amdgpu: prefer hwmon over gpu_metrics for gpu temp

Co-authored-by: Martin Roukala <martin.roukala@mupuf.org>

src/amdgpu.cpp

@@ -194,7 +194,8 @@ void amdgpu_get_metrics(){
 	gpu_info.CoreClock = amdgpu_common_metrics.current_gfxclk_mhz;
 	gpu_info.MemClock = amdgpu_common_metrics.current_uclk_mhz;
 
-	gpu_info.temp = amdgpu_common_metrics.gpu_temp_c;
+	// Use hwmon instead, see gpu.cpp
+	// gpu_info.temp = amdgpu_common_metrics.gpu_temp_c;
 	gpu_info.apu_cpu_power = amdgpu_common_metrics.average_cpu_power_w;
 	gpu_info.apu_cpu_temp = amdgpu_common_metrics.apu_cpu_temp_c;
 

src/gpu.cpp

@@ -82,15 +82,6 @@ void getAmdGpuInfo(){
             gpu_info.load = value;
         }
 
-        if (amdgpu.temp) {
-            rewind(amdgpu.temp);
-            fflush(amdgpu.temp);
-            int value = 0;
-            if (fscanf(amdgpu.temp, "%d", &value) != 1)
-                value = 0;
-            gpu_info.temp = value / 1000;
-        }
-
         if (amdgpu.core_clock) {
             rewind(amdgpu.core_clock);
             fflush(amdgpu.core_clock);
@@ -134,4 +125,14 @@ void getAmdGpuInfo(){
             value = 0;
         gpu_info.memoryUsed = float(value) / (1024 * 1024 * 1024);
     }
+    // On some GPUs SMU can sometimes return the wrong temperature.
+    // As HWMON is way more visible than the SMU metrics, let's always trust it as it is the most likely to work 
+    if (amdgpu.temp){
+        rewind(amdgpu.temp);
+        fflush(amdgpu.temp);
+        int value = 0;
+        if (fscanf(amdgpu.temp, "%d", &value) != 1)
+            value = 0;
+        gpu_info.temp = value / 1000;
+    }
 }

src/overlay.cpp

@@ -679,6 +679,12 @@ void init_gpu_stats(uint32_t& vendorID, uint32_t reported_deviceID, overlay_para
             amdgpu.vram_total = fopen((path + "/mem_info_vram_total").c_str(), "r");
          if (!amdgpu.vram_used)
             amdgpu.vram_used = fopen((path + "/mem_info_vram_used").c_str(), "r");
+         
+         path += "/hwmon/";
+         auto dirs = ls(path.c_str(), "hwmon", LS_DIRS);
+         for (auto& dir : dirs)
+            if (!amdgpu.temp)
+               amdgpu.temp = fopen((path + dir + "/temp1_input").c_str(), "r");
 
          if (!metrics_path.empty())
             break;
@@ -691,15 +697,11 @@ void init_gpu_stats(uint32_t& vendorID, uint32_t reported_deviceID, overlay_para
 
          SPDLOG_DEBUG("using amdgpu path: {}", path);
 
-         path += "/hwmon/";
-         auto dirs = ls(path.c_str(), "hwmon", LS_DIRS);
          for (auto& dir : dirs) {
             if (!amdgpu.core_clock)
                amdgpu.core_clock = fopen((path + dir + "/freq1_input").c_str(), "r");
             if (!amdgpu.memory_clock)
                amdgpu.memory_clock = fopen((path + dir + "/freq2_input").c_str(), "r");
-            if (!amdgpu.temp)
-               amdgpu.temp = fopen((path + dir + "/temp1_input").c_str(), "r");
             if (!amdgpu.power_usage)
                amdgpu.power_usage = fopen((path + dir + "/power1_average").c_str(), "r");
          }