xref: /linux/tools/perf/util/env.c (revision be59dba332e1e8edd3e88d991ba0e4795ae2bcb2) 
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 "util/rwsem.h"
7 #include <linux/compiler.h>
8 #include <linux/ctype.h>
9 #include <linux/rbtree.h>
10 #include <linux/string.h>
11 #include <linux/zalloc.h>
12 #include "cgroup.h"
13 #include <errno.h>
14 #include <sys/utsname.h>
15 #include <stdlib.h>
16 #include <string.h>
17 #include "pmu.h"
18 #include "pmus.h"
19 #include "strbuf.h"
20 #include "trace/beauty/beauty.h"
21 
22 struct perf_env perf_env;
23 
24 #ifdef HAVE_LIBBPF_SUPPORT
25 #include "bpf-event.h"
26 #include "bpf-utils.h"
27 #include <bpf/libbpf.h>
28 
29 bool perf_env__insert_bpf_prog_info(struct perf_env *env,
30 				    struct bpf_prog_info_node *info_node)
31 {
32 	bool ret;
33 
34 	down_write(&env->bpf_progs.lock);
35 	ret = __perf_env__insert_bpf_prog_info(env, info_node);
36 	up_write(&env->bpf_progs.lock);
37 
38 	return ret;
39 }
40 
41 bool __perf_env__insert_bpf_prog_info(struct perf_env *env, struct bpf_prog_info_node *info_node)
42 {
43 	__u32 prog_id = info_node->info_linear->info.id;
44 	struct bpf_prog_info_node *node;
45 	struct rb_node *parent = NULL;
46 	struct rb_node **p;
47 
48 	p = &env->bpf_progs.infos.rb_node;
49 
50 	while (*p != NULL) {
51 		parent = *p;
52 		node = rb_entry(parent, struct bpf_prog_info_node, rb_node);
53 		if (prog_id < node->info_linear->info.id) {
54 			p = &(*p)->rb_left;
55 		} else if (prog_id > node->info_linear->info.id) {
56 			p = &(*p)->rb_right;
57 		} else {
58 			pr_debug("duplicated bpf prog info %u\n", prog_id);
59 			return false;
60 		}
61 	}
62 
63 	rb_link_node(&info_node->rb_node, parent, p);
64 	rb_insert_color(&info_node->rb_node, &env->bpf_progs.infos);
65 	env->bpf_progs.infos_cnt++;
66 	return true;
67 }
68 
69 struct bpf_prog_info_node *perf_env__find_bpf_prog_info(struct perf_env *env,
70 							__u32 prog_id)
71 {
72 	struct bpf_prog_info_node *node = NULL;
73 	struct rb_node *n;
74 
75 	down_read(&env->bpf_progs.lock);
76 	n = env->bpf_progs.infos.rb_node;
77 
78 	while (n) {
79 		node = rb_entry(n, struct bpf_prog_info_node, rb_node);
80 		if (prog_id < node->info_linear->info.id)
81 			n = n->rb_left;
82 		else if (prog_id > node->info_linear->info.id)
83 			n = n->rb_right;
84 		else
85 			goto out;
86 	}
87 	node = NULL;
88 
89 out:
90 	up_read(&env->bpf_progs.lock);
91 	return node;
92 }
93 
94 void perf_env__iterate_bpf_prog_info(struct perf_env *env,
95 				     void (*cb)(struct bpf_prog_info_node *node,
96 						void *data),
97 				     void *data)
98 {
99 	struct rb_node *first;
100 
101 	down_read(&env->bpf_progs.lock);
102 	first = rb_first(&env->bpf_progs.infos);
103 	for (struct rb_node *node = first; node != NULL; node = rb_next(node))
104 		(*cb)(rb_entry(node, struct bpf_prog_info_node, rb_node), data);
105 	up_read(&env->bpf_progs.lock);
106 }
107 
108 bool perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node)
109 {
110 	bool ret;
111 
112 	down_write(&env->bpf_progs.lock);
113 	ret = __perf_env__insert_btf(env, btf_node);
114 	up_write(&env->bpf_progs.lock);
115 	return ret;
116 }
117 
118 bool __perf_env__insert_btf(struct perf_env *env, struct btf_node *btf_node)
119 {
120 	struct rb_node *parent = NULL;
121 	__u32 btf_id = btf_node->id;
122 	struct btf_node *node;
123 	struct rb_node **p;
124 
125 	p = &env->bpf_progs.btfs.rb_node;
126 
127 	while (*p != NULL) {
128 		parent = *p;
129 		node = rb_entry(parent, struct btf_node, rb_node);
130 		if (btf_id < node->id) {
131 			p = &(*p)->rb_left;
132 		} else if (btf_id > node->id) {
133 			p = &(*p)->rb_right;
134 		} else {
135 			pr_debug("duplicated btf %u\n", btf_id);
136 			return false;
137 		}
138 	}
139 
140 	rb_link_node(&btf_node->rb_node, parent, p);
141 	rb_insert_color(&btf_node->rb_node, &env->bpf_progs.btfs);
142 	env->bpf_progs.btfs_cnt++;
143 	return true;
144 }
145 
146 struct btf_node *perf_env__find_btf(struct perf_env *env, __u32 btf_id)
147 {
148 	struct btf_node *res;
149 
150 	down_read(&env->bpf_progs.lock);
151 	res = __perf_env__find_btf(env, btf_id);
152 	up_read(&env->bpf_progs.lock);
153 	return res;
154 }
155 
156 struct btf_node *__perf_env__find_btf(struct perf_env *env, __u32 btf_id)
157 {
158 	struct btf_node *node = NULL;
159 	struct rb_node *n;
160 
161 	n = env->bpf_progs.btfs.rb_node;
162 
163 	while (n) {
164 		node = rb_entry(n, struct btf_node, rb_node);
165 		if (btf_id < node->id)
166 			n = n->rb_left;
167 		else if (btf_id > node->id)
168 			n = n->rb_right;
169 		else
170 			return node;
171 	}
172 	return NULL;
173 }
174 
175 /* purge data in bpf_progs.infos tree */
176 static void perf_env__purge_bpf(struct perf_env *env)
177 {
178 	struct rb_root *root;
179 	struct rb_node *next;
180 
181 	down_write(&env->bpf_progs.lock);
182 
183 	root = &env->bpf_progs.infos;
184 	next = rb_first(root);
185 
186 	while (next) {
187 		struct bpf_prog_info_node *node;
188 
189 		node = rb_entry(next, struct bpf_prog_info_node, rb_node);
190 		next = rb_next(&node->rb_node);
191 		rb_erase(&node->rb_node, root);
192 		zfree(&node->info_linear);
193 		bpf_metadata_free(node->metadata);
194 		free(node);
195 	}
196 
197 	env->bpf_progs.infos_cnt = 0;
198 
199 	root = &env->bpf_progs.btfs;
200 	next = rb_first(root);
201 
202 	while (next) {
203 		struct btf_node *node;
204 
205 		node = rb_entry(next, struct btf_node, rb_node);
206 		next = rb_next(&node->rb_node);
207 		rb_erase(&node->rb_node, root);
208 		free(node);
209 	}
210 
211 	env->bpf_progs.btfs_cnt = 0;
212 
213 	up_write(&env->bpf_progs.lock);
214 }
215 #else // HAVE_LIBBPF_SUPPORT
216 static void perf_env__purge_bpf(struct perf_env *env __maybe_unused)
217 {
218 }
219 #endif // HAVE_LIBBPF_SUPPORT
220 
221 void perf_env__exit(struct perf_env *env)
222 {
223 	int i, j;
224 
225 	perf_env__purge_bpf(env);
226 	perf_env__purge_cgroups(env);
227 	zfree(&env->hostname);
228 	zfree(&env->os_release);
229 	zfree(&env->version);
230 	zfree(&env->arch);
231 	zfree(&env->cpu_desc);
232 	zfree(&env->cpuid);
233 	zfree(&env->cmdline);
234 	zfree(&env->cmdline_argv);
235 	zfree(&env->sibling_dies);
236 	zfree(&env->sibling_cores);
237 	zfree(&env->sibling_threads);
238 	zfree(&env->pmu_mappings);
239 	zfree(&env->cpu);
240 	for (i = 0; i < env->nr_cpu_pmu_caps; i++)
241 		zfree(&env->cpu_pmu_caps[i]);
242 	zfree(&env->cpu_pmu_caps);
243 	zfree(&env->numa_map);
244 
245 	for (i = 0; i < env->nr_numa_nodes; i++)
246 		perf_cpu_map__put(env->numa_nodes[i].map);
247 	zfree(&env->numa_nodes);
248 
249 	for (i = 0; i < env->caches_cnt; i++)
250 		cpu_cache_level__free(&env->caches[i]);
251 	zfree(&env->caches);
252 
253 	for (i = 0; i < env->nr_memory_nodes; i++)
254 		zfree(&env->memory_nodes[i].set);
255 	zfree(&env->memory_nodes);
256 
257 	for (i = 0; i < env->nr_hybrid_nodes; i++) {
258 		zfree(&env->hybrid_nodes[i].pmu_name);
259 		zfree(&env->hybrid_nodes[i].cpus);
260 	}
261 	zfree(&env->hybrid_nodes);
262 
263 	for (i = 0; i < env->nr_pmus_with_caps; i++) {
264 		for (j = 0; j < env->pmu_caps[i].nr_caps; j++)
265 			zfree(&env->pmu_caps[i].caps[j]);
266 		zfree(&env->pmu_caps[i].caps);
267 		zfree(&env->pmu_caps[i].pmu_name);
268 	}
269 	zfree(&env->pmu_caps);
270 }
271 
272 void perf_env__init(struct perf_env *env)
273 {
274 #ifdef HAVE_LIBBPF_SUPPORT
275 	env->bpf_progs.infos = RB_ROOT;
276 	env->bpf_progs.btfs = RB_ROOT;
277 	init_rwsem(&env->bpf_progs.lock);
278 #endif
279 	env->kernel_is_64_bit = -1;
280 }
281 
282 static void perf_env__init_kernel_mode(struct perf_env *env)
283 {
284 	const char *arch = perf_env__raw_arch(env);
285 
286 	if (!strncmp(arch, "x86_64", 6) || !strncmp(arch, "aarch64", 7) ||
287 	    !strncmp(arch, "arm64", 5) || !strncmp(arch, "mips64", 6) ||
288 	    !strncmp(arch, "parisc64", 8) || !strncmp(arch, "riscv64", 7) ||
289 	    !strncmp(arch, "s390x", 5) || !strncmp(arch, "sparc64", 7))
290 		env->kernel_is_64_bit = 1;
291 	else
292 		env->kernel_is_64_bit = 0;
293 }
294 
295 int perf_env__kernel_is_64_bit(struct perf_env *env)
296 {
297 	if (env->kernel_is_64_bit == -1)
298 		perf_env__init_kernel_mode(env);
299 
300 	return env->kernel_is_64_bit;
301 }
302 
303 int perf_env__set_cmdline(struct perf_env *env, int argc, const char *argv[])
304 {
305 	int i;
306 
307 	/* do not include NULL termination */
308 	env->cmdline_argv = calloc(argc, sizeof(char *));
309 	if (env->cmdline_argv == NULL)
310 		goto out_enomem;
311 
312 	/*
313 	 * Must copy argv contents because it gets moved around during option
314 	 * parsing:
315 	 */
316 	for (i = 0; i < argc ; i++) {
317 		env->cmdline_argv[i] = argv[i];
318 		if (env->cmdline_argv[i] == NULL)
319 			goto out_free;
320 	}
321 
322 	env->nr_cmdline = argc;
323 
324 	return 0;
325 out_free:
326 	zfree(&env->cmdline_argv);
327 out_enomem:
328 	return -ENOMEM;
329 }
330 
331 int perf_env__read_cpu_topology_map(struct perf_env *env)
332 {
333 	int idx, nr_cpus;
334 
335 	if (env->cpu != NULL)
336 		return 0;
337 
338 	if (env->nr_cpus_avail == 0)
339 		env->nr_cpus_avail = cpu__max_present_cpu().cpu;
340 
341 	nr_cpus = env->nr_cpus_avail;
342 	if (nr_cpus == -1)
343 		return -EINVAL;
344 
345 	env->cpu = calloc(nr_cpus, sizeof(env->cpu[0]));
346 	if (env->cpu == NULL)
347 		return -ENOMEM;
348 
349 	for (idx = 0; idx < nr_cpus; ++idx) {
350 		struct perf_cpu cpu = { .cpu = idx };
351 		int core_id   = cpu__get_core_id(cpu);
352 		int socket_id = cpu__get_socket_id(cpu);
353 		int die_id    = cpu__get_die_id(cpu);
354 
355 		env->cpu[idx].core_id	= core_id >= 0 ? core_id : -1;
356 		env->cpu[idx].socket_id	= socket_id >= 0 ? socket_id : -1;
357 		env->cpu[idx].die_id	= die_id >= 0 ? die_id : -1;
358 	}
359 
360 	env->nr_cpus_avail = nr_cpus;
361 	return 0;
362 }
363 
364 int perf_env__read_pmu_mappings(struct perf_env *env)
365 {
366 	struct perf_pmu *pmu = NULL;
367 	u32 pmu_num = 0;
368 	struct strbuf sb;
369 
370 	while ((pmu = perf_pmus__scan(pmu)))
371 		pmu_num++;
372 
373 	if (!pmu_num) {
374 		pr_debug("pmu mappings not available\n");
375 		return -ENOENT;
376 	}
377 	env->nr_pmu_mappings = pmu_num;
378 
379 	if (strbuf_init(&sb, 128 * pmu_num) < 0)
380 		return -ENOMEM;
381 
382 	while ((pmu = perf_pmus__scan(pmu))) {
383 		if (strbuf_addf(&sb, "%u:%s", pmu->type, pmu->name) < 0)
384 			goto error;
385 		/* include a NULL character at the end */
386 		if (strbuf_add(&sb, "", 1) < 0)
387 			goto error;
388 	}
389 
390 	env->pmu_mappings = strbuf_detach(&sb, NULL);
391 
392 	return 0;
393 
394 error:
395 	strbuf_release(&sb);
396 	return -1;
397 }
398 
399 int perf_env__read_cpuid(struct perf_env *env)
400 {
401 	char cpuid[128];
402 	struct perf_cpu cpu = {-1};
403 	int err = get_cpuid(cpuid, sizeof(cpuid), cpu);
404 
405 	if (err)
406 		return err;
407 
408 	free(env->cpuid);
409 	env->cpuid = strdup(cpuid);
410 	if (env->cpuid == NULL)
411 		return ENOMEM;
412 	return 0;
413 }
414 
415 static int perf_env__read_arch(struct perf_env *env)
416 {
417 	struct utsname uts;
418 
419 	if (env->arch)
420 		return 0;
421 
422 	if (!uname(&uts))
423 		env->arch = strdup(uts.machine);
424 
425 	return env->arch ? 0 : -ENOMEM;
426 }
427 
428 static int perf_env__read_nr_cpus_avail(struct perf_env *env)
429 {
430 	if (env->nr_cpus_avail == 0)
431 		env->nr_cpus_avail = cpu__max_present_cpu().cpu;
432 
433 	return env->nr_cpus_avail ? 0 : -ENOENT;
434 }
435 
436 const char *perf_env__raw_arch(struct perf_env *env)
437 {
438 	return env && !perf_env__read_arch(env) ? env->arch : "unknown";
439 }
440 
441 int perf_env__nr_cpus_avail(struct perf_env *env)
442 {
443 	return env && !perf_env__read_nr_cpus_avail(env) ? env->nr_cpus_avail : 0;
444 }
445 
446 void cpu_cache_level__free(struct cpu_cache_level *cache)
447 {
448 	zfree(&cache->type);
449 	zfree(&cache->map);
450 	zfree(&cache->size);
451 }
452 
453 /*
454  * Return architecture name in a normalized form.
455  * The conversion logic comes from the Makefile.
456  */
457 static const char *normalize_arch(char *arch)
458 {
459 	if (!strcmp(arch, "x86_64"))
460 		return "x86";
461 	if (arch[0] == 'i' && arch[2] == '8' && arch[3] == '6')
462 		return "x86";
463 	if (!strcmp(arch, "sun4u") || !strncmp(arch, "sparc", 5))
464 		return "sparc";
465 	if (!strncmp(arch, "aarch64", 7) || !strncmp(arch, "arm64", 5))
466 		return "arm64";
467 	if (!strncmp(arch, "arm", 3) || !strcmp(arch, "sa110"))
468 		return "arm";
469 	if (!strncmp(arch, "s390", 4))
470 		return "s390";
471 	if (!strncmp(arch, "parisc", 6))
472 		return "parisc";
473 	if (!strncmp(arch, "powerpc", 7) || !strncmp(arch, "ppc", 3))
474 		return "powerpc";
475 	if (!strncmp(arch, "mips", 4))
476 		return "mips";
477 	if (!strncmp(arch, "sh", 2) && isdigit(arch[2]))
478 		return "sh";
479 	if (!strncmp(arch, "loongarch", 9))
480 		return "loongarch";
481 
482 	return arch;
483 }
484 
485 const char *perf_env__arch(struct perf_env *env)
486 {
487 	char *arch_name;
488 
489 	if (!env || !env->arch) { /* Assume local operation */
490 		static struct utsname uts = { .machine[0] = '\0', };
491 		if (uts.machine[0] == '\0' && uname(&uts) < 0)
492 			return NULL;
493 		arch_name = uts.machine;
494 	} else
495 		arch_name = env->arch;
496 
497 	return normalize_arch(arch_name);
498 }
499 
500 #if defined(HAVE_LIBTRACEEVENT)
501 #include "trace/beauty/arch_errno_names.c"
502 #endif
503 
504 const char *perf_env__arch_strerrno(struct perf_env *env __maybe_unused, int err __maybe_unused)
505 {
506 #if defined(HAVE_LIBTRACEEVENT)
507 	if (env->arch_strerrno == NULL)
508 		env->arch_strerrno = arch_syscalls__strerrno_function(perf_env__arch(env));
509 
510 	return env->arch_strerrno ? env->arch_strerrno(err) : "no arch specific strerrno function";
511 #else
512 	return "!HAVE_LIBTRACEEVENT";
513 #endif
514 }
515 
516 const char *perf_env__cpuid(struct perf_env *env)
517 {
518 	int status;
519 
520 	if (!env->cpuid) { /* Assume local operation */
521 		status = perf_env__read_cpuid(env);
522 		if (status)
523 			return NULL;
524 	}
525 
526 	return env->cpuid;
527 }
528 
529 int perf_env__nr_pmu_mappings(struct perf_env *env)
530 {
531 	int status;
532 
533 	if (!env->nr_pmu_mappings) { /* Assume local operation */
534 		status = perf_env__read_pmu_mappings(env);
535 		if (status)
536 			return 0;
537 	}
538 
539 	return env->nr_pmu_mappings;
540 }
541 
542 const char *perf_env__pmu_mappings(struct perf_env *env)
543 {
544 	int status;
545 
546 	if (!env->pmu_mappings) { /* Assume local operation */
547 		status = perf_env__read_pmu_mappings(env);
548 		if (status)
549 			return NULL;
550 	}
551 
552 	return env->pmu_mappings;
553 }
554 
555 int perf_env__numa_node(struct perf_env *env, struct perf_cpu cpu)
556 {
557 	if (!env->nr_numa_map) {
558 		struct numa_node *nn;
559 		int i, nr = 0;
560 
561 		for (i = 0; i < env->nr_numa_nodes; i++) {
562 			nn = &env->numa_nodes[i];
563 			nr = max(nr, (int)perf_cpu_map__max(nn->map).cpu);
564 		}
565 
566 		nr++;
567 
568 		/*
569 		 * We initialize the numa_map array to prepare
570 		 * it for missing cpus, which return node -1
571 		 */
572 		env->numa_map = malloc(nr * sizeof(int));
573 		if (!env->numa_map)
574 			return -1;
575 
576 		for (i = 0; i < nr; i++)
577 			env->numa_map[i] = -1;
578 
579 		env->nr_numa_map = nr;
580 
581 		for (i = 0; i < env->nr_numa_nodes; i++) {
582 			struct perf_cpu tmp;
583 			int j;
584 
585 			nn = &env->numa_nodes[i];
586 			perf_cpu_map__for_each_cpu(tmp, j, nn->map)
587 				env->numa_map[tmp.cpu] = i;
588 		}
589 	}
590 
591 	return cpu.cpu >= 0 && cpu.cpu < env->nr_numa_map ? env->numa_map[cpu.cpu] : -1;
592 }
593 
594 bool perf_env__has_pmu_mapping(struct perf_env *env, const char *pmu_name)
595 {
596 	char *pmu_mapping = env->pmu_mappings, *colon;
597 
598 	for (int i = 0; i < env->nr_pmu_mappings; ++i) {
599 		if (strtoul(pmu_mapping, &colon, 0) == ULONG_MAX || *colon != ':')
600 			goto out_error;
601 
602 		pmu_mapping = colon + 1;
603 		if (strcmp(pmu_mapping, pmu_name) == 0)
604 			return true;
605 
606 		pmu_mapping += strlen(pmu_mapping) + 1;
607 	}
608 out_error:
609 	return false;
610 }
611 
612 char *perf_env__find_pmu_cap(struct perf_env *env, const char *pmu_name,
613 			     const char *cap)
614 {
615 	char *cap_eq;
616 	int cap_size;
617 	char **ptr;
618 	int i, j;
619 
620 	if (!pmu_name || !cap)
621 		return NULL;
622 
623 	cap_size = strlen(cap);
624 	cap_eq = zalloc(cap_size + 2);
625 	if (!cap_eq)
626 		return NULL;
627 
628 	memcpy(cap_eq, cap, cap_size);
629 	cap_eq[cap_size] = '=';
630 
631 	if (!strcmp(pmu_name, "cpu")) {
632 		for (i = 0; i < env->nr_cpu_pmu_caps; i++) {
633 			if (!strncmp(env->cpu_pmu_caps[i], cap_eq, cap_size + 1)) {
634 				free(cap_eq);
635 				return &env->cpu_pmu_caps[i][cap_size + 1];
636 			}
637 		}
638 		goto out;
639 	}
640 
641 	for (i = 0; i < env->nr_pmus_with_caps; i++) {
642 		if (strcmp(env->pmu_caps[i].pmu_name, pmu_name))
643 			continue;
644 
645 		ptr = env->pmu_caps[i].caps;
646 
647 		for (j = 0; j < env->pmu_caps[i].nr_caps; j++) {
648 			if (!strncmp(ptr[j], cap_eq, cap_size + 1)) {
649 				free(cap_eq);
650 				return &ptr[j][cap_size + 1];
651 			}
652 		}
653 	}
654 
655 out:
656 	free(cap_eq);
657 	return NULL;
658 }
659 
660 void perf_env__find_br_cntr_info(struct perf_env *env,
661 				 unsigned int *nr,
662 				 unsigned int *width)
663 {
664 	if (nr) {
665 		*nr = env->cpu_pmu_caps ? env->br_cntr_nr :
666 					  env->pmu_caps->br_cntr_nr;
667 	}
668 
669 	if (width) {
670 		*width = env->cpu_pmu_caps ? env->br_cntr_width :
671 					     env->pmu_caps->br_cntr_width;
672 	}
673 }
674 
675 bool perf_env__is_x86_amd_cpu(struct perf_env *env)
676 {
677 	static int is_amd; /* 0: Uninitialized, 1: Yes, -1: No */
678 
679 	if (is_amd == 0)
680 		is_amd = env->cpuid && strstarts(env->cpuid, "AuthenticAMD") ? 1 : -1;
681 
682 	return is_amd >= 1 ? true : false;
683 }
684 
685 bool x86__is_amd_cpu(void)
686 {
687 	struct perf_env env = { .total_mem = 0, };
688 	bool is_amd;
689 
690 	perf_env__cpuid(&env);
691 	is_amd = perf_env__is_x86_amd_cpu(&env);
692 	perf_env__exit(&env);
693 
694 	return is_amd;
695 }
696