b.liu | e958203 | 2025-04-17 19:18:16 +0800 | [diff] [blame^] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | #include <api/fs/fs.h> |
| 3 | #include "cpumap.h" |
| 4 | #include "debug.h" |
| 5 | #include "event.h" |
| 6 | #include <assert.h> |
| 7 | #include <dirent.h> |
| 8 | #include <stdio.h> |
| 9 | #include <stdlib.h> |
| 10 | #include <linux/bitmap.h> |
| 11 | #include "asm/bug.h" |
| 12 | |
| 13 | #include <linux/ctype.h> |
| 14 | #include <linux/zalloc.h> |
| 15 | |
| 16 | static int max_cpu_num; |
| 17 | static int max_present_cpu_num; |
| 18 | static int max_node_num; |
| 19 | static int *cpunode_map; |
| 20 | |
| 21 | static struct perf_cpu_map *cpu_map__from_entries(struct cpu_map_entries *cpus) |
| 22 | { |
| 23 | struct perf_cpu_map *map; |
| 24 | |
| 25 | map = perf_cpu_map__empty_new(cpus->nr); |
| 26 | if (map) { |
| 27 | unsigned i; |
| 28 | |
| 29 | for (i = 0; i < cpus->nr; i++) { |
| 30 | /* |
| 31 | * Special treatment for -1, which is not real cpu number, |
| 32 | * and we need to use (int) -1 to initialize map[i], |
| 33 | * otherwise it would become 65535. |
| 34 | */ |
| 35 | if (cpus->cpu[i] == (u16) -1) |
| 36 | map->map[i] = -1; |
| 37 | else |
| 38 | map->map[i] = (int) cpus->cpu[i]; |
| 39 | } |
| 40 | } |
| 41 | |
| 42 | return map; |
| 43 | } |
| 44 | |
| 45 | static struct perf_cpu_map *cpu_map__from_mask(struct perf_record_record_cpu_map *mask) |
| 46 | { |
| 47 | struct perf_cpu_map *map; |
| 48 | int nr, nbits = mask->nr * mask->long_size * BITS_PER_BYTE; |
| 49 | |
| 50 | nr = bitmap_weight(mask->mask, nbits); |
| 51 | |
| 52 | map = perf_cpu_map__empty_new(nr); |
| 53 | if (map) { |
| 54 | int cpu, i = 0; |
| 55 | |
| 56 | for_each_set_bit(cpu, mask->mask, nbits) |
| 57 | map->map[i++] = cpu; |
| 58 | } |
| 59 | return map; |
| 60 | |
| 61 | } |
| 62 | |
| 63 | struct perf_cpu_map *cpu_map__new_data(struct perf_record_cpu_map_data *data) |
| 64 | { |
| 65 | if (data->type == PERF_CPU_MAP__CPUS) |
| 66 | return cpu_map__from_entries((struct cpu_map_entries *)data->data); |
| 67 | else |
| 68 | return cpu_map__from_mask((struct perf_record_record_cpu_map *)data->data); |
| 69 | } |
| 70 | |
| 71 | size_t cpu_map__fprintf(struct perf_cpu_map *map, FILE *fp) |
| 72 | { |
| 73 | #define BUFSIZE 1024 |
| 74 | char buf[BUFSIZE]; |
| 75 | |
| 76 | cpu_map__snprint(map, buf, sizeof(buf)); |
| 77 | return fprintf(fp, "%s\n", buf); |
| 78 | #undef BUFSIZE |
| 79 | } |
| 80 | |
| 81 | struct perf_cpu_map *perf_cpu_map__empty_new(int nr) |
| 82 | { |
| 83 | struct perf_cpu_map *cpus = malloc(sizeof(*cpus) + sizeof(int) * nr); |
| 84 | |
| 85 | if (cpus != NULL) { |
| 86 | int i; |
| 87 | |
| 88 | cpus->nr = nr; |
| 89 | for (i = 0; i < nr; i++) |
| 90 | cpus->map[i] = -1; |
| 91 | |
| 92 | refcount_set(&cpus->refcnt, 1); |
| 93 | } |
| 94 | |
| 95 | return cpus; |
| 96 | } |
| 97 | |
| 98 | static int cpu__get_topology_int(int cpu, const char *name, int *value) |
| 99 | { |
| 100 | char path[PATH_MAX]; |
| 101 | |
| 102 | snprintf(path, PATH_MAX, |
| 103 | "devices/system/cpu/cpu%d/topology/%s", cpu, name); |
| 104 | |
| 105 | return sysfs__read_int(path, value); |
| 106 | } |
| 107 | |
| 108 | int cpu_map__get_socket_id(int cpu) |
| 109 | { |
| 110 | int value, ret = cpu__get_topology_int(cpu, "physical_package_id", &value); |
| 111 | return ret ?: value; |
| 112 | } |
| 113 | |
| 114 | int cpu_map__get_socket(struct perf_cpu_map *map, int idx, void *data __maybe_unused) |
| 115 | { |
| 116 | int cpu; |
| 117 | |
| 118 | if (idx > map->nr) |
| 119 | return -1; |
| 120 | |
| 121 | cpu = map->map[idx]; |
| 122 | |
| 123 | return cpu_map__get_socket_id(cpu); |
| 124 | } |
| 125 | |
| 126 | static int cmp_ids(const void *a, const void *b) |
| 127 | { |
| 128 | return *(int *)a - *(int *)b; |
| 129 | } |
| 130 | |
| 131 | int cpu_map__build_map(struct perf_cpu_map *cpus, struct perf_cpu_map **res, |
| 132 | int (*f)(struct perf_cpu_map *map, int cpu, void *data), |
| 133 | void *data) |
| 134 | { |
| 135 | struct perf_cpu_map *c; |
| 136 | int nr = cpus->nr; |
| 137 | int cpu, s1, s2; |
| 138 | |
| 139 | /* allocate as much as possible */ |
| 140 | c = calloc(1, sizeof(*c) + nr * sizeof(int)); |
| 141 | if (!c) |
| 142 | return -1; |
| 143 | |
| 144 | for (cpu = 0; cpu < nr; cpu++) { |
| 145 | s1 = f(cpus, cpu, data); |
| 146 | for (s2 = 0; s2 < c->nr; s2++) { |
| 147 | if (s1 == c->map[s2]) |
| 148 | break; |
| 149 | } |
| 150 | if (s2 == c->nr) { |
| 151 | c->map[c->nr] = s1; |
| 152 | c->nr++; |
| 153 | } |
| 154 | } |
| 155 | /* ensure we process id in increasing order */ |
| 156 | qsort(c->map, c->nr, sizeof(int), cmp_ids); |
| 157 | |
| 158 | refcount_set(&c->refcnt, 1); |
| 159 | *res = c; |
| 160 | return 0; |
| 161 | } |
| 162 | |
| 163 | int cpu_map__get_die_id(int cpu) |
| 164 | { |
| 165 | int value, ret = cpu__get_topology_int(cpu, "die_id", &value); |
| 166 | |
| 167 | return ret ?: value; |
| 168 | } |
| 169 | |
| 170 | int cpu_map__get_die(struct perf_cpu_map *map, int idx, void *data) |
| 171 | { |
| 172 | int cpu, die_id, s; |
| 173 | |
| 174 | if (idx > map->nr) |
| 175 | return -1; |
| 176 | |
| 177 | cpu = map->map[idx]; |
| 178 | |
| 179 | die_id = cpu_map__get_die_id(cpu); |
| 180 | /* There is no die_id on legacy system. */ |
| 181 | if (die_id == -1) |
| 182 | die_id = 0; |
| 183 | |
| 184 | s = cpu_map__get_socket(map, idx, data); |
| 185 | if (s == -1) |
| 186 | return -1; |
| 187 | |
| 188 | /* |
| 189 | * Encode socket in bit range 15:8 |
| 190 | * die_id is relative to socket, and |
| 191 | * we need a global id. So we combine |
| 192 | * socket + die id |
| 193 | */ |
| 194 | if (WARN_ONCE(die_id >> 8, "The die id number is too big.\n")) |
| 195 | return -1; |
| 196 | |
| 197 | if (WARN_ONCE(s >> 8, "The socket id number is too big.\n")) |
| 198 | return -1; |
| 199 | |
| 200 | return (s << 8) | (die_id & 0xff); |
| 201 | } |
| 202 | |
| 203 | int cpu_map__get_core_id(int cpu) |
| 204 | { |
| 205 | int value, ret = cpu__get_topology_int(cpu, "core_id", &value); |
| 206 | return ret ?: value; |
| 207 | } |
| 208 | |
| 209 | int cpu_map__get_core(struct perf_cpu_map *map, int idx, void *data) |
| 210 | { |
| 211 | int cpu, s_die; |
| 212 | |
| 213 | if (idx > map->nr) |
| 214 | return -1; |
| 215 | |
| 216 | cpu = map->map[idx]; |
| 217 | |
| 218 | cpu = cpu_map__get_core_id(cpu); |
| 219 | |
| 220 | /* s_die is the combination of socket + die id */ |
| 221 | s_die = cpu_map__get_die(map, idx, data); |
| 222 | if (s_die == -1) |
| 223 | return -1; |
| 224 | |
| 225 | /* |
| 226 | * encode socket in bit range 31:24 |
| 227 | * encode die id in bit range 23:16 |
| 228 | * core_id is relative to socket and die, |
| 229 | * we need a global id. So we combine |
| 230 | * socket + die id + core id |
| 231 | */ |
| 232 | if (WARN_ONCE(cpu >> 16, "The core id number is too big.\n")) |
| 233 | return -1; |
| 234 | |
| 235 | return (s_die << 16) | (cpu & 0xffff); |
| 236 | } |
| 237 | |
| 238 | int cpu_map__build_socket_map(struct perf_cpu_map *cpus, struct perf_cpu_map **sockp) |
| 239 | { |
| 240 | return cpu_map__build_map(cpus, sockp, cpu_map__get_socket, NULL); |
| 241 | } |
| 242 | |
| 243 | int cpu_map__build_die_map(struct perf_cpu_map *cpus, struct perf_cpu_map **diep) |
| 244 | { |
| 245 | return cpu_map__build_map(cpus, diep, cpu_map__get_die, NULL); |
| 246 | } |
| 247 | |
| 248 | int cpu_map__build_core_map(struct perf_cpu_map *cpus, struct perf_cpu_map **corep) |
| 249 | { |
| 250 | return cpu_map__build_map(cpus, corep, cpu_map__get_core, NULL); |
| 251 | } |
| 252 | |
| 253 | /* setup simple routines to easily access node numbers given a cpu number */ |
| 254 | static int get_max_num(char *path, int *max) |
| 255 | { |
| 256 | size_t num; |
| 257 | char *buf; |
| 258 | int err = 0; |
| 259 | |
| 260 | if (filename__read_str(path, &buf, &num)) |
| 261 | return -1; |
| 262 | |
| 263 | buf[num] = '\0'; |
| 264 | |
| 265 | /* start on the right, to find highest node num */ |
| 266 | while (--num) { |
| 267 | if ((buf[num] == ',') || (buf[num] == '-')) { |
| 268 | num++; |
| 269 | break; |
| 270 | } |
| 271 | } |
| 272 | if (sscanf(&buf[num], "%d", max) < 1) { |
| 273 | err = -1; |
| 274 | goto out; |
| 275 | } |
| 276 | |
| 277 | /* convert from 0-based to 1-based */ |
| 278 | (*max)++; |
| 279 | |
| 280 | out: |
| 281 | free(buf); |
| 282 | return err; |
| 283 | } |
| 284 | |
| 285 | /* Determine highest possible cpu in the system for sparse allocation */ |
| 286 | static void set_max_cpu_num(void) |
| 287 | { |
| 288 | const char *mnt; |
| 289 | char path[PATH_MAX]; |
| 290 | int ret = -1; |
| 291 | |
| 292 | /* set up default */ |
| 293 | max_cpu_num = 4096; |
| 294 | max_present_cpu_num = 4096; |
| 295 | |
| 296 | mnt = sysfs__mountpoint(); |
| 297 | if (!mnt) |
| 298 | goto out; |
| 299 | |
| 300 | /* get the highest possible cpu number for a sparse allocation */ |
| 301 | ret = snprintf(path, PATH_MAX, "%s/devices/system/cpu/possible", mnt); |
| 302 | if (ret >= PATH_MAX) { |
| 303 | pr_err("sysfs path crossed PATH_MAX(%d) size\n", PATH_MAX); |
| 304 | goto out; |
| 305 | } |
| 306 | |
| 307 | ret = get_max_num(path, &max_cpu_num); |
| 308 | if (ret) |
| 309 | goto out; |
| 310 | |
| 311 | /* get the highest present cpu number for a sparse allocation */ |
| 312 | ret = snprintf(path, PATH_MAX, "%s/devices/system/cpu/present", mnt); |
| 313 | if (ret >= PATH_MAX) { |
| 314 | pr_err("sysfs path crossed PATH_MAX(%d) size\n", PATH_MAX); |
| 315 | goto out; |
| 316 | } |
| 317 | |
| 318 | ret = get_max_num(path, &max_present_cpu_num); |
| 319 | |
| 320 | out: |
| 321 | if (ret) |
| 322 | pr_err("Failed to read max cpus, using default of %d\n", max_cpu_num); |
| 323 | } |
| 324 | |
| 325 | /* Determine highest possible node in the system for sparse allocation */ |
| 326 | static void set_max_node_num(void) |
| 327 | { |
| 328 | const char *mnt; |
| 329 | char path[PATH_MAX]; |
| 330 | int ret = -1; |
| 331 | |
| 332 | /* set up default */ |
| 333 | max_node_num = 8; |
| 334 | |
| 335 | mnt = sysfs__mountpoint(); |
| 336 | if (!mnt) |
| 337 | goto out; |
| 338 | |
| 339 | /* get the highest possible cpu number for a sparse allocation */ |
| 340 | ret = snprintf(path, PATH_MAX, "%s/devices/system/node/possible", mnt); |
| 341 | if (ret >= PATH_MAX) { |
| 342 | pr_err("sysfs path crossed PATH_MAX(%d) size\n", PATH_MAX); |
| 343 | goto out; |
| 344 | } |
| 345 | |
| 346 | ret = get_max_num(path, &max_node_num); |
| 347 | |
| 348 | out: |
| 349 | if (ret) |
| 350 | pr_err("Failed to read max nodes, using default of %d\n", max_node_num); |
| 351 | } |
| 352 | |
| 353 | int cpu__max_node(void) |
| 354 | { |
| 355 | if (unlikely(!max_node_num)) |
| 356 | set_max_node_num(); |
| 357 | |
| 358 | return max_node_num; |
| 359 | } |
| 360 | |
| 361 | int cpu__max_cpu(void) |
| 362 | { |
| 363 | if (unlikely(!max_cpu_num)) |
| 364 | set_max_cpu_num(); |
| 365 | |
| 366 | return max_cpu_num; |
| 367 | } |
| 368 | |
| 369 | int cpu__max_present_cpu(void) |
| 370 | { |
| 371 | if (unlikely(!max_present_cpu_num)) |
| 372 | set_max_cpu_num(); |
| 373 | |
| 374 | return max_present_cpu_num; |
| 375 | } |
| 376 | |
| 377 | |
| 378 | int cpu__get_node(int cpu) |
| 379 | { |
| 380 | if (unlikely(cpunode_map == NULL)) { |
| 381 | pr_debug("cpu_map not initialized\n"); |
| 382 | return -1; |
| 383 | } |
| 384 | |
| 385 | return cpunode_map[cpu]; |
| 386 | } |
| 387 | |
| 388 | static int init_cpunode_map(void) |
| 389 | { |
| 390 | int i; |
| 391 | |
| 392 | set_max_cpu_num(); |
| 393 | set_max_node_num(); |
| 394 | |
| 395 | cpunode_map = calloc(max_cpu_num, sizeof(int)); |
| 396 | if (!cpunode_map) { |
| 397 | pr_err("%s: calloc failed\n", __func__); |
| 398 | return -1; |
| 399 | } |
| 400 | |
| 401 | for (i = 0; i < max_cpu_num; i++) |
| 402 | cpunode_map[i] = -1; |
| 403 | |
| 404 | return 0; |
| 405 | } |
| 406 | |
| 407 | int cpu__setup_cpunode_map(void) |
| 408 | { |
| 409 | struct dirent *dent1, *dent2; |
| 410 | DIR *dir1, *dir2; |
| 411 | unsigned int cpu, mem; |
| 412 | char buf[PATH_MAX]; |
| 413 | char path[PATH_MAX]; |
| 414 | const char *mnt; |
| 415 | int n; |
| 416 | |
| 417 | /* initialize globals */ |
| 418 | if (init_cpunode_map()) |
| 419 | return -1; |
| 420 | |
| 421 | mnt = sysfs__mountpoint(); |
| 422 | if (!mnt) |
| 423 | return 0; |
| 424 | |
| 425 | n = snprintf(path, PATH_MAX, "%s/devices/system/node", mnt); |
| 426 | if (n >= PATH_MAX) { |
| 427 | pr_err("sysfs path crossed PATH_MAX(%d) size\n", PATH_MAX); |
| 428 | return -1; |
| 429 | } |
| 430 | |
| 431 | dir1 = opendir(path); |
| 432 | if (!dir1) |
| 433 | return 0; |
| 434 | |
| 435 | /* walk tree and setup map */ |
| 436 | while ((dent1 = readdir(dir1)) != NULL) { |
| 437 | if (dent1->d_type != DT_DIR || sscanf(dent1->d_name, "node%u", &mem) < 1) |
| 438 | continue; |
| 439 | |
| 440 | n = snprintf(buf, PATH_MAX, "%s/%s", path, dent1->d_name); |
| 441 | if (n >= PATH_MAX) { |
| 442 | pr_err("sysfs path crossed PATH_MAX(%d) size\n", PATH_MAX); |
| 443 | continue; |
| 444 | } |
| 445 | |
| 446 | dir2 = opendir(buf); |
| 447 | if (!dir2) |
| 448 | continue; |
| 449 | while ((dent2 = readdir(dir2)) != NULL) { |
| 450 | if (dent2->d_type != DT_LNK || sscanf(dent2->d_name, "cpu%u", &cpu) < 1) |
| 451 | continue; |
| 452 | cpunode_map[cpu] = mem; |
| 453 | } |
| 454 | closedir(dir2); |
| 455 | } |
| 456 | closedir(dir1); |
| 457 | return 0; |
| 458 | } |
| 459 | |
| 460 | bool cpu_map__has(struct perf_cpu_map *cpus, int cpu) |
| 461 | { |
| 462 | return perf_cpu_map__idx(cpus, cpu) != -1; |
| 463 | } |
| 464 | |
| 465 | int cpu_map__cpu(struct perf_cpu_map *cpus, int idx) |
| 466 | { |
| 467 | return cpus->map[idx]; |
| 468 | } |
| 469 | |
| 470 | size_t cpu_map__snprint(struct perf_cpu_map *map, char *buf, size_t size) |
| 471 | { |
| 472 | int i, cpu, start = -1; |
| 473 | bool first = true; |
| 474 | size_t ret = 0; |
| 475 | |
| 476 | #define COMMA first ? "" : "," |
| 477 | |
| 478 | for (i = 0; i < map->nr + 1; i++) { |
| 479 | bool last = i == map->nr; |
| 480 | |
| 481 | cpu = last ? INT_MAX : map->map[i]; |
| 482 | |
| 483 | if (start == -1) { |
| 484 | start = i; |
| 485 | if (last) { |
| 486 | ret += snprintf(buf + ret, size - ret, |
| 487 | "%s%d", COMMA, |
| 488 | map->map[i]); |
| 489 | } |
| 490 | } else if (((i - start) != (cpu - map->map[start])) || last) { |
| 491 | int end = i - 1; |
| 492 | |
| 493 | if (start == end) { |
| 494 | ret += snprintf(buf + ret, size - ret, |
| 495 | "%s%d", COMMA, |
| 496 | map->map[start]); |
| 497 | } else { |
| 498 | ret += snprintf(buf + ret, size - ret, |
| 499 | "%s%d-%d", COMMA, |
| 500 | map->map[start], map->map[end]); |
| 501 | } |
| 502 | first = false; |
| 503 | start = i; |
| 504 | } |
| 505 | } |
| 506 | |
| 507 | #undef COMMA |
| 508 | |
| 509 | pr_debug2("cpumask list: %s\n", buf); |
| 510 | return ret; |
| 511 | } |
| 512 | |
| 513 | static char hex_char(unsigned char val) |
| 514 | { |
| 515 | if (val < 10) |
| 516 | return val + '0'; |
| 517 | if (val < 16) |
| 518 | return val - 10 + 'a'; |
| 519 | return '?'; |
| 520 | } |
| 521 | |
| 522 | size_t cpu_map__snprint_mask(struct perf_cpu_map *map, char *buf, size_t size) |
| 523 | { |
| 524 | int i, cpu; |
| 525 | char *ptr = buf; |
| 526 | unsigned char *bitmap; |
| 527 | int last_cpu = cpu_map__cpu(map, map->nr - 1); |
| 528 | |
| 529 | if (buf == NULL) |
| 530 | return 0; |
| 531 | |
| 532 | bitmap = zalloc(last_cpu / 8 + 1); |
| 533 | if (bitmap == NULL) { |
| 534 | buf[0] = '\0'; |
| 535 | return 0; |
| 536 | } |
| 537 | |
| 538 | for (i = 0; i < map->nr; i++) { |
| 539 | cpu = cpu_map__cpu(map, i); |
| 540 | bitmap[cpu / 8] |= 1 << (cpu % 8); |
| 541 | } |
| 542 | |
| 543 | for (cpu = last_cpu / 4 * 4; cpu >= 0; cpu -= 4) { |
| 544 | unsigned char bits = bitmap[cpu / 8]; |
| 545 | |
| 546 | if (cpu % 8) |
| 547 | bits >>= 4; |
| 548 | else |
| 549 | bits &= 0xf; |
| 550 | |
| 551 | *ptr++ = hex_char(bits); |
| 552 | if ((cpu % 32) == 0 && cpu > 0) |
| 553 | *ptr++ = ','; |
| 554 | } |
| 555 | *ptr = '\0'; |
| 556 | free(bitmap); |
| 557 | |
| 558 | buf[size - 1] = '\0'; |
| 559 | return ptr - buf; |
| 560 | } |
| 561 | |
| 562 | const struct perf_cpu_map *cpu_map__online(void) /* thread unsafe */ |
| 563 | { |
| 564 | static const struct perf_cpu_map *online = NULL; |
| 565 | |
| 566 | if (!online) |
| 567 | online = perf_cpu_map__new(NULL); /* from /sys/devices/system/cpu/online */ |
| 568 | |
| 569 | return online; |
| 570 | } |