Jonas Jelonek [Fri, 27 Feb 2026 23:24:29 +0000 (23:24 +0000)]
realtek: dts: drop SYS LED of XikeStor SKS8310-8X
The sys_led node in the device tree is wrong and doesn't work. On GPIO
23 nothing seems to be connected. The switch has a SYS LED but as with a
lot other Realtek switches, this LED is connected to the pin A0, which
can be driven as GPIO0 or system LED. The hardware-assisted behavior is
configured during boot by u-boot.
Though it is be possible to deactivate the hardware-assisted behavior
and give userspace control over the LED, it doesn't work for this
switch. XikeStor apparently decided to hook the SYS LED to some kind of
external watchdog. This one seems to expect a LOW signal on that line
periodically. This leaves room for two options:
1) keep behavior and let the LED blink as configured by u-boot
2) turn off LED completely
Option 2) is rather odd and may confuse users, thus it is better to keep
the behavior as-is for now. u-boot configures the LED to blink fast
which might be annoying. As soon as we have a way in the device tree to
specify different blinking frequencies as supported by the hardware,
this can be smoothened.
Jonas Jelonek [Fri, 27 Feb 2026 22:38:02 +0000 (22:38 +0000)]
realtek: dts: add monitor IC node for XikeStor SKS8310-8X
The XikeStor SKS8310-8X has a monitoring IC LM75B from National
Instruments on board. This was missed when support for the device was
added. Add it now and also add the corresponding kmod to the device
packages.
Fixes: 62d50fb196 ("realtek: add support for XikeStor SKS8310-8X") Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com> Link: https://github.com/openwrt/openwrt/pull/22211 Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Jonas Jelonek [Fri, 27 Feb 2026 22:31:55 +0000 (22:31 +0000)]
realtek: use nvmem for mac address for XikeStor SKS8310-8X
Define an nvmem cell in the device tree of XikeStor SKS8310-8X to
reference that as a source for the MAC address of the switch (ports).
This eliminates the need for the userspace script to read and parse the
MAC address.
The driver covers phys and not ports. Adapt the iterator to
align with this naming convention. While we are here convert
some open coding to this iterator macro.
The mdio driver is about phys and not about the (dsa) port number
of the cpu. This was a generic alias when all of this was part
of a big large mdio/ethernet driver. Give the cpu_port variable
a new name that better fits to the mdio context.
In most drivers upstream use "ethernet-ports" instead of "ports"
in dts. Especially the upstream rtl9300 mdio driver uses this to
lookup the port/phy mapping. Do the same downstream. There is no
need to adapt the dsa driver because it scans the dts via
for_each_node_by_name(dn, "port").
Bevan Weiss [Wed, 25 Feb 2026 08:49:50 +0000 (19:49 +1100)]
packages: add PSE (PoE) packages
There are a number of pse chips already present in upstream Linux.
OpenWrt is starting to support a number of devices can contain various pse
chips. But having all the pse chip combinations defined at the target
level will result in bloat on images.
Present the current upstream Linux pse drivers as packages so that they
can be selectively included per board (rather than per target).
Signed-off-by: Bevan Weiss <bevan.weiss@gmail.com>
[Make it depend on REGULATOR_SUPPORT, and !SMALL_FLASH] Link: https://github.com/openwrt/openwrt/pull/22172 Signed-off-by: Robert Marko <robimarko@gmail.com>
Robert Marko [Fri, 27 Feb 2026 11:08:56 +0000 (12:08 +0100)]
targets: add REGULATOR_SUPPORT auto-feature
Not all targets need regulator support, so they dont enable it as its
disabled in the generic config by default.
So, in order to allow kernel modules to depend on regulator support lets
add a new feature flag "regulator" and set it automatically if target
kernel config enables CONFIG_REGULATOR.
Hauke Mehrtens [Thu, 26 Feb 2026 19:49:17 +0000 (20:49 +0100)]
lantiq: dm200: Fix loading PHY firmware
The device has 1 100MBit/s port. By default the PHY firmware is running
in 1GBit/s mode. The driver will try to load the 1GBit/s firmware and
fail if it is not there. Set the GPHY0 also to 100MBit/s mode.
The driver uses all nodes independent of the status attribute.
Do the same fix for AVM FRITZ!Box 7412 too.
Reported-by: Achelon in OpenWrt forum Fixes: https://github.com/openwrt/openwrt/issues/21836 Link: https://github.com/openwrt/openwrt/pull/22188 Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Linus Lüssing [Wed, 28 Jan 2026 00:52:15 +0000 (01:52 +0100)]
realtek: remove currently unused VLAN profile_id 1
So far we are only associating and using a VLAN profile ID of 0 to any
VLAN ID, VLAN profile ID 1 is unused. Therefore we can safely remove its
initialization, just like we don't initialize any of VLAN profile IDs
2 to 7 on rtl83xx/rtl930x or 2 to 15 on rtl931x.
Linus Lüssing [Sat, 10 May 2025 13:38:49 +0000 (15:38 +0200)]
realtek: setup special portmasks only once
The special all-ports multicast portmasks table entry would be written
twice, redundantly as vlan_profile_setup() is called twice. Which is
unnecessary. Writing this reserved multicast portmask entry is
independent of a VLAN profile, therefore let's move this to a more
suitable position.
Bevan Weiss [Fri, 27 Feb 2026 22:39:54 +0000 (09:39 +1100)]
kernel: lift CONFIG_REGULATOR_VEXPRESS 'not set' to generic
Advertise the default 'not set' of this CONFIG higher in the tree since
it's already referenced by several other targets, with a few more targets
missing this reference (which fails build to prompt for N/y/m).
Make use of the new fast_age() helpers for the RTL83xx targets.
For this:
- split the existing rtldsa_83xx_fast_age() into two device
specific functions. This removes a family_id check.
- change the callbacks in the config structures
realtek: dsa: provide generic fast_age for RTL93xx
Fast ageing of L2 entries is supported by DSA with two callbacks.
- port_fast_age(): age out for one port
- port_vlan_fast_age(): age out for one vlan on one port
Independent from the SoC it always boils down to issue a command
to the L2_TBL_FLUSH_CTRL register. Nevertheless the current
implemententation is repeated multiple times and makes use of
the family_id.
As a first refactoring step provide generic fast_age() functions
for RTL930x and RTL931x by rearranging the existing definitions
of vlan_port_fast_age().
The logic is as follows:
- provide a SoC dependent function that works with or without VLAN.
When VLAN/VID = -1 only flush the specific port otherwise only
flush given VLAN on port.
- provide a port_fast_age() helper that calls the SoC specific
functions with VLAN = -1.
- provide a port_vlan_fast_age() helper that calls the SOC specific
functions and handing over the given VLAN.
rteth_hw_receive() already has a loop that checks how much work
may be done. No need to repeat that in the caller too. While we
are here adapt the function prefix.
realtek: eth: be defensive in rteth_confirm_and_disable_irqs()
With latest refactoring irq activation and deactivation is side-by-side
in the code. This makes it easier to align these functions. Current
assumption is, that the ethernet irq is only called on one cpu and
napi handling does not interfere. So it should be totally fine to
run irq disabling (called from interrupt handler) without locks.
Nevertheless be defensive and add a lock(). So in the case the ethernet
irq is fired twice on two cpus (e.g. RTL931x) the code is on the
safe side.
Once again there is a family_id check. This time around the irq
reactivation in the receive path. For cleanup and better readability
add a helper directly besides its already existing counterpart
rteth_confirm_and_disable_irqs(). With that drop unneeded defines.
Compared to ralink,mtd-eeprom , the nvmewm binding ends up byteswapping
the data on big endian hosts. Meaning on big endian, the nvmwem binding
is equivalent to:
ralink,mtd-eeprom +
ralink,eeprom-wrap
Revert as a result since there's no eeprom-swap here.
Robert Marko [Fri, 27 Feb 2026 10:09:00 +0000 (11:09 +0100)]
generic: add CONFIG_REGULATOR_QCOM_USB_VBUS
CONFIG_REGULATOR_QCOM_USB_VBUS showed up on Layerscape recently, and it
looks like multiple targets disable it manually, so simply disable it in
generic configuration instead.
Hauke Mehrtens [Wed, 25 Feb 2026 00:26:02 +0000 (01:26 +0100)]
mac80211: activate BRCMFMAC_SDIO on x86
Activate the option BRCMFMAC_SDIO by default on x86 too. x86 already
compiles MMC support into the kernel. This will just compile brcmfmac
with MMC support.
Hauke Mehrtens [Wed, 25 Feb 2026 23:41:11 +0000 (00:41 +0100)]
wifi-scripts: fix handling of 64 character WPA key
The key variable is not defined in the scope when setting wpa_psk. Use
config.key instead.
This fixes configuration the 64 characters wpa_psk directly.
Hauke Mehrtens [Wed, 25 Feb 2026 21:42:56 +0000 (22:42 +0100)]
ramips: mt7621: fix network configuration
The configuration for the dlink,dir-1360-a1 also changed the settings
for the devices defined on top of it. "lan1 lan2 lan3 lan4" "wan" is
the default configuration, no need to add it here.
Fixes: 7a8e2efed587 ("ramips: add support for D-Link DIR-1360 A1") Reported-by: schmars in IRC Link: https://github.com/openwrt/openwrt/pull/22179 Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Sander Vanheule [Sat, 21 Feb 2026 14:00:39 +0000 (15:00 +0100)]
realtek: hog the GS1900-24E external IC reset line
The GPIO line connecting to the reset signals of the GS1900-24E(A1)'s
external ICs (RTL8218B phys and RTL8231 expander) cannot be asserted by
the MDIO subsystem, as the reset is shared between busses.
To prevent users from accidentally asserting the reset line, a GPIO hog
is created to permanently de-assert the signal, reliably keeping the
phys and GPIO expanders on.
Tested-by: Simon Fischer <simi.fischa@gmail.com> Signed-off-by: Sander Vanheule <sander@svanheule.net>
Sander Vanheule [Sat, 21 Feb 2026 13:44:23 +0000 (14:44 +0100)]
realtek: rtl838x: drop GS1900 MDIO reset GPIO
The reset line wired to the RTL8231 on the GS1900 series may also
connect to other external ICs on the board. On the GS1900-24E, the
reset line is wired (via buffers) to the board's RTL8231 expanders and
the RTL8218 phys. As these external devices (phys) are on different
busses, the reset line shouldn't be specified on one bus or the other.
Drop the reset specification from the generic GPIO description, so it
can be added back on a per-device basis after confirming the behavior.
Hauke Mehrtens [Sun, 22 Feb 2026 22:11:47 +0000 (23:11 +0100)]
wireless-regdb: update to version 2026.02.04
75bedc5 wireless-regdb: Update regulatory info for Australia (AU) for 2025 a6e5195 wireless-regdb: Update broken link in regulatory.bin(5) manpage 9e8c67f wireless-regdb: Update regulatory info for Malaysia (MY) for 2024 61a4637 wireless-regdb: Update regulatory info for Malaysia (MY) for 2025 5cefe55 wireless-regdb: Update regulatory info for Tunisia (TN) on 6GHz for 2025 1a729ae wireless-regdb: Update regulatory info for Canada (CA) for 2025 ea20dfa wireless-regdb: update regulatory database based on preceding changes
Jonas Lochmann [Mon, 16 Feb 2026 00:00:00 +0000 (01:00 +0100)]
iproute2: include upstream patch for musl libc
Due to a missing include, the constant UINT_MAX is undefined. This
fixes issues when building v25.12.0-rc5. Including a newer version of
iproute2 would include the patch, but causes other building issues.
Tested-by: Arif Rahman Hakim <arif2785id@gmail.com> Signed-off-by: Ahmed Naseef <naseefkm@gmail.com> Link: https://github.com/openwrt/openwrt/pull/22095 Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
mt7620: workaround jal imm26 and redundant PAGE_ALIGN
On MT7620-class platforms (CONFIG_NET_RALINK_MT7620) we observe sporadic
wrong-jump-targets, kernel oopses, hanging, corrupted backtraces or even
"half-written" instructions when the compiler emits a direct 'jal imm26'
call.
This is triggered in:
- the small random helpers inside get_random_u32_below(), and
- the blkcg_maybe_throttle_current() call in resume_user_mode_work().
This patch forces those two call sites to use an indirect call via
a volatile function pointer (load into register + jalr) when building
for MT7620, avoiding embedding a 26-bit immediate jump target.
Additionally, on MT7620 builds the exec path in fs/exec.c is modified:
- skip arch_align_stack() + PAGE_ALIGN() in setup_arg_pages()
because the micro-randomization (< PAGE_SIZE) implemented by many
ports (including MT7620) is negated immediately by PAGE_ALIGN().
Skipping the redundant PAGE_ALIGN() reduces exposure to the
problematic code pattern.
These changes are targeted workarounds for MT7620; behavioral logic is unchanged.
Tim Harvey [Fri, 20 Feb 2026 01:01:48 +0000 (17:01 -0800)]
imx: cortexa53: remove KSZ9477 static driver
The KSZ9477 driver was added to the cortexa53 kernel to support the
Gateworks Venice product family which has a board with this switch. Now
that the kmod-dsa-ksz9477 driver is available as a package remove the
static configuration ad add the package.
This resolves an issue caused by having the switch driver static and the
PHY driver as a module such that the PHY driver was not registered early
enough to be used causing some errata to not be worked around.
The makefile recipe is wrong and a module compilation tries
to build each object individually. Fix that. This allows to
build the dsa driver as a module.
Suggested-by: Balázs Triszka <info@balika011.hu> Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de> Link: https://github.com/openwrt/openwrt/pull/22121 Signed-off-by: Robert Marko <robimarko@gmail.com>
The ethernet driver has a hard link to the dsa driver. Especially
the setup_tc() function must be available when the driver loads.
Decouple it by using a dsa callback recipe.
Suggested-by: Balázs Triszka <info@balika011.hu> Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de> Link: https://github.com/openwrt/openwrt/pull/22121 Signed-off-by: Robert Marko <robimarko@gmail.com>
We do not want to rely on the soc_info structure. But at the moment
it is still referenced in several places. Add an EXPORT() to it.
So drivers that need access to this structure can be build as modules.
Suggested-by: Balázs Triszka <info@balika011.hu> Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de> Link: https://github.com/openwrt/openwrt/pull/22121 Signed-off-by: Robert Marko <robimarko@gmail.com>
Eicke Herbertz [Sat, 21 Feb 2026 16:35:54 +0000 (17:35 +0100)]
ath79: remove kmod-ath9k and wpad from Mikrotik RB750r2
The RB750r2 (HEXLite) does not have wifi and those packages bloat the image
by a significant amount. When building a custom image with WireGuard and
booting that from initramfs, there wasn't enough space left in tmpfs to
upload and flash the squashfs image. Investigating what packages I could
remove, I discovered these unneeded ones.
Shiji Yang [Sun, 22 Feb 2026 02:52:20 +0000 (10:52 +0800)]
kernel: net: fix deadlock caused by phy led trigger
Register phy led_triggers earlier to avoid AB-BA deadlock.
Fixes: https://github.com/openwrt/openwrt/issues/18472 Signed-off-by: Shiji Yang <yangshiji66@outlook.com> Link: https://github.com/openwrt/openwrt/pull/22136 Signed-off-by: Robert Marko <robimarko@gmail.com>
Two different irq handlers exist for RTL83xx and RTL93xx. Basically
they do always the same.
- Check transmit interrupts (not needed anymore)
- Check rx overflow interrupts (not needed anymore)
- Determine rx interrupts and queues that must be processed.
- In case of RTL839x check for L2 interrupts
With all the recent refactoring their logic is more or less the
same. Merge them into one handler. For better readability add a
helper that determines the work (aka rings) that needs to be
processed.
There is a wrong port assignment for the 4 SFP+ ports
on that device. Additionally the transmit polarity
change option was missed. Fix that.
Fixes: f88135b ("realtek: add support for Linksys LGS352C") Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de> Link: https://github.com/openwrt/openwrt/pull/22119 Signed-off-by: Robert Marko <robimarko@gmail.com>
Jonas Jelonek [Wed, 18 Feb 2026 20:18:21 +0000 (20:18 +0000)]
realtek: pcs: rtl931x: configure SerDes-backing MAC groups
RTL931x organizes MACs into 12 groups (one per SerDes) that must be
explicitly enabled before link establishment. Without initialization,
link may fail or packets may be corrupted, especially in USXGMII/XSGMII
modes. This is the case for devices which lack initialization by the
bootloader.
Simply enable all MACs in all groups by writing 0xffffffff to the
registers. Unused MACs and reserved bits are harmless, avoiding complex
logic to always set only needed MACs.
This is placed in the PCS driver since the MAC groups are assigned per
SerDes and the DSA driver lacks SerDes awareness (on purpose)
* the device uses 9600 Baud by default but this extremely slows down the
device when using the serial console. OpenWrt is configured to use
115200 Baud. If you need to enter the bootloader, you need to use 9600
Baud.
* PT7A7514WE watchdog is fed through hardware-assisted SYS_LED. However,
the bootloader seems to deactivate that again during autoboot. There's
a quirk in early arch setup for this.
* a kernel binary "nos.img" needs to be stored into JFFS2 filesystem
using 4KiB erase block instead of 64KiB.
* V1 is the version with the 19"-sized case. As of 2026, there's a newer
version with a narrow case.
Flash instructions using initramfs image:
-----------------------------------------
1. Prepare TFTP server with an IP address in 192.168.2.0/24.
2. Connect your PC to Port 1 on SKS8300-12X.
3. Power on SKS8300-12X and interrupt autoboot by Ctrl + B.
4. Login to the vendor CLI by Ctrl + F and "diagshell_unipoe_env" as password.
5. Switch baudrate to 115200 by running a command and then reconnect
with different settings:
baudrate 115200
6. Switch to U-Boot CLI by "debug_unish_env".
7. Enable Port 1 with the following commands:
rtk 10g 0 fiber1g # (or fiber10g if 10GBase-*R)
rtk ext-devInit 0 # init RTL8231 that holds SFP GPIOs
rtk ext-pinSet 2 0 # set tx-disable of port 1 to LOW
9. On the initramfs image, backup the stock firmware if needed.
10. Upload (or download) sysupgrade image to the device.
11. Erase "firmware" partition to cleanup JFFS2 of stock FW:
mtd erase firmware
12. Perform sysupgrade with the sysupgrade image.
13. Wait until the flash completes and the system reboots into OpenWrt.
Reverting to stock firmware:
----------------------------
1. Prepare OpenWrt SDK to use the mkfs.jffs2 tool contained in it
Note: the official mkfs.jffs2 tool in mtd-utils doesn't support 4KiB
erase size and not usable for SKS8300-8X
2. Create a directory for working
3. Download official firmware for SKS8300-8X from XikeStor's official
website
4. Rename the downloaded firmware to "nos.img" and place it to the
working directory
5. Create a JFFS2 filesystem binary with the working directory
Jonas Jelonek [Sat, 7 Feb 2026 23:36:16 +0000 (23:36 +0000)]
realtek: arch: add early watchdog quirk
On some XikeStor switches (SKS8300-8X, SKS8300-12X), the SYS_LED/GPIO0 is
used to feed an external PT7A7514WE watchdog. While this was no issue on
SKS8300-8X, the bootloader on the SKS8300-12X seems to deactivate the
automatic feeding on purpose by setting the pin function of to GPIO0
instead of SYS_LED. This kills the periodic signal generated on that pin.
This causes the kernel to just stop quite early and reset the system
entirely.
Because this happens very early, it doesn't work to define this as a
pinctrl entry or GPIO hog. The drivers aren't even loaded at that stage.
To work around the issue, we need to configure this in the arch-specific
early setup. An affected device needs to have a corresponding node in
the DTS that is picked up then.
Startup of mdio on a RTL8216 based device currently shows the
following warnings.
[1.948608] skip polling setup for unknown PHY 001ccaf3 on port 0
[1.968920] skip polling setup for unknown PHY 001ccaf3 on port 8
[1.989171] skip polling setup for unknown PHY 001ccaf3 on port 16
[2.009704] skip polling setup for unknown PHY 001ccaf3 on port 20
[2.030209] skip polling setup for unknown PHY 001ccaf3 on port 24
[2.052270] realtek-otto-serdes-mdio 1b000000.switchcore:mdio-serdes:
Realtek SerDes mdio bus initialized, 12 SerDes, 64 pages
Add the phy detection to the mdio bus so that polling setup works.
While we are here sort the phy ids alphabetically.
The NVMEM codepath does not perform automatic byte conversion. It can be
fixed but the upstream version is quite different from the local
mac80211 patch. Revert until mac80211 gets updated and the whole mess
can get squared away.
realtek: mdio: taker over RTL931x PHY polling from DSA
PHY polling setup has found a home in the mdio driver. For RTL931x
there still exists a setup sequence for polling type (serdes/mdio)
in the DSA driver. Put it where it belongs.
Allow building imagebuilder and/or sdk with dependency on the same
external toolchain as used to build the imagebuilder or sdk (so that ib
and sdk may be built using an external toolchain).
Signed-off-by: Daniel Dickinson <openwrt@daniel.thecshore.com>
Original patch:
https://lists.openwrt.org/pipermail/openwrt-devel/2016-January/012552.html
[rebased to current main, impvove commit title] Signed-off-by: Pawel Dembicki <paweldembicki@gmail.com> Link: https://github.com/openwrt/openwrt/pull/22089 Signed-off-by: Robert Marko <robimarko@gmail.com>
Daniel Golle [Fri, 20 Feb 2026 14:49:02 +0000 (14:49 +0000)]
Revert "package: kernel: dtc: Add DTO support"
It looks like commit 6d2f3b1b19 ("package: kernel: dtc: Add DTO support")
added this patch file 9 years ago without it ever being applied anywhere.
Back then there wasn't even a 'dtc' package, but we just used 'dtc' from
the Linux kernel sources.
Nowadays there is package/utils/dtc which is used to build dtc to be used
on the target (*not* a host-build!), and it of course already contains
support for device tree overlays since v1.4.3 from 2017...
The MAC_FORCE_MODE_CTRL register is only used for the CPU port.
No need to repeat the port register calculation for each usage.
Simply point to the cpu port register directly.
Felix Fietkau [Fri, 20 Feb 2026 18:05:45 +0000 (18:05 +0000)]
wireguard-tools: fix string indexing in endpoint host check
Use substr() instead of array index syntax to access the first
character of the endpoint host string, as ucode does not support
array-style indexing on strings.
Fixes: https://github.com/openwrt/openwrt/issues/22116 Fixes: 8f977b4a4055 ("wireguard-tools: fix handling of multi-value config options") Signed-off-by: Felix Fietkau <nbd@nbd.name>
Ahmed Naseef [Fri, 20 Feb 2026 05:10:31 +0000 (09:10 +0400)]
qualcommbe: fix rdp433 build after DTS directory move
Commit 041ef9648d ("qualcommbe: move Device DTS to dedicated DTS
directory") introduced a dedicated DTS directory and set DEVICE_DTS_DIR
to ../dts as default. This broke the rdp433 device build since its DTS
resides in the kernel tree (applied via patches), not in the target dts
directory.
Fix this by overriding DEVICE_DTS_DIR for rdp433 to point to the kernel
DTS directory.
Fixes: 041ef9648d ("qualcommbe: move Device DTS to dedicated DTS directory") Signed-off-by: Ahmed Naseef <naseefkm@gmail.com> Link: https://github.com/openwrt/openwrt/pull/22096 Signed-off-by: Robert Marko <robimarko@gmail.com>
Felix Fietkau [Fri, 20 Feb 2026 08:50:44 +0000 (08:50 +0000)]
wireguard-tools: fix handling of multi-value config options
Config options like addresses and ip6prefix can be passed as either a
space-separated string or an array. Add a to_array() helper and use it
consistently for all multi-value options (addresses, ip6prefix,
allowed_ips).
Fixes: https://github.com/openwrt/openwrt/issues/22102 Fixes: 41bc454602f1 ("wireguard-tools: rewrite proto handler in ucode") Signed-off-by: Felix Fietkau <nbd@nbd.name>
projects / thirdparty / openwrt.git / log
