Skip to content

nRF52840 Renode smoke — direct-probe failures (2026-11-08)

Direct labwired test runs of the new firmware-nrf52840-renode-smoke bin against configs/systems/onboarding/nrf52840.yaml, before the final pipeline run. The files in this directory are from probe 1.

  1. max_steps 400,000 — FAIL 1/2. UARTE0 lines take ~100k cycles each at the configured baud. The run stopped while waiting on EVENTS_ENDTX after the GPIO line (stalled_poll: uart0 +0x120). This was a step budget problem, not a model problem. The final script uses 4,000,000.
  2. max_steps 4,000,000, first firmware — FAIL 1/2, summary d7:
    gpiote out0 p0.14=00 FAIL
    rtc0 cmp=0 counter>=4 1 FAIL
    
  3. GPIOTE: the firmware read GPIO OUT. On silicon a task-mode channel drives the pad and leaves OUT alone, so the firmware was wrong. Fixed to read IN.
  4. RTC0: COUNTER advanced, but EVENTS_COMPARE[0] stayed 0. The model latches the compare event only when EVTEN.COMPARE0 is set. The firmware now enables it, as nrfx does. The INTEN-only path is not exercised.
  5. IN read, P0.14 PIN_CNF = 1 (DIR = output) — FAIL, gpiote out0 p0.14 in=00. Engine gap: Nrf52Gpio::effective_in returns OUT & DIR for an output pin and ignores the GPIOTE pad latch. On silicon, task mode overrides the pin's direction and level. nrfx configures task pins as outputs, so this configuration matters. It is NOT claimed by the final smoke: P0.14 is left as an input (PIN_CNF = 0, buffer connected), and the model handles that correctly. Fixing it means changing the shared GPIO IN fast path.