Merge branch 'master' of github.com:RRZE-HPC/likwid

src/includes/sysFeatures.h

@@ -39,5 +39,4 @@
 #define HWFEATURES_MAX_STR_LENGTH 100
 #define HWFEATURES_MIN_STRLEN(a, b) ((a) > (b) ? (b) : (a))
 
-
 #endif

src/includes/sysFeatures_intel_thermal.h

@@ -0,0 +1,8 @@
+#ifndef SYSFEATURES_INTEL_THERMAL_H
+#define SYSFEATURES_INTEL_THERMAL_H
+
+#include <sysFeatures_types.h>
+
+extern const _SysFeatureList likwid_sysft_intel_cpu_thermal_feature_list;
+
+#endif //SYSFEATURES_INTEL_THERMAL_H

src/sysFeatures_common.c

@@ -362,16 +362,46 @@ static int readmsr_socket(const LikwidDevice_t device, uint64_t reg, uint64_t *m
         if ((int)t->packageId == device->id.simple.id && t->inCpuSet)
         {
             err = HPMaddThread(t->apicId);
-            if (err < 0) continue;
+            if (err < 0)
+                continue;
             err = HPMread(t->apicId, MSR_DEV, reg, msrData);
-            if (err < 0) continue;
+            if (err < 0)
+                continue;
             return 0;
         }
     }
+
+    if (err < 0)
+        return err;
+    return -ENODEV;
+}
+
+static int readmsr_core(const LikwidDevice_t device, uint64_t reg, uint64_t *msrData)
+{
+    assert(device->type == DEVICE_TYPE_CORE);
+
+    int err = topology_init();
     if (err < 0)
-    {
         return err;
+
+    CpuTopology_t topo = get_cpuTopology();
+    for (unsigned i = 0; i < topo->numHWThreads; i++)
+    {
+        HWThread* t = &topo->threadPool[i];
+        if ((int)t->coreId == device->id.simple.id && t->inCpuSet)
+        {
+            err = HPMaddThread(t->apicId);
+            if (err < 0)
+                continue;
+            err = HPMread(t->apicId, MSR_DEV, reg, msrData);
+            if (err < 0)
+                continue;
+            return 0;
+        }
     }
+
+    if (err < 0)
+        return err;
     return -ENODEV;
 }
 
@@ -398,6 +428,9 @@ int likwid_sysft_readmsr(const LikwidDevice_t device, uint64_t reg, uint64_t *ms
     case DEVICE_TYPE_SOCKET:
         err = readmsr_socket(device, reg, msrData);
         break;
+    case DEVICE_TYPE_CORE:
+        err = readmsr_core(device, reg, msrData);
+        break;
     case DEVICE_TYPE_HWTHREAD:
         err = readmsr_hwthread(device, reg, msrData);
         break;

src/sysFeatures_intel.c

@@ -48,6 +48,7 @@
 #include <sysFeatures_intel_turbo.h>
 #include <sysFeatures_intel_uncorefreq.h>
 #include <sysFeatures_intel_spec_ctrl.h>
+#include <sysFeatures_intel_thermal.h>
 
 static const _HWArchFeatures intel_arch_features[];
 
@@ -81,6 +82,7 @@ static const _SysFeatureList* intel_arch_feature_inputs[] = {
     &likwid_sysft_intel_cpu_turbo_feature_list,
     &likwid_sysft_intel_uncorefreq_feature_list,
     &likwid_sysft_intel_cpu_spec_ctrl_feature_list,
+    &likwid_sysft_intel_cpu_thermal_feature_list,
     NULL,
 };
 

src/sysFeatures_intel_thermal.c

@@ -0,0 +1,91 @@
+#include <sysFeatures_intel_thermal.h>
+
+#include <string.h>
+#include <stdbool.h>
+
+#include <sysFeatures_common.h>
+#include <topology.h>
+#include <registers.h>
+
+static int intel_thermal_temperature_getter(const LikwidDevice_t device, bool core, char **value)
+{
+    if (!device || !value)
+        return -EINVAL;
+
+    int err;
+    _LikwidDevice socketDev;
+    if (core)
+    {
+        /* If we read temperature from the core, we need to determine its socket.
+         * That is because the TEMPERATURE_TARGET is per socket only. */
+        err = topology_init();
+        if (err < 0)
+            return err;
+
+        bool found = false;
+        CpuTopology_t topo = get_cpuTopology();
+        for (unsigned i = 0; i < topo->numHWThreads; i++)
+        {
+            const HWThread *t = &topo->threadPool[i];
+            if (t->packageId != i)
+                continue;
+
+            memset(&socketDev, 0, sizeof(socketDev));
+            socketDev.type = DEVICE_TYPE_SOCKET;
+            socketDev.id.simple.id = t->packageId;
+            socketDev.internal_id = t->packageId;
+            found = true;
+            break;
+        }
+
+        if (!found)
+            return -EINVAL;
+    } else {
+        socketDev = *device;
+    }
+
+    uint64_t therm_status_raw;
+    err = likwid_sysft_readmsr(device, core ? IA32_THERM_STATUS : IA32_PACKAGE_THERM_STATUS, &therm_status_raw);
+    if (err < 0)
+        return err;
+
+    const int readout = (int)field64(therm_status_raw, 16, 7);
+
+    uint64_t temp_target_raw;
+    err = likwid_sysft_readmsr(&socketDev, MSR_TEMPERATURE_TARGET, &temp_target_raw);
+    if (err < 0)
+        return err;
+
+    const int temp_target = (int)field64(temp_target_raw, 16, 8);
+    const int temp_offset = (int)field64(temp_target_raw, 24, 6);
+
+    const int final_temp = temp_target - temp_offset - readout;
+
+    return likwid_sysft_uint64_to_string(final_temp, value);
+}
+
+static int intel_thermal_temperature_core_getter(LikwidDevice_t dev, char **value)
+{
+    return intel_thermal_temperature_getter(dev, DEVICE_TYPE_CORE, value);
+}
+
+static int intel_thermal_temperature_socket_getter(LikwidDevice_t dev, char **value)
+{
+    return intel_thermal_temperature_getter(dev, DEVICE_TYPE_SOCKET, value);
+}
+
+static int intel_thermal_tester(void)
+{
+    return cpuid_hasFeature(TM2);
+}
+
+static _SysFeature intel_thermal_features[] = {
+    {"core_temp", "thermal", "Current CPU temperature (core)", intel_thermal_temperature_core_getter, NULL, DEVICE_TYPE_CORE, NULL, "degrees C"},
+    {"pkg_temp", "thermal", "Current CPU temperature (package)", intel_thermal_temperature_socket_getter, NULL, DEVICE_TYPE_SOCKET, NULL, "degrees C"},
+};
+
+const _SysFeatureList likwid_sysft_intel_cpu_thermal_feature_list = {
+    .num_features = ARRAY_COUNT(intel_thermal_features),
+    .tester = intel_thermal_tester,
+    .features = intel_thermal_features,
+};