q/uzbek-plus
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
uzbek-plus/b6p/B6P-DESIGN.md

3.8 KiB
Raw Permalink Blame History

B6P — persistent modem integration (P0–P5). Nothing written to flash.

P0 freeze — proven source

Proven B6N implementation = b6j/aurora-modem aa9e9aa2… (b6j-in6.tar d5b551b5) + n.conf aa9e27af… + b6j/n.conf (J1 lifecycle, MODEM_REG_SM=1, T_REG=600). Production controller rootfs/usr/sbin/aurora-modem = that file with diagnostic code REMOVED only (diff-b6n-vs-b6p.diff):

  • removed: MODEM_RMTFS_PERSIST (J3), MODEM_HOOK_PRE_ONLINE/PRE_MM (B6K AT), MODEM_REG_SNAP, MODEM_STOP_DISABLE branch (mmcli --disable / qmicli low-power; n.conf had it OFF), re-attach holdoff (n.conf had 0), old 180 s wait_for path (REG_SM is now the only path), rmtfs-stop command.
  • kept verbatim: start / disconnect / stop / status / addroute, registration state machine reg_sm (local-failure checks, transition log), safety gates, rollback-on-failure (MODEM_KEEP_ON_FAIL=0 default), stop order.
  • changed: config file /etc/aurora/modem.conf (APN/USER/PASSWORD/AUTH/IP_TYPE/REG_TIMEOUT/ROUTES → mapped to the internal MODEM_* names); state dir /run/aurora-modem (was /tmp/aurora-modem.d); rmtfs start check counts new 'registering services' lines (old: grep of an appended log could match a previous run's line).
  • diff-old-b6-vs-b6p.diff: vs the original b6/ (session 28) for reference.

P1 runtime

boot → /init: eMMC all RO → NV snapshot → /firmware vfat RO + firmware path → USB NCM 172.16.42.1 → /etc/init.d/aurora-modem start (background) → controller: RO gates → rmtfs -r -P -s -v → MSS → wwan0qmi0 +2 s → SIM ready → DMS online → reg_sm (registered + PS attached, ≤600 s) → udevd/dbus/polkitd/ModemManager → simple-connect → IPv4 on wwanN + /32 routes (bearer DNS + ROUTES), no default route, usb0 untouched. MSS/rmtfs/MM stay up. Reconnect: aurora-modem disconnect + aurora-modem start (MSS not restarted). Shutdown: /etc/init.d/aurora-modem stop = bearer disconnect → routes → wwan DOWN → MM stop → BAM suspended → rmtfs SIGTERM → MSS offline → rmtfs exits last. No automatic retry of MSS/rmtfs. On failure the controller rolls back with a full stop and the service stays idle.

P2 audit (golden mm1, initramfs 250bf8ee)

  • OpenRC: NOT present (no openrc, no /etc/init.d, no runlevels). PID 1 = custom /init (busybox sh): proc/sys/devtmpfs/configfs, eMMC setro + verify, NV snapshot (/tmp/nv-before.txt), by-partlabel symlinks, modem FAT RO at /firmware, firmware_class path, USB gadget NCM, eMMC report, then endless UART shell.
  • udev/dbus/polkit/MM: not started at boot (only by bin/mm-up or the controller). Packages: 36 Alpine v3.24 apks (linux/initramfs-mm1/dl), rmtfs 1.3, busybox-static 1.37.
  • Decision (user, 2026-09-30): no OpenRC; one service script /etc/init.d/aurora-modem (start|stop|status) called once from /init. Kill switch: cmdline aurora.modem=off.

P3 config — /etc/aurora/modem.conf (Beeline test profile; operator-published default credentials, not private)

P4 time

Controller timing uses /proc/uptime only; LTE start/bearer do not need wall time. RAM tests check DNS/ICMP/HTTP; HTTPS optional (skipped by datatest when year < 2026). RTC untouched, no NTP.

P5 candidate

out/aurora-mm1-b6p.img (build-candidate.sh): kernel payload = golden Image.gz-dtb (0ff9c802), same header/addresses/cmdline, ramdisk = golden initramfs.cpio.gz (exact 11 944 942 bytes, 250bf8ee) + appended gzip member overlay.cpio.gz (kernel unpacks concatenated archives; later entry replaces /init). Golden itself is not rebuildable bit-for-bit from the current tree (built without -t → source mtimes; bin/b4b2-mm edited later), hence the append approach. mkbootimg.py reproduces golden 38c960ef bit-for-bit from golden inputs (checked in the build). Manifest diff (out/manifest-diff.txt): REPLACES /init; NEW /usr/sbin/aurora-modem, /etc/init.d/aurora-modem, /etc/aurora/modem.conf, dirs etc/aurora, etc/init.d.