Monitoring & data

A field guide to the subsystems that sense board state and store platform data. Each entry gives the role and the key files; the fan/GPU/telemetry actuation path is covered in the companion Fan, power & telemetry deep-dive.

Voltage & temperature monitor
Perf monitor
SoC power smoothing
GPIO & IOX
Timers & watchdog
Flash
FRU
Register datastore & register table
VPP
See also

Voltage & temperature monitor

Role: ADC-based monitoring of leak-detect rails, busbar temperature, MCU internal temperature, and power-good voltages. The subsystem exposes a single unified entry point, nv::volt_mon::init() (src/nv/volt_mon/volt_mon.h:61), which brings up only the modules a project enables (via LeakDetectSensorNum, BusBarTempSensorNum, etc.) and starts ADC scanning in a conflict-free order. The header is explicit that this is the only function main.cpp should call (src/nv/volt_mon/volt_mon.h:53).

Key files (src/nv/volt_mon/):

adc.h — the shared Adc class (src/nv/volt_mon/adc.h:89) with scanning vs. one-shot conversion, an RAII OneShotGuard lock, and per-instance error tracking.
busbar_temp.h — the BusbarTemp NTC driver (src/nv/volt_mon/busbar_temp.h:48), including threshold config (overridable at runtime via NSM SetThermalParameter) and a 2-bit virtual-GPIO fault state.
common.h — sensor structs (VoltMon, LeakDetectSensor, BusBarTempSensor, McuInternalTempSensor), the State/VrGpioState enums, and ADC↔voltage conversion helpers. The ADC reference is AdcVolVref = 3317 mV at 16-bit resolution (src/nv/volt_mon/common.h:29).
ntc_table.h, leak_detect.h, mcu_internal_temp.h, pgood_volt.h — the per-sensor families. MCU internal temp uses NXP’s VBE bandgap formula (src/nv/volt_mon/adc.h:75).
legacy/ and timer/ — two implementations: the legacy ISR-driven path and a timer-driven path (void init(TimerId, period), src/nv/volt_mon/volt_mon.h:75) that drains ADC FIFOs on a periodic timer.

Readings from this subsystem feed the telemetry cache and NSM thermal events.

Perf monitor

Role: Firmware performance and out-of-band (OOB) bus error monitoring. nv::perf_mon::Driver (src/nv/perf_mon/perf_mon.h:91) is a singleton (src/nv/perf_mon/perf_mon.cpp:52) that records per-task CPU execution time, packet latency across the USB↔I3C path, interface throughput counters, and firmware/RAM size usage derived from linker symbols.

CPU accounting is driven by task_switch_hook() (src/nv/perf_mon/perf_mon.cpp:31), a FreeRTOS trace hook that charges elapsed CTIMER ticks to the outgoing task on every context switch. OOB per-bus error counters (OobBus enum) are maintained in a separate file, perf_mon_oob_err.cpp — e.g. Driver::set_transaction_error() (src/nv/perf_mon/perf_mon_oob_err.cpp:23) — and surfaced to hosts through NSM.

Key files (src/nv/perf_mon/): perf_mon.h, perf_mon.cpp, perf_mon_oob_err.cpp.

SoC power smoothing

Role: Real-time SoC power smoothing (the “MPF” — Modular Power Filter — control chain) that reacts to the module’s state-of-charge every ADC conversion. nv::soc_pwr_smoothing::PowerSmoothing (src/nv/soc_pwr_smoothing/soc_pwr_smoothing.h:63) is the public façade over a shared PowerManager and PresetManager.

The hot path is PowerSmoothing::adc_isr() (src/nv/soc_pwr_smoothing/soc_pwr_smoothing.cpp:326) — placed in RAMX SRAM for speed — which applies any pending preset change and then runs one filter iteration via PowerManager::run_iteration() (src/nv/soc_pwr_smoothing/power_manager.h:114), called roughly every 100 µs. The filter is built from composable function blocks whose base class is FuncBlock (src/nv/soc_pwr_smoothing/mpf.h:27). Settings that must survive a power cycle (enable flags, active preset, max AC ramp rate) are persisted to the flash PDS.

Key files (src/nv/soc_pwr_smoothing/): soc_pwr_smoothing.{h,cpp}, power_manager.h, mpf.h, presets.{h,cpp}, the policy blocks (power_brake_policy.h, thermal_brake_policy.h, offset_policy.h), runtime_cfg.h, and the NSM handlers in nsm_pwr_smoothing_handlers.cpp.

GPIO & IOX

Role: Pin-level I/O and I/O-expander emulation.

GPIO (src/nv/gpio/): nv::gpio::Driver (src/nv/gpio/driver.h:24) wraps the SoC GPIO peripheral with pin init, interrupt setup, and read()/write() including virtual-physical GPIO reads (src/nv/gpio/driver.h:52). The telemetry cache uses this driver to sample the board’s GPIO status word.
IOX (src/nv/iox/): nv::iox::Iox (src/nv/iox/iox.h:41) is a PCA9555 I/O-expander emulation — it presents the standard PCA9555 register model over I2C while backing it with real SoC GPIO. It runs as its own RTOS task: Task::make() (src/nv/iox/task.cpp:39) with the work loop in Task::main() (src/nv/iox/task.cpp:79).

Timers & watchdog

Role: Timing primitives and task liveness enforcement.

CTIMER (src/nv/ctimer/): nv::ctimer::Driver (src/nv/ctimer/ctimer.h:27) is a thin wrapper over the SoC counter/timer, offering read_ticks(), ticks_to_us(), and delay_for_us() (src/nv/ctimer/ctimer.h:33). It is the tick source used by perf monitoring and the power-smoothing ISR timing checks.
Watchdog (src/nv/watchdog/): nv::watchdog::Runtime (src/nv/watchdog/runtime.h:25) extends the SoC WWDT driver with init()/update_timeout()/feed() (src/nv/watchdog/runtime.h:28) and a task-liveness scheme: start_task_status_query() (src/nv/watchdog/runtime.cpp:31) periodically polls monitored tasks and only feeds the hardware watchdog when they report alive (RetryThreshold = 2, src/nv/watchdog/runtime.h:37). boot.h and notify_interface.h cover boot-time watchdog state and the task-notify API.

Flash

Role: Non-volatile storage management — firmware slots, a persistent data-store (PDS), OTP/eFuse, and CFPA/CMPA secure regions. nv::flash::Flash (src/nv/flash/flash.h:28) is a request-based API whose calls (read, write, erase, get_data/set_data) are marshaled to a dedicated flash task (Task::make(), src/nv/flash/task.cpp:73) so callers never touch flash hardware directly.

Two features stand out:

Background copy for A/B firmware updates: background_copy_start() (src/nv/flash/flash.cpp:138) and background_copy_query() report progress without blocking.
Persistent data store (PDS): pds.{h,cpp} / npds.{h,cpp} back the key/value settings used by power smoothing and other modules; pds_pwr_smoothing_defaults.h seeds power-smoothing defaults.

Key files (src/nv/flash/): flash.{h,cpp}, task.{h,cpp}, background_copy.{h,cpp}, datastore.h, pds.{h,cpp}, npds.{h,cpp}.

FRU

Role: Read and parse FRU (Field Replaceable Unit) inventory EEPROMs over I2C. The flow, per the header, is “FRU → EEPROM → I2C → Parser → API” (src/nv/fru/fru.h:30). Two one-shot calls extract structured data into fixed-size, allocation-free buffers: get_chassis_info() (declaration src/nv/fru/fru.h:93, implementation src/nv/fru/fru.cpp:292) returns chassis type, serial, and custom fields; get_board_info() (src/nv/fru/fru.cpp:413) returns the board serial. A rich Status enum distinguishes I2C-layer failures from parse/checksum errors (src/nv/fru/fru.h:36).

Key files (src/nv/fru/): fru.h, fru.cpp.

Register datastore & register table

Role: A generic, host-facing register abstraction for telemetry and RAS.

reg_table (src/nv/reg_table/): nv::reg_table::Table (src/nv/reg_table/table.h:83) holds an array of Entry records — each with a telemetry entry_id, a downstream handle, RAS/event control bits, and an API-type route (src/nv/reg_table/table.h:47). Host access is dispatched through Hooks::on_register_table_access() (src/nv/reg_table/hooks.cpp:24). The whole module compiles out to common::Empty when UBS_FEATURES_reg_table is off (src/nv/reg_table/table.h:101).
reg_ds (src/nv/reg_ds/): the downstream resource providers a table entry can route to, implementing the DsInterface contract — reg_ds::Info for read-only table/metadata (src/nv/reg_ds/info.h:28) and reg_ds::Gpio for GPIO-backed entries with read/write (src/nv/reg_ds/gpio.h:32).

(Interpretation: reg_table is the host-visible index; reg_ds supplies the actual values behind each index. Together they let a host read heterogeneous resources through one uniform register interface.)

VPP

Role: Voltage/Power-Protection for E1.S NVMe drive bays — hot-swap control and presence/power-good handling, presented to a host as PCA9555 GPIO expanders. nv::vpp::VPP (src/nv/vpp/vpp.h:31) owns a PCA9555 emulator and a per-drive hot-swap state machine (E1sHotSwap, from hssm.h), driven by GPIO interrupts and ADC voltage-monitor readings (gpio_interrupt()/adc_interrupt(), src/nv/vpp/vpp.h:57). It runs as an RTOS task (nv::vpp::Task, src/nv/vpp/task.h:46) with GPIO and ADC interrupt callbacks that fan out to the right drive instance.

Key files (src/nv/vpp/): vpp.{h,cpp}, task.{h,cpp}, hssm.{h,cpp}, common.h.