Loading...
1#include "cpumap.h"
2#include "env.h"
3#include "util.h"
4
5struct perf_env perf_env;
6
7void perf_env__exit(struct perf_env *env)
8{
9 int i;
10
11 zfree(&env->hostname);
12 zfree(&env->os_release);
13 zfree(&env->version);
14 zfree(&env->arch);
15 zfree(&env->cpu_desc);
16 zfree(&env->cpuid);
17 zfree(&env->cmdline);
18 zfree(&env->cmdline_argv);
19 zfree(&env->sibling_cores);
20 zfree(&env->sibling_threads);
21 zfree(&env->numa_nodes);
22 zfree(&env->pmu_mappings);
23 zfree(&env->cpu);
24
25 for (i = 0; i < env->caches_cnt; i++)
26 cpu_cache_level__free(&env->caches[i]);
27 zfree(&env->caches);
28}
29
30int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
31{
32 int i;
33
34 /* do not include NULL termination */
35 env->cmdline_argv = calloc(argc, sizeof(char *));
36 if (env->cmdline_argv == NULL)
37 goto out_enomem;
38
39 /*
40 * Must copy argv contents because it gets moved around during option
41 * parsing:
42 */
43 for (i = 0; i < argc ; i++) {
44 env->cmdline_argv[i] = argv[i];
45 if (env->cmdline_argv[i] == NULL)
46 goto out_free;
47 }
48
49 env->nr_cmdline = argc;
50
51 return 0;
52out_free:
53 zfree(&env->cmdline_argv);
54out_enomem:
55 return -ENOMEM;
56}
57
58int perf_env__read_cpu_topology_map(struct perf_env *env)
59{
60 int cpu, nr_cpus;
61
62 if (env->cpu != NULL)
63 return 0;
64
65 if (env->nr_cpus_avail == 0)
66 env->nr_cpus_avail = sysconf(_SC_NPROCESSORS_CONF);
67
68 nr_cpus = env->nr_cpus_avail;
69 if (nr_cpus == -1)
70 return -EINVAL;
71
72 env->cpu = calloc(nr_cpus, sizeof(env->cpu[0]));
73 if (env->cpu == NULL)
74 return -ENOMEM;
75
76 for (cpu = 0; cpu < nr_cpus; ++cpu) {
77 env->cpu[cpu].core_id = cpu_map__get_core_id(cpu);
78 env->cpu[cpu].socket_id = cpu_map__get_socket_id(cpu);
79 }
80
81 env->nr_cpus_avail = nr_cpus;
82 return 0;
83}
84
85void cpu_cache_level__free(struct cpu_cache_level *cache)
86{
87 free(cache->type);
88 free(cache->map);
89 free(cache->size);
90}
1// SPDX-License-Identifier: GPL-2.0
2#include "cpumap.h"
3#include "debug.h"
4#include "env.h"
5#include "util/header.h"
6#include "linux/compiler.h"
7#include <linux/ctype.h>
8#include <linux/string.h>
9#include <linux/zalloc.h>
10#include "cgroup.h"
11#include <errno.h>
12#include <sys/utsname.h>
13#include <stdlib.h>
14#include <string.h>
15#include "pmu.h"
16#include "pmus.h"
17#include "strbuf.h"
18#include "trace/beauty/beauty.h"
19
20struct perf_env perf_env;
21
22#ifdef HAVE_LIBBPF_SUPPORT
23#include "bpf-event.h"
24#include "bpf-utils.h"
25#include <bpf/libbpf.h>
26
27bool perf_env__insert_bpf_prog_info(struct perf_env *env,
28 struct bpf_prog_info_node *info_node)
29{
30 bool ret;
31
32 down_write(&env->bpf_progs.lock);
33 ret = __perf_env__insert_bpf_prog_info(env, info_node);
34 up_write(&env->bpf_progs.lock);
35
36 return ret;
37}
38
39bool __perf_env__insert_bpf_prog_info(struct perf_env *env, struct bpf_prog_info_node *info_node)
40{
41 __u32 prog_id = info_node->info_linear->info.id;
42 struct bpf_prog_info_node *node;
43 struct rb_node *parent = NULL;
44 struct rb_node **p;
45
46 p = &env->bpf_progs.infos.rb_node;
47
48 while (*p != NULL) {
49 parent = *p;
50 node = rb_entry(parent, struct bpf_prog_info_node, rb_node);
51 if (prog_id < node->info_linear->info.id) {
52 p = &(*p)->rb_left;
53 } else if (prog_id > node->info_linear->info.id) {
54 p = &(*p)->rb_right;
55 } else {
56 pr_debug("duplicated bpf prog info %u\n", prog_id);
57 return false;
58 }
59 }
60
61 rb_link_node(&info_node->rb_node, parent, p);
62 rb_insert_color(&info_node->rb_node, &env->bpf_progs.infos);
63 env->bpf_progs.infos_cnt++;
64 return true;
65}
66
67struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
68 __u32 prog_id)
69{
70 struct bpf_prog_info_node *node = NULL;
71 struct rb_node *n;
72
73 down_read(&env->bpf_progs.lock);
74 n = env->bpf_progs.infos.rb_node;
75
76 while (n) {
77 node = rb_entry(n, struct bpf_prog_info_node, rb_node);
78 if (prog_id < node->info_linear->info.id)
79 n = n->rb_left;
80 else if (prog_id > node->info_linear->info.id)
81 n = n->rb_right;
82 else
83 goto out;
84 }
85 node = NULL;
86
87out:
88 up_read(&env->bpf_progs.lock);
89 return node;
90}
91
92bool perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node)
93{
94 bool ret;
95
96 down_write(&env->bpf_progs.lock);
97 ret = __perf_env__insert_btf(env, btf_node);
98 up_write(&env->bpf_progs.lock);
99 return ret;
100}
101
102bool __perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node)
103{
104 struct rb_node *parent = NULL;
105 __u32 btf_id = btf_node->id;
106 struct btf_node *node;
107 struct rb_node **p;
108
109 p = &env->bpf_progs.btfs.rb_node;
110
111 while (*p != NULL) {
112 parent = *p;
113 node = rb_entry(parent, struct btf_node, rb_node);
114 if (btf_id < node->id) {
115 p = &(*p)->rb_left;
116 } else if (btf_id > node->id) {
117 p = &(*p)->rb_right;
118 } else {
119 pr_debug("duplicated btf %u\n", btf_id);
120 return false;
121 }
122 }
123
124 rb_link_node(&btf_node->rb_node, parent, p);
125 rb_insert_color(&btf_node->rb_node, &env->bpf_progs.btfs);
126 env->bpf_progs.btfs_cnt++;
127 return true;
128}
129
130struct btf_node *perf_env__find_btf(struct perf_env *env, __u32 btf_id)
131{
132 struct btf_node *res;
133
134 down_read(&env->bpf_progs.lock);
135 res = __perf_env__find_btf(env, btf_id);
136 up_read(&env->bpf_progs.lock);
137 return res;
138}
139
140struct btf_node *__perf_env__find_btf(struct perf_env *env, __u32 btf_id)
141{
142 struct btf_node *node = NULL;
143 struct rb_node *n;
144
145 n = env->bpf_progs.btfs.rb_node;
146
147 while (n) {
148 node = rb_entry(n, struct btf_node, rb_node);
149 if (btf_id < node->id)
150 n = n->rb_left;
151 else if (btf_id > node->id)
152 n = n->rb_right;
153 else
154 return node;
155 }
156 return NULL;
157}
158
159/* purge data in bpf_progs.infos tree */
160static void perf_env__purge_bpf(struct perf_env *env)
161{
162 struct rb_root *root;
163 struct rb_node *next;
164
165 down_write(&env->bpf_progs.lock);
166
167 root = &env->bpf_progs.infos;
168 next = rb_first(root);
169
170 while (next) {
171 struct bpf_prog_info_node *node;
172
173 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
174 next = rb_next(&node->rb_node);
175 rb_erase(&node->rb_node, root);
176 zfree(&node->info_linear);
177 free(node);
178 }
179
180 env->bpf_progs.infos_cnt = 0;
181
182 root = &env->bpf_progs.btfs;
183 next = rb_first(root);
184
185 while (next) {
186 struct btf_node *node;
187
188 node = rb_entry(next, struct btf_node, rb_node);
189 next = rb_next(&node->rb_node);
190 rb_erase(&node->rb_node, root);
191 free(node);
192 }
193
194 env->bpf_progs.btfs_cnt = 0;
195
196 up_write(&env->bpf_progs.lock);
197}
198#else // HAVE_LIBBPF_SUPPORT
199static void perf_env__purge_bpf(struct perf_env *env __maybe_unused)
200{
201}
202#endif // HAVE_LIBBPF_SUPPORT
203
204void perf_env__exit(struct perf_env *env)
205{
206 int i, j;
207
208 perf_env__purge_bpf(env);
209 perf_env__purge_cgroups(env);
210 zfree(&env->hostname);
211 zfree(&env->os_release);
212 zfree(&env->version);
213 zfree(&env->arch);
214 zfree(&env->cpu_desc);
215 zfree(&env->cpuid);
216 zfree(&env->cmdline);
217 zfree(&env->cmdline_argv);
218 zfree(&env->sibling_dies);
219 zfree(&env->sibling_cores);
220 zfree(&env->sibling_threads);
221 zfree(&env->pmu_mappings);
222 zfree(&env->cpu);
223 for (i = 0; i < env->nr_cpu_pmu_caps; i++)
224 zfree(&env->cpu_pmu_caps[i]);
225 zfree(&env->cpu_pmu_caps);
226 zfree(&env->numa_map);
227
228 for (i = 0; i < env->nr_numa_nodes; i++)
229 perf_cpu_map__put(env->numa_nodes[i].map);
230 zfree(&env->numa_nodes);
231
232 for (i = 0; i < env->caches_cnt; i++)
233 cpu_cache_level__free(&env->caches[i]);
234 zfree(&env->caches);
235
236 for (i = 0; i < env->nr_memory_nodes; i++)
237 zfree(&env->memory_nodes[i].set);
238 zfree(&env->memory_nodes);
239
240 for (i = 0; i < env->nr_hybrid_nodes; i++) {
241 zfree(&env->hybrid_nodes[i].pmu_name);
242 zfree(&env->hybrid_nodes[i].cpus);
243 }
244 zfree(&env->hybrid_nodes);
245
246 for (i = 0; i < env->nr_pmus_with_caps; i++) {
247 for (j = 0; j < env->pmu_caps[i].nr_caps; j++)
248 zfree(&env->pmu_caps[i].caps[j]);
249 zfree(&env->pmu_caps[i].caps);
250 zfree(&env->pmu_caps[i].pmu_name);
251 }
252 zfree(&env->pmu_caps);
253}
254
255void perf_env__init(struct perf_env *env)
256{
257#ifdef HAVE_LIBBPF_SUPPORT
258 env->bpf_progs.infos = RB_ROOT;
259 env->bpf_progs.btfs = RB_ROOT;
260 init_rwsem(&env->bpf_progs.lock);
261#endif
262 env->kernel_is_64_bit = -1;
263}
264
265static void perf_env__init_kernel_mode(struct perf_env *env)
266{
267 const char *arch = perf_env__raw_arch(env);
268
269 if (!strncmp(arch, "x86_64", 6) || !strncmp(arch, "aarch64", 7) ||
270 !strncmp(arch, "arm64", 5) || !strncmp(arch, "mips64", 6) ||
271 !strncmp(arch, "parisc64", 8) || !strncmp(arch, "riscv64", 7) ||
272 !strncmp(arch, "s390x", 5) || !strncmp(arch, "sparc64", 7))
273 env->kernel_is_64_bit = 1;
274 else
275 env->kernel_is_64_bit = 0;
276}
277
278int perf_env__kernel_is_64_bit(struct perf_env *env)
279{
280 if (env->kernel_is_64_bit == -1)
281 perf_env__init_kernel_mode(env);
282
283 return env->kernel_is_64_bit;
284}
285
286int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
287{
288 int i;
289
290 /* do not include NULL termination */
291 env->cmdline_argv = calloc(argc, sizeof(char *));
292 if (env->cmdline_argv == NULL)
293 goto out_enomem;
294
295 /*
296 * Must copy argv contents because it gets moved around during option
297 * parsing:
298 */
299 for (i = 0; i < argc ; i++) {
300 env->cmdline_argv[i] = argv[i];
301 if (env->cmdline_argv[i] == NULL)
302 goto out_free;
303 }
304
305 env->nr_cmdline = argc;
306
307 return 0;
308out_free:
309 zfree(&env->cmdline_argv);
310out_enomem:
311 return -ENOMEM;
312}
313
314int perf_env__read_cpu_topology_map(struct perf_env *env)
315{
316 int idx, nr_cpus;
317
318 if (env->cpu != NULL)
319 return 0;
320
321 if (env->nr_cpus_avail == 0)
322 env->nr_cpus_avail = cpu__max_present_cpu().cpu;
323
324 nr_cpus = env->nr_cpus_avail;
325 if (nr_cpus == -1)
326 return -EINVAL;
327
328 env->cpu = calloc(nr_cpus, sizeof(env->cpu[0]));
329 if (env->cpu == NULL)
330 return -ENOMEM;
331
332 for (idx = 0; idx < nr_cpus; ++idx) {
333 struct perf_cpu cpu = { .cpu = idx };
334
335 env->cpu[idx].core_id = cpu__get_core_id(cpu);
336 env->cpu[idx].socket_id = cpu__get_socket_id(cpu);
337 env->cpu[idx].die_id = cpu__get_die_id(cpu);
338 }
339
340 env->nr_cpus_avail = nr_cpus;
341 return 0;
342}
343
344int perf_env__read_pmu_mappings(struct perf_env *env)
345{
346 struct perf_pmu *pmu = NULL;
347 u32 pmu_num = 0;
348 struct strbuf sb;
349
350 while ((pmu = perf_pmus__scan(pmu)))
351 pmu_num++;
352
353 if (!pmu_num) {
354 pr_debug("pmu mappings not available\n");
355 return -ENOENT;
356 }
357 env->nr_pmu_mappings = pmu_num;
358
359 if (strbuf_init(&sb, 128 * pmu_num) < 0)
360 return -ENOMEM;
361
362 while ((pmu = perf_pmus__scan(pmu))) {
363 if (strbuf_addf(&sb, "%u:%s", pmu->type, pmu->name) < 0)
364 goto error;
365 /* include a NULL character at the end */
366 if (strbuf_add(&sb, "", 1) < 0)
367 goto error;
368 }
369
370 env->pmu_mappings = strbuf_detach(&sb, NULL);
371
372 return 0;
373
374error:
375 strbuf_release(&sb);
376 return -1;
377}
378
379int perf_env__read_cpuid(struct perf_env *env)
380{
381 char cpuid[128];
382 struct perf_cpu cpu = {-1};
383 int err = get_cpuid(cpuid, sizeof(cpuid), cpu);
384
385 if (err)
386 return err;
387
388 free(env->cpuid);
389 env->cpuid = strdup(cpuid);
390 if (env->cpuid == NULL)
391 return ENOMEM;
392 return 0;
393}
394
395static int perf_env__read_arch(struct perf_env *env)
396{
397 struct utsname uts;
398
399 if (env->arch)
400 return 0;
401
402 if (!uname(&uts))
403 env->arch = strdup(uts.machine);
404
405 return env->arch ? 0 : -ENOMEM;
406}
407
408static int perf_env__read_nr_cpus_avail(struct perf_env *env)
409{
410 if (env->nr_cpus_avail == 0)
411 env->nr_cpus_avail = cpu__max_present_cpu().cpu;
412
413 return env->nr_cpus_avail ? 0 : -ENOENT;
414}
415
416const char *perf_env__raw_arch(struct perf_env *env)
417{
418 return env && !perf_env__read_arch(env) ? env->arch : "unknown";
419}
420
421int perf_env__nr_cpus_avail(struct perf_env *env)
422{
423 return env && !perf_env__read_nr_cpus_avail(env) ? env->nr_cpus_avail : 0;
424}
425
426void cpu_cache_level__free(struct cpu_cache_level *cache)
427{
428 zfree(&cache->type);
429 zfree(&cache->map);
430 zfree(&cache->size);
431}
432
433/*
434 * Return architecture name in a normalized form.
435 * The conversion logic comes from the Makefile.
436 */
437static const char *normalize_arch(char *arch)
438{
439 if (!strcmp(arch, "x86_64"))
440 return "x86";
441 if (arch[0] == 'i' && arch[2] == '8' && arch[3] == '6')
442 return "x86";
443 if (!strcmp(arch, "sun4u") || !strncmp(arch, "sparc", 5))
444 return "sparc";
445 if (!strncmp(arch, "aarch64", 7) || !strncmp(arch, "arm64", 5))
446 return "arm64";
447 if (!strncmp(arch, "arm", 3) || !strcmp(arch, "sa110"))
448 return "arm";
449 if (!strncmp(arch, "s390", 4))
450 return "s390";
451 if (!strncmp(arch, "parisc", 6))
452 return "parisc";
453 if (!strncmp(arch, "powerpc", 7) || !strncmp(arch, "ppc", 3))
454 return "powerpc";
455 if (!strncmp(arch, "mips", 4))
456 return "mips";
457 if (!strncmp(arch, "sh", 2) && isdigit(arch[2]))
458 return "sh";
459 if (!strncmp(arch, "loongarch", 9))
460 return "loongarch";
461
462 return arch;
463}
464
465const char *perf_env__arch(struct perf_env *env)
466{
467 char *arch_name;
468
469 if (!env || !env->arch) { /* Assume local operation */
470 static struct utsname uts = { .machine[0] = '\0', };
471 if (uts.machine[0] == '\0' && uname(&uts) < 0)
472 return NULL;
473 arch_name = uts.machine;
474 } else
475 arch_name = env->arch;
476
477 return normalize_arch(arch_name);
478}
479
480const char *perf_env__arch_strerrno(struct perf_env *env __maybe_unused, int err __maybe_unused)
481{
482#if defined(HAVE_SYSCALL_TABLE_SUPPORT) && defined(HAVE_LIBTRACEEVENT)
483 if (env->arch_strerrno == NULL)
484 env->arch_strerrno = arch_syscalls__strerrno_function(perf_env__arch(env));
485
486 return env->arch_strerrno ? env->arch_strerrno(err) : "no arch specific strerrno function";
487#else
488 return "!(HAVE_SYSCALL_TABLE_SUPPORT && HAVE_LIBTRACEEVENT)";
489#endif
490}
491
492const char *perf_env__cpuid(struct perf_env *env)
493{
494 int status;
495
496 if (!env->cpuid) { /* Assume local operation */
497 status = perf_env__read_cpuid(env);
498 if (status)
499 return NULL;
500 }
501
502 return env->cpuid;
503}
504
505int perf_env__nr_pmu_mappings(struct perf_env *env)
506{
507 int status;
508
509 if (!env->nr_pmu_mappings) { /* Assume local operation */
510 status = perf_env__read_pmu_mappings(env);
511 if (status)
512 return 0;
513 }
514
515 return env->nr_pmu_mappings;
516}
517
518const char *perf_env__pmu_mappings(struct perf_env *env)
519{
520 int status;
521
522 if (!env->pmu_mappings) { /* Assume local operation */
523 status = perf_env__read_pmu_mappings(env);
524 if (status)
525 return NULL;
526 }
527
528 return env->pmu_mappings;
529}
530
531int perf_env__numa_node(struct perf_env *env, struct perf_cpu cpu)
532{
533 if (!env->nr_numa_map) {
534 struct numa_node *nn;
535 int i, nr = 0;
536
537 for (i = 0; i < env->nr_numa_nodes; i++) {
538 nn = &env->numa_nodes[i];
539 nr = max(nr, perf_cpu_map__max(nn->map).cpu);
540 }
541
542 nr++;
543
544 /*
545 * We initialize the numa_map array to prepare
546 * it for missing cpus, which return node -1
547 */
548 env->numa_map = malloc(nr * sizeof(int));
549 if (!env->numa_map)
550 return -1;
551
552 for (i = 0; i < nr; i++)
553 env->numa_map[i] = -1;
554
555 env->nr_numa_map = nr;
556
557 for (i = 0; i < env->nr_numa_nodes; i++) {
558 struct perf_cpu tmp;
559 int j;
560
561 nn = &env->numa_nodes[i];
562 perf_cpu_map__for_each_cpu(tmp, j, nn->map)
563 env->numa_map[tmp.cpu] = i;
564 }
565 }
566
567 return cpu.cpu >= 0 && cpu.cpu < env->nr_numa_map ? env->numa_map[cpu.cpu] : -1;
568}
569
570bool perf_env__has_pmu_mapping(struct perf_env *env, const char *pmu_name)
571{
572 char *pmu_mapping = env->pmu_mappings, *colon;
573
574 for (int i = 0; i < env->nr_pmu_mappings; ++i) {
575 if (strtoul(pmu_mapping, &colon, 0) == ULONG_MAX || *colon != ':')
576 goto out_error;
577
578 pmu_mapping = colon + 1;
579 if (strcmp(pmu_mapping, pmu_name) == 0)
580 return true;
581
582 pmu_mapping += strlen(pmu_mapping) + 1;
583 }
584out_error:
585 return false;
586}
587
588char *perf_env__find_pmu_cap(struct perf_env *env, const char *pmu_name,
589 const char *cap)
590{
591 char *cap_eq;
592 int cap_size;
593 char **ptr;
594 int i, j;
595
596 if (!pmu_name || !cap)
597 return NULL;
598
599 cap_size = strlen(cap);
600 cap_eq = zalloc(cap_size + 2);
601 if (!cap_eq)
602 return NULL;
603
604 memcpy(cap_eq, cap, cap_size);
605 cap_eq[cap_size] = '=';
606
607 if (!strcmp(pmu_name, "cpu")) {
608 for (i = 0; i < env->nr_cpu_pmu_caps; i++) {
609 if (!strncmp(env->cpu_pmu_caps[i], cap_eq, cap_size + 1)) {
610 free(cap_eq);
611 return &env->cpu_pmu_caps[i][cap_size + 1];
612 }
613 }
614 goto out;
615 }
616
617 for (i = 0; i < env->nr_pmus_with_caps; i++) {
618 if (strcmp(env->pmu_caps[i].pmu_name, pmu_name))
619 continue;
620
621 ptr = env->pmu_caps[i].caps;
622
623 for (j = 0; j < env->pmu_caps[i].nr_caps; j++) {
624 if (!strncmp(ptr[j], cap_eq, cap_size + 1)) {
625 free(cap_eq);
626 return &ptr[j][cap_size + 1];
627 }
628 }
629 }
630
631out:
632 free(cap_eq);
633 return NULL;
634}
635
636void perf_env__find_br_cntr_info(struct perf_env *env,
637 unsigned int *nr,
638 unsigned int *width)
639{
640 if (nr) {
641 *nr = env->cpu_pmu_caps ? env->br_cntr_nr :
642 env->pmu_caps->br_cntr_nr;
643 }
644
645 if (width) {
646 *width = env->cpu_pmu_caps ? env->br_cntr_width :
647 env->pmu_caps->br_cntr_width;
648 }
649}
650
651bool perf_env__is_x86_amd_cpu(struct perf_env *env)
652{
653 static int is_amd; /* 0: Uninitialized, 1: Yes, -1: No */
654
655 if (is_amd == 0)
656 is_amd = env->cpuid && strstarts(env->cpuid, "AuthenticAMD") ? 1 : -1;
657
658 return is_amd >= 1 ? true : false;
659}
660
661bool x86__is_amd_cpu(void)
662{
663 struct perf_env env = { .total_mem = 0, };
664 bool is_amd;
665
666 perf_env__cpuid(&env);
667 is_amd = perf_env__is_x86_amd_cpu(&env);
668 perf_env__exit(&env);
669
670 return is_amd;
671}