Add support for NUMA nodes

Related to #599.

6 files changed, 75 insertions, 39 deletions

diff --git a/hardinfo/cpu_util.c b/hardinfo/cpu_util.c
index 5064c173..60f2bbae 100644
--- a/hardinfo/cpu_util.c
+++ b/hardinfo/cpu_util.c
@@ -71,37 +71,58 @@ gint get_cpu_int(const char* item, int cpuid, int null_val) {
 #define MAX_CORES_PER_PACK 256
 #define MAX_PACKS 128
 
-int cpu_procs_cores_threads(int *p, int *c, int *t) {
+int cpu_procs_cores_threads_nodes(int *p, int *c, int *t, int *n)
+{
     cpubits *threads, *cores, *packs;
     char *tmp;
     int i, m, pack_id, core_id;
+
     g_file_get_contents("/sys/devices/system/cpu/present", &tmp, NULL, NULL);
-    if (tmp != NULL) {
-        threads = cpubits_from_str(tmp);
-        cores = cpubits_from_str("");
-        packs = cpubits_from_str("");
-        m = cpubits_max(threads);
-        for (i = 0; i <= m; i++) {
-            pack_id = get_cpu_int("topology/physical_package_id", i, CPU_TOPO_NULL);
-            core_id = get_cpu_int("topology/core_id", i, CPU_TOPO_NULL);
-            if (pack_id < 0) pack_id = 0;
-            CPUBIT_SET(packs, pack_id);
-            if (core_id >= 0) { CPUBIT_SET(cores, (pack_id * MAX_CORES_PER_PACK) + core_id ); }
-        }
-        *t = cpubits_count(threads);
-        *c = cpubits_count(cores);
-        *p = cpubits_count(packs);
-        if (!*c) *c = 1;
-        if (!*p) *p = 1;
-        free(threads);
-        free(cores);
-        free(packs);
-        free(tmp);
-        return 1;
-    } else {
-        *p = *c = *t = -1;
+    if (tmp == NULL) {
+        *p = *c = *t = *n = -1;
         return 0;
     }
+
+    threads = cpubits_from_str(tmp);
+    cores = cpubits_from_str("");
+    packs = cpubits_from_str("");
+    m = cpubits_max(threads);
+    for (i = 0; i <= m; i++) {
+        pack_id = get_cpu_int("topology/physical_package_id", i, CPU_TOPO_NULL);
+        core_id = get_cpu_int("topology/core_id", i, CPU_TOPO_NULL);
+        if (pack_id < 0)
+            pack_id = 0;
+        CPUBIT_SET(packs, pack_id);
+        if (core_id >= 0) {
+            CPUBIT_SET(cores, (pack_id * MAX_CORES_PER_PACK) + core_id);
+        }
+    }
+
+    *t = cpubits_count(threads);
+    *c = cpubits_count(cores);
+    *p = cpubits_count(packs);
+
+    g_free(tmp);
+    g_file_get_contents("/sys/devices/system/node/possible", &tmp, NULL, NULL);
+    if (tmp != NULL) {
+        cpubits *nodes = cpubits_from_str(tmp);
+        if (nodes)
+            *n = cpubits_count(nodes);
+        free(nodes);
+    }
+
+    if (!*c)
+        *c = 1;
+    if (!*p)
+        *p = 1;
+    if (!*n)
+        *n = 1;
+
+    g_free(threads);
+    g_free(cores);
+    g_free(packs);
+    g_free(tmp);
+    return 1;
 }
 
 cpufreq_data *cpufreq_new(gint id)
diff --git a/includes/cpu_util.h b/includes/cpu_util.h
index c11d3f69..1adbf491 100644
--- a/includes/cpu_util.h
+++ b/includes/cpu_util.h
@@ -47,4 +47,6 @@ void cputopo_free(cpu_topology_data *cputd);
 
 gchar *cputopo_section_str(cpu_topology_data *cputd);
 
+int cpu_procs_cores_threads_nodes(int *p, int *c, int *t, int *n);
+
 #endif
diff --git a/modules/benchmark.c b/modules/benchmark.c
index 6d36100f..4b536d42 100644
--- a/modules/benchmark.c
+++ b/modules/benchmark.c
@@ -135,14 +135,15 @@ bench_value benchmark_crunch_for(float seconds,
                                  gpointer callback,
                                  gpointer callback_data)
 {
-    int cpu_procs, cpu_cores, cpu_threads, thread_number, stop = 0;
+    int cpu_procs, cpu_cores, cpu_threads, cpu_nodes;
+    int thread_number, stop = 0;
     GSList *threads = NULL, *t;
     GTimer *timer;
     bench_value ret = EMPTY_BENCH_VALUE;
 
     timer = g_timer_new();
 
-    cpu_procs_cores_threads(&cpu_procs, &cpu_cores, &cpu_threads);
+    cpu_procs_cores_threads_nodes(&cpu_procs, &cpu_cores, &cpu_threads, &cpu_nodes);
     if (n_threads > 0)
         ret.threads_used = n_threads;
     else if (n_threads < 0)
@@ -222,12 +223,14 @@ static gpointer benchmark_parallel_for_dispatcher(gpointer data)
 bench_value
 benchmark_parallel(gint n_threads, gpointer callback, gpointer callback_data)
 {
-    int cpu_procs, cpu_cores, cpu_threads;
-    cpu_procs_cores_threads(&cpu_procs, &cpu_cores, &cpu_threads);
+    int cpu_procs, cpu_cores, cpu_threads, cpu_nodes;
+    cpu_procs_cores_threads_nodes(&cpu_procs, &cpu_cores, &cpu_threads, &cpu_nodes);
+
     if (n_threads == 0)
         n_threads = cpu_threads;
     else if (n_threads == -1)
         n_threads = cpu_cores;
+
     return benchmark_parallel_for(n_threads, 0, n_threads, callback,
                                   callback_data);
 }
@@ -244,7 +247,7 @@ bench_value benchmark_parallel_for(gint n_threads,
                                    gpointer callback_data)
 {
     gchar *temp;
-    int cpu_procs, cpu_cores, cpu_threads;
+    int cpu_procs, cpu_cores, cpu_threads, cpu_nodes;
     guint iter_per_thread, iter, thread_number = 0;
     GSList *threads = NULL, *t;
     GTimer *timer;
@@ -253,7 +256,7 @@ bench_value benchmark_parallel_for(gint n_threads,
 
     timer = g_timer_new();
 
-    cpu_procs_cores_threads(&cpu_procs, &cpu_cores, &cpu_threads);
+    cpu_procs_cores_threads_nodes(&cpu_procs, &cpu_cores, &cpu_threads, &cpu_nodes);
 
     if (n_threads > 0)
         ret.threads_used = n_threads;
diff --git a/modules/benchmark/bench_results.c b/modules/benchmark/bench_results.c
index 40345aa5..6ba475ed 100644
--- a/modules/benchmark/bench_results.c
+++ b/modules/benchmark/bench_results.c
@@ -40,6 +40,7 @@ typedef struct {
     int processors;
     int cores;
     int threads;
+    int nodes;
     char *mid;
     int ptr_bits;             /* 32, 64... BENCH_PTR_BITS; 0 for unspecified */
     int is_su_data;           /* 1 = data collected as root */
@@ -183,7 +184,7 @@ bench_machine *bench_machine_this()
         m->machine_type = module_call_method("computer::getMachineType");
         free(tmp);
 
-        cpu_procs_cores_threads(&m->processors, &m->cores, &m->threads);
+        cpu_procs_cores_threads_nodes(&m->processors, &m->cores, &m->threads, &m->nodes);
         gen_machine_id(m);
     }
     return m;
@@ -355,6 +356,11 @@ bench_result *bench_result_benchmarkjson(const gchar *bench_name,
              json_get_string(machine, "UserNote"));
     filter_invalid_chars(b->bvalue.user_note);
 
+    int nodes = json_get_int(machine, "NumNodes");
+
+    if (nodes == 0)
+        nodes = 1;
+
     b->machine = bench_machine_new();
     *b->machine = (bench_machine){
         .board = json_get_string_dup(machine, "Board"),
@@ -367,6 +373,7 @@ bench_result *bench_result_benchmarkjson(const gchar *bench_name,
         .processors = json_get_int(machine, "NumCpus"),
         .cores = json_get_int(machine, "NumCores"),
         .threads = json_get_int(machine, "NumThreads"),
+        .nodes = nodes,
         .mid = json_get_string_dup(machine, "MachineId"),
         .ptr_bits = json_get_int(machine, "PointerBits"),
         .is_su_data = json_get_boolean(machine, "DataFromSuperUser"),
diff --git a/modules/benchmark/sysbench.c b/modules/benchmark/sysbench.c
index 9b8a4a38..75008804 100644
--- a/modules/benchmark/sysbench.c
+++ b/modules/benchmark/sysbench.c
@@ -230,8 +230,10 @@ void benchmark_sbcpu_single(void) {
 }
 
 void benchmark_sbcpu_all(void) {
-    int cpu_procs, cpu_cores, cpu_threads;
-    cpu_procs_cores_threads(&cpu_procs, &cpu_cores, &cpu_threads);
+    int cpu_procs, cpu_cores, cpu_threads, cpu_nodes;
+
+    cpu_procs_cores_threads_nodes(&cpu_procs, &cpu_cores, &cpu_threads, &cpu_nodes);
+
     struct sysbench_ctx ctx = {
         .test = "cpu",
         .threads = cpu_threads,
diff --git a/modules/devices.c b/modules/devices.c
index 578cce3f..d6f30aef 100644
--- a/modules/devices.c
+++ b/modules/devices.c
@@ -166,19 +166,20 @@ static gint proc_cmp_max_freq(Processor *a, Processor *b) {
 
 gchar *processor_describe_default(GSList * processors)
 {
-    int packs, cores, threads;
-    const gchar  *packs_fmt, *cores_fmt, *threads_fmt;
+    int packs, cores, threads, nodes;
+    const gchar  *packs_fmt, *cores_fmt, *threads_fmt, *nodes_fmt;
     gchar *ret, *full_fmt;
 
-    cpu_procs_cores_threads(&packs, &cores, &threads);
+    cpu_procs_cores_threads_nodes(&packs, &cores, &threads, &nodes);
 
     /* if topology info was available, else fallback to old method */
     if (cores > 0) {
         packs_fmt = ngettext("%d physical processor", "%d physical processors", packs);
         cores_fmt = ngettext("%d core", "%d cores", cores);
         threads_fmt = ngettext("%d thread", "%d threads", threads);
-        full_fmt = g_strdup_printf(_(/*/NP procs; NC cores; NT threads*/ "%s; %s; %s"), packs_fmt, cores_fmt, threads_fmt);
-        ret = g_strdup_printf(full_fmt, packs, cores, threads);
+        nodes_fmt = ngettext("%d NUMA node", "%d NUMA nodes", nodes);
+        full_fmt = g_strdup_printf(_(/*/NP procs; NC cores; NN nodes; NT threads*/ "%s; %s, %s; %s"), packs_fmt, cores_fmt, nodes_fmt, threads_fmt);
+        ret = g_strdup_printf(full_fmt, packs, cores, nodes, threads);
         g_free(full_fmt);
         return ret;
     } else {