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