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 {"unhalted-cycles", "CPU_CLK_UNHALTED.THREAD_P_ANY"},
73 {"instructions", "inst-retired.any_p"},
74 {"branch-mispredicts", "br_misp_retired.all_branches"},
75 {"branches", "br_inst_retired.all_branches"},
76 {"interrupts", "hw_interrupts.received"},
77 {"ic-misses", "frontend_retired.l1i_miss"},
81 static struct pmu_alias pmu_amd_alias_table[] = {
82 {"UNHALTED_CORE_CYCLES", "ls_not_halted_cyc"},
83 {"UNHALTED-CORE-CYCLES", "ls_not_halted_cyc"},
95 if (sysctlbyname("kern.hwpmc.cpuid", (void *)NULL, &s,
96 (void *)NULL, 0) == -1)
98 if ((buf = malloc(s + 1)) == NULL)
100 if (sysctlbyname("kern.hwpmc.cpuid", buf, &s,
101 (void *)NULL, 0) == -1) {
103 return (PMU_INVALID);
105 if (strcasestr(buf, "AuthenticAMD") != NULL)
107 else if (strcasestr(buf, "GenuineIntel") != NULL)
116 * The Intel fixed mode counters are:
117 * "inst_retired.any",
118 * "cpu_clk_unhalted.thread",
119 * "cpu_clk_unhalted.thread_any",
120 * "cpu_clk_unhalted.ref_tsc",
125 pmu_alias_get(const char *name)
128 struct pmu_alias *pa;
129 struct pmu_alias *pmu_alias_table;
131 if ((mfr = pmu_events_mfr()) == PMU_INVALID)
134 pmu_alias_table = pmu_amd_alias_table;
135 else if (mfr == PMU_INTEL)
136 pmu_alias_table = pmu_intel_alias_table;
140 for (pa = pmu_alias_table; pa->pa_alias != NULL; pa++)
141 if (strcasecmp(name, pa->pa_alias) == 0)
142 return (pa->pa_name);
147 struct pmu_event_desc {
149 uint64_t ped_offcore_rsp;
150 uint64_t ped_l3_thread;
151 uint64_t ped_l3_slice;
153 uint32_t ped_frontend;
155 uint32_t ped_config1;
165 static const struct pmu_events_map *
166 pmu_events_map_get(const char *cpuid)
170 const struct pmu_events_map *pme;
173 memcpy(buf, cpuid, 64);
175 if (sysctlbyname("kern.hwpmc.cpuid", (void *)NULL, &s,
176 (void *)NULL, 0) == -1)
178 if (sysctlbyname("kern.hwpmc.cpuid", buf, &s,
179 (void *)NULL, 0) == -1)
182 for (pme = pmu_events_map; pme->cpuid != NULL; pme++)
183 if (strcmp(buf, pme->cpuid) == 0)
188 static const struct pmu_event *
189 pmu_event_get(const char *cpuid, const char *event_name, int *idx)
191 const struct pmu_events_map *pme;
192 const struct pmu_event *pe;
195 if ((pme = pmu_events_map_get(cpuid)) == NULL)
197 for (i = 0, pe = pme->table; pe->name || pe->desc || pe->event; pe++, i++) {
198 if (pe->name == NULL)
200 if (strcasecmp(pe->name, event_name) == 0) {
210 pmc_pmu_idx_get_by_event(const char *cpuid, const char *event)
213 const char *realname;
215 realname = pmu_alias_get(event);
216 if (pmu_event_get(cpuid, realname, &idx) == NULL)
222 pmc_pmu_event_get_by_idx(const char *cpuid, int idx)
224 const struct pmu_events_map *pme;
226 if ((pme = pmu_events_map_get(cpuid)) == NULL)
228 assert(pme->table[idx].name);
229 return (pme->table[idx].name);
233 pmu_parse_event(struct pmu_event_desc *ped, const char *eventin)
236 char *kvp, *key, *value, *r;
239 if ((event = strdup(eventin)) == NULL)
242 bzero(ped, sizeof(*ped));
243 ped->ped_period = DEFAULT_SAMPLE_COUNT;
245 while ((kvp = strsep(&event, ",")) != NULL) {
246 key = strsep(&kvp, "=");
250 if (strcmp(key, "umask") == 0)
251 ped->ped_umask = strtol(value, NULL, 16);
252 else if (strcmp(key, "event") == 0)
253 ped->ped_event = strtol(value, NULL, 16);
254 else if (strcmp(key, "period") == 0)
255 ped->ped_period = strtol(value, NULL, 10);
256 else if (strcmp(key, "offcore_rsp") == 0)
257 ped->ped_offcore_rsp = strtol(value, NULL, 16);
258 else if (strcmp(key, "any") == 0)
259 ped->ped_any = strtol(value, NULL, 10);
260 else if (strcmp(key, "cmask") == 0)
261 ped->ped_cmask = strtol(value, NULL, 10);
262 else if (strcmp(key, "inv") == 0)
263 ped->ped_inv = strtol(value, NULL, 10);
264 else if (strcmp(key, "edge") == 0)
265 ped->ped_edge = strtol(value, NULL, 10);
266 else if (strcmp(key, "frontend") == 0)
267 ped->ped_frontend = strtol(value, NULL, 16);
268 else if (strcmp(key, "ldlat") == 0)
269 ped->ped_ldlat = strtol(value, NULL, 16);
270 else if (strcmp(key, "fc_mask") == 0)
271 ped->ped_fc_mask = strtol(value, NULL, 16);
272 else if (strcmp(key, "ch_mask") == 0)
273 ped->ped_ch_mask = strtol(value, NULL, 16);
274 else if (strcmp(key, "config1") == 0)
275 ped->ped_config1 = strtol(value, NULL, 16);
276 else if (strcmp(key, "l3_thread_mask") == 0)
277 ped->ped_l3_thread = strtol(value, NULL, 16);
278 else if (strcmp(key, "l3_slice_mask") == 0)
279 ped->ped_l3_slice = strtol(value, NULL, 16);
281 debug = getenv("PMUDEBUG");
282 if (debug != NULL && strcmp(debug, "true") == 0 && value != NULL)
283 printf("unrecognized kvpair: %s:%s\n", key, value);
291 pmc_pmu_sample_rate_get(const char *event_name)
293 const struct pmu_event *pe;
294 struct pmu_event_desc ped;
296 event_name = pmu_alias_get(event_name);
297 if ((pe = pmu_event_get(NULL, event_name, NULL)) == NULL)
298 return (DEFAULT_SAMPLE_COUNT);
299 if (pe->alias && (pe = pmu_event_get(NULL, pe->alias, NULL)) == NULL)
300 return (DEFAULT_SAMPLE_COUNT);
301 if (pe->event == NULL)
302 return (DEFAULT_SAMPLE_COUNT);
303 if (pmu_parse_event(&ped, pe->event))
304 return (DEFAULT_SAMPLE_COUNT);
305 return (ped.ped_period);
309 pmc_pmu_enabled(void)
312 return (pmu_events_map_get(NULL) != NULL);
316 pmc_pmu_print_counters(const char *event_name)
318 const struct pmu_events_map *pme;
319 const struct pmu_event *pe;
320 struct pmu_event_desc ped;
324 debug = getenv("PMUDEBUG");
327 if (debug != NULL && strcmp(debug, "true") == 0)
329 if ((pme = pmu_events_map_get(NULL)) == NULL)
331 for (pe = pme->table; pe->name || pe->desc || pe->event; pe++) {
332 if (pe->name == NULL)
334 if (event_name != NULL && strcasestr(pe->name, event_name) == NULL)
336 printf("\t%s\n", pe->name);
338 pmu_parse_event(&ped, pe->event);
343 pmc_pmu_print_counter_desc(const char *ev)
345 const struct pmu_events_map *pme;
346 const struct pmu_event *pe;
348 if ((pme = pmu_events_map_get(NULL)) == NULL)
350 for (pe = pme->table; pe->name || pe->desc || pe->event; pe++) {
351 if (pe->name == NULL)
353 if (strcasestr(pe->name, ev) != NULL &&
355 printf("%s:\t%s\n", pe->name, pe->desc);
360 pmc_pmu_print_counter_desc_long(const char *ev)
362 const struct pmu_events_map *pme;
363 const struct pmu_event *pe;
365 if ((pme = pmu_events_map_get(NULL)) == NULL)
367 for (pe = pme->table; pe->name || pe->desc || pe->event; pe++) {
368 if (pe->name == NULL)
370 if (strcasestr(pe->name, ev) != NULL) {
371 if (pe->long_desc != NULL)
372 printf("%s:\n%s\n", pe->name, pe->long_desc);
373 else if (pe->desc != NULL)
374 printf("%s:\t%s\n", pe->name, pe->desc);
380 pmc_pmu_print_counter_full(const char *ev)
382 const struct pmu_events_map *pme;
383 const struct pmu_event *pe;
385 if ((pme = pmu_events_map_get(NULL)) == NULL)
387 for (pe = pme->table; pe->name || pe->desc || pe->event; pe++) {
388 if (pe->name == NULL)
390 if (strcasestr(pe->name, ev) == NULL)
392 printf("name: %s\n", pe->name);
393 if (pe->long_desc != NULL)
394 printf("desc: %s\n", pe->long_desc);
395 else if (pe->desc != NULL)
396 printf("desc: %s\n", pe->desc);
397 if (pe->event != NULL)
398 printf("event: %s\n", pe->event);
399 if (pe->topic != NULL)
400 printf("topic: %s\n", pe->topic);
402 printf("pmu: %s\n", pe->pmu);
403 if (pe->unit != NULL)
404 printf("unit: %s\n", pe->unit);
405 if (pe->perpkg != NULL)
406 printf("perpkg: %s\n", pe->perpkg);
407 if (pe->metric_expr != NULL)
408 printf("metric_expr: %s\n", pe->metric_expr);
409 if (pe->metric_name != NULL)
410 printf("metric_name: %s\n", pe->metric_name);
411 if (pe->metric_group != NULL)
412 printf("metric_group: %s\n", pe->metric_group);
417 pmc_pmu_amd_pmcallocate(const char *event_name, struct pmc_op_pmcallocate *pm,
418 struct pmu_event_desc *ped)
420 struct pmc_md_amd_op_pmcallocate *amd;
421 const struct pmu_event *pe;
424 amd = &pm->pm_md.pm_amd;
425 if (ped->ped_umask > 0) {
426 pm->pm_caps |= PMC_CAP_QUALIFIER;
427 amd->pm_amd_config |= AMD_PMC_TO_UNITMASK(ped->ped_umask);
429 pm->pm_class = PMC_CLASS_K8;
430 pe = pmu_event_get(NULL, event_name, &idx);
432 if (strcmp("l3cache", pe->topic) == 0){
433 amd->pm_amd_config |= AMD_PMC_TO_EVENTMASK(ped->ped_event);
434 amd->pm_amd_sub_class = PMC_AMD_SUB_CLASS_L3_CACHE;
435 amd->pm_amd_config |= AMD_PMC_TO_L3SLICE(ped->ped_l3_slice);
436 amd->pm_amd_config |= AMD_PMC_TO_L3CORE(ped->ped_l3_thread);
438 else if (strcmp("data fabric", pe->topic) == 0){
440 amd->pm_amd_config |= AMD_PMC_TO_EVENTMASK_DF(ped->ped_event);
441 amd->pm_amd_sub_class = PMC_AMD_SUB_CLASS_DATA_FABRIC;
444 amd->pm_amd_config |= AMD_PMC_TO_EVENTMASK(ped->ped_event);
445 amd->pm_amd_sub_class = PMC_AMD_SUB_CLASS_CORE;
446 if ((pm->pm_caps & (PMC_CAP_USER|PMC_CAP_SYSTEM)) == 0 ||
447 (pm->pm_caps & (PMC_CAP_USER|PMC_CAP_SYSTEM)) ==
448 (PMC_CAP_USER|PMC_CAP_SYSTEM))
449 amd->pm_amd_config |= (AMD_PMC_USR | AMD_PMC_OS);
450 else if (pm->pm_caps & PMC_CAP_USER)
451 amd->pm_amd_config |= AMD_PMC_USR;
452 else if (pm->pm_caps & PMC_CAP_SYSTEM)
453 amd->pm_amd_config |= AMD_PMC_OS;
455 amd->pm_amd_config |= AMD_PMC_EDGE;
457 amd->pm_amd_config |= AMD_PMC_EDGE;
458 if (pm->pm_caps & PMC_CAP_INTERRUPT)
459 amd->pm_amd_config |= AMD_PMC_INT;
465 pmc_pmu_intel_pmcallocate(const char *event_name, struct pmc_op_pmcallocate *pm,
466 struct pmu_event_desc *ped)
468 struct pmc_md_iap_op_pmcallocate *iap;
472 iap = &pm->pm_md.pm_iap;
473 if (strcasestr(event_name, "UNC_") == event_name ||
474 strcasestr(event_name, "uncore") != NULL) {
475 pm->pm_class = PMC_CLASS_UCP;
476 pm->pm_caps |= PMC_CAP_QUALIFIER;
477 } else if ((ped->ped_umask == -1) ||
478 (ped->ped_event == 0x0 && ped->ped_umask == 0x3)) {
479 pm->pm_class = PMC_CLASS_IAF;
481 pm->pm_class = PMC_CLASS_IAP;
482 pm->pm_caps |= PMC_CAP_QUALIFIER;
484 iap->pm_iap_config |= IAP_EVSEL(ped->ped_event);
485 if (ped->ped_umask > 0)
486 iap->pm_iap_config |= IAP_UMASK(ped->ped_umask);
487 iap->pm_iap_config |= IAP_CMASK(ped->ped_cmask);
488 iap->pm_iap_rsp = ped->ped_offcore_rsp;
490 if ((pm->pm_caps & (PMC_CAP_USER|PMC_CAP_SYSTEM)) == 0 ||
491 (pm->pm_caps & (PMC_CAP_USER|PMC_CAP_SYSTEM)) ==
492 (PMC_CAP_USER|PMC_CAP_SYSTEM))
493 iap->pm_iap_config |= (IAP_USR | IAP_OS);
494 else if (pm->pm_caps & PMC_CAP_USER)
495 iap->pm_iap_config |= IAP_USR;
496 else if (pm->pm_caps & PMC_CAP_SYSTEM)
497 iap->pm_iap_config |= IAP_OS;
499 iap->pm_iap_config |= IAP_EDGE;
501 iap->pm_iap_config |= IAP_ANY;
503 iap->pm_iap_config |= IAP_EDGE;
504 if (pm->pm_caps & PMC_CAP_INTERRUPT)
505 iap->pm_iap_config |= IAP_INT;
510 pmc_pmu_pmcallocate(const char *event_name, struct pmc_op_pmcallocate *pm)
512 const struct pmu_event *pe;
513 struct pmu_event_desc ped;
517 if ((mfr = pmu_events_mfr()) == PMU_INVALID)
520 bzero(&pm->pm_md, sizeof(pm->pm_md));
521 pm->pm_caps |= (PMC_CAP_READ | PMC_CAP_WRITE);
522 event_name = pmu_alias_get(event_name);
523 if ((pe = pmu_event_get(NULL, event_name, &idx)) == NULL)
525 if (pe->alias && (pe = pmu_event_get(NULL, pe->alias, &idx)) == NULL)
530 if (pe->event == NULL)
532 if (pmu_parse_event(&ped, pe->event))
535 if (mfr == PMU_INTEL)
536 return (pmc_pmu_intel_pmcallocate(event_name, pm, &ped));
538 return (pmc_pmu_amd_pmcallocate(event_name, pm, &ped));
542 * Ultimately rely on AMD calling theirs the same
544 static const char *stat_mode_cntrs[] = {
545 "cpu_clk_unhalted.thread",
547 "br_inst_retired.all_branches",
548 "br_misp_retired.all_branches",
549 "longest_lat_cache.reference",
550 "longest_lat_cache.miss",
554 pmc_pmu_stat_mode(const char ***cntrs)
556 if (pmc_pmu_enabled()) {
557 *cntrs = stat_mode_cntrs;
566 pmc_pmu_sample_rate_get(const char *event_name __unused)
568 return (DEFAULT_SAMPLE_COUNT);
572 pmc_pmu_print_counters(const char *event_name __unused)
577 pmc_pmu_print_counter_desc(const char *e __unused)
582 pmc_pmu_print_counter_desc_long(const char *e __unused)
587 pmc_pmu_print_counter_full(const char *e __unused)
593 pmc_pmu_enabled(void)
599 pmc_pmu_pmcallocate(const char *e __unused, struct pmc_op_pmcallocate *p __unused)
605 pmc_pmu_event_get_by_idx(const char *c __unused, int idx __unused)
611 pmc_pmu_stat_mode(const char ***a __unused)
617 pmc_pmu_idx_get_by_event(const char *c __unused, const char *e __unused)