#include "cat.h"
#include "cm2dm_msg.h"
#include "reg.h"
#include "telemetry.h"
#include "timer.h"
#include <stdbool.h>
#include <tenstorrent/post_code.h>
#include <tenstorrent/sys_init_defines.h>
#include <zephyr/kernel.h>
#include <zephyr/sys/util.h>
#include <zephyr/drivers/gpio.h>
#include <zephyr/drivers/misc/bh_fwtable.h>
#include <zephyr/init.h>
#include <zephyr/drivers/sensor.h>
#include <zephyr/drivers/sensor/tenstorrent/pvt_tt_bh.h>
#include <zephyr/logging/log.h>
#include <zephyr/logging/log_ctrl.h>

Include dependency graph for cat.c:

Data Structures
struct	RESET_UNIT_CATMON_THERM_TRIP_CNTL_reg_t
union	RESET_UNIT_CATMON_THERM_TRIP_CNTL_reg_u

Macros
#define	RESET_UNIT_CATMON_THERM_TRIP_STATUS_REG_ADDR 0x80030164
#define	RESET_UNIT_CATMON_THERM_TRIP_CNTL_REG_ADDR 0x80030168
#define	RESET_UNIT_CATMON_THERM_TRIP_CNTL_REG_DEFAULT 0x00000318
#define	CAT_EARLY_TRIP_TEMP 100
#define	DEFAULT_CALIBRATION (CAT_EARLY_TRIP_TEMP - T_J_SHUTDOWN)
#define	TRIM_CODE_BITS 6
#define	GDDR_THERM_TRIP_DMC_NOTIFY_MS 50

Functions
	LOG_MODULE_REGISTER (cat)
	SENSOR_DT_READ_IODEV (cat_ts_avg_iodev, DT_NODELABEL(pvt), {SENSOR_CHAN_PVT_TT_BH_TS_AVG, 0})
	RTIO_DEFINE (cat_ts_avg_ctx, 1, 1)
static uint8_t	TempToTrimCode (float temp)
static float	TrimCodeToTemp (int32_t trim_code)
static void	WaitCATUpdate (void)
static void	EnableCAT (uint8_t trim_code, bool shutdown_on_trip)
static int	CATEarlyInit (void)
	SYS_INIT_APP (CATEarlyInit)
static float	CalibrateCAT (void)
static int	CATInit (void)
	SYS_INIT_APP (CATInit)
int	EvaluateGddrThermTrip (int64_t now, int max_temp, int trip_temp, int critical_temp, int64_t duration_ms)
	Evaluate the GDDR thermal-trip state machine for one sample against the supplied thresholds.
static int	MonitorGddrThermTrip (int64_t now, int max_temp)
static void	TriggerThermTrip (void)
static void	gddr_therm_trip_trigger_handler (struct k_work *work)
static	K_WORK_DELAYABLE_DEFINE (gddr_therm_trip_trigger_work, gddr_therm_trip_trigger_handler)
uint8_t	CatSetGddrThermTripEnabled (bool enabled)
	Enable or disable the GDDR thermal-trip action at runtime.
static void	gddr_therm_trip_work_handler (struct k_work *work)
static	K_WORK_DEFINE (gddr_therm_trip_worker, gddr_therm_trip_work_handler)
static void	gddr_therm_trip_timer_handler (struct k_timer *timer)
static	K_TIMER_DEFINE (gddr_therm_trip_timer, gddr_therm_trip_timer_handler, NULL)
int	StartGddrThermTripMonitor (void)
	SYS_INIT_APP (StartGddrThermTripMonitor)

Variables
static const int	gddr_therm_trip_interval = 100
static int	gddr_therm_trip_temp
static int	gddr_therm_trip_critical_temp
static int64_t	gddr_therm_trip_duration_ms
static const struct device *const	fwtable_dev = DEVICE_DT_GET(DT_NODELABEL(fwtable))
static const struct device *const	pvt = DEVICE_DT_GET(DT_NODELABEL(pvt))
static uint8_t	cat_ts_avg_buf [sizeof(struct pvt_tt_bh_rtio_data)]
static const struct device *	gpio1 = DEVICE_DT_GET_OR_NULL(DT_NODELABEL(gpio1))

Macro Definition Documentation

◆ CAT_EARLY_TRIP_TEMP

#define CAT_EARLY_TRIP_TEMP 100

◆ DEFAULT_CALIBRATION

#define DEFAULT_CALIBRATION (CAT_EARLY_TRIP_TEMP - T_J_SHUTDOWN)

◆ GDDR_THERM_TRIP_DMC_NOTIFY_MS

#define GDDR_THERM_TRIP_DMC_NOTIFY_MS 50

◆ RESET_UNIT_CATMON_THERM_TRIP_CNTL_REG_ADDR

#define RESET_UNIT_CATMON_THERM_TRIP_CNTL_REG_ADDR 0x80030168

◆ RESET_UNIT_CATMON_THERM_TRIP_CNTL_REG_DEFAULT

#define RESET_UNIT_CATMON_THERM_TRIP_CNTL_REG_DEFAULT 0x00000318

◆ RESET_UNIT_CATMON_THERM_TRIP_STATUS_REG_ADDR

#define RESET_UNIT_CATMON_THERM_TRIP_STATUS_REG_ADDR 0x80030164

◆ TRIM_CODE_BITS

#define TRIM_CODE_BITS 6

Function Documentation

◆ CalibrateCAT()

float CalibrateCAT ( void )

static

◆ CATEarlyInit()

int CATEarlyInit ( void )

static

◆ CATInit()

int CATInit ( void )

static

◆ CatSetGddrThermTripEnabled()

uint8_t CatSetGddrThermTripEnabled ( bool enabled )

Enable or disable the GDDR thermal-trip action at runtime.

Parameters

enabled false to disable, true to enable.

Returns: 0 on success.

◆ EnableCAT()

void EnableCAT	(	uint8_t	trim_code,
		bool	shutdown_on_trip )

static

◆ EvaluateGddrThermTrip()

int EvaluateGddrThermTrip	(	int64_t	now,
		int	max_temp,
		int	trip_temp,
		int	critical_temp,
		int64_t	duration_ms )

Evaluate the GDDR thermal-trip state machine for one sample against the supplied thresholds.

Production code uses the internal MonitorGddrThermTrip wrapper, which supplies the thresholds seeded from the firmware table at init.

Parameters

now	Current uptime in milliseconds.
max_temp	Highest observed GDDR temperature in degrees Celsius.
trip_temp	Sustained over-temp threshold in degrees Celsius.
critical_temp	Instantaneous (no-dwell) trip threshold in degrees Celsius.
duration_ms	Sustained dwell time before tripping, in milliseconds.

Returns: The tripping temperature (non-zero) when a trip fires, otherwise 0.

◆ gddr_therm_trip_timer_handler()

void gddr_therm_trip_timer_handler ( struct k_timer * timer )

static

◆ gddr_therm_trip_trigger_handler()

void gddr_therm_trip_trigger_handler ( struct k_work * work )

static

◆ gddr_therm_trip_work_handler()

void gddr_therm_trip_work_handler ( struct k_work * work )

static

◆ K_TIMER_DEFINE()

K_TIMER_DEFINE	(	gddr_therm_trip_timer	,
		gddr_therm_trip_timer_handler	,
		NULL	)

static

◆ K_WORK_DEFINE()

K_WORK_DEFINE	(	gddr_therm_trip_worker	,
		gddr_therm_trip_work_handler	)

static

◆ K_WORK_DELAYABLE_DEFINE()

K_WORK_DELAYABLE_DEFINE	(	gddr_therm_trip_trigger_work	,
		gddr_therm_trip_trigger_handler	)

static

◆ LOG_MODULE_REGISTER()

LOG_MODULE_REGISTER ( cat )

◆ MonitorGddrThermTrip()

int MonitorGddrThermTrip	(	int64_t	now,
		int	max_temp )

static

◆ RTIO_DEFINE()

RTIO_DEFINE	(	cat_ts_avg_ctx	,
		1	,
		1	)

◆ SENSOR_DT_READ_IODEV()

SENSOR_DT_READ_IODEV	(	cat_ts_avg_iodev	,
		DT_NODELABEL(pvt)	,
		{SENSOR_CHAN_PVT_TT_BH_TS_AVG, 0}	)

◆ StartGddrThermTripMonitor()

int StartGddrThermTripMonitor ( void )

◆ SYS_INIT_APP() [1/3]

SYS_INIT_APP ( CATEarlyInit )

◆ SYS_INIT_APP() [2/3]

SYS_INIT_APP ( CATInit )

◆ SYS_INIT_APP() [3/3]

SYS_INIT_APP ( StartGddrThermTripMonitor )

◆ TempToTrimCode()

uint8_t TempToTrimCode ( float temp )

static

◆ TriggerThermTrip()

void TriggerThermTrip ( void )

static

◆ TrimCodeToTemp()

float TrimCodeToTemp ( int32_t trim_code )

static

◆ WaitCATUpdate()

void WaitCATUpdate ( void )

static

Variable Documentation

◆ cat_ts_avg_buf

uint8_t cat_ts_avg_buf[sizeof(struct pvt_tt_bh_rtio_data)]

static

◆ fwtable_dev

const struct device* const fwtable_dev = DEVICE_DT_GET(DT_NODELABEL(fwtable))

static

◆ gddr_therm_trip_critical_temp

int gddr_therm_trip_critical_temp

static

◆ gddr_therm_trip_duration_ms

int64_t gddr_therm_trip_duration_ms

static

◆ gddr_therm_trip_interval

const int gddr_therm_trip_interval = 100

static

◆ gddr_therm_trip_temp

int gddr_therm_trip_temp

static

◆ gpio1

const struct device* gpio1 = DEVICE_DT_GET_OR_NULL(DT_NODELABEL(gpio1))

static

◆ pvt

const struct device* const pvt = DEVICE_DT_GET(DT_NODELABEL(pvt))

static

Data Structures

Macros

Functions

Variables

Macro Definition Documentation

◆ CAT_EARLY_TRIP_TEMP

◆ DEFAULT_CALIBRATION

◆ GDDR_THERM_TRIP_DMC_NOTIFY_MS

◆ RESET_UNIT_CATMON_THERM_TRIP_CNTL_REG_ADDR

◆ RESET_UNIT_CATMON_THERM_TRIP_CNTL_REG_DEFAULT

◆ RESET_UNIT_CATMON_THERM_TRIP_STATUS_REG_ADDR

◆ TRIM_CODE_BITS

Function Documentation

◆ CalibrateCAT()

◆ CATEarlyInit()

◆ CATInit()

◆ CatSetGddrThermTripEnabled()

◆ EnableCAT()

◆ EvaluateGddrThermTrip()

◆ gddr_therm_trip_timer_handler()

◆ gddr_therm_trip_trigger_handler()

◆ gddr_therm_trip_work_handler()

◆ K_TIMER_DEFINE()

◆ K_WORK_DEFINE()

◆ K_WORK_DELAYABLE_DEFINE()

◆ LOG_MODULE_REGISTER()

◆ MonitorGddrThermTrip()

◆ RTIO_DEFINE()

◆ SENSOR_DT_READ_IODEV()

◆ StartGddrThermTripMonitor()

◆ SYS_INIT_APP() [1/3]

◆ SYS_INIT_APP() [2/3]

◆ SYS_INIT_APP() [3/3]

◆ TempToTrimCode()

◆ TriggerThermTrip()

◆ TrimCodeToTemp()

◆ WaitCATUpdate()

Variable Documentation

◆ cat_ts_avg_buf

◆ fwtable_dev

◆ gddr_therm_trip_critical_temp

◆ gddr_therm_trip_duration_ms

◆ gddr_therm_trip_interval

◆ gddr_therm_trip_temp

◆ gpio1

◆ pvt