SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 %pF 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         int hardirq, softirq;
 371
 372         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
 373         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
 374
 375         if (!trace_seq_printf(s, " %c%c%c",
 376                               (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
 377                                 (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ?
 378                                   'X' : '.',
 379                               (entry->flags & TRACE_FLAG_NEED_RESCHED) ?
 380                                 'N' : '.',
 381                               (hardirq && softirq) ? 'H' :
 382                                 hardirq ? 'h' : softirq ? 's' : '.'))
 383                 return 0;
 384
 385         if (entry->preempt_count)
 386                 return trace_seq_printf(s, "%x", entry->preempt_count);
 387         return trace_seq_puts(s, ".");
 388 }
 389
 390 /* If the pid changed since the last trace, output this event */
 391 static enum print_line_t
 392 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
 393 {
 394         pid_t prev_pid;
 395         pid_t *last_pid;
 396         int ret;
 397
 398         if (!data)
 399                 return TRACE_TYPE_HANDLED;
 400
 401         last_pid = &(per_cpu_ptr(data, cpu)->last_pid);
 402
 403         if (*last_pid == pid)
 404                 return TRACE_TYPE_HANDLED;
 405
 406         prev_pid = *last_pid;
 407         *last_pid = pid;
 408
 409         if (prev_pid == -1)
 410                 return TRACE_TYPE_HANDLED;
 411 /*
 412  * Context-switch trace line:
 413
 414  ------------------------------------------
 415  | 1)  migration/0--1  =>  sshd-1755
 416  ------------------------------------------
 417
 418  */
 419         ret = trace_seq_printf(s,
 420                 " ------------------------------------------\n");
 421         if (!ret)
 422                 return TRACE_TYPE_PARTIAL_LINE;
 423
 424         ret = print_graph_cpu(s, cpu);
 425         if (ret == TRACE_TYPE_PARTIAL_LINE)
 426                 return TRACE_TYPE_PARTIAL_LINE;
 427
 428         ret = print_graph_proc(s, prev_pid);
 429         if (ret == TRACE_TYPE_PARTIAL_LINE)
 430                 return TRACE_TYPE_PARTIAL_LINE;
 431
 432         ret = trace_seq_printf(s, " => ");
 433         if (!ret)
 434                 return TRACE_TYPE_PARTIAL_LINE;
 435
 436         ret = print_graph_proc(s, pid);
 437         if (ret == TRACE_TYPE_PARTIAL_LINE)
 438                 return TRACE_TYPE_PARTIAL_LINE;
 439
 440         ret = trace_seq_printf(s,
 441                 "\n ------------------------------------------\n\n");
 442         if (!ret)
 443                 return TRACE_TYPE_PARTIAL_LINE;
 444
 445         return TRACE_TYPE_HANDLED;
 446 }
 447
 448 static struct ftrace_graph_ret_entry *
 449 get_return_for_leaf(struct trace_iterator *iter,
 450                 struct ftrace_graph_ent_entry *curr)
 451 {
 452         struct ring_buffer_iter *ring_iter;
 453         struct ring_buffer_event *event;
 454         struct ftrace_graph_ret_entry *next;
 455
 456         ring_iter = iter->buffer_iter[iter->cpu];
 457
 458         /* First peek to compare current entry and the next one */
 459         if (ring_iter)
 460                 event = ring_buffer_iter_peek(ring_iter, NULL);
 461         else {
 462         /* We need to consume the current entry to see the next one */
 463                 ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
 464                 event = ring_buffer_peek(iter->tr->buffer, iter->cpu,
 465                                         NULL);
 466         }
 467
 468         if (!event)
 469                 return NULL;
 470
 471         next = ring_buffer_event_data(event);
 472
 473         if (next->ent.type != TRACE_GRAPH_RET)
 474                 return NULL;
 475
 476         if (curr->ent.pid != next->ent.pid ||
 477                         curr->graph_ent.func != next->ret.func)
 478                 return NULL;
 479
 480         /* this is a leaf, now advance the iterator */
 481         if (ring_iter)
 482                 ring_buffer_read(ring_iter, NULL);
 483
 484         return next;
 485 }
 486
 487 /* Signal a overhead of time execution to the output */
 488 static int
 489 print_graph_overhead(unsigned long long duration, struct trace_seq *s)
 490 {
 491         /* If duration disappear, we don't need anything */
 492         if (!(tracer_flags.val & TRACE_GRAPH_PRINT_DURATION))
 493                 return 1;
 494
 495         /* Non nested entry or return */
 496         if (duration == -1)
 497                 return trace_seq_printf(s, "  ");
 498
 499         if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) {
 500                 /* Duration exceeded 100 msecs */
 501                 if (duration > 100000ULL)
 502                         return trace_seq_printf(s, "! ");
 503
 504                 /* Duration exceeded 10 msecs */
 505                 if (duration > 10000ULL)
 506                         return trace_seq_printf(s, "+ ");
 507         }
 508
 509         return trace_seq_printf(s, "  ");
 510 }
 511
 512 static int print_graph_abs_time(u64 t, struct trace_seq *s)
 513 {
 514         unsigned long usecs_rem;
 515
 516         usecs_rem = do_div(t, NSEC_PER_SEC);
 517         usecs_rem /= 1000;
 518
 519         return trace_seq_printf(s, "%5lu.%06lu |  ",
 520                         (unsigned long)t, usecs_rem);
 521 }
 522
 523 static enum print_line_t
 524 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
 525                 enum trace_type type, int cpu, pid_t pid)
 526 {
 527         int ret;
 528         struct trace_seq *s = &iter->seq;
 529
 530         if (addr < (unsigned long)__irqentry_text_start ||
 531                 addr >= (unsigned long)__irqentry_text_end)
 532                 return TRACE_TYPE_UNHANDLED;
 533
 534         /* Absolute time */
 535         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME) {
 536                 ret = print_graph_abs_time(iter->ts, s);
 537                 if (!ret)
 538                         return TRACE_TYPE_PARTIAL_LINE;
 539         }
 540
 541         /* Cpu */
 542         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
 543                 ret = print_graph_cpu(s, cpu);
 544                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 545                         return TRACE_TYPE_PARTIAL_LINE;
 546         }
 547
 548         /* Proc */
 549         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) {
 550                 ret = print_graph_proc(s, pid);
 551                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 552                         return TRACE_TYPE_PARTIAL_LINE;
 553                 ret = trace_seq_printf(s, " | ");
 554                 if (!ret)
 555                         return TRACE_TYPE_PARTIAL_LINE;
 556         }
 557
 558         /* No overhead */
 559         ret = print_graph_overhead(-1, s);
 560         if (!ret)
 561                 return TRACE_TYPE_PARTIAL_LINE;
 562
 563         if (type == TRACE_GRAPH_ENT)
 564                 ret = trace_seq_printf(s, "==========>");
 565         else
 566                 ret = trace_seq_printf(s, "<==========");
 567
 568         if (!ret)
 569                 return TRACE_TYPE_PARTIAL_LINE;
 570
 571         /* Don't close the duration column if haven't one */
 572         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
 573                 trace_seq_printf(s, " |");
 574         ret = trace_seq_printf(s, "\n");
 575
 576         if (!ret)
 577                 return TRACE_TYPE_PARTIAL_LINE;
 578         return TRACE_TYPE_HANDLED;
 579 }
 580
 581 enum print_line_t
 582 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
 583 {
 584         unsigned long nsecs_rem = do_div(duration, 1000);
 585         /* log10(ULONG_MAX) + '\0' */
 586         char msecs_str[21];
 587         char nsecs_str[5];
 588         int ret, len;
 589         int i;
 590
 591         sprintf(msecs_str, "%lu", (unsigned long) duration);
 592
 593         /* Print msecs */
 594         ret = trace_seq_printf(s, "%s", msecs_str);
 595         if (!ret)
 596                 return TRACE_TYPE_PARTIAL_LINE;
 597
 598         len = strlen(msecs_str);
 599
 600         /* Print nsecs (we don't want to exceed 7 numbers) */
 601         if (len < 7) {
 602                 snprintf(nsecs_str, 8 - len, "%03lu", nsecs_rem);
 603                 ret = trace_seq_printf(s, ".%s", nsecs_str);
 604                 if (!ret)
 605                         return TRACE_TYPE_PARTIAL_LINE;
 606                 len += strlen(nsecs_str);
 607         }
 608
 609         ret = trace_seq_printf(s, " us ");
 610         if (!ret)
 611                 return TRACE_TYPE_PARTIAL_LINE;
 612
 613         /* Print remaining spaces to fit the row's width */
 614         for (i = len; i < 7; i++) {
 615                 ret = trace_seq_printf(s, " ");
 616                 if (!ret)
 617                         return TRACE_TYPE_PARTIAL_LINE;
 618         }
 619         return TRACE_TYPE_HANDLED;
 620 }
 621
 622 static enum print_line_t
 623 print_graph_duration(unsigned long long duration, struct trace_seq *s)
 624 {
 625         int ret;
 626
 627         ret = trace_print_graph_duration(duration, s);
 628         if (ret != TRACE_TYPE_HANDLED)
 629                 return ret;
 630
 631         ret = trace_seq_printf(s, "|  ");
 632         if (!ret)
 633                 return TRACE_TYPE_PARTIAL_LINE;
 634
 635         return TRACE_TYPE_HANDLED;
 636 }
 637
 638 /* Case of a leaf function on its call entry */
 639 static enum print_line_t
 640 print_graph_entry_leaf(struct trace_iterator *iter,
 641                 struct ftrace_graph_ent_entry *entry,
 642                 struct ftrace_graph_ret_entry *ret_entry, struct trace_seq *s)
 643 {
 644         struct fgraph_data *data = iter->private;
 645         struct ftrace_graph_ret *graph_ret;
 646         struct ftrace_graph_ent *call;
 647         unsigned long long duration;
 648         int ret;
 649         int i;
 650
 651         graph_ret = &ret_entry->ret;
 652         call = &entry->graph_ent;
 653         duration = graph_ret->rettime - graph_ret->calltime;
 654
 655         if (data) {
 656                 int cpu = iter->cpu;
 657                 int *depth = &(per_cpu_ptr(data, cpu)->depth);
 658
 659                 /*
 660                  * Comments display at + 1 to depth. Since
 661                  * this is a leaf function, keep the comments
 662                  * equal to this depth.
 663                  */
 664                 *depth = call->depth - 1;
 665         }
 666
 667         /* Overhead */
 668         ret = print_graph_overhead(duration, s);
 669         if (!ret)
 670                 return TRACE_TYPE_PARTIAL_LINE;
 671
 672         /* Duration */
 673         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 674                 ret = print_graph_duration(duration, s);
 675                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 676                         return TRACE_TYPE_PARTIAL_LINE;
 677         }
 678
 679         /* Function */
 680         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 681                 ret = trace_seq_printf(s, " ");
 682                 if (!ret)
 683                         return TRACE_TYPE_PARTIAL_LINE;
 684         }
 685
 686         ret = trace_seq_printf(s, "%pf();\n", (void *)call->func);
 687         if (!ret)
 688                 return TRACE_TYPE_PARTIAL_LINE;
 689
 690         return TRACE_TYPE_HANDLED;
 691 }
 692
 693 static enum print_line_t
 694 print_graph_entry_nested(struct trace_iterator *iter,
 695                          struct ftrace_graph_ent_entry *entry,
 696                          struct trace_seq *s, int cpu)
 697 {
 698         struct ftrace_graph_ent *call = &entry->graph_ent;
 699         struct fgraph_data *data = iter->private;
 700         int ret;
 701         int i;
 702
 703         if (data) {
 704                 int cpu = iter->cpu;
 705                 int *depth = &(per_cpu_ptr(data, cpu)->depth);
 706
 707                 *depth = call->depth;
 708         }
 709
 710         /* No overhead */
 711         ret = print_graph_overhead(-1, s);
 712         if (!ret)
 713                 return TRACE_TYPE_PARTIAL_LINE;
 714
 715         /* No time */
 716         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 717                 ret = trace_seq_printf(s, "            |  ");
 718                 if (!ret)
 719                         return TRACE_TYPE_PARTIAL_LINE;
 720         }
 721
 722         /* Function */
 723         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 724                 ret = trace_seq_printf(s, " ");
 725                 if (!ret)
 726                         return TRACE_TYPE_PARTIAL_LINE;
 727         }
 728
 729         ret = trace_seq_printf(s, "%pf() {\n", (void *)call->func);
 730         if (!ret)
 731                 return TRACE_TYPE_PARTIAL_LINE;
 732
 733         /*
 734          * we already consumed the current entry to check the next one
 735          * and see if this is a leaf.
 736          */
 737         return TRACE_TYPE_NO_CONSUME;
 738 }
 739
 740 static enum print_line_t
 741 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
 742                      int type, unsigned long addr)
 743 {
 744         struct fgraph_data *data = iter->private;
 745         struct trace_entry *ent = iter->ent;
 746         int cpu = iter->cpu;
 747         int ret;
 748
 749         /* Pid */
 750         if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
 751                 return TRACE_TYPE_PARTIAL_LINE;
 752
 753         if (type) {
 754                 /* Interrupt */
 755                 ret = print_graph_irq(iter, addr, type, cpu, ent->pid);
 756                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 757                         return TRACE_TYPE_PARTIAL_LINE;
 758         }
 759
 760         /* Absolute time */
 761         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME) {
 762                 ret = print_graph_abs_time(iter->ts, s);
 763                 if (!ret)
 764                         return TRACE_TYPE_PARTIAL_LINE;
 765         }
 766
 767         /* Cpu */
 768         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
 769                 ret = print_graph_cpu(s, cpu);
 770                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 771                         return TRACE_TYPE_PARTIAL_LINE;
 772         }
 773
 774         /* Proc */
 775         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) {
 776                 ret = print_graph_proc(s, ent->pid);
 777                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 778                         return TRACE_TYPE_PARTIAL_LINE;
 779
 780                 ret = trace_seq_printf(s, " | ");
 781                 if (!ret)
 782                         return TRACE_TYPE_PARTIAL_LINE;
 783         }
 784
 785         /* Latency format */
 786         if (trace_flags & TRACE_ITER_LATENCY_FMT) {
 787                 ret = print_graph_lat_fmt(s, ent);
 788                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 789                         return TRACE_TYPE_PARTIAL_LINE;
 790         }
 791
 792         return 0;
 793 }
 794
 795 static enum print_line_t
 796 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
 797                         struct trace_iterator *iter)
 798 {
 799         int cpu = iter->cpu;
 800         struct ftrace_graph_ent *call = &field->graph_ent;
 801         struct ftrace_graph_ret_entry *leaf_ret;
 802
 803         if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func))
 804                 return TRACE_TYPE_PARTIAL_LINE;
 805
 806         leaf_ret = get_return_for_leaf(iter, field);
 807         if (leaf_ret)
 808                 return print_graph_entry_leaf(iter, field, leaf_ret, s);
 809         else
 810                 return print_graph_entry_nested(iter, field, s, cpu);
 811
 812 }
 813
 814 static enum print_line_t
 815 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
 816                    struct trace_entry *ent, struct trace_iterator *iter)
 817 {
 818         unsigned long long duration = trace->rettime - trace->calltime;
 819         struct fgraph_data *data = iter->private;
 820         pid_t pid = ent->pid;
 821         int cpu = iter->cpu;
 822         int ret;
 823         int i;
 824
 825         if (data) {
 826                 int cpu = iter->cpu;
 827                 int *depth = &(per_cpu_ptr(data, cpu)->depth);
 828
 829                 /*
 830                  * Comments display at + 1 to depth. This is the
 831                  * return from a function, we now want the comments
 832                  * to display at the same level of the bracket.
 833                  */
 834                 *depth = trace->depth - 1;
 835         }
 836
 837         if (print_graph_prologue(iter, s, 0, 0))
 838                 return TRACE_TYPE_PARTIAL_LINE;
 839
 840         /* Overhead */
 841         ret = print_graph_overhead(duration, s);
 842         if (!ret)
 843                 return TRACE_TYPE_PARTIAL_LINE;
 844
 845         /* Duration */
 846         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 847                 ret = print_graph_duration(duration, s);
 848                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 849                         return TRACE_TYPE_PARTIAL_LINE;
 850         }
 851
 852         /* Closing brace */
 853         for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
 854                 ret = trace_seq_printf(s, " ");
 855                 if (!ret)
 856                         return TRACE_TYPE_PARTIAL_LINE;
 857         }
 858
 859         ret = trace_seq_printf(s, "}\n");
 860         if (!ret)
 861                 return TRACE_TYPE_PARTIAL_LINE;
 862
 863         /* Overrun */
 864         if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERRUN) {
 865                 ret = trace_seq_printf(s, " (Overruns: %lu)\n",
 866                                         trace->overrun);
 867                 if (!ret)
 868                         return TRACE_TYPE_PARTIAL_LINE;
 869         }
 870
 871         ret = print_graph_irq(iter, trace->func, TRACE_GRAPH_RET, cpu, pid);
 872         if (ret == TRACE_TYPE_PARTIAL_LINE)
 873                 return TRACE_TYPE_PARTIAL_LINE;
 874
 875         return TRACE_TYPE_HANDLED;
 876 }
 877
 878 static enum print_line_t
 879 print_graph_comment(struct trace_seq *s,  struct trace_entry *ent,
 880                     struct trace_iterator *iter)
 881 {
 882         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 883         struct fgraph_data *data = iter->private;
 884         struct trace_event *event;
 885         int depth = 0;
 886         int ret;
 887         int i;
 888
 889         if (data)
 890                 depth = per_cpu_ptr(data, iter->cpu)->depth;
 891
 892         if (print_graph_prologue(iter, s, 0, 0))
 893                 return TRACE_TYPE_PARTIAL_LINE;
 894
 895         /* No overhead */
 896         ret = print_graph_overhead(-1, s);
 897         if (!ret)
 898                 return TRACE_TYPE_PARTIAL_LINE;
 899
 900         /* No time */
 901         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 902                 ret = trace_seq_printf(s, "            |  ");
 903                 if (!ret)
 904                         return TRACE_TYPE_PARTIAL_LINE;
 905         }
 906
 907         /* Indentation */
 908         if (depth > 0)
 909                 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++) {
 910                         ret = trace_seq_printf(s, " ");
 911                         if (!ret)
 912                                 return TRACE_TYPE_PARTIAL_LINE;
 913                 }
 914
 915         /* The comment */
 916         ret = trace_seq_printf(s, "/* ");
 917         if (!ret)
 918                 return TRACE_TYPE_PARTIAL_LINE;
 919
 920         switch (iter->ent->type) {
 921         case TRACE_BPRINT:
 922                 ret = trace_print_bprintk_msg_only(iter);
 923                 if (ret != TRACE_TYPE_HANDLED)
 924                         return ret;
 925                 break;
 926         case TRACE_PRINT:
 927                 ret = trace_print_printk_msg_only(iter);
 928                 if (ret != TRACE_TYPE_HANDLED)
 929                         return ret;
 930                 break;
 931         default:
 932                 event = ftrace_find_event(ent->type);
 933                 if (!event)
 934                         return TRACE_TYPE_UNHANDLED;
 935
 936                 ret = event->trace(iter, sym_flags);
 937                 if (ret != TRACE_TYPE_HANDLED)
 938                         return ret;
 939         }
 940
 941         /* Strip ending newline */
 942         if (s->buffer[s->len - 1] == '\n') {
 943                 s->buffer[s->len - 1] = '\0';
 944                 s->len--;
 945         }
 946
 947         ret = trace_seq_printf(s, " */\n");
 948         if (!ret)
 949                 return TRACE_TYPE_PARTIAL_LINE;
 950
 951         return TRACE_TYPE_HANDLED;
 952 }
 953
 954
 955 enum print_line_t
 956 print_graph_function(struct trace_iterator *iter)
 957 {
 958         struct trace_entry *entry = iter->ent;
 959         struct trace_seq *s = &iter->seq;
 960
 961         switch (entry->type) {
 962         case TRACE_GRAPH_ENT: {
 963                 /*
 964                  * print_graph_entry() may consume the current event,
 965                  * thus @field may become invalid, so we need to save it.
 966                  * sizeof(struct ftrace_graph_ent_entry) is very small,
 967                  * it can be safely saved at the stack.
 968                  */
 969                 struct ftrace_graph_ent_entry *field, saved;
 970                 trace_assign_type(field, entry);
 971                 saved = *field;
 972                 return print_graph_entry(&saved, s, iter);
 973         }
 974         case TRACE_GRAPH_RET: {
 975                 struct ftrace_graph_ret_entry *field;
 976                 trace_assign_type(field, entry);
 977                 return print_graph_return(&field->ret, s, entry, iter);
 978         }
 979         default:
 980                 return print_graph_comment(s, entry, iter);
 981         }
 982
 983         return TRACE_TYPE_HANDLED;
 984 }
 985
 986 static void print_lat_header(struct seq_file *s)
 987 {
 988         static const char spaces[] = "                " /* 16 spaces */
 989                 "    "                                  /* 4 spaces */
 990                 "                 ";                    /* 17 spaces */
 991         int size = 0;
 992
 993         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
 994                 size += 16;
 995         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
 996                 size += 4;
 997         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
 998                 size += 17;
 999
1000         seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1001         seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1002         seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1003         seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1004         seq_printf(s, "#%.*s||| /                      \n", size, spaces);
1005         seq_printf(s, "#%.*s||||                       \n", size, spaces);
1006 }
1007
1008 static void print_graph_headers(struct seq_file *s)
1009 {
1010         int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
1011
1012         if (lat)
1013                 print_lat_header(s);
1014
1015         /* 1st line */
1016         seq_printf(s, "#");
1017         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
1018                 seq_printf(s, "     TIME       ");
1019         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
1020                 seq_printf(s, " CPU");
1021         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
1022                 seq_printf(s, "  TASK/PID       ");
1023         if (lat)
1024                 seq_printf(s, "||||");
1025         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
1026                 seq_printf(s, "  DURATION   ");
1027         seq_printf(s, "               FUNCTION CALLS\n");
1028
1029         /* 2nd line */
1030         seq_printf(s, "#");
1031         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
1032                 seq_printf(s, "      |         ");
1033         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
1034                 seq_printf(s, " |  ");
1035         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
1036                 seq_printf(s, "   |    |        ");
1037         if (lat)
1038                 seq_printf(s, "||||");
1039         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
1040                 seq_printf(s, "   |   |      ");
1041         seq_printf(s, "               |   |   |   |\n");
1042 }
1043
1044 static void graph_trace_open(struct trace_iterator *iter)
1045 {
1046         /* pid and depth on the last trace processed */
1047         struct fgraph_data *data = alloc_percpu(struct fgraph_data);
1048         int cpu;
1049
1050         if (!data)
1051                 pr_warning("function graph tracer: not enough memory\n");
1052         else
1053                 for_each_possible_cpu(cpu) {
1054                         pid_t *pid = &(per_cpu_ptr(data, cpu)->last_pid);
1055                         int *depth = &(per_cpu_ptr(data, cpu)->depth);
1056                         *pid = -1;
1057                         *depth = 0;
1058                 }
1059
1060         iter->private = data;
1061 }
1062
1063 static void graph_trace_close(struct trace_iterator *iter)
1064 {
1065         free_percpu(iter->private);
1066 }
1067
1068 static struct tracer graph_trace __read_mostly = {
1069         .name           = "function_graph",
1070         .open           = graph_trace_open,
1071         .close          = graph_trace_close,
1072         .wait_pipe      = poll_wait_pipe,
1073         .init           = graph_trace_init,
1074         .reset          = graph_trace_reset,
1075         .print_line     = print_graph_function,
1076         .print_header   = print_graph_headers,
1077         .flags          = &tracer_flags,
1078 #ifdef CONFIG_FTRACE_SELFTEST
1079         .selftest       = trace_selftest_startup_function_graph,
1080 #endif
1081 };
1082
1083 static __init int init_graph_trace(void)
1084 {
1085         max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1086
1087         return register_tracer(&graph_trace);
1088 }
1089
1090 device_initcall(init_graph_trace);