refactor: split motor control out of main.c#33
Merged
Conversation
Extract behavior-neutral control path units so demag, faults, and port work can land without editing a 2500-line god file: - commutation.c: getBemfState, commutate, advanceincrement, zcfoundroutine - bemf_zc.c: PeriodElapsedCallback, interruptRoutine, startMotor - control_loop.c: setInput, tenKhzRoutine, PID/current helpers - motor_runtime.h: extern map for shared runtime state still defined in main F051 duty/COM timer macros use statement form for GCC 15 -Werror. Builds verified: ARK_4IN1_F051 (HWCI_PERF=1) and SEQURE_4IN1_F421.
Behavior-neutral move of stuck-rotor, signal-timeout, BEMF timeout policy, and interval-stall handling out of main/control_loop so the control path is easier to reason about and demag work can layer on top.
Further behavior-neutral split of main.c: - settings.c: load/save EEPROM apply + bootloader checkDeviceInfo - brushed.c: BRUSHED_MODE control loop - runtime_loop.c: variable PWM, desync, IRQ priority, telemetry, ADC/LVC, and sine/stepper vs brushless running branch main.c drops to ~840 lines (init + thin main loop).
Ownership of shared motor-control state now lives with motor_runtime (definitions in .c, declarations in .h). main.c is init + thin main loop only (~535 lines). Expand motor_runtime.h for symbols modules previously pulled via local externs.
Introduce esc_state (DISARMED → ARMING → ARMED_IDLE → SINE/OPEN/CLOSED → BRAKE / FAULT_*) as the named policy mode. Transitions update both the enum and legacy flags so ISRs and existing modules keep reading armed/ running/stepper_sine/old_routine without hot-path changes. Wire arming, stuck/signal/LVC faults, desync/stall, sine handoff, and main-loop reconcile. Smoke: noprop 100% on 3 A bench (pass).
- Transition allow-list with escCommitState; illegal edges counted (optional ESC_STATE_STRICT hang); reconcile forces without the table - Policy helpers mirror legacy flags for 20 kHz/ISR-coupled paths; reconcile at tenKhz + main loop keeps esc_state aligned for logs - hwci_perf v5: esc_state + esc_illegal_edge_count; host decoder, CSV columns, sim, and unit tests - test_esc_state.py mirrors the C edge table Bench: noprop 100% hold, CLOSED_LOOP, 0 illegal edges, 0 BEMF timeouts
Fail noprop smoke when the motor never really spun: require high-throttle running fraction, zero stuck-rotor latch (bemf>=102), OPEN/CLOSED esc_state, zero illegal edges, and min RPM floors. Gates are profile-configured, independent of baseline regression, and drive hwci analyze/ci exit codes. Verified: good free-run exit 0; stuck-latch run exit 1.
Add scripts/cppcheck-ark.sh (make cppcheck) with max-configs=1 so targets.h does not explode config space, suppressions for CMSIS/HAL, and a GitHub Actions workflow that installs cppcheck and runs on every PR. CI fails on error/warning only; style findings are advisory.
1) Size / sources - Exclude brushed.c unless target name contains BRUSHED; exclude hwci_perf.c unless HWCI_PERF=1 (F051 production drops ~0.8% flash) - scripts/check-size-ark.sh + make size-check-ark enforce 98% flash / 97% RAM on ARK F051 with HWCI_PERF - CI job size-ark in static-analysis workflow 2) HAL-facing extract - adc_app.c: vendor ADC kick/temp/bus/current (out of runtime_loop) - pwm_app.c: setPwmDeadTime() (out of settings load path)
- Host model (esc_state_model.py) parses esc_allowed[] from firmware C (single source of truth) and mirrors reconcile/named transitions - Expanded tests: arm→open→closed, sine handoff, stall, stuck/signal/LVC, illegal-edge counter, flag-backed predicates before reconcile - Policy call sites use escIsArmed/escIsDriving/escInSineStart/ escInPollZcDrive/escInOpenLoop/escInClosedLoop after tenKhz reconcile - Size gate ceilings 99% flash / 98% RAM for tight F051+HWCI builds
Drop unused heap reservation, keep the sine table in flash, and inline esc_state predicates so RAM-resident callers avoid long-call veneers. HWCI RAM ~97.6%→82.1%; flash region ~98.5%→98.1%. Tighten size-gate RAM budget to 90%.
Shrink F051 flash by compiling most code for size. Keep commutation, BEMF/ZC, 20 kHz control, and F051 phaseouts at -O3 via RAM_FUNC optimize attributes and a phaseouts pragma (single-shot link cannot mix trailing -O3 per file). HWCI ~98%→88% flash; size gate flash 95%.
Restore MCU_SITL checkDeviceInfo / poles guard in settings.c, portable flash section for filename, uint32 desync_happened for DroneCAN/SITL, serial_telemetry include for SITL telemetry path, and longer DShot signal-loss coast settle for CI.
AlexKlimaj
force-pushed
the
feat/split-main-control
branch
from
aef460e to
68494d7
Compare
The 3 A supply brick current-limits; a host max_current near 3 A was aborting clean free-run holds on stand glitches. Raise host current limit and leave temp/rpm/thrust gates in place.
Global -O3 again for main-loop throughput (was -Os for flash). Stop calling escReconcileFromFlags from tenKhzRoutine — flag-backed predicates already drive policy there; reconcile once per main loop only, removing the F051 RAM→flash veneer on every 20 kHz tick. Size gate flash limit 99% for HWCI_PERF images.
- getAbsDif is static inline (desync and other hot callers). - faultSignalTimeoutTick is RAM_FUNC so tenKhzRoutine does not veneer into flash every 50 us on F051. Tried inlining variable-PWM/desync into main; rolled back after free-run 100% stalls. Prefer same-TU/out-of-line for those policies.
Bring main_instrumented BDShot baseline onto feat/split-main-control: HWCI_PERF v6 fields (rx/tx/telem/EDT), dshot.c hooks, host decoder/CSV/sim, PX4 capture/sustain scripts, and docs. Version is v6 here (v3–v5 already used for zc/phase/esc_state). Hardware: ARK FPV BDShot300 free-run on battery — RPM tracks throttle (~7.7k@12% … ~18k@40%), soft ramp-down, unit tests green.
Two exclusive setups used at different times: - SETUP A: Flight Stand drives uni DShot (rig.yaml / rig.flightstand.yaml, hwci run + noprop profiles). No PX4 on the signal pin. - SETUP B: ARK FPV BDShot owns the pin (rig.px4_bdshot.yaml with throttle_backend none, px4_motor_stream / capture scripts). Add NullThrottle, BENCH_SETUPS.md, and clear labels on scripts/profiles so the two paths cannot be mixed by accident.
Member
Author
Hardware validation (feat/split-main-control @
|
| Test | Setup | Result |
|---|---|---|
| BDShot free-run ladder to 100% vs baseline | B (PX4) | PASS — no regression vs main_instrumented |
| Prop smoke to 30% | A (Flight Stand) | PASS |
| Efficiency sweep 10→100% | A (Flight Stand) | PASS |
low_rpm_crawl (8→3%) |
A (Flight Stand) | PASS |
startup_reliability (10 cold starts) |
A (Flight Stand) | PASS |
1. BDShot A/B free-run to 100% (SETUP B — PX4 owns signal)
Artifacts: hwci/runs/ab-bdshot100-20260710_160854/
| cmd | A max RPM | B max RPM | ΔB−A | A mean | B mean |
|---|---|---|---|---|---|
| 10% | 8 842 | 8 800 | −42 | 6 879 | 6 917 |
| 20% | 12 242 | 12 285 | +43 | 10 752 | 10 753 |
| 40% | 18 428 | 18 200 | −228 | 18 140 | 18 014 |
| 60% | 25 971 | 26 214 | +243 | 25 561 | 25 837 |
| 80% | 33 485 | 33 485 | 0 | 32 820 | 32 814 |
| 100% free-run | ~3 k mean | ~3 k mean | — | both collapse | same on A & B |
| 50% after | recovers | recovers | — | ~21.7 k | ~21.8 k |
- Peak RPM both: 33 485; spin_frac A=0.952 / B=0.959
error_countΔ: A=+2143, B=0- No high-throttle armed drop on either side
- Note: 100% free-run RPM collapse (~3 k while servo=full) is shared by baseline and split, not a modularization regression. After step-down to 50%, both recover.
Verdict: no BDShot free-run regression of the refactor vs instrumented upstream.
2. Prop smoke to 30% (SETUP A — Flight Stand)
Run: hwci/runs/split-prop-smoke30-20260710_161808
| Hold | RPM | Thrust | Current |
|---|---|---|---|
| 10% | ~4 632 | ~19 gf | ~0.23 A |
| 20% | ~8 383 | ~57 gf | ~0.71 A |
| 30% | ~11 542 | ~140 gf | ~1.55 A |
Abort: none · armed drop: none · demag: 0 · max I: 1.67 A
3. Efficiency sweep 10→100% (SETUP A)
Run: hwci/runs/split-efficiency-sweep-100-20260710_161927
Profile: efficiency_sweep
| Throttle | RPM | Thrust (gf) | I (A) | g/W |
|---|---|---|---|---|
| 10% | 3 591 | 11 | 0.15 | 3.69 |
| 20% | 8 381 | 68 | 0.72 | 4.00 |
| 30% | 11 526 | 139 | 1.55 | 3.80 |
| 40% | 14 348 | 229 | 2.76 | 3.54 |
| 50% | 17 134 | 349 | 4.60 | 3.28 |
| 60% | 19 690 | 444 | 7.01 | 2.79 |
| 70% | 21 849 | 545 | 9.93 | 2.46 |
| 80% | 23 790 | 649 | 13.16 | 2.27 |
| 90% | 25 434 | 748 | 16.83 | 2.09 |
| 100% | 26 809 | 806 | 20.79 | 1.88 |
- Abort: none · armed drop: none · demag: 0 · t100 run_frac=1.0
- Peak ~1044 gf / 22.2 A · idle loop ~38.3 kHz · CPU max ~74% · FET ~39 °C
4. low_rpm_crawl (SETUP A)
Run: hwci/hwci/runs/split-crawl-startup-20260710_162126/low_rpm_crawl
(weak-BEMF: spin 10% then 8→6→4→3%)
| Hold | RPM | Thrust (gf) | run_frac | zc_jitter % |
|---|---|---|---|---|
| 8% | 3 801 | 10.8 | 1.000 | 0.06 |
| 6% | 2 989 | 7.1 | 1.000 | 3.18 |
| 4% | 2 117 | 3.0 | 1.000 | 1.12 |
| 3% | 1 682 | 1.3 | 1.000 | 1.45 |
Abort: none · demag: 0 · BEMF timeout samples: 0 · armed drop: none
5. startup_reliability (SETUP A)
Run: …/startup_reliability — 10× cold start at 8%
| Metric | Result |
|---|---|
| Attempts | 10 |
| Failures | 0 |
| Time-to-run max | 16.3 ms |
| Time-to-run mean | ~11 ms |
| Demag | 0 |
Setups (do not mix signal wiring)
| SETUP A | SETUP B | |
|---|---|---|
| Signal | Flight Stand ESC out | ARK FPV BDShot |
| Config | rig.yaml / config/rig.flightstand.yaml |
config/rig.px4_bdshot.yaml |
| Command | hwci run --profile … |
scripts/px4_motor_stream.py |
| Docs | README | docs/BENCH_SETUPS.md |
Bottom line: modular split holds prop-on thrust-stand sweep through 100%, weak-BEMF crawl, cold starts, and matches main_instrumented on BDShot free-run through 80% (shared free-run quirk at 100% on both trees).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Behavior-neutral modularization of motor control out of
main.c, plus a top-level drive state machine:commutation.cbemf_zc.ccontrol_loop.cfaults.csettings.cruntime_loop.cbrushed.cBRUSHED_MODE)motor_runtime.cesc_state.cmain.cis init + thin main loop. Legacy flags stay for ISR hot path; SM is the named policy mode.Bench
noprop_smoke_100pct_3aon 25 V / 3 A no-prop — full hold at 100% (exit 0)Test plan
make ARK_4IN1_F051 HWCI_PERF=1make SEQURE_4IN1_F421