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>
29 #include "utils_config_cores.h"
34 #define RDT_PLUGIN "intel_rdt"
36 #define RDT_MAX_SOCKETS 8
37 #define RDT_MAX_SOCKET_CORES 64
38 #define RDT_MAX_CORES (RDT_MAX_SOCKET_CORES * RDT_MAX_SOCKETS)
46 core_groups_list_t cores;
47 enum pqos_mon_event events[RDT_MAX_CORES];
48 struct pqos_mon_data *pgroups[RDT_MAX_CORES];
50 const struct pqos_cpuinfo *pqos_cpu;
51 const struct pqos_cap *pqos_cap;
52 const struct pqos_capability *cap_mon;
54 typedef struct rdt_ctx_s rdt_ctx_t;
56 static rdt_ctx_t *g_rdt;
58 static rdt_config_status g_state = UNKNOWN;
61 static void rdt_dump_cgroups(void) {
62 char cores[RDT_MAX_CORES * 4];
67 DEBUG(RDT_PLUGIN ": Core Groups Dump");
68 DEBUG(RDT_PLUGIN ": groups count: %" PRIsz, g_rdt->num_groups);
70 for (size_t i = 0; i < g_rdt->num_groups; i++) {
71 core_group_t *cgroup = g_rdt->cores.cgroups + i;
73 memset(cores, 0, sizeof(cores));
74 for (int j = 0; j < cgroup->num_cores; j++) {
75 snprintf(cores + strlen(cores), sizeof(cores) - strlen(cores) - 1, " %d",
79 DEBUG(RDT_PLUGIN ": group[%zu]:", i);
80 DEBUG(RDT_PLUGIN ": description: %s", cgroup->desc);
81 DEBUG(RDT_PLUGIN ": cores: %s", cores);
82 DEBUG(RDT_PLUGIN ": events: 0x%X", g_rdt->events[i]);
88 static inline double bytes_to_kb(const double bytes) { return bytes / 1024.0; }
90 static inline double bytes_to_mb(const double bytes) {
91 return bytes / (1024.0 * 1024.0);
94 static void rdt_dump_data(void) {
96 * CORE - monitored group of cores
97 * RMID - Resource Monitoring ID associated with the monitored group
98 * LLC - last level cache occupancy
99 * MBL - local memory bandwidth
100 * MBR - remote memory bandwidth
102 DEBUG(" CORE RMID LLC[KB] MBL[MB] MBR[MB]");
103 for (int i = 0; i < g_rdt->num_groups; i++) {
105 const struct pqos_event_values *pv = &g_rdt->pgroups[i]->values;
107 double llc = bytes_to_kb(pv->llc);
108 double mbr = bytes_to_mb(pv->mbm_remote_delta);
109 double mbl = bytes_to_mb(pv->mbm_local_delta);
111 DEBUG(" [%s] %8u %10.1f %10.1f %10.1f", g_rdt->cores.cgroups[i].desc,
112 g_rdt->pgroups[i]->poll_ctx[0].rmid, llc, mbl, mbr);
115 #endif /* COLLECT_DEBUG */
117 static void rdt_free_cgroups(void) {
118 config_cores_cleanup(&g_rdt->cores);
119 for (int i = 0; i < RDT_MAX_CORES; i++) {
120 sfree(g_rdt->pgroups[i]);
124 static int rdt_default_cgroups(void) {
125 unsigned num_cores = g_rdt->pqos_cpu->num_cores;
127 g_rdt->cores.cgroups = calloc(num_cores, sizeof(*(g_rdt->cores.cgroups)));
128 if (g_rdt->cores.cgroups == NULL) {
129 ERROR(RDT_PLUGIN ": Error allocating core groups array");
132 g_rdt->cores.num_cgroups = num_cores;
134 /* configure each core in separate group */
135 for (unsigned i = 0; i < num_cores; i++) {
136 core_group_t *cgroup = g_rdt->cores.cgroups + i;
137 char desc[DATA_MAX_NAME_LEN];
139 /* set core group info */
140 cgroup->cores = calloc(1, sizeof(*(cgroup->cores)));
141 if (cgroup->cores == NULL) {
142 ERROR(RDT_PLUGIN ": Error allocating cores array");
146 cgroup->num_cores = 1;
147 cgroup->cores[0] = i;
149 snprintf(desc, sizeof(desc), "%d", g_rdt->pqos_cpu->cores[i].lcore);
150 cgroup->desc = strdup(desc);
151 if (cgroup->desc == NULL) {
152 ERROR(RDT_PLUGIN ": Error allocating core group description");
161 static int rdt_is_core_id_valid(unsigned int core_id) {
163 for (unsigned int i = 0; i < g_rdt->pqos_cpu->num_cores; i++)
164 if (core_id == g_rdt->pqos_cpu->cores[i].lcore)
170 static int rdt_config_cgroups(oconfig_item_t *item) {
172 enum pqos_mon_event events = 0;
174 if (config_cores_parse(item, &g_rdt->cores) < 0) {
176 ERROR(RDT_PLUGIN ": Error parsing core groups configuration.");
179 n = g_rdt->cores.num_cgroups;
181 /* validate configured core id values */
182 for (size_t group_idx = 0; group_idx < n; group_idx++) {
183 core_group_t *cgroup = g_rdt->cores.cgroups + group_idx;
184 for (size_t core_idx = 0; core_idx < cgroup->num_cores; core_idx++) {
185 if (!rdt_is_core_id_valid(cgroup->cores[core_idx])) {
186 ERROR(RDT_PLUGIN ": Core group '%s' contains invalid core id '%u'",
187 cgroup->desc, cgroup->cores[core_idx]);
195 /* create default core groups if "Cores" config option is empty */
196 int ret = rdt_default_cgroups();
199 ERROR(RDT_PLUGIN ": Error creating default core groups configuration.");
204 ": No core groups configured. Default core groups created.");
207 /* Get all available events on this platform */
208 for (unsigned int i = 0; i < g_rdt->cap_mon->u.mon->num_events; i++)
209 events |= g_rdt->cap_mon->u.mon->events[i].type;
211 events &= ~(PQOS_PERF_EVENT_LLC_MISS);
213 DEBUG(RDT_PLUGIN ": Number of cores in the system: %u",
214 g_rdt->pqos_cpu->num_cores);
215 DEBUG(RDT_PLUGIN ": Available events to monitor: %#x", events);
217 g_rdt->num_groups = n;
218 for (size_t i = 0; i < n; i++) {
219 for (size_t j = 0; j < i; j++) {
221 found = config_cores_cmp_cgroups(&g_rdt->cores.cgroups[j],
222 &g_rdt->cores.cgroups[i]);
225 ERROR(RDT_PLUGIN ": Cannot monitor same cores in different groups.");
230 g_rdt->events[i] = events;
231 g_rdt->pgroups[i] = calloc(1, sizeof(*g_rdt->pgroups[i]));
232 if (g_rdt->pgroups[i] == NULL) {
234 ERROR(RDT_PLUGIN ": Failed to allocate memory for monitoring data.");
242 static void rdt_pqos_log(void *context, const size_t size, const char *msg) {
243 DEBUG(RDT_PLUGIN ": %s", msg);
246 static int rdt_preinit(void) {
250 /* already initialized if config callback was called before init callback */
254 g_rdt = calloc(1, sizeof(*g_rdt));
256 ERROR(RDT_PLUGIN ": Failed to allocate memory for rdt context.");
260 struct pqos_config pqos = {.fd_log = -1,
261 .callback_log = rdt_pqos_log,
265 ret = pqos_init(&pqos);
266 if (ret != PQOS_RETVAL_OK) {
267 ERROR(RDT_PLUGIN ": Error initializing PQoS library!");
268 goto rdt_preinit_error1;
271 ret = pqos_cap_get(&g_rdt->pqos_cap, &g_rdt->pqos_cpu);
272 if (ret != PQOS_RETVAL_OK) {
273 ERROR(RDT_PLUGIN ": Error retrieving PQoS capabilities.");
274 goto rdt_preinit_error2;
277 ret = pqos_cap_get_type(g_rdt->pqos_cap, PQOS_CAP_TYPE_MON, &g_rdt->cap_mon);
278 if (ret == PQOS_RETVAL_PARAM) {
279 ERROR(RDT_PLUGIN ": Error retrieving monitoring capabilities.");
280 goto rdt_preinit_error2;
283 if (g_rdt->cap_mon == NULL) {
286 ": Monitoring capability not detected. Nothing to do for the plugin.");
287 goto rdt_preinit_error2;
290 /* Reset pqos monitoring groups registers */
305 static int rdt_config(oconfig_item_t *ci) {
306 if (rdt_preinit() != 0) {
307 g_state = CONFIGURATION_ERROR;
308 /* if we return -1 at this point collectd
309 reports a failure in configuration and
315 for (int i = 0; i < ci->children_num; i++) {
316 oconfig_item_t *child = ci->children + i;
318 if (strcasecmp("Cores", child->key) == 0) {
319 if (rdt_config_cgroups(child) != 0) {
320 g_state = CONFIGURATION_ERROR;
321 /* if we return -1 at this point collectd
322 reports a failure in configuration and
330 #endif /* COLLECT_DEBUG */
332 ERROR(RDT_PLUGIN ": Unknown configuration parameter \"%s\".", child->key);
339 static void rdt_submit_derive(char *cgroup, char *type, char *type_instance,
341 value_list_t vl = VALUE_LIST_INIT;
343 vl.values = &(value_t){.derive = value};
346 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
347 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
348 sstrncpy(vl.type, type, sizeof(vl.type));
350 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
352 plugin_dispatch_values(&vl);
355 static void rdt_submit_gauge(char *cgroup, char *type, char *type_instance,
357 value_list_t vl = VALUE_LIST_INIT;
359 vl.values = &(value_t){.gauge = value};
362 sstrncpy(vl.plugin, RDT_PLUGIN, sizeof(vl.plugin));
363 snprintf(vl.plugin_instance, sizeof(vl.plugin_instance), "%s", cgroup);
364 sstrncpy(vl.type, type, sizeof(vl.type));
366 sstrncpy(vl.type_instance, type_instance, sizeof(vl.type_instance));
368 plugin_dispatch_values(&vl);
371 static int rdt_read(__attribute__((unused)) user_data_t *ud) {
375 ERROR(RDT_PLUGIN ": rdt_read: plugin not initialized.");
379 ret = pqos_mon_poll(&g_rdt->pgroups[0], (unsigned)g_rdt->num_groups);
380 if (ret != PQOS_RETVAL_OK) {
381 ERROR(RDT_PLUGIN ": Failed to poll monitoring data.");
387 #endif /* COLLECT_DEBUG */
389 for (size_t i = 0; i < g_rdt->num_groups; i++) {
390 core_group_t *cgroup = g_rdt->cores.cgroups + i;
392 enum pqos_mon_event mbm_events =
393 (PQOS_MON_EVENT_LMEM_BW | PQOS_MON_EVENT_TMEM_BW |
394 PQOS_MON_EVENT_RMEM_BW);
396 const struct pqos_event_values *pv = &g_rdt->pgroups[i]->values;
398 /* Submit only monitored events data */
400 if (g_rdt->events[i] & PQOS_MON_EVENT_L3_OCCUP)
401 rdt_submit_gauge(cgroup->desc, "bytes", "llc", pv->llc);
403 if (g_rdt->events[i] & PQOS_PERF_EVENT_IPC)
404 rdt_submit_gauge(cgroup->desc, "ipc", NULL, pv->ipc);
406 if (g_rdt->events[i] & mbm_events) {
407 rdt_submit_derive(cgroup->desc, "memory_bandwidth", "local",
408 pv->mbm_local_delta);
409 rdt_submit_derive(cgroup->desc, "memory_bandwidth", "remote",
410 pv->mbm_remote_delta);
417 static int rdt_init(void) {
420 if (g_state == CONFIGURATION_ERROR)
427 /* Start monitoring */
428 for (size_t i = 0; i < g_rdt->num_groups; i++) {
429 core_group_t *cg = g_rdt->cores.cgroups + i;
431 ret = pqos_mon_start(cg->num_cores, cg->cores, g_rdt->events[i],
432 (void *)cg->desc, g_rdt->pgroups[i]);
434 if (ret != PQOS_RETVAL_OK)
435 ERROR(RDT_PLUGIN ": Error starting monitoring group %s (pqos status=%d)",
442 static int rdt_shutdown(void) {
445 DEBUG(RDT_PLUGIN ": rdt_shutdown.");
450 /* Stop monitoring */
451 for (size_t i = 0; i < g_rdt->num_groups; i++) {
452 pqos_mon_stop(g_rdt->pgroups[i]);
456 if (ret != PQOS_RETVAL_OK)
457 ERROR(RDT_PLUGIN ": Error shutting down PQoS library.");
465 void module_register(void) {
466 plugin_register_init(RDT_PLUGIN, rdt_init);
467 plugin_register_complex_config(RDT_PLUGIN, rdt_config);
468 plugin_register_complex_read(NULL, RDT_PLUGIN, rdt_read, 0, NULL);
469 plugin_register_shutdown(RDT_PLUGIN, rdt_shutdown);