1 /**
2 * collectd - src/intel_rdt.c
3 *
4 * Copyright(c) 2016 Intel Corporation. All rights reserved.
5 *
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:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
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
22 * SOFTWARE.
23 *
24 * Authors:
25 * Serhiy Pshyk <serhiyx.pshyk@intel.com>
26 **/
28 #include "collectd.h"
29 #include "common.h"
31 #include <pqos.h>
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)
39 struct rdt_core_group_s {
40 char *desc;
41 size_t num_cores;
42 unsigned *cores;
43 enum pqos_mon_event events;
44 };
45 typedef struct rdt_core_group_s rdt_core_group_t;
47 struct rdt_ctx_s {
48 rdt_core_group_t cgroups[RDT_MAX_CORES];
49 struct pqos_mon_data *pgroups[RDT_MAX_CORES];
50 size_t num_groups;
51 const struct pqos_cpuinfo *pqos_cpu;
52 const struct pqos_cap *pqos_cap;
53 const struct pqos_capability *cap_mon;
54 };
55 typedef struct rdt_ctx_s rdt_ctx_t;
57 static rdt_ctx_t *g_rdt = NULL;
59 static int isdup(const uint64_t *nums, size_t size, uint64_t val) {
60 for (size_t i = 0; i < size; i++)
61 if (nums[i] == val)
62 return 1;
63 return 0;
64 }
66 static int strtouint64(const char *s, uint64_t *n) {
67 char *endptr = NULL;
69 assert(s != NULL);
70 assert(n != NULL);
72 *n = strtoull(s, &endptr, 0);
74 if (!(*s != '\0' && *endptr == '\0')) {
75 DEBUG(RDT_PLUGIN ": Error converting '%s' to unsigned number.", s);
76 return (-EINVAL);
77 }
79 return (0);
80 }
82 /*
83 * NAME
84 * strlisttonums
85 *
86 * DESCRIPTION
87 * Converts string of characters representing list of numbers into array of
88 * numbers. Allowed formats are:
89 * 0,1,2,3
90 * 0-10,20-18
91 * 1,3,5-8,10,0x10-12
92 *
93 * Numbers can be in decimal or hexadecimal format.
94 *
95 * PARAMETERS
96 * `s' String representing list of unsigned numbers.
97 * `nums' Array to put converted numeric values into.
98 * `max' Maximum number of elements that nums can accommodate.
99 *
100 * RETURN VALUE
101 * Number of elements placed into nums.
102 */
103 static size_t strlisttonums(char *s, uint64_t *nums, size_t max) {
104 int ret;
105 size_t index = 0;
106 char *saveptr = NULL;
108 if (s == NULL || nums == NULL || max == 0)
109 return index;
111 for (;;) {
112 char *p = NULL;
113 char *token = NULL;
115 token = strtok_r(s, ",", &saveptr);
116 if (token == NULL)
117 break;
119 s = NULL;
121 while (isspace(*token))
122 token++;
123 if (*token == '\0')
124 continue;
126 p = strchr(token, '-');
127 if (p != NULL) {
128 uint64_t n, start, end;
129 *p = '\0';
130 ret = strtouint64(token, &start);
131 if (ret < 0)
132 return (0);
133 ret = strtouint64(p + 1, &end);
134 if (ret < 0)
135 return (0);
136 if (start > end) {
137 return (0);
138 }
139 for (n = start; n <= end; n++) {
140 if (!(isdup(nums, index, n))) {
141 nums[index] = n;
142 index++;
143 }
144 if (index >= max)
145 return index;
146 }
147 } else {
148 uint64_t val;
150 ret = strtouint64(token, &val);
151 if (ret < 0)
152 return (0);
154 if (!(isdup(nums, index, val))) {
155 nums[index] = val;
156 index++;
157 }
158 if (index >= max)
159 return index;
160 }
161 }
163 return index;
164 }
166 /*
167 * NAME
168 * cgroup_cmp
169 *
170 * DESCRIPTION
171 * Function to compare cores in 2 core groups.
172 *
173 * PARAMETERS
174 * `cg_a' Pointer to core group a.
175 * `cg_b' Pointer to core group b.
176 *
177 * RETURN VALUE
178 * 1 if both groups contain the same cores
179 * 0 if none of their cores match
180 * -1 if some but not all cores match
181 */
182 static int cgroup_cmp(const rdt_core_group_t *cg_a,
183 const rdt_core_group_t *cg_b) {
184 int found = 0;
186 assert(cg_a != NULL);
187 assert(cg_b != NULL);
189 const int sz_a = cg_a->num_cores;
190 const int sz_b = cg_b->num_cores;
191 const unsigned *tab_a = cg_a->cores;
192 const unsigned *tab_b = cg_b->cores;
194 for (int i = 0; i < sz_a; i++) {
195 for (int j = 0; j < sz_b; j++)
196 if (tab_a[i] == tab_b[j])
197 found++;
198 }
199 /* if no cores are the same */
200 if (!found)
201 return 0;
202 /* if group contains same cores */
203 if (sz_a == sz_b && sz_b == found)
204 return 1;
205 /* if not all cores are the same */
206 return -1;
207 }
209 static int cgroup_set(rdt_core_group_t *cg, char *desc, uint64_t *cores,
210 size_t num_cores) {
211 assert(cg != NULL);
212 assert(desc != NULL);
213 assert(cores != NULL);
214 assert(num_cores > 0);
216 cg->cores = calloc(num_cores, sizeof(unsigned));
217 if (cg->cores == NULL) {
218 ERROR(RDT_PLUGIN ": Error allocating core group table");
219 return (-ENOMEM);
220 }
221 cg->num_cores = num_cores;
222 cg->desc = strdup(desc);
223 if (cg->desc == NULL) {
224 ERROR(RDT_PLUGIN ": Error allocating core group description");
225 sfree(cg->cores);
226 return (-ENOMEM);
227 }
229 for (size_t i = 0; i < num_cores; i++)
230 cg->cores[i] = (unsigned)cores[i];
232 return 0;
233 }
235 /*
236 * NAME
237 * oconfig_to_cgroups
238 *
239 * DESCRIPTION
240 * Function to set the descriptions and cores for each core group.
241 * Takes a config option containing list of strings that are used to set
242 * core group values.
243 *
244 * PARAMETERS
245 * `item' Config option containing core groups.
246 * `groups' Table of core groups to set values in.
247 * `max_groups' Maximum number of core groups allowed.
248 * `max_core' Maximum allowed core value.
249 *
250 * RETURN VALUE
251 * On success, the number of core groups set up. On error, appropriate
252 * negative error value.
253 */
254 static int oconfig_to_cgroups(oconfig_item_t *item, rdt_core_group_t *groups,
255 size_t max_groups, uint64_t max_core) {
256 int index = 0;
258 assert(groups != NULL);
259 assert(max_groups > 0);
260 assert(item != NULL);
262 for (int j = 0; j < item->values_num; j++) {
263 int ret;
264 size_t n;
265 uint64_t cores[RDT_MAX_CORES] = {0};
266 char value[DATA_MAX_NAME_LEN];
268 if ((item->values[j].value.string == NULL) ||
269 (strlen(item->values[j].value.string) == 0))
270 continue;
272 sstrncpy(value, item->values[j].value.string, sizeof(value));
274 n = strlisttonums(value, cores, STATIC_ARRAY_SIZE(cores));
275 if (n == 0) {
276 ERROR(RDT_PLUGIN ": Error parsing core group (%s)",
277 item->values[j].value.string);
278 return (-EINVAL);
279 }
281 for (int i = 0; i < n; i++) {
282 if (cores[i] > max_core) {
283 ERROR(RDT_PLUGIN ": Core group (%s) contains invalid core id (%d)",
284 item->values[j].value.string, (int)cores[i]);
285 return (-EINVAL);
286 }
287 }
289 /* set core group info */
290 ret = cgroup_set(&groups[index], item->values[j].value.string, cores, n);
291 if (ret < 0)
292 return ret;
294 index++;
296 if (index >= max_groups) {
297 WARNING(RDT_PLUGIN ": Too many core groups configured");
298 return index;
299 }
300 }
302 return index;
303 }
305 #if COLLECT_DEBUG
306 static void rdt_dump_cgroups(void) {
307 char cores[RDT_MAX_CORES * 4];
309 if (g_rdt == NULL)
310 return;
312 DEBUG(RDT_PLUGIN ": Core Groups Dump");
313 DEBUG(RDT_PLUGIN ": groups count: %zu", g_rdt->num_groups);
315 for (int i = 0; i < g_rdt->num_groups; i++) {
317 memset(cores, 0, sizeof(cores));
318 for (int j = 0; j < g_rdt->cgroups[i].num_cores; j++) {
319 snprintf(cores + strlen(cores), sizeof(cores) - strlen(cores) - 1, " %d",
320 g_rdt->cgroups[i].cores[j]);
321 }
323 DEBUG(RDT_PLUGIN ": group[%d]:", i);
324 DEBUG(RDT_PLUGIN ": description: %s", g_rdt->cgroups[i].desc);
325 DEBUG(RDT_PLUGIN ": cores: %s", cores);
326 DEBUG(RDT_PLUGIN ": events: 0x%X", g_rdt->cgroups[i].events);
327 }
329 return;
330 }
332 static inline double bytes_to_kb(const double bytes) { return bytes / 1024.0; }
334 static inline double bytes_to_mb(const double bytes) {
335 return bytes / (1024.0 * 1024.0);
336 }
338 static void rdt_dump_data(void) {
339 /*
340 * CORE - monitored group of cores
341 * RMID - Resource Monitoring ID associated with the monitored group
342 * LLC - last level cache occupancy
343 * MBL - local memory bandwidth
344 * MBR - remote memory bandwidth
345 */
346 DEBUG(" CORE RMID LLC[KB] MBL[MB] MBR[MB]");
347 for (int i = 0; i < g_rdt->num_groups; i++) {
349 const struct pqos_event_values *pv = &g_rdt->pgroups[i]->values;
351 double llc = bytes_to_kb(pv->llc);
352 double mbr = bytes_to_mb(pv->mbm_remote_delta);
353 double mbl = bytes_to_mb(pv->mbm_local_delta);
355 DEBUG(" [%s] %8u %10.1f %10.1f %10.1f", g_rdt->cgroups[i].desc,
356 g_rdt->pgroups[i]->poll_ctx[0].rmid, llc, mbl, mbr);
357 }
358 }
359 #endif /* COLLECT_DEBUG */
361 static void rdt_free_cgroups(void) {
362 for (int i = 0; i < RDT_MAX_CORES; i++) {
363 sfree(g_rdt->cgroups[i].desc);
365 sfree(g_rdt->cgroups[i].cores);
366 g_rdt->cgroups[i].num_cores = 0;
368 sfree(g_rdt->pgroups[i]);
369 }
370 }
372 static int rdt_default_cgroups(void) {
373 int ret;
375 /* configure each core in separate group */
376 for (unsigned i = 0; i < g_rdt->pqos_cpu->num_cores; i++) {
377 char desc[DATA_MAX_NAME_LEN];
378 uint64_t core = i;
380 ssnprintf(desc, sizeof(desc), "%d", g_rdt->pqos_cpu->cores[i].lcore);
382 /* set core group info */
383 ret = cgroup_set(&g_rdt->cgroups[i], desc, &core, 1);
384 if (ret < 0)
385 return ret;
386 }
388 return g_rdt->pqos_cpu->num_cores;
389 }
391 static int rdt_config_cgroups(oconfig_item_t *item) {
392 int n = 0;
393 enum pqos_mon_event events = 0;
395 if (item == NULL) {
396 DEBUG(RDT_PLUGIN ": cgroups_config: Invalid argument.");
397 return (-EINVAL);
398 }
400 DEBUG(RDT_PLUGIN ": Core groups [%d]:", item->values_num);
401 for (int j = 0; j < item->values_num; j++) {
402 if (item->values[j].type != OCONFIG_TYPE_STRING) {
403 ERROR(RDT_PLUGIN ": given core group value is not a string [idx=%d]", j);
404 return (-EINVAL);
405 }
406 DEBUG(RDT_PLUGIN ": [%d]: %s", j, item->values[j].value.string);
407 }
409 n = oconfig_to_cgroups(item, g_rdt->cgroups, RDT_MAX_CORES,
410 g_rdt->pqos_cpu->num_cores - 1);
411 if (n < 0) {
412 rdt_free_cgroups();
413 ERROR(RDT_PLUGIN ": Error parsing core groups configuration.");
414 return (-EINVAL);
415 }
417 if (n == 0) {
418 /* create default core groups if "Cores" config option is empty */
419 n = rdt_default_cgroups();
420 if (n < 0) {
421 rdt_free_cgroups();
422 ERROR(RDT_PLUGIN ": Error creating default core groups configuration.");
423 return n;
424 }
425 INFO(RDT_PLUGIN
426 ": No core groups configured. Default core groups created.");
427 }
429 /* Get all available events on this platform */
430 for (int i = 0; i < g_rdt->cap_mon->u.mon->num_events; i++)
431 events |= g_rdt->cap_mon->u.mon->events[i].type;
433 events &= ~(PQOS_PERF_EVENT_LLC_MISS);
435 DEBUG(RDT_PLUGIN ": Number of cores in the system: %u",
436 g_rdt->pqos_cpu->num_cores);
437 DEBUG(RDT_PLUGIN ": Available events to monitor: %#x", events);
439 g_rdt->num_groups = n;
440 for (int i = 0; i < n; i++) {
441 for (int j = 0; j < i; j++) {
442 int found = 0;
443 found = cgroup_cmp(&g_rdt->cgroups[j], &g_rdt->cgroups[i]);
444 if (found != 0) {
445 rdt_free_cgroups();
446 ERROR(RDT_PLUGIN ": Cannot monitor same cores in different groups.");
447 return (-EINVAL);
448 }
449 }
451 g_rdt->cgroups[i].events = events;
452 g_rdt->pgroups[i] = calloc(1, sizeof(*g_rdt->pgroups[i]));
453 if (g_rdt->pgroups[i] == NULL) {
454 rdt_free_cgroups();
455 ERROR(RDT_PLUGIN ": Failed to allocate memory for monitoring data.");
456 return (-ENOMEM);
457 }
458 }
460 return (0);
461 }
463 static int rdt_preinit(void) {
464 int ret;
466 if (g_rdt != NULL) {
467 /* already initialized if config callback was called before init callback */
468 return (0);
469 }
471 g_rdt = calloc(1, sizeof(*g_rdt));
472 if (g_rdt == NULL) {
473 ERROR(RDT_PLUGIN ": Failed to allocate memory for rdt context.");
474 return (-ENOMEM);
475 }
477 /* In case previous instance of the application was not closed properly
478 * call fini and ignore return code. */
479 pqos_fini();
481 /* TODO:
482 * stdout should not be used here. Will be reworked when support of log
483 * callback is added to PQoS library.
484 */
485 ret = pqos_init(&(struct pqos_config){.fd_log = STDOUT_FILENO});
486 if (ret != PQOS_RETVAL_OK) {
487 ERROR(RDT_PLUGIN ": Error initializing PQoS library!");
488 goto rdt_preinit_error1;
489 }
491 ret = pqos_cap_get(&g_rdt->pqos_cap, &g_rdt->pqos_cpu);
492 if (ret != PQOS_RETVAL_OK) {
493 ERROR(RDT_PLUGIN ": Error retrieving PQoS capabilities.");
494 goto rdt_preinit_error2;
495 }
497 ret = pqos_cap_get_type(g_rdt->pqos_cap, PQOS_CAP_TYPE_MON, &g_rdt->cap_mon);
498 if (ret == PQOS_RETVAL_PARAM) {
499 ERROR(RDT_PLUGIN ": Error retrieving monitoring capabilities.");
500 goto rdt_preinit_error2;
501 }
503 if (g_rdt->cap_mon == NULL) {
504 ERROR(
505 RDT_PLUGIN
506 ": Monitoring capability not detected. Nothing to do for the plugin.");
507 goto rdt_preinit_error2;
508 }
510 return (0);
512 rdt_preinit_error2:
513 pqos_fini();
515 rdt_preinit_error1:
517 sfree(g_rdt);
519 return (-1);
520 }
522 static int rdt_config(oconfig_item_t *ci) {
523 int ret = 0;
525 ret = rdt_preinit();
526 if (ret != 0)
527 return ret;
529 for (int i = 0; i < ci->children_num; i++) {
530 oconfig_item_t *child = ci->children + i;
532 if (strcasecmp("Cores", child->key) == 0) {
534 ret = rdt_config_cgroups(child);
535 if (ret != 0)
536 return ret;
538 #if COLLECT_DEBUG
539 rdt_dump_cgroups();
540 #endif /* COLLECT_DEBUG */
542 } else {
543 ERROR(RDT_PLUGIN ": Unknown configuration parameter \"%s\".", child->key);
544 }
545 }
547 return (0);
548 }
550 static void rdt_submit_derive(char *cgroup, char *type, char *type_instance,
551 derive_t value) {
552 value_list_t vl = VALUE_LIST_INIT;
554 vl.values = &(value_t){.derive = value};
555 vl.values_len = 1;
557 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
558 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
559 sstrncpy(vl.type, type, sizeof(vl.type));
560 if (type_instance)
561 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
563 plugin_dispatch_values(&vl);
564 }
566 static void rdt_submit_gauge(char *cgroup, char *type, char *type_instance,
567 gauge_t value) {
568 value_list_t vl = VALUE_LIST_INIT;
570 vl.values = &(value_t){.gauge = value};
571 vl.values_len = 1;
573 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
574 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
575 sstrncpy(vl.type, type, sizeof(vl.type));
576 if (type_instance)
577 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
579 plugin_dispatch_values(&vl);
580 }
582 static int rdt_read(__attribute__((unused)) user_data_t *ud) {
583 int ret;
585 if (g_rdt == NULL) {
586 ERROR(RDT_PLUGIN ": rdt_read: plugin not initialized.");
587 return (-EINVAL);
588 }
590 ret = pqos_mon_poll(&g_rdt->pgroups[0], (unsigned)g_rdt->num_groups);
591 if (ret != PQOS_RETVAL_OK) {
592 ERROR(RDT_PLUGIN ": Failed to poll monitoring data.");
593 return (-1);
594 }
596 #if COLLECT_DEBUG
597 rdt_dump_data();
598 #endif /* COLLECT_DEBUG */
600 for (int i = 0; i < g_rdt->num_groups; i++) {
601 enum pqos_mon_event mbm_events =
602 (PQOS_MON_EVENT_LMEM_BW | PQOS_MON_EVENT_TMEM_BW |
603 PQOS_MON_EVENT_RMEM_BW);
605 const struct pqos_event_values *pv = &g_rdt->pgroups[i]->values;
607 /* Submit only monitored events data */
609 if (g_rdt->cgroups[i].events & PQOS_MON_EVENT_L3_OCCUP)
610 rdt_submit_gauge(g_rdt->cgroups[i].desc, "bytes", "llc", pv->llc);
612 if (g_rdt->cgroups[i].events & PQOS_PERF_EVENT_IPC)
613 rdt_submit_gauge(g_rdt->cgroups[i].desc, "ipc", NULL, pv->ipc);
615 if (g_rdt->cgroups[i].events & mbm_events) {
616 rdt_submit_derive(g_rdt->cgroups[i].desc, "memory_bandwidth", "local",
617 pv->mbm_local_delta);
618 rdt_submit_derive(g_rdt->cgroups[i].desc, "memory_bandwidth", "remote",
619 pv->mbm_remote_delta);
620 }
621 }
623 return (0);
624 }
626 static int rdt_init(void) {
627 int ret;
629 ret = rdt_preinit();
630 if (ret != 0)
631 return ret;
633 /* Start monitoring */
634 for (int i = 0; i < g_rdt->num_groups; i++) {
635 rdt_core_group_t *cg = &g_rdt->cgroups[i];
637 ret = pqos_mon_start(cg->num_cores, cg->cores, cg->events, (void *)cg->desc,
638 g_rdt->pgroups[i]);
640 if (ret != PQOS_RETVAL_OK)
641 ERROR(RDT_PLUGIN ": Error starting monitoring group %s (pqos status=%d)",
642 cg->desc, ret);
643 }
645 return (0);
646 }
648 static int rdt_shutdown(void) {
649 int ret;
651 DEBUG(RDT_PLUGIN ": rdt_shutdown.");
653 if (g_rdt == NULL)
654 return (0);
656 /* Stop monitoring */
657 for (int i = 0; i < g_rdt->num_groups; i++) {
658 pqos_mon_stop(g_rdt->pgroups[i]);
659 }
661 ret = pqos_fini();
662 if (ret != PQOS_RETVAL_OK)
663 ERROR(RDT_PLUGIN ": Error shutting down PQoS library.");
665 rdt_free_cgroups();
666 sfree(g_rdt);
668 return (0);
669 }
671 void module_register(void) {
672 plugin_register_init(RDT_PLUGIN, rdt_init);
673 plugin_register_complex_config(RDT_PLUGIN, rdt_config);
674 plugin_register_complex_read(NULL, RDT_PLUGIN, rdt_read, 0, NULL);
675 plugin_register_shutdown(RDT_PLUGIN, rdt_shutdown);
676 }