Get kernel source code

I use Linux mainline kernel source. My Debian has 6.12.12 kernel.

wget https://mirror.nju.edu.cn/kernel.org/linux/kernel/v6.x/linux-6.12.12.tar.xz

unpack to ~/fromsource/linux-6.12.12/

I think the debian source pkg should also work.

Modify the cp210x driver’s source code

You need to add support for your device in the source code. First get the USB vendor ID and product ID. You can get it from kernel log (journalctl -k |tail) or lsusb command. Kernel log is usually like “idVendor=11ca, idProduct=0211, bcdDevice=1.00”, lsusb output is like “Bus 007 Device 005: ID 11ca:0211 VeriFone Inc Verifone USB to Modem”. In my example, vendor ID is 11ca, product ID is 0211. Now you need to modify the source code at ~/fromsource/linux-6.12.12/drivers/usb/serial/cp210x.c

add
{ USB_DEVICE(0x11CA, 0x0211) },

between all those device vendor ID, product ID lines.

Build the cp210x module

Install some dependencies: pahole and resolve_btfids.

sudo apt install pahole libelf-dev

cd ~/fromsource/linux-6.12.12/
cd ./tools/bpf/resolve_btfids
make
ls -l resolve_btfids
sudo mkdir -p /usr/src/linux-headers-`uname -r`/tools/bpf/resolve_btfids
sudo ln -s $(realpath resolve_btfids) /usr/src/linux-headers-`uname -r`/tools/bpf/resolve_btfids

To avoid Skipping BTF generation for xxx Error:

sudo cp /sys/kernel/btf/vmlinux /usr/lib/modules/`uname -r`/build/

Now it’s time to build the cp210x kernel module.

cd ~/fromsource/linux-6.12.12/drivers/usb/serial/
make -C /lib/modules/`uname -r`/build M=$PWD
ls -l cp210x.ko

Install and load cp210x module

# install cp210x
sudo cp usbserial.ko /lib/modules/`uname -r`/kernel/drivers/usb/serial/usbserial.ko
sudo cp cp210x.ko /lib/modules/`uname -r`/kernel/drivers/usb/serial/cp210x.ko
# you must update module dependency map after cp new file to kernel
# module dir. otherwise, you will get "unknown symbol in module" error.
sudo depmod -a

# load cp210x
sudo modprobe cp210x

Now plugin your device, you should have /dev/ttyUSB0

In kernel log you should see

Nov 29 00:25:26 agem10.dev.emacsos.com kernel: usbcore: registered new interface driver usbserial_generic
Nov 29 00:25:26 agem10.dev.emacsos.com kernel: usbserial: USB Serial support registered for generic
Nov 29 00:25:28 agem10.dev.emacsos.com kernel: usbcore: registered new interface driver cp210x
Nov 29 00:25:28 agem10.dev.emacsos.com kernel: usbserial: USB Serial support registered for cp210x
Nov 29 00:25:28 agem10.dev.emacsos.com kernel: cp210x 7-1.2:1.0: cp210x converter detected
Nov 29 00:25:28 agem10.dev.emacsos.com kernel: usb 7-1.2: cp210x converter now attached to ttyUSB0

Now the device is fully working.

Before You Start

Before you try to access the UART via TTL, check whether telnet is enabled on the current firmware. If telnet is enabled, you don’t need to use UART. Also check if a 3rd party u-boot is installed on the device. If it is, you don’t need to use UART.

Check your USB TTL device is working properly. Plug in your device, you should see /dev/ttyUSB0 shown up in your device tree. In my case, I use a CP2102 USB device, I need to build my own kernel module for it to work.

Get latest OpenWRT files for this device. You can find them at https://firmware-selector.openwrt.org/, type in 360t7, the files will show up. You need the SYSUPGRADE, PRELOADER.BIN, BL31-UBOOT.FIP files.

Connect the UART TTL Cables

It is possible to connect 3 cables to the device without opening the case. Put the device on the table, face down. Put antenna on the left side, put RJ45 sockets on the right side. Put a cellphone flashlight on the back of the device, adjust the flashlight spotlight so you can see 4 holes on the board. The first hole is square; the rest are circles. Ignore the first hole, the remaining 3 circle holes are GND, TX, RX. You need to connect them to your TTL device’s GND, RX, TX.

Get 3 thin metal wire. Those used to fasten charging cables works fine. Remove the outside of the wire. Fold them to Z shape so they can be insert to the 360t7 device through the louver holes on the case. Use tape to fix each wire once they are in position. Connect all 3 wires to duPont wire on your USB TTL device. GND-GND, TX-RX, RX-TX.

Prepare Terminal to Connect to Serial

On Linux, install minicom package. Run minicom with

sudo minicom -D /dev/ttyUSB0

Configure minicom by pressing Ctrl-A O

 Serial port setup
 F - Hardware Flow Control
 G - Software Flow Control

Make sure both says No. Otherwise, keyboard input won’t work.

Save this configure as default in Ctrl-A O menu. Otherwise, each time you start minicom, you need to config this again.

Go To OpenWRT Failsafe Mode

With minicom running, all cables connected, now connect power cable to boot your device. You should see u-boot and kernel output in minicom terminal. After kernel boot, wait for 3s, then press F and Enter key repeatedly to enter failsafe mode.

If you never saw failsafe mode text prompt, let the router boot normally, then flash 360T7-v4.2.4.7959_upgrade.bin in stock firmware web UI. In newer firmware, failsafe mode is disabled, revert to this older version to allow OpenWRT failsafe mode.

Enable Telnet and Set Root Password in Failsafe Mode

Once you get to failsafe mode root shell, run these commands:

fw_setenv bootmenu_delay 3
mount_root
# Enable telnet service
sed -i 's/.*local debug=.*/\tlocal debug=1/' /etc/init.d/telnet
# Set password for root, type a password of your choice twice.
passwd root
cat /proc/mtd
# you should see stock Qihoo 360t7 layout.

root@(none):/# cat /proc/mtd
dev:    size   erasesize  name
mtd0: 08000000 00020000 "spi0.0"
mtd1: 00100000 00020000 "bl2"
mtd2: 00080000 00020000 "u-boot-env"
mtd3: 00200000 00020000 "Factory"
mtd4: 00200000 00020000 "fip"
mtd5: 02400000 00020000 "ubi"
mtd6: 02400000 00020000 "firmware-1"
mtd7: 02400000 00020000 "plugin"
mtd8: 00100000 00020000 "config"
mtd9: 00080000 00020000 "factory"
mtd10: 00700000 00020000 "log"

Run “reboot” to reboot the device. Now you should use stock u-boot menu. You can use u-boot menu to flash u-boot and firmware, or you can do it in telnet session after the device is fully boot. I choose to use u-boot menu.

On u-boot menu, select “Upgrade ATF BL2”, serve OpenWRT preloader.bin via TFTP. Run mtkautoboot to show u-boot menu again.

Select “Upgrade ATF FIP”, serve OpenWRT bl31-uboot.fip via TFTP. Run “reset” to reboot device. Now you should see OpenWRT’s u-boot menu. It is different than stock u-boot menu.

  ( ( ( OpenWrt ) ) )       U-Boot 2024.10-OpenWrt-r28872-daca7c049b (Sep 19 20

1. Run default boot command.
2. Boot system via TFTP.
3. Boot production system from NAND.
4. Boot recovery system from NAND.
5. Load production system via TFTP then write to NAND.
6. Load recovery system via TFTP then write to NAND.
7. Load BL31+U-Boot FIP via TFTP then write to NAND.
8. Load BL2 preloader via TFTP then write to NAND.
9. Reboot.
a. Reset all settings to factory defaults.
0. Exit

Press UP/DOWN to move, ENTER to select, ESC to quit

Flash OpenWRT

On Linux computer, serve openwrt-mediatek-filogic-qihoo_360t7-initramfs-recovery.itb via 192.168.1.254 TFTP server. In OpenWRT u-boot menu, select “Boot system via TFTP.” It will boot to OpenWRT initramfs-recovery. When initramfs-recovery boots, use ssh or luci to flash OpenWRT sysupgrade image.

After that, router will auto reboot, you will have latest OpenWRT running on Qihoo 360T7.

Could not open mtd device: fip
Can't open device for writing!

You may use this trick to allow write to the FIP and other MTD partitions.

opkg update
opkg install kmod-mtd-rw

insmod /lib/modules/$(uname -r)/mtd-rw.ko i_want_a_brick=1
# the i_want_a_brick=1 part is required! Otherwise, the command will fail.

Now you can run the mtd write command again.

If write still fail, make sure the partition name is correct. The partition name is case sensitive. You can check the partition name using

cat /proc/mtd

check kernel boot param

Check these two files

cat /proc/cmdline
cat /etc/default/grub

If ipv6.disable=1 is set, the kernel IPv6 stack will be disabled. If ipv6.disable_ipv6=1 is set, the kernel keeps the IPv6 stack functional but will not assign IPv6 addresses to any of your network devices.

Make sure you do not use any of these kernel options if you want to use IPv6.

check interface disable_ipv6 settings

sysctl net.ipv6.conf.lo.disable_ipv6
sysctl net.ipv6.conf.eno1.disable_ipv6
# -r list all values matching this pattern
sysctl -r 'net.ipv6.conf.*.disable_ipv6' net.ipv6.conf

Make sure all of the interfaces that you want to have ipv6 enabled has disable_ipv6=0. Specifically, you SHOULD also enable ipv6 on lo interface. You should also enable IPv6 on Linux bridge and the physical interface that it is connected to.

check interface RA settings if you use SLAAC

RA stands for router advertisement. It is a broadcast packet that allow configuration of IP address and routing. SLAAC relies on RA to work.

sysctl net.ipv6.conf.all.accept_ra
# replace vmbr0 with your interface
sysctl net.ipv6.conf.vmbr0.accept_ra

It should be 1 or 2 depending on your device.
If it’s 0, you need to set this at boot time. e.g. in /etc/network/interfaces

To learn more about accept_ra param, see Linux kernel document. Search accept_ra in that page.

For example, on one of my promox VE node, I have

auto vmbr0
iface vmbr0 inet static
        address 192.168.2.47/24
        gateway 192.168.2.1
        bridge-ports eno1
        bridge-stp off
        bridge-fd 0
iface vmbr0 inet6 auto
        accept_ra 2

If you doubt whether RA is working properly, run tcpdump to capture the packet and view it in wireshark.

# replace INTERFACE with your interface name
sudo tcpdump -i INTERFACE -w /tmp/ra_packets.pcap 'icmp6 and (ip6[40] = 134 or ip6[40] = 133)'

check lo ip6-loopback ping

ping6 ::1
ping6 ip6-loopback

check local hostname is ipv6 resolvable

in /etc/hosts
::1 entry should include your fqdn

check whether ipv6 address is assigned on interface

ip -6 a
ip -6 a vmbr0

If your IPv6 network use SLAAC, IP address should be assigned automatically. One interface usually has one public IPv6 address (starts with 2x:xx:xx) and more than one private IPv6 address (starts with fx:xx:xx).

If your IPv6 network use DHCPv6 only, you should config DHCP in /etc/network/interface or corresponding RHEL config file or Network Manager config file. To just test DHCPv6 is working, you can run:

dhclient -6 INTERFACE

check ipv6 routes is added

ip -6 r
ip -6 r | grep default

IPv6 routes are added by clients via RA (router advertisement).

Make sure the default route either has no expire field or has a good expire field. In openwrt default config, default route expire should be between 1200s and 1800s. If it’s less than 1200s, RA is not properly received and processed by host.

In some home server, if you enable IPv6 on IPMI/SBC/iLO/iDrac and use the shared network port, host OS will no longer receive the RA packet. You may see low expire on default route or missing a default route. In that case, you should disable IPv6 on IPMI.

check IPv6 neighbors for nodes in the same LAN

ip -6 n show|grep -v -E '(FAILED|STALE)'

At least one router should be listed when IPv6 is working. You can compare results on different nodes in the same LAN.

check you can visit other IPv6 nodes on the Internet

ping6 aliyun.com
curl -6 https://aliyun.com
curl -I -6 https://mirrors.tuna.tsinghua.edu.cn/
# test whether your OS prefer IPv6 or IPv4 when both are available
curl test.ipw.cn

Of the choices in title, I have used Ubuntu, Debian and FreeBSD. I had evaluated Arch Linux and Gentoo to a great extend and decide not to use them. I have started to use Linux since around 2005. The first distros that got me interested in Linux was Red Hat Linux 9 and Fedora Core 4. I remember the days when I burned Fedora Core 4 and Ubuntu 6.10 CDs and gave them to others when I was in college. And Ubuntu sent free CDs with free shipping all over the world at that time.

Without further ado, let’s talk about features I care in Linux distro.

Stable Release Instead of Rolling Release

Actually rolling release means no release at all. In stable release, the kernel and package are tested, work together well and the feature set is frozen. Security or critical bug fix patches are provided without breaking backward compatibility. Stable release also means there is a documented process of upgrading old release to a new one. All backward incompatible changes are well documented.

Using a stable release doesn’t mean you have to stick with old versions of software. You can always compile the software you do want. For example, I used to compile mplayer and Emacs in Debian stable. Some software encourage you to use an official build, such as bugzilla. Some software have latest build maintained for Debian stable, such as postgresql. You are definitely not constrained by the software version that is in Debian stable’s apt repository. A stable release will make sure the underlying system and library is solid, if you want to get latest applications, go ahead. On the Linux desktop, AppImage and Flatpak is also becoming more popular. If you want to get OBS or DigiKam, there is a build for any Linux. Personally I still prefer OS native builds though.

If you want to learn more about how a stable release is made. Read Debian Release Management or FreeBSD’s release engineering. If you are a developer or project manager, you know making a good release is hard.

This rules out Arch Linux, which is a rolling release. Rolling release doesn’t mean install once and upgrade forever. It’s more like every package may break at every upgrade. You need to test software you use in a stage environment by yourself if you don’t want surprises. This also rules out Ubuntu non-LTS releases. I don’t call it a release anymore.

Binary Packages Instead of Source Packages

Unless you only use a fixed set of packages and rarely upgrade them, binary packages will greatly simplify and speed up software installation. Even if you have a fairly powerful machine, it still cause great pain if a low level library requires an update, which triggers a chain of recompiles. I was bitten by this once while I was using FreeBSD. For some dedicated web or database server, this may not be a big problem.

A side note for python developers. Since this year (2016), most PyPI packages are in wheel binary format, which makes installing faster and easier. It’s amazing that PyPI is developed and maintained by a single person.

This rules out Gentoo and FreeBSD ports.

Large Selection of Software and Large User Base

The core team should maintain a large selection of software, so if there are problems, they can be exposed and handled inside the project. Large user base means if you hit a problem, it’s likely there is already a fix or a workaround.

This rules out FreeBSD. Its binary package system is not as sophisticated as Debian’s. Binary packages and ports are not part of the official release.

Edit: In 2025, FreeBSD 14 offers significantly more binary packages than it did in 2016. If your hardware runs well on FreeBSD and you don’t need any Linux-only non-free software, you should give FreeBSD a try.

Worldwide Mirrors for Packages

This will speed up operating system install and upgrade, and package install greatly.

Honor Mentions About Other Distro

RedHat Enterprise Linux (RHEL)/Oracle Linux/Amazon Linux, Ubuntu LTS, OpenSUSE.

RHEL is very similar to Debian, if you prefer RedHat’s way of doing things, they are good choices. Also, RHEL has the best documentation between RHEL, Ubuntu and Debian. The document quality order from high to low is RHEL, Debian, Ubuntu. You do need to register an account to view most of the documents though.

Oracle Linux and Amazon Linux are derived from RHEL (or CentOS-stream, Fedora). If you run on Oracle Cloud or AWS, you may use these flavors. They behave similarly to RHEL and do not require a paid license to use. In comparison, the RHEL does require a license to use in mass production. There is a no-cost RHEL developer license that you can use in development environment and small production environment. There are a few community versions of RHEL re-pack when CentOS is acquired and discontinued by RedHat, such as RockyLinux and AlmaLinux. The offer binary compatibility with RHEL. But the default dnf repo and management tools are not all the same.

Ubuntu LTS, although the release is not as good as Debian, does have a large user base and a long term support (LTS). That makes it a much better choice than Ubuntu non-LTS release. Ubuntu LTS also got commercial support if you needs that. I found Ubuntu’s release not as stable as Debian. That’s why I always choose Debian over Ubuntu LTS. Only choose Ubuntu LTS only if you want to purchase support from Canonical.

I have not used OpenSUSE for real, so I won’t remark on them. Based on short experience of OpenSUSE Leap, the installer is very enterprisy. Side note: I think Open Build Service (e.g. OpenSUSE OBS) is a cool project, which works for all Linux projects and distros. But I haven’t tried it yet.

Edit: CentOS is acquired by RedHat and rebrand to CentOS stream. Added notes on Oracle Linux, Amazon Linux and some CentOS remakes.

If you need advice on choosing a Linux distro, you can leave a comment or email me. Do tell me what you want to do with Linux so the advice could be specific to your use case.