2 * collectd - src/intel_rdt.c
4 * Copyright(c) 2016-2018 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>
26 * Starzyk, Mateusz <mateuszx.starzyk@intel.com>
27 * Wojciech Andralojc <wojciechx.andralojc@intel.com>
31 #include "utils/common/common.h"
32 #include "utils/config_cores/config_cores.h"
35 #define RDT_PLUGIN "intel_rdt"
37 /* libpqos v2.0 or newer is required for process monitoring*/
39 #if defined(PQOS_VERSION) && PQOS_VERSION >= 20000
43 #define RDT_PLUGIN "intel_rdt"
45 #define RDT_MAX_SOCKETS 8
46 #define RDT_MAX_SOCKET_CORES 64
47 #define RDT_MAX_CORES (RDT_MAX_SOCKET_CORES * RDT_MAX_SOCKETS)
51 * Process name inside comm file is limited to 16 chars.
52 * More info here: http://man7.org/linux/man-pages/man5/proc.5.html
54 #define RDT_MAX_NAME_LEN 16
55 #define RDT_MAX_NAMES_GROUPS 64
57 #define RDT_PROC_PATH "/proc"
66 /* Helper typedef for process name array
67 * Extra 1 char is added for string null termination.
69 typedef char proc_comm_t[RDT_MAX_NAME_LEN + 1];
71 /* Linked one-way list of pids. */
72 typedef struct pids_list_s {
74 struct pids_list_s *next;
77 /* Holds process name and list of pids assigned to that name */
78 typedef struct proc_pids_s {
79 proc_comm_t proccess_name;
83 struct rdt_name_group_s {
87 proc_pids_t *proc_pids_array;
88 size_t monitored_pids_count;
89 enum pqos_mon_event events;
91 typedef struct rdt_name_group_s rdt_name_group_t;
95 core_groups_list_t cores;
96 enum pqos_mon_event events[RDT_MAX_CORES];
97 struct pqos_mon_data *pcgroups[RDT_MAX_CORES];
99 rdt_name_group_t ngroups[RDT_MAX_NAMES_GROUPS];
100 struct pqos_mon_data *pngroups[RDT_MAX_NAMES_GROUPS];
102 #endif /* LIBPQOS2 */
103 const struct pqos_cpuinfo *pqos_cpu;
104 const struct pqos_cap *pqos_cap;
105 const struct pqos_capability *cap_mon;
107 typedef struct rdt_ctx_s rdt_ctx_t;
109 static rdt_ctx_t *g_rdt;
111 static rdt_config_status g_state = UNKNOWN;
113 static int g_interface = -1;
116 static int isdupstr(const char *names[], const size_t size, const char *name) {
117 for (size_t i = 0; i < size; i++)
118 if (strncmp(names[i], name, (size_t)RDT_MAX_NAME_LEN) == 0)
129 * Converts string representing list of strings into array of strings.
131 * name,name1,name2,name3
134 * `str_list' String representing list of strings.
135 * `names' Array to put extracted strings into.
136 * `names_num' Variable to put number of extracted strings.
139 * Number of elements placed into names.
141 static int strlisttoarray(char *str_list, char ***names, size_t *names_num) {
142 char *saveptr = NULL;
144 if (str_list == NULL || names == NULL)
148 char *token = strtok_r(str_list, ",", &saveptr);
154 while (isspace(*token))
160 if (!(isdupstr((const char **)*names, *names_num, token)))
161 if (0 != strarray_add(names, names_num, token)) {
162 ERROR(RDT_PLUGIN ": Error allocating process name string");
175 * Function to compare names in two name groups.
178 * `ng_a' Pointer to name group a.
179 * `ng_b' Pointer to name group b.
182 * 1 if both groups contain the same names
183 * 0 if none of their names match
184 * -1 if some but not all names match
186 static int ngroup_cmp(const rdt_name_group_t *ng_a,
187 const rdt_name_group_t *ng_b) {
190 assert(ng_a != NULL);
191 assert(ng_b != NULL);
193 const size_t sz_a = (unsigned)ng_a->num_names;
194 const size_t sz_b = (unsigned)ng_b->num_names;
195 const char **tab_a = (const char **)ng_a->names;
196 const char **tab_b = (const char **)ng_b->names;
198 for (size_t i = 0; i < sz_a; i++) {
199 for (size_t j = 0; j < sz_b; j++)
200 if (strncmp(tab_a[i], tab_b[j], (size_t)RDT_MAX_NAME_LEN) == 0)
203 /* if no names are the same */
206 /* if group contains same names */
207 if (sz_a == sz_b && sz_b == (size_t)found)
209 /* if not all names are the same */
218 * Function to set the descriptions and names for each process names group.
219 * Takes a config option containing list of strings that are used to set
220 * process group values.
223 * `item' Config option containing process names groups.
224 * `groups' Table of process name groups to set values in.
225 * `max_groups' Maximum number of process name groups allowed.
228 * On success, the number of name groups set up. On error, appropriate
229 * negative error value.
231 static int oconfig_to_ngroups(const oconfig_item_t *item,
232 rdt_name_group_t *groups,
233 const size_t max_groups) {
236 assert(groups != NULL);
237 assert(max_groups > 0);
238 assert(item != NULL);
240 for (int j = 0; j < item->values_num; j++) {
242 char value[DATA_MAX_NAME_LEN];
244 if ((item->values[j].value.string == NULL) ||
245 (strlen(item->values[j].value.string) == 0))
248 sstrncpy(value, item->values[j].value.string, sizeof(value));
250 ret = strlisttoarray(value, &groups[index].names, &groups[index].num_names);
251 if (ret != 0 || groups[index].num_names == 0) {
252 ERROR(RDT_PLUGIN ": Error parsing process names group (%s)",
253 item->values[j].value.string);
257 /* set group description info */
258 groups[index].desc = sstrdup(item->values[j].value.string);
259 if (groups[index].desc == NULL) {
260 ERROR(RDT_PLUGIN ": Error allocating name group description");
264 groups[index].proc_pids_array = NULL;
265 groups[index].monitored_pids_count = 0;
269 if (index >= (const int)max_groups) {
270 WARNING(RDT_PLUGIN ": Too many process names groups configured");
277 #endif /* LIBPQOS2 */
280 static void rdt_dump_cgroups(void) {
281 char cores[RDT_MAX_CORES * 4];
286 DEBUG(RDT_PLUGIN ": Core Groups Dump");
287 DEBUG(RDT_PLUGIN ": groups count: %" PRIsz, g_rdt->cores.num_cgroups);
289 for (size_t i = 0; i < g_rdt->cores.num_cgroups; i++) {
290 core_group_t *cgroup = g_rdt->cores.cgroups + i;
292 memset(cores, 0, sizeof(cores));
293 for (size_t j = 0; j < cgroup->num_cores; j++) {
294 snprintf(cores + strlen(cores), sizeof(cores) - strlen(cores) - 1, " %d",
298 DEBUG(RDT_PLUGIN ": group[%zu]:", i);
299 DEBUG(RDT_PLUGIN ": description: %s", cgroup->desc);
300 DEBUG(RDT_PLUGIN ": cores: %s", cores);
301 DEBUG(RDT_PLUGIN ": events: 0x%X", g_rdt->events[i]);
308 static void rdt_dump_ngroups(void) {
310 char names[DATA_MAX_NAME_LEN];
315 DEBUG(RDT_PLUGIN ": Process Names Groups Dump");
316 DEBUG(RDT_PLUGIN ": groups count: %" PRIsz, g_rdt->num_ngroups);
318 for (size_t i = 0; i < g_rdt->num_ngroups; i++) {
319 memset(names, 0, sizeof(names));
320 for (size_t j = 0; j < g_rdt->ngroups[i].num_names; j++)
321 snprintf(names + strlen(names), sizeof(names) - strlen(names) - 1, " %s",
322 g_rdt->ngroups[i].names[j]);
324 DEBUG(RDT_PLUGIN ": group[%d]:", (int)i);
325 DEBUG(RDT_PLUGIN ": description: %s", g_rdt->ngroups[i].desc);
326 DEBUG(RDT_PLUGIN ": process names:%s", names);
327 DEBUG(RDT_PLUGIN ": events: 0x%X", g_rdt->ngroups[i].events);
332 #endif /* LIBPQOS2 */
334 static inline double bytes_to_kb(const double bytes) { return bytes / 1024.0; }
336 static inline double bytes_to_mb(const double bytes) {
337 return bytes / (1024.0 * 1024.0);
340 static void rdt_dump_cores_data(void) {
342 * CORE - monitored group of cores
343 * RMID - Resource Monitoring ID associated with the monitored group
344 * This is not available for monitoring with resource control
345 * LLC - last level cache occupancy
346 * MBL - local memory bandwidth
347 * MBR - remote memory bandwidth
350 if (g_interface == PQOS_INTER_OS_RESCTRL_MON) {
351 DEBUG(RDT_PLUGIN ": CORE LLC[KB] MBL[MB] MBR[MB]");
353 DEBUG(RDT_PLUGIN ": CORE RMID LLC[KB] MBL[MB] MBR[MB]");
356 DEBUG(RDT_PLUGIN ": CORE RMID LLC[KB] MBL[MB] MBR[MB]");
357 #endif /* LIBPQOS2 */
359 for (size_t i = 0; i < g_rdt->cores.num_cgroups; i++) {
360 const struct pqos_event_values *pv = &g_rdt->pcgroups[i]->values;
362 double llc = bytes_to_kb(pv->llc);
363 double mbr = bytes_to_mb(pv->mbm_remote_delta);
364 double mbl = bytes_to_mb(pv->mbm_local_delta);
366 if (g_interface == PQOS_INTER_OS_RESCTRL_MON) {
367 DEBUG(RDT_PLUGIN ": [%s] %10.1f %10.1f %10.1f",
368 g_rdt->cores.cgroups[i].desc, llc, mbl, mbr);
370 DEBUG(RDT_PLUGIN ": [%s] %8u %10.1f %10.1f %10.1f",
371 g_rdt->cores.cgroups[i].desc, g_rdt->pcgroups[i]->poll_ctx[0].rmid,
375 DEBUG(RDT_PLUGIN ": [%s] %8u %10.1f %10.1f %10.1f",
376 g_rdt->cores.cgroups[i].desc, g_rdt->pcgroups[i]->poll_ctx[0].rmid,
378 #endif /* LIBPQOS2 */
383 static void rdt_dump_pids_data(void) {
385 * NAME - monitored group of processes
386 * PIDs - list of PID numbers in the NAME group
387 * LLC - last level cache occupancy
388 * MBL - local memory bandwidth
389 * MBR - remote memory bandwidth
392 DEBUG(RDT_PLUGIN ": NAME PIDs");
393 char pids[DATA_MAX_NAME_LEN];
394 for (size_t i = 0; i < g_rdt->num_ngroups; ++i) {
395 memset(pids, 0, sizeof(pids));
396 for (size_t j = 0; j < g_rdt->ngroups[i].num_names; ++j) {
397 pids_list_t *list = g_rdt->ngroups[i].proc_pids_array[j].pids;
398 while (list != NULL) {
399 snprintf(pids + strlen(pids), sizeof(pids) - strlen(pids) - 1, " %u",
404 DEBUG(RDT_PLUGIN ": [%s] %s", g_rdt->ngroups[i].desc, pids);
407 DEBUG(RDT_PLUGIN ": NAME LLC[KB] MBL[MB] MBR[MB]");
408 for (size_t i = 0; i < g_rdt->num_ngroups; i++) {
410 const struct pqos_event_values *pv = &g_rdt->pngroups[i]->values;
412 double llc = bytes_to_kb(pv->llc);
413 double mbr = bytes_to_mb(pv->mbm_remote_delta);
414 double mbl = bytes_to_mb(pv->mbm_local_delta);
416 DEBUG(RDT_PLUGIN ": [%s] %10.1f %10.1f %10.1f", g_rdt->ngroups[i].desc,
420 #endif /* LIBPQOS2 */
421 #endif /* COLLECT_DEBUG */
423 static void rdt_free_cgroups(void) {
424 config_cores_cleanup(&g_rdt->cores);
425 for (int i = 0; i < RDT_MAX_CORES; i++) {
426 sfree(g_rdt->pcgroups[i]);
431 static int pids_list_free(pids_list_t *list) {
434 pids_list_t *current = list;
435 while (current != NULL) {
436 pids_list_t *previous = current;
437 current = current->next;
443 static void rdt_free_ngroups(void) {
444 for (int i = 0; i < RDT_MAX_NAMES_GROUPS; i++) {
445 if (g_rdt->ngroups[i].desc)
446 DEBUG(RDT_PLUGIN ": Freeing pids \'%s\' group\'s data...",
447 g_rdt->ngroups[i].desc);
448 sfree(g_rdt->ngroups[i].desc);
449 strarray_free(g_rdt->ngroups[i].names, g_rdt->ngroups[i].num_names);
451 if (g_rdt->ngroups[i].proc_pids_array) {
452 for (size_t j = 0; j < g_rdt->ngroups[i].num_names; ++j) {
453 if (NULL == g_rdt->ngroups[i].proc_pids_array[j].pids)
455 pids_list_free(g_rdt->ngroups[i].proc_pids_array[j].pids);
458 sfree(g_rdt->ngroups[i].proc_pids_array);
461 g_rdt->ngroups[i].num_names = 0;
462 sfree(g_rdt->pngroups[i]);
465 #endif /* LIBPQOS2 */
467 static int rdt_default_cgroups(void) {
468 unsigned num_cores = g_rdt->pqos_cpu->num_cores;
470 g_rdt->cores.cgroups = calloc(num_cores, sizeof(*g_rdt->cores.cgroups));
471 if (g_rdt->cores.cgroups == NULL) {
472 ERROR(RDT_PLUGIN ": Error allocating core groups array");
475 g_rdt->cores.num_cgroups = num_cores;
477 /* configure each core in separate group */
478 for (unsigned i = 0; i < num_cores; i++) {
479 core_group_t *cgroup = g_rdt->cores.cgroups + i;
480 char desc[DATA_MAX_NAME_LEN];
482 /* set core group info */
483 cgroup->cores = calloc(1, sizeof(*cgroup->cores));
484 if (cgroup->cores == NULL) {
485 ERROR(RDT_PLUGIN ": Error allocating cores array");
489 cgroup->num_cores = 1;
490 cgroup->cores[0] = i;
492 snprintf(desc, sizeof(desc), "%d", g_rdt->pqos_cpu->cores[i].lcore);
493 cgroup->desc = strdup(desc);
494 if (cgroup->desc == NULL) {
495 ERROR(RDT_PLUGIN ": Error allocating core group description");
504 static int rdt_is_core_id_valid(unsigned int core_id) {
506 for (unsigned int i = 0; i < g_rdt->pqos_cpu->num_cores; i++)
507 if (core_id == g_rdt->pqos_cpu->cores[i].lcore)
514 static int rdt_is_proc_name_valid(const char *name) {
517 unsigned len = strlen(name);
518 if (len > 0 && len <= RDT_MAX_NAME_LEN)
522 ": Process name \'%s\' is too long. Max supported len is %d chars.",
523 name, RDT_MAX_NAME_LEN);
529 #endif /* LIBPQOS2 */
531 static int rdt_config_cgroups(oconfig_item_t *item) {
533 enum pqos_mon_event events = 0;
535 if (config_cores_parse(item, &g_rdt->cores) < 0) {
537 ERROR(RDT_PLUGIN ": Error parsing core groups configuration.");
540 n = g_rdt->cores.num_cgroups;
542 /* validate configured core id values */
543 for (size_t group_idx = 0; group_idx < n; group_idx++) {
544 core_group_t *cgroup = g_rdt->cores.cgroups + group_idx;
545 for (size_t core_idx = 0; core_idx < cgroup->num_cores; core_idx++) {
546 if (!rdt_is_core_id_valid(cgroup->cores[core_idx])) {
547 ERROR(RDT_PLUGIN ": Core group '%s' contains invalid core id '%u'",
548 cgroup->desc, cgroup->cores[core_idx]);
556 /* create default core groups if "Cores" config option is empty */
557 int ret = rdt_default_cgroups();
560 ERROR(RDT_PLUGIN ": Error creating default core groups configuration.");
565 ": No core groups configured. Default core groups created.");
568 /* Get all available events on this platform */
569 for (unsigned int i = 0; i < g_rdt->cap_mon->u.mon->num_events; i++)
570 events |= g_rdt->cap_mon->u.mon->events[i].type;
572 events &= ~(PQOS_PERF_EVENT_LLC_MISS);
574 DEBUG(RDT_PLUGIN ": Number of cores in the system: %u",
575 g_rdt->pqos_cpu->num_cores);
576 DEBUG(RDT_PLUGIN ": Available events to monitor: %#x", events);
578 g_rdt->cores.num_cgroups = n;
579 for (int i = 0; i < n; i++) {
580 for (int j = 0; j < i; j++) {
582 found = config_cores_cmp_cgroups(&g_rdt->cores.cgroups[j],
583 &g_rdt->cores.cgroups[i]);
586 ERROR(RDT_PLUGIN ": Cannot monitor same cores in different groups.");
591 g_rdt->events[i] = events;
592 g_rdt->pcgroups[i] = calloc(1, sizeof(*g_rdt->pcgroups[i]));
593 if (g_rdt->pcgroups[i] == NULL) {
595 ERROR(RDT_PLUGIN ": Failed to allocate memory for monitoring data.");
604 static int rdt_config_ngroups(const oconfig_item_t *item) {
606 enum pqos_mon_event events = 0;
609 DEBUG(RDT_PLUGIN ": ngroups_config: Invalid argument.");
613 DEBUG(RDT_PLUGIN ": Process names groups [%d]:", item->values_num);
614 for (int j = 0; j < item->values_num; j++) {
615 if (item->values[j].type != OCONFIG_TYPE_STRING) {
617 ": given process names group value is not a string [idx=%d]",
621 DEBUG(RDT_PLUGIN ": [%d]: %s", j, item->values[j].value.string);
624 n = oconfig_to_ngroups(item, g_rdt->ngroups, RDT_MAX_NAMES_GROUPS);
627 ERROR(RDT_PLUGIN ": Error parsing process name groups configuration.");
631 /* validate configured process name values */
632 for (int group_idx = 0; group_idx < n; group_idx++) {
633 for (size_t name_idx = 0; name_idx < g_rdt->ngroups[group_idx].num_names;
635 if (!rdt_is_proc_name_valid(g_rdt->ngroups[group_idx].names[name_idx])) {
636 ERROR(RDT_PLUGIN ": Process name group '%s' contains invalid name '%s'",
637 g_rdt->ngroups[group_idx].desc,
638 g_rdt->ngroups[group_idx].names[name_idx]);
646 ERROR(RDT_PLUGIN ": Empty process name groups configured.");
650 /* Get all available events on this platform */
651 for (unsigned i = 0; i < g_rdt->cap_mon->u.mon->num_events; i++)
652 events |= g_rdt->cap_mon->u.mon->events[i].type;
654 events &= ~(PQOS_PERF_EVENT_LLC_MISS);
656 DEBUG(RDT_PLUGIN ": Available events to monitor: %#x", events);
658 g_rdt->num_ngroups = n;
659 for (int i = 0; i < n; i++) {
660 for (int j = 0; j < i; j++) {
661 int found = ngroup_cmp(&g_rdt->ngroups[j], &g_rdt->ngroups[i]);
665 ": Cannot monitor same process name in different groups.");
670 g_rdt->ngroups[i].events = events;
671 g_rdt->pngroups[i] = calloc(1, sizeof(*g_rdt->pngroups[i]));
672 if (g_rdt->pngroups[i] == NULL) {
675 ": Failed to allocate memory for process name monitoring data.");
688 * Adds pid at the end of the pids list.
689 * Allocates memory for new pid element, it is up to user to free it.
692 * `list' Head of target pids_list.
693 * `pid' Pid to be added.
696 * On success, returns 0.
697 * -1 on memory allocation error.
699 static int pids_list_add_pid(pids_list_t **list, const pid_t pid) {
702 pids_list_t *new_element = calloc(1, sizeof(*new_element));
704 if (new_element == NULL) {
705 ERROR(RDT_PLUGIN ": Alloc error\n");
708 new_element->pid = pid;
709 new_element->next = NULL;
711 pids_list_t **current = list;
712 while (*current != NULL) {
713 current = &((*current)->next);
715 *current = new_element;
721 * pids_list_contains_pid
724 * Tests if pids list contains specific pid.
727 * `list' Head of pids_list.
728 * `pid' Pid to be searched for.
731 * If PID found in list, returns 1,
732 * Otherwise returns 0.
734 static int pids_list_contains_pid(pids_list_t *list, const pid_t pid) {
737 pids_list_t *current = list;
738 while (current != NULL) {
739 if (current->pid == pid)
741 current = current->next;
748 * pids_list_add_pids_list
751 * Adds pids list at the end of the pids list.
752 * Allocates memory for new pid elements, it is up to user to free it.
753 * Increases dst_num by a number of added PIDs.
756 * `dst' Head of target PIDs list.
757 * `src' Head of source PIDs list.
758 * `dst_num' Variable to be increased by a number of appended PIDs.
761 * On success, returns 0.
762 * -1 on memory allocation error.
764 static int pids_list_add_pids_list(pids_list_t **dst, pids_list_t *src,
770 pids_list_t *current = src;
773 while (current != NULL) {
774 ret = pids_list_add_pid(dst, current->pid);
779 current = current->next;
790 * Reads process name from given pid directory.
791 * Strips new-line character (\n).
794 * `procfs_path` Path to systems proc directory (e.g. /proc)
795 * `pid_entry' Dirent for PID directory
796 * `name' Output buffer for process name, recommended proc_comm.
797 * `out_size' Output buffer size, recommended sizeof(proc_comm)
800 * On success, the number of read bytes (includes stripped \n).
801 * -1 on file open error
803 static int read_proc_name(const char *procfs_path,
804 const struct dirent *pid_entry, char *name,
805 const size_t out_size) {
810 memset(name, 0, out_size);
812 const char *comm_file_name = "comm";
814 char *path = ssnprintf_alloc("%s/%s/%s", procfs_path, pid_entry->d_name,
817 FILE *f = fopen(path, "r");
819 ERROR(RDT_PLUGIN ": Failed to open comm file, error: %d\n", errno);
823 size_t read_length = fread(name, sizeof(char), out_size, f);
826 /* strip new line ending */
827 char *newline = strchr(name, '\n');
840 * Gets pid number for given /proc/pid directory entry or
841 * returns error if input directory does not hold PID information.
844 * `entry' Dirent for PID directory
845 * `pid' PID number to be filled
848 * 0 on success. -1 on error.
850 static int get_pid_number(struct dirent *entry, pid_t *pid) {
851 char *tmp_end; /* used for strtoul error check*/
853 if (pid == NULL || entry == NULL)
856 if (entry->d_type != DT_DIR)
859 /* trying to get pid number from directory name*/
860 *pid = strtoul(entry->d_name, &tmp_end, 10);
861 if (*tmp_end != '\0') {
862 return -1; /* conversion failed, not proc-pid */
864 /* all checks passed, marking as success */
873 * Copies element from list to array. Assumes the space for the array is
877 * `array' First element of target array
878 * `list' Head of the list
879 * `array_length' Length (element count) of the target array
881 static void pids_list_to_array(pid_t *array, pids_list_t *list,
882 const size_t array_length) {
886 assert(array_length > 0);
890 while (list != NULL && current < array_length) {
891 array[current] = list->pid;
899 * initialize_proc_pids
902 * Helper function to properly initialize array of proc_pids.
903 * Allocates memory for proc_pids structs.
906 * `procs_names_array' Array of null-terminated strings with
907 * process' names to be copied to new array
908 * `procs_names_array_size' procs_names_array element count
909 * `proc_pids_array' Address of pointer, under which new
910 * array of proc_pids will be allocated.
913 * 0 on success. Negative number on error:
914 * -1: allocation error
916 static int initialize_proc_pids(const char **procs_names_array,
917 const size_t procs_names_array_size,
918 proc_pids_t **proc_pids_array) {
920 assert(proc_pids_array);
921 assert(NULL == *proc_pids_array);
923 /* Copy procs names to output array. Initialize pids list with NULL value. */
924 *proc_pids_array = calloc(procs_names_array_size, sizeof(**proc_pids_array));
926 if (NULL == *proc_pids_array)
929 for (size_t i = 0; i < procs_names_array_size; ++i) {
930 sstrncpy((*proc_pids_array)[i].proccess_name, procs_names_array[i],
931 STATIC_ARRAY_SIZE((*proc_pids_array)[i].proccess_name));
932 (*proc_pids_array)[i].pids = NULL;
940 * fetch_pids_for_procs
943 * Finds PIDs matching given process's names.
944 * Searches all PID directories in /proc fs and
945 * allocates memory for proc_pids structs, it is up to user to free it.
946 * Output array will have same element count as input array.
949 * `procfs_path' Path to systems proc directory (e.g. /proc)
950 * `procs_names_array' Array of null-terminated strings with
951 * process' names to be copied to new array
952 * `procs_names_array_size' procs_names_array element count
953 * `proc_pids_array' Address of pointer, under which new
954 * array of proc_pids will be allocated.
958 * 0 on success. -1 on error.
960 static int fetch_pids_for_procs(const char *procfs_path,
961 const char **procs_names_array,
962 const size_t procs_names_array_size,
963 proc_pids_t **proc_pids_array) {
965 assert(procs_names_array);
966 assert(procs_names_array_size);
968 DIR *proc_dir = opendir(procfs_path);
969 if (proc_dir == NULL) {
970 ERROR(RDT_PLUGIN ": Could not open %s directory, error: %d", procfs_path,
975 int init_result = initialize_proc_pids(
976 procs_names_array, procs_names_array_size, proc_pids_array);
977 if (0 != init_result)
980 /* Go through procfs and find PIDS and their comms */
981 struct dirent *entry;
982 while ((entry = readdir(proc_dir)) != NULL) {
985 int pid_conversion = get_pid_number(entry, &pid);
986 if (pid_conversion < 0)
991 read_proc_name(procfs_path, entry, comm, sizeof(proc_comm_t));
992 if (read_result <= 0) {
993 ERROR(RDT_PLUGIN ": Comm file skipped. Read result: %d", read_result);
997 /* Try to find comm in input procs array (proc_pids_array has same names) */
998 for (size_t i = 0; i < procs_names_array_size; ++i) {
999 if (0 == strncmp(comm, (*proc_pids_array)[i].proccess_name,
1000 STATIC_ARRAY_SIZE(comm)))
1001 pids_list_add_pid(&((*proc_pids_array)[i].pids), pid);
1005 int close_result = closedir(proc_dir);
1006 if (0 != close_result) {
1007 ERROR(RDT_PLUGIN ": failed to close %s directory, error: %d", procfs_path,
1009 sfree(*proc_pids_array);
1014 #endif /* LIBPQOS2 */
1016 static void rdt_pqos_log(void *context, const size_t size, const char *msg) {
1017 DEBUG(RDT_PLUGIN ": %s", msg);
1020 static int rdt_preinit(void) {
1023 if (g_rdt != NULL) {
1024 /* already initialized if config callback was called before init callback */
1028 g_rdt = calloc(1, sizeof(*g_rdt));
1029 if (g_rdt == NULL) {
1030 ERROR(RDT_PLUGIN ": Failed to allocate memory for rdt context.");
1034 struct pqos_config pqos = {.fd_log = -1,
1035 .callback_log = rdt_pqos_log,
1036 .context_log = NULL,
1039 .interface = PQOS_INTER_OS_RESCTRL_MON};
1040 DEBUG(RDT_PLUGIN ": Initializing PQoS with RESCTRL interface");
1042 .interface = PQOS_INTER_MSR};
1043 DEBUG(RDT_PLUGIN ": Initializing PQoS with MSR interface");
1046 ret = pqos_init(&pqos);
1047 DEBUG(RDT_PLUGIN ": PQoS initialization result: [%d]", ret);
1050 if (ret == PQOS_RETVAL_INTER) {
1051 pqos.interface = PQOS_INTER_MSR;
1052 DEBUG(RDT_PLUGIN ": Initializing PQoS with MSR interface");
1053 ret = pqos_init(&pqos);
1054 DEBUG(RDT_PLUGIN ": PQoS initialization result: [%d]", ret);
1058 if (ret != PQOS_RETVAL_OK) {
1059 ERROR(RDT_PLUGIN ": Error initializing PQoS library!");
1060 goto rdt_preinit_error1;
1063 g_interface = pqos.interface;
1065 ret = pqos_cap_get(&g_rdt->pqos_cap, &g_rdt->pqos_cpu);
1066 if (ret != PQOS_RETVAL_OK) {
1067 ERROR(RDT_PLUGIN ": Error retrieving PQoS capabilities.");
1068 goto rdt_preinit_error2;
1071 ret = pqos_cap_get_type(g_rdt->pqos_cap, PQOS_CAP_TYPE_MON, &g_rdt->cap_mon);
1072 if (ret == PQOS_RETVAL_PARAM) {
1073 ERROR(RDT_PLUGIN ": Error retrieving monitoring capabilities.");
1074 goto rdt_preinit_error2;
1077 if (g_rdt->cap_mon == NULL) {
1080 ": Monitoring capability not detected. Nothing to do for the plugin.");
1081 goto rdt_preinit_error2;
1084 /* Reset pqos monitoring groups registers */
1098 static int rdt_config(oconfig_item_t *ci) {
1099 if (rdt_preinit() != 0) {
1100 g_state = CONFIGURATION_ERROR;
1101 /* if we return -1 at this point collectd
1102 reports a failure in configuration and
1108 for (int i = 0; i < ci->children_num; i++) {
1109 oconfig_item_t *child = ci->children + i;
1111 if (strncasecmp("Cores", child->key, (size_t)strlen("Cores")) == 0) {
1112 if (rdt_config_cgroups(child) != 0) {
1113 g_state = CONFIGURATION_ERROR;
1114 /* if we return -1 at this point collectd
1115 reports a failure in configuration and
1123 #endif /* COLLECT_DEBUG */
1124 } else if (strncasecmp("Processes", child->key,
1125 (size_t)strlen("Processes")) == 0) {
1127 if (g_interface != PQOS_INTER_OS_RESCTRL_MON) {
1128 ERROR(RDT_PLUGIN ": Configuration parameter \"%s\" not supported. "
1129 "Resctrl monitoring is needed for PIDs monitoring.",
1131 g_state = CONFIGURATION_ERROR;
1132 /* if we return -1 at this point collectd
1133 reports a failure in configuration and
1139 if (rdt_config_ngroups(child) != 0) {
1140 g_state = CONFIGURATION_ERROR;
1141 /* if we return -1 at this point collectd
1142 reports a failure in configuration and
1150 #endif /* COLLECT_DEBUG */
1151 #else /* !LIBPQOS2 */
1152 ERROR(RDT_PLUGIN ": Configuration parameter \"%s\" not supported, please "
1153 "recompile collectd with libpqos version 2.0 or newer.",
1155 #endif /* LIBPQOS2 */
1157 ERROR(RDT_PLUGIN ": Unknown configuration parameter \"%s\".", child->key);
1164 static void rdt_submit_derive(const char *cgroup, const char *type,
1165 const char *type_instance, derive_t value) {
1166 value_list_t vl = VALUE_LIST_INIT;
1168 vl.values = &(value_t){.derive = value};
1171 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
1172 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
1173 sstrncpy(vl.type, type, sizeof(vl.type));
1175 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
1177 plugin_dispatch_values(&vl);
1180 static void rdt_submit_gauge(const char *cgroup, const char *type,
1181 const char *type_instance, gauge_t value) {
1182 value_list_t vl = VALUE_LIST_INIT;
1184 vl.values = &(value_t){.gauge = value};
1187 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
1188 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
1189 sstrncpy(vl.type, type, sizeof(vl.type));
1191 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
1193 plugin_dispatch_values(&vl);
1197 static int rdt_pid_list_diff(pids_list_t *prev, pids_list_t *curr,
1198 pids_list_t **added, size_t *added_num,
1199 pids_list_t **removed, size_t *removed_num) {
1200 assert(prev || curr);
1205 /* append all PIDs from curr to added*/
1206 return pids_list_add_pids_list(added, curr, added_num);
1207 } else if (NULL == curr) {
1208 /* append all PIDs from prev to removed*/
1209 return pids_list_add_pids_list(removed, prev, removed_num);
1212 pids_list_t *item = prev;
1213 while (item != NULL) {
1214 if (0 == pids_list_contains_pid(curr, item->pid)) {
1215 pids_list_add_pid(removed, item->pid);
1222 while (item != NULL) {
1223 if (0 == pids_list_contains_pid(prev, item->pid)) {
1224 pids_list_add_pid(added, item->pid);
1233 static int rdt_refresh_ngroup(rdt_name_group_t *ngroup,
1234 struct pqos_mon_data *group_mon_data) {
1241 if (NULL == ngroup->proc_pids_array) {
1243 ": rdt_refresh_ngroup: \'%s\' uninitialized process pids array.",
1249 DEBUG(RDT_PLUGIN ": rdt_refresh_ngroup: \'%s\' process names group.",
1252 proc_pids_t *proc_pids_array_prev = ngroup->proc_pids_array;
1253 proc_pids_t *proc_pids_array_curr = NULL;
1256 fetch_pids_for_procs(RDT_PROC_PATH, (const char **)ngroup->names,
1257 ngroup->num_names, &proc_pids_array_curr);
1259 if (0 != fetch_result) {
1260 ERROR(RDT_PLUGIN ": rdt_refresh_ngroup: \'%s\' failed to fetch PIDs.",
1262 return fetch_result;
1265 pids_list_t *new_pids = NULL;
1266 size_t new_pids_count = 0;
1268 pids_list_t *lost_pids = NULL;
1269 size_t lost_pids_count = 0;
1271 for (size_t i = 0; i < ngroup->num_names; ++i) {
1272 if (NULL == proc_pids_array_prev[i].pids &&
1273 NULL == proc_pids_array_curr[i].pids)
1275 int diff_result = rdt_pid_list_diff(
1276 proc_pids_array_prev[i].pids, proc_pids_array_curr[i].pids, &new_pids,
1277 &new_pids_count, &lost_pids, &lost_pids_count);
1278 if (0 != diff_result) {
1280 ": rdt_refresh_ngroup: \'%s\'. Error [%d] during PID diff.",
1281 ngroup->desc, diff_result);
1287 DEBUG(RDT_PLUGIN ": rdt_refresh_ngroup: \'%s\' process names group, added: "
1289 ngroup->desc, (unsigned)new_pids_count, (unsigned)lost_pids_count);
1291 if (new_pids_count != 0 || lost_pids_count != 0) {
1294 pid_t new_pids_array[new_pids_count];
1295 pids_list_to_array(new_pids_array, new_pids,
1296 STATIC_ARRAY_SIZE(new_pids_array));
1298 /* no pids are monitored for this group yet: start monitoring */
1299 if (0 == ngroup->monitored_pids_count) {
1302 pqos_mon_start_pids(new_pids_count, new_pids_array, ngroup->events,
1303 (void *)ngroup->desc, group_mon_data);
1304 if (PQOS_RETVAL_OK == start_result) {
1305 ngroup->monitored_pids_count = new_pids_count;
1307 ERROR(RDT_PLUGIN ": rdt_refresh_ngroup: \'%s\'. Error [%d] while "
1308 "STARTING pids monitoring",
1309 ngroup->desc, start_result);
1311 goto pqos_error_recovery;
1317 pqos_mon_add_pids(new_pids_count, new_pids_array, group_mon_data);
1318 if (PQOS_RETVAL_OK == add_result)
1319 ngroup->monitored_pids_count += new_pids_count;
1322 ": rdt_refresh_ngroup: \'%s\'. Error [%d] while ADDING pids.",
1323 ngroup->desc, add_result);
1325 goto pqos_error_recovery;
1331 pid_t lost_pids_array[lost_pids_count];
1332 pids_list_to_array(lost_pids_array, lost_pids,
1333 STATIC_ARRAY_SIZE(lost_pids_array));
1335 if (lost_pids_count == ngroup->monitored_pids_count) {
1336 /* all pids for this group are lost: stop monitoring */
1337 int stop_result = pqos_mon_stop(group_mon_data);
1338 if (PQOS_RETVAL_OK != stop_result) {
1339 ERROR(RDT_PLUGIN ": rdt_refresh_ngroup: \'%s\'. Error [%d] while "
1340 "STOPPING monitoring",
1341 ngroup->desc, stop_result);
1343 goto pqos_error_recovery;
1345 ngroup->monitored_pids_count = 0;
1347 assert(lost_pids_count < ngroup->monitored_pids_count);
1348 int remove_result = pqos_mon_remove_pids(
1349 lost_pids_count, lost_pids_array, group_mon_data);
1350 if (PQOS_RETVAL_OK == remove_result) {
1351 ngroup->monitored_pids_count -= lost_pids_count;
1354 ": rdt_refresh_ngroup: \'%s\'. Error [%d] while REMOVING pids.",
1355 ngroup->desc, remove_result);
1357 goto pqos_error_recovery;
1362 ngroup->proc_pids_array = proc_pids_array_curr;
1367 pqos_error_recovery:
1369 * Resources might be temporary unavailable.
1372 * Collectd will halt the reading thread for this
1373 * plugin if it returns an error.
1374 * Consecutive errors will be increasing the read period
1375 * up to 1 day interval.
1376 * On pqos error stop monitoring current group
1377 * and reset the proc_pids array
1378 * monitoring will be restarted on next collectd read cycle
1380 DEBUG(RDT_PLUGIN ": rdt_refresh_ngroup: \'%s\' group RESET after error.",
1382 pqos_mon_stop(group_mon_data);
1383 for (size_t i = 0; i < ngroup->num_names; ++i) {
1384 if (ngroup->proc_pids_array[i].pids)
1385 pids_list_free(ngroup->proc_pids_array[i].pids);
1387 sfree(ngroup->proc_pids_array);
1389 initialize_proc_pids((const char **)ngroup->names, ngroup->num_names,
1390 &ngroup->proc_pids_array);
1391 ngroup->monitored_pids_count = 0;
1394 if (ngroup->proc_pids_array == proc_pids_array_curr) {
1395 assert(proc_pids_array_curr);
1396 /* new list was successfully saved, free the old one */
1397 for (size_t i = 0; i < ngroup->num_names; ++i)
1398 if (proc_pids_array_prev[i].pids)
1399 pids_list_free(proc_pids_array_prev[i].pids);
1401 sfree(proc_pids_array_prev);
1404 /* new list was not saved. Free the new list, keep the old one*/
1405 for (size_t i = 0; i < ngroup->num_names; ++i)
1406 if (proc_pids_array_curr[i].pids)
1407 pids_list_free(proc_pids_array_curr[i].pids);
1409 sfree(proc_pids_array_curr);
1413 pids_list_free(new_pids);
1416 pids_list_free(lost_pids);
1421 static int read_pids_data() {
1423 if (0 == g_rdt->num_ngroups) {
1424 DEBUG(RDT_PLUGIN ": read_pids_data: not configured - PIDs read skipped");
1428 DEBUG(RDT_PLUGIN ": read_pids_data: Scanning active groups");
1429 struct pqos_mon_data *active_groups[RDT_MAX_NAMES_GROUPS] = {0};
1430 size_t active_group_idx = 0;
1431 for (size_t pngroups_idx = 0;
1432 pngroups_idx < STATIC_ARRAY_SIZE(g_rdt->pngroups); ++pngroups_idx)
1433 if (0 != g_rdt->ngroups[pngroups_idx].monitored_pids_count)
1434 active_groups[active_group_idx++] = g_rdt->pngroups[pngroups_idx];
1438 if (0 == active_group_idx) {
1439 DEBUG(RDT_PLUGIN ": read_pids_data: no active groups - PIDs read skipped");
1440 goto groups_refresh;
1443 DEBUG(RDT_PLUGIN ": read_pids_data: PIDs data polling");
1445 int poll_result = pqos_mon_poll(active_groups, active_group_idx);
1446 if (poll_result != PQOS_RETVAL_OK) {
1447 ERROR(RDT_PLUGIN ": read_pids_data: Failed to poll monitoring data for "
1448 "pids. Error [%d].",
1451 goto groups_refresh;
1454 for (size_t i = 0; i < g_rdt->num_ngroups; i++) {
1455 enum pqos_mon_event mbm_events =
1456 (PQOS_MON_EVENT_LMEM_BW | PQOS_MON_EVENT_TMEM_BW |
1457 PQOS_MON_EVENT_RMEM_BW);
1459 if (g_rdt->pngroups[i] == NULL ||
1460 g_rdt->ngroups[i].monitored_pids_count == 0)
1463 const struct pqos_event_values *pv = &g_rdt->pngroups[i]->values;
1465 /* Submit only monitored events data */
1467 if (g_rdt->ngroups[i].events & PQOS_MON_EVENT_L3_OCCUP)
1468 rdt_submit_gauge(g_rdt->ngroups[i].desc, "bytes", "llc", pv->llc);
1470 if (g_rdt->ngroups[i].events & PQOS_PERF_EVENT_IPC)
1471 rdt_submit_gauge(g_rdt->ngroups[i].desc, "ipc", NULL, pv->ipc);
1473 if (g_rdt->ngroups[i].events & mbm_events) {
1474 rdt_submit_derive(g_rdt->ngroups[i].desc, "memory_bandwidth", "local",
1475 pv->mbm_local_delta);
1476 rdt_submit_derive(g_rdt->ngroups[i].desc, "memory_bandwidth", "remote",
1477 pv->mbm_remote_delta);
1482 rdt_dump_pids_data();
1483 #endif /* COLLECT_DEBUG */
1486 for (size_t i = 0; i < g_rdt->num_ngroups; i++) {
1487 int refresh_result =
1488 rdt_refresh_ngroup(&(g_rdt->ngroups[i]), g_rdt->pngroups[i]);
1490 if (0 != refresh_result) {
1491 ERROR(RDT_PLUGIN ": read_pids_data: NGroup %zu refresh failed. Error: %d",
1494 /* refresh error will be escalated only if there were no
1497 ret = refresh_result;
1506 static void rdt_init_pids_monitoring() {
1507 for (size_t group_idx = 0; group_idx < g_rdt->num_ngroups; group_idx++) {
1509 * Each group must have not-null proc_pids array.
1510 * Initial refresh is not mandatory for proper
1511 * PIDs statistics detection.
1513 rdt_name_group_t *ng = &g_rdt->ngroups[group_idx];
1514 int init_result = initialize_proc_pids((const char **)ng->names,
1515 ng->num_names, &ng->proc_pids_array);
1516 if (0 != init_result) {
1518 ": Initialization of proc_pids for group %zu failed. Error: %d",
1519 group_idx, init_result);
1523 int refresh_result = rdt_refresh_ngroup(&(g_rdt->ngroups[group_idx]),
1524 g_rdt->pngroups[group_idx]);
1525 if (0 != refresh_result)
1526 ERROR(RDT_PLUGIN ": Initial refresh of group %zu failed. Error: %d",
1527 group_idx, refresh_result);
1530 #endif /* LIBPQOS2 */
1532 static int read_cores_data() {
1534 if (0 == g_rdt->cores.num_cgroups) {
1535 DEBUG(RDT_PLUGIN ": read_cores_data: not configured - Cores read skipped");
1538 DEBUG(RDT_PLUGIN ": read_cores_data: Cores data poll");
1541 pqos_mon_poll(&g_rdt->pcgroups[0], (unsigned)g_rdt->cores.num_cgroups);
1542 if (ret != PQOS_RETVAL_OK) {
1543 ERROR(RDT_PLUGIN ": read_cores_data: Failed to poll monitoring data for "
1544 "cores. Error [%d].",
1549 for (size_t i = 0; i < g_rdt->cores.num_cgroups; i++) {
1550 core_group_t *cgroup = g_rdt->cores.cgroups + i;
1551 enum pqos_mon_event mbm_events =
1552 (PQOS_MON_EVENT_LMEM_BW | PQOS_MON_EVENT_TMEM_BW |
1553 PQOS_MON_EVENT_RMEM_BW);
1555 const struct pqos_event_values *pv = &g_rdt->pcgroups[i]->values;
1557 /* Submit only monitored events data */
1559 if (g_rdt->events[i] & PQOS_MON_EVENT_L3_OCCUP)
1560 rdt_submit_gauge(cgroup->desc, "bytes", "llc", pv->llc);
1562 if (g_rdt->events[i] & PQOS_PERF_EVENT_IPC)
1563 rdt_submit_gauge(cgroup->desc, "ipc", NULL, pv->ipc);
1565 if (g_rdt->events[i] & mbm_events) {
1566 rdt_submit_derive(cgroup->desc, "memory_bandwidth", "local",
1567 pv->mbm_local_delta);
1568 rdt_submit_derive(cgroup->desc, "memory_bandwidth", "remote",
1569 pv->mbm_remote_delta);
1574 rdt_dump_cores_data();
1575 #endif /* COLLECT_DEBUG */
1580 static int rdt_read(__attribute__((unused)) user_data_t *ud) {
1582 if (g_rdt == NULL) {
1583 ERROR(RDT_PLUGIN ": rdt_read: plugin not initialized.");
1587 int cores_read_result = read_cores_data();
1590 int pids_read_result = read_pids_data();
1591 #endif /* LIBPQOS2 */
1593 if (0 != cores_read_result)
1594 return cores_read_result;
1597 if (0 != pids_read_result)
1598 return pids_read_result;
1599 #endif /* LIBPQOS2 */
1604 static void rdt_init_cores_monitoring() {
1605 for (size_t i = 0; i < g_rdt->cores.num_cgroups; i++) {
1606 core_group_t *cg = g_rdt->cores.cgroups + i;
1608 int mon_start_result =
1609 pqos_mon_start(cg->num_cores, cg->cores, g_rdt->events[i],
1610 (void *)cg->desc, g_rdt->pcgroups[i]);
1612 if (mon_start_result != PQOS_RETVAL_OK)
1614 ": Error starting cores monitoring group %s (pqos status=%d)",
1615 cg->desc, mon_start_result);
1619 static int rdt_init(void) {
1621 if (g_state == CONFIGURATION_ERROR)
1624 int rdt_preinint_result = rdt_preinit();
1625 if (rdt_preinint_result != 0)
1626 return rdt_preinint_result;
1628 rdt_init_cores_monitoring();
1630 rdt_init_pids_monitoring();
1631 #endif /* LIBPQOS2 */
1636 static int rdt_shutdown(void) {
1639 DEBUG(RDT_PLUGIN ": rdt_shutdown.");
1644 /* Stop monitoring cores */
1645 for (size_t i = 0; i < g_rdt->cores.num_cgroups; i++) {
1646 pqos_mon_stop(g_rdt->pcgroups[i]);
1649 /* Stop pids monitoring */
1651 for (size_t i = 0; i < g_rdt->num_ngroups; i++)
1652 pqos_mon_stop(g_rdt->pngroups[i]);
1656 if (ret != PQOS_RETVAL_OK)
1657 ERROR(RDT_PLUGIN ": Error shutting down PQoS library.");
1662 #endif /* LIBPQOS2 */
1668 void module_register(void) {
1669 plugin_register_init(RDT_PLUGIN, rdt_init);
1670 plugin_register_complex_config(RDT_PLUGIN, rdt_config);
1671 plugin_register_complex_read(NULL, RDT_PLUGIN, rdt_read, 0, NULL);
1672 plugin_register_shutdown(RDT_PLUGIN, rdt_shutdown);