Fix int8_t overflow in AAF frequency selection for ICM-42688P/42686P

[sensei-hacker]

Summary

getGyroAafConfig() in src/main/drivers/accgyro/accgyro_icm42605.c tracks the running "best candidate" Anti-Alias Filter (AAF) frequency in a local variable selectedFreq declared as int8_t (max value 127). The aafLUT42688[] table (used for ICM-42688P/42686P) contains freq values up to 1962, all stored as uint16_t in aafConfig_t. Any LUT value above 127 silently truncates on assignment to selectedFreq, corrupting the closest-frequency comparison and causing the wrong AAF register configuration to be selected.

This affects every board using the ICM-42688P/42686P gyro at INAV's default hardware LPF setting (GYRO_LPF_256HZ, set unconditionally in gyro.c): the function selects 303Hz instead of the correct 258Hz AAF cutoff.

Changes

• Changed selectedFreq from int8_t to uint16_t in getGyroAafConfig(), matching the type of aafConfig_t.freq (the field it tracks) and desiredFreq (the value it's compared against).

Testing

• Wrote a standalone test harness reproducing the exact struct/LUT/algorithm from the source file, confirming the bug: for GYRO_LPF_256HZ (default) on the 42688P table, the buggy code selects 303Hz instead of the correct 258Hz; also reproduced for GYRO_LPF_188HZ on both the 42688P and 42605 tables.
• Re-ran the same harness with the fix applied — all cases now select the correct closest-frequency LUT entry.
• Built the AETH743Basic target (uses ICM-42688P, confirmed via WHOAMI-detected ICM42605_VARIANT_42688P at runtime, exercising aafLUT42688[] directly) — build succeeds with zero warnings on the changed file.
• Verified via manual trace of C integer promotion rules that the ABS() macro (src/main/common/maths.h) has no signed/unsigned hazard now that desiredFreq, selectedFreq, and aafConfigs[i].freq are all uint16_t — both operands promote to signed int before subtraction on all INAV-supported 32-bit targets.

Code Review

Reviewed with inav-code-review agent — approved, no issues found.

Related

This is a long-standing pre-existing bug present on master, maintenance-9.x, maintenance-10.x, and release/9.1. Flagged by Qodo bot review on PR #11681 (a release/9.1 → maintenance-10.x branch-sync merge) and independently verified against the actual file content. Targeting release/9.1 per current branch-basing guidance; this will flow up to maintenance-10.x via the normal release/9.1 → maintenance-10.x merge process.

[qodo-code-review]

PR Summary by Qodo

Fix AAF frequency selection overflow for ICM-42688P/42686P

🐞 Bug fix 🕐 Less than 10 minutes

AI Description

• Fix AAF closest-frequency selection by preventing int8 truncation of LUT frequencies.
• Ensure ICM-42688P/42686P selects the correct AAF cutoff for default LPF settings.

Diagram

graph TD
  A["Gyro LPF setting"] --> B["getGyroAafConfig()"] --> C["AAF LUT (42688/42605)"] --> D["selectedFreq (uint16_t)"] --> E["Closest match"] --> F["AAF registers"]
  subgraph Legend
    direction LR
    _cfg["Config/input"] ~~~ _fn["Function"] ~~~ _tbl["Lookup table"]
  end

Loading

High-Level Assessment

The PR’s approach is the correct minimal fix: use a wide enough type (uint16_t) for selectedFreq to match the LUT frequency domain and prevent truncation. Alternatives like casting during comparisons or using a separate delta-only variable add complexity without improving correctness.

Files changed (1) +1 / -1

Bug fix (1) +1 / -1

accgyro_icm42605.cPrevent AAF LUT frequency truncation during closest-match selection +1/-1

Prevent AAF LUT frequency truncation during closest-match selection

• Changes getGyroAafConfig() to track the running selected frequency as uint16_t instead of int8_t. This prevents overflow/truncation when LUT entries exceed 127Hz (notably on the ICM-42688P/42686P table), restoring correct closest-frequency selection.

src/main/drivers/accgyro/accgyro_icm42605.c

[qodo-code-review]

Code Review by Qodo

🐞 Bugs (0) 📘 Rule violations (0) 📎 Requirement gaps (0)

Great, no issues found!

Qodo reviewed your code and found no material issues that require review

[github-actions]

Test firmware build ready — commit 1ffa80b

Download firmware for PR #11682

240 targets built. Find your board's .hex file by name on that page (e.g. MATEKF405SE.hex). Files are individually downloadable — no GitHub login required.

Development build for testing only. Use Full Chip Erase when flashing.