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
20 struct 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
perf_env__insert_bpf_prog_info(struct perf_env * env,struct bpf_prog_info_node * info_node)27 bool 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
__perf_env__insert_bpf_prog_info(struct perf_env * env,struct bpf_prog_info_node * info_node)39 bool __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
perf_env__find_bpf_prog_info(struct perf_env * env,__u32 prog_id)67 struct 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
87 out:
88 up_read(&env->bpf_progs.lock);
89 return node;
90 }
91
perf_env__insert_btf(struct perf_env * env,struct btf_node * btf_node)92 bool 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
__perf_env__insert_btf(struct perf_env * env,struct btf_node * btf_node)102 bool __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
perf_env__find_btf(struct perf_env * env,__u32 btf_id)130 struct 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
__perf_env__find_btf(struct perf_env * env,__u32 btf_id)140 struct 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 */
perf_env__purge_bpf(struct perf_env * env)160 static 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
perf_env__purge_bpf(struct perf_env * env __maybe_unused)199 static void perf_env__purge_bpf(struct perf_env *env __maybe_unused)
200 {
201 }
202 #endif // HAVE_LIBBPF_SUPPORT
203
perf_env__exit(struct perf_env * env)204 void 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
perf_env__init(struct perf_env * env)255 void 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
perf_env__init_kernel_mode(struct perf_env * env)265 static 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
perf_env__kernel_is_64_bit(struct perf_env * env)278 int 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
perf_env__set_cmdline(struct perf_env * env,int argc,const char * argv[])286 int 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;
308 out_free:
309 zfree(&env->cmdline_argv);
310 out_enomem:
311 return -ENOMEM;
312 }
313
perf_env__read_cpu_topology_map(struct perf_env * env)314 int 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 int core_id = cpu__get_core_id(cpu);
335 int socket_id = cpu__get_socket_id(cpu);
336 int die_id = cpu__get_die_id(cpu);
337
338 env->cpu[idx].core_id = core_id >= 0 ? core_id : -1;
339 env->cpu[idx].socket_id = socket_id >= 0 ? socket_id : -1;
340 env->cpu[idx].die_id = die_id >= 0 ? die_id : -1;
341 }
342
343 env->nr_cpus_avail = nr_cpus;
344 return 0;
345 }
346
perf_env__read_pmu_mappings(struct perf_env * env)347 int perf_env__read_pmu_mappings(struct perf_env *env)
348 {
349 struct perf_pmu *pmu = NULL;
350 u32 pmu_num = 0;
351 struct strbuf sb;
352
353 while ((pmu = perf_pmus__scan(pmu)))
354 pmu_num++;
355
356 if (!pmu_num) {
357 pr_debug("pmu mappings not available\n");
358 return -ENOENT;
359 }
360 env->nr_pmu_mappings = pmu_num;
361
362 if (strbuf_init(&sb, 128 * pmu_num) < 0)
363 return -ENOMEM;
364
365 while ((pmu = perf_pmus__scan(pmu))) {
366 if (strbuf_addf(&sb, "%u:%s", pmu->type, pmu->name) < 0)
367 goto error;
368 /* include a NULL character at the end */
369 if (strbuf_add(&sb, "", 1) < 0)
370 goto error;
371 }
372
373 env->pmu_mappings = strbuf_detach(&sb, NULL);
374
375 return 0;
376
377 error:
378 strbuf_release(&sb);
379 return -1;
380 }
381
perf_env__read_cpuid(struct perf_env * env)382 int perf_env__read_cpuid(struct perf_env *env)
383 {
384 char cpuid[128];
385 struct perf_cpu cpu = {-1};
386 int err = get_cpuid(cpuid, sizeof(cpuid), cpu);
387
388 if (err)
389 return err;
390
391 free(env->cpuid);
392 env->cpuid = strdup(cpuid);
393 if (env->cpuid == NULL)
394 return ENOMEM;
395 return 0;
396 }
397
perf_env__read_arch(struct perf_env * env)398 static int perf_env__read_arch(struct perf_env *env)
399 {
400 struct utsname uts;
401
402 if (env->arch)
403 return 0;
404
405 if (!uname(&uts))
406 env->arch = strdup(uts.machine);
407
408 return env->arch ? 0 : -ENOMEM;
409 }
410
perf_env__read_nr_cpus_avail(struct perf_env * env)411 static int perf_env__read_nr_cpus_avail(struct perf_env *env)
412 {
413 if (env->nr_cpus_avail == 0)
414 env->nr_cpus_avail = cpu__max_present_cpu().cpu;
415
416 return env->nr_cpus_avail ? 0 : -ENOENT;
417 }
418
perf_env__raw_arch(struct perf_env * env)419 const char *perf_env__raw_arch(struct perf_env *env)
420 {
421 return env && !perf_env__read_arch(env) ? env->arch : "unknown";
422 }
423
perf_env__nr_cpus_avail(struct perf_env * env)424 int perf_env__nr_cpus_avail(struct perf_env *env)
425 {
426 return env && !perf_env__read_nr_cpus_avail(env) ? env->nr_cpus_avail : 0;
427 }
428
cpu_cache_level__free(struct cpu_cache_level * cache)429 void cpu_cache_level__free(struct cpu_cache_level *cache)
430 {
431 zfree(&cache->type);
432 zfree(&cache->map);
433 zfree(&cache->size);
434 }
435
436 /*
437 * Return architecture name in a normalized form.
438 * The conversion logic comes from the Makefile.
439 */
normalize_arch(char * arch)440 static const char *normalize_arch(char *arch)
441 {
442 if (!strcmp(arch, "x86_64"))
443 return "x86";
444 if (arch[0] == 'i' && arch[2] == '8' && arch[3] == '6')
445 return "x86";
446 if (!strcmp(arch, "sun4u") || !strncmp(arch, "sparc", 5))
447 return "sparc";
448 if (!strncmp(arch, "aarch64", 7) || !strncmp(arch, "arm64", 5))
449 return "arm64";
450 if (!strncmp(arch, "arm", 3) || !strcmp(arch, "sa110"))
451 return "arm";
452 if (!strncmp(arch, "s390", 4))
453 return "s390";
454 if (!strncmp(arch, "parisc", 6))
455 return "parisc";
456 if (!strncmp(arch, "powerpc", 7) || !strncmp(arch, "ppc", 3))
457 return "powerpc";
458 if (!strncmp(arch, "mips", 4))
459 return "mips";
460 if (!strncmp(arch, "sh", 2) && isdigit(arch[2]))
461 return "sh";
462 if (!strncmp(arch, "loongarch", 9))
463 return "loongarch";
464
465 return arch;
466 }
467
perf_env__arch(struct perf_env * env)468 const char *perf_env__arch(struct perf_env *env)
469 {
470 char *arch_name;
471
472 if (!env || !env->arch) { /* Assume local operation */
473 static struct utsname uts = { .machine[0] = '\0', };
474 if (uts.machine[0] == '\0' && uname(&uts) < 0)
475 return NULL;
476 arch_name = uts.machine;
477 } else
478 arch_name = env->arch;
479
480 return normalize_arch(arch_name);
481 }
482
483 #if defined(HAVE_LIBTRACEEVENT)
484 #include "trace/beauty/arch_errno_names.c"
485 #endif
486
perf_env__arch_strerrno(struct perf_env * env __maybe_unused,int err __maybe_unused)487 const char *perf_env__arch_strerrno(struct perf_env *env __maybe_unused, int err __maybe_unused)
488 {
489 #if defined(HAVE_LIBTRACEEVENT)
490 if (env->arch_strerrno == NULL)
491 env->arch_strerrno = arch_syscalls__strerrno_function(perf_env__arch(env));
492
493 return env->arch_strerrno ? env->arch_strerrno(err) : "no arch specific strerrno function";
494 #else
495 return "!HAVE_LIBTRACEEVENT";
496 #endif
497 }
498
perf_env__cpuid(struct perf_env * env)499 const char *perf_env__cpuid(struct perf_env *env)
500 {
501 int status;
502
503 if (!env->cpuid) { /* Assume local operation */
504 status = perf_env__read_cpuid(env);
505 if (status)
506 return NULL;
507 }
508
509 return env->cpuid;
510 }
511
perf_env__nr_pmu_mappings(struct perf_env * env)512 int perf_env__nr_pmu_mappings(struct perf_env *env)
513 {
514 int status;
515
516 if (!env->nr_pmu_mappings) { /* Assume local operation */
517 status = perf_env__read_pmu_mappings(env);
518 if (status)
519 return 0;
520 }
521
522 return env->nr_pmu_mappings;
523 }
524
perf_env__pmu_mappings(struct perf_env * env)525 const char *perf_env__pmu_mappings(struct perf_env *env)
526 {
527 int status;
528
529 if (!env->pmu_mappings) { /* Assume local operation */
530 status = perf_env__read_pmu_mappings(env);
531 if (status)
532 return NULL;
533 }
534
535 return env->pmu_mappings;
536 }
537
perf_env__numa_node(struct perf_env * env,struct perf_cpu cpu)538 int perf_env__numa_node(struct perf_env *env, struct perf_cpu cpu)
539 {
540 if (!env->nr_numa_map) {
541 struct numa_node *nn;
542 int i, nr = 0;
543
544 for (i = 0; i < env->nr_numa_nodes; i++) {
545 nn = &env->numa_nodes[i];
546 nr = max(nr, perf_cpu_map__max(nn->map).cpu);
547 }
548
549 nr++;
550
551 /*
552 * We initialize the numa_map array to prepare
553 * it for missing cpus, which return node -1
554 */
555 env->numa_map = malloc(nr * sizeof(int));
556 if (!env->numa_map)
557 return -1;
558
559 for (i = 0; i < nr; i++)
560 env->numa_map[i] = -1;
561
562 env->nr_numa_map = nr;
563
564 for (i = 0; i < env->nr_numa_nodes; i++) {
565 struct perf_cpu tmp;
566 int j;
567
568 nn = &env->numa_nodes[i];
569 perf_cpu_map__for_each_cpu(tmp, j, nn->map)
570 env->numa_map[tmp.cpu] = i;
571 }
572 }
573
574 return cpu.cpu >= 0 && cpu.cpu < env->nr_numa_map ? env->numa_map[cpu.cpu] : -1;
575 }
576
perf_env__has_pmu_mapping(struct perf_env * env,const char * pmu_name)577 bool perf_env__has_pmu_mapping(struct perf_env *env, const char *pmu_name)
578 {
579 char *pmu_mapping = env->pmu_mappings, *colon;
580
581 for (int i = 0; i < env->nr_pmu_mappings; ++i) {
582 if (strtoul(pmu_mapping, &colon, 0) == ULONG_MAX || *colon != ':')
583 goto out_error;
584
585 pmu_mapping = colon + 1;
586 if (strcmp(pmu_mapping, pmu_name) == 0)
587 return true;
588
589 pmu_mapping += strlen(pmu_mapping) + 1;
590 }
591 out_error:
592 return false;
593 }
594
perf_env__find_pmu_cap(struct perf_env * env,const char * pmu_name,const char * cap)595 char *perf_env__find_pmu_cap(struct perf_env *env, const char *pmu_name,
596 const char *cap)
597 {
598 char *cap_eq;
599 int cap_size;
600 char **ptr;
601 int i, j;
602
603 if (!pmu_name || !cap)
604 return NULL;
605
606 cap_size = strlen(cap);
607 cap_eq = zalloc(cap_size + 2);
608 if (!cap_eq)
609 return NULL;
610
611 memcpy(cap_eq, cap, cap_size);
612 cap_eq[cap_size] = '=';
613
614 if (!strcmp(pmu_name, "cpu")) {
615 for (i = 0; i < env->nr_cpu_pmu_caps; i++) {
616 if (!strncmp(env->cpu_pmu_caps[i], cap_eq, cap_size + 1)) {
617 free(cap_eq);
618 return &env->cpu_pmu_caps[i][cap_size + 1];
619 }
620 }
621 goto out;
622 }
623
624 for (i = 0; i < env->nr_pmus_with_caps; i++) {
625 if (strcmp(env->pmu_caps[i].pmu_name, pmu_name))
626 continue;
627
628 ptr = env->pmu_caps[i].caps;
629
630 for (j = 0; j < env->pmu_caps[i].nr_caps; j++) {
631 if (!strncmp(ptr[j], cap_eq, cap_size + 1)) {
632 free(cap_eq);
633 return &ptr[j][cap_size + 1];
634 }
635 }
636 }
637
638 out:
639 free(cap_eq);
640 return NULL;
641 }
642
perf_env__find_br_cntr_info(struct perf_env * env,unsigned int * nr,unsigned int * width)643 void perf_env__find_br_cntr_info(struct perf_env *env,
644 unsigned int *nr,
645 unsigned int *width)
646 {
647 if (nr) {
648 *nr = env->cpu_pmu_caps ? env->br_cntr_nr :
649 env->pmu_caps->br_cntr_nr;
650 }
651
652 if (width) {
653 *width = env->cpu_pmu_caps ? env->br_cntr_width :
654 env->pmu_caps->br_cntr_width;
655 }
656 }
657
perf_env__is_x86_amd_cpu(struct perf_env * env)658 bool perf_env__is_x86_amd_cpu(struct perf_env *env)
659 {
660 static int is_amd; /* 0: Uninitialized, 1: Yes, -1: No */
661
662 if (is_amd == 0)
663 is_amd = env->cpuid && strstarts(env->cpuid, "AuthenticAMD") ? 1 : -1;
664
665 return is_amd >= 1 ? true : false;
666 }
667
x86__is_amd_cpu(void)668 bool x86__is_amd_cpu(void)
669 {
670 struct perf_env env = { .total_mem = 0, };
671 bool is_amd;
672
673 perf_env__cpuid(&env);
674 is_amd = perf_env__is_x86_amd_cpu(&env);
675 perf_env__exit(&env);
676
677 return is_amd;
678 }
679