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