Skip to content

Changelog

What changed in each release of the driver. Entries under Breaking need source changes in code that used the affected API.

[Unreleased]

Added

  • First version of the PF1550 driver (NXP PF1550 PMIC, I2C): typed, limit-checked and read-back configuration of the charger, the three buck regulators, the three LDOs, VREFDDR, VSNVS, the coin cell charger, the USBPHY LDO and the charger LED; the state machine, SHIP and CORE_OFF behind confirmation tokens; the power-down order behind an acknowledgement; the interrupt groups with clear-after-copy; and the power-up and power-down sequencer as constexpr data for the pre-programmed options A1 to A9.
  • Mandatory Limits and Acknowledgements; no public raw register write.
  • A host-test model of the chip, a fuzz target, compile-fail tests, a host example and ESP-IDF examples (configure and charge, interrupts and the charger watchdog).
  • Not verified on hardware.
  • Ninth review: a CHG_INT byte read by ReadRegister() also owes the input limit check (the flag is raised where every status byte is noted, so the next ServiceInterrupts() restores the limit; tests/test_round10.cpp); the fuzz target drops its last exemption (a raw CHG_INT read) and owes the check for every charger byte the latch shows; the round-9 sweeps take their bound from a clean run of the same call; a corpus input walks charging, a cable removal with its interrupt, a failure at the second-pass INT_CATEGORY read, ClearLatchedEvents() and a clean service, so the corpus replay finds a lost pending check; contract and testing pages list the INTB shortcut's pending-check condition and every fuzz promise.
  • Eighth review: a charger interrupt read by a ServiceInterrupts() call that then failed (a later pass, the clear write) owes the input limit check, so the next successful call restores a limit that went back to its OTP value, with or without ClearLatchedEvents() between; tests/test_round9.cpp fails each transfer of the call in turn and pins the five untested guards (mask write failure, latch kept after an error, failed switch-off is Bus, ClearLatchedEvents() empties the latch, a failed VBUS_SNS read is an error); the fuzz target calls ClearLatchedEvents() and no longer exempts a failed service from the limit invariant.
  • Fuzz finding after the seventh review: the reset-seen invariant armed bits the driver had not changed (a rail mode bit left alone is no sentinel); it now arms only the changed bits (all mode bits for a switch-off), documented and pinned by tests/test_round8.cpp; the input is in the corpus.
  • Seventh review: the idle shortcut of ServiceInterrupts() is not taken while an input limit check is pending (a ClearLatchedEvents() between a failed and the next call no longer drops it); the ServiceInterrupts() header says "no source masked"; the VBUS operating range follows Table 5 (UVLO rising 3.8 to 4.2 V); the unseen-reset behaviour of the status and interrupt paths is documented; the fuzz target asserts the input limit invariant after ReadChargerStatus() and ServiceInterrupts() with its own VBUS_SNS decode, and no limit write in a call that returns DeviceReset; its invariants ignore a line that was stuck during the call.
  • Sixth review: the input limit restore of ReadChargerStatus() and ServiceInterrupts() looks for a chip reset first (DeviceReset, nothing written); a failed check in ServiceInterrupts() stays pending until it passes; the (void) casts of results in the tests are static_cast<void>; the remaining comparisons under && and || in the examples and compile-fail tests are parenthesised; ambiguity C60 (Table 163 default 400 mA against OTP 1500 mA).
  • Fifth review: the input limit check after a cable removal also runs from ReadChargerStatus() and ServiceInterrupts() (on a charger interrupt), so it does not need watchdog.enable, and from ServiceChargerWatchdog() before the BATFET_CNFG checks; while VBUSIN is invalid the limit field is held in reset and the restore waits for a valid supply (no VerifyFailed on every call); a configured limit is kept with LinearOnly too; a DEVICE_ID line stuck at 0x00 is a Bus error and does not take the switch-off away; tests for a limit inside Limits that is not the configured one, for a refused charge start with an unacknowledged protection off and for single reads; the error-name test covers every enumerator.
  • Fourth review: with the charger in Charge, ServiceChargerWatchdog() reads the input limit a cable removal reverts and writes the configured one again, or switches the charger off with OutOfLimits; SelfCheck() reports input_limit_ok; a refused charge start leaves the charger off; DEVICE_ID is read twice before another device is assumed; the fuzz target injects cable removals; comparisons under && and || are parenthesised; the examples show the DeviceReset recovery.
  • Third review: a cable removal (VBUS_LIN_ILIM reverts, Table 163) is not reported as a chip reset (VbusRemoved() in the model); a switch-off is sent even after a re-start that failed on a bus error; the register transfer buffers are members of the driver object; the shutdown gates read twice; argument errors send nothing; a rail the driver switched off is remembered as off; Configure() looks for a reset before the charger block; the fuzz target checks that a reset of the chip is seen by SelfCheck().
  • Safety fixes from the independent review, each with a model mode and a test: a read that is written back must be stable (two reads agree, no all-ones) and every typed write is compared over the whole byte; ServiceChargerWatchdog() refuses when BATFET_CNFG reads with the overcurrent protection off; status bits read by any call are kept in a latch and reported by ServiceInterrupts() (GetLatchedEvents(), ClearLatchedEvents()); a rail is enabled only at a held voltage inside its Limits window; the DVS declaration is checked against Table 85 (Acknowledgements::dvs_on_unlisted_part for other parts); an unsafe charger found at start-up can be switched off (SetChargerMode(Off), CheckChargerAgainstLimits() work once DEVICE_ID has passed); CancelShutdownRequest() clears GOTO_SHIP and GOTO_CORE_OFF, and the CORE_OFF documentation now says it is armed, not immediate; LoadSwitchConfirm{.input_mv} carries the LDO's input voltage and is checked against Limits; a chip reset in the session is detected (a changed register reads at its start-up value again; PWRUP_I is not a reset flag, C59) and reported as Error::DeviceReset; switching off is never refused for a reset, a failed start or an unreadable register; DISBATFET found at start-up is reported.
  • PowerConfig writes its unguarded fields only with their own flags (write_standby, write_debounce, write_regscp, write_low_sys_warn); BatfetConfig::oc_without_vbus defaults to true and false needs allow_batfet_oc_off; standby or sleep without run is InvalidArgument.
  • The coverage matrix (1090 rows) is in the repository as docs/coverage-matrix.csv with a generated page and a check (tools/coverage_matrix.py, CTest pf1550_coverage_matrix). Ambiguities C57 and C58 added.
  • The fuzz target also checks that no write newly switches off a protection (BATFET_OC, the fast-charge timer, the thermistor, DISBATFET), that guarded power fields change only with their acknowledgements, that a rail enabled by a write held a voltage inside its window, and it runs with read corruption.