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>
33 #define RDT_PLUGIN "intel_rdt"
35 #define RDT_MAX_SOCKETS 8
36 #define RDT_MAX_SOCKET_CORES 64
37 #define RDT_MAX_CORES (RDT_MAX_SOCKET_CORES * RDT_MAX_SOCKETS)
44 struct rdt_core_group_s {
48 enum pqos_mon_event events;
50 typedef struct rdt_core_group_s rdt_core_group_t;
53 rdt_core_group_t cgroups[RDT_MAX_CORES];
54 struct pqos_mon_data *pgroups[RDT_MAX_CORES];
56 const struct pqos_cpuinfo *pqos_cpu;
57 const struct pqos_cap *pqos_cap;
58 const struct pqos_capability *cap_mon;
60 typedef struct rdt_ctx_s rdt_ctx_t;
62 static rdt_ctx_t *g_rdt = NULL;
64 static rdt_config_status g_state = UNKNOWN;
66 static int isdup(const uint64_t *nums, size_t size, uint64_t val) {
67 for (size_t i = 0; i < size; i++)
73 static int strtouint64(const char *s, uint64_t *n) {
79 *n = strtoull(s, &endptr, 0);
81 if (!(*s != '\0' && *endptr == '\0')) {
82 DEBUG(RDT_PLUGIN ": Error converting '%s' to unsigned number.", s);
94 * Converts string of characters representing list of numbers into array of
95 * numbers. Allowed formats are:
100 * Numbers can be in decimal or hexadecimal format.
103 * `s' String representing list of unsigned numbers.
104 * `nums' Array to put converted numeric values into.
105 * `max' Maximum number of elements that nums can accommodate.
108 * Number of elements placed into nums.
110 static size_t strlisttonums(char *s, uint64_t *nums, size_t max) {
113 char *saveptr = NULL;
115 if (s == NULL || nums == NULL || max == 0)
122 token = strtok_r(s, ",", &saveptr);
128 while (isspace(*token))
133 p = strchr(token, '-');
135 uint64_t n, start, end;
137 ret = strtouint64(token, &start);
140 ret = strtouint64(p + 1, &end);
146 for (n = start; n <= end; n++) {
147 if (!(isdup(nums, index, n))) {
157 ret = strtouint64(token, &val);
161 if (!(isdup(nums, index, val))) {
178 * Function to compare cores in 2 core groups.
181 * `cg_a' Pointer to core group a.
182 * `cg_b' Pointer to core group b.
185 * 1 if both groups contain the same cores
186 * 0 if none of their cores match
187 * -1 if some but not all cores match
189 static int cgroup_cmp(const rdt_core_group_t *cg_a,
190 const rdt_core_group_t *cg_b) {
193 assert(cg_a != NULL);
194 assert(cg_b != NULL);
196 const int sz_a = cg_a->num_cores;
197 const int sz_b = cg_b->num_cores;
198 const unsigned *tab_a = cg_a->cores;
199 const unsigned *tab_b = cg_b->cores;
201 for (int i = 0; i < sz_a; i++) {
202 for (int j = 0; j < sz_b; j++)
203 if (tab_a[i] == tab_b[j])
206 /* if no cores are the same */
209 /* if group contains same cores */
210 if (sz_a == sz_b && sz_b == found)
212 /* if not all cores are the same */
216 static int cgroup_set(rdt_core_group_t *cg, char *desc, uint64_t *cores,
219 assert(desc != NULL);
220 assert(cores != NULL);
221 assert(num_cores > 0);
223 cg->cores = calloc(num_cores, sizeof(unsigned));
224 if (cg->cores == NULL) {
225 ERROR(RDT_PLUGIN ": Error allocating core group table");
228 cg->num_cores = num_cores;
229 cg->desc = strdup(desc);
230 if (cg->desc == NULL) {
231 ERROR(RDT_PLUGIN ": Error allocating core group description");
236 for (size_t i = 0; i < num_cores; i++)
237 cg->cores[i] = (unsigned)cores[i];
247 * Function to set the descriptions and cores for each core group.
248 * Takes a config option containing list of strings that are used to set
252 * `item' Config option containing core groups.
253 * `groups' Table of core groups to set values in.
254 * `max_groups' Maximum number of core groups allowed.
255 * `max_core' Maximum allowed core value.
258 * On success, the number of core groups set up. On error, appropriate
259 * negative error value.
261 static int oconfig_to_cgroups(oconfig_item_t *item, rdt_core_group_t *groups,
262 size_t max_groups, uint64_t max_core) {
265 assert(groups != NULL);
266 assert(max_groups > 0);
267 assert(item != NULL);
269 for (int j = 0; j < item->values_num; j++) {
272 uint64_t cores[RDT_MAX_CORES] = {0};
273 char value[DATA_MAX_NAME_LEN];
275 if ((item->values[j].value.string == NULL) ||
276 (strlen(item->values[j].value.string) == 0))
279 sstrncpy(value, item->values[j].value.string, sizeof(value));
281 n = strlisttonums(value, cores, STATIC_ARRAY_SIZE(cores));
283 ERROR(RDT_PLUGIN ": Error parsing core group (%s)",
284 item->values[j].value.string);
288 for (int i = 0; i < n; i++) {
289 if (cores[i] > max_core) {
290 ERROR(RDT_PLUGIN ": Core group (%s) contains invalid core id (%d)",
291 item->values[j].value.string, (int)cores[i]);
296 /* set core group info */
297 ret = cgroup_set(&groups[index], item->values[j].value.string, cores, n);
303 if (index >= max_groups) {
304 WARNING(RDT_PLUGIN ": Too many core groups configured");
313 static void rdt_dump_cgroups(void) {
314 char cores[RDT_MAX_CORES * 4];
319 DEBUG(RDT_PLUGIN ": Core Groups Dump");
320 DEBUG(RDT_PLUGIN ": groups count: %zu", g_rdt->num_groups);
322 for (int i = 0; i < g_rdt->num_groups; i++) {
324 memset(cores, 0, sizeof(cores));
325 for (int j = 0; j < g_rdt->cgroups[i].num_cores; j++) {
326 snprintf(cores + strlen(cores), sizeof(cores) - strlen(cores) - 1, " %d",
327 g_rdt->cgroups[i].cores[j]);
330 DEBUG(RDT_PLUGIN ": group[%d]:", i);
331 DEBUG(RDT_PLUGIN ": description: %s", g_rdt->cgroups[i].desc);
332 DEBUG(RDT_PLUGIN ": cores: %s", cores);
333 DEBUG(RDT_PLUGIN ": events: 0x%X", g_rdt->cgroups[i].events);
339 static inline double bytes_to_kb(const double bytes) { return bytes / 1024.0; }
341 static inline double bytes_to_mb(const double bytes) {
342 return bytes / (1024.0 * 1024.0);
345 static void rdt_dump_data(void) {
347 * CORE - monitored group of cores
348 * RMID - Resource Monitoring ID associated with the monitored group
349 * LLC - last level cache occupancy
350 * MBL - local memory bandwidth
351 * MBR - remote memory bandwidth
353 DEBUG(" CORE RMID LLC[KB] MBL[MB] MBR[MB]");
354 for (int i = 0; i < g_rdt->num_groups; i++) {
356 const struct pqos_event_values *pv = &g_rdt->pgroups[i]->values;
358 double llc = bytes_to_kb(pv->llc);
359 double mbr = bytes_to_mb(pv->mbm_remote_delta);
360 double mbl = bytes_to_mb(pv->mbm_local_delta);
362 DEBUG(" [%s] %8u %10.1f %10.1f %10.1f", g_rdt->cgroups[i].desc,
363 g_rdt->pgroups[i]->poll_ctx[0].rmid, llc, mbl, mbr);
366 #endif /* COLLECT_DEBUG */
368 static void rdt_free_cgroups(void) {
369 for (int i = 0; i < RDT_MAX_CORES; i++) {
370 sfree(g_rdt->cgroups[i].desc);
372 sfree(g_rdt->cgroups[i].cores);
373 g_rdt->cgroups[i].num_cores = 0;
375 sfree(g_rdt->pgroups[i]);
379 static int rdt_default_cgroups(void) {
382 /* configure each core in separate group */
383 for (unsigned i = 0; i < g_rdt->pqos_cpu->num_cores; i++) {
384 char desc[DATA_MAX_NAME_LEN];
387 ssnprintf(desc, sizeof(desc), "%d", g_rdt->pqos_cpu->cores[i].lcore);
389 /* set core group info */
390 ret = cgroup_set(&g_rdt->cgroups[i], desc, &core, 1);
395 return g_rdt->pqos_cpu->num_cores;
398 static int rdt_config_cgroups(oconfig_item_t *item) {
400 enum pqos_mon_event events = 0;
403 DEBUG(RDT_PLUGIN ": cgroups_config: Invalid argument.");
407 DEBUG(RDT_PLUGIN ": Core groups [%d]:", item->values_num);
408 for (int j = 0; j < item->values_num; j++) {
409 if (item->values[j].type != OCONFIG_TYPE_STRING) {
410 ERROR(RDT_PLUGIN ": given core group value is not a string [idx=%d]", j);
413 DEBUG(RDT_PLUGIN ": [%d]: %s", j, item->values[j].value.string);
416 n = oconfig_to_cgroups(item, g_rdt->cgroups, RDT_MAX_CORES,
417 g_rdt->pqos_cpu->num_cores - 1);
420 ERROR(RDT_PLUGIN ": Error parsing core groups configuration.");
425 /* create default core groups if "Cores" config option is empty */
426 n = rdt_default_cgroups();
429 ERROR(RDT_PLUGIN ": Error creating default core groups configuration.");
433 ": No core groups configured. Default core groups created.");
436 /* Get all available events on this platform */
437 for (int i = 0; i < g_rdt->cap_mon->u.mon->num_events; i++)
438 events |= g_rdt->cap_mon->u.mon->events[i].type;
440 events &= ~(PQOS_PERF_EVENT_LLC_MISS);
442 DEBUG(RDT_PLUGIN ": Number of cores in the system: %u",
443 g_rdt->pqos_cpu->num_cores);
444 DEBUG(RDT_PLUGIN ": Available events to monitor: %#x", events);
446 g_rdt->num_groups = n;
447 for (int i = 0; i < n; i++) {
448 for (int j = 0; j < i; j++) {
450 found = cgroup_cmp(&g_rdt->cgroups[j], &g_rdt->cgroups[i]);
453 ERROR(RDT_PLUGIN ": Cannot monitor same cores in different groups.");
458 g_rdt->cgroups[i].events = events;
459 g_rdt->pgroups[i] = calloc(1, sizeof(*g_rdt->pgroups[i]));
460 if (g_rdt->pgroups[i] == NULL) {
462 ERROR(RDT_PLUGIN ": Failed to allocate memory for monitoring data.");
470 static void rdt_pqos_log(void *context, const size_t size, const char *msg) {
471 DEBUG(RDT_PLUGIN ": %s", msg);
474 static int rdt_preinit(void) {
478 /* already initialized if config callback was called before init callback */
482 g_rdt = calloc(1, sizeof(*g_rdt));
484 ERROR(RDT_PLUGIN ": Failed to allocate memory for rdt context.");
488 struct pqos_config pqos = {.fd_log = -1,
489 .callback_log = rdt_pqos_log,
493 ret = pqos_init(&pqos);
494 if (ret != PQOS_RETVAL_OK) {
495 ERROR(RDT_PLUGIN ": Error initializing PQoS library!");
496 goto rdt_preinit_error1;
499 ret = pqos_cap_get(&g_rdt->pqos_cap, &g_rdt->pqos_cpu);
500 if (ret != PQOS_RETVAL_OK) {
501 ERROR(RDT_PLUGIN ": Error retrieving PQoS capabilities.");
502 goto rdt_preinit_error2;
505 ret = pqos_cap_get_type(g_rdt->pqos_cap, PQOS_CAP_TYPE_MON, &g_rdt->cap_mon);
506 if (ret == PQOS_RETVAL_PARAM) {
507 ERROR(RDT_PLUGIN ": Error retrieving monitoring capabilities.");
508 goto rdt_preinit_error2;
511 if (g_rdt->cap_mon == NULL) {
514 ": Monitoring capability not detected. Nothing to do for the plugin.");
515 goto rdt_preinit_error2;
518 /* Reset pqos monitoring groups registers */
533 static int rdt_config(oconfig_item_t *ci) {
538 g_state = CONFIGURATION_ERROR;
539 /* if we return -1 at this point collectd
540 reports a failure in configuration and
546 for (int i = 0; i < ci->children_num; i++) {
547 oconfig_item_t *child = ci->children + i;
549 if (strcasecmp("Cores", child->key) == 0) {
551 ret = rdt_config_cgroups(child);
553 g_state = CONFIGURATION_ERROR;
554 /* if we return -1 at this point collectd
555 reports a failure in configuration and
563 #endif /* COLLECT_DEBUG */
566 ERROR(RDT_PLUGIN ": Unknown configuration parameter \"%s\".", child->key);
574 static void rdt_submit_derive(char *cgroup, char *type, char *type_instance,
576 value_list_t vl = VALUE_LIST_INIT;
578 vl.values = &(value_t){.derive = value};
581 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
582 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
583 sstrncpy(vl.type, type, sizeof(vl.type));
585 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
587 plugin_dispatch_values(&vl);
590 static void rdt_submit_gauge(char *cgroup, char *type, char *type_instance,
592 value_list_t vl = VALUE_LIST_INIT;
594 vl.values = &(value_t){.gauge = value};
597 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
598 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
599 sstrncpy(vl.type, type, sizeof(vl.type));
601 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
603 plugin_dispatch_values(&vl);
606 static int rdt_read(__attribute__((unused)) user_data_t *ud) {
610 ERROR(RDT_PLUGIN ": rdt_read: plugin not initialized.");
614 ret = pqos_mon_poll(&g_rdt->pgroups[0], (unsigned)g_rdt->num_groups);
615 if (ret != PQOS_RETVAL_OK) {
616 ERROR(RDT_PLUGIN ": Failed to poll monitoring data.");
622 #endif /* COLLECT_DEBUG */
624 for (int i = 0; i < g_rdt->num_groups; i++) {
625 enum pqos_mon_event mbm_events =
626 (PQOS_MON_EVENT_LMEM_BW | PQOS_MON_EVENT_TMEM_BW |
627 PQOS_MON_EVENT_RMEM_BW);
629 const struct pqos_event_values *pv = &g_rdt->pgroups[i]->values;
631 /* Submit only monitored events data */
633 if (g_rdt->cgroups[i].events & PQOS_MON_EVENT_L3_OCCUP)
634 rdt_submit_gauge(g_rdt->cgroups[i].desc, "bytes", "llc", pv->llc);
636 if (g_rdt->cgroups[i].events & PQOS_PERF_EVENT_IPC)
637 rdt_submit_gauge(g_rdt->cgroups[i].desc, "ipc", NULL, pv->ipc);
639 if (g_rdt->cgroups[i].events & mbm_events) {
640 rdt_submit_derive(g_rdt->cgroups[i].desc, "memory_bandwidth", "local",
641 pv->mbm_local_delta);
642 rdt_submit_derive(g_rdt->cgroups[i].desc, "memory_bandwidth", "remote",
643 pv->mbm_remote_delta);
650 static int rdt_init(void) {
653 if(g_state == CONFIGURATION_ERROR)
660 /* Start monitoring */
661 for (int i = 0; i < g_rdt->num_groups; i++) {
662 rdt_core_group_t *cg = &g_rdt->cgroups[i];
664 ret = pqos_mon_start(cg->num_cores, cg->cores, cg->events, (void *)cg->desc,
667 if (ret != PQOS_RETVAL_OK)
668 ERROR(RDT_PLUGIN ": Error starting monitoring group %s (pqos status=%d)",
675 static int rdt_shutdown(void) {
678 DEBUG(RDT_PLUGIN ": rdt_shutdown.");
683 /* Stop monitoring */
684 for (int i = 0; i < g_rdt->num_groups; i++) {
685 pqos_mon_stop(g_rdt->pgroups[i]);
689 if (ret != PQOS_RETVAL_OK)
690 ERROR(RDT_PLUGIN ": Error shutting down PQoS library.");
698 void module_register(void) {
699 plugin_register_init(RDT_PLUGIN, rdt_init);
700 plugin_register_complex_config(RDT_PLUGIN, rdt_config);
701 plugin_register_complex_read(NULL, RDT_PLUGIN, rdt_read, 0, NULL);
702 plugin_register_shutdown(RDT_PLUGIN, rdt_shutdown);