Three defects prevented the board from working at all:
- INT0/INT1 were falling-edge triggered. Edge detection needs the I/O
clock, which SLEEP_MODE_PWR_DOWN stops, so neither the reed switch nor
the RTC alarm could wake the MCU. Both are now low-level triggered (the
only asynchronous mode), and each handler masks its own interrupt while
the source is still asserted so the low level cannot retrigger. The reed
and RTC pins also get their pull-ups; they were explicitly driven low.
- Statics were 1440 B of 2048 with a 538 B main frame, so the first NFC
read ran the stack into .data. Shrank the oversized buffers and made the
NFC scratch buffer static: statics 1440 -> 1038 B, main frame -> 204 B.
- The FIFO was filled after entering TX mode with TxStart = FifoNotEmpty,
so transmission began before the payload was loaded. Load in standby.
Memory safety: clamp the unvalidated RX length (len - 4 underflowed to
>=252 into a 60-byte buffer), fix writes one byte past DATA_BUFFER_65,
fix the diameter copy length in st25dv.c, NUL-terminate remove_spaces,
and bounds-check the NDEF parser (dropping its tag-sized VLA and its
unchecked payload_length decrements).
Hangs: add bail-outs to every peripheral poll loop - RFM69 mode/TX/RX
waits, the EEPROM WIP poll, all six I2C TWINT spins, and the ADC. The
LDO is cut before sleeping, so a slow peripheral hung the firmware with
no watchdog armed.
Correctness: boot no longer wipes the EEPROM spool; the replayed packet
is sent once and deleted only on success; short ATOMIC_BLOCK sections
replace the blanket cli() that lost reed pulses during the radio window;
the I2C rail comes up before the RTC is touched; sleep_bod_disable() moves
into the timed sequence with the sleep race closed; sei() no longer runs
inside ISRs; REG_FDEV_MSB was 0x06 twice so deviation was 0; ADC uses
return ADC and a /64 prescaler; SS1 is an output before SPE is set.
VAL_DATA_MODUL_OOK was misnamed rather than wrong - 0x01 lands in
ModulationShaping, not ModulationType - so the register value is
unchanged and on-air behavior still matches the base station.
Verified: builds clean under -Wall -Wextra on both gnu17 and c23.
Not yet run on hardware.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01YVJKatfeMJjAmuH9KYiLuv
thebears
ed476473b6