2 * collectd - src/intel_rdt.c
4 * Copyright(c) 2016 Intel Corporation. All rights reserved.
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
25 * Serhiy Pshyk <serhiyx.pshyk@intel.com>
31 #include "utils_config_cores.h"
35 #define RDT_PLUGIN "intel_rdt"
37 #define RDT_MAX_SOCKETS 8
38 #define RDT_MAX_SOCKET_CORES 64
39 #define RDT_MAX_CORES (RDT_MAX_SOCKET_CORES * RDT_MAX_SOCKETS)
47 core_groups_list_t cores;
48 enum pqos_mon_event events[RDT_MAX_CORES];
49 struct pqos_mon_data *pgroups[RDT_MAX_CORES];
51 const struct pqos_cpuinfo *pqos_cpu;
52 const struct pqos_cap *pqos_cap;
53 const struct pqos_capability *cap_mon;
55 typedef struct rdt_ctx_s rdt_ctx_t;
57 static rdt_ctx_t *g_rdt = NULL;
59 static rdt_config_status g_state = UNKNOWN;
62 static void rdt_dump_cgroups(void) {
63 char cores[RDT_MAX_CORES * 4];
68 DEBUG(RDT_PLUGIN ": Core Groups Dump");
69 DEBUG(RDT_PLUGIN ": groups count: %zu", g_rdt->num_groups);
71 for (int i = 0; i < g_rdt->num_groups; i++) {
72 core_group_t *cgroup = g_rdt->cores.cgroups + i;
74 memset(cores, 0, sizeof(cores));
75 for (int j = 0; j < cgroup->num_cores; j++) {
76 snprintf(cores + strlen(cores), sizeof(cores) - strlen(cores) - 1, " %d",
80 DEBUG(RDT_PLUGIN ": group[%d]:", i);
81 DEBUG(RDT_PLUGIN ": description: %s", cgroup->desc);
82 DEBUG(RDT_PLUGIN ": cores: %s", cores);
83 DEBUG(RDT_PLUGIN ": events: 0x%X", g_rdt->events[i]);
89 static inline double bytes_to_kb(const double bytes) { return bytes / 1024.0; }
91 static inline double bytes_to_mb(const double bytes) {
92 return bytes / (1024.0 * 1024.0);
95 static void rdt_dump_data(void) {
97 * CORE - monitored group of cores
98 * RMID - Resource Monitoring ID associated with the monitored group
99 * LLC - last level cache occupancy
100 * MBL - local memory bandwidth
101 * MBR - remote memory bandwidth
103 DEBUG(" CORE RMID LLC[KB] MBL[MB] MBR[MB]");
104 for (int i = 0; i < g_rdt->num_groups; i++) {
106 const struct pqos_event_values *pv = &g_rdt->pgroups[i]->values;
108 double llc = bytes_to_kb(pv->llc);
109 double mbr = bytes_to_mb(pv->mbm_remote_delta);
110 double mbl = bytes_to_mb(pv->mbm_local_delta);
112 DEBUG(" [%s] %8u %10.1f %10.1f %10.1f", g_rdt->cores.cgroups[i].desc,
113 g_rdt->pgroups[i]->poll_ctx[0].rmid, llc, mbl, mbr);
116 #endif /* COLLECT_DEBUG */
118 static void rdt_free_cgroups(void) {
119 config_cores_cleanup(&g_rdt->cores);
120 for (int i = 0; i < RDT_MAX_CORES; i++) {
121 sfree(g_rdt->pgroups[i]);
125 static int rdt_default_cgroups(void) {
126 unsigned num_cores = g_rdt->pqos_cpu->num_cores;
128 g_rdt->cores.cgroups = calloc(num_cores, sizeof(*(g_rdt->cores.cgroups)));
129 if (g_rdt->cores.cgroups == NULL) {
130 ERROR(RDT_PLUGIN ": Error allocating core groups array");
133 g_rdt->cores.num_cgroups = num_cores;
135 /* configure each core in separate group */
136 for (unsigned i = 0; i < num_cores; i++) {
137 core_group_t *cgroup = g_rdt->cores.cgroups + i;
138 char desc[DATA_MAX_NAME_LEN];
140 /* set core group info */
141 cgroup->cores = calloc(1, sizeof(*(cgroup->cores)));
142 if (cgroup->cores == NULL) {
143 ERROR(RDT_PLUGIN ": Error allocating cores array");
147 cgroup->num_cores = 1;
148 cgroup->cores[0] = i;
150 snprintf(desc, sizeof(desc), "%d", g_rdt->pqos_cpu->cores[i].lcore);
151 cgroup->desc = strdup(desc);
152 if (cgroup->desc == NULL) {
153 ERROR(RDT_PLUGIN ": Error allocating core group description");
162 static int rdt_is_core_id_valid(int core_id) {
164 for (int i = 0; i < g_rdt->pqos_cpu->num_cores; i++)
165 if (core_id == g_rdt->pqos_cpu->cores[i].lcore)
171 static int rdt_config_cgroups(oconfig_item_t *item) {
173 enum pqos_mon_event events = 0;
175 if (config_cores_parse(item, &g_rdt->cores) < 0) {
177 ERROR(RDT_PLUGIN ": Error parsing core groups configuration.");
180 n = g_rdt->cores.num_cgroups;
182 /* validate configured core id values */
183 for (size_t group_idx = 0; group_idx < n; group_idx++) {
184 core_group_t *cgroup = g_rdt->cores.cgroups + group_idx;
185 for (size_t core_idx = 0; core_idx < cgroup->num_cores; core_idx++) {
186 if (!rdt_is_core_id_valid((int) cgroup->cores[core_idx])) {
187 ERROR(RDT_PLUGIN ": Core group '%s' contains invalid core id '%d'",
188 cgroup->desc, (int)cgroup->cores[core_idx]);
196 /* create default core groups if "Cores" config option is empty */
197 int ret = rdt_default_cgroups();
200 ERROR(RDT_PLUGIN ": Error creating default core groups configuration.");
205 ": No core groups configured. Default core groups created.");
208 /* Get all available events on this platform */
209 for (int i = 0; i < g_rdt->cap_mon->u.mon->num_events; i++)
210 events |= g_rdt->cap_mon->u.mon->events[i].type;
212 events &= ~(PQOS_PERF_EVENT_LLC_MISS);
214 DEBUG(RDT_PLUGIN ": Number of cores in the system: %u",
215 g_rdt->pqos_cpu->num_cores);
216 DEBUG(RDT_PLUGIN ": Available events to monitor: %#x", events);
218 g_rdt->num_groups = n;
219 for (size_t i = 0; i < n; i++) {
220 for (size_t j = 0; j < i; j++) {
222 found = config_cores_cmp_cgroups(&g_rdt->cores.cgroups[j],
223 &g_rdt->cores.cgroups[i]);
226 ERROR(RDT_PLUGIN ": Cannot monitor same cores in different groups.");
231 g_rdt->events[i] = events;
232 g_rdt->pgroups[i] = calloc(1, sizeof(*g_rdt->pgroups[i]));
233 if (g_rdt->pgroups[i] == NULL) {
235 ERROR(RDT_PLUGIN ": Failed to allocate memory for monitoring data.");
243 static void rdt_pqos_log(void *context, const size_t size, const char *msg) {
244 DEBUG(RDT_PLUGIN ": %s", msg);
247 static int rdt_preinit(void) {
251 /* already initialized if config callback was called before init callback */
255 g_rdt = calloc(1, sizeof(*g_rdt));
257 ERROR(RDT_PLUGIN ": Failed to allocate memory for rdt context.");
261 struct pqos_config pqos = {.fd_log = -1,
262 .callback_log = rdt_pqos_log,
266 ret = pqos_init(&pqos);
267 if (ret != PQOS_RETVAL_OK) {
268 ERROR(RDT_PLUGIN ": Error initializing PQoS library!");
269 goto rdt_preinit_error1;
272 ret = pqos_cap_get(&g_rdt->pqos_cap, &g_rdt->pqos_cpu);
273 if (ret != PQOS_RETVAL_OK) {
274 ERROR(RDT_PLUGIN ": Error retrieving PQoS capabilities.");
275 goto rdt_preinit_error2;
278 ret = pqos_cap_get_type(g_rdt->pqos_cap, PQOS_CAP_TYPE_MON, &g_rdt->cap_mon);
279 if (ret == PQOS_RETVAL_PARAM) {
280 ERROR(RDT_PLUGIN ": Error retrieving monitoring capabilities.");
281 goto rdt_preinit_error2;
284 if (g_rdt->cap_mon == NULL) {
287 ": Monitoring capability not detected. Nothing to do for the plugin.");
288 goto rdt_preinit_error2;
291 /* Reset pqos monitoring groups registers */
306 static int rdt_config(oconfig_item_t *ci) {
307 if (rdt_preinit() != 0) {
308 g_state = CONFIGURATION_ERROR;
309 /* if we return -1 at this point collectd
310 reports a failure in configuration and
316 for (int i = 0; i < ci->children_num; i++) {
317 oconfig_item_t *child = ci->children + i;
319 if (strcasecmp("Cores", child->key) == 0) {
320 if (rdt_config_cgroups(child) != 0) {
321 g_state = CONFIGURATION_ERROR;
322 /* if we return -1 at this point collectd
323 reports a failure in configuration and
331 #endif /* COLLECT_DEBUG */
333 ERROR(RDT_PLUGIN ": Unknown configuration parameter \"%s\".", child->key);
340 static void rdt_submit_derive(char *cgroup, char *type, char *type_instance,
342 value_list_t vl = VALUE_LIST_INIT;
344 vl.values = &(value_t){.derive = value};
347 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
348 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
349 sstrncpy(vl.type, type, sizeof(vl.type));
351 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
353 plugin_dispatch_values(&vl);
356 static void rdt_submit_gauge(char *cgroup, char *type, char *type_instance,
358 value_list_t vl = VALUE_LIST_INIT;
360 vl.values = &(value_t){.gauge = value};
363 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
364 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
365 sstrncpy(vl.type, type, sizeof(vl.type));
367 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
369 plugin_dispatch_values(&vl);
372 static int rdt_read(__attribute__((unused)) user_data_t *ud) {
376 ERROR(RDT_PLUGIN ": rdt_read: plugin not initialized.");
380 ret = pqos_mon_poll(&g_rdt->pgroups[0], (unsigned)g_rdt->num_groups);
381 if (ret != PQOS_RETVAL_OK) {
382 ERROR(RDT_PLUGIN ": Failed to poll monitoring data.");
388 #endif /* COLLECT_DEBUG */
390 for (int i = 0; i < g_rdt->num_groups; i++) {
391 core_group_t *cgroup = g_rdt->cores.cgroups + i;
393 enum pqos_mon_event mbm_events =
394 (PQOS_MON_EVENT_LMEM_BW | PQOS_MON_EVENT_TMEM_BW |
395 PQOS_MON_EVENT_RMEM_BW);
397 const struct pqos_event_values *pv = &g_rdt->pgroups[i]->values;
399 /* Submit only monitored events data */
401 if (g_rdt->events[i] & PQOS_MON_EVENT_L3_OCCUP)
402 rdt_submit_gauge(cgroup->desc, "bytes", "llc", pv->llc);
404 if (g_rdt->events[i] & PQOS_PERF_EVENT_IPC)
405 rdt_submit_gauge(cgroup->desc, "ipc", NULL, pv->ipc);
407 if (g_rdt->events[i] & mbm_events) {
408 rdt_submit_derive(cgroup->desc, "memory_bandwidth", "local",
409 pv->mbm_local_delta);
410 rdt_submit_derive(cgroup->desc, "memory_bandwidth", "remote",
411 pv->mbm_remote_delta);
418 static int rdt_init(void) {
421 if (g_state == CONFIGURATION_ERROR)
428 /* Start monitoring */
429 for (int i = 0; i < g_rdt->num_groups; i++) {
430 core_group_t *cg = g_rdt->cores.cgroups + i;
432 ret = pqos_mon_start(cg->num_cores, cg->cores, g_rdt->events[i],
433 (void *)cg->desc, g_rdt->pgroups[i]);
435 if (ret != PQOS_RETVAL_OK)
436 ERROR(RDT_PLUGIN ": Error starting monitoring group %s (pqos status=%d)",
443 static int rdt_shutdown(void) {
446 DEBUG(RDT_PLUGIN ": rdt_shutdown.");
451 /* Stop monitoring */
452 for (int i = 0; i < g_rdt->num_groups; i++) {
453 pqos_mon_stop(g_rdt->pgroups[i]);
457 if (ret != PQOS_RETVAL_OK)
458 ERROR(RDT_PLUGIN ": Error shutting down PQoS library.");
466 void module_register(void) {
467 plugin_register_init(RDT_PLUGIN, rdt_init);
468 plugin_register_complex_config(RDT_PLUGIN, rdt_config);
469 plugin_register_complex_read(NULL, RDT_PLUGIN, rdt_read, 0, NULL);
470 plugin_register_shutdown(RDT_PLUGIN, rdt_shutdown);