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/pmus.h" 21 #include "util/sample.h" 22 #include "util/string2.h" 23 #include "util/util.h" 24 #include <linux/err.h> 25 26 #define MEM_OPERATION_LOAD 0x1 27 #define MEM_OPERATION_STORE 0x2 28 29 struct perf_mem { 30 struct perf_tool tool; 31 char const *input_name; 32 bool hide_unresolved; 33 bool dump_raw; 34 bool force; 35 bool phys_addr; 36 bool data_page_size; 37 int operation; 38 const char *cpu_list; 39 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 40 }; 41 42 static int parse_record_events(const struct option *opt, 43 const char *str, int unset __maybe_unused) 44 { 45 struct perf_mem *mem = *(struct perf_mem **)opt->value; 46 47 if (!strcmp(str, "list")) { 48 perf_mem_events__list(); 49 exit(0); 50 } 51 if (perf_mem_events__parse(str)) 52 exit(-1); 53 54 mem->operation = 0; 55 return 0; 56 } 57 58 static const char * const __usage[] = { 59 "perf mem record [<options>] [<command>]", 60 "perf mem record [<options>] -- <command> [<options>]", 61 NULL 62 }; 63 64 static const char * const *record_mem_usage = __usage; 65 66 static int __cmd_record(int argc, const char **argv, struct perf_mem *mem) 67 { 68 int rec_argc, i = 0, j, tmp_nr = 0; 69 int start, end; 70 const char **rec_argv; 71 char **rec_tmp; 72 int ret; 73 bool all_user = false, all_kernel = false; 74 struct perf_mem_event *e; 75 struct option options[] = { 76 OPT_CALLBACK('e', "event", &mem, "event", 77 "event selector. use 'perf mem record -e list' to list available events", 78 parse_record_events), 79 OPT_UINTEGER(0, "ldlat", &perf_mem_events__loads_ldlat, "mem-loads latency"), 80 OPT_INCR('v', "verbose", &verbose, 81 "be more verbose (show counter open errors, etc)"), 82 OPT_BOOLEAN('U', "all-user", &all_user, "collect only user level data"), 83 OPT_BOOLEAN('K', "all-kernel", &all_kernel, "collect only kernel level data"), 84 OPT_END() 85 }; 86 87 if (perf_mem_events__init()) { 88 pr_err("failed: memory events not supported\n"); 89 return -1; 90 } 91 92 argc = parse_options(argc, argv, options, record_mem_usage, 93 PARSE_OPT_KEEP_UNKNOWN); 94 95 /* Max number of arguments multiplied by number of PMUs that can support them. */ 96 rec_argc = argc + 9 * perf_pmus__num_mem_pmus(); 97 98 if (mem->cpu_list) 99 rec_argc += 2; 100 101 rec_argv = calloc(rec_argc + 1, sizeof(char *)); 102 if (!rec_argv) 103 return -1; 104 105 /* 106 * Save the allocated event name strings. 107 */ 108 rec_tmp = calloc(rec_argc + 1, sizeof(char *)); 109 if (!rec_tmp) { 110 free(rec_argv); 111 return -1; 112 } 113 114 rec_argv[i++] = "record"; 115 116 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD_STORE); 117 118 /* 119 * The load and store operations are required, use the event 120 * PERF_MEM_EVENTS__LOAD_STORE if it is supported. 121 */ 122 if (e->tag && 123 (mem->operation & MEM_OPERATION_LOAD) && 124 (mem->operation & MEM_OPERATION_STORE)) { 125 e->record = true; 126 rec_argv[i++] = "-W"; 127 } else { 128 if (mem->operation & MEM_OPERATION_LOAD) { 129 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD); 130 e->record = true; 131 } 132 133 if (mem->operation & MEM_OPERATION_STORE) { 134 e = perf_mem_events__ptr(PERF_MEM_EVENTS__STORE); 135 e->record = true; 136 } 137 } 138 139 e = perf_mem_events__ptr(PERF_MEM_EVENTS__LOAD); 140 if (e->record) 141 rec_argv[i++] = "-W"; 142 143 rec_argv[i++] = "-d"; 144 145 if (mem->phys_addr) 146 rec_argv[i++] = "--phys-data"; 147 148 if (mem->data_page_size) 149 rec_argv[i++] = "--data-page-size"; 150 151 start = i; 152 ret = perf_mem_events__record_args(rec_argv, &i, rec_tmp, &tmp_nr); 153 if (ret) 154 goto out; 155 end = i; 156 157 if (all_user) 158 rec_argv[i++] = "--all-user"; 159 160 if (all_kernel) 161 rec_argv[i++] = "--all-kernel"; 162 163 if (mem->cpu_list) { 164 rec_argv[i++] = "-C"; 165 rec_argv[i++] = mem->cpu_list; 166 } 167 168 for (j = 0; j < argc; j++, i++) 169 rec_argv[i] = argv[j]; 170 171 if (verbose > 0) { 172 pr_debug("calling: record "); 173 174 for (j = start; j < end; j++) 175 pr_debug("%s ", rec_argv[j]); 176 177 pr_debug("\n"); 178 } 179 180 ret = cmd_record(i, rec_argv); 181 out: 182 for (i = 0; i < tmp_nr; i++) 183 free(rec_tmp[i]); 184 185 free(rec_tmp); 186 free(rec_argv); 187 return ret; 188 } 189 190 static int 191 dump_raw_samples(struct perf_tool *tool, 192 union perf_event *event, 193 struct perf_sample *sample, 194 struct machine *machine) 195 { 196 struct perf_mem *mem = container_of(tool, struct perf_mem, tool); 197 struct addr_location al; 198 const char *fmt, *field_sep; 199 char str[PAGE_SIZE_NAME_LEN]; 200 struct dso *dso = NULL; 201 202 addr_location__init(&al); 203 if (machine__resolve(machine, &al, sample) < 0) { 204 fprintf(stderr, "problem processing %d event, skipping it.\n", 205 event->header.type); 206 addr_location__exit(&al); 207 return -1; 208 } 209 210 if (al.filtered || (mem->hide_unresolved && al.sym == NULL)) 211 goto out_put; 212 213 if (al.map != NULL) { 214 dso = map__dso(al.map); 215 if (dso) 216 dso->hit = 1; 217 } 218 219 field_sep = symbol_conf.field_sep; 220 if (field_sep) { 221 fmt = "%d%s%d%s0x%"PRIx64"%s0x%"PRIx64"%s"; 222 } else { 223 fmt = "%5d%s%5d%s0x%016"PRIx64"%s0x016%"PRIx64"%s"; 224 symbol_conf.field_sep = " "; 225 } 226 printf(fmt, 227 sample->pid, 228 symbol_conf.field_sep, 229 sample->tid, 230 symbol_conf.field_sep, 231 sample->ip, 232 symbol_conf.field_sep, 233 sample->addr, 234 symbol_conf.field_sep); 235 236 if (mem->phys_addr) { 237 printf("0x%016"PRIx64"%s", 238 sample->phys_addr, 239 symbol_conf.field_sep); 240 } 241 242 if (mem->data_page_size) { 243 printf("%s%s", 244 get_page_size_name(sample->data_page_size, str), 245 symbol_conf.field_sep); 246 } 247 248 if (field_sep) 249 fmt = "%"PRIu64"%s0x%"PRIx64"%s%s:%s\n"; 250 else 251 fmt = "%5"PRIu64"%s0x%06"PRIx64"%s%s:%s\n"; 252 253 printf(fmt, 254 sample->weight, 255 symbol_conf.field_sep, 256 sample->data_src, 257 symbol_conf.field_sep, 258 dso ? dso->long_name : "???", 259 al.sym ? al.sym->name : "???"); 260 out_put: 261 addr_location__exit(&al); 262 return 0; 263 } 264 265 static int process_sample_event(struct perf_tool *tool, 266 union perf_event *event, 267 struct perf_sample *sample, 268 struct evsel *evsel __maybe_unused, 269 struct machine *machine) 270 { 271 return dump_raw_samples(tool, event, sample, machine); 272 } 273 274 static int report_raw_events(struct perf_mem *mem) 275 { 276 struct itrace_synth_opts itrace_synth_opts = { 277 .set = true, 278 .mem = true, /* Only enable memory event */ 279 .default_no_sample = true, 280 }; 281 282 struct perf_data data = { 283 .path = input_name, 284 .mode = PERF_DATA_MODE_READ, 285 .force = mem->force, 286 }; 287 int ret; 288 struct perf_session *session = perf_session__new(&data, &mem->tool); 289 290 if (IS_ERR(session)) 291 return PTR_ERR(session); 292 293 session->itrace_synth_opts = &itrace_synth_opts; 294 295 if (mem->cpu_list) { 296 ret = perf_session__cpu_bitmap(session, mem->cpu_list, 297 mem->cpu_bitmap); 298 if (ret < 0) 299 goto out_delete; 300 } 301 302 ret = symbol__init(&session->header.env); 303 if (ret < 0) 304 goto out_delete; 305 306 printf("# PID, TID, IP, ADDR, "); 307 308 if (mem->phys_addr) 309 printf("PHYS ADDR, "); 310 311 if (mem->data_page_size) 312 printf("DATA PAGE SIZE, "); 313 314 printf("LOCAL WEIGHT, DSRC, SYMBOL\n"); 315 316 ret = perf_session__process_events(session); 317 318 out_delete: 319 perf_session__delete(session); 320 return ret; 321 } 322 static char *get_sort_order(struct perf_mem *mem) 323 { 324 bool has_extra_options = (mem->phys_addr | mem->data_page_size) ? true : false; 325 char sort[128]; 326 327 /* 328 * there is no weight (cost) associated with stores, so don't print 329 * the column 330 */ 331 if (!(mem->operation & MEM_OPERATION_LOAD)) { 332 strcpy(sort, "--sort=mem,sym,dso,symbol_daddr," 333 "dso_daddr,tlb,locked"); 334 } else if (has_extra_options) { 335 strcpy(sort, "--sort=local_weight,mem,sym,dso,symbol_daddr," 336 "dso_daddr,snoop,tlb,locked,blocked"); 337 } else 338 return NULL; 339 340 if (mem->phys_addr) 341 strcat(sort, ",phys_daddr"); 342 343 if (mem->data_page_size) 344 strcat(sort, ",data_page_size"); 345 346 return strdup(sort); 347 } 348 349 static int report_events(int argc, const char **argv, struct perf_mem *mem) 350 { 351 const char **rep_argv; 352 int ret, i = 0, j, rep_argc; 353 char *new_sort_order; 354 355 if (mem->dump_raw) 356 return report_raw_events(mem); 357 358 rep_argc = argc + 3; 359 rep_argv = calloc(rep_argc + 1, sizeof(char *)); 360 if (!rep_argv) 361 return -1; 362 363 rep_argv[i++] = "report"; 364 rep_argv[i++] = "--mem-mode"; 365 rep_argv[i++] = "-n"; /* display number of samples */ 366 367 new_sort_order = get_sort_order(mem); 368 if (new_sort_order) 369 rep_argv[i++] = new_sort_order; 370 371 for (j = 1; j < argc; j++, i++) 372 rep_argv[i] = argv[j]; 373 374 ret = cmd_report(i, rep_argv); 375 free(new_sort_order); 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