First successful NixOS userspace on the T-Display-K230

Captured 2026-09-20 with tools/capture-boot.py over the CH342 console (/dev/ttyACM0, 115200 8N1). The board reached a root shell; these are commands run on it, not on the build host.

␛[?2004l␛[?2004h
[root@nixos:~]# 
␛[?2004l␛[?2004h
[root@nixos:~]# [   0.204] --- sending: cat /proc/cpuinfo ---
cat /proc/cpuinfo
␛[?2004lprocessor	: 0
hart		: 0
isa		: rv64imafdcv_zicbom_zicboz_zicntr_zicsr_zifencei_zihpm_zba_zbb_zbs_svpbmt
mmu		: sv39
uarch		: thead,c908
mvendorid	: 0x5b7
marchid		: 0x8000000009140d00
mimpid		: 0x50000
␛[?2004h
[root@nixos:~]# 
␛[?2004l␛[?2004h
[root@nixos:~]# [   0.404] --- sending: uname -a ---
uname -a
␛[?2004lLinux nixos 6.6.36 #1-NixOS SMP Tue Jan  1 00:00:00 UTC 1980 riscv64 GNU/Linux
␛[?2004h
[root@nixos:~]# 
␛[?2004l␛[?2004h
[root@nixos:~]# [   0.604] --- sending: cat /proc/cmdline ---
cat /proc/cmdline
␛[?2004lconsole=ttyS0,115200n8 root=fstab loglevel=4 lsm=landlock,yama,bpf init=/nix/store/2h0xrlf5n74fjjhxqd18y6vl55w4zyrr-nixos-system-nixos-26.11.20260919.20b1ddd/init
␛[?2004h
[root@nixos:~]# 
␛[?2004l␛[?2004h
[root@nixos:~]# [   0.805] --- sending: findmnt -no SOURCE,FSTYPE / ---
findmnt -no SOURCE,FSTYPE /
␛[?2004l/dev/mmcblk1p2 ext4
␛[?2004h
[root@nixos:~]# 
␛[?2004l␛[?2004h
[root@nixos:~]# [   1.005] --- sending: nproc; free -m | head -2 ---
nproc; free -m | head -2
␛[?2004l1
               total        used        free      shared  buff/cache   available
Mem:             935         136         773           6         101         798
␛[?2004h
[root@nixos:~]# 
␛[?2004l␛[?2004h
[root@nixos:~]# [  90.179] --- capture end ---

What this proves

• We are on the real silicon. uarch: thead,c908, mvendorid: 0x5b7 (T-Head), marchid: 0x8000000009140d00. This cannot be produced by the x86 build host or by QEMU’s virt.
• Our kernel, not the vendor’s. Linux nixos 6.6.36 #1-NixOS.
• Our command line reached the kernel. /proc/cmdline is exactly the string bootargs.txt carries, init= included. This is the fix in stage1-emergency-mode.md, confirmed end to end.
• root=fstab works on this board. There is no explicit root= on the command line, yet / is mounted from /dev/mmcblk1p2 ext4. Root was therefore resolved through the initrd’s fstab and the NIXOS_SD label, which boot-path-differences.md had flagged as never yet exercised — every previous boot got an explicit root= from the vendor fallback. That risk is now closed.
• 935 MB usable of the 1 GiB LPDDR.

The board runs Linux on ONE hart, and that is by design

nproc reports 1, and /proc/cpuinfo lists only processor: 0. This is not a misconfiguration on our side:

• arch/riscv/boot/dts/canaan/k230.dtsi declares cpu@0 and nothing else.
• No K230 device tree in the Xuantie tree declares a cpu@1. Every cpu@1 in dts/canaan/ belongs to the K210, an unrelated dual-core part.
• CONFIG_SMP=y is set, so the kernel is capable of SMP; it is the device tree that describes a single hart.

The K230D does have two C908 cores. The second is the one Canaan’s AMP split gives to RT-Smart — which is consistent with the board having shipped running RT-Smart on it. Bringing it under Linux would mean describing it in the device tree and taking it away from that firmware, which is a separate piece of work with its own risks, not a detail of this change.

Consequence for the-board-boots-what-we-built task 4.2. Its verification clause asks for cat /proc/cpuinfo “showing two C908 harts”. That is not achievable on this hardware under Linux, so the clause is wrong rather than unmet. The substance of the task — reach an interactive prompt, run a command that could only run on this hardware — is satisfied above. The clause needs restating to “showing a thead,c908 hart”, and the single-hart fact belongs in the spec. Flagged rather than silently rewritten.