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 "common.h"
29 #include "collectd.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 typedef enum {
40 UNKNOWN = 0,
41 CONFIGURATION_ERROR,
42 } rdt_config_status;
44 struct rdt_core_group_s {
45 char *desc;
46 size_t num_cores;
47 unsigned *cores;
48 enum pqos_mon_event events;
49 };
50 typedef struct rdt_core_group_s rdt_core_group_t;
52 struct rdt_ctx_s {
53 rdt_core_group_t cgroups[RDT_MAX_CORES];
54 struct pqos_mon_data *pgroups[RDT_MAX_CORES];
55 size_t num_groups;
56 const struct pqos_cpuinfo *pqos_cpu;
57 const struct pqos_cap *pqos_cap;
58 const struct pqos_capability *cap_mon;
59 };
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++)
68 if (nums[i] == val)
69 return 1;
70 return 0;
71 }
73 static int strtouint64(const char *s, uint64_t *n) {
74 char *endptr = NULL;
76 assert(s != NULL);
77 assert(n != NULL);
79 *n = strtoull(s, &endptr, 0);
81 if (!(*s != '\0' && *endptr == '\0')) {
82 DEBUG(RDT_PLUGIN ": Error converting '%s' to unsigned number.", s);
83 return (-EINVAL);
84 }
86 return (0);
87 }
89 /*
90 * NAME
91 * strlisttonums
92 *
93 * DESCRIPTION
94 * Converts string of characters representing list of numbers into array of
95 * numbers. Allowed formats are:
96 * 0,1,2,3
97 * 0-10,20-18
98 * 1,3,5-8,10,0x10-12
99 *
100 * Numbers can be in decimal or hexadecimal format.
101 *
102 * PARAMETERS
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.
106 *
107 * RETURN VALUE
108 * Number of elements placed into nums.
109 */
110 static size_t strlisttonums(char *s, uint64_t *nums, size_t max) {
111 int ret;
112 size_t index = 0;
113 char *saveptr = NULL;
115 if (s == NULL || nums == NULL || max == 0)
116 return index;
118 for (;;) {
119 char *p = NULL;
120 char *token = NULL;
122 token = strtok_r(s, ",", &saveptr);
123 if (token == NULL)
124 break;
126 s = NULL;
128 while (isspace(*token))
129 token++;
130 if (*token == '\0')
131 continue;
133 p = strchr(token, '-');
134 if (p != NULL) {
135 uint64_t n, start, end;
136 *p = '\0';
137 ret = strtouint64(token, &start);
138 if (ret < 0)
139 return (0);
140 ret = strtouint64(p + 1, &end);
141 if (ret < 0)
142 return (0);
143 if (start > end) {
144 return (0);
145 }
146 for (n = start; n <= end; n++) {
147 if (!(isdup(nums, index, n))) {
148 nums[index] = n;
149 index++;
150 }
151 if (index >= max)
152 return index;
153 }
154 } else {
155 uint64_t val;
157 ret = strtouint64(token, &val);
158 if (ret < 0)
159 return (0);
161 if (!(isdup(nums, index, val))) {
162 nums[index] = val;
163 index++;
164 }
165 if (index >= max)
166 return index;
167 }
168 }
170 return index;
171 }
173 /*
174 * NAME
175 * cgroup_cmp
176 *
177 * DESCRIPTION
178 * Function to compare cores in 2 core groups.
179 *
180 * PARAMETERS
181 * `cg_a' Pointer to core group a.
182 * `cg_b' Pointer to core group b.
183 *
184 * RETURN VALUE
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
188 */
189 static int cgroup_cmp(const rdt_core_group_t *cg_a,
190 const rdt_core_group_t *cg_b) {
191 int found = 0;
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])
204 found++;
205 }
206 /* if no cores are the same */
207 if (!found)
208 return 0;
209 /* if group contains same cores */
210 if (sz_a == sz_b && sz_b == found)
211 return 1;
212 /* if not all cores are the same */
213 return -1;
214 }
216 static int cgroup_set(rdt_core_group_t *cg, char *desc, uint64_t *cores,
217 size_t num_cores) {
218 assert(cg != NULL);
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");
226 return (-ENOMEM);
227 }
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");
232 sfree(cg->cores);
233 return (-ENOMEM);
234 }
236 for (size_t i = 0; i < num_cores; i++)
237 cg->cores[i] = (unsigned)cores[i];
239 return 0;
240 }
242 /*
243 * NAME
244 * oconfig_to_cgroups
245 *
246 * DESCRIPTION
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
249 * core group values.
250 *
251 * PARAMETERS
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.
256 *
257 * RETURN VALUE
258 * On success, the number of core groups set up. On error, appropriate
259 * negative error value.
260 */
261 static int oconfig_to_cgroups(oconfig_item_t *item, rdt_core_group_t *groups,
262 size_t max_groups, uint64_t max_core) {
263 int index = 0;
265 assert(groups != NULL);
266 assert(max_groups > 0);
267 assert(item != NULL);
269 for (int j = 0; j < item->values_num; j++) {
270 int ret;
271 size_t n;
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))
277 continue;
279 sstrncpy(value, item->values[j].value.string, sizeof(value));
281 n = strlisttonums(value, cores, STATIC_ARRAY_SIZE(cores));
282 if (n == 0) {
283 ERROR(RDT_PLUGIN ": Error parsing core group (%s)",
284 item->values[j].value.string);
285 return (-EINVAL);
286 }
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]);
292 return (-EINVAL);
293 }
294 }
296 /* set core group info */
297 ret = cgroup_set(&groups[index], item->values[j].value.string, cores, n);
298 if (ret < 0)
299 return ret;
301 index++;
303 if (index >= max_groups) {
304 WARNING(RDT_PLUGIN ": Too many core groups configured");
305 return index;
306 }
307 }
309 return index;
310 }
312 #if COLLECT_DEBUG
313 static void rdt_dump_cgroups(void) {
314 char cores[RDT_MAX_CORES * 4];
316 if (g_rdt == NULL)
317 return;
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]);
328 }
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);
334 }
336 return;
337 }
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);
343 }
345 static void rdt_dump_data(void) {
346 /*
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
352 */
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);
364 }
365 }
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]);
376 }
377 }
379 static int rdt_default_cgroups(void) {
380 int ret;
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];
385 uint64_t core = i;
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);
391 if (ret < 0)
392 return ret;
393 }
395 return g_rdt->pqos_cpu->num_cores;
396 }
398 static int rdt_config_cgroups(oconfig_item_t *item) {
399 int n = 0;
400 enum pqos_mon_event events = 0;
402 if (item == NULL) {
403 DEBUG(RDT_PLUGIN ": cgroups_config: Invalid argument.");
404 return (-EINVAL);
405 }
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);
411 return (-EINVAL);
412 }
413 DEBUG(RDT_PLUGIN ": [%d]: %s", j, item->values[j].value.string);
414 }
416 n = oconfig_to_cgroups(item, g_rdt->cgroups, RDT_MAX_CORES,
417 g_rdt->pqos_cpu->num_cores - 1);
418 if (n < 0) {
419 rdt_free_cgroups();
420 ERROR(RDT_PLUGIN ": Error parsing core groups configuration.");
421 return (-EINVAL);
422 }
424 if (n == 0) {
425 /* create default core groups if "Cores" config option is empty */
426 n = rdt_default_cgroups();
427 if (n < 0) {
428 rdt_free_cgroups();
429 ERROR(RDT_PLUGIN ": Error creating default core groups configuration.");
430 return n;
431 }
432 INFO(RDT_PLUGIN
433 ": No core groups configured. Default core groups created.");
434 }
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++) {
449 int found = 0;
450 found = cgroup_cmp(&g_rdt->cgroups[j], &g_rdt->cgroups[i]);
451 if (found != 0) {
452 rdt_free_cgroups();
453 ERROR(RDT_PLUGIN ": Cannot monitor same cores in different groups.");
454 return (-EINVAL);
455 }
456 }
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) {
461 rdt_free_cgroups();
462 ERROR(RDT_PLUGIN ": Failed to allocate memory for monitoring data.");
463 return (-ENOMEM);
464 }
465 }
467 return (0);
468 }
470 static int rdt_preinit(void) {
471 int ret;
473 if (g_rdt != NULL) {
474 /* already initialized if config callback was called before init callback */
475 return (0);
476 }
478 g_rdt = calloc(1, sizeof(*g_rdt));
479 if (g_rdt == NULL) {
480 ERROR(RDT_PLUGIN ": Failed to allocate memory for rdt context.");
481 return (-ENOMEM);
482 }
484 /* In case previous instance of the application was not closed properly
485 * call fini and ignore return code. */
486 pqos_fini();
488 /* TODO:
489 * stdout should not be used here. Will be reworked when support of log
490 * callback is added to PQoS library.
491 */
492 ret = pqos_init(&(struct pqos_config){.fd_log = STDOUT_FILENO});
493 if (ret != PQOS_RETVAL_OK) {
494 ERROR(RDT_PLUGIN ": Error initializing PQoS library!");
495 goto rdt_preinit_error1;
496 }
498 ret = pqos_cap_get(&g_rdt->pqos_cap, &g_rdt->pqos_cpu);
499 if (ret != PQOS_RETVAL_OK) {
500 ERROR(RDT_PLUGIN ": Error retrieving PQoS capabilities.");
501 goto rdt_preinit_error2;
502 }
504 ret = pqos_cap_get_type(g_rdt->pqos_cap, PQOS_CAP_TYPE_MON, &g_rdt->cap_mon);
505 if (ret == PQOS_RETVAL_PARAM) {
506 ERROR(RDT_PLUGIN ": Error retrieving monitoring capabilities.");
507 goto rdt_preinit_error2;
508 }
510 if (g_rdt->cap_mon == NULL) {
511 ERROR(
512 RDT_PLUGIN
513 ": Monitoring capability not detected. Nothing to do for the plugin.");
514 goto rdt_preinit_error2;
515 }
517 return (0);
519 rdt_preinit_error2:
520 pqos_fini();
522 rdt_preinit_error1:
524 sfree(g_rdt);
526 return (-1);
527 }
529 static int rdt_config(oconfig_item_t *ci) {
530 int ret = 0;
532 ret = rdt_preinit();
533 if (ret != 0) {
534 g_state = CONFIGURATION_ERROR;
535 /* if we return -1 at this point collectd
536 reports a failure in configuration and
537 aborts
538 */
539 goto exit;
540 }
542 for (int i = 0; i < ci->children_num; i++) {
543 oconfig_item_t *child = ci->children + i;
545 if (strcasecmp("Cores", child->key) == 0) {
547 ret = rdt_config_cgroups(child);
548 if (ret != 0) {
549 g_state = CONFIGURATION_ERROR;
550 /* if we return -1 at this point collectd
551 reports a failure in configuration and
552 aborts
553 */
554 goto exit;
555 }
557 #if COLLECT_DEBUG
558 rdt_dump_cgroups();
559 #endif /* COLLECT_DEBUG */
561 } else {
562 ERROR(RDT_PLUGIN ": Unknown configuration parameter \"%s\".", child->key);
563 }
564 }
566 exit:
567 return (0);
568 }
570 static void rdt_submit_derive(char *cgroup, char *type, char *type_instance,
571 derive_t value) {
572 value_list_t vl = VALUE_LIST_INIT;
574 vl.values = &(value_t){.derive = value};
575 vl.values_len = 1;
577 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
578 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
579 sstrncpy(vl.type, type, sizeof(vl.type));
580 if (type_instance)
581 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
583 plugin_dispatch_values(&vl);
584 }
586 static void rdt_submit_gauge(char *cgroup, char *type, char *type_instance,
587 gauge_t value) {
588 value_list_t vl = VALUE_LIST_INIT;
590 vl.values = &(value_t){.gauge = value};
591 vl.values_len = 1;
593 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
594 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
595 sstrncpy(vl.type, type, sizeof(vl.type));
596 if (type_instance)
597 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
599 plugin_dispatch_values(&vl);
600 }
602 static int rdt_read(__attribute__((unused)) user_data_t *ud) {
603 int ret;
605 if (g_rdt == NULL) {
606 ERROR(RDT_PLUGIN ": rdt_read: plugin not initialized.");
607 return (-EINVAL);
608 }
610 ret = pqos_mon_poll(&g_rdt->pgroups[0], (unsigned)g_rdt->num_groups);
611 if (ret != PQOS_RETVAL_OK) {
612 ERROR(RDT_PLUGIN ": Failed to poll monitoring data.");
613 return (-1);
614 }
616 #if COLLECT_DEBUG
617 rdt_dump_data();
618 #endif /* COLLECT_DEBUG */
620 for (int i = 0; i < g_rdt->num_groups; i++) {
621 enum pqos_mon_event mbm_events =
622 (PQOS_MON_EVENT_LMEM_BW | PQOS_MON_EVENT_TMEM_BW |
623 PQOS_MON_EVENT_RMEM_BW);
625 const struct pqos_event_values *pv = &g_rdt->pgroups[i]->values;
627 /* Submit only monitored events data */
629 if (g_rdt->cgroups[i].events & PQOS_MON_EVENT_L3_OCCUP)
630 rdt_submit_gauge(g_rdt->cgroups[i].desc, "bytes", "llc", pv->llc);
632 if (g_rdt->cgroups[i].events & PQOS_PERF_EVENT_IPC)
633 rdt_submit_gauge(g_rdt->cgroups[i].desc, "ipc", NULL, pv->ipc);
635 if (g_rdt->cgroups[i].events & mbm_events) {
636 rdt_submit_derive(g_rdt->cgroups[i].desc, "memory_bandwidth", "local",
637 pv->mbm_local_delta);
638 rdt_submit_derive(g_rdt->cgroups[i].desc, "memory_bandwidth", "remote",
639 pv->mbm_remote_delta);
640 }
641 }
643 return (0);
644 }
646 static int rdt_init(void) {
647 int ret;
649 if(g_state == CONFIGURATION_ERROR)
650 return (-1);
652 ret = rdt_preinit();
653 if (ret != 0)
654 return ret;
656 /* Start monitoring */
657 for (int i = 0; i < g_rdt->num_groups; i++) {
658 rdt_core_group_t *cg = &g_rdt->cgroups[i];
660 ret = pqos_mon_start(cg->num_cores, cg->cores, cg->events, (void *)cg->desc,
661 g_rdt->pgroups[i]);
663 if (ret != PQOS_RETVAL_OK)
664 ERROR(RDT_PLUGIN ": Error starting monitoring group %s (pqos status=%d)",
665 cg->desc, ret);
666 }
668 return (0);
669 }
671 static int rdt_shutdown(void) {
672 int ret;
674 DEBUG(RDT_PLUGIN ": rdt_shutdown.");
676 if (g_rdt == NULL)
677 return (0);
679 /* Stop monitoring */
680 for (int i = 0; i < g_rdt->num_groups; i++) {
681 pqos_mon_stop(g_rdt->pgroups[i]);
682 }
684 ret = pqos_fini();
685 if (ret != PQOS_RETVAL_OK)
686 ERROR(RDT_PLUGIN ": Error shutting down PQoS library.");
688 rdt_free_cgroups();
689 sfree(g_rdt);
691 return (0);
692 }
694 void module_register(void) {
695 plugin_register_init(RDT_PLUGIN, rdt_init);
696 plugin_register_complex_config(RDT_PLUGIN, rdt_config);
697 plugin_register_complex_read(NULL, RDT_PLUGIN, rdt_read, 0, NULL);
698 plugin_register_shutdown(RDT_PLUGIN, rdt_shutdown);
699 }