kernel/trace/trace_functions_graph.c

   1 /*
   2  *
   3  * Function graph tracer.
   4  * Copyright (c) 2008-2009 Frederic Weisbecker <fweisbec@gmail.com>
   5  * Mostly borrowed from function tracer which
   6  * is Copyright (c) Steven Rostedt <srostedt@redhat.com>
   7  *
   8  */
   9 #include <linux/debugfs.h>
  10 #include <linux/uaccess.h>
  11 #include <linux/ftrace.h>
  12 #include <linux/fs.h>
  13
  14 #include "trace.h"
  15 #include "trace_output.h"
  16
  17 struct fgraph_data {
  18         pid_t           last_pid;
  19         int             depth;
  20 };
  21
  22 #define TRACE_GRAPH_INDENT      2
  23
  24 /* Flag options */
  25 #define TRACE_GRAPH_PRINT_OVERRUN       0x1
  26 #define TRACE_GRAPH_PRINT_CPU           0x2
  27 #define TRACE_GRAPH_PRINT_OVERHEAD      0x4
  28 #define TRACE_GRAPH_PRINT_PROC          0x8
  29 #define TRACE_GRAPH_PRINT_DURATION      0x10
  30 #define TRACE_GRAPH_PRINT_ABS_TIME      0X20
  31
  32 static struct tracer_opt trace_opts[] = {
  33         /* Display overruns? (for self-debug purpose) */
  34         { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
  35         /* Display CPU ? */
  36         { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
  37         /* Display Overhead ? */
  38         { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
  39         /* Display proc name/pid */
  40         { TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
  41         /* Display duration of execution */
  42         { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
  43         /* Display absolute time of an entry */
  44         { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
  45         { } /* Empty entry */
  46 };
  47
  48 static struct tracer_flags tracer_flags = {
  49         /* Don't display overruns and proc by default */
  50         .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
  51                TRACE_GRAPH_PRINT_DURATION,
  52         .opts = trace_opts
  53 };
  54
  55 static struct trace_array *graph_array;
  56
  57
  58 /* Add a function return address to the trace stack on thread info.*/
  59 int
  60 ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
  61                          unsigned long frame_pointer)
  62 {
  63         unsigned long long calltime;
  64         int index;
  65
  66         if (!current->ret_stack)
  67                 return -EBUSY;
  68
  69         /*
  70          * We must make sure the ret_stack is tested before we read
  71          * anything else.
  72          */
  73         smp_rmb();
  74
  75         /* The return trace stack is full */
  76         if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
  77                 atomic_inc(&current->trace_overrun);
  78                 return -EBUSY;
  79         }
  80
  81         calltime = trace_clock_local();
  82
  83         index = ++current->curr_ret_stack;
  84         barrier();
  85         current->ret_stack[index].ret = ret;
  86         current->ret_stack[index].func = func;
  87         current->ret_stack[index].calltime = calltime;
  88         current->ret_stack[index].subtime = 0;
  89         current->ret_stack[index].fp = frame_pointer;
  90         *depth = index;
  91
  92         return 0;
  93 }
  94
  95 /* Retrieve a function return address to the trace stack on thread info.*/
  96 static void
  97 ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
  98                         unsigned long frame_pointer)
  99 {
 100         int index;
 101
 102         index = current->curr_ret_stack;
 103
 104         if (unlikely(index < 0)) {
 105                 ftrace_graph_stop();
 106                 WARN_ON(1);
 107                 /* Might as well panic, otherwise we have no where to go */
 108                 *ret = (unsigned long)panic;
 109                 return;
 110         }
 111
 112 #ifdef CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST
 113         /*
 114          * The arch may choose to record the frame pointer used
 115          * and check it here to make sure that it is what we expect it
 116          * to be. If gcc does not set the place holder of the return
 117          * address in the frame pointer, and does a copy instead, then
 118          * the function graph trace will fail. This test detects this
 119          * case.
 120          *
 121          * Currently, x86_32 with optimize for size (-Os) makes the latest
 122          * gcc do the above.
 123          */
 124         if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
 125                 ftrace_graph_stop();
 126                 WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
 127                      "  from func %ps return to %lx\n",
 128                      current->ret_stack[index].fp,
 129                      frame_pointer,
 130                      (void *)current->ret_stack[index].func,
 131                      current->ret_stack[index].ret);
 132                 *ret = (unsigned long)panic;
 133                 return;
 134         }
 135 #endif
 136
 137         *ret = current->ret_stack[index].ret;
 138         trace->func = current->ret_stack[index].func;
 139         trace->calltime = current->ret_stack[index].calltime;
 140         trace->overrun = atomic_read(&current->trace_overrun);
 141         trace->depth = index;
 142 }
 143
 144 /*
 145  * Send the trace to the ring-buffer.
 146  * @return the original return address.
 147  */
 148 unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
 149 {
 150         struct ftrace_graph_ret trace;
 151         unsigned long ret;
 152
 153         ftrace_pop_return_trace(&trace, &ret, frame_pointer);
 154         trace.rettime = trace_clock_local();
 155         ftrace_graph_return(&trace);
 156         barrier();
 157         current->curr_ret_stack--;
 158
 159         if (unlikely(!ret)) {
 160                 ftrace_graph_stop();
 161                 WARN_ON(1);
 162                 /* Might as well panic. What else to do? */
 163                 ret = (unsigned long)panic;
 164         }
 165
 166         return ret;
 167 }
 168
 169 static int __trace_graph_entry(struct trace_array *tr,
 170                                 struct ftrace_graph_ent *trace,
 171                                 unsigned long flags,
 172                                 int pc)
 173 {
 174         struct ftrace_event_call *call = &event_funcgraph_entry;
 175         struct ring_buffer_event *event;
 176         struct ring_buffer *buffer = tr->buffer;
 177         struct ftrace_graph_ent_entry *entry;
 178
 179         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 180                 return 0;
 181
 182         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
 183                                           sizeof(*entry), flags, pc);
 184         if (!event)
 185                 return 0;
 186         entry   = ring_buffer_event_data(event);
 187         entry->graph_ent                        = *trace;
 188         if (!filter_current_check_discard(buffer, call, entry, event))
 189                 ring_buffer_unlock_commit(buffer, event);
 190
 191         return 1;
 192 }
 193
 194 int trace_graph_entry(struct ftrace_graph_ent *trace)
 195 {
 196         struct trace_array *tr = graph_array;
 197         struct trace_array_cpu *data;
 198         unsigned long flags;
 199         long disabled;
 200         int ret;
 201         int cpu;
 202         int pc;
 203
 204         if (unlikely(!tr))
 205                 return 0;
 206
 207         if (!ftrace_trace_task(current))
 208                 return 0;
 209
 210         if (!ftrace_graph_addr(trace->func))
 211                 return 0;
 212
 213         local_irq_save(flags);
 214         cpu = raw_smp_processor_id();
 215         data = tr->data[cpu];
 216         disabled = atomic_inc_return(&data->disabled);
 217         if (likely(disabled == 1)) {
 218                 pc = preempt_count();
 219                 ret = __trace_graph_entry(tr, trace, flags, pc);
 220         } else {
 221                 ret = 0;
 222         }
 223         /* Only do the atomic if it is not already set */
 224         if (!test_tsk_trace_graph(current))
 225                 set_tsk_trace_graph(current);
 226
 227         atomic_dec(&data->disabled);
 228         local_irq_restore(flags);
 229
 230         return ret;
 231 }
 232
 233 static void __trace_graph_return(struct trace_array *tr,
 234                                 struct ftrace_graph_ret *trace,
 235                                 unsigned long flags,
 236                                 int pc)
 237 {
 238         struct ftrace_event_call *call = &event_funcgraph_exit;
 239         struct ring_buffer_event *event;
 240         struct ring_buffer *buffer = tr->buffer;
 241         struct ftrace_graph_ret_entry *entry;
 242
 243         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
 244                 return;
 245
 246         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
 247                                           sizeof(*entry), flags, pc);
 248         if (!event)
 249                 return;
 250         entry   = ring_buffer_event_data(event);
 251         entry->ret                              = *trace;
 252         if (!filter_current_check_discard(buffer, call, entry, event))
 253                 ring_buffer_unlock_commit(buffer, event);
 254 }
 255
 256 void trace_graph_return(struct ftrace_graph_ret *trace)
 257 {
 258         struct trace_array *tr = graph_array;
 259         struct trace_array_cpu *data;
 260         unsigned long flags;
 261         long disabled;
 262         int cpu;
 263         int pc;
 264
 265         local_irq_save(flags);
 266         cpu = raw_smp_processor_id();
 267         data = tr->data[cpu];
 268         disabled = atomic_inc_return(&data->disabled);
 269         if (likely(disabled == 1)) {
 270                 pc = preempt_count();
 271                 __trace_graph_return(tr, trace, flags, pc);
 272         }
 273         if (!trace->depth)
 274                 clear_tsk_trace_graph(current);
 275         atomic_dec(&data->disabled);
 276         local_irq_restore(flags);
 277 }
 278
 279 static int graph_trace_init(struct trace_array *tr)
 280 {
 281         int ret;
 282
 283         graph_array = tr;
 284         ret = register_ftrace_graph(&trace_graph_return,
 285                                     &trace_graph_entry);
 286         if (ret)
 287                 return ret;
 288         tracing_start_cmdline_record();
 289
 290         return 0;
 291 }
 292
 293 void set_graph_array(struct trace_array *tr)
 294 {
 295         graph_array = tr;
 296 }
 297
 298 static void graph_trace_reset(struct trace_array *tr)
 299 {
 300         tracing_stop_cmdline_record();
 301         unregister_ftrace_graph();
 302 }
 303
 304 static int max_bytes_for_cpu;
 305
 306 static enum print_line_t
 307 print_graph_cpu(struct trace_seq *s, int cpu)
 308 {
 309         int ret;
 310
 311         /*
 312          * Start with a space character - to make it stand out
 313          * to the right a bit when trace output is pasted into
 314          * email:
 315          */
 316         ret = trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
 317         if (!ret)
 318                 return TRACE_TYPE_PARTIAL_LINE;
 319
 320         return TRACE_TYPE_HANDLED;
 321 }
 322
 323 #define TRACE_GRAPH_PROCINFO_LENGTH     14
 324
 325 static enum print_line_t
 326 print_graph_proc(struct trace_seq *s, pid_t pid)
 327 {
 328         char comm[TASK_COMM_LEN];
 329         /* sign + log10(MAX_INT) + '\0' */
 330         char pid_str[11];
 331         int spaces = 0;
 332         int ret;
 333         int len;
 334         int i;
 335
 336         trace_find_cmdline(pid, comm);
 337         comm[7] = '\0';
 338         sprintf(pid_str, "%d", pid);
 339
 340         /* 1 stands for the "-" character */
 341         len = strlen(comm) + strlen(pid_str) + 1;
 342
 343         if (len < TRACE_GRAPH_PROCINFO_LENGTH)
 344                 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
 345
 346         /* First spaces to align center */
 347         for (i = 0; i < spaces / 2; i++) {
 348                 ret = trace_seq_printf(s, " ");
 349                 if (!ret)
 350                         return TRACE_TYPE_PARTIAL_LINE;
 351         }
 352
 353         ret = trace_seq_printf(s, "%s-%s", comm, pid_str);
 354         if (!ret)
 355                 return TRACE_TYPE_PARTIAL_LINE;
 356
 357         /* Last spaces to align center */
 358         for (i = 0; i < spaces - (spaces / 2); i++) {
 359                 ret = trace_seq_printf(s, " ");
 360                 if (!ret)
 361                         return TRACE_TYPE_PARTIAL_LINE;
 362         }
 363         return TRACE_TYPE_HANDLED;
 364 }
 365
 366
 367 static enum print_line_t
 368 print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
 369 {
 370         if (!trace_seq_putc(s, ' '))
 371                 return 0;
 372
 373         return trace_print_lat_fmt(s, entry);
 374 }
 375
 376 /* If the pid changed since the last trace, output this event */
 377 static enum print_line_t
 378 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
 379 {
 380         pid_t prev_pid;
 381         pid_t *last_pid;
 382         int ret;
 383
 384         if (!data)
 385                 return TRACE_TYPE_HANDLED;
 386
 387         last_pid = &(per_cpu_ptr(data, cpu)->last_pid);
 388
 389         if (*last_pid == pid)
 390                 return TRACE_TYPE_HANDLED;
 391
 392         prev_pid = *last_pid;
 393         *last_pid = pid;
 394
 395         if (prev_pid == -1)
 396                 return TRACE_TYPE_HANDLED;
 397 /*
 398  * Context-switch trace line:
 399
 400  ------------------------------------------
 401  | 1)  migration/0--1  =>  sshd-1755
 402  ------------------------------------------
 403
 404  */
 405         ret = trace_seq_printf(s,
 406                 " ------------------------------------------\n");
 407         if (!ret)
 408                 return TRACE_TYPE_PARTIAL_LINE;
 409
 410         ret = print_graph_cpu(s, cpu);
 411         if (ret == TRACE_TYPE_PARTIAL_LINE)
 412                 return TRACE_TYPE_PARTIAL_LINE;
 413
 414         ret = print_graph_proc(s, prev_pid);
 415         if (ret == TRACE_TYPE_PARTIAL_LINE)
 416                 return TRACE_TYPE_PARTIAL_LINE;
 417
 418         ret = trace_seq_printf(s, " => ");
 419         if (!ret)
 420                 return TRACE_TYPE_PARTIAL_LINE;
 421
 422         ret = print_graph_proc(s, pid);
 423         if (ret == TRACE_TYPE_PARTIAL_LINE)
 424                 return TRACE_TYPE_PARTIAL_LINE;
 425
 426         ret = trace_seq_printf(s,
 427                 "\n ------------------------------------------\n\n");
 428         if (!ret)
 429                 return TRACE_TYPE_PARTIAL_LINE;
 430
 431         return TRACE_TYPE_HANDLED;
 432 }
 433
 434 static struct ftrace_graph_ret_entry *
 435 get_return_for_leaf(struct trace_iterator *iter,
 436                 struct ftrace_graph_ent_entry *curr)
 437 {
 438         struct ring_buffer_iter *ring_iter;
 439         struct ring_buffer_event *event;
 440         struct ftrace_graph_ret_entry *next;
 441
 442         ring_iter = iter->buffer_iter[iter->cpu];
 443
 444         /* First peek to compare current entry and the next one */
 445         if (ring_iter)
 446                 event = ring_buffer_iter_peek(ring_iter, NULL);
 447         else {
 448         /* We need to consume the current entry to see the next one */
 449                 ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
 450                 event = ring_buffer_peek(iter->tr->buffer, iter->cpu,
 451                                         NULL);
 452         }
 453
 454         if (!event)
 455                 return NULL;
 456
 457         next = ring_buffer_event_data(event);
 458
 459         if (next->ent.type != TRACE_GRAPH_RET)
 460                 return NULL;
 461
 462         if (curr->ent.pid != next->ent.pid ||
 463                         curr->graph_ent.func != next->ret.func)
 464                 return NULL;
 465
 466         /* this is a leaf, now advance the iterator */
 467         if (ring_iter)
 468                 ring_buffer_read(ring_iter, NULL);
 469
 470         return next;
 471 }
 472
 473 /* Signal a overhead of time execution to the output */
 474 static int
 475 print_graph_overhead(unsigned long long duration, struct trace_seq *s)
 476 {
 477         /* If duration disappear, we don't need anything */
 478         if (!(tracer_flags.val & TRACE_GRAPH_PRINT_DURATION))
 479                 return 1;
 480
 481         /* Non nested entry or return */
 482         if (duration == -1)
 483                 return trace_seq_printf(s, "  ");
 484
 485         if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) {
 486                 /* Duration exceeded 100 msecs */
 487                 if (duration > 100000ULL)
 488                         return trace_seq_printf(s, "! ");
 489
 490                 /* Duration exceeded 10 msecs */
 491                 if (duration > 10000ULL)
 492                         return trace_seq_printf(s, "+ ");
 493         }
 494
 495         return trace_seq_printf(s, "  ");
 496 }
 497
 498 static int print_graph_abs_time(u64 t, struct trace_seq *s)
 499 {
 500         unsigned long usecs_rem;
 501
 502         usecs_rem = do_div(t, NSEC_PER_SEC);
 503         usecs_rem /= 1000;
 504
 505         return trace_seq_printf(s, "%5lu.%06lu |  ",
 506                         (unsigned long)t, usecs_rem);
 507 }
 508
 509 static enum print_line_t
 510 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
 511                 enum trace_type type, int cpu, pid_t pid)
 512 {
 513         int ret;
 514         struct trace_seq *s = &iter->seq;
 515
 516         if (addr < (unsigned long)__irqentry_text_start ||
 517                 addr >= (unsigned long)__irqentry_text_end)
 518                 return TRACE_TYPE_UNHANDLED;
 519
 520         /* Absolute time */
 521         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME) {
 522                 ret = print_graph_abs_time(iter->ts, s);
 523                 if (!ret)
 524                         return TRACE_TYPE_PARTIAL_LINE;
 525         }
 526
 527         /* Cpu */
 528         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
 529                 ret = print_graph_cpu(s, cpu);
 530                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 531                         return TRACE_TYPE_PARTIAL_LINE;
 532         }
 533
 534         /* Proc */
 535         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) {
 536                 ret = print_graph_proc(s, pid);
 537                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 538                         return TRACE_TYPE_PARTIAL_LINE;
 539                 ret = trace_seq_printf(s, " | ");
 540                 if (!ret)
 541                         return TRACE_TYPE_PARTIAL_LINE;
 542         }
 543
 544         /* No overhead */
 545         ret = print_graph_overhead(-1, s);
 546         if (!ret)
 547                 return TRACE_TYPE_PARTIAL_LINE;
 548
 549         if (type == TRACE_GRAPH_ENT)
 550                 ret = trace_seq_printf(s, "==========>");
 551         else
 552                 ret = trace_seq_printf(s, "<==========");
 553
 554         if (!ret)
 555                 return TRACE_TYPE_PARTIAL_LINE;
 556
 557         /* Don't close the duration column if haven't one */
 558         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
 559                 trace_seq_printf(s, " |");
 560         ret = trace_seq_printf(s, "\n");
 561
 562         if (!ret)
 563                 return TRACE_TYPE_PARTIAL_LINE;
 564         return TRACE_TYPE_HANDLED;
 565 }
 566
 567 enum print_line_t
 568 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
 569 {
 570         unsigned long nsecs_rem = do_div(duration, 1000);
 571         /* log10(ULONG_MAX) + '\0' */
 572         char msecs_str[21];
 573         char nsecs_str[5];
 574         int ret, len;
 575         int i;
 576
 577         sprintf(msecs_str, "%lu", (unsigned long) duration);
 578
 579         /* Print msecs */
 580         ret = trace_seq_printf(s, "%s", msecs_str);
 581         if (!ret)
 582                 return TRACE_TYPE_PARTIAL_LINE;
 583
 584         len = strlen(msecs_str);
 585
 586         /* Print nsecs (we don't want to exceed 7 numbers) */
 587         if (len < 7) {
 588                 snprintf(nsecs_str, 8 - len, "%03lu", nsecs_rem);
 589                 ret = trace_seq_printf(s, ".%s", nsecs_str);
 590                 if (!ret)
 591                         return TRACE_TYPE_PARTIAL_LINE;
 592                 len += strlen(nsecs_str);
 593         }
 594
 595         ret = trace_seq_printf(s, " us ");
 596         if (!ret)
 597                 return TRACE_TYPE_PARTIAL_LINE;
 598
 599         /* Print remaining spaces to fit the row's width */
 600         for (i = len; i < 7; i++) {
 601                 ret = trace_seq_printf(s, " ");
 602                 if (!ret)
 603                         return TRACE_TYPE_PARTIAL_LINE;
 604         }
 605         return TRACE_TYPE_HANDLED;
 606 }
 607
 608 static enum print_line_t
 609 print_graph_duration(unsigned long long duration, struct trace_seq *s)
 610 {
 611         int ret;
 612
 613         ret = trace_print_graph_duration(duration, s);
 614         if (ret != TRACE_TYPE_HANDLED)
 615                 return ret;
 616
 617         ret = trace_seq_printf(s, "|  ");
 618         if (!ret)
 619                 return TRACE_TYPE_PARTIAL_LINE;
 620
 621         return TRACE_TYPE_HANDLED;
 622 }
 623
 624 /* Case of a leaf function on its call entry */
 625 static enum print_line_t
 626 print_graph_entry_leaf(struct trace_iterator *iter,
 627                 struct ftrace_graph_ent_entry *entry,
 628                 struct ftrace_graph_ret_entry *ret_entry, struct trace_seq *s)
 629 {
 630         struct fgraph_data *data = iter->private;
 631         struct ftrace_graph_ret *graph_ret;
 632         struct ftrace_graph_ent *call;
 633         unsigned long long duration;
 634         int ret;
 635         int i;
 636
 637         graph_ret = &ret_entry->ret;
 638         call = &entry->graph_ent;
 639         duration = graph_ret->rettime - graph_ret->calltime;
 640
 641         if (data) {
 642                 int cpu = iter->cpu;
 643                 int *depth = &(per_cpu_ptr(data, cpu)->depth);
 644
 645                 /*
 646                  * Comments display at + 1 to depth. Since
 647                  * this is a leaf function, keep the comments
 648                  * equal to this depth.
 649                  */
 650                 *depth = call->depth - 1;
 651         }
 652
 653         /* Overhead */
 654         ret = print_graph_overhead(duration, s);
 655         if (!ret)
 656                 return TRACE_TYPE_PARTIAL_LINE;
 657
 658         /* Duration */
 659         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 660                 ret = print_graph_duration(duration, s);
 661                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 662                         return TRACE_TYPE_PARTIAL_LINE;
 663         }
 664
 665         /* Function */
 666         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 667                 ret = trace_seq_printf(s, " ");
 668                 if (!ret)
 669                         return TRACE_TYPE_PARTIAL_LINE;
 670         }
 671
 672         ret = trace_seq_printf(s, "%ps();\n", (void *)call->func);
 673         if (!ret)
 674                 return TRACE_TYPE_PARTIAL_LINE;
 675
 676         return TRACE_TYPE_HANDLED;
 677 }
 678
 679 static enum print_line_t
 680 print_graph_entry_nested(struct trace_iterator *iter,
 681                          struct ftrace_graph_ent_entry *entry,
 682                          struct trace_seq *s, int cpu)
 683 {
 684         struct ftrace_graph_ent *call = &entry->graph_ent;
 685         struct fgraph_data *data = iter->private;
 686         int ret;
 687         int i;
 688
 689         if (data) {
 690                 int cpu = iter->cpu;
 691                 int *depth = &(per_cpu_ptr(data, cpu)->depth);
 692
 693                 *depth = call->depth;
 694         }
 695
 696         /* No overhead */
 697         ret = print_graph_overhead(-1, s);
 698         if (!ret)
 699                 return TRACE_TYPE_PARTIAL_LINE;
 700
 701         /* No time */
 702         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 703                 ret = trace_seq_printf(s, "            |  ");
 704                 if (!ret)
 705                         return TRACE_TYPE_PARTIAL_LINE;
 706         }
 707
 708         /* Function */
 709         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 710                 ret = trace_seq_printf(s, " ");
 711                 if (!ret)
 712                         return TRACE_TYPE_PARTIAL_LINE;
 713         }
 714
 715         ret = trace_seq_printf(s, "%ps() {\n", (void *)call->func);
 716         if (!ret)
 717                 return TRACE_TYPE_PARTIAL_LINE;
 718
 719         /*
 720          * we already consumed the current entry to check the next one
 721          * and see if this is a leaf.
 722          */
 723         return TRACE_TYPE_NO_CONSUME;
 724 }
 725
 726 static enum print_line_t
 727 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
 728                      int type, unsigned long addr)
 729 {
 730         struct fgraph_data *data = iter->private;
 731         struct trace_entry *ent = iter->ent;
 732         int cpu = iter->cpu;
 733         int ret;
 734
 735         /* Pid */
 736         if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
 737                 return TRACE_TYPE_PARTIAL_LINE;
 738
 739         if (type) {
 740                 /* Interrupt */
 741                 ret = print_graph_irq(iter, addr, type, cpu, ent->pid);
 742                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 743                         return TRACE_TYPE_PARTIAL_LINE;
 744         }
 745
 746         /* Absolute time */
 747         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME) {
 748                 ret = print_graph_abs_time(iter->ts, s);
 749                 if (!ret)
 750                         return TRACE_TYPE_PARTIAL_LINE;
 751         }
 752
 753         /* Cpu */
 754         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
 755                 ret = print_graph_cpu(s, cpu);
 756                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 757                         return TRACE_TYPE_PARTIAL_LINE;
 758         }
 759
 760         /* Proc */
 761         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) {
 762                 ret = print_graph_proc(s, ent->pid);
 763                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 764                         return TRACE_TYPE_PARTIAL_LINE;
 765
 766                 ret = trace_seq_printf(s, " | ");
 767                 if (!ret)
 768                         return TRACE_TYPE_PARTIAL_LINE;
 769         }
 770
 771         /* Latency format */
 772         if (trace_flags & TRACE_ITER_LATENCY_FMT) {
 773                 ret = print_graph_lat_fmt(s, ent);
 774                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 775                         return TRACE_TYPE_PARTIAL_LINE;
 776         }
 777
 778         return 0;
 779 }
 780
 781 static enum print_line_t
 782 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
 783                         struct trace_iterator *iter)
 784 {
 785         int cpu = iter->cpu;
 786         struct ftrace_graph_ent *call = &field->graph_ent;
 787         struct ftrace_graph_ret_entry *leaf_ret;
 788
 789         if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func))
 790                 return TRACE_TYPE_PARTIAL_LINE;
 791
 792         leaf_ret = get_return_for_leaf(iter, field);
 793         if (leaf_ret)
 794                 return print_graph_entry_leaf(iter, field, leaf_ret, s);
 795         else
 796                 return print_graph_entry_nested(iter, field, s, cpu);
 797
 798 }
 799
 800 static enum print_line_t
 801 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
 802                    struct trace_entry *ent, struct trace_iterator *iter)
 803 {
 804         unsigned long long duration = trace->rettime - trace->calltime;
 805         struct fgraph_data *data = iter->private;
 806         pid_t pid = ent->pid;
 807         int cpu = iter->cpu;
 808         int ret;
 809         int i;
 810
 811         if (data) {
 812                 int cpu = iter->cpu;
 813                 int *depth = &(per_cpu_ptr(data, cpu)->depth);
 814
 815                 /*
 816                  * Comments display at + 1 to depth. This is the
 817                  * return from a function, we now want the comments
 818                  * to display at the same level of the bracket.
 819                  */
 820                 *depth = trace->depth - 1;
 821         }
 822
 823         if (print_graph_prologue(iter, s, 0, 0))
 824                 return TRACE_TYPE_PARTIAL_LINE;
 825
 826         /* Overhead */
 827         ret = print_graph_overhead(duration, s);
 828         if (!ret)
 829                 return TRACE_TYPE_PARTIAL_LINE;
 830
 831         /* Duration */
 832         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 833                 ret = print_graph_duration(duration, s);
 834                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 835                         return TRACE_TYPE_PARTIAL_LINE;
 836         }
 837
 838         /* Closing brace */
 839         for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
 840                 ret = trace_seq_printf(s, " ");
 841                 if (!ret)
 842                         return TRACE_TYPE_PARTIAL_LINE;
 843         }
 844
 845         ret = trace_seq_printf(s, "}\n");
 846         if (!ret)
 847                 return TRACE_TYPE_PARTIAL_LINE;
 848
 849         /* Overrun */
 850         if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERRUN) {
 851                 ret = trace_seq_printf(s, " (Overruns: %lu)\n",
 852                                         trace->overrun);
 853                 if (!ret)
 854                         return TRACE_TYPE_PARTIAL_LINE;
 855         }
 856
 857         ret = print_graph_irq(iter, trace->func, TRACE_GRAPH_RET, cpu, pid);
 858         if (ret == TRACE_TYPE_PARTIAL_LINE)
 859                 return TRACE_TYPE_PARTIAL_LINE;
 860
 861         return TRACE_TYPE_HANDLED;
 862 }
 863
 864 static enum print_line_t
 865 print_graph_comment(struct trace_seq *s,  struct trace_entry *ent,
 866                     struct trace_iterator *iter)
 867 {
 868         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 869         struct fgraph_data *data = iter->private;
 870         struct trace_event *event;
 871         int depth = 0;
 872         int ret;
 873         int i;
 874
 875         if (data)
 876                 depth = per_cpu_ptr(data, iter->cpu)->depth;
 877
 878         if (print_graph_prologue(iter, s, 0, 0))
 879                 return TRACE_TYPE_PARTIAL_LINE;
 880
 881         /* No overhead */
 882         ret = print_graph_overhead(-1, s);
 883         if (!ret)
 884                 return TRACE_TYPE_PARTIAL_LINE;
 885
 886         /* No time */
 887         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 888                 ret = trace_seq_printf(s, "            |  ");
 889                 if (!ret)
 890                         return TRACE_TYPE_PARTIAL_LINE;
 891         }
 892
 893         /* Indentation */
 894         if (depth > 0)
 895                 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++) {
 896                         ret = trace_seq_printf(s, " ");
 897                         if (!ret)
 898                                 return TRACE_TYPE_PARTIAL_LINE;
 899                 }
 900
 901         /* The comment */
 902         ret = trace_seq_printf(s, "/* ");
 903         if (!ret)
 904                 return TRACE_TYPE_PARTIAL_LINE;
 905
 906         switch (iter->ent->type) {
 907         case TRACE_BPRINT:
 908                 ret = trace_print_bprintk_msg_only(iter);
 909                 if (ret != TRACE_TYPE_HANDLED)
 910                         return ret;
 911                 break;
 912         case TRACE_PRINT:
 913                 ret = trace_print_printk_msg_only(iter);
 914                 if (ret != TRACE_TYPE_HANDLED)
 915                         return ret;
 916                 break;
 917         default:
 918                 event = ftrace_find_event(ent->type);
 919                 if (!event)
 920                         return TRACE_TYPE_UNHANDLED;
 921
 922                 ret = event->trace(iter, sym_flags);
 923                 if (ret != TRACE_TYPE_HANDLED)
 924                         return ret;
 925         }
 926
 927         /* Strip ending newline */
 928         if (s->buffer[s->len - 1] == '\n') {
 929                 s->buffer[s->len - 1] = '\0';
 930                 s->len--;
 931         }
 932
 933         ret = trace_seq_printf(s, " */\n");
 934         if (!ret)
 935                 return TRACE_TYPE_PARTIAL_LINE;
 936
 937         return TRACE_TYPE_HANDLED;
 938 }
 939
 940
 941 enum print_line_t
 942 print_graph_function(struct trace_iterator *iter)
 943 {
 944         struct trace_entry *entry = iter->ent;
 945         struct trace_seq *s = &iter->seq;
 946
 947         switch (entry->type) {
 948         case TRACE_GRAPH_ENT: {
 949                 /*
 950                  * print_graph_entry() may consume the current event,
 951                  * thus @field may become invalid, so we need to save it.
 952                  * sizeof(struct ftrace_graph_ent_entry) is very small,
 953                  * it can be safely saved at the stack.
 954                  */
 955                 struct ftrace_graph_ent_entry *field, saved;
 956                 trace_assign_type(field, entry);
 957                 saved = *field;
 958                 return print_graph_entry(&saved, s, iter);
 959         }
 960         case TRACE_GRAPH_RET: {
 961                 struct ftrace_graph_ret_entry *field;
 962                 trace_assign_type(field, entry);
 963                 return print_graph_return(&field->ret, s, entry, iter);
 964         }
 965         default:
 966                 return print_graph_comment(s, entry, iter);
 967         }
 968
 969         return TRACE_TYPE_HANDLED;
 970 }
 971
 972 static void print_lat_header(struct seq_file *s)
 973 {
 974         static const char spaces[] = "                " /* 16 spaces */
 975                 "    "                                  /* 4 spaces */
 976                 "                 ";                    /* 17 spaces */
 977         int size = 0;
 978
 979         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
 980                 size += 16;
 981         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
 982                 size += 4;
 983         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
 984                 size += 17;
 985
 986         seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
 987         seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
 988         seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
 989         seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
 990         seq_printf(s, "#%.*s||| / _-=> lock-depth      \n", size, spaces);
 991         seq_printf(s, "#%.*s|||| /                     \n", size, spaces);
 992 }
 993
 994 static void print_graph_headers(struct seq_file *s)
 995 {
 996         int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
 997
 998         if (lat)
 999                 print_lat_header(s);
1000
1001         /* 1st line */
1002         seq_printf(s, "#");
1003         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
1004                 seq_printf(s, "     TIME       ");
1005         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
1006                 seq_printf(s, " CPU");
1007         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
1008                 seq_printf(s, "  TASK/PID       ");
1009         if (lat)
1010                 seq_printf(s, "|||||");
1011         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
1012                 seq_printf(s, "  DURATION   ");
1013         seq_printf(s, "               FUNCTION CALLS\n");
1014
1015         /* 2nd line */
1016         seq_printf(s, "#");
1017         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
1018                 seq_printf(s, "      |         ");
1019         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
1020                 seq_printf(s, " |  ");
1021         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
1022                 seq_printf(s, "   |    |        ");
1023         if (lat)
1024                 seq_printf(s, "|||||");
1025         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
1026                 seq_printf(s, "   |   |      ");
1027         seq_printf(s, "               |   |   |   |\n");
1028 }
1029
1030 static void graph_trace_open(struct trace_iterator *iter)
1031 {
1032         /* pid and depth on the last trace processed */
1033         struct fgraph_data *data = alloc_percpu(struct fgraph_data);
1034         int cpu;
1035
1036         if (!data)
1037                 pr_warning("function graph tracer: not enough memory\n");
1038         else
1039                 for_each_possible_cpu(cpu) {
1040                         pid_t *pid = &(per_cpu_ptr(data, cpu)->last_pid);
1041                         int *depth = &(per_cpu_ptr(data, cpu)->depth);
1042                         *pid = -1;
1043                         *depth = 0;
1044                 }
1045
1046         iter->private = data;
1047 }
1048
1049 static void graph_trace_close(struct trace_iterator *iter)
1050 {
1051         free_percpu(iter->private);
1052 }
1053
1054 static struct tracer graph_trace __read_mostly = {
1055         .name           = "function_graph",
1056         .open           = graph_trace_open,
1057         .close          = graph_trace_close,
1058         .wait_pipe      = poll_wait_pipe,
1059         .init           = graph_trace_init,
1060         .reset          = graph_trace_reset,
1061         .print_line     = print_graph_function,
1062         .print_header   = print_graph_headers,
1063         .flags          = &tracer_flags,
1064 #ifdef CONFIG_FTRACE_SELFTEST
1065         .selftest       = trace_selftest_startup_function_graph,
1066 #endif
1067 };
1068
1069 static __init int init_graph_trace(void)
1070 {
1071         max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1072
1073         return register_tracer(&graph_trace);
1074 }
1075
1076 device_initcall(init_graph_trace);