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