Kernel 6.6 added dynamic SWIOTLB allocation, but with it also started
allocating 64MB of the SWIOTLB bounce buffer by default which is quite a
lot of memory on most OpenWrt devices.
Luckily in kernel 6.7 arm64 received an optimization that reduces that
default size to 1MB per 1GB of RAM if certain criteria was met.
So in order to reclaim back 63MB of RAM which brought some ipq807x devices
close to OOM under load lets backport the upstream commit.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Refresh backport patches for kernel 6.1.82 with make target/linux/refresh.
Fixes: 06cdc07f8c ("ath79: add support for Huawei AP5030DN")
Signed-off-by: Mieczyslaw Nalewaj <namiltd@yahoo.com>
Huawei AP5030DN is a dual-band, dual-radio 802.11ac Wave 1 3x3 MIMO
enterprise access point with two Gigabit Ethernet ports and PoE
support.
Hardware highlights:
- CPU: QCA9550 SoC at 720MHz
- RAM: 256MB DDR2
- Flash: 32MB SPI-NOR
- Wi-Fi 2.4GHz: QCA9550-internal radio
- Wi-Fi 5GHz: QCA9880 PCIe WLAN SoC
- Ethernet 1: 10/100/1000 Mbps Ethernet through Broadcom B50612E PHY
- Ethernet 2: 10/100/1000 Mbps Ethernet through Marvell 88E1510 PHY
- PoE: input through Ethernet 1 port
- Standalone 12V/2A power input
- Serial console externally available through RJ45 port
- External watchdog: SGM706 (1.6s timeout)
Serial console:
9600n8 (9600 baud, no stop bits, no parity, 8 data bits)
MAC addresses:
Each device has 32 consecutive MAC addresses allocated by
the vendor, which don't overlap between devices.
This was confirmed with multiple devices with consecutive
serial numbers.
The MAC address range starts with the address on the label.
To be able to distinguish between the interfaces,
the following MAC address scheme is used:
- eth0 = label MAC
- eth1 = label MAC + 1
- radio0 (Wi-Fi 5GHz) = label MAC + 2
- radio1 (Wi-Fi 2.4GHz) = label MAC + 3
Installation:
0. Connect some sort of RJ45-to-USB adapter to "Console" port of the AP
1. Power up the AP
2. At prompt "Press f or F to stop Auto-Boot in 3 seconds",
do what they say.
Log in with default admin password "admin@huawei.com".
3. Boot the OpenWrt initramfs from TFTP using the hidden script
"run ramboot". Replace IP address as needed:
> setenv serverip 192.168.1.10
> setenv ipaddr 192.168.1.1
> setenv rambootfile
openwrt-ath79-generic-huawei_ap5030dn-initramfs-kernel.bin
> saveenv
> run ramboot
4. Optional but recommended as the factory firmware cannot
be downloaded publicly:
Back up contents of "firmware" partition using the web interface or ssh:
$ ssh root@192.168.1.1 cat /dev/mtd11 > huawei_ap5030dn_fw_backup.bin
5. Run sysupgrade using sysupgrade image. OpenWrt
shall boot from flash afterwards.
Return to factory firmware (using firmware upgrade package downloaded from
non-public Huawei website):
1. Start a TFTP server in the directory where
the firmware upgrade package is located
2. Boot to u-boot as described above
3. Install firmware upgrade package and format the config partitions:
> update system FatAP5X30XN_SOMEVERSION.bin
> format_fs
Return to factory firmware (from previously created backup):
1. Copy over the firmware partition backup to /tmp,
for example using scp
2. Use sysupgrade with force to restore the backup:
sysupgrade -F huawei_ap5030dn_fw_backup.bin
3. Boot AP to U-Boot as described above
Quirks and known issues
-----------------------
- On initial power-up, the Huawei-modified bootloader suspends both
ethernet PHYs (it sets the "Power Down" bit in the MII control
register). Unfortunately, at the time of the initial port, the kernel
driver for the B50612E/BCM54612E PHY behind eth0 doesn't have a resume
callback defined which would clear this bit. This makes the PHY unusable
since it remains suspended forever. This is why the backported kernel
patches in this commit are required which add this callback and for
completeness also a suspend callback.
- The stock firmware has a semi dual boot concept where the primary
kernel uses a squashfs as root partition and the secondary kernel uses
an initramfs. This dual boot concept is circumvented on purpose to gain
more flash space and since the stock firmware's flash layout isn't
compatible with mtdsplit.
- The external watchdog's timeout of 1.6s is very hard to satisfy
during bootup. This is why the GPIO15 pin connected to the watchdog input
is configured directly in the LZMA loader to output the CPU_CLK/4 signal
which keeps the watchdog happy until the wdt-gpio kernel driver takes
over. Because it would also take too long to read the whole kernel image
from flash, the uImage header only includes the loader which then reads
the kernel image from flash after GPIO15 is configured.
Signed-off-by: Marco von Rosenberg <marcovr@selfnet.de>
[fixed 6.6 backport patch naming]
Signed-off-by: David Bauer <mail@david-bauer.net>
On x86, the build failed while trying to compile tools/lib/string.c because
of a clash with the system provided implementation for strlcpy
Add ifdefs to prevent the conflict.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
It's required to support NAND controllers with WP pin on boards that
don't have it connected to NAND chip.
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Backport AXP15060, AXP313a and AXP192 support.
The AXP15060 PMIC is used for starfive boards,
and the AXP313a PMIC is used for sunxi boards.
Remove conflicting patches from starfive target.
Signed-off-by: Chukun Pan <amadeus@jmu.edu.cn>
Fix issue of transmitting abnormal data which leads to link problems
in 1G and 2.5G SerDes modes (SGMII, 1000Base-X, 2500Base-X) on the
MediaTek MT7988 SoC.
Link: b72d6cba92
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
When rebasing patches on top of 6.1, a change that removed NF_TABLES
from NF_FLOW_TABLE's dependency was dropped accidentally, and iptables
flow offload module can only be built when nftables is selected.
Restore the change to fix it.
Fixes: 19a246bb65 ("generic: 6.1: manually refresh hack patches")
Signed-off-by: Qingfang Deng <dqfext@gmail.com>
Disable CONFIG_CRYPTO_SM4_ARM64_CE_CCM and CONFIG_CRYPTO_SM4_ARM64_CE_GCM
in the generic config like we do with similar CE symbols.
This avoids the need to disable them in individual targets.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Upstream commit e298d8a38b23 [0] changed method how to blink delays are pased
to function. Downstream commit must follow it.
[0] https://lore.kernel.org/r/20230510162234.291439-2-hdegoede@redhat.com
Reported-by: Mieczyslaw Nalewaj <namiltd@yahoo.com>
Signed-off-by: Pawel Dembicki <paweldembicki@gmail.com>
The dir variable has been used uninitialized since the port to 5.10, and
somehow this remains undetected by GCC.
Fixes: b10d604459 ("kernel: add linux 5.10 support")
Signed-off-by: Qingfang Deng <qingfang.deng@siflower.com.cn>
Also backport most recent MHI modem additions to Linux 6.6.
Adds support for generic SDX75-based modems.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
CONFIG_NET_VENDOR_DAVICOM is needed to activate the CONFIG_DM9000 option
which builds the kmod-dm9000. This fixes the following warning:
logs/package/kernel/linux/compile.txt:WARNING: kmod-dm9000 is not available in the kernel config - generating empty package
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Many can kernel modules are now gated by the newly introduced
CONFIG_CAN_NETLINK configuration option. Activate it to build the can
drivers again.
This was changed in this upstream Linux commit:
https://git.kernel.org/linus/df6ad5dd838e0fa543ca28ca6154901fa65a9443
This should fix these warnings with kernel 6.1 and 6.6:
logs/package/kernel/linux/compile.txt:WARNING: kmod-can-c-can is not available in the kernel config - generating empty package
logs/package/kernel/linux/compile.txt:WARNING: kmod-can-c-can-pci is not available in the kernel config - generating empty package
logs/package/kernel/linux/compile.txt:WARNING: kmod-can-c-can-platform is not available in the kernel config - generating empty package
logs/package/kernel/linux/compile.txt:WARNING: kmod-can-mcp251x is not available in the kernel config - generating empty package
logs/package/kernel/linux/compile.txt:WARNING: kmod-can-slcan is not available in the kernel config - generating empty package
logs/package/kernel/linux/compile.txt:WARNING: kmod-can-usb-8dev is not available in the kernel config - generating empty package
logs/package/kernel/linux/compile.txt:WARNING: kmod-can-usb-ems is not available in the kernel config - generating empty package
logs/package/kernel/linux/compile.txt:WARNING: kmod-can-usb-kvaser is not available in the kernel config - generating empty package
logs/package/kernel/linux/compile.txt:WARNING: kmod-can-usb-peak is not available in the kernel config - generating empty package
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
A patch to resolve an issue was found in MediaTek's GPL-licensed SDK:
In the mtk_ppe_stop() function, the PPE scan mode is not disabled before
disabling the PPE. This can potentially lead to a hang during the process
of disabling the PPE.
Without this patch, the PPE may experience a hang during the reboot test.
Reference: b40da332df
Suggested-by: Bc-bocun Chen <bc-bocun.chen@mediatek.com>
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Backport commit faa5f17fe2 ("kernel: mtk_eth_soc: release
MAC_MCR_FORCE_LINK only when MAC is up") to Linux 5.15 as well.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
When kernel 6.6 is being compiled with GCC13 it will prompt for:
Compressed Debug information
> 1. Don't compress debug information (DEBUG_INFO_COMPRESSED_NONE)
2. Compress debugging information with zlib (DEBUG_INFO_COMPRESSED_ZLIB)
3. Compress debugging information with zstd (DEBUG_INFO_COMPRESSED_ZSTD) (NEW)
This is because kernel now also has support for compressing debug info
with ZSTD, so add the missing symbol and disable it by default.
Signed-off-by: Robert Marko <robimarko@gmail.com>
CONFIG_HID_SUPPORT:
Set default value for kmod-hid
CONFIG_USB_CDNS3_IMX:
CONFIG_USB_CDNS3_PCI_WRAP:
CONFIG_USB_CDNSP_PCI:
Introduced by db0d7cf6a1 (usb: add cdns3 support)
Signed-off-by: Weijie Gao <hackpascal@gmail.com>
This reverts commit 7ea82bc17d.
This commit does not meet formal requirments, and has been merged due to
my error.
Signed-off-by: Robert Marko <robimarko@gmail.com>
CONFIG_DEBUG_INFO_COMPRESSED_ZSTD, required by GCC 13.0+
Compress the debug information using zstd. This may provide better compression than zlib, for about the same time costs, but requires newer toolchain support. Requires GCC 13.0+ or Clang 16.0+, binutils 2.40+, and zstd.
Signed-off-by: CharlesMengCA <58993776+CharlesMengCA@users.noreply.github.com>
A new patch was added meanwhile which will go as fix via netdev tree.
Copy & refresh it for Linux 6.6 as well.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Avoid crashing the kernel when trying to detect early versions of
RealTek RTL8221B 2.5G Ethernet PHY.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Currently, the existing uncompressed kallsym support is causing qualcommax
boards to hang on boot, and only after earlycon and verbose BUG() prints
are enabled the trace is visible:
[ 0.000000] ------------[ cut here ]------------
[ 0.000000] kernel BUG at kernel/kallsyms.c:340!
[ 0.000000] Internal error: Oops - BUG: 00000000f2000800 [#1] SMP
Felix has fixed up the uncompressed kallsyms support so modify the current
patch with the fix.
All credits for the code go to Felix.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Update fitblk driver which has previously been backported to Linux 6.1
so it can build and work with Linux 6.6.
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Add missing symbols to config for 6.6.
Signed-off-by: Robert Marko <robimarko@gmail.com>
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Signed-off-by: Weijie Gao <hackpascal@gmail.com>
