Work log: phone-unavailable bench prep

The vibration motor moved from a DTS guess to a real FreeBSD candidate: gpio_vibrator(4) attaches to the standalone pinephonepro-vibrator node, advertises EV_FF / FF_RUMBLE, and exposes passive sysctls for active, play_count, stop_count, last_error, and suspend/resume state. 5e5c8f8 vibrator: add gpio rumble candidate added the driver and evdev force-feedback plumbing. d3857f0 vibrator: add rumble bench verifier added the userland rumble verifier. e3b9325 vibrator: quiesce on suspend added suspend quiesce so a motor cannot stay enabled across a sleep transition.

Two camera-adjacent flash paths got bench helpers. The SGM3140 helper keeps the default mode status-only, then offers bounded torch and flash pulses when explicitly requested. 47cd7e0 sgm3140: add bounded flash bench helper added the helper, and the later bench-analyzer pass gave its status path stable PASS/WARN markers.

The GD25LQ128 SPI NOR work got a read-only verifier instead of another note-to-self. b957995 gd25lq128: add read-only bench verifier reads the rk2aw/idbloader and U-Boot FIT offsets through /dev/flash0, hashes the samples, rejects all-zero bootloader data, and checks the FIT magic at sector 20480.

The sensor receipt became one command. 70763a9 sensors: add combined phone receipt replaced the old two-sensor status script with mise run debug:sensors:phone, a combined read-only receipt for STK3311, MPU-6500, magnetometer, RK3399 TSADC, and RK3399 eFuse. It prints raw sysctl trees plus a compact predicate block: attached, expected IDs, plausible ranges, last_sample_error, and TSADC shutdown state.

The per-component receipts were wrapped into a phone bench bundle. 9009f9f bench: add phone receipt bundle added mise run bench:phone, which writes timestamped logs under logs/phone-bench/<timestamp>-<name>/. The default path is non-destructive: sensors, SPI NOR read-only verifier, SGM3140 status, and modem status. Physical outputs stay behind --with-actuators, which adds vibrator rumble and bounded SGM3140 pulse checks.

41c2bfa bench: add phone receipt analyzer added mise run bench:phone:analyze and made the bench runner emit analysis.txt automatically. The analyzer turns the saved bundle into a PASS/WARN/MISSING/FAIL table so the next hardware session starts with a one-screen receipt rather than a pile of logs.

The next requested step was a Coppice build smoke for current master. The local Coppice CLI was not installed, so the run used the gateway API at http://192.168.122.181:3000 directly. Auth was disabled, /templates reported only python-bhyve, and /snapshots reported three python-bhyve checkpoints. Cold create, snapshot fork, and a jail-template probe all failed to produce a live sandbox: the gateway tasks stayed at backend create started / backend fork started with no entry in GET /sandboxes. The stuck tasks were cancelled through POST /tasks/:id/cancel.

The local receipt for the blocked substrate attempt is logs/coppice-build/20260501-024215-coppice-substrate-blocked.log. Node2 showed the same behavior against its own gateway and was captured in logs/coppice-build/20260501-024357-coppice-node2-substrate-blocked.log.

Coppice came back far enough to run real build work. mise run coppice:preflight succeeded against both bhyve gateway nodes, and the build-smoke helper was corrected to match the honor pipeline more closely: it now fetches source archives with base fetch / tar, copies jsm222/drm-subtree before local overlays, skips package LLVM, and can run buildkernel with COPPICE_KERNEL_NO_MODULES=1 for constrained bhyve guests.

The best build attempt was on node2 in logs/coppice-build/20260501-095947-build-smoke.log. It fetched the project archive and FreeBSD stable/15, copied the DRM subtree, applied the local overlays and patches, and reached:

make -C /work/freebsd-src -DWITHOUT_CLEAN WERROR= NO_MODULES=yes -j1 buildkernel

At the last live poll it was compiling kernel objects from the PINEPHONE_PRO config with /work/freebsd-src at 1.7 GB and /work/obj started. The run ended when the Coppice gateway returned curl: (52) Empty reply from server; the sandbox was gone after the API came back. A patch-only retry on node2 later rebooted the guest mid-run, and node1 still showed resolver failures from inside the bhyve guest.

With the phone still unavailable, the next requested step was to build on honor. SSH to honor failed with No route to host for 192.168.1.149:22, so the no-phone work stayed local: source-audit the hardware map against the in-tree DTS and the component pages, then correct stale spec/status claims that would waste the next bench session.

The fixes were documentation-only in this pass: HP detect is already named in DTS on GPIO4_D4 and now correctly reads as a policy gap, not a DTS gap; TSADC/eFuse/PL330/vibrator are partial candidates rather than missing; SARADC is working, not missing; the component matrix now lists IMX258 on DTS &i2c1 addr 0x1a; the FUSB302 note now says DTS &i2c4 addr 0x22 / FreeBSD shifted addr 0x44 on iicbus3; and the broad root summary no longer publishes stale exact totals.

The Coppice build smoke now survives a full PINEPHONE_PRO kernel compile. fda814e build: detach coppice kernel smoke changed tools/coppice-build-smoke.sh so long kernel builds run detached inside the bhyve guest and the local task polls /work/logs/coppice-build.log through short execs. That avoided the gateway empty-reply failure mode without losing the build transcript.

The first detached full-kernel run reached the final link and exposed a real source issue instead of an infrastructure issue: fusb302.o referenced rk818_charger_set_input_current_ma, but rk818_battery.c was not listed in sys/conf/files.arm64. 41aead6 build: include rk818 charger helper adds the helper object to the arm64 file list. The retry in logs/coppice-build/20260501-105104-build-smoke.log fetched commit 41aead6, applied Honeyguide + DRM + local overlays + patches, and completed:

PASS: Coppice build smoke completed

The next Coppice improvement is snapshot reuse, so fresh sandboxes do not repeat the slow FreeBSD and DRM archive setup. mise run coppice:prepare-snapshot now creates a bhyve sandbox, caches FreeBSD stable/15 and jsm222/drm-subtree under /work/cache, snapshots that prepared filesystem, and deletes the source sandbox. coppice:build-smoke then recognizes /work/cache/freebsd-src and /work/cache/drm-subtree in a forked snapshot, while still fetching the current pine64-freebsd archive for each run.

The first prepared source-cache snapshot is 903b69a29ce3, from logs/coppice-build/20260501-110823-prepare-snapshot.log. A forked patch smoke in logs/coppice-build/20260501-111422-build-smoke.log confirmed the cache path:

== copying cached FreeBSD source archive ==
== copying cached drm-subtree ==
PASS: coppice build smoke completed