8d7721c775
B7A: QMI DMS time vs VNIIFTRI NTP (DMS = UTC - 0.28 s, sigma 1 ms; NITZ = DMS truncated to 1 s). B7B: a 1970 -> 2026 date -s step is harmless for the modem stack. B7C: aurora-modem probes DMS/NITZ along the start flow and steps CLOCK_REALTIME once at the first valid sample (REG registered/attached), never RTC; background SNTP check (query only) after START OK. 3/3 class-A cold boots PASS (residual +0.04/+0.36/+0.84 s), cache 14fe453a written (B7C_WRITE_VERIFIED) and verified by a normal power-on (V1, +0.26 s). Also publishes the prerequisite R2 (boot-hang / lk eMMC investigation, T4 telnet baseline that B7C builds on) and R3 (autonomous cold boot, NCM loss) material, and extends tools/publish-sanitize.py to r2/, r3/, b7/. Binary images, initramfs, busybox and raw logs stay out (see b7/*/SHA256SUMS).
3.3 KiB
B7C audit + proposed patch (not built, nothing written)
Live board files == b6p/rootfs (aurora-modem 6e922dec…, init.d 533f47ea…, modem.conf 6ffb206e…) + r2/t4/rootfs/init (13721ca5…).
Clock usage in the current controller / init
| Mechanism | Clock | Affected by a REALTIME step? |
|---|---|---|
| now(), since(), wait_for deadlines, reg_sm T_REG deadline, log timestamps, diag file names | /proc/uptime (CLOCK_BOOTTIME) | no |
sleep N / sleep 0.2 (busybox, nanosleep) |
relative | no |
timeout N (busybox) |
relative | no |
ping -W, udevadm settle --timeout |
relative/monotonic | no |
| qmicli / mmcli / ModemManager timeouts | GLib monotonic | no (B7B: no reaction) |
/init: only sleep loops |
relative | no |
wall clock (date) |
not used anywhere today | — |
| Conclusion: no change needed for timeouts; the patch itself uses only uptime for waits/alignment. |
Kernel
out-bam1 .config: CONFIG_RTC_HCTOSYS=y, CONFIG_RTC_SYSTOHC=y (rtc0 = rtc-pm8xxx, DT without allow-set-time).
SYSTOHC writes the RTC every 11 min only while STA_UNSYNC is cleared (NTP-disciplined). date -s keeps STA_UNSYNC (B7B: status 64).
→ The SNTP stage must never discipline/slew (busybox ntpd without -w would clear UNSYNC): query-only ntpd -w, step via date -s only if |offset| ≥ 1 s.
Insertion points (fresh start path)
QMI_READY (first moment DMS answers) → SIM_READY → RADIO_ONLINE (before reg_sm) → each reg_sm state transition → REGISTERED (before killall qmi-proxy) → BEARER_CONNECTED (after START OK) + background SNTP. Each probe: 2 qmicli calls (~40 ms each), raw kept in /run/aurora-modem/time-probe-N.{dms,nitz}, table /run/aurora-modem/time.tsv, one lifecycle/kmsg line. Set happens at most once ($D/time-set), only while the clock is before the build floor. Policy in the test image: TIME_REQUIRE_NITZ=1 (DMS without NITZ is logged as INVALID(no-nitz), i.e. plausible but not yet trusted) — this still answers "when is DMS first valid"; relax only after evidence.
Patch
b7/b7c/aurora-modem-b7c.diff (b6p → b7c, +55 lines), patched controller sha 932c9ff2… Locally tested under busybox sh with stubbed qmicli/date (real B7A outputs): VALID → one date -s, second probe no re-set; 1980 → INVALID(year<2026); no NITZ → INVALID(no-nitz); SNTP/nslookup parsers on real outputs.
Planned RAM image (after GO)
initramfs-t4.cpio.gz (unchanged) + appended overlay: /usr/sbin/aurora-modem (patched), /etc/aurora/build-epoch (build time),
/usr/libexec/aurora/busybox (Alpine busybox-static 1.37.0-r20 aarch64, static, 1115944 B, sha a6bb1e6e…, used only as busybox ntpd -w).
/init, init.d, modem.conf unchanged. Boot: RESET-held power-on → lk1st fastboot → fastboot boot. No cache/eMMC write.
Build (2026-10-02, GO "сборка")
b7/b7c/build-b7c.sh: T4 initramfs db47835b verified, mkbootimg reproduces T4 bit-for-bit, overlay (aurora-modem 932c9ff2, build-epoch 1790935600 = 2026-10-02T10:06:40Z,
/usr/libexec/aurora/busybox a6bb1e6e) → out/aurora-mm1-b7c.img d0c4cc00… (21291008 B; ramdisk 12628896 B in the image == initramfs-b7c f219b04d).
Layer extraction check: final /usr/sbin/aurora-modem, build-epoch, busybox = overlay; /init, init.d, modem.conf = T4. ntpd applet runs (qemu --help).
Copied to 480s, sha256sum -c SHA256SUMS OK. Host driver for one cold boot: b7/b7c/b7c-boot.sh K (not run yet).