next update to EcoNet Ethernet driver
which will be using real resets rather than bit-bashing. Patches are
backported up to the current state of clk-next because I intend on
upstreaming these patches soon.
Tianling Shen [Tue, 20 Jan 2026 02:12:12 +0000 (10:12 +0800)]
rockchip: fix set default serial_{addr,port} in boot script
On some SoCs like RK3399, U-Boot no longer export the serial
address in $stdout variable, it has been changed to something like
`stdout=serial,vidconsole` since video supported was added. So
detect $soc variable instead, this also makes the script more clear.
Addtional note: rk3566 is reported as rk3568, rk3582 is reported as
rk3588 in $soc.
realtek: mdio: derive phy smi-bus from parent node
The dts for RTL93xx devices has duplicate data about the
smi bus of a phy node. The parent node declares the number
of the bus and the realtek,smi-address attribute does the
same.
Remove the bus part from the realtek,smi-address attribute
and lookup the bus from the parent node. While we are here
remove all realtek,smi-address attributes where phy id
matches the bus address. The driver will use that as a
fallback.
Switch the mdio bus topology for devices that have their phys
attached to bus 1-3. This does not have any impact because
the mdio driver was completly redisgned
With this commit the bus id is stored twice. Once in the (new)
bus and in the (old) realtek,smi-address property. E.g.
&mdio_bus1 {
reg = <1>; <<< bus id
phy24: ethernet-phy@24 {
reg = <26>;
compatible = "ethernet-phy-ieee802.3-c22";
realtek,smi-address = <1 2>; <<< bus & address id
};
};
realtek: mdio/dsa: refactor mdio bus initialization
The mdio driver currently determines the smi bus and address from the
realtek,smi-address attribute of the phy. To better reflect the
topology and align with upstream, the phys should be relocated below
their associated bus. As an interim solution the following dts notation
is in focus.
- the phy reg property still denotes the port number
- the bus number can be derived from the parent bus node.
- the bus address is taken from realtek,smi-address
Refactor bus initialization so it can handle phy nodes below
multiple bus nodes.
During initialization the mdio driver registers phys with the
iterator "pn". To make clear that it is a phy address rename it
to "addr".
While we are here improve the upper bound check of the maximum
possible address. This is the family specific cpu port and not
the generic upper bound constant for all devices.
realtek: mdio: register mdio bus at controller node
In the future the mdio controller will have multiple busses
defined in the dts below the controller node. Nevertheless
it will still hand out only one single bus to the kernel.
Attach the (exported) bus to the controller node instead of
the single (dts) bus subnode.
With this change the mdio lookup in the dsa driver must
be changed to point to the mdio controller node too.
realtek: dsa: Remove family check from port_get_stp_state()
The device specific stp_get() functions can return the state
of a given port individually. No need to disassemble the
device specific state table. Additionally change function
prefix.
The family specific stp_get() functions have a new return value
that gives the stp status of a given port. Instead of a constant
zero value provide the real data.
While we are here align the RTL930x implementation with the
other ones.
- Drop the debug output
- Make use of priv->r->tbl_access_data_0()
The device specific stp_get() function returns a device specific
state to the generic caller. So the caller must use a if/else
statement depending on the family to look into the retrieved data.
Change the signature of the function so it can return a device
independent state of a given port in the future. Additionally
use the new function prefix.
Ryan Chen [Sun, 18 Jan 2026 02:51:14 +0000 (20:51 -0600)]
wifi-scripts: fix WiFi 6E discovery for 6GHz 320MHz operation
WiFi 6E (802.11ax) clients cannot discover 6GHz APs operating at
320MHz because the HE Operation element contains uninitialized
center frequency values.
For EHT320 mode, the code sets eht_oper_centr_freq_seg0_idx but not
the corresponding HE values. Later, the HE values are copied from
VHT values, but VHT is not used on 6GHz, leaving he_oper_chwidth
and he_oper_centr_freq_seg0_idx at 0. This causes WiFi 6E clients
to see incorrect channel width information, making the AP invisible
to them during scanning.
Fix this by:
1. Setting he_oper_chwidth to 3 (160MHz) for EHT320 mode
2. Computing he_oper_centr_freq_seg0_idx based on the 160MHz segment
that contains the primary channel
3. Preserving these pre-set values instead of overwriting them with
uninitialized VHT values
WiFi 7 clients continue to see 320MHz operation via the EHT Operation
element, while WiFi 6E clients can now discover and connect at 160MHz.
Robert Marko [Wed, 21 Jan 2026 14:27:45 +0000 (15:27 +0100)]
microchipsw: lan969x: use network driver as a module
Due to PTP conflicts with CONFIG_ALL_KMODS which causes the end images
to end up with no switchdev driver included, lets switch to using the
switchdev and PHY drivers as modules instead.
Fixes: 88a404a2d177 ("microchipsw: add support for Microchip LAN969x switches") Signed-off-by: Robert Marko <robert.marko@sartura.hr>
Jonas Gorski [Wed, 14 Jan 2026 09:30:55 +0000 (10:30 +0100)]
strace: use bundled kernel headers
Build against the bundled kernel headers instead of using the current
kernel ones. This ensures strace is using the kernel headers it is
written against, and not a random one that may contain breaking uapi
changes (which happen from time to time).
Fixes build against 6.18 final and recent LTS/stable kernels that got
minor uapi breakages (rename of a 6.18 introduced #define and a struct
field).
Signed-off-by: Jonas Gorski <jonas.gorski@gmail.com>
Jan Hoffmann [Mon, 19 Jan 2026 17:45:46 +0000 (18:45 +0100)]
realtek: make all link speeds work for RTL8221B in Zyxel XGS1210-12 rev B1
SGMII only works correctly on this device if inband auto-negotiation is
enabled. Configure the PHY for SGMII and in-band mode in the device tree
to make this happen.
For 2.5G link speeds the PHY will still switch to 2500Base-X without AN.
The same configuration also works on RTL8226, so it is fine to apply
this change to the A1 revision of XGS1010-12/XGS1210-12 as well.
Kamil Jońca [Tue, 6 Jan 2026 08:03:33 +0000 (09:03 +0100)]
ramips: mt76x8: add support for Cudy LT400E v1
This commit adds support for the Cudy LT400E v1,
an indoor 4G/LTE WiFi router.
It has 4x10/100 Mbps Ethernet ports, a 2.4 GHz radio,
and a Quectel 4G/LTE EC200A modem.
Installation:
The stock firmware is a customized variant of OpenWrt, which implements
a signature check that only allows flashing official firmware. Cudy
offers, on their website [1][6], intermediate OpenWrt firmware images
which do not implement the signature check. After flashing the
intermediate image, the upstream (i.e., official) OpenWrt image can be
installed. See the device's entry in OpenWrt's Wiki for more details [2].
Cellular network:
The Quectel EC200A modem is internally connected to the SoC via USB, and
operates in ECM mode, supported by the CDC Ether driver [3]. When
inserting a SIM card in the slot, if no PIN is needed, the network
interface "usb0" is brought up and the cellular connection works out of
the box. To interact with the modem directly (unlock the PIN, change APN
and PDP settings, etc.), AT commands can be sent with picocom to serial
port /dev/ttyUSB2 [4].
Recovery:
The stock firmware can be recovered via TFTP, as the bootloader (based
on U-Boot) implements a recovery client. Set up a TFTP server on your
computer with IP 192.168.88.1/24 serving the stock firmware from Cudy's
website renamed to "recovery.bin". Press the Reset button (keep it
pressed), power on the device, wait for the TFTP server to send the
recovery.bin file, then release the Reset button. The router will take a
couple of minutes to reboot and set up the stock firmware. See the
device's entry in OpenWrt's Wiki for more details [2].
Discussion:
Use the dedicated thread in OpenWrt's forum to discuss any aspects [5].
Acknowledgments:
The code to support this device is largely based on the sources kindly
provided by Cudy in compliance with the GPL license. Thanks for
supporting the OpenWrt community.
Chen Minqiang [Fri, 12 Dec 2025 02:58:16 +0000 (10:58 +0800)]
mediatek: filogic: add support for Tenda BE12 Pro
This patch adds support for Tenda BE12 Pro.
The stock bootloader requires a specific 16-byte header (Magic: "God1")
to correctly verify and boot the Linux kernel from flash. A new image
command `tenda-mkdualimageheader` is added to generate this header and
append the checksum/size information required for the device to boot
successfully.
MAC Addresses (NVMEM):
- Base MAC located at Factory partition offset 0x4
- gmac2 (WAN) : Base - 2 (Label MAC)
- gmac0 (LAN) : Base - 1
- gmac1 (LAN) : Base - 3
- 2.4 GHz : Base
- 5 GHz : Base + 1
Installation:
1. Connect via UART and interrupt U-Boot by pressing 'Ctrl+C' during boot.
2. Set up a TFTP server on a computer (IP 192.168.1.2) containing
'openwrt-initramfs.bin'.
3. Load the initramfs image:
tftpboot openwrt-initramfs.bin
4. Boot the image:
bootm
5. Once booted into OpenWrt initramfs, perform a sysupgrade with the
sysupgrade image.
Paul Donald [Tue, 6 Jan 2026 00:53:25 +0000 (01:53 +0100)]
ucode: enable ucode-mod-zlib
The module exists in ucode and has been present for a while
but has not been enabled. It provides the ucode zlib module
for handling gzip and zlib compression in ucode scripts.
Chukun Pan [Thu, 10 Jul 2025 14:15:05 +0000 (22:15 +0800)]
generic: restore network port name for r8169 LED
Currently, the LED names generated by the r8169 driver are the
network port names renamed by udev, such as enp1s0. Add a hack
patch to restore the interface name to match OpenWrt.
The gmac definition has an offset of 1 compared to the label MAC. Fixes: a55ab9e1343e85021253e9c55f67adf33ad68f09 ("mediatek: filogic: prevent faulty mac address assignment") Signed-off-by: Maximilian Baumgartner <aufhaxer@googlemail.com> Link: https://github.com/openwrt/openwrt/pull/21560 Signed-off-by: Robert Marko <robimarko@gmail.com>
Jonas Jelonek [Sun, 11 Jan 2026 22:28:35 +0000 (22:28 +0000)]
realtek: pcs: rtl930x: rework SerDes mode setting
Rework the SerDes mode setting to have a better logical structure,
similar to how RTL931x currently does it. Mode setting is grouped into a
MAC setup and IP core setup. Most modes just need to trigger the MAC
setup and then they work, otherwise some complex sequence is needed.
To achieve that, rename several functions and group their content
differently. While at it, rename some constants to use the RTPCS_ prefix.
RTL930x has some helpers which take care of writing the actual mode
value to the corresponding register. They exist for the MAC setup SerDes
mode, the USXGMII submode and the SerDes IP core mode.
To reflect that, adjust the naming of these helpers accordingly. Most
importantly, prefix the helper names with '__'. This is a common
convention to denote helpers/functions which are internal, not intended
for general use and skip certain checks. Those conditions apply to these
helpers. Though they are still used quite often here, this might change
with further cleanups.
projects / thirdparty / openwrt.git / log
