Temperature messages for error buffer

common/power_ctl.c

@@ -283,7 +283,7 @@ disable_ps(void)
   // to finish their activity. For Rev2 of the CM (when the I2C
   // pullups are from management power) this delay can be reduced or
   // removed.
-  vTaskDelay(pdMS_TO_TICKS(200));
+  vTaskDelay(pdMS_TO_TICKS(500));
 
   // disable in reverse order
   for (int prio = num_priorities; prio > 0;  --prio) {

common/utils.c

@@ -165,7 +165,7 @@ void errbuffer_reset(errbuf_handle_t ebuf){
 	ebuf->head = ebuf->minaddr;
 	ebuf->counter=0;
 	ebuf->last=0;
-	errbuffer_put(ebuf, RESET_BUFFER,0);
+	errbuffer_put(ebuf, EBUF_RESET_BUFFER,0);
 	return;
 }
 
@@ -174,17 +174,17 @@ void errbuffer_put(errbuf_handle_t ebuf, uint16_t errcode, uint16_t errdata){
 	// If duplicated error code...
 	if(errcode == ebuf->last){
 
-		// if counter is not a multiple of 4, don't write new entry
-		if(oldcount%4!=0){ ebuf->counter=ebuf->counter+1; }
+		// if counter is not a multiple of COUNTER_UPDATE, don't write new entry
+		if(oldcount%COUNTER_UPDATE!=0){ ebuf->counter=ebuf->counter+1; }
 
 		// if counter has already reached max value, increment head
-		if(oldcount%16==0){	//Change this to use COUNTER_OFFSET
+		if(oldcount%(1<<COUNTER_OFFSET)==0){	//Change this to use COUNTER_OFFSET
 			ebuf->counter=0;
 			ebuf->head = increase_head(ebuf);
 			write_eeprom(0,increase_head(ebuf)); }
 
-		// if counter is multiple of 4, write entry and increment counter
-		if(oldcount%4==0){
+		// if counter is multiple of COUNTER_UPDATE, write entry and increment counter
+		if(oldcount%COUNTER_UPDATE==0){
 			ebuf->counter=ebuf->counter+1;
 			write_eeprom(errbuffer_entry(errcode,errdata),decrease_head(ebuf)); }
 	}
@@ -198,8 +198,8 @@ void errbuffer_put(errbuf_handle_t ebuf, uint16_t errcode, uint16_t errdata){
 	return;
 }
 
-void errbuffer_get(errbuf_handle_t ebuf, uint32_t (*arrptr)[EBUF_NGET]){
-	int i=0,j=0,max=EBUF_NGET;
+void errbuffer_get(errbuf_handle_t ebuf, uint32_t num, uint32_t (*arrptr)[num]){
+	int i=0,j=0,max=num;
 	while(i<max){i++; ebuf->head = decrease_head(ebuf);}
 	while(j<max){
 		(*arrptr)[j] = read_eeprom_single(ebuf->head);

common/utils.h

@@ -12,12 +12,6 @@
 #include <stdbool.h>
 
 
-// write to and read from eeprom
-void write_eeprom(uint32_t data, uint32_t addr);
-uint32_t read_eeprom_single(uint32_t addr);
-uint64_t read_eeprom_multi(uint32_t addr);
-
-
 // write, read or toggle GPIO pin by name
 void write_gpio_pin(int pin, uint8_t value);
 uint8_t read_gpio_pin(int pin);
@@ -26,25 +20,38 @@ uint8_t toggle_gpio_pin(int pin);
 
 void setupActiveLowPins(void);
 
+
+// EEPROM
+#define SN_ADDR 64
+#define FF_ADDR (SN_ADDR+4)
+
+void write_eeprom(uint32_t data, uint32_t addr);
+uint32_t read_eeprom_single(uint32_t addr);
+uint64_t read_eeprom_multi(uint32_t addr);
+
 // EEPROM buffer
 #define EBUF_MINBLK 2
 #define EBUF_MAXBLK 5
-#define EBUF_NGET 5			// Number of entries returned with errbuffer_get
 
-#define ERRDATA_OFFSET 8	// Number of bits reserved for error data
-#define ERRCODE_OFFSET 4	// Number of bits reserved for error codes
+#define ERRDATA_OFFSET 7	// Number of bits reserved for error data
+#define ERRCODE_OFFSET 5	// Number of bits reserved for error codes
 #define COUNTER_OFFSET (16-ERRDATA_OFFSET-ERRCODE_OFFSET)	// Number of bits reserved for message counter
 
-#define ERRDATA_MASK 255
-#define ERRCODE_MASK 3840
-#define COUNTER_MASK 61440
+#define ERRDATA_MASK ((1<<(ERRDATA_OFFSET))-1)
+#define ERRCODE_MASK ((1<<(ERRDATA_OFFSET+ERRCODE_OFFSET))-1-ERRDATA_MASK)
+#define COUNTER_MASK ((1<<(ERRDATA_OFFSET+ERRCODE_OFFSET+COUNTER_OFFSET))-1-ERRDATA_MASK-ERRCODE_MASK)
+
+#define COUNTER_UPDATE 4	//Number of repeated entries that initiates a hardware counter update (re-write entry)
 
 // error codes
-#define RESTART 1
-#define RESET_BUFFER 2
-#define TEMP_HIGH 3
-#define TEMP_NORMAL 4
-#define PWR_OFF_TEMP 5 // add other pwr_off codes for different reasons
+#define EBUF_RESTART 1
+#define EBUF_RESET_BUFFER 2
+#define EBUF_POWER_OFF 3
+#define EBUF_POWER_OFF_TEMP 4
+#define EBUF_POWER_ON 5
+
+#define EBUF_TEMP_HIGH(status) ((1<<(ERRCODE_OFFSET-1))|status)
+#define EBUF_TEMP_NORMAL EBUF_TEMP_HIGH(0)
 
 typedef struct error_buffer_t error_buffer_t;
 typedef error_buffer_t* errbuf_handle_t;
@@ -55,7 +62,7 @@ void errbuffer_init(errbuf_handle_t ebuf, uint8_t minblk, uint8_t maxblk);
 void errbuffer_reset(errbuf_handle_t ebuf);
 
 void errbuffer_put(errbuf_handle_t ebuf, uint16_t errcode, uint16_t errdata);
-void errbuffer_get(errbuf_handle_t ebuf, uint32_t (*arrptr)[EBUF_NGET]);
+void errbuffer_get(errbuf_handle_t ebuf, uint32_t num, uint32_t (*arrptr)[num]);
 
 uint32_t errbuffer_capacity(errbuf_handle_t ebuf);
 uint32_t errbuffer_minaddr(errbuf_handle_t ebuf);

projects/cm_mcu/AlarmTask.c

@@ -24,28 +24,45 @@ enum temp_state {TEMP_UNKNOWN, TEMP_GOOD, TEMP_BAD};
 enum alarm_state {ALM_UNKNOWN, ALM_GOOD, ALM_BAD};
 
 
-
 // Status of the alarm task
 static uint32_t status = 0x0;
+uint32_t oldstatus;
 
 // read-only, so no need to use queue
 uint32_t getAlarmStatus()
 {
   return status;
 }
 
-#define INITIAL_ALARM_TEMP 65.0 // in Celsius duh
-static float alarm_temp = INITIAL_ALARM_TEMP;
+#define INITIAL_ALARM_TEMP_FF 50.0 // in Celsius duh
+#define INITIAL_ALARM_TEMP_DCDC 75.0
+#define INITIAL_ALARM_TEMP_TM4C 75.0
+#define INITIAL_ALARM_TEMP_FPGA 75.0
+static float alarm_temp_ff = INITIAL_ALARM_TEMP_FF;
+static float alarm_temp_dcdc = INITIAL_ALARM_TEMP_DCDC;
+static float alarm_temp_tm4c = INITIAL_ALARM_TEMP_TM4C;
+static float alarm_temp_fpga = INITIAL_ALARM_TEMP_FPGA;
 
-float getAlarmTemperature()
+float getAlarmTemperature(enum device device_name)
 {
-  return alarm_temp;
+	switch(device_name){
+		case TM4C: return alarm_temp_tm4c;
+		case DCDC: return alarm_temp_dcdc;
+		case FPGA: return alarm_temp_fpga;
+		case FF: return alarm_temp_ff;
+		default: return 0;
+	}
 }
 
-void setAlarmTemperature(const float newtemp)
+void setAlarmTemperature(enum device device_name,const float newtemp)
 {
-  alarm_temp = newtemp;
-  return;
+	switch(device_name){
+		case TM4C: alarm_temp_tm4c = newtemp; return;
+		case DCDC: alarm_temp_dcdc = newtemp; return;
+		case FPGA:   alarm_temp_fpga = newtemp; return;
+		case FF: alarm_temp_ff = newtemp; return;
+		default: return;
+	}
 }
 
 
@@ -54,8 +71,10 @@ void AlarmTask(void *parameters)
   // initialize to the current tick time
   TickType_t xLastWakeTime = xTaskGetTickCount();
   uint32_t message; // this must be in a semi-permanent scope
+  uint16_t errbuf_data,errbuf_olddata=0;
   enum temp_state current_temp_state = TEMP_UNKNOWN;
-  float buffer_maxtemp = INITIAL_ALARM_TEMP;
+  float temp_over_ff, temp_over_fpga, temp_over_tm4c, temp_over_dcdc, temp_over_max, worst_temp;
+  // todo: should be able to do this w just max_temp_over and worst_temp, just to clean it up
 
   vTaskDelayUntil( &xLastWakeTime, pdMS_TO_TICKS( 2500 ) );
 
@@ -81,11 +100,17 @@ void AlarmTask(void *parameters)
 
     // microcontroller
     float tm4c_temp = getADCvalue(ADC_INFO_TEMP_ENTRY);
-    if ( tm4c_temp > alarm_temp ) status |= ALM_STAT_TM4C_OVERTEMP;
-    if ((tm4c_temp-buffer_maxtemp)/5>0) buffer_maxtemp=tm4c_temp;
+    if ( tm4c_temp > alarm_temp_tm4c ) status |= ALM_STAT_TM4C_OVERTEMP;
+    temp_over_tm4c = alarm_temp_tm4c-tm4c_temp;
+    temp_over_max = temp_over_tm4c;
+    worst_temp=tm4c_temp;
     // FPGA
     float max_fpga = MAX(fpga_args.pm_values[0], fpga_args.pm_values[1]);
-    if ( max_fpga > alarm_temp) status |= ALM_STAT_FPGA_OVERTEMP;
+    if ( max_fpga > alarm_temp_fpga) status |= ALM_STAT_FPGA_OVERTEMP;
+    temp_over_fpga = alarm_temp_fpga-max_fpga;
+    if (temp_over_fpga>temp_over_max){
+    	temp_over_max = temp_over_fpga;
+    	worst_temp=max_fpga;}
 
     // DCDC. The first command is READ_TEMPERATURE_1.
     // I am assuming it stays that way!!!!!!!!
@@ -98,7 +123,11 @@ void AlarmTask(void *parameters)
           max_dcdc_temp = thistemp;
       }
     }
-    if ( max_dcdc_temp > alarm_temp ) status |= ALM_STAT_DCDC_OVERTEMP;
+    if ( max_dcdc_temp > alarm_temp_tm4c ) status |= ALM_STAT_DCDC_OVERTEMP;
+    temp_over_dcdc = alarm_temp_dcdc-max_dcdc_temp;
+    if (temp_over_dcdc>temp_over_max){
+    	temp_over_max = temp_over_dcdc;
+    	worst_temp=max_dcdc_temp;}
     // Fireflies. These are reported as ints but we are asked
     // to report a float.
     int8_t imax_ff_temp = -99;
@@ -107,21 +136,28 @@ void AlarmTask(void *parameters)
       if ( v > imax_ff_temp )
         imax_ff_temp = v;
     }
-    if ( (float)imax_ff_temp > alarm_temp ) status |= ALM_STAT_FIREFLY_OVERTEMP;
+    if ( (float)imax_ff_temp > alarm_temp_ff ) status |= ALM_STAT_FIREFLY_OVERTEMP;
+    temp_over_ff = alarm_temp_ff-imax_ff_temp;
+    if (temp_over_ff>temp_over_max){
+    	temp_over_max=temp_over_ff;
+    	worst_temp=imax_ff_temp;}
 
     if ( status && current_temp_state != TEMP_BAD ) {
     	// If temp goes from good to bad, turn on alarm, send error message to buffer
-    	// data field takes the status bitmask plus temp divided by 4 (rounded)
-      uint16_t errbuf_data=((uint16_t)status<<4)|((uint16_t)(buffer_maxtemp+2)>>2);
-      errbuffer_put(ebuf, TEMP_HIGH,errbuf_data);
+      errbuf_data=(0x0FFFFFFFU)&(uint8_t)worst_temp;
+      if ((errbuf_data!=errbuf_olddata)||(status!=oldstatus)){
+    	  // only send message when status or temp have changed, to avoid filling up buffer
+    	  errbuffer_put(ebuf, EBUF_TEMP_HIGH(status),errbuf_data);
+      	  errbuf_olddata=errbuf_data;
+      	  oldstatus=status;}
       message = TEMP_ALARM;
       xQueueSendToFront(xPwrQueue, &message, pdMS_TO_TICKS(100));
       current_temp_state = TEMP_BAD;
     }
     else if ( !status && current_temp_state == TEMP_BAD ) {
     	// If temp goes from bad to good, turn off alarm, send message to buffer
-      errbuffer_put(ebuf, TEMP_NORMAL, 0);
-      buffer_maxtemp = INITIAL_ALARM_TEMP;
+      errbuf_data=(0x0FFFFFFFU&(uint8_t)worst_temp);
+      errbuffer_put(ebuf, EBUF_TEMP_NORMAL, errbuf_data);
       message = TEMP_ALARM_CLEAR;
       xQueueSendToFront(xPwrQueue, &message, pdMS_TO_TICKS(100));
       current_temp_state = TEMP_GOOD;