2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2018, Matthew Macy
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 #include <sys/types.h>
32 #include <sys/errno.h>
33 #include <sys/sysctl.h>
41 #include <libpmcstat.h>
42 #include "pmu-events/pmu-events.h"
44 #if defined(__amd64__) || defined(__i386__)
56 static struct pmu_alias pmu_intel_alias_table[] = {
57 {"UNHALTED_CORE_CYCLES", "CPU_CLK_UNHALTED.THREAD_P_ANY"},
58 {"UNHALTED-CORE-CYCLES", "CPU_CLK_UNHALTED.THREAD_P_ANY"},
59 {"LLC_MISSES", "LONGEST_LAT_CACHE.MISS"},
60 {"LLC-MISSES", "LONGEST_LAT_CACHE.MISS"},
61 {"LLC_REFERENCE", "LONGEST_LAT_CACHE.REFERENCE"},
62 {"LLC-REFERENCE", "LONGEST_LAT_CACHE.REFERENCE"},
63 {"LLC_MISS_RHITM", "mem_load_l3_miss_retired.remote_hitm"},
64 {"LLC-MISS-RHITM", "mem_load_l3_miss_retired.remote_hitm"},
65 {"RESOURCE_STALL", "RESOURCE_STALLS.ANY"},
66 {"RESOURCE_STALLS_ANY", "RESOURCE_STALLS.ANY"},
67 {"BRANCH_INSTRUCTION_RETIRED", "BR_INST_RETIRED.ALL_BRANCHES"},
68 {"BRANCH-INSTRUCTION-RETIRED", "BR_INST_RETIRED.ALL_BRANCHES"},
69 {"BRANCH_MISSES_RETIRED", "BR_MISP_RETIRED.ALL_BRANCHES"},
70 {"BRANCH-MISSES-RETIRED", "BR_MISP_RETIRED.ALL_BRANCHES"},
71 {"cycles", "tsc-tsc"},
72 {"instructions", "inst-retired.any_p"},
73 {"branch-mispredicts", "br_misp_retired.all_branches"},
74 {"branches", "br_inst_retired.all_branches"},
75 {"interrupts", "hw_interrupts.received"},
76 {"ic-misses", "frontend_retired.l1i_miss"},
80 static struct pmu_alias pmu_amd_alias_table[] = {
81 {"UNHALTED_CORE_CYCLES", "ls_not_halted_cyc"},
82 {"UNHALTED-CORE-CYCLES", "ls_not_halted_cyc"},
94 if (sysctlbyname("kern.hwpmc.cpuid", (void *)NULL, &s,
95 (void *)NULL, 0) == -1)
97 if ((buf = malloc(s + 1)) == NULL)
99 if (sysctlbyname("kern.hwpmc.cpuid", buf, &s,
100 (void *)NULL, 0) == -1) {
102 return (PMU_INVALID);
104 if (strcasestr(buf, "AuthenticAMD") != NULL)
106 else if (strcasestr(buf, "GenuineIntel") != NULL)
115 * The Intel fixed mode counters are:
116 * "inst_retired.any",
117 * "cpu_clk_unhalted.thread",
118 * "cpu_clk_unhalted.thread_any",
119 * "cpu_clk_unhalted.ref_tsc",
124 pmu_alias_get(const char *name)
127 struct pmu_alias *pa;
128 struct pmu_alias *pmu_alias_table;
130 if ((mfr = pmu_events_mfr()) == PMU_INVALID)
133 pmu_alias_table = pmu_amd_alias_table;
134 else if (mfr == PMU_INTEL)
135 pmu_alias_table = pmu_intel_alias_table;
139 for (pa = pmu_alias_table; pa->pa_alias != NULL; pa++)
140 if (strcasecmp(name, pa->pa_alias) == 0)
141 return (pa->pa_name);
146 struct pmu_event_desc {
148 uint64_t ped_offcore_rsp;
149 uint64_t ped_l3_thread;
150 uint64_t ped_l3_slice;
152 uint32_t ped_frontend;
154 uint32_t ped_config1;
164 static const struct pmu_events_map *
165 pmu_events_map_get(const char *cpuid)
169 const struct pmu_events_map *pme;
172 memcpy(buf, cpuid, 64);
174 if (sysctlbyname("kern.hwpmc.cpuid", (void *)NULL, &s,
175 (void *)NULL, 0) == -1)
177 if (sysctlbyname("kern.hwpmc.cpuid", buf, &s,
178 (void *)NULL, 0) == -1)
181 for (pme = pmu_events_map; pme->cpuid != NULL; pme++)
182 if (strcmp(buf, pme->cpuid) == 0)
187 static const struct pmu_event *
188 pmu_event_get(const char *cpuid, const char *event_name, int *idx)
190 const struct pmu_events_map *pme;
191 const struct pmu_event *pe;
194 if ((pme = pmu_events_map_get(cpuid)) == NULL)
196 for (i = 0, pe = pme->table; pe->name || pe->desc || pe->event; pe++, i++) {
197 if (pe->name == NULL)
199 if (strcasecmp(pe->name, event_name) == 0) {
209 pmc_pmu_idx_get_by_event(const char *cpuid, const char *event)
212 const char *realname;
214 realname = pmu_alias_get(event);
215 if (pmu_event_get(cpuid, realname, &idx) == NULL)
221 pmc_pmu_event_get_by_idx(const char *cpuid, int idx)
223 const struct pmu_events_map *pme;
225 if ((pme = pmu_events_map_get(cpuid)) == NULL)
227 assert(pme->table[idx].name);
228 return (pme->table[idx].name);
232 pmu_parse_event(struct pmu_event_desc *ped, const char *eventin)
235 char *kvp, *key, *value, *r;
238 if ((event = strdup(eventin)) == NULL)
241 bzero(ped, sizeof(*ped));
242 ped->ped_period = DEFAULT_SAMPLE_COUNT;
244 while ((kvp = strsep(&event, ",")) != NULL) {
245 key = strsep(&kvp, "=");
249 if (strcmp(key, "umask") == 0)
250 ped->ped_umask = strtol(value, NULL, 16);
251 else if (strcmp(key, "event") == 0)
252 ped->ped_event = strtol(value, NULL, 16);
253 else if (strcmp(key, "period") == 0)
254 ped->ped_period = strtol(value, NULL, 10);
255 else if (strcmp(key, "offcore_rsp") == 0)
256 ped->ped_offcore_rsp = strtol(value, NULL, 16);
257 else if (strcmp(key, "any") == 0)
258 ped->ped_any = strtol(value, NULL, 10);
259 else if (strcmp(key, "cmask") == 0)
260 ped->ped_cmask = strtol(value, NULL, 10);
261 else if (strcmp(key, "inv") == 0)
262 ped->ped_inv = strtol(value, NULL, 10);
263 else if (strcmp(key, "edge") == 0)
264 ped->ped_edge = strtol(value, NULL, 10);
265 else if (strcmp(key, "frontend") == 0)
266 ped->ped_frontend = strtol(value, NULL, 16);
267 else if (strcmp(key, "ldlat") == 0)
268 ped->ped_ldlat = strtol(value, NULL, 16);
269 else if (strcmp(key, "fc_mask") == 0)
270 ped->ped_fc_mask = strtol(value, NULL, 16);
271 else if (strcmp(key, "ch_mask") == 0)
272 ped->ped_ch_mask = strtol(value, NULL, 16);
273 else if (strcmp(key, "config1") == 0)
274 ped->ped_config1 = strtol(value, NULL, 16);
275 else if (strcmp(key, "l3_thread_mask") == 0)
276 ped->ped_l3_thread = strtol(value, NULL, 16);
277 else if (strcmp(key, "l3_slice_mask") == 0)
278 ped->ped_l3_slice = strtol(value, NULL, 16);
280 debug = getenv("PMUDEBUG");
281 if (debug != NULL && strcmp(debug, "true") == 0 && value != NULL)
282 printf("unrecognized kvpair: %s:%s\n", key, value);
290 pmc_pmu_sample_rate_get(const char *event_name)
292 const struct pmu_event *pe;
293 struct pmu_event_desc ped;
295 event_name = pmu_alias_get(event_name);
296 if ((pe = pmu_event_get(NULL, event_name, NULL)) == NULL)
297 return (DEFAULT_SAMPLE_COUNT);
298 if (pe->alias && (pe = pmu_event_get(NULL, pe->alias, NULL)) == NULL)
299 return (DEFAULT_SAMPLE_COUNT);
300 if (pe->event == NULL)
301 return (DEFAULT_SAMPLE_COUNT);
302 if (pmu_parse_event(&ped, pe->event))
303 return (DEFAULT_SAMPLE_COUNT);
304 return (ped.ped_period);
308 pmc_pmu_enabled(void)
311 return (pmu_events_map_get(NULL) != NULL);
315 pmc_pmu_print_counters(const char *event_name)
317 const struct pmu_events_map *pme;
318 const struct pmu_event *pe;
319 struct pmu_event_desc ped;
323 debug = getenv("PMUDEBUG");
326 if (debug != NULL && strcmp(debug, "true") == 0)
328 if ((pme = pmu_events_map_get(NULL)) == NULL)
330 for (pe = pme->table; pe->name || pe->desc || pe->event; pe++) {
331 if (pe->name == NULL)
333 if (event_name != NULL && strcasestr(pe->name, event_name) == NULL)
335 printf("\t%s\n", pe->name);
337 pmu_parse_event(&ped, pe->event);
342 pmc_pmu_print_counter_desc(const char *ev)
344 const struct pmu_events_map *pme;
345 const struct pmu_event *pe;
347 if ((pme = pmu_events_map_get(NULL)) == NULL)
349 for (pe = pme->table; pe->name || pe->desc || pe->event; pe++) {
350 if (pe->name == NULL)
352 if (strcasestr(pe->name, ev) != NULL &&
354 printf("%s:\t%s\n", pe->name, pe->desc);
359 pmc_pmu_print_counter_desc_long(const char *ev)
361 const struct pmu_events_map *pme;
362 const struct pmu_event *pe;
364 if ((pme = pmu_events_map_get(NULL)) == NULL)
366 for (pe = pme->table; pe->name || pe->desc || pe->event; pe++) {
367 if (pe->name == NULL)
369 if (strcasestr(pe->name, ev) != NULL) {
370 if (pe->long_desc != NULL)
371 printf("%s:\n%s\n", pe->name, pe->long_desc);
372 else if (pe->desc != NULL)
373 printf("%s:\t%s\n", pe->name, pe->desc);
379 pmc_pmu_print_counter_full(const char *ev)
381 const struct pmu_events_map *pme;
382 const struct pmu_event *pe;
384 if ((pme = pmu_events_map_get(NULL)) == NULL)
386 for (pe = pme->table; pe->name || pe->desc || pe->event; pe++) {
387 if (pe->name == NULL)
389 if (strcasestr(pe->name, ev) == NULL)
391 printf("name: %s\n", pe->name);
392 if (pe->long_desc != NULL)
393 printf("desc: %s\n", pe->long_desc);
394 else if (pe->desc != NULL)
395 printf("desc: %s\n", pe->desc);
396 if (pe->event != NULL)
397 printf("event: %s\n", pe->event);
398 if (pe->topic != NULL)
399 printf("topic: %s\n", pe->topic);
401 printf("pmu: %s\n", pe->pmu);
402 if (pe->unit != NULL)
403 printf("unit: %s\n", pe->unit);
404 if (pe->perpkg != NULL)
405 printf("perpkg: %s\n", pe->perpkg);
406 if (pe->metric_expr != NULL)
407 printf("metric_expr: %s\n", pe->metric_expr);
408 if (pe->metric_name != NULL)
409 printf("metric_name: %s\n", pe->metric_name);
410 if (pe->metric_group != NULL)
411 printf("metric_group: %s\n", pe->metric_group);
416 pmc_pmu_amd_pmcallocate(const char *event_name, struct pmc_op_pmcallocate *pm,
417 struct pmu_event_desc *ped)
419 struct pmc_md_amd_op_pmcallocate *amd;
420 const struct pmu_event *pe;
423 amd = &pm->pm_md.pm_amd;
424 if (ped->ped_umask > 0) {
425 pm->pm_caps |= PMC_CAP_QUALIFIER;
426 amd->pm_amd_config |= AMD_PMC_TO_UNITMASK(ped->ped_umask);
428 pm->pm_class = PMC_CLASS_K8;
429 pe = pmu_event_get(NULL, event_name, &idx);
431 if (strcmp("l3cache", pe->topic) == 0){
432 amd->pm_amd_config |= AMD_PMC_TO_EVENTMASK(ped->ped_event);
433 amd->pm_amd_sub_class = PMC_AMD_SUB_CLASS_L3_CACHE;
434 amd->pm_amd_config |= AMD_PMC_TO_L3SLICE(ped->ped_l3_slice);
435 amd->pm_amd_config |= AMD_PMC_TO_L3CORE(ped->ped_l3_thread);
437 else if (strcmp("data fabric", pe->topic) == 0){
439 amd->pm_amd_config |= AMD_PMC_TO_EVENTMASK_DF(ped->ped_event);
440 amd->pm_amd_sub_class = PMC_AMD_SUB_CLASS_DATA_FABRIC;
443 amd->pm_amd_config |= AMD_PMC_TO_EVENTMASK(ped->ped_event);
444 amd->pm_amd_sub_class = PMC_AMD_SUB_CLASS_CORE;
445 if ((pm->pm_caps & (PMC_CAP_USER|PMC_CAP_SYSTEM)) == 0 ||
446 (pm->pm_caps & (PMC_CAP_USER|PMC_CAP_SYSTEM)) ==
447 (PMC_CAP_USER|PMC_CAP_SYSTEM))
448 amd->pm_amd_config |= (AMD_PMC_USR | AMD_PMC_OS);
449 else if (pm->pm_caps & PMC_CAP_USER)
450 amd->pm_amd_config |= AMD_PMC_USR;
451 else if (pm->pm_caps & PMC_CAP_SYSTEM)
452 amd->pm_amd_config |= AMD_PMC_OS;
454 amd->pm_amd_config |= AMD_PMC_EDGE;
456 amd->pm_amd_config |= AMD_PMC_EDGE;
457 if (pm->pm_caps & PMC_CAP_INTERRUPT)
458 amd->pm_amd_config |= AMD_PMC_INT;
464 pmc_pmu_intel_pmcallocate(const char *event_name, struct pmc_op_pmcallocate *pm,
465 struct pmu_event_desc *ped)
467 struct pmc_md_iap_op_pmcallocate *iap;
471 iap = &pm->pm_md.pm_iap;
472 if (strcasestr(event_name, "UNC_") == event_name ||
473 strcasestr(event_name, "uncore") != NULL) {
474 pm->pm_class = PMC_CLASS_UCP;
475 pm->pm_caps |= PMC_CAP_QUALIFIER;
476 } else if ((ped->ped_umask == -1) ||
477 (ped->ped_event == 0x0 && ped->ped_umask == 0x3)) {
478 pm->pm_class = PMC_CLASS_IAF;
480 pm->pm_class = PMC_CLASS_IAP;
481 pm->pm_caps |= PMC_CAP_QUALIFIER;
483 iap->pm_iap_config |= IAP_EVSEL(ped->ped_event);
484 if (ped->ped_umask > 0)
485 iap->pm_iap_config |= IAP_UMASK(ped->ped_umask);
486 iap->pm_iap_config |= IAP_CMASK(ped->ped_cmask);
487 iap->pm_iap_rsp = ped->ped_offcore_rsp;
489 if ((pm->pm_caps & (PMC_CAP_USER|PMC_CAP_SYSTEM)) == 0 ||
490 (pm->pm_caps & (PMC_CAP_USER|PMC_CAP_SYSTEM)) ==
491 (PMC_CAP_USER|PMC_CAP_SYSTEM))
492 iap->pm_iap_config |= (IAP_USR | IAP_OS);
493 else if (pm->pm_caps & PMC_CAP_USER)
494 iap->pm_iap_config |= IAP_USR;
495 else if (pm->pm_caps & PMC_CAP_SYSTEM)
496 iap->pm_iap_config |= IAP_OS;
498 iap->pm_iap_config |= IAP_EDGE;
500 iap->pm_iap_config |= IAP_ANY;
502 iap->pm_iap_config |= IAP_EDGE;
503 if (pm->pm_caps & PMC_CAP_INTERRUPT)
504 iap->pm_iap_config |= IAP_INT;
509 pmc_pmu_pmcallocate(const char *event_name, struct pmc_op_pmcallocate *pm)
511 const struct pmu_event *pe;
512 struct pmu_event_desc ped;
516 if ((mfr = pmu_events_mfr()) == PMU_INVALID)
519 bzero(&pm->pm_md, sizeof(pm->pm_md));
520 pm->pm_caps |= (PMC_CAP_READ | PMC_CAP_WRITE);
521 event_name = pmu_alias_get(event_name);
522 if ((pe = pmu_event_get(NULL, event_name, &idx)) == NULL)
524 if (pe->alias && (pe = pmu_event_get(NULL, pe->alias, &idx)) == NULL)
529 if (pe->event == NULL)
531 if (pmu_parse_event(&ped, pe->event))
534 if (mfr == PMU_INTEL)
535 return (pmc_pmu_intel_pmcallocate(event_name, pm, &ped));
537 return (pmc_pmu_amd_pmcallocate(event_name, pm, &ped));
541 * Ultimately rely on AMD calling theirs the same
543 static const char *stat_mode_cntrs[] = {
544 "cpu_clk_unhalted.thread",
546 "br_inst_retired.all_branches",
547 "br_misp_retired.all_branches",
548 "longest_lat_cache.reference",
549 "longest_lat_cache.miss",
553 pmc_pmu_stat_mode(const char ***cntrs)
555 if (pmc_pmu_enabled()) {
556 *cntrs = stat_mode_cntrs;
565 pmc_pmu_sample_rate_get(const char *event_name __unused)
567 return (DEFAULT_SAMPLE_COUNT);
571 pmc_pmu_print_counters(const char *event_name __unused)
576 pmc_pmu_print_counter_desc(const char *e __unused)
581 pmc_pmu_print_counter_desc_long(const char *e __unused)
586 pmc_pmu_print_counter_full(const char *e __unused)
592 pmc_pmu_enabled(void)
598 pmc_pmu_pmcallocate(const char *e __unused, struct pmc_op_pmcallocate *p __unused)
604 pmc_pmu_event_get_by_idx(const char *c __unused, int idx __unused)
610 pmc_pmu_stat_mode(const char ***a __unused)
616 pmc_pmu_idx_get_by_event(const char *c __unused, const char *e __unused)