Added support for some NVML sensors.

GPU temp sensor and clock reasons.
2025-02-02 21:42:30 -08:00
parent 4cb9ef42a6
commit 4c5321bc4e
6 changed files with 185 additions and 22 deletions
@@ -0,0 +1,95 @@
+#include <stdio.h>
+
+#include "nvidia-sensors.h"
+
+int init_nvml()
+{
+	nvmlReturn_t result;
+	result = nvmlInit_v2();
+	if (result != NVML_SUCCESS) {
+		printf("Could not init NVML: %s\n", nvmlErrorString(result));
+		return 0;
+	}
+}
+
+void print_nvml_temp(int compact, struct card_info *card)
+{
+	if (compact)
+		printf(" %3d", get_nvml_temp(card));
+	else
+		printf("GPU1: %+d°C\n", get_nvml_temp(card));
+}
+
+void print_nvml_clock_reason(int compact, struct card_info *card)
+{
+	unsigned long long reasons = get_nvml_clock_reasons(card);
+	int single_reason = 1;
+	
+	if (compact)
+		printf(" CLK ");
+	else
+		printf("Clock reasons: ");
+	
+	for (int i = 0; i < (sizeof(clock_reason_names) / sizeof(struct clock_reason)); i++) {
+		if (reasons & clock_reason_names[i].mask) {
+			if (!single_reason) {
+				if (compact)
+					printf(",");
+				else
+					printf(", ");
+			}	
+			single_reason = 0;
+			
+			if (compact)
+				printf("%s", clock_reason_names[i].short_name);
+			else
+				printf("%s", clock_reason_names[i].long_name);
+		}
+	}
+	
+	if (single_reason)
+		printf("None");
+	
+	if (!compact)
+		printf(" (0x%llx)\n", reasons);
+}
+
+unsigned int get_nvml_temp(struct card_info *card)
+{
+	nvmlReturn_t result;
+	nvmlDevice_t nvml_device;
+	result = nvmlDeviceGetHandleByPciBusId_v2(card->pci_id, &nvml_device);
+	if (result != NVML_SUCCESS) {
+		printf("Failed to get device handle for card at %s: %s\n", card->pci_id, nvmlErrorString(result));
+		return 0;
+	}
+	
+	unsigned int temp;
+	result = nvmlDeviceGetTemperature(nvml_device, NVML_TEMPERATURE_GPU, &temp); 
+	if (result != NVML_SUCCESS) {
+		printf("Failed to get temperature for card at %s: %s\n", card->pci_id, nvmlErrorString(result));
+		return 0;
+	}
+	return temp;
+}
+
+unsigned long long get_nvml_clock_reasons(struct card_info *card)
+{
+	nvmlReturn_t result;
+	nvmlDevice_t nvml_device;
+	result = nvmlDeviceGetHandleByPciBusId_v2(card->pci_id, &nvml_device);
+	if (result != NVML_SUCCESS) {
+		printf("Failed to get device handle for card at %s: %s\n", card->pci_id, nvmlErrorString(result));
+		return 0;
+	}
+	
+	unsigned long long reasons;
+	result = nvmlDeviceGetCurrentClocksEventReasons(nvml_device, &reasons) ; 
+	if (result != NVML_SUCCESS) {
+		printf("Failed to get clock reasons for card at %s: %s\n", card->pci_id, nvmlErrorString(result));
+		return 0;
+	}
+	
+	return reasons;
+}
+