Jonas Jelonek [Sat, 6 Jun 2026 21:51:47 +0000 (21:51 +0000)]
realtek: pcs: add post_config sds_ops hook
Some variants need finalization that must run only after the SerDes is
powered up: RTL838X performs a one-shot switch queue reset on first
start, RTL930X runs RX calibration and the TX config. So far this lived
at the tail of the variant setup_serdes() functions.
Add an optional post_config hook to rtpcs_sds_ops and call it from
rtpcs_pcs_config() right after setup_serdes(), then relocate the
RTL838X and RTL930X tail work into it. As activate() still runs at the
end of setup_serdes() here, the execution order is unchanged.
Jonas Jelonek [Thu, 11 Jun 2026 20:05:28 +0000 (20:05 +0000)]
realtek: pcs: rtl930x: apply tx config before activation
In RTL930x setup, tx config was called at then end of the procedure
after configuration and calibration ran. This is still a leftover from
the old code located in DSA/PHY. However, applying TX configuration like
amplification factors, etc. doesn't make sense after calibration, it
should run before. Moreover the call was commented with "leave loopback
mode" which is just wrong and doesn't describe what the function does.
Fix this misery.
Testing on device with different interface modes shows no difference so
far, especially no negative effects.
Ryan Leung [Sat, 6 Jun 2026 11:19:46 +0000 (19:19 +0800)]
mediatek: wavlink wl-wn536ax6 rev a: add "network" LED trigger
Currently the "WiFi" LED indicator light is triggered only by activity on the 5 GHz PHY.
Change it to be triggered by WLAN activity on either Wi-Fi PHY, matching stock behaviour, by using
`kmod-ledtrig-network` added in 2aa1185fb05e ('leds: add "network" LED trigger (lan/wan/wlan)').
Ryan Leung [Sat, 6 Jun 2026 11:19:24 +0000 (19:19 +0800)]
mediatek: wavlink wl-wn536ax6 rev a: fix WLAN 5GHz MAC address
The second WLAN MAC address in the "Factory" partition at 0x0a is not the same as the
WLAN 5GHz BSSID observed when using the manufacturer's stock firmware, which is derived from
the 2.4GHz/label MAC address by setting bits 26 and 7 (Locally Administered).
While at it, also fix alphabetical ordering of some other device names.
Fixes: 1748ce829537 ("mediatek: add support for WAVLINK WL-WN536AX6 Rev a") Signed-off-by: Ryan Leung <untilscour@protonmail.com> Link: https://github.com/openwrt/openwrt/pull/23682 Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Ryan Leung [Sat, 6 Jun 2026 11:19:12 +0000 (19:19 +0800)]
mediatek: wavlink wl-wn536ax6 rev a: change nvmem layout for MAC addresses
Currently the WAN MAC address is read from the first (LAN) MAC address in the "HW" partition and
then incremented by 1 instead of being read directly from the second MAC address.
Change to reading the two MAC addresses (LAN/WAN) directly from "Factory" partition.
Fixes: 1748ce829537 ("mediatek: add support for WAVLINK WL-WN536AX6 Rev a") Signed-off-by: Ryan Leung <untilscour@protonmail.com> Link: https://github.com/openwrt/openwrt/pull/23682 Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Ryan Leung [Sat, 6 Jun 2026 11:18:45 +0000 (19:18 +0800)]
mediatek: wavlink wl-wn536ax6 rev a: enable nmbm bad block management
Enable nmbm bad block management with up to 8 MiB reserved. The manufacturer's stock device tree
contains the property `compatible = "generic,nmbm";` in node `nmbm_spim_nand` and
manufacturer's stock bootlog contains "nmbm nmbm_spim_nand: NMBM has been successfully attached"
Fixes: 1748ce829537 ("mediatek: add support for WAVLINK WL-WN536AX6 Rev a") Signed-off-by: Ryan Leung <untilscour@protonmail.com> Link: https://github.com/openwrt/openwrt/pull/23682 Signed-off-by: Jonas Jelonek <jelonek.jonas@gmail.com>
Qingfang Deng [Thu, 21 May 2026 01:53:46 +0000 (09:53 +0800)]
kernel: modules: mppe: clean up
Remove kmod-crypto-arc4 and kmod-crypto-ecb from depends, as it no
longer uses skcipher API for encryption.
Remove the non-existent CONFIG_PPP_MPPE_MPPC symbol.
In previous versions of OpenWrt, ethernet was partially working,
sometimes depending on initialization state left by bootloader.
Since the switch to NSS drivers, it is completely broken.
- swap GMAC to PHY address mapping
- use rgmii internal delay
- drop `fixed-link` rates
- add pinctrl settings for rgmii0
- declare reset GPIO 51 (used for both PHYs)
- disable hibernation
Paul Spooren [Fri, 12 Jun 2026 14:49:44 +0000 (16:49 +0200)]
build: fixup version.date creation for source archives
Published sources archives may contain a mixture of MTIMEs, so taking the time
of the first file found varies based on filesystem order. To fix this, sort all
MTIMEs and take the latest.
Since piping into a file directly creates that file, we also need to
specifically tell `find` to ignore `version.date`.
Furthermore, move this prior to the ./src copy step, since for some packages,
the OpenWrt build system ships extra ./src files, which would introduce the
wrong timestamps again.
generic: mxl862xx: allow CPU/SerDes ports to probe on firmware < 1.0.84
The firmware-version gate in mxl862xx_phylink_get_caps() leaves
config->supported_interfaces empty for the CPU/SerDes ports (9, 13) when
the switch runs firmware older than 1.0.84. phylink rejects an empty
supported_interfaces bitmap, so the switch fails to probe at all:
mxl862xx mdio-bus:10: switch ready after 2480ms, firmware 1.0.70 (build 70)
mxl862xx mdio-bus:10: phylink: error: empty supported_interfaces
error creating PHYLINK: -22
mxl862xx mdio-bus:10: probe with driver mxl862xx failed with error -22
This regresses boards still shipping the older vendor firmware (e.g. the
BananaPi BPi-R4 Pro 8X, which ships 1.0.70) where all LAN ports disappear.
Ungate only the CPU/SerDes ports (9, 13) so the switch probes and the user
can update the firmware; ports 10..12 and 14..16 stay gated as they are
genuinely unsupported by the old firmware.
Tested on a BPi-R4 Pro 8X with switch firmware 1.0.70 (kernel 6.18); the
6.12 patch receives the identical change for parity.
Michał Kępień [Fri, 12 Jun 2026 07:28:03 +0000 (09:28 +0200)]
ath79: mikrotik: also compile AG71XX_LEGACY as a module
Commit 9091c9f8cbd9 ("ath79: mikrotik: compile SWCONFIG and AR8216_PHY
as modules") caused the hybrid PHY/MDIO ar8xxx driver to be built as a
module instead of built-in. On at least MikroTik RouterBOARD 951G-2HnD,
this prevents the kernel from binding the correct PHY driver at boot
because the ar8xxx driver is not yet available when the MDIO bus is
probed:
- before:
ag71xx-legacy 19000000.eth: invalid MAC address, using random address
switch0: Atheros AR8327 rev. 4 switch registered on mdio.0
ag71xx-legacy 19000000.eth: connected to PHY at mdio.0:00 [uid=004dd034, driver=Atheros AR8216/AR8236/AR8316]
eth0: Atheros AG71xx at 0xb9000000, irq 4, mode: rgmii
- after:
ag71xx-legacy 19000000.eth: invalid MAC address, using random address
ag71xx-legacy 19000000.eth: connected to PHY at mdio.0:00 [uid=004dd034, driver=Generic PHY]
eth0: Atheros AG71xx at 0xb9000000, irq 4, mode: rgmii
As the PHY is already bound to the fallback "Generic PHY" driver when
the ar8xxx module is loaded, the switch remains non-functional and the
router has no network connectivity.
Fix by also compiling the MAC driver (ag71xx-legacy) as a module,
deferring switch initialization until control is transferred to
userspace during boot. The configured module autoload priorities
(ar8xxx: 43, ag71xx-legacy: 50) ensure that the ar8xxx driver is loaded
before ag71xx-legacy.
Fixes: https://github.com/openwrt/openwrt/issues/23739 Fixes: 9091c9f8cbd9 ("ath79: mikrotik: compile SWCONFIG and AR8216_PHY as modules") Suggested-by: Daniel Golle <daniel@makrotopia.org> Signed-off-by: Michał Kępień <openwrt@kempniu.pl> Link: https://github.com/openwrt/openwrt/pull/23748 Signed-off-by: Robert Marko <robimarko@gmail.com>
Jonas Jelonek [Thu, 11 Jun 2026 22:05:16 +0000 (22:05 +0000)]
realtek: pcs: rtl838x: drop redundant reset writes
Sometimes redundancy is so obvious that one likely misses it. Trying to
bring all SerDes setup into a unified shape, one thinks twice if
something is really needed or not.
For the RTL838x case, the single writes for take/release reset were still
an outlier. Looking closer at them one can see that the same bits are all
covered already in deactivate/activate. So before, they have been put
into the desired state but those outlier writes mess with them again.
[0, 3, 0x7146/0x7106] just deals with the SOFT_RST bit that is already
covered by rtpcs_838x_sds_reset.
[0, 0, 0xc00] touches multiple bits, amongst the EN_RX/EN_TX bits
already covered in deactivate/active. Moreover, it potentially forces
other bits into a state causing broken functionality, e.g.
INV_HSI/INV_HSO which deal with the polarity. This has no effect right
no right now but might be a latent issue in the future.
Also move the reset call down to the end of the function, doing a
soft/RX reset after every configuration is done. This is likely what the
SDK also intended, and mirrors the 839x behavior.
Jonas Jelonek [Thu, 11 Jun 2026 18:40:32 +0000 (18:40 +0000)]
realtek: pcs: rtl838x: replace literal with macro
Replace 0 with MII_BMCR because in this context, it maps cleanly to
normal PHY register structure. BMCR_PDOWN is already used there to show
which bits are set.
Jonas Jelonek [Sat, 6 Jun 2026 21:03:07 +0000 (21:03 +0000)]
realtek: pcs: rtl838x: merge and simplify patching
Reduce the per-SerDes patching to a per-mode patching since they share a
lot of similarities, especially for QSGMII.
For fiber patching, the differences are bigger. This takes an approach
to merge the sequences completely despite applying some settings which
weren't applied before on one of the SerDes. Testing and SerDes dumps
showed no or minimal differences, and no practical effect on
functionality. A look at the SerDes register documentation we have shows
reordering the writes isn't problematic because most of the bits aren't
any trigger bits.
With all its users gone, we can now drop also the workaround SDS macro.
Jonas Jelonek [Thu, 11 Jun 2026 08:28:27 +0000 (08:28 +0000)]
realtek: pcs: rtl838x: carve out writes for SerDes 0/1
For an attempt to simplify and reduce patching, carve out two special
writes for SerDes 0/1. While their exact purpose isn't fully known, they
need to match to make QSGMII on those two SerDes work. In order to be
able to merge the per-SerDes QSGMII patching sequences, carve them out
to the patching function one level above.
Jonas Jelonek [Sat, 6 Jun 2026 19:17:32 +0000 (19:17 +0000)]
realtek: pcs: rtl838x: set mode after patching
In order to have a unified SerDes setup sequence, align RTL838x to all
other variants. Set the mode after patching has been applied. Testing on
devices showed the order is rather irrelevant.
Jonas Jelonek [Wed, 22 Apr 2026 13:54:15 +0000 (13:54 +0000)]
realtek: pcs: rtl838x: fold lane enables and fiber path into {de,}activate
Expand the 838x deactivate/activate helpers to also cover the analog lane
enables (REG0 EN_RX/EN_TX) and the fiber path power-down (FIB_REG0
CFG_FIB_PDOWN), which are operational-state concerns rather than reset
concerns:
- _deactivate: clear EN_RX/EN_TX and set CFG_FIB_PDOWN in addition to
block-level power off.
- _activate: clear CFG_FIB_PDOWN and set EN_RX/EN_TX before block-level
power on. The pulse-start EN=0 is omitted since _deactivate already
established EN=0 and nothing in between touches REG0[1:0].
Shrink rtpcs_838x_sds_reset() accordingly: it now owns only the SOFT_RST
toggle, making it a pure digital reset primitive. The stale "SerDes %d
reset" dev_info was dropped along with the analog/fiber writes.
Behavioural note: the EN and FIB_PDOWN writes now happen *after* the
SOFT_RST release inside rtpcs_838x_sds_reset(), rather than before it as
in the old combined sds_reset. This creates a brief window where the
SerDes is out of digital reset with lanes still disabled, before
activate re-enables them. Consistent with the new
power_off/deactivate -> configure -> reset -> activate/power_on phase
structure, but diverges from the SDK ordering (EN pulse before SOFT_RST
release). Testing on real hardware shows no difference.
Backport several fixes and enhancements for the rtl8365mb DSA driver
from upstream.
The backport includes the following changes:
- Fix for rtl8_4 tag (from v7.1)
- VLAN and Forwarding offload (from v7.2)
Before this backport, all bridge forwarding was handled by the CPU,
causing a significant performance drop between LAN ports, similar
to standard WAN/LAN routing. Enabling hardware forwarding offload
alleviates CPU overhead and restores line-rate switching performance.
Shine [Sun, 31 May 2026 13:16:09 +0000 (15:16 +0200)]
scripts: dhcp/dhcpv6: handling of invalid client ID values
Verify and clean up client ID and global DUID config values before use, in
order to prevent DHCP client malfunction and loss of IPv4 and/or IPv6
connectivity.
- Accept common separators in the string (colon, dash, spaces/tabs, lf)
- Ignore invalid values (non-hex or odd-numbered length)
- Log a warning for invalid settings
The fall back mechanism is as follows:
If a (user-configured) network.<ifname>.clientid setting is found invalid,
ignore and fall back to using network.globals.dhcp_default_duid. In case
that's also invalid (e.g. has been tampered with), don't pass a client
ID to updhc/odhcpc. In that case, udhcpc/odhcpc will use their defaults,
which currently is the i/f MAC address resp. the i/f DUID-LL.
Only error handling is introduced, no behavior change for valid settings.
Paul Spooren [Wed, 10 Jun 2026 14:21:40 +0000 (16:21 +0200)]
package: nftables backport of reproducible builds patches
Building `nftables` twice results in slightly different binaries and since
it's including in the OpenWrt default firmware, breaks reproducibility of
shipped artifacts.
Enabling autoreconf, since the patches introduce a `nftvbersion.h.in` file.
This commit backports patches netfilter to fix this:
Paul Spooren [Wed, 10 Jun 2026 15:18:37 +0000 (17:18 +0200)]
package: make APK embedded help gzip reproducible
APK compresses it's helptext using LUA and require `zlib`, which isn't
available on the Buildbots. It thens falls back to `gzip`, which embeds the
MTIME, making the binary itself unreproducible.
This commits adds a downstream patch to run `gzip` with `-n`, setting the time
to 0.
Rosen Penev [Thu, 4 Dec 2025 03:15:05 +0000 (19:15 -0800)]
ath79: fix label-mac-device
It appears 683-of_net-add-mac-address-to-of-tree.patch relies on the
mac-address nvmem property being present. These nodes don't need it as
they take it from the eeprom but label-mac-device needs it.
1. Attach to serial console using a Cisco cable, 115200 8n1
2. Power on the switch and rapidly press Esc to interrupt U-Boot
3. Initialize the RJ45 ports by running `rtk network on`
4. Serve the initramfs-kernel.bin image on TFTP, IP 192.168.1.111
5. Use any RJ45 port to connect to your TFTP server
6. Run `tftpboot 0x81000000 initramfs-kernel.bin; bootm`
7. Use `mtd dump` or `dd` to backup the original firmware at `mtd5` first
8. Run sysupgrade to flash sysupgrade.bin into the switch and reboot
Restoring original firmware
-----------------------------
Vendor does not provide firmware images, so you should have made a backup.
Use `mtd write` to restore your backed up copy into the `mtd5` partition.
MAC Address
------------
All interfaces are assigned the same MAC address, from the U-Boot env
`ethaddr`. This MAC address is also printed on the device label.
Thomas Richard [Sat, 18 Apr 2026 12:37:33 +0000 (14:37 +0200)]
stm32: 6.18: disable some unnecessary options
The COMMON_CLK_STM32MP215 option is for clock support on MP215 platforms,
and PINCTRL_STM32_HDP is for Hardware Debug Port pin control. They can be
safely removed.
Thomas Richard [Mon, 18 May 2026 08:36:04 +0000 (10:36 +0200)]
stm32: fix return value of phy_power_on() in dwmac-stm32 driver
Patch 700-net-stmmac-dwmac-stm32-add-support-of-phy-supply-pro.patch
introduces phy-supply support in dwmac-stm32 driver. But the phy_power_on()
function always returns 0 even if it failed to drive the regulator. So fix
phy_power_on() to return an error if something wrong happened.
Thomas Richard [Fri, 17 Apr 2026 18:55:39 +0000 (20:55 +0200)]
kernel/stm32: restore files for v6.12
This is an automatically generated commit which aids following Kernel patch
history, as git will see the move and copy as a rename thus defeating the
purpose.
For the original discussion see:
https://lists.openwrt.org/pipermail/openwrt-devel/2023-October/041673.html
Hauke Mehrtens [Sun, 7 Jun 2026 11:31:36 +0000 (13:31 +0200)]
uclient: update to Git HEAD (2026-06-07)
f227ab4f1285 uclient: fix memory leak of url when backend alloc fails 645236607a3d uclient: free proxy_url in uclient_free 53d2d11fb00d uclient-http: fix NULL deref when digest WWW-Authenticate lacks params 0ba1b8718be5 ucode: check calloc return in uc_uclient_new 5cb19466d076 uclient-fetch: reset redirect counter per request 568c447950c1 uclient-fetch: retry short writes and surface errors 4c4a61a69ac3 uclient-fetch: use strtoull to parse Content-Length 440ca260622d uclient-http: validate Content-Length and chunk sizes from the server 8658324bd234 uclient: cast to unsigned char before ctype classifications bf403fa0f8e5 uclient-fetch: reject CR in --header values ec47f4176b1e uclient-fetch: always allocate auth_str and free it on exit 391dacd10d31 uclient-http: fail digest auth when the cnonce cannot be randomized ba1f4311b099 ucode: clamp read length to the size of the static buffer 0ba47f319b09 uclient-fetch: advance to the next URL between requests 9dd0055d527d uclient: initialize *port for unknown address families f816506651d5 uclient-http: start digest nonce count at 1 9b9b2da40835 uclient-http: fix 2-byte buffer undercount in digest add_field 80fb1e5ee68b ucode: avoid double-free of SSL context on ssl_init error d44fb561bbcf ucode: don't clear an unrelated registry slot on free 02562403d2a3 uclient-fetch: close the output file before fetching the next URL 4a63561d0012 uclient-http: initialize fd to -1 so a failed connect can't close stdin 428ae8342118 uclient: cancel pending timers in uclient_free() ffd8db308289 uclient-fetch: guard SSL debug setup against a missing SSL context 9f4db039a263 ucode: release callback argument references d2af92b0aad9 uclient-http: resolve redirects against the proxy target URL 5dddda023267 uclient-utils: avoid out-of-bounds pointer in get_url_filename 8507588e7ce7 uclient-http: use the proxy target URL for HTTP authentication 322c89c6ca0a uclient-http: retry 401 auth for all body-less request methods db3bb00ddda7 uclient-http: accept unquoted digest auth parameter values 5ce9983b2a11 uclient-http: follow 303 and 308 redirects 070c868486dc uclient-fetch: finalize request on invalid 206 Content-Range 83659a5a9bc1 uclient-fetch: concatenate multiple URLs into a single -O file 43bed2753998 ucode: don't leak string references in status()/get_headers() 03b17e4fc793 uclient-fetch: attach credentials to the proxy target, not the proxy 74c07b12f7ab uclient-fetch: don't run the progress meter in spider mode daad21fa2c17 uclient-fetch: handle init_request failure on the 204 resume retry
Anna Kiri [Mon, 8 Jun 2026 09:07:05 +0000 (11:07 +0200)]
mtd: fix buffer leak and fd leak in mtd_dump()
Two leaks in mtd_dump():
- The buffer allocated with malloc(erasesize) is never freed before
returning, leaking erasesize bytes on every call.
- The pre-existing malloc-NULL early return path also leaked the just-
opened fd by returning directly instead of going through cleanup.
Initialize buf to NULL, route the malloc-NULL case through the
existing 'out:' label, and add free(buf) on the cleanup path so both
fd and buf are released consistently on every exit.
Anna Kiri [Mon, 8 Jun 2026 09:06:32 +0000 (11:06 +0200)]
mtd: check malloc() return value in mtd_check()
After malloc(erasesize) in mtd_check(), the result was never checked
for NULL. On allocation failure, the buf pointer remained NULL and
would later be used in image_check() via read(imagefd, buf + buflen,
...), causing a NULL pointer dereference.
Add a NULL check after the allocation and return early. The early
return path also closes the just-opened fd and frees the strdup'd
colon-separated device-list copy to avoid leaks.
Anna Kiri [Mon, 8 Jun 2026 09:05:55 +0000 (11:05 +0200)]
mtd: check return values of lseek() and write() in mtd_write_buffer()
Two ignored return values in mtd_write_buffer() caused silent failures:
- lseek() return value was ignored. A failed seek (e.g. EBADF, ESPIPE)
followed by write() would silently write at the wrong offset.
- write() return value was ignored, silently discarding write errors.
This could lead to data corruption on the MTD device without any
indication.
Check both return values, report errors to stderr and return -1 on
failure.
Improve some descriptions so that they make the purpose of the
particular firmware package clearer to the user (e.g. as viewed in
verbose package listings).
Grische [Sun, 3 May 2026 14:34:46 +0000 (16:34 +0200)]
ipq806x: ap3935: disable hibernation on LAN1
The Extreme Networks AP3935's LAN1 port (gmac0/RGMII to AR8035 at MDIO
addr 1) fails to acquire a link if no Ethernet cable is connected at
power-on; the port stays dead even after a cable is later inserted, and
only a network restart recovers it.
The AR803x family enables hibernation mode by default, this setting
overrides it.
Jonas Jelonek [Tue, 2 Jun 2026 21:44:22 +0000 (21:44 +0000)]
realtek: pcs: replace __free() auto-cleanup stuff
Upstream netdev maintainers disregard both variable declarations
mid-function and usage of __free() for auto-cleanup [1]. To avoid any
pain when trying to upstream the PCS driver at some point, adjust that
now before forgetting it.
John Audia [Sat, 6 Jun 2026 12:17:46 +0000 (08:17 -0400)]
x86/64: config-6.18 remove obsolete options from config
Remove some dead weight from the config due to changes in the CRYPTO API
in 6.18. These are now internal library optimizations rather than selectable
algorithms, the corresponding Kconfig options were removed.
Andre Heider [Wed, 1 Apr 2026 13:13:40 +0000 (15:13 +0200)]
kernel/omap: restore files for v6.12
This is an automatically generated commit which aids following Kernel patch
history, as git will see the move and copy as a rename thus defeating the
purpose.
For the original discussion see:
https://lists.openwrt.org/pipermail/openwrt-devel/2023-October/041673.html
projects / thirdparty / openwrt.git / log
