1 // SPDX-License-Identifier: GPL-2.0 2 #include <inttypes.h> 3 #include <sys/types.h> 4 #include <sys/stat.h> 5 #include <unistd.h> 6 #include "builtin.h" 7 8 #include <subcmd/parse-options.h> 9 #include "util/auxtrace.h" 10 #include "util/trace-event.h" 11 #include "util/tool.h" 12 #include "util/session.h" 13 #include "util/data.h" 14 #include "util/map_symbol.h" 15 #include "util/mem-events.h" 16 #include "util/debug.h" 17 #include "util/dso.h" 18 #include "util/map.h" 19 #include "util/symbol.h" 20 #include "util/pmu.h" 21 #include "util/pmu-hybrid.h" 22 #include "util/sample.h" 23 #include "util/string2.h" 24 #include "util/util.h" 25 #include <linux/err.h> 26 27 #define MEM_OPERATION_LOAD 0x1 28 #define MEM_OPERATION_STORE 0x2 29 30 struct perf_mem { 31 struct perf_tool tool; 32 char const *input_name; 33 bool hide_unresolved; 34 bool dump_raw; 35 bool force; 36 bool phys_addr; 37 bool data_page_size; 38 int operation; 39 const char *cpu_list; 40 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 41 }; 42 43 static int parse_record_events(const struct option *opt, 44 const char *str, int unset __maybe_unused) 45 { 46 struct perf_mem *mem = *(struct perf_mem **)opt->value; 47 48 if (!strcmp(str, "list")) { 49 perf_mem_events__list(); 50 exit(0); 51 } 52 if (perf_mem_events__parse(str)) 53 exit(-1); 54 55 mem->operation = 0; 56 return 0; 57 } 58 59 static const char * const __usage[] = { 60 "perf mem record [<options>] [<command>]", 61 "perf mem record [<options>] -- <command> [<options>]", 62 NULL 63 }; 64 65 static const char * const *record_mem_usage = __usage; 66 67 static int __cmd_record(int argc, const char **argv, struct perf_mem *mem) 68 { 69 int rec_argc, i = 0, j, tmp_nr = 0; 70 int start, end; 71 const char **rec_argv; 72 char **rec_tmp; 73 int ret; 74 bool all_user = false, all_kernel = false; 75 struct perf_mem_event *e; 76 struct option options[] = { 77 OPT_CALLBACK('e', "event", &mem, "event", 78 "event selector. use 'perf mem record -e list' to list available events", 79 parse_record_events), 80 OPT_UINTEGER(0, "ldlat", &perf_mem_events__loads_ldlat, "mem-loads latency"), 81 OPT_INCR('v', "verbose", &verbose, 82 "be more verbose (show counter open errors, etc)"), 83 OPT_BOOLEAN('U', "all-user", &all_user, "collect only user level data"), 84 OPT_BOOLEAN('K', "all-kernel", &all_kernel, "collect only kernel level data"), 85 OPT_END() 86 }; 87 88 if (perf_mem_events__init()) { 89 pr_err("failed: memory events not supported\n"); 90 return -1; 91 } 92 93 argc = parse_options(argc, argv, options, record_mem_usage, 94 PARSE_OPT_KEEP_UNKNOWN); 95 96 if (!perf_pmu__has_hybrid()) 97 rec_argc = argc + 9; /* max number of arguments */ 98 else 99 rec_argc = argc + 9 * perf_pmu__hybrid_pmu_num(); 100 101 if (mem->cpu_list) 102 rec_argc += 2; 103 104 rec_argv = calloc(rec_argc + 1, sizeof(char *)); 105 if (!rec_argv) 106 return -1; 107 108 /* 109 * Save the allocated event name strings. 110 */ 111 rec_tmp = calloc(rec_argc + 1, sizeof(char *)); 112 if (!rec_tmp) { 113 free(rec_argv); 114 return -1; 115 } 116 117 rec_argv[i++] = "record"; 118 119 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD_STORE); 120 121 /* 122 * The load and store operations are required, use the event 123 * PERF_MEM_EVENTS__LOAD_STORE if it is supported. 124 */ 125 if (e->tag && 126 (mem->operation & MEM_OPERATION_LOAD) && 127 (mem->operation & MEM_OPERATION_STORE)) { 128 e->record = true; 129 rec_argv[i++] = "-W"; 130 } else { 131 if (mem->operation & MEM_OPERATION_LOAD) { 132 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD); 133 e->record = true; 134 } 135 136 if (mem->operation & MEM_OPERATION_STORE) { 137 e = perf_mem_events__ptr(PERF_MEM_EVENTS__STORE); 138 e->record = true; 139 } 140 } 141 142 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD); 143 if (e->record) 144 rec_argv[i++] = "-W"; 145 146 rec_argv[i++] = "-d"; 147 148 if (mem->phys_addr) 149 rec_argv[i++] = "--phys-data"; 150 151 if (mem->data_page_size) 152 rec_argv[i++] = "--data-page-size"; 153 154 start = i; 155 ret = perf_mem_events__record_args(rec_argv, &i, rec_tmp, &tmp_nr); 156 if (ret) 157 goto out; 158 end = i; 159 160 if (all_user) 161 rec_argv[i++] = "--all-user"; 162 163 if (all_kernel) 164 rec_argv[i++] = "--all-kernel"; 165 166 if (mem->cpu_list) { 167 rec_argv[i++] = "-C"; 168 rec_argv[i++] = mem->cpu_list; 169 } 170 171 for (j = 0; j < argc; j++, i++) 172 rec_argv[i] = argv[j]; 173 174 if (verbose > 0) { 175 pr_debug("calling: record "); 176 177 for (j = start; j < end; j++) 178 pr_debug("%s ", rec_argv[j]); 179 180 pr_debug("\n"); 181 } 182 183 ret = cmd_record(i, rec_argv); 184 out: 185 for (i = 0; i < tmp_nr; i++) 186 free(rec_tmp[i]); 187 188 free(rec_tmp); 189 free(rec_argv); 190 return ret; 191 } 192 193 static int 194 dump_raw_samples(struct perf_tool *tool, 195 union perf_event *event, 196 struct perf_sample *sample, 197 struct machine *machine) 198 { 199 struct perf_mem *mem = container_of(tool, struct perf_mem, tool); 200 struct addr_location al; 201 const char *fmt, *field_sep; 202 char str[PAGE_SIZE_NAME_LEN]; 203 struct dso *dso = NULL; 204 205 if (machine__resolve(machine, &al, sample) < 0) { 206 fprintf(stderr, "problem processing %d event, skipping it.\n", 207 event->header.type); 208 return -1; 209 } 210 211 if (al.filtered || (mem->hide_unresolved && al.sym == NULL)) 212 goto out_put; 213 214 if (al.map != NULL) { 215 dso = map__dso(al.map); 216 if (dso) 217 dso->hit = 1; 218 } 219 220 field_sep = symbol_conf.field_sep; 221 if (field_sep) { 222 fmt = "%d%s%d%s0x%"PRIx64"%s0x%"PRIx64"%s"; 223 } else { 224 fmt = "%5d%s%5d%s0x%016"PRIx64"%s0x016%"PRIx64"%s"; 225 symbol_conf.field_sep = " "; 226 } 227 printf(fmt, 228 sample->pid, 229 symbol_conf.field_sep, 230 sample->tid, 231 symbol_conf.field_sep, 232 sample->ip, 233 symbol_conf.field_sep, 234 sample->addr, 235 symbol_conf.field_sep); 236 237 if (mem->phys_addr) { 238 printf("0x%016"PRIx64"%s", 239 sample->phys_addr, 240 symbol_conf.field_sep); 241 } 242 243 if (mem->data_page_size) { 244 printf("%s%s", 245 get_page_size_name(sample->data_page_size, str), 246 symbol_conf.field_sep); 247 } 248 249 if (field_sep) 250 fmt = "%"PRIu64"%s0x%"PRIx64"%s%s:%s\n"; 251 else 252 fmt = "%5"PRIu64"%s0x%06"PRIx64"%s%s:%s\n"; 253 254 printf(fmt, 255 sample->weight, 256 symbol_conf.field_sep, 257 sample->data_src, 258 symbol_conf.field_sep, 259 dso ? dso->long_name : "???", 260 al.sym ? al.sym->name : "???"); 261 out_put: 262 addr_location__put(&al); 263 return 0; 264 } 265 266 static int process_sample_event(struct perf_tool *tool, 267 union perf_event *event, 268 struct perf_sample *sample, 269 struct evsel *evsel __maybe_unused, 270 struct machine *machine) 271 { 272 return dump_raw_samples(tool, event, sample, machine); 273 } 274 275 static int report_raw_events(struct perf_mem *mem) 276 { 277 struct itrace_synth_opts itrace_synth_opts = { 278 .set = true, 279 .mem = true, /* Only enable memory event */ 280 .default_no_sample = true, 281 }; 282 283 struct perf_data data = { 284 .path = input_name, 285 .mode = PERF_DATA_MODE_READ, 286 .force = mem->force, 287 }; 288 int ret; 289 struct perf_session *session = perf_session__new(&data, &mem->tool); 290 291 if (IS_ERR(session)) 292 return PTR_ERR(session); 293 294 session->itrace_synth_opts = &itrace_synth_opts; 295 296 if (mem->cpu_list) { 297 ret = perf_session__cpu_bitmap(session, mem->cpu_list, 298 mem->cpu_bitmap); 299 if (ret < 0) 300 goto out_delete; 301 } 302 303 ret = symbol__init(&session->header.env); 304 if (ret < 0) 305 goto out_delete; 306 307 printf("# PID, TID, IP, ADDR, "); 308 309 if (mem->phys_addr) 310 printf("PHYS ADDR, "); 311 312 if (mem->data_page_size) 313 printf("DATA PAGE SIZE, "); 314 315 printf("LOCAL WEIGHT, DSRC, SYMBOL\n"); 316 317 ret = perf_session__process_events(session); 318 319 out_delete: 320 perf_session__delete(session); 321 return ret; 322 } 323 static char *get_sort_order(struct perf_mem *mem) 324 { 325 bool has_extra_options = (mem->phys_addr | mem->data_page_size) ? true : false; 326 char sort[128]; 327 328 /* 329 * there is no weight (cost) associated with stores, so don't print 330 * the column 331 */ 332 if (!(mem->operation & MEM_OPERATION_LOAD)) { 333 strcpy(sort, "--sort=mem,sym,dso,symbol_daddr," 334 "dso_daddr,tlb,locked"); 335 } else if (has_extra_options) { 336 strcpy(sort, "--sort=local_weight,mem,sym,dso,symbol_daddr," 337 "dso_daddr,snoop,tlb,locked,blocked"); 338 } else 339 return NULL; 340 341 if (mem->phys_addr) 342 strcat(sort, ",phys_daddr"); 343 344 if (mem->data_page_size) 345 strcat(sort, ",data_page_size"); 346 347 return strdup(sort); 348 } 349 350 static int report_events(int argc, const char **argv, struct perf_mem *mem) 351 { 352 const char **rep_argv; 353 int ret, i = 0, j, rep_argc; 354 char *new_sort_order; 355 356 if (mem->dump_raw) 357 return report_raw_events(mem); 358 359 rep_argc = argc + 3; 360 rep_argv = calloc(rep_argc + 1, sizeof(char *)); 361 if (!rep_argv) 362 return -1; 363 364 rep_argv[i++] = "report"; 365 rep_argv[i++] = "--mem-mode"; 366 rep_argv[i++] = "-n"; /* display number of samples */ 367 368 new_sort_order = get_sort_order(mem); 369 if (new_sort_order) 370 rep_argv[i++] = new_sort_order; 371 372 for (j = 1; j < argc; j++, i++) 373 rep_argv[i] = argv[j]; 374 375 ret = cmd_report(i, rep_argv); 376 free(rep_argv); 377 return ret; 378 } 379 380 struct mem_mode { 381 const char *name; 382 int mode; 383 }; 384 385 #define MEM_OPT(n, m) \ 386 { .name = n, .mode = (m) } 387 388 #define MEM_END { .name = NULL } 389 390 static const struct mem_mode mem_modes[]={ 391 MEM_OPT("load", MEM_OPERATION_LOAD), 392 MEM_OPT("store", MEM_OPERATION_STORE), 393 MEM_END 394 }; 395 396 static int 397 parse_mem_ops(const struct option *opt, const char *str, int unset) 398 { 399 int *mode = (int *)opt->value; 400 const struct mem_mode *m; 401 char *s, *os = NULL, *p; 402 int ret = -1; 403 404 if (unset) 405 return 0; 406 407 /* str may be NULL in case no arg is passed to -t */ 408 if (str) { 409 /* because str is read-only */ 410 s = os = strdup(str); 411 if (!s) 412 return -1; 413 414 /* reset mode */ 415 *mode = 0; 416 417 for (;;) { 418 p = strchr(s, ','); 419 if (p) 420 *p = '\0'; 421 422 for (m = mem_modes; m->name; m++) { 423 if (!strcasecmp(s, m->name)) 424 break; 425 } 426 if (!m->name) { 427 fprintf(stderr, "unknown sampling op %s," 428 " check man page\n", s); 429 goto error; 430 } 431 432 *mode |= m->mode; 433 434 if (!p) 435 break; 436 437 s = p + 1; 438 } 439 } 440 ret = 0; 441 442 if (*mode == 0) 443 *mode = MEM_OPERATION_LOAD; 444 error: 445 free(os); 446 return ret; 447 } 448 449 int cmd_mem(int argc, const char **argv) 450 { 451 struct stat st; 452 struct perf_mem mem = { 453 .tool = { 454 .sample = process_sample_event, 455 .mmap = perf_event__process_mmap, 456 .mmap2 = perf_event__process_mmap2, 457 .comm = perf_event__process_comm, 458 .lost = perf_event__process_lost, 459 .fork = perf_event__process_fork, 460 .attr = perf_event__process_attr, 461 .build_id = perf_event__process_build_id, 462 .namespaces = perf_event__process_namespaces, 463 .auxtrace_info = perf_event__process_auxtrace_info, 464 .auxtrace = perf_event__process_auxtrace, 465 .auxtrace_error = perf_event__process_auxtrace_error, 466 .ordered_events = true, 467 }, 468 .input_name = "perf.data", 469 /* 470 * default to both load an store sampling 471 */ 472 .operation = MEM_OPERATION_LOAD | MEM_OPERATION_STORE, 473 }; 474 const struct option mem_options[] = { 475 OPT_CALLBACK('t', "type", &mem.operation, 476 "type", "memory operations(load,store) Default load,store", 477 parse_mem_ops), 478 OPT_BOOLEAN('D', "dump-raw-samples", &mem.dump_raw, 479 "dump raw samples in ASCII"), 480 OPT_BOOLEAN('U', "hide-unresolved", &mem.hide_unresolved, 481 "Only display entries resolved to a symbol"), 482 OPT_STRING('i', "input", &input_name, "file", 483 "input file name"), 484 OPT_STRING('C', "cpu", &mem.cpu_list, "cpu", 485 "list of cpus to profile"), 486 OPT_STRING_NOEMPTY('x', "field-separator", &symbol_conf.field_sep, 487 "separator", 488 "separator for columns, no spaces will be added" 489 " between columns '.' is reserved."), 490 OPT_BOOLEAN('f', "force", &mem.force, "don't complain, do it"), 491 OPT_BOOLEAN('p', "phys-data", &mem.phys_addr, "Record/Report sample physical addresses"), 492 OPT_BOOLEAN(0, "data-page-size", &mem.data_page_size, "Record/Report sample data address page size"), 493 OPT_END() 494 }; 495 const char *const mem_subcommands[] = { "record", "report", NULL }; 496 const char *mem_usage[] = { 497 NULL, 498 NULL 499 }; 500 501 argc = parse_options_subcommand(argc, argv, mem_options, mem_subcommands, 502 mem_usage, PARSE_OPT_KEEP_UNKNOWN); 503 504 if (!argc || !(strncmp(argv[0], "rec", 3) || mem.operation)) 505 usage_with_options(mem_usage, mem_options); 506 507 if (!mem.input_name || !strlen(mem.input_name)) { 508 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 509 mem.input_name = "-"; 510 else 511 mem.input_name = "perf.data"; 512 } 513 514 if (strlen(argv[0]) > 2 && strstarts("record", argv[0])) 515 return __cmd_record(argc, argv, &mem); 516 else if (strlen(argv[0]) > 2 && strstarts("report", argv[0])) 517 return report_events(argc, argv, &mem); 518 else 519 usage_with_options(mem_usage, mem_options); 520 521 return 0; 522 } 523