tools/perf/builtin-mem.c

   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/sort.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         const char              *input_name;
  33         const char              *sort_key;
  34         bool                    hide_unresolved;
  35         bool                    dump_raw;
  36         bool                    force;
  37         bool                    phys_addr;
  38         bool                    data_page_size;
  39         bool                    all_kernel;
  40         bool                    all_user;
  41         bool                    data_type;
  42         int                     operation;
  43         const char              *cpu_list;
  44         DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
  45 };
  46
  47 static int parse_record_events(const struct option *opt,
  48                                const char *str, int unset __maybe_unused)
  49 {
  50         struct perf_mem *mem = (struct perf_mem *)opt->value;
  51         struct perf_pmu *pmu;
  52
  53         pmu = perf_mem_events_find_pmu();
  54         if (!pmu) {
  55                 pr_err("failed: there is no PMU that supports perf mem\n");
  56                 exit(-1);
  57         }
  58
  59         if (!strcmp(str, "list")) {
  60                 perf_pmu__mem_events_list(pmu);
  61                 exit(0);
  62         }
  63         if (perf_pmu__mem_events_parse(pmu, str))
  64                 exit(-1);
  65
  66         mem->operation = 0;
  67         return 0;
  68 }
  69
  70 static int __cmd_record(int argc, const char **argv, struct perf_mem *mem,
  71                         const struct option *options)
  72 {
  73         int rec_argc, i = 0, j;
  74         int start, end;
  75         const char **rec_argv;
  76         int ret;
  77         struct perf_mem_event *e;
  78         struct perf_pmu *pmu;
  79         const char * const record_usage[] = {
  80                 "perf mem record [<options>] [<command>]",
  81                 "perf mem record [<options>] -- <command> [<options>]",
  82                 NULL
  83         };
  84
  85         pmu = perf_mem_events_find_pmu();
  86         if (!pmu) {
  87                 pr_err("failed: no PMU supports the memory events\n");
  88                 return -1;
  89         }
  90
  91         if (perf_pmu__mem_events_init()) {
  92                 pr_err("failed: memory events not supported\n");
  93                 return -1;
  94         }
  95
  96         argc = parse_options(argc, argv, options, record_usage,
  97                              PARSE_OPT_KEEP_UNKNOWN);
  98
  99         /* Max number of arguments multiplied by number of PMUs that can support them. */
 100         rec_argc = argc + 9 * (perf_pmu__mem_events_num_mem_pmus(pmu) + 1);
 101
 102         if (mem->cpu_list)
 103                 rec_argc += 2;
 104
 105         rec_argv = calloc(rec_argc + 1, sizeof(char *));
 106         if (!rec_argv)
 107                 return -1;
 108
 109         rec_argv[i++] = "record";
 110
 111         e = perf_pmu__mem_events_ptr(pmu, PERF_MEM_EVENTS__LOAD_STORE);
 112
 113         /*
 114          * The load and store operations are required, use the event
 115          * PERF_MEM_EVENTS__LOAD_STORE if it is supported.
 116          */
 117         if (e->tag &&
 118             (mem->operation & MEM_OPERATION_LOAD) &&
 119             (mem->operation & MEM_OPERATION_STORE)) {
 120                 perf_mem_record[PERF_MEM_EVENTS__LOAD_STORE] = true;
 121                 rec_argv[i++] = "-W";
 122         } else {
 123                 if (mem->operation & MEM_OPERATION_LOAD)
 124                         perf_mem_record[PERF_MEM_EVENTS__LOAD] = true;
 125
 126                 if (mem->operation & MEM_OPERATION_STORE)
 127                         perf_mem_record[PERF_MEM_EVENTS__STORE] = true;
 128         }
 129
 130         if (perf_mem_record[PERF_MEM_EVENTS__LOAD])
 131                 rec_argv[i++] = "-W";
 132
 133         rec_argv[i++] = "-d";
 134
 135         if (mem->phys_addr)
 136                 rec_argv[i++] = "--phys-data";
 137
 138         if (mem->data_page_size)
 139                 rec_argv[i++] = "--data-page-size";
 140
 141         start = i;
 142         ret = perf_mem_events__record_args(rec_argv, &i);
 143         if (ret)
 144                 goto out;
 145         end = i;
 146
 147         if (mem->all_user)
 148                 rec_argv[i++] = "--all-user";
 149
 150         if (mem->all_kernel)
 151                 rec_argv[i++] = "--all-kernel";
 152
 153         if (mem->cpu_list) {
 154                 rec_argv[i++] = "-C";
 155                 rec_argv[i++] = mem->cpu_list;
 156         }
 157
 158         for (j = 0; j < argc; j++, i++)
 159                 rec_argv[i] = argv[j];
 160
 161         if (verbose > 0) {
 162                 pr_debug("calling: record ");
 163
 164                 for (j = start; j < end; j++)
 165                         pr_debug("%s ", rec_argv[j]);
 166
 167                 pr_debug("\n");
 168         }
 169
 170         ret = cmd_record(i, rec_argv);
 171 out:
 172         free(rec_argv);
 173         return ret;
 174 }
 175
 176 static int
 177 dump_raw_samples(const struct perf_tool *tool,
 178                  union perf_event *event,
 179                  struct perf_sample *sample,
 180                  struct machine *machine)
 181 {
 182         struct perf_mem *mem = container_of(tool, struct perf_mem, tool);
 183         struct addr_location al;
 184         const char *fmt, *field_sep;
 185         char str[PAGE_SIZE_NAME_LEN];
 186         struct dso *dso = NULL;
 187
 188         addr_location__init(&al);
 189         if (machine__resolve(machine, &al, sample) < 0) {
 190                 fprintf(stderr, "problem processing %d event, skipping it.\n",
 191                                 event->header.type);
 192                 addr_location__exit(&al);
 193                 return -1;
 194         }
 195
 196         if (al.filtered || (mem->hide_unresolved && al.sym == NULL))
 197                 goto out_put;
 198
 199         if (al.map != NULL) {
 200                 dso = map__dso(al.map);
 201                 if (dso)
 202                         dso__set_hit(dso);
 203         }
 204
 205         field_sep = symbol_conf.field_sep;
 206         if (field_sep) {
 207                 fmt = "%d%s%d%s0x%"PRIx64"%s0x%"PRIx64"%s";
 208         } else {
 209                 fmt = "%5d%s%5d%s0x%016"PRIx64"%s0x016%"PRIx64"%s";
 210                 symbol_conf.field_sep = " ";
 211         }
 212         printf(fmt,
 213                 sample->pid,
 214                 symbol_conf.field_sep,
 215                 sample->tid,
 216                 symbol_conf.field_sep,
 217                 sample->ip,
 218                 symbol_conf.field_sep,
 219                 sample->addr,
 220                 symbol_conf.field_sep);
 221
 222         if (mem->phys_addr) {
 223                 printf("0x%016"PRIx64"%s",
 224                         sample->phys_addr,
 225                         symbol_conf.field_sep);
 226         }
 227
 228         if (mem->data_page_size) {
 229                 printf("%s%s",
 230                         get_page_size_name(sample->data_page_size, str),
 231                         symbol_conf.field_sep);
 232         }
 233
 234         if (field_sep)
 235                 fmt = "%"PRIu64"%s0x%"PRIx64"%s%s:%s\n";
 236         else
 237                 fmt = "%5"PRIu64"%s0x%06"PRIx64"%s%s:%s\n";
 238
 239         printf(fmt,
 240                 sample->weight,
 241                 symbol_conf.field_sep,
 242                 sample->data_src,
 243                 symbol_conf.field_sep,
 244                 dso ? dso__long_name(dso) : "???",
 245                 al.sym ? al.sym->name : "???");
 246 out_put:
 247         addr_location__exit(&al);
 248         return 0;
 249 }
 250
 251 static int process_sample_event(const struct perf_tool *tool,
 252                                 union perf_event *event,
 253                                 struct perf_sample *sample,
 254                                 struct evsel *evsel __maybe_unused,
 255                                 struct machine *machine)
 256 {
 257         return dump_raw_samples(tool, event, sample, machine);
 258 }
 259
 260 static int report_raw_events(struct perf_mem *mem)
 261 {
 262         struct itrace_synth_opts itrace_synth_opts = {
 263                 .set = true,
 264                 .mem = true,    /* Only enable memory event */
 265                 .default_no_sample = true,
 266         };
 267
 268         struct perf_data data = {
 269                 .path  = input_name,
 270                 .mode  = PERF_DATA_MODE_READ,
 271                 .force = mem->force,
 272         };
 273         int ret;
 274         struct perf_session *session;
 275
 276         perf_tool__init(&mem->tool, /*ordered_events=*/true);
 277         mem->tool.sample                = process_sample_event;
 278         mem->tool.mmap          = perf_event__process_mmap;
 279         mem->tool.mmap2         = perf_event__process_mmap2;
 280         mem->tool.comm          = perf_event__process_comm;
 281         mem->tool.lost          = perf_event__process_lost;
 282         mem->tool.fork          = perf_event__process_fork;
 283         mem->tool.attr          = perf_event__process_attr;
 284         mem->tool.build_id      = perf_event__process_build_id;
 285         mem->tool.namespaces    = perf_event__process_namespaces;
 286         mem->tool.auxtrace_info  = perf_event__process_auxtrace_info;
 287         mem->tool.auxtrace       = perf_event__process_auxtrace;
 288         mem->tool.auxtrace_error = perf_event__process_auxtrace_error;
 289
 290         session = perf_session__new(&data, &mem->tool);
 291
 292         if (IS_ERR(session))
 293                 return PTR_ERR(session);
 294
 295         session->itrace_synth_opts = &itrace_synth_opts;
 296
 297         if (mem->cpu_list) {
 298                 ret = perf_session__cpu_bitmap(session, mem->cpu_list,
 299                                                mem->cpu_bitmap);
 300                 if (ret < 0)
 301                         goto out_delete;
 302         }
 303
 304         ret = symbol__init(&session->header.env);
 305         if (ret < 0)
 306                 goto out_delete;
 307
 308         printf("# PID, TID, IP, ADDR, ");
 309
 310         if (mem->phys_addr)
 311                 printf("PHYS ADDR, ");
 312
 313         if (mem->data_page_size)
 314                 printf("DATA PAGE SIZE, ");
 315
 316         printf("LOCAL WEIGHT, DSRC, SYMBOL\n");
 317
 318         ret = perf_session__process_events(session);
 319
 320 out_delete:
 321         perf_session__delete(session);
 322         return ret;
 323 }
 324
 325 static char *get_sort_order(struct perf_mem *mem)
 326 {
 327         bool has_extra_options = (mem->phys_addr | mem->data_page_size) ? true : false;
 328         char sort[128];
 329
 330         if (mem->sort_key)
 331                 scnprintf(sort, sizeof(sort), "--sort=%s", mem->sort_key);
 332         else if (mem->data_type)
 333                 strcpy(sort, "--sort=mem,snoop,tlb,type");
 334         /*
 335          * there is no weight (cost) associated with stores, so don't print
 336          * the column
 337          */
 338         else if (!(mem->operation & MEM_OPERATION_LOAD)) {
 339                 strcpy(sort, "--sort=mem,sym,dso,symbol_daddr,"
 340                              "dso_daddr,tlb,locked");
 341         } else if (has_extra_options) {
 342                 strcpy(sort, "--sort=local_weight,mem,sym,dso,symbol_daddr,"
 343                              "dso_daddr,snoop,tlb,locked,blocked");
 344         } else
 345                 return NULL;
 346
 347         if (mem->phys_addr)
 348                 strcat(sort, ",phys_daddr");
 349
 350         if (mem->data_page_size)
 351                 strcat(sort, ",data_page_size");
 352
 353         /* make sure it has 'type' sort key even -s option is used */
 354         if (mem->data_type && !strstr(sort, "type"))
 355                 strcat(sort, ",type");
 356
 357         return strdup(sort);
 358 }
 359
 360 static int __cmd_report(int argc, const char **argv, struct perf_mem *mem,
 361                         const struct option *options)
 362 {
 363         const char **rep_argv;
 364         int ret, i = 0, j, rep_argc;
 365         char *new_sort_order;
 366         const char * const report_usage[] = {
 367                 "perf mem report [<options>]",
 368                 NULL
 369         };
 370
 371         argc = parse_options(argc, argv, options, report_usage,
 372                              PARSE_OPT_KEEP_UNKNOWN);
 373
 374         if (mem->dump_raw)
 375                 return report_raw_events(mem);
 376
 377         rep_argc = argc + 3;
 378         rep_argv = calloc(rep_argc + 1, sizeof(char *));
 379         if (!rep_argv)
 380                 return -1;
 381
 382         rep_argv[i++] = "report";
 383         rep_argv[i++] = "--mem-mode";
 384         rep_argv[i++] = "-n"; /* display number of samples */
 385
 386         new_sort_order = get_sort_order(mem);
 387         if (new_sort_order)
 388                 rep_argv[i++] = new_sort_order;
 389
 390         for (j = 0; j < argc; j++, i++)
 391                 rep_argv[i] = argv[j];
 392
 393         ret = cmd_report(i, rep_argv);
 394         free(new_sort_order);
 395         free(rep_argv);
 396         return ret;
 397 }
 398
 399 struct mem_mode {
 400         const char *name;
 401         int mode;
 402 };
 403
 404 #define MEM_OPT(n, m) \
 405         { .name = n, .mode = (m) }
 406
 407 #define MEM_END { .name = NULL }
 408
 409 static const struct mem_mode mem_modes[]={
 410         MEM_OPT("load", MEM_OPERATION_LOAD),
 411         MEM_OPT("store", MEM_OPERATION_STORE),
 412         MEM_END
 413 };
 414
 415 static int
 416 parse_mem_ops(const struct option *opt, const char *str, int unset)
 417 {
 418         int *mode = (int *)opt->value;
 419         const struct mem_mode *m;
 420         char *s, *os = NULL, *p;
 421         int ret = -1;
 422
 423         if (unset)
 424                 return 0;
 425
 426         /* str may be NULL in case no arg is passed to -t */
 427         if (str) {
 428                 /* because str is read-only */
 429                 s = os = strdup(str);
 430                 if (!s)
 431                         return -1;
 432
 433                 /* reset mode */
 434                 *mode = 0;
 435
 436                 for (;;) {
 437                         p = strchr(s, ',');
 438                         if (p)
 439                                 *p = '\0';
 440
 441                         for (m = mem_modes; m->name; m++) {
 442                                 if (!strcasecmp(s, m->name))
 443                                         break;
 444                         }
 445                         if (!m->name) {
 446                                 fprintf(stderr, "unknown sampling op %s,"
 447                                             " check man page\n", s);
 448                                 goto error;
 449                         }
 450
 451                         *mode |= m->mode;
 452
 453                         if (!p)
 454                                 break;
 455
 456                         s = p + 1;
 457                 }
 458         }
 459         ret = 0;
 460
 461         if (*mode == 0)
 462                 *mode = MEM_OPERATION_LOAD;
 463 error:
 464         free(os);
 465         return ret;
 466 }
 467
 468 int cmd_mem(int argc, const char **argv)
 469 {
 470         struct stat st;
 471         struct perf_mem mem = {
 472                 .input_name              = "perf.data",
 473                 /*
 474                  * default to both load an store sampling
 475                  */
 476                 .operation               = MEM_OPERATION_LOAD | MEM_OPERATION_STORE,
 477         };
 478         char *sort_order_help = sort_help("sort by key(s):", SORT_MODE__MEMORY);
 479         const struct option mem_options[] = {
 480         OPT_CALLBACK('t', "type", &mem.operation,
 481                    "type", "memory operations(load,store) Default load,store",
 482                     parse_mem_ops),
 483         OPT_STRING('C', "cpu", &mem.cpu_list, "cpu",
 484                    "list of cpus to profile"),
 485         OPT_BOOLEAN('f', "force", &mem.force, "don't complain, do it"),
 486         OPT_INCR('v', "verbose", &verbose,
 487                  "be more verbose (show counter open errors, etc)"),
 488         OPT_BOOLEAN('p', "phys-data", &mem.phys_addr, "Record/Report sample physical addresses"),
 489         OPT_BOOLEAN(0, "data-page-size", &mem.data_page_size, "Record/Report sample data address page size"),
 490         OPT_END()
 491         };
 492         const struct option record_options[] = {
 493         OPT_CALLBACK('e', "event", &mem, "event",
 494                      "event selector. use 'perf mem record -e list' to list available events",
 495                      parse_record_events),
 496         OPT_UINTEGER(0, "ldlat", &perf_mem_events__loads_ldlat, "mem-loads latency"),
 497         OPT_BOOLEAN('U', "all-user", &mem.all_user, "collect only user level data"),
 498         OPT_BOOLEAN('K', "all-kernel", &mem.all_kernel, "collect only kernel level data"),
 499         OPT_PARENT(mem_options)
 500         };
 501         const struct option report_options[] = {
 502         OPT_BOOLEAN('D', "dump-raw-samples", &mem.dump_raw,
 503                     "dump raw samples in ASCII"),
 504         OPT_BOOLEAN('U', "hide-unresolved", &mem.hide_unresolved,
 505                     "Only display entries resolved to a symbol"),
 506         OPT_STRING('i', "input", &input_name, "file",
 507                    "input file name"),
 508         OPT_STRING_NOEMPTY('x', "field-separator", &symbol_conf.field_sep,
 509                    "separator",
 510                    "separator for columns, no spaces will be added"
 511                    " between columns '.' is reserved."),
 512         OPT_STRING('s', "sort", &mem.sort_key, "key[,key2...]",
 513                    sort_order_help),
 514         OPT_BOOLEAN('T', "type-profile", &mem.data_type,
 515                     "Show data-type profile result"),
 516         OPT_PARENT(mem_options)
 517         };
 518         const char *const mem_subcommands[] = { "record", "report", NULL };
 519         const char *mem_usage[] = {
 520                 NULL,
 521                 NULL
 522         };
 523
 524         argc = parse_options_subcommand(argc, argv, mem_options, mem_subcommands,
 525                                         mem_usage, PARSE_OPT_STOP_AT_NON_OPTION);
 526
 527         if (!argc || !(strncmp(argv[0], "rec", 3) || mem.operation))
 528                 usage_with_options(mem_usage, mem_options);
 529
 530         if (!mem.input_name || !strlen(mem.input_name)) {
 531                 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
 532                         mem.input_name = "-";
 533                 else
 534                         mem.input_name = "perf.data";
 535         }
 536
 537         if (strlen(argv[0]) > 2 && strstarts("record", argv[0]))
 538                 return __cmd_record(argc, argv, &mem, record_options);
 539         else if (strlen(argv[0]) > 2 && strstarts("report", argv[0]))
 540                 return __cmd_report(argc, argv, &mem, report_options);
 541         else
 542                 usage_with_options(mem_usage, mem_options);
 543
 544         /* free usage string allocated by parse_options_subcommand */
 545         free((void *)mem_usage[0]);
 546
 547         return 0;
 548 }