Rework CompileImage/Initramfs locking logic to fix race condition on
concurrent access/compilation.
Rework each intermediate step and group them under one single execution
block. Protect this with a new lock, 'gen-initramfs' and rename the
compile initramfs to compile-initramfs lock name.
This is done to handle corner case scenario where a rootfs with the same
hash ID is generated at the same time. To handle this, we execute
everything under lock and use a specific hash ID lock to prevent
concurrent access/compilation.
Robert Marko [Mon, 16 Sep 2024 09:51:45 +0000 (11:51 +0200)]
generic: enable RELR relocations
RELR is a relocation packing format for relative relocations.
Linux has supported using it for ARM64 since 5.4, and more recently
for LoongArch as well.
It requires compatible tooling as well, so only with binutils 2.43
TOOLS_SUPPORT_RELR is set and thus RELR available.
RELR should always decrease the binary size and while testing this
the default build for Dynalink WRX-36 (qualcommax/ipq807x) reduced
the compressed kernel size by 38k and uncompressed one by 640k.
Merging of the realtek 6.6 series forgot to include some final fixes
for the new MDIO driver. What was changed in last second?
1. The MDIO driver used wrong constants to make use of the raw
page (for direct register access). Provide a rawpage variable in
the bus private structure, populate it during initialization and
make use of it at the proper places
2. We always used the variable portaddr for the bus index. Usually
our driver uses either addr or port for the same meaning. Remove the
duplication and reuse the normal addr variable.
3. Drop functions rtmdio_write_page() and rtmdio_read_page(). These
only call the PHY driver read/write page functions. We know that
these will only access page 0x1f. As we have only Realtek PHYs
and our driver only reacts to this special page, just hardcode it.
Benefit is that we can use these functions for PHY detection when
read/write page functions are not yet assigned.
4. Add two new helper functions phy_port_read_paged() and
phy_port_write_paged(). These allow to access arbitrary ports on
the MDIO bus when the packages are not initialized. These will be
needed for proper RTL8218B and RTL8214FC detection in forthcoming
patches.
5. The port tracking wrongly used index 0 to mark "normal" access.
This does not allow to make a "special" access to port 0. Use
index -1 to mark "normal" access.
Provide the fix for 5.15 and 6.6 to allow for easy version
comparison.
With commit a22d359fa56fe0 VLAN handling was fixed for kernel 6.6.
This restored network connectivity of the devices. For easy testing
backport the fix for 5.15 too.
Ivan Pavlov [Thu, 15 Aug 2024 18:47:25 +0000 (21:47 +0300)]
hostapd: update to 2.11 release tag
Release 2.11 has been quite a few new features and fixes since the 2.10
release. The following ChangeLog entries highlight some of the main
changes:
* Wi-Fi Easy Connect
- add support for DPP release 3
- allow Configurator parameters to be provided during config exchange
* HE/IEEE 802.11ax/Wi-Fi 6
- various fixes
* EHT/IEEE 802.11be/Wi-Fi 7
- add preliminary support
* SAE: add support for fetching the password from a RADIUS server
* support OpenSSL 3.0 API changes
* support background radar detection and CAC with some additional
drivers
* support RADIUS ACL/PSK check during 4-way handshake (wpa_psk_radius=3)
* EAP-SIM/AKA: support IMSI privacy
* improve 4-way handshake operations
- use Secure=1 in message 3 during PTK rekeying
Firmware:
The vendor firmware is a fork of OpenWrt (Reboot) with a kernel version
4.4.93. The flash is arranged as below and there is a dual boot
mechanism alternating between rootfs_0 and rootfs_1.
In OpenWrt rootfs_0 will be used as a boot partition that will contain the
kernel and the dtb. The squashfs rootfs and overlay are standard OpenWrt
behaviour.
U-boot:
With proper serial access booting can be halted to U-boot by pressing any
key. TFTP and flash writes are available, but only the first one has been
tested.
NOTE: Recovery mode can be accessed by holding down the reset button while
powering on the device. The led 'Update' will show a solid green light
once ready. A web server will be running at 192.168.1.1:80 and it will
allow flashing a firmware package. You can cycle between rootfs_0 and
rootfs_1 by pressing the reset button once.
Root password:
With the vendor web UI create a backup of your settings and download the
archive to your computer. Within the archive in the file
/etc/shadow replace the password hash for root with that of a password you
know. Restore the configuration with the vendor web UI and you will have
changed the root password.
SSH access:
You might need to enable the SSH service for LAN interface as by default
it's enabled for WAN only.
Installing OpenWrt:
With the vendor web UI install the OpenWrt factory image. Alternatively,
ssh to the device and use sysupgrade -n from cli.
Finalize by installing the OpenWrt sysupgrade image to get a fully
functioning system.
Reverting to the vendor firmware:
Boot with OpenWrt initramfs image
- Remove volumes rootfs_0, rootfs and rootfs_data and create vendor
volumes.
Andre Heider [Sun, 4 Aug 2024 14:06:56 +0000 (16:06 +0200)]
omap: refresh kernel 6.6 config
Unset CONFIG_CRYPTO_MANAGER_DISABLE_TESTS and
CONFIG_CRYPTO_MANAGER_EXTRA_TESTS manually to fix these linker errors:
crypto/crypto_engine.o: in function `crypto_engine_register_akcipher': undefined reference to `crypto_register_akcipher'
crypto/crypto_engine.o: in function `crypto_engine_unregister_akcipher': undefined reference to `crypto_unregister_akcipher'
crypto/crypto_engine.o: in function `crypto_engine_register_kpp': undefined reference to `crypto_register_kpp'
crypto/crypto_engine.o: in function `crypto_engine_unregister_kpp': undefined reference to `crypto_unregister_kpp'
Andre Heider [Fri, 9 Aug 2024 06:34:29 +0000 (08:34 +0200)]
kernel/omap: Restore kernel files for v6.1
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
Andreas Böhler [Sat, 1 Jun 2024 17:28:22 +0000 (19:28 +0200)]
realtek: add fan controller support to D-Link DGS-1210-28MP
The DGS-1210-28MP has a LM63 fan controller connected via i2c of the
RTL8231. The clock line is always low if the property
i2c-gpio,scl-open-drain is not set; with this property, the GPIO pin is
force-drive and the clock works as expected.
The LM63 is not configured by U-Boot, thus only manual fan control is
possible by settings pwm1_enable to "1" and writing the desired values to
pwm1.
The OEM firmware drives the fan from user mode and sets it up like this:
// PWM LUT/value r/w, PWM Clock = 1.4kHz
0x4a 0x28
// Tachometer spinup disabled, spin-up cycles bypassed
0x4b 0x00
// PWM Frequency = default
0x4d 0x17
// PWM Value (28)
0x4c 0x1c
// If > 0 C, use
0x50 0x00
// PWM = 28
0x51 0x1c
// If > 51 C, use
0x52 0x33
// PWM = 44
0x53 0x2e
// Set hysteresis to 100 = default
0x4f 0x03
// Turn on automatic mode and w/p the LUT values
0x4a 0x08
A thread in the OEM firmware polls the ALERT status register for fan
failures.
Unfortunately, the lm63 kernel driver does not perform any initialization
of the chip and it does not support changing some config registers (like
PWM frequency or LUT). Hence, we are stuck with the defaults and need to do
fan control in software.
Andreas Böhler [Sat, 1 Jun 2024 17:14:39 +0000 (19:14 +0200)]
realtek: add full SFP support to D-Link DGS-1210-28 series
The DGS-1210-28 series was lacking full SFP support due to missing GPIOs.
Fortunately, the existing GPIO definitions of DGS-1210-52 match, this adds
the required i2c-gpio nodes to the DTS and allows hotplug SFP support.
The CPU port of realtek switches needs some proper PVID set to handle
untagged packets. Because the ethernet driver does no special VLAN
handling (see CPU tag RVID/RVID_SEL) as of now we can only steer
untagged packets by setting PVID for the CPU port. VLAN handling has
never been perfect but 3 events made things worse.
- Commit a37650821644 ("rtl83xx: dsa: Do nothing when vid 0")
- Commit e691e2b302d9 ("rtl83xx: dsa: reset PVID to 1 instead of 0")
- Upgrade to kernel 6.6
Reasons are:
- Rejecting VID 0 disabled Linux initialization routines
- Initialization for PVID forgot to set priv->ports[port].pvid
- Kernel 6.6 does no longer clarify CPU port as untagged
To fix this prepare the VID 0 setup inside the driver. Join all ports
to VID 0 and let no one from outsinde interfere with this setup.
Especially ignore PVID settings for the CPU port for all further
VLAN commands.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de> Suggested-by: Bjørn Mork <bjorn@mork.no>
realtek: 6.6: refactor mac config and link up for RTL83xx
Since kernel commit c5714f68a76bcad3d ("net: phylink: explicitly invalidate
link_state members in mac_config") it should be clear that link data can
only be used in mac_link_up(). Refactor that for the RTL83xx targets.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
Especially fw_arg2 should be the pointer to the environment and it looks
very suspicous. It is not aligned and the address is outside KSEG0 and
KSEG1. Booting the device will result in a hang. Do better at verifying
the address.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de> Suggested-by: Bjørn Mork <bjorn@mork.no>
The supported_interfaces bitmap cannot be empty since mainline kernel
commit de5c9bf40c45 ("net: phylink: require supported_interfaces to
be filled"). Fix the dsa and ethernet driver accordingly.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
The DSA framework has changed a bit since 6.1, lets adapt to match.
Currently there is no one-patch-fits-all solution to directly fix
all errors up to 6.6. So cover the final differences with this
second patch.
The DSA framework has changed a bit since 5.15, lets adapt to match.
Currently there is no one-patch-fits-all solution to directly fix
all errors up to 6.6. So at least take all the already known changes
that cover differences between 5.15 and 6.1
Most notable upstream changes are:
- d3eed0e57d5d ("net: dsa: keep the bridge_dev and bridge_num as part
of the same structure")
Update of port_bridge_{join,leave}: use same helper as upstream
- c26933639b54 ("net: dsa: request drivers to perform FDB isolation")
Update of port_fdb_{add,del}, port_mdb_{add,del}
- dedd6a009f41 ("net: dsa: create a dsa_lag structure")
Update of port_lag_{join,leave}
Signed-off-by: Olliver Schinagl <oliver@schinagl.nl>
[align updates with upstream, add references to upstream commits] Signed-off-by: Sander Vanheule <sander@svanheule.net> Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
No content changes. Only take over the new patch locations. All errors
that wil arise from compiling with the phy driver will be covered by
follow up patches.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
No content changes. Only adapt the failing hooks and take over the
new patch locations. All errors that wil arise from compiling with
the dsa driver will be covered by follow up patches.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
This is not a surprise. Before upgrade to 6.6 we refactored the mdio part of
the ethernet driver and knew that changes will come. Drop all unnecessary
stuff from the old world and adapt to the new kernel.
- remove legacy functions
- directly link new functions
- adapt to new shared base address
- remove references to old MDIO bus capabilities
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
No content changes. Only take over the new patch locations. All errors
that will arise from compiling with the ethernet driver will be covered
by follow up patches.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
With the new kernel the MDIO bus gets created after the smbus
read/write functions are used. Make use of native functions.
Relocate bus initialization into a separate function to make
patch easier to read.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
No content changes. Only take over the new patch locations.
With this patch all platform specific changes for kernel 6.6 are in place.
Realtek devices are bootable from serial console. VPE is fully functional
on devices that support it. The switch functions (ethernet and DSA) are
not enabled yet.
Boot tested on RTL8380 (Linksys LGS310C) and RTL8393 (Zyxel GS1920-24).
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
VPE in mainline kernel has changed a lot. This patch wraps up the 5.15
patch files and rebases them in one single patch on top of kernel 6.6.
Former patches are
Copy files and config from 5.15 kernel version. Because of the big version jump
leave out the patches for now so we can treat them individually later on.
Signed-off-by: Markus Stockhausen <markus.stockhausen@gmx.de>
With commit b53202a8c3f7 ("realtek: switch to use generic MDIO accessor functions")
the phy access logic was enhanced. A quite big kernel patch was introduced that
allowed to make use of hardware-assisted page access like we have in the realtek
target. Basically it works the following way
- The enhanced bus intercepts page write accesses
- Currently selected pages are stored in internal vars.
- Finally only all-in-one-consistent bus/register accesses are issued
- It intercepted page changes and ensured that only a complete bus call
For the details see https://github.com/openwrt/openwrt/pull/16172
Switching over to newer kernels this patch is really hard to maintain. Its heavy
modifcations exist only in the realtek target. So it does not matter if we keep
it as an kernel modification or directly include it in our driver. To make it the
future brighter we drop this patch and take over its logic. Thus the kernel will
stay totally modification-free. What do we do?
1. Up to now the bus->priv structure directly pointed to ethernet->priv. Create an
explicit private structure rtl838x_bus_priv that not only holds the ethernet->priv
pointer but also space for some bus status tracking vars as well.
2. Wherever we use a reference to ethernet->priv directly replace that by an
additional indirection over the new rtl838x_bus_priv structure.
3. Up to now the phy flag PHY_HAS_REALTEK_PAGES identified that we can use the
alternative paged access from the patch. As this will be no longer available
remove it and provide read_page/write_page functions for each possible PHY.
These functions will be pretty standard as for other Realtek PHYs.
4. The existing mdio bus read/write function rely on the classic MII_ADDR_C45
flag - one interface for two access types. This mixup will be removed on the way
to kernel 6.6. In the future there will be two pairs of access functions. One for
classic access one for c45 style access. Rewrite our functions into 3 parts:
- a classic read/write function: ready for kernel 6.6
- a new c45 read/write function: ready for kernel 6.6
- a legacy read/write wrapper: for current 5.15 for the time being
When we switch to 6.6 we only need to remove the legacy wrappers and link the
new functions. Life can be so easy.
5. The classic read/write functions will incorporate the interception logic that
was originally in the patch.
6. The package convenience functions that were embedded in the patch get lost as
well. Rewrite them inside our phy driver.
projects / thirdparty / openwrt.git / log
