GPU: support Nvidia GPU temp detection

CHANGELOG.md

@@ -15,6 +15,7 @@ Features:
 * Support st version detection (Terminal, Linux)
 * Support st terminal font detection (TerminalFont, Linux)
 * Add option `--migrate-config <?target-file-path>`
+* Support Nvidia GPU temp detection via nvml (GPU)
 
 Bugfixes:
 * Better GPU memory and type detection (GPU, Windows)

CMakeLists.txt

@@ -412,6 +412,7 @@ if(LINUX)
         src/detection/sound/sound_linux.c
         src/detection/swap/swap_linux.c
         src/detection/temps/temps_linux.c
+        src/detection/temps/temps_nvidia.c
         src/detection/terminalfont/terminalfont_linux.c
         src/detection/terminalshell/terminalshell_linux.c
         src/detection/terminalsize/terminalsize_linux.c
@@ -518,6 +519,7 @@ elseif(BSD)
         src/detection/sound/sound_linux.c
         src/detection/swap/swap_bsd.c
         src/detection/temps/temps_bsd.c
+        src/detection/temps/temps_nvidia.c
         src/detection/terminalfont/terminalfont_linux.c
         src/detection/terminalshell/terminalshell_linux.c
         src/detection/terminalsize/terminalsize_linux.c
@@ -622,6 +624,7 @@ elseif(WIN32)
         src/detection/terminalshell/terminalshell_windows.c
         src/detection/terminalsize/terminalsize_windows.c
         src/detection/temps/temps_windows.cpp
+        src/detection/temps/temps_nvidia.c
         src/detection/theme/theme_nosupport.c
         src/detection/uptime/uptime_windows.c
         src/detection/users/users_windows.c

src/3rdparty/nvml/repo.json

@@ -0,0 +1,6 @@
+{
+    "home": "https://developer.nvidia.com/gpu-deployment-kit",
+    "license": "Embed in source",
+    "version": "v9",
+    "author": "NVIDIA Corporation"
+}

src/detection/gpu/gpu_linux.c

@@ -1,5 +1,6 @@
 #include "detection/gpu/gpu.h"
 #include "detection/vulkan/vulkan.h"
+#include "detection/temps/temps_nvidia.h"
 
 #ifdef FF_HAVE_LIBPCI
 #include "common/library.h"
@@ -35,7 +36,7 @@ typedef struct PCIData
 
 static void pciDetectVendorName(FFGPUResult* gpu, PCIData* pci, struct pci_dev* device)
 {
-    ffStrbufAppendS(&gpu->vendor, ffGetGPUVendorString(device->vendor_id));
+    ffStrbufSetStatic(&gpu->vendor, ffGetGPUVendorString(device->vendor_id));
 
     if(gpu->vendor.length > 0)
         return;
@@ -45,11 +46,11 @@ static void pciDetectVendorName(FFGPUResult* gpu, PCIData* pci, struct pci_dev*
     ffStrbufRecalculateLength(&gpu->vendor);
 
     if(ffStrbufContainS(&gpu->vendor, "AMD") || ffStrbufContainS(&gpu->vendor, "ATI"))
-        ffStrbufSetS(&gpu->vendor, FF_GPU_VENDOR_NAME_AMD);
+        ffStrbufSetStatic(&gpu->vendor, FF_GPU_VENDOR_NAME_AMD);
     else if(ffStrbufContainS(&gpu->vendor, "Intel"))
-        ffStrbufSetS(&gpu->vendor, FF_GPU_VENDOR_NAME_INTEL);
+        ffStrbufSetStatic(&gpu->vendor, FF_GPU_VENDOR_NAME_INTEL);
     else if(ffStrbufContainS(&gpu->vendor, "NVIDIA"))
-        ffStrbufSetS(&gpu->vendor, FF_GPU_VENDOR_NAME_NVIDIA);
+        ffStrbufSetStatic(&gpu->vendor, FF_GPU_VENDOR_NAME_NVIDIA);
 }
 
 static void drmDetectDeviceName(FFGPUResult* gpu, PCIData* pci, struct pci_dev* device)
@@ -240,6 +241,9 @@ static void pciHandleDevice(FF_MAYBE_UNUSED const FFGPUOptions* options, FFlist*
     if(options->temp)
         pciDetectTemp(gpu, device);
     #endif
+
+    if (gpu->temperature != gpu->temperature && gpu->vendor.chars == FF_GPU_VENDOR_NAME_NVIDIA)
+        ffDetectNvidiaGpuTemp(&gpu->temperature, device->device_id);
 }
 
 jmp_buf pciInitJmpBuf;

src/detection/gpu/gpu_windows.c

@@ -1,4 +1,5 @@
 #include "gpu.h"
+#include "detection/temps/temps_nvidia.h"
 #include "util/windows/unicode.h"
 #include "util/windows/registry.h"
 
@@ -53,11 +54,11 @@ const char* ffDetectGPUImpl(FF_MAYBE_UNUSED const FFGPUOptions* options, FFlist*
             ffRegReadStrbuf(hKey, L"ProviderName", &gpu->vendor, NULL);
 
             if(ffStrbufContainS(&gpu->vendor, "AMD") || ffStrbufContainS(&gpu->vendor, "ATI"))
-                ffStrbufSetS(&gpu->vendor, FF_GPU_VENDOR_NAME_AMD);
+                ffStrbufSetStatic(&gpu->vendor, FF_GPU_VENDOR_NAME_AMD);
             else if(ffStrbufContainS(&gpu->vendor, "Intel"))
-                ffStrbufSetS(&gpu->vendor, FF_GPU_VENDOR_NAME_INTEL);
+                ffStrbufSetStatic(&gpu->vendor, FF_GPU_VENDOR_NAME_INTEL);
             else if(ffStrbufContainS(&gpu->vendor, "NVIDIA"))
-                ffStrbufSetS(&gpu->vendor, FF_GPU_VENDOR_NAME_NVIDIA);
+                ffStrbufSetStatic(&gpu->vendor, FF_GPU_VENDOR_NAME_NVIDIA);
 
             wmemcpy(regDirectxKey + regDirectxKeyPrefixLength, displayDevice.DeviceKey + deviceKeyPrefixLength, strlen("00000000-0000-0000-0000-000000000000}"));
             FF_HKEY_AUTO_DESTROY hDirectxKey = NULL;
@@ -74,6 +75,13 @@ const char* ffDetectGPUImpl(FF_MAYBE_UNUSED const FFGPUOptions* options, FFlist*
                     gpu->dedicated.total = dedicatedVideoMemory + dedicatedSystemMemory;
                     gpu->shared.total = sharedSystemMemory;
                 }
+
+                if (options->temp && gpu->vendor.chars == FF_GPU_VENDOR_NAME_NVIDIA)
+                {
+                    uint32_t deviceId;
+                    if(ffRegReadUint(hDirectxKey, L"DeviceId", &deviceId, NULL))
+                        ffDetectNvidiaGpuTemp(&gpu->temperature, deviceId);
+                }
             }
             else if (!ffRegReadUint64(hKey, L"HardwareInformation.qwMemorySize", &gpu->dedicated.total, NULL))
             {

src/detection/temps/temps_nvidia.c

@@ -0,0 +1,73 @@
+#include "temps_nvidia.h"
+
+#include "3rdparty/nvml/nvml.h"
+#include "common/library.h"
+
+struct FFNvmlData {
+    FF_LIBRARY_SYMBOL(nvmlDeviceGetCount_v2)
+    FF_LIBRARY_SYMBOL(nvmlDeviceGetHandleByIndex_v2)
+    FF_LIBRARY_SYMBOL(nvmlDeviceGetPciInfo_v3)
+    FF_LIBRARY_SYMBOL(nvmlDeviceGetTemperature)
+
+    bool inited;
+} nvmlData;
+
+const char* ffDetectNvidiaGpuTemp(double* temp, uint32_t deviceId)
+{
+    if (!nvmlData.inited)
+    {
+        nvmlData.inited = true;
+        FF_LIBRARY_LOAD(libnvml, NULL, "dlopen nvml failed",
+#ifdef _WIN32
+            "nvml"
+#else
+            "libnvidia-ml"
+#endif
+        FF_LIBRARY_EXTENSION, -1
+        );
+        FF_LIBRARY_LOAD_SYMBOL_MESSAGE(libnvml, nvmlInit_v2)
+        FF_LIBRARY_LOAD_SYMBOL_MESSAGE(libnvml, nvmlShutdown)
+        FF_LIBRARY_LOAD_SYMBOL_VAR_MESSAGE(libnvml, nvmlData, nvmlDeviceGetCount_v2)
+        FF_LIBRARY_LOAD_SYMBOL_VAR_MESSAGE(libnvml, nvmlData, nvmlDeviceGetHandleByIndex_v2)
+        FF_LIBRARY_LOAD_SYMBOL_VAR_MESSAGE(libnvml, nvmlData, nvmlDeviceGetPciInfo_v3)
+        FF_LIBRARY_LOAD_SYMBOL_VAR_MESSAGE(libnvml, nvmlData, nvmlDeviceGetTemperature)
+
+        if (ffnvmlInit_v2() != NVML_SUCCESS)
+        {
+            nvmlData.ffnvmlDeviceGetTemperature = NULL;
+            return "nvmlInit_v2() failed";
+        }
+        atexit((void*) ffnvmlShutdown);
+        libnvml = NULL; // don't close nvml
+    }
+
+    if (nvmlData.ffnvmlDeviceGetTemperature == NULL)
+        return "loading nvml library failed";
+
+    uint32_t count;
+    if (nvmlData.ffnvmlDeviceGetCount_v2(&count) != NVML_SUCCESS)
+        return "nvmlDeviceGetCount_v2() failed";
+
+    for (uint32_t i = 0; i < count; i++)
+    {
+        nvmlDevice_t device;
+        if (nvmlData.ffnvmlDeviceGetHandleByIndex_v2(i, &device) != NVML_SUCCESS)
+            continue;
+
+        nvmlPciInfo_t pciInfo;
+        if (nvmlData.ffnvmlDeviceGetPciInfo_v3(device, &pciInfo) != NVML_SUCCESS)
+            continue;
+
+        if (pciInfo.pciDeviceId >> 16 != deviceId)
+            continue;
+
+        uint32_t value;
+        if (nvmlData.ffnvmlDeviceGetTemperature(device, NVML_TEMPERATURE_GPU, &value) != NVML_SUCCESS)
+            return "nvmlDeviceGetTemperature(device, NVML_TEMPERATURE_GPU, &value) failed";
+
+        *temp = value;
+        return NULL;
+    }
+
+    return "Device not found";
+}

src/detection/temps/temps_nvidia.h

@@ -0,0 +1,5 @@
+#pragma once
+
+#include <stdint.h>
+
+const char* ffDetectNvidiaGpuTemp(double* temp, uint32_t deviceId);