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