aalc: add sensor level 3 and event log (#2056)

Summary: - modify fan table - block auto tune command util bic is ready - disable auto tune will not set duty to 0 - add more error log event - add sensor COOLING_CAPACITY - add sensor BPB_RACK_LEVEL_3 - extend COOLING_CAPACITY modbus command to 2 register - add modbus shell command for test - modify the threshold of y_filter to be dynamic - fix threshold struct member last_status to the correct size Pull Request resolved: #2056 Test Plan: - Build code: PASS Reviewed By: waffle2k Differential Revision: D65333687 fbshipit-source-id: 38af4196585393c7ac438fef4fb5cd4130b05115
facebook · Nov 15, 2024 · ddf9fc9 · ddf9fc9
1 parent 33ebec3
commit ddf9fc9
Show file tree

Hide file tree

Showing 16 changed files with 292 additions and 147 deletions.
diff --git a/meta-facebook/aalc-rpu/src/platform/plat_def.h b/meta-facebook/aalc-rpu/src/platform/plat_def.h
@@ -22,5 +22,6 @@
 #define ADC_CALIBRATION 1
 #define MORE_THAN_ONE_ADM1272
 #define ENABLE_PLAT_DEF_SENSOR
+#define BIC_UPDATE_MAX_OFFSET 0x60000
 
 #endif
diff --git a/meta-facebook/aalc-rpu/src/platform/plat_fsc.c b/meta-facebook/aalc-rpu/src/platform/plat_fsc.c
@@ -22,35 +22,6 @@ static uint8_t fsc_poll_flag = 1;
 extern zone_cfg zone_table[];
 extern uint32_t zone_table_size;
 
-static int get_flow_rate_setpoint(void)
-{
-	const int worst_case = 200; // TODO: TBD
-	float flow_rate_val = 0.0;
-	float tout_val = 0.0;
-	float tin_val = 0.0;
-
-	uint8_t flow_rate_status = get_sensor_reading_to_real_val(
-		SENSOR_NUM_BPB_RPU_COOLANT_FLOW_RATE_LPM, &flow_rate_val);
-	uint8_t tout_status =
-		get_sensor_reading_to_real_val(SENSOR_NUM_BPB_HEX_WATER_INLET_TEMP_C, &tout_val);
-	uint8_t tin_status =
-		get_sensor_reading_to_real_val(SENSOR_NUM_BPB_RPU_COOLANT_INLET_TEMP_C, &tin_val);
-
-	if (flow_rate_status != SENSOR_READ_4BYTE_ACUR_SUCCESS)
-		return worst_case;
-
-	if (tout_status != SENSOR_READ_4BYTE_ACUR_SUCCESS)
-		return worst_case;
-
-	if (tin_status != SENSOR_READ_4BYTE_ACUR_SUCCESS)
-		return worst_case;
-
-	FSC_PRINTF("flow_rate_val %f, tout_val %f, tin_val %f\n", flow_rate_val, tout_val, tin_val);
-	FSC_PRINTF("1.00815 * flow_rate_val * (tout_val - tin_val) = %f\n",
-		   1.00815 * flow_rate_val * (tout_val - tin_val));
-	return (int)(1.00815 * flow_rate_val * (tout_val - tin_val));
-}
-
 uint8_t fsc_debug_set(uint8_t enable)
 {
 	enable_debug_print = enable;
@@ -159,19 +130,6 @@ static uint8_t calculatePID(zone_cfg *zone_p, uint8_t *duty)
 
 		int16_t temp = (int16_t)tmp;
 
-		tmp = 0.0;
-		if (p->setpoint_type == SETPOINT_AIR_INTEL_AVG_C) {
-			if (get_sensor_reading_to_real_val(SENSOR_NUM_SB_HEX_AIR_INLET_AVG_TEMP_C,
-							   &tmp) !=
-			    SENSOR_READ_4BYTE_ACUR_SUCCESS) {
-				tmp = 40.0; // worst case
-			}
-			tmp = CLAMP(tmp, -20, 100);
-			p->setpoint = (int16_t)tmp + 10;
-		} else if (p->setpoint_type == SETPOINT_FLOW_RATE_LPM) {
-			p->setpoint = get_flow_rate_setpoint();
-		}
-
 		FSC_PRINTF("\t\t----- sensor_num %x, temp = %d, p->setpoint %d\n", p->sensor_num,
 			   temp, p->setpoint);
 

diff --git a/meta-facebook/aalc-rpu/src/platform/plat_fsc.h b/meta-facebook/aalc-rpu/src/platform/plat_fsc.h
@@ -46,11 +46,6 @@ enum FSC_TYPE {
 	FSC_TYPE_DEFAULT,
 };
 
-enum SETPOINT_TYPE {
-	SETPOINT_AIR_INTEL_AVG_C = 0,
-	SETPOINT_FLOW_RATE_LPM,
-};
-
 /* stepwise */
 typedef struct {
 	uint8_t temp;
@@ -71,7 +66,7 @@ typedef struct {
 typedef struct {
 	uint8_t sensor_num;
 	int setpoint;
-	uint8_t setpoint_type;
+	/*uint8_t setpoint_type;*/
 	float kp;
 	float ki;
 	float kd;

diff --git a/meta-facebook/aalc-rpu/src/platform/plat_fsc_table.c b/meta-facebook/aalc-rpu/src/platform/plat_fsc_table.c
@@ -7,7 +7,7 @@
 pid_cfg hex_fan_pid_table[] = {
 	{
 		.sensor_num = SENSOR_NUM_BPB_RPU_COOLANT_OUTLET_TEMP_C,
-		.setpoint_type = SETPOINT_AIR_INTEL_AVG_C,
+		.setpoint = 40,
 		.kp = -7,
 		.ki = -0.1,
 		.kd = -3,
@@ -63,6 +63,7 @@ stepwise_cfg hex_fan_stepwise_table[] = {
 	},
 };
 
+/*
 pid_cfg pump_pid_table[] = {
 	{
 		.sensor_num = SENSOR_NUM_BPB_RPU_COOLANT_OUTLET_TEMP_C,
@@ -120,6 +121,18 @@ stepwise_cfg pump_stepwise_table[] = {
 			{45, 35},
 		},
 	},
+};*/
+
+stepwise_cfg pump_stepwise_table[] = {
+	{
+		.sensor_num = SENSOR_NUM_COOLING_CAPACITY,
+		.step = {
+			{30, 45},
+			{45, 55},
+			{60, 65},
+			{61, 100},
+		},
+	},
 };
 
 stepwise_cfg rpu_fan_stepwise_table[] = {
@@ -163,8 +176,9 @@ zone_cfg zone_table[] = {
 		// zone 2 - pump
 		.sw_tbl = pump_stepwise_table,
 		.sw_tbl_num = ARRAY_SIZE(pump_stepwise_table),
+		/*
 		.pid_tbl = pump_pid_table,
-		.pid_tbl_num = ARRAY_SIZE(pump_pid_table),
+		.pid_tbl_num = ARRAY_SIZE(pump_pid_table),*/
 		.interval = 1,
 		.set_duty = pwm_control,
 		.set_duty_arg = PWM_GROUP_E_PUMP,

diff --git a/meta-facebook/aalc-rpu/src/platform/plat_hwmon.c b/meta-facebook/aalc-rpu/src/platform/plat_hwmon.c
@@ -145,6 +145,9 @@ bool pump_setting_set_auto_tune_flag(pump_reset_struct *data, uint8_t bit_val)
 {
 	CHECK_NULL_ARG_WITH_RETURN(data, false);
 
+	if (!get_status_flag(STATUS_FLAG_SYSTEM_READY))
+		return false;
+
 	if (bit_val) {
 		for (uint8_t i = MANUAL_PWM_E_HEX_FAN; i <= MANUAL_PWM_E_RPU_FAN; i++) {
 			if (get_manual_pwm_flag(i))
@@ -415,7 +418,13 @@ uint8_t pwm_control(uint8_t group, uint8_t duty)
 	static int64_t auto_tune_time;
 	if (!get_status_flag(STATUS_FLAG_AUTO_TUNE)) {
 		auto_tune_time = k_uptime_get();
-		set_pwm_group(group, 0);
+		return 0;
+	}
+
+	// keep 0 duty when first enabling auto tune
+	static bool is_init = false;
+	if (!is_init) {
+		is_init = true;
 		return 0;
 	}
 

diff --git a/meta-facebook/aalc-rpu/src/platform/plat_log.c b/meta-facebook/aalc-rpu/src/platform/plat_log.c
@@ -52,6 +52,30 @@ const err_sensor_mapping sensor_err_codes[] = {
 	{ HIGH_PRESS_DETECTED, SENSOR_NUM_BPB_RPU_COOLANT_OUTLET_P_KPA },
 	{ FLOW_RATE_SENSOR_TRIGGERED, SENSOR_NUM_BPB_RPU_COOLANT_FLOW_RATE_LPM },
 	{ EMERGENCY_BUTTON_TRIGGERED, SENSOR_NUM_EMERGENCY_BUTTON_TRIGGERED },
+	{ LOG_ERR_HEX_EXTERNAL_Y_FILTER, SENSOR_NUM_HEX_EXTERNAL_Y_FILTER },
+	{ LOG_ERR_BPB_RPU_COOLANT_INLET_P_KPA, SENSOR_NUM_BPB_RPU_COOLANT_INLET_P_KPA },
+	{ LOG_ERR_BPB_RACK_PRESSURE_3_P_KPA, SENSOR_NUM_BPB_RACK_PRESSURE_3_P_KPA },
+	{ LOG_ERR_BPB_RACK_PRESSURE_4_P_KPA, SENSOR_NUM_BPB_RACK_PRESSURE_4_P_KPA },
+	{ LOG_ERR_SB_HEX_PRESSURE_1_P_KPA, SENSOR_NUM_SB_HEX_PRESSURE_1_P_KPA },
+	{ LOG_ERR_SB_HEX_PRESSURE_2_P_KPA, SENSOR_NUM_SB_HEX_PRESSURE_2_P_KPA },
+	{ LOG_ERR_BPB_RPU_COOLANT_INLET_TEMP_C, SENSOR_NUM_BPB_RPU_COOLANT_INLET_TEMP_C },
+	{ LOG_ERR_BPB_RPU_COOLANT_OUTLET_TEMP_C, SENSOR_NUM_BPB_RPU_COOLANT_OUTLET_TEMP_C },
+	{ LOG_ERR_BPB_HEX_WATER_INLET_TEMP_C, SENSOR_NUM_BPB_HEX_WATER_INLET_TEMP_C },
+	{ LOG_ERR_SB_HEX_AIR_INLET_AVG_TEMP_C, SENSOR_NUM_SB_HEX_AIR_INLET_AVG_TEMP_C },
+	{ LOG_ERR_FB_1_FAN_TACH_RPM, SENSOR_NUM_FB_1_FAN_TACH_RPM },
+	{ LOG_ERR_FB_2_FAN_TACH_RPM, SENSOR_NUM_FB_2_FAN_TACH_RPM },
+	{ LOG_ERR_FB_3_FAN_TACH_RPM, SENSOR_NUM_FB_3_FAN_TACH_RPM },
+	{ LOG_ERR_FB_4_FAN_TACH_RPM, SENSOR_NUM_FB_4_FAN_TACH_RPM },
+	{ LOG_ERR_FB_5_FAN_TACH_RPM, SENSOR_NUM_FB_5_FAN_TACH_RPM },
+	{ LOG_ERR_FB_6_FAN_TACH_RPM, SENSOR_NUM_FB_6_FAN_TACH_RPM },
+	{ LOG_ERR_FB_7_FAN_TACH_RPM, SENSOR_NUM_FB_7_FAN_TACH_RPM },
+	{ LOG_ERR_FB_8_FAN_TACH_RPM, SENSOR_NUM_FB_8_FAN_TACH_RPM },
+	{ LOG_ERR_FB_9_FAN_TACH_RPM, SENSOR_NUM_FB_9_FAN_TACH_RPM },
+	{ LOG_ERR_FB_10_FAN_TACH_RPM, SENSOR_NUM_FB_10_FAN_TACH_RPM },
+	{ LOG_ERR_FB_11_FAN_TACH_RPM, SENSOR_NUM_FB_11_FAN_TACH_RPM },
+	{ LOG_ERR_FB_12_FAN_TACH_RPM, SENSOR_NUM_FB_12_FAN_TACH_RPM },
+	{ LOG_ERR_FB_13_FAN_TACH_RPM, SENSOR_NUM_FB_13_FAN_TACH_RPM },
+	{ LOG_ERR_FB_14_FAN_TACH_RPM, SENSOR_NUM_FB_14_FAN_TACH_RPM },
 };
 
 const err_sensor_mapping sensor_normal_codes[] = {

diff --git a/meta-facebook/aalc-rpu/src/platform/plat_log.h b/meta-facebook/aalc-rpu/src/platform/plat_log.h
@@ -60,6 +60,30 @@ enum LOG_ERROR_CODE {
 	HIGH_PRESS_DETECTED = 0xA0,
 	FLOW_RATE_SENSOR_TRIGGERED = 0xA1,
 	EMERGENCY_BUTTON_TRIGGERED = 0xA2,
+	LOG_ERR_HEX_EXTERNAL_Y_FILTER = 0xA3,
+	LOG_ERR_BPB_RPU_COOLANT_INLET_P_KPA = 0xA4,
+	LOG_ERR_BPB_RACK_PRESSURE_3_P_KPA = 0xA5,
+	LOG_ERR_BPB_RACK_PRESSURE_4_P_KPA = 0xA6,
+	LOG_ERR_SB_HEX_PRESSURE_1_P_KPA = 0xA7,
+	LOG_ERR_SB_HEX_PRESSURE_2_P_KPA = 0xA8,
+	LOG_ERR_BPB_RPU_COOLANT_INLET_TEMP_C = 0xA9,
+	LOG_ERR_BPB_RPU_COOLANT_OUTLET_TEMP_C = 0xAA,
+	LOG_ERR_BPB_HEX_WATER_INLET_TEMP_C = 0xAB,
+	LOG_ERR_SB_HEX_AIR_INLET_AVG_TEMP_C = 0xAC,
+	LOG_ERR_FB_1_FAN_TACH_RPM = 0xAD,
+	LOG_ERR_FB_2_FAN_TACH_RPM = 0xAE,
+	LOG_ERR_FB_3_FAN_TACH_RPM = 0xAF,
+	LOG_ERR_FB_4_FAN_TACH_RPM = 0xB0,
+	LOG_ERR_FB_5_FAN_TACH_RPM = 0xB1,
+	LOG_ERR_FB_6_FAN_TACH_RPM = 0xB2,
+	LOG_ERR_FB_7_FAN_TACH_RPM = 0xB3,
+	LOG_ERR_FB_8_FAN_TACH_RPM = 0xB4,
+	LOG_ERR_FB_9_FAN_TACH_RPM = 0xB5,
+	LOG_ERR_FB_10_FAN_TACH_RPM = 0xB6,
+	LOG_ERR_FB_11_FAN_TACH_RPM = 0xB7,
+	LOG_ERR_FB_12_FAN_TACH_RPM = 0xB8,
+	LOG_ERR_FB_13_FAN_TACH_RPM = 0xB9,
+	LOG_ERR_FB_14_FAN_TACH_RPM = 0xBA,
 };
 
 typedef struct _err_sensor_mapping {

diff --git a/meta-facebook/aalc-rpu/src/platform/plat_modbus.c b/meta-facebook/aalc-rpu/src/platform/plat_modbus.c
@@ -97,7 +97,18 @@ uint8_t modbus_get_senser_reading(modbus_command_mapping *cmd)
 
 	if (status == SENSOR_READ_4BYTE_ACUR_SUCCESS) {
 		float r = pow_of_10(cmd->arg2);
-		int16_t byte_val = val / cmd->arg1 / r; // scale
+		int16_t byte_val = 0;
+		// capacity scale kW to W
+		if (cmd->addr == MODBUS_AALC_COOLING_CAPACITY_W_ADDR) {
+			int32_t scaled_val = (int32_t)(val * 1000 / cmd->arg1 / r);
+			byte_val = scaled_val & 0xFFFF;
+		} else if (cmd->addr == MODBUS_AALC_COOLING_CAPACITY_W_EXT_ADDR) {
+			int32_t scaled_val = (int32_t)(val * 1000 / cmd->arg1 / r);
+			byte_val = (scaled_val >> 16) & 0xFFFF;
+		} else {
+			byte_val = val / cmd->arg1 / r; // scale
+		}
+
 		memcpy(cmd->data, &byte_val, sizeof(uint16_t) * cmd->cmd_size);
 		return MODBUS_EXC_NONE;
 	}
@@ -458,39 +469,6 @@ uint8_t modbus_get_sticky_sensor_status(modbus_command_mapping *cmd)
 	return MODBUS_EXC_NONE;
 }
 
-uint8_t modbus_get_aalc_cooling_capacity(modbus_command_mapping *cmd)
-{
-	CHECK_NULL_ARG_WITH_RETURN(cmd, MODBUS_EXC_ILLEGAL_DATA_VAL);
-	/*
-	*	AALC_Cooling_Capacity_W  = 67.21*Flow rate*(Tout-Tin)
-	*	Flow Rate = Flow sensor reading (LPM)
-	*	Tout = Rack Coolant temperature sensor reading (°C)
-	*	Tin = RPU Coolant outlet temperature sensor reading(°C)
-	*/
-
-	float flow_rate_val, tout_val, tin_val;
-
-	uint8_t flow_rate_status = get_sensor_reading_to_real_val(
-		SENSOR_NUM_BPB_RPU_COOLANT_FLOW_RATE_LPM, &flow_rate_val);
-	uint8_t tout_status =
-		get_sensor_reading_to_real_val(SENSOR_NUM_BPB_HEX_WATER_INLET_TEMP_C, &tout_val);
-	uint8_t tin_status =
-		get_sensor_reading_to_real_val(SENSOR_NUM_BPB_RPU_COOLANT_INLET_TEMP_C, &tin_val);
-
-	if (flow_rate_status != SENSOR_READ_4BYTE_ACUR_SUCCESS)
-		return MODBUS_EXC_ILLEGAL_DATA_VAL;
-
-	if (tout_status != SENSOR_READ_4BYTE_ACUR_SUCCESS)
-		return MODBUS_EXC_ILLEGAL_DATA_VAL;
-
-	if (tin_status != SENSOR_READ_4BYTE_ACUR_SUCCESS)
-		return MODBUS_EXC_ILLEGAL_DATA_VAL;
-
-	cmd->data[0] = (uint16_t)(67.21 * flow_rate_val * (tout_val - tin_val));
-
-	return MODBUS_EXC_NONE;
-}
-
 uint8_t modbus_get_manual_flag(modbus_command_mapping *cmd)
 {
 	CHECK_NULL_ARG_WITH_RETURN(cmd, MODBUS_EXC_ILLEGAL_DATA_VAL);
@@ -713,8 +691,10 @@ modbus_command_mapping modbus_command_table[] = {
 	  SENSOR_NUM_MB_FAN1_TACH_RPM, 1, 0, 1 },
 	{ MODBUS_MB_FAN2_TACH_RPM_ADDR, NULL, modbus_get_senser_reading,
 	  SENSOR_NUM_MB_FAN2_TACH_RPM, 1, 0, 1 },
-	{ MODBUS_AALC_COOLING_CAPACITY_W_ADDR, NULL, modbus_get_aalc_cooling_capacity, 0, 1, -1,
-	  1 },
+	{ MODBUS_AALC_COOLING_CAPACITY_W_ADDR, NULL, modbus_get_senser_reading,
+	  SENSOR_NUM_COOLING_CAPACITY, 1, -1, 1 },
+	{ MODBUS_AALC_COOLING_CAPACITY_W_EXT_ADDR, NULL, modbus_get_senser_reading,
+	  SENSOR_NUM_COOLING_CAPACITY, 1, -1, 1 },
 	{ MODBUS_RPU_PUMP1_STATUS_ADDR, NULL, modbus_get_aalc_sensor_status, RPU_PUMP1_STATUS, 0, 0,
 	  1 },
 	{ MODBUS_RPU_PUMP2_STATUS_ADDR, NULL, modbus_get_aalc_sensor_status, RPU_PUMP2_STATUS, 0, 0,
@@ -803,6 +783,8 @@ modbus_command_mapping modbus_command_table[] = {
 	  SENSOR_NUM_BPB_RACK_LEVEL_1, 1, 0, 1 },
 	{ MODBUS_BPB_RACK_LEVEL_2_ADDR, NULL, modbus_get_senser_reading,
 	  SENSOR_NUM_BPB_RACK_LEVEL_2, 1, 0, 1 },
+	{ MODBUS_BPB_RACK_LEVEL_3_ADDR, NULL, modbus_get_senser_reading,
+	  SENSOR_NUM_BPB_RACK_LEVEL_3, 1, 0, 1 },
 	{ MODBUS_MB_HUM_PCT_RH_ADDR, NULL, modbus_get_senser_reading, SENSOR_NUM_MB_HUM_PCT_RH, 1,
 	  0, 1 },
 	{ MODBUS_PDB_HUM_PCT_RH_ADDR, NULL, modbus_get_senser_reading, SENSOR_NUM_PDB_HUM_PCT_RH, 1,
@@ -1242,7 +1224,7 @@ modbus_command_mapping modbus_command_table[] = {
 	{ MODBUS_HEAT_EXCHANGER_FAN_CONTROL_BOX_FBPN_ADDR, NULL, NULL, 0, 0, 0, 4 },
 };
 
-static modbus_command_mapping *ptr_to_modbus_table(uint16_t addr)
+modbus_command_mapping *ptr_to_modbus_table(uint16_t addr)
 {
 	for (uint16_t i = 0; i < ARRAY_SIZE(modbus_command_table); i++) {
 		if ((addr >= modbus_command_table[i].addr) &&

diff --git a/meta-facebook/aalc-rpu/src/platform/plat_modbus.h b/meta-facebook/aalc-rpu/src/platform/plat_modbus.h
@@ -313,6 +313,8 @@ enum read_fru_data_type {
 #define MODBUS_V_3_3_AUX_ADDR 0xA030
 #define MODBUS_V_1_2_AUX_ADDR 0xA031
 #define MODBUS_V_5_USB_ADDR 0xA032
+#define MODBUS_BPB_RACK_LEVEL_3_ADDR 0xA033
+#define MODBUS_AALC_COOLING_CAPACITY_W_EXT_ADDR 0xA034
 #define MODBUS_MODBUS_ADDR_PATH_WITH_WEDGE400_ADDR 0xA401
 #define MODBUS_MANUAL_CONTROL_RPU_FAN_ON_OFF_ADDR 0xA480
 #define MODBUS_ERROR_LOG_COUNT_ADDR 0x1A28
@@ -388,4 +390,6 @@ enum read_fru_data_type {
 
 #define MODBUS_test_for_write_uptime_ADDR 0x1A0C
 
+modbus_command_mapping *ptr_to_modbus_table(uint16_t addr);
+
 #endif
diff --git a/meta-facebook/aalc-rpu/src/platform/plat_sdr_table.c b/meta-facebook/aalc-rpu/src/platform/plat_sdr_table.c
@@ -10613,6 +10613,16 @@ SDR_Full_sensor plat_sdr_table[] = {
 		.sensor_num = SENSOR_NUM_SB_HEX_AIR_INLET_AVG_TEMP_C,
 		.ID_str = "SB_HEX_AIR_INLET_AVG_TEMP_C",
 	},
+	// SENSOR_NUM_COOLING_CAPACITY
+	{
+		.sensor_num = SENSOR_NUM_COOLING_CAPACITY,
+		.ID_str = "COOLING_CAPACITY",
+	},
+	// SENSOR_NUM_BPB_RACK_LEVEL_3
+	{
+		.sensor_num = SENSOR_NUM_BPB_RACK_LEVEL_3,
+		.ID_str = "BPB_RACK_LEVEL_3",
+	},
 };
 
 const int SDR_TABLE_SIZE = ARRAY_SIZE(plat_sdr_table);