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