Added fan sensors

Makefile

@@ -1,7 +1,7 @@
 .PHONY : clean debug
 
-OBJS = evga-icx.o evga-card.o icx3.o
+OBJS = evga-icx.o evga-card.o icx3.o board-sensors.o zen3-rapl.o
-LDLIBS = -li2c
+LDLIBS = -li2c -lm
 CFLAGS = -MD
 
 ifdef USE_NVML
@@ -18,7 +18,7 @@ endif
 
 evga-icx : $(OBJS)
 
-debug : CFLAGS += -g -O0
+debug : CFLAGS += -g -Og
 debug : evga-icx
 
 clean :

README.md

@@ -55,6 +55,8 @@ Available options:
 --watch N    : Keep printing output every N seconds
 --overwrite  : Overwrite previously displayed info with --watch and --compact instead of continuously logging new lines
 --color      : Print headers in color in --compact mode for better readability
+--no-reasons : Do not query NVML for clock reasons (can cause stuttering)
+--board      : Also print temperatures from the CPU, motherboard, and other sensors
 ```
 
 ### Examples:

board-sensors.c

@@ -0,0 +1,122 @@
+#include <dirent.h>
+#include <string.h>
+#include <stdlib.h>
+
+#include "board-sensors.h"
+
+int find_board_sensors(struct hwmon_avail_sensor *board_sensors, int max_sensors)
+{
+	const char *hwmon_path = "/sys/class/hwmon/";
+	char device_path[NAME_MAX];
+	char sensor_path[NAME_MAX];
+	
+	char driver_name[256];
+	
+	FILE *file;
+	DIR *dir;
+	struct dirent *ent;
+	
+	int num_sensors = 0;
+	
+	/* Start looking for hwmon devices in /sys/class/hwmon/ */
+	dir = opendir(hwmon_path);
+	
+	/* make sure we can open the device directory */
+	if(dir == NULL)
+		return 0;
+	
+	/* loop over all hwmon devices */
+	while((ent = readdir(dir)) != NULL)
+	{
+		/* Ignore any non-hwmon dirs */
+		if(strncmp(ent->d_name, "hwmon", 5) != 0) 
+			continue;
+		
+		strcpy(device_path, hwmon_path);
+		strcat(device_path, ent->d_name);
+		
+		/* Read in the name of the device */
+		strcpy(sensor_path, device_path);
+		strcat(sensor_path, "/name");
+		
+		file = fopen(sensor_path, "r");
+		if (file == NULL)
+			continue;
+		
+		if (fgets(driver_name, sizeof(driver_name), file) == NULL) {
+			fclose(file);
+			continue;
+		}
+		
+		fclose(file);
+
+		/* Driver names have a linebreak at the end so let's remove that for comparison*/
+		driver_name[strlen(driver_name) - 1] = '\0';
+		
+		/* Loop through all supported sensors and see if any are present in this device */
+		for (int i = 0; i < (sizeof(hwmon_sensor_info) / sizeof(struct hwmon_sensor)); i++) {
+			if (strcmp(driver_name, hwmon_sensor_info[i].driver_name) == 0) {
+				/* We matched the driver name, try to open the files */
+				strcpy(sensor_path, device_path);
+				strcat(sensor_path, "/");
+				strcat(sensor_path, hwmon_sensor_info[i].sensor_file_name);
+				strcat(sensor_path, "_input");
+				
+				file = fopen(sensor_path, "r");
+				if (file != NULL) {
+					fclose(file);
+					/* Good open of the sensor file */
+					board_sensors[num_sensors].file = calloc(NAME_MAX, sizeof(char));
+					strcpy(board_sensors[num_sensors].file, sensor_path);
+					board_sensors[num_sensors].sort_index = i;
+					board_sensors[num_sensors].sensor_info = &hwmon_sensor_info[i];
+					
+					/* Read in the sensor name */
+					board_sensors[num_sensors].sensor_name = calloc(MAX_SENSOR_NAME_LENGTH, sizeof(char));
+					strcpy(sensor_path, device_path);
+					strcat(sensor_path, "/");
+					strcat(sensor_path, hwmon_sensor_info[i].sensor_file_name);
+					strcat(sensor_path, "_label");
+					
+					file = fopen(sensor_path, "r");
+					if (file != NULL)
+						fgets(board_sensors[num_sensors].sensor_name, MAX_SENSOR_NAME_LENGTH, file);
+
+					/* Sensor name seems to always have a trailing newline we don't want */
+					size_t len_without_newline = strcspn(board_sensors[num_sensors].sensor_name, "\n");
+					board_sensors[num_sensors].sensor_name[len_without_newline] = '\0';
+					
+					if (num_sensors == max_sensors)
+						return num_sensors;
+					num_sensors++;
+				}
+			}
+		}
+		
+	}
+	
+	return num_sensors;
+}
+
+/* Returns 0 on a bad read or missing sensor, 1 on OK */
+int get_sensor_reading(struct hwmon_avail_sensor *sensor, float *reading) {
+	char buf[256] = {0};
+	long int raw;
+	FILE *file;
+	
+	file = fopen(sensor->file, "r");
+	if (file == NULL)
+		return 0;
+
+	fgets(buf, 256, file);
+	raw = strtol(buf, NULL, 10);
+	
+	fclose(file);
+	
+	*reading = (float)raw / sensor->sensor_info->divisor;
+
+	if (*reading == sensor->sensor_info->bad_value)
+		return 0;
+	
+	return 1;
+}

board-sensors.h

@@ -0,0 +1,60 @@
+#ifndef BOARD_SENSORS_H
+#define BOARD_SENSORS_H
+
+#include <stdio.h>
+
+#define MAX_SENSOR_NAME_LENGTH 256
+
+struct hwmon_sensor {
+	char *driver_name;				/* Contents of /sys/class/hwmon/hwmonX/name */
+	char *sensor_file_name;			/* Sysfs file to read */
+	char *name_prefix;				/* Prefix to attach to temp*_label for clarity */
+	char *short_name;				/* 'Category' name when using compact mode */
+	char *units;					/* Units string */
+	float divisor;					/* Divisor to convert temp* to units */
+	float bad_value;	/* Raw value that indicates a bad (missing) sensor TODO: verfiy most of these*/
+};
+
+/* Note the order here matters, it's the order these will be printed in */
+static struct hwmon_sensor hwmon_sensor_info[] = 
+{
+	{"zenpower",	"temp1",	"CPU ",			"CPU",	"°C",	1000.0,	-40.0	},	/* Tdie */
+	{"asusec",		"temp2",	"Motherboard ",	"CPU",	"°C",	1000.0,	-40.0	},	/* CPU */
+	{"zenpower",	"temp3",	"CPU ",			"CCX",	"°C",	1000.0,	-40.0	},	/* Tccd1 */
+	{"zenpower",	"temp4",	"CPU ",			"CCX",	"°C",	1000.0,	-40.0	},	/* Tccd2 */
+	{"zenpower",	"temp5",	"CPU ",			"CCX",	"°C",	1000.0,	-40.0	},	/* Tccd3 */
+	{"zenpower",	"temp6",	"CPU ",			"CCX",	"°C",	1000.0,	-40.0	},	/* Tccd4 */
+	{"zenpower",	"temp7",	"CPU ",			"CCX",	"°C",	1000.0,	-40.0	},	/* Tccd5 */
+	{"zenpower",	"temp8",	"CPU ",			"CCX",	"°C",	1000.0,	-40.0	},	/* Tccd6 */
+	{"zenpower",	"temp9",	"CPU ",			"CCX",	"°C",	1000.0,	-40.0	},	/* Tccd7 */
+	{"zenpower",	"temp10",	"CPU ",			"CCX",	"°C",	1000.0,	-40.0	},	/* Tccd8 */
+	{"zen-rapl",	"",			"CPU ",			"POW",	" W",	0.0,	0.0		},	/* Zen RAPL placeholder */
+	{"asusec",		"temp1",	"Motherboard ",	"CHIP",	"°C",	1000.0,	-40.0	},	/* Chipset */
+	{"asusec",		"temp5",	"Motherboard ",	"VRM",	"°C",	1000.0,	-40.0	},	/* VRM */
+	{"asusec",		"temp3",	"",				"MOBO",	"°C",	1000.0,	-40.0	},	/* Motherboard */
+	{"asusec",		"temp4",	"Motherboard ",	"SENS",	"°C",	1000.0,	-40.0	},	/* T_Sensor */
+	{"asusec",		"temp6",	"Motherboard ",	"H2O",	"°C",	1000.0,	-40.0	},	/* Water_In */
+	{"asusec",		"temp7",	"Motherboard ",	"H2O",	"°C",	1000.0,	-40.0	},	/* Water_Out */
+	{"nvme",		"temp1",	"NVMe ",		"NVME",	"°C",	1000.0,	-40.0	},	/* NVME Composite */
+	{"nct6798",		"fan2",		"CPU fan",		"CPU",	"%",	15.0,	0.0		},	/* cpu_fan */
+	{"asusec",		"fan1",		"",				"CPU",	"%",	15.0,	0.0		},	/* cpu_opt */
+	{"nct6798",		"fan1",		"Chassis 1",	"CHA",	"%",	12.0,	0.0		},	/* cha1? */
+	{"nct6798",		"fan3",		"Chassis 2",	"CHA",	"%",	12.0,	0.0		},	/* cha2? */
+	{"nct6798",		"fan4",		"Chassis 3",	"CHA",	"%",	13.0,	0.0		},	/* cha3 */
+	{"nct6798",		"fan5",		"H amp",		"CHA",	"%",	12.0,	0.0		},	/* h_amp? */
+	{"nct6798",		"fan6",		"AIO pump",		"CHA",	"%",	12.0,	0.0		},	/* aio_pump? */
+	{"nct6798",		"fan7",		"W pump",		"CHA",	"%",	12.0,	0.0		},	/* w_pump+? */
+	{"asusec",		"fan2",		"",				"CHIP",	"%",	35.0,	0.0		},	/* chipset */
+};
+
+struct hwmon_avail_sensor {
+	char *sensor_name;					/* Sensor name as read from the sysfs file */
+	int sort_index;						/* Sort index for order to display in */
+	char *file;							/* File to read from */
+	struct hwmon_sensor *sensor_info;	/* Associated sensor info struct */
+};
+
+int find_board_sensors(struct hwmon_avail_sensor *board_sensors, int max_sensors);
+int get_sensor_reading(struct hwmon_avail_sensor *sensor, float *reading);
+
+#endif

evga-icx.c

@@ -15,13 +15,20 @@
 
 #include "icx3.h"
 #include "evga-card.h"
+#include "board-sensors.h"
+#include "zen3-rapl.h"
 
 #define MAX_GPUS 16
+#define MAX_BOARD_SENSORS 256
 #define HEADER_COLOR_START "\x1b[36m"
 #define HEADER_COLOR_END "\x1b[39m"
 
-char *header_start = "";
+static char *header_start = "";
-char *header_end = "";
+static char *header_end = "";
+
+static int zen3_rapl_sensor = -1;	/* Board sensor number for the RAPL sensor */
+static int compact = 0;				/* Compact one-line per GPU display */
+static int no_reasons = 0;			/* Don't probe or display NVML clock reasons */
 
 static const char helpstring[] = "Available options:\n"
 								 "--i2c N      : Only probe I2C bus N instead of all (may help with stuttering or freezing when probing I2C devices)\n"
@@ -37,20 +44,24 @@ static const char helpstring[] = "Available options:\n"
 								 "--compact    : Print sensor reading in a compact one-line per card format\n"
 								 "--watch N    : Keep printing output every N seconds\n"
 								 "--overwrite  : Overwrite previously displayed info with --watch and --compact instead of continuously logging\n"
-								 "--color      : Print headers in color in --compact mode for better readability\n";
+								 "--color      : Print headers in color in --compact mode for better readability\n"
+								 "--no-reasons : Do not query NVML for clock reasons (can cause stuttering)\n"
+								 "--board      : Also print temperatures from the CPU, motherboard, and other sensors";
 								 
-void print_gpu_info(int gpu_num, struct card_info gpus[], int compact);
+void print_gpu_info(int gpu_num, struct card_info gpus[]);
+void print_board_info(struct hwmon_avail_sensor *board_sensors, int num_sensors);
 
 int main (int argc, char **argv)
 {
 	struct card_info gpus[MAX_GPUS];
-	int gpu_count;
+	struct hwmon_avail_sensor board_sensors[MAX_BOARD_SENSORS];
-	int print_info = 0;
+	int gpu_count, board_sensor_count;
-	int compact = 0;
+	int print_info = 0;			
-	int gpu_num = -1;    /* Card to control */
+	int gpu_num = -1;				/* Card to control */
-	int i2c_bus = -1;
+	int i2c_bus = -1;				/* Specific i2c bus to probe instead of all */
-	int overwrite = 0;
+	int overwrite = 0;				/* Overwrite printed console info in compact mode */		
-	unsigned int watch = 0;
+	unsigned int watch = 0;			/* Refresh display every this many seconds */
+	int print_board_sensors = 0;	/* Print CPU/motherbord/other sensors as well */
 	char *fan_speed[ICX3_MAX_FANS] = {NULL}; 
 	
 	/* Input parsing */
@@ -110,6 +121,10 @@ int main (int argc, char **argv)
 		} else if (strcmp(argv[i], "--color") == 0) {
 			header_start = HEADER_COLOR_START;
 			header_end = HEADER_COLOR_END;
+		} else if (strcmp(argv[i], "--no-reasons") == 0) {
+			no_reasons = 1;
+		} else if (strcmp(argv[i], "--board") == 0) {
+			print_board_sensors = 1;
 		} else {
 			printf(helpstring);
 			return 0;
@@ -129,8 +144,10 @@ int main (int argc, char **argv)
 	if (overwrite && !compact)
 		overwrite = 0;
 	
+	/* Scan for supported GPUs */
 	gpu_count = find_evga_gpu_i2cs(gpus, MAX_GPUS, i2c_bus);
 
+	/* Check for no GPUs found or other errors */
 	if (gpu_count == -1) {
 		printf("Error scanning I2C devices\n");
 		return -1;
@@ -138,12 +155,21 @@ int main (int argc, char **argv)
 		printf("No supported GPUs found.\nAre you root or do you have udev access to i2c devices?\nDo you need to run `modprobe i2c-dev`?\n");
 		return -1;
 	}
-	
+	/* Check for invalid GPUs */
 	if (gpu_num > gpu_count - 1) {
 		printf("Invalid GPU number specified (%d, max %d)\n", gpu_num, gpu_count - 1);
 		return -1;
 	}
 		
+	/* Scan for motherboard/CPU/other sensors */
+	if (print_board_sensors) {
+		board_sensor_count = find_board_sensors(board_sensors, MAX_BOARD_SENSORS);
+		if (init_rapl() && board_sensor_count < MAX_BOARD_SENSORS) {
+			board_sensors[board_sensor_count] = rapl_sensor;
+			zen3_rapl_sensor = board_sensor_count;
+			board_sensor_count++;
+		}
+	}
 	
 	/* execute fan commands */
 	if (gpu_num == -1) {
@@ -163,8 +189,10 @@ int main (int argc, char **argv)
 	/* NVML init */
 #ifdef USE_NVML
     init_nvml();
-	for (int i = 0; i < gpu_count; i++)
+	for (int i = 0; i < gpu_count; i++) {
+		gpus[i].nvml_device = malloc(sizeof(nvmlDevice_t));
 		get_nvml_handle(&gpus[i]);
+	}
 #endif
 
 	/* PCI init for VRAM/hotspot temps */
@@ -179,15 +207,18 @@ int main (int argc, char **argv)
 			if (overwrite)
 				printf("\x1b[K");	/* Clear current console line */
 			
+			if (print_board_sensors)
+				print_board_info(board_sensors, board_sensor_count);
+
 			if (gpu_num == -1) {
 				/* No GPU specified on command line, loop over all supported GPUs */
 				for (int i = 0; i < gpu_count; i++){
 					if (i > 0)
 						printf("\n");
-					print_gpu_info(i, &gpus[i], compact);
+					print_gpu_info(i, &gpus[i]);
 				}		
 			} else if (gpu_num <= gpu_count - 1) {
-				print_gpu_info(gpu_num, &gpus[gpu_num], compact);
+				print_gpu_info(gpu_num, &gpus[gpu_num]);
 			}
 
 			if (!overwrite)
@@ -195,8 +226,8 @@ int main (int argc, char **argv)
 			
 			if (overwrite && compact) {
 				printf("\x1b[1G");						/* Move cursor back to column 1 */
-				if (gpu_count > 1)
+				if (gpu_count > 1 || print_board_sensors)
-					printf("\x1b[%dA", gpu_count-1);	/* Move cursor back up to the top of gpu list */
+					printf("\x1b[%dA", gpu_count-1+print_board_sensors);	/* Move cursor back up to the top of gpu list */
 			}
 			
 			fflush(stdout);
@@ -209,7 +240,81 @@ int main (int argc, char **argv)
 #endif
 }
 
-void print_gpu_info(int gpu_num, struct card_info *gpu, int compact) {
+void print_board_info(struct hwmon_avail_sensor *board_sensors, int num_sensors)
+{
+	int printed_sensors = 0;
+	int current_sort_index = 0;
+	float current_reading = 0.0;
+	int good_reading = 0;
+	/* These allow us to 'summarize' units and categories by only printing them when they change */
+	char *last_short_name = NULL;
+	char *last_units = NULL;
+	
+	float rapl_power;
+	
+	for (int i = 0; i < (sizeof(hwmon_sensor_info) / sizeof(struct hwmon_sensor)); i++) {
+		
+		/* Inject our Zen RAPL power reading here */
+		if (strcmp(hwmon_sensor_info[i].driver_name, "zen-rapl") == 0) {
+			board_sensors[zen3_rapl_sensor].sort_index = current_sort_index;
+			board_sensors[zen3_rapl_sensor].sensor_info = &hwmon_sensor_info[i];
+		}
+		
+		for (int j=0; j < num_sensors; j++) {
+			struct hwmon_avail_sensor *current_sensor = &board_sensors[j];
+
+			/* Loop over all sensors, but only output those with the current sort index so they come out sort of sorted
+			Duplicates (e.g.) multiple NVMe will come out in whatever sort of order the directory listing happened to */
+			if (current_sensor->sort_index == current_sort_index) {
+				printed_sensors++;
+				
+				if (j == zen3_rapl_sensor) {
+					good_reading = 1;
+					current_reading = get_rapl_package_power();
+				} else {
+					good_reading = get_sensor_reading(current_sensor, &current_reading);
+				}
+				
+				if (!good_reading)
+					continue;
+				
+				if (compact) {
+					/* Print units if needed */
+					if (last_units != NULL && strcmp(current_sensor->sensor_info->units, last_units))
+						printf("%s", last_units);
+					
+					/* Print new section header if needed */
+					if (last_short_name == NULL || strcmp(current_sensor->sensor_info->short_name, last_short_name)) {
+						if (last_short_name != NULL) /* Spacer for all headings not the first one */
+							printf(" ");
+						printf("%s%s%s", header_start, current_sensor->sensor_info->short_name, header_end);
+					}
+					printf(" %3.0f", current_reading);
+					
+					last_short_name = current_sensor->sensor_info->short_name;
+					last_units = current_sensor->sensor_info->units;
+					
+				} else {
+					printf("%s%s: %+.1f%s\n", 
+						current_sensor->sensor_info->name_prefix,
+						current_sensor->sensor_name,
+						current_reading,
+						current_sensor->sensor_info->units);
+				}
+				
+			}
+		}
+		current_sort_index++;
+	}
+	
+	if (compact && last_units != NULL)
+		printf("%s", last_units);
+	
+	printf("\n");
+}
+
+void print_gpu_info(int gpu_num, struct card_info *gpu)
+{
 	if (compact) {
 		/* One line per GPU */
 		printf("%s#%d FAN%s", header_start, gpu_num, header_end);
@@ -241,8 +346,10 @@ void print_gpu_info(int gpu_num, struct card_info *gpu, int compact) {
 #ifdef USE_NVML
 		printf("%s MEM %s", header_start, header_end);
 		printf("%3d%%", get_nvml_mem_util(gpu));
-		printf("%s CLK %s", header_start, header_end);
+		if (!no_reasons) {
-		print_nvml_clock_reason(1, gpu);
+			printf("%s CLK %s", header_start, header_end);
+			print_nvml_clock_reason(1, gpu);
+		}
 #endif
 
 
@@ -262,8 +369,8 @@ void print_gpu_info(int gpu_num, struct card_info *gpu, int compact) {
 				printf("VRAM: +%.0f°C\n", get_vram_temp(gpu));			/* Print the VRAM temp before the rest of the memory sensors */
 #endif
 			printf("%s: %+.1f°C\n",
-			icx3_temp_sensor_names[i],
+				icx3_temp_sensor_names[i],
-			icx_temp_sensors[i]);
+				icx_temp_sensors[i]);
 		}
 		
 #ifdef USE_LIBPCI
@@ -272,11 +379,12 @@ void print_gpu_info(int gpu_num, struct card_info *gpu, int compact) {
 
 #ifdef USE_NVML
 		printf("Mem util: %d%%\n", get_nvml_mem_util(gpu));
-		printf("Clock reasons: ");
+		if (!no_reasons) {
-		print_nvml_clock_reason(0, gpu);
+			printf("Clock reasons: ");
+			print_nvml_clock_reason(0, gpu);
+		}
 		printf("\n");
 #endif
 	}
-	
 }
 	

zen3-rapl.c

@@ -0,0 +1,112 @@
+#include <stdio.h>
+#include <unistd.h>
+#include <cpuid.h>
+#include <math.h>
+#include <string.h>
+#include <fcntl.h>
+
+#include "zen3-rapl.h"
+
+#define AMD_STRING "AuthenticAMD"
+#define ZEN_FAMILY 0x17
+#define ZEN3_FAMILY 0x19
+
+#define MEASUREMENT_TIME 0.25
+
+static float energy_unit = 0;
+
+static int msr_file = -1;
+
+static float package_power = 0.0;
+
+static int check_zen()
+{
+    unsigned int eax = 0, ebx = 0, ecx = 0, edx = 0, ext_family;
+    char vendor[13];
+
+    __get_cpuid(0, &eax, &ebx, &ecx, &edx);
+
+    memcpy(vendor, &ebx, 4);
+    memcpy(vendor+4, &edx, 4);
+    memcpy(vendor+8, &ecx, 4);
+    vendor[12] = 0;
+
+    if (strcmp(vendor, AMD_STRING) != 0){
+        return 0;
+    }
+
+    __get_cpuid(1, &eax, &ebx, &ecx, &edx);
+
+    ext_family = ((eax >> 8) & 0xF) + ((eax >> 20) & 0xFF);
+    if (ext_family != ZEN_FAMILY && ext_family != ZEN3_FAMILY){
+        return 0;
+    }
+
+    return 1;
+}
+
+static int read_msr(int file, unsigned int index, unsigned long long *data)
+{
+    if (file < 0)
+        return 0;
+
+    return pread(file, data, sizeof *data, index) == sizeof *data;
+}
+
+static float get_energy_unit()
+{
+    unsigned long long data;
+    // AMD OSRR: page 139 - MSRC001_0299
+    if (!read_msr(msr_file, 0xC0010299, &data))
+        return 0.0;
+
+    return pow(1.0/2.0, (float)((data >> 8) & 0x1F));
+}
+
+static unsigned long get_package_energy()
+{
+    unsigned long long data;
+    // AMD OSRR: page 139 - MSRC001_029B
+    if (!read_msr(msr_file, 0xC001029B, &data))
+        return 0;
+
+    return data;
+}
+
+int init_rapl()
+{
+	/* Check for supported Zen CPU */
+	if (!check_zen())
+		return 0;
+	
+	/* Open MSR file */
+	char *msr_path = "/dev/cpu/0/msr";
+    msr_file = open(msr_path, O_RDONLY);
+	if (msr_file < 0)
+		return 0;
+	
+	/* Energy unit */
+	energy_unit = get_energy_unit();
+    if (energy_unit == 0)
+        return 0;
+	
+	return 1;
+}
+
+float get_rapl_package_power()
+{
+	unsigned long package_eng_b = 0;
+	unsigned long package_eng_a = 0;
+
+    package_eng_b = get_package_energy();
+
+    usleep(MEASUREMENT_TIME*1000000);
+
+    package_eng_a = get_package_energy();
+
+	/* Only update if we computed a good power reading, otherwise return the most recent measurement */
+    if (package_eng_a >= package_eng_b)
+        package_power = (package_eng_a - package_eng_b) * energy_unit / MEASUREMENT_TIME;
+	
+	return package_power;
+}

zen3-rapl.h

@@ -0,0 +1,6 @@
+#include "board-sensors.h"
+
+static struct hwmon_avail_sensor rapl_sensor = {"Package power", -1, NULL, NULL};
+
+int init_rapl();
+float get_rapl_package_power();