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/slab.h>
  13 #include <linux/fs.h>
  14
  15 #include "trace.h"
  16 #include "trace_output.h"
  17
  18 struct fgraph_cpu_data {
  19         pid_t           last_pid;
  20         int             depth;
  21         int             ignore;
  22         unsigned long   enter_funcs[FTRACE_RETFUNC_DEPTH];
  23 };
  24
  25 struct fgraph_data {
  26         struct fgraph_cpu_data          *cpu_data;
  27
  28         /* Place to preserve last processed entry. */
  29         struct ftrace_graph_ent_entry   ent;
  30         struct ftrace_graph_ret_entry   ret;
  31         int                             failed;
  32         int                             cpu;
  33 };
  34
  35 #define TRACE_GRAPH_INDENT      2
  36
  37 /* Flag options */
  38 #define TRACE_GRAPH_PRINT_OVERRUN       0x1
  39 #define TRACE_GRAPH_PRINT_CPU           0x2
  40 #define TRACE_GRAPH_PRINT_OVERHEAD      0x4
  41 #define TRACE_GRAPH_PRINT_PROC          0x8
  42 #define TRACE_GRAPH_PRINT_DURATION      0x10
  43 #define TRACE_GRAPH_PRINT_ABS_TIME      0X20
  44
  45 static struct tracer_opt trace_opts[] = {
  46         /* Display overruns? (for self-debug purpose) */
  47         { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
  48         /* Display CPU ? */
  49         { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
  50         /* Display Overhead ? */
  51         { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
  52         /* Display proc name/pid */
  53         { TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
  54         /* Display duration of execution */
  55         { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
  56         /* Display absolute time of an entry */
  57         { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
  58         { } /* Empty entry */
  59 };
  60
  61 static struct tracer_flags tracer_flags = {
  62         /* Don't display overruns and proc by default */
  63         .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
  64                TRACE_GRAPH_PRINT_DURATION,
  65         .opts = trace_opts
  66 };
  67
  68 static struct trace_array *graph_array;
  69
  70
  71 /* Add a function return address to the trace stack on thread info.*/
  72 int
  73 ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
  74                          unsigned long frame_pointer)
  75 {
  76         unsigned long long calltime;
  77         int index;
  78
  79         if (!current->ret_stack)
  80                 return -EBUSY;
  81
  82         /*
  83          * We must make sure the ret_stack is tested before we read
  84          * anything else.
  85          */
  86         smp_rmb();
  87
  88         /* The return trace stack is full */
  89         if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
  90                 atomic_inc(&current->trace_overrun);
  91                 return -EBUSY;
  92         }
  93
  94         calltime = trace_clock_local();
  95
  96         index = ++current->curr_ret_stack;
  97         barrier();
  98         current->ret_stack[index].ret = ret;
  99         current->ret_stack[index].func = func;
 100         current->ret_stack[index].calltime = calltime;
 101         current->ret_stack[index].subtime = 0;
 102         current->ret_stack[index].fp = frame_pointer;
 103         *depth = index;
 104
 105         return 0;
 106 }
 107
 108 /* Retrieve a function return address to the trace stack on thread info.*/
 109 static void
 110 ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
 111                         unsigned long frame_pointer)
 112 {
 113         int index;
 114
 115         index = current->curr_ret_stack;
 116
 117         if (unlikely(index < 0)) {
 118                 ftrace_graph_stop();
 119                 WARN_ON(1);
 120                 /* Might as well panic, otherwise we have no where to go */
 121                 *ret = (unsigned long)panic;
 122                 return;
 123         }
 124
 125 #ifdef CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST
 126         /*
 127          * The arch may choose to record the frame pointer used
 128          * and check it here to make sure that it is what we expect it
 129          * to be. If gcc does not set the place holder of the return
 130          * address in the frame pointer, and does a copy instead, then
 131          * the function graph trace will fail. This test detects this
 132          * case.
 133          *
 134          * Currently, x86_32 with optimize for size (-Os) makes the latest
 135          * gcc do the above.
 136          */
 137         if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
 138                 ftrace_graph_stop();
 139                 WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
 140                      "  from func %ps return to %lx\n",
 141                      current->ret_stack[index].fp,
 142                      frame_pointer,
 143                      (void *)current->ret_stack[index].func,
 144                      current->ret_stack[index].ret);
 145                 *ret = (unsigned long)panic;
 146                 return;
 147         }
 148 #endif
 149
 150         *ret = current->ret_stack[index].ret;
 151         trace->func = current->ret_stack[index].func;
 152         trace->calltime = current->ret_stack[index].calltime;
 153         trace->overrun = atomic_read(&current->trace_overrun);
 154         trace->depth = index;
 155 }
 156
 157 /*
 158  * Send the trace to the ring-buffer.
 159  * @return the original return address.
 160  */
 161 unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
 162 {
 163         struct ftrace_graph_ret trace;
 164         unsigned long ret;
 165
 166         ftrace_pop_return_trace(&trace, &ret, frame_pointer);
 167         trace.rettime = trace_clock_local();
 168         ftrace_graph_return(&trace);
 169         barrier();
 170         current->curr_ret_stack--;
 171
 172         if (unlikely(!ret)) {
 173                 ftrace_graph_stop();
 174                 WARN_ON(1);
 175                 /* Might as well panic. What else to do? */
 176                 ret = (unsigned long)panic;
 177         }
 178
 179         return ret;
 180 }
 181
 182 static int __trace_graph_entry(struct trace_array *tr,
 183                                 struct ftrace_graph_ent *trace,
 184                                 unsigned long flags,
 185                                 int pc)
 186 {
 187         struct ftrace_event_call *call = &event_funcgraph_entry;
 188         struct ring_buffer_event *event;
 189         struct ring_buffer *buffer = tr->buffer;
 190         struct ftrace_graph_ent_entry *entry;
 191
 192         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
 193                 return 0;
 194
 195         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
 196                                           sizeof(*entry), flags, pc);
 197         if (!event)
 198                 return 0;
 199         entry   = ring_buffer_event_data(event);
 200         entry->graph_ent                        = *trace;
 201         if (!filter_current_check_discard(buffer, call, entry, event))
 202                 ring_buffer_unlock_commit(buffer, event);
 203
 204         return 1;
 205 }
 206
 207 int trace_graph_entry(struct ftrace_graph_ent *trace)
 208 {
 209         struct trace_array *tr = graph_array;
 210         struct trace_array_cpu *data;
 211         unsigned long flags;
 212         long disabled;
 213         int ret;
 214         int cpu;
 215         int pc;
 216
 217         if (!ftrace_trace_task(current))
 218                 return 0;
 219
 220         /* trace it when it is-nested-in or is a function enabled. */
 221         if (!(trace->depth || ftrace_graph_addr(trace->func)))
 222                 return 0;
 223
 224         local_irq_save(flags);
 225         cpu = raw_smp_processor_id();
 226         data = tr->data[cpu];
 227         disabled = atomic_inc_return(&data->disabled);
 228         if (likely(disabled == 1)) {
 229                 pc = preempt_count();
 230                 ret = __trace_graph_entry(tr, trace, flags, pc);
 231         } else {
 232                 ret = 0;
 233         }
 234
 235         atomic_dec(&data->disabled);
 236         local_irq_restore(flags);
 237
 238         return ret;
 239 }
 240
 241 int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
 242 {
 243         if (tracing_thresh)
 244                 return 1;
 245         else
 246                 return trace_graph_entry(trace);
 247 }
 248
 249 static void __trace_graph_return(struct trace_array *tr,
 250                                 struct ftrace_graph_ret *trace,
 251                                 unsigned long flags,
 252                                 int pc)
 253 {
 254         struct ftrace_event_call *call = &event_funcgraph_exit;
 255         struct ring_buffer_event *event;
 256         struct ring_buffer *buffer = tr->buffer;
 257         struct ftrace_graph_ret_entry *entry;
 258
 259         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
 260                 return;
 261
 262         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
 263                                           sizeof(*entry), flags, pc);
 264         if (!event)
 265                 return;
 266         entry   = ring_buffer_event_data(event);
 267         entry->ret                              = *trace;
 268         if (!filter_current_check_discard(buffer, call, entry, event))
 269                 ring_buffer_unlock_commit(buffer, event);
 270 }
 271
 272 void trace_graph_return(struct ftrace_graph_ret *trace)
 273 {
 274         struct trace_array *tr = graph_array;
 275         struct trace_array_cpu *data;
 276         unsigned long flags;
 277         long disabled;
 278         int cpu;
 279         int pc;
 280
 281         local_irq_save(flags);
 282         cpu = raw_smp_processor_id();
 283         data = tr->data[cpu];
 284         disabled = atomic_inc_return(&data->disabled);
 285         if (likely(disabled == 1)) {
 286                 pc = preempt_count();
 287                 __trace_graph_return(tr, trace, flags, pc);
 288         }
 289         atomic_dec(&data->disabled);
 290         local_irq_restore(flags);
 291 }
 292
 293 void set_graph_array(struct trace_array *tr)
 294 {
 295         graph_array = tr;
 296
 297         /* Make graph_array visible before we start tracing */
 298
 299         smp_mb();
 300 }
 301
 302 void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
 303 {
 304         if (tracing_thresh &&
 305             (trace->rettime - trace->calltime < tracing_thresh))
 306                 return;
 307         else
 308                 trace_graph_return(trace);
 309 }
 310
 311 static int graph_trace_init(struct trace_array *tr)
 312 {
 313         int ret;
 314
 315         set_graph_array(tr);
 316         if (tracing_thresh)
 317                 ret = register_ftrace_graph(&trace_graph_thresh_return,
 318                                             &trace_graph_thresh_entry);
 319         else
 320                 ret = register_ftrace_graph(&trace_graph_return,
 321                                             &trace_graph_entry);
 322         if (ret)
 323                 return ret;
 324         tracing_start_cmdline_record();
 325
 326         return 0;
 327 }
 328
 329 static void graph_trace_reset(struct trace_array *tr)
 330 {
 331         tracing_stop_cmdline_record();
 332         unregister_ftrace_graph();
 333 }
 334
 335 static int max_bytes_for_cpu;
 336
 337 static enum print_line_t
 338 print_graph_cpu(struct trace_seq *s, int cpu)
 339 {
 340         int ret;
 341
 342         /*
 343          * Start with a space character - to make it stand out
 344          * to the right a bit when trace output is pasted into
 345          * email:
 346          */
 347         ret = trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
 348         if (!ret)
 349                 return TRACE_TYPE_PARTIAL_LINE;
 350
 351         return TRACE_TYPE_HANDLED;
 352 }
 353
 354 #define TRACE_GRAPH_PROCINFO_LENGTH     14
 355
 356 static enum print_line_t
 357 print_graph_proc(struct trace_seq *s, pid_t pid)
 358 {
 359         char comm[TASK_COMM_LEN];
 360         /* sign + log10(MAX_INT) + '\0' */
 361         char pid_str[11];
 362         int spaces = 0;
 363         int ret;
 364         int len;
 365         int i;
 366
 367         trace_find_cmdline(pid, comm);
 368         comm[7] = '\0';
 369         sprintf(pid_str, "%d", pid);
 370
 371         /* 1 stands for the "-" character */
 372         len = strlen(comm) + strlen(pid_str) + 1;
 373
 374         if (len < TRACE_GRAPH_PROCINFO_LENGTH)
 375                 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
 376
 377         /* First spaces to align center */
 378         for (i = 0; i < spaces / 2; i++) {
 379                 ret = trace_seq_printf(s, " ");
 380                 if (!ret)
 381                         return TRACE_TYPE_PARTIAL_LINE;
 382         }
 383
 384         ret = trace_seq_printf(s, "%s-%s", comm, pid_str);
 385         if (!ret)
 386                 return TRACE_TYPE_PARTIAL_LINE;
 387
 388         /* Last spaces to align center */
 389         for (i = 0; i < spaces - (spaces / 2); i++) {
 390                 ret = trace_seq_printf(s, " ");
 391                 if (!ret)
 392                         return TRACE_TYPE_PARTIAL_LINE;
 393         }
 394         return TRACE_TYPE_HANDLED;
 395 }
 396
 397
 398 static enum print_line_t
 399 print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
 400 {
 401         if (!trace_seq_putc(s, ' '))
 402                 return 0;
 403
 404         return trace_print_lat_fmt(s, entry);
 405 }
 406
 407 /* If the pid changed since the last trace, output this event */
 408 static enum print_line_t
 409 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
 410 {
 411         pid_t prev_pid;
 412         pid_t *last_pid;
 413         int ret;
 414
 415         if (!data)
 416                 return TRACE_TYPE_HANDLED;
 417
 418         last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
 419
 420         if (*last_pid == pid)
 421                 return TRACE_TYPE_HANDLED;
 422
 423         prev_pid = *last_pid;
 424         *last_pid = pid;
 425
 426         if (prev_pid == -1)
 427                 return TRACE_TYPE_HANDLED;
 428 /*
 429  * Context-switch trace line:
 430
 431  ------------------------------------------
 432  | 1)  migration/0--1  =>  sshd-1755
 433  ------------------------------------------
 434
 435  */
 436         ret = trace_seq_printf(s,
 437                 " ------------------------------------------\n");
 438         if (!ret)
 439                 return TRACE_TYPE_PARTIAL_LINE;
 440
 441         ret = print_graph_cpu(s, cpu);
 442         if (ret == TRACE_TYPE_PARTIAL_LINE)
 443                 return TRACE_TYPE_PARTIAL_LINE;
 444
 445         ret = print_graph_proc(s, prev_pid);
 446         if (ret == TRACE_TYPE_PARTIAL_LINE)
 447                 return TRACE_TYPE_PARTIAL_LINE;
 448
 449         ret = trace_seq_printf(s, " => ");
 450         if (!ret)
 451                 return TRACE_TYPE_PARTIAL_LINE;
 452
 453         ret = print_graph_proc(s, pid);
 454         if (ret == TRACE_TYPE_PARTIAL_LINE)
 455                 return TRACE_TYPE_PARTIAL_LINE;
 456
 457         ret = trace_seq_printf(s,
 458                 "\n ------------------------------------------\n\n");
 459         if (!ret)
 460                 return TRACE_TYPE_PARTIAL_LINE;
 461
 462         return TRACE_TYPE_HANDLED;
 463 }
 464
 465 static struct ftrace_graph_ret_entry *
 466 get_return_for_leaf(struct trace_iterator *iter,
 467                 struct ftrace_graph_ent_entry *curr)
 468 {
 469         struct fgraph_data *data = iter->private;
 470         struct ring_buffer_iter *ring_iter = NULL;
 471         struct ring_buffer_event *event;
 472         struct ftrace_graph_ret_entry *next;
 473
 474         /*
 475          * If the previous output failed to write to the seq buffer,
 476          * then we just reuse the data from before.
 477          */
 478         if (data && data->failed) {
 479                 curr = &data->ent;
 480                 next = &data->ret;
 481         } else {
 482
 483                 ring_iter = iter->buffer_iter[iter->cpu];
 484
 485                 /* First peek to compare current entry and the next one */
 486                 if (ring_iter)
 487                         event = ring_buffer_iter_peek(ring_iter, NULL);
 488                 else {
 489                         /*
 490                          * We need to consume the current entry to see
 491                          * the next one.
 492                          */
 493                         ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
 494                         event = ring_buffer_peek(iter->tr->buffer, iter->cpu,
 495                                                  NULL);
 496                 }
 497
 498                 if (!event)
 499                         return NULL;
 500
 501                 next = ring_buffer_event_data(event);
 502
 503                 if (data) {
 504                         /*
 505                          * Save current and next entries for later reference
 506                          * if the output fails.
 507                          */
 508                         data->ent = *curr;
 509                         data->ret = *next;
 510                 }
 511         }
 512
 513         if (next->ent.type != TRACE_GRAPH_RET)
 514                 return NULL;
 515
 516         if (curr->ent.pid != next->ent.pid ||
 517                         curr->graph_ent.func != next->ret.func)
 518                 return NULL;
 519
 520         /* this is a leaf, now advance the iterator */
 521         if (ring_iter)
 522                 ring_buffer_read(ring_iter, NULL);
 523
 524         return next;
 525 }
 526
 527 /* Signal a overhead of time execution to the output */
 528 static int
 529 print_graph_overhead(unsigned long long duration, struct trace_seq *s)
 530 {
 531         /* If duration disappear, we don't need anything */
 532         if (!(tracer_flags.val & TRACE_GRAPH_PRINT_DURATION))
 533                 return 1;
 534
 535         /* Non nested entry or return */
 536         if (duration == -1)
 537                 return trace_seq_printf(s, "  ");
 538
 539         if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) {
 540                 /* Duration exceeded 100 msecs */
 541                 if (duration > 100000ULL)
 542                         return trace_seq_printf(s, "! ");
 543
 544                 /* Duration exceeded 10 msecs */
 545                 if (duration > 10000ULL)
 546                         return trace_seq_printf(s, "+ ");
 547         }
 548
 549         return trace_seq_printf(s, "  ");
 550 }
 551
 552 static int print_graph_abs_time(u64 t, struct trace_seq *s)
 553 {
 554         unsigned long usecs_rem;
 555
 556         usecs_rem = do_div(t, NSEC_PER_SEC);
 557         usecs_rem /= 1000;
 558
 559         return trace_seq_printf(s, "%5lu.%06lu |  ",
 560                         (unsigned long)t, usecs_rem);
 561 }
 562
 563 static enum print_line_t
 564 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
 565                 enum trace_type type, int cpu, pid_t pid)
 566 {
 567         int ret;
 568         struct trace_seq *s = &iter->seq;
 569
 570         if (addr < (unsigned long)__irqentry_text_start ||
 571                 addr >= (unsigned long)__irqentry_text_end)
 572                 return TRACE_TYPE_UNHANDLED;
 573
 574         /* Absolute time */
 575         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME) {
 576                 ret = print_graph_abs_time(iter->ts, s);
 577                 if (!ret)
 578                         return TRACE_TYPE_PARTIAL_LINE;
 579         }
 580
 581         /* Cpu */
 582         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
 583                 ret = print_graph_cpu(s, cpu);
 584                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 585                         return TRACE_TYPE_PARTIAL_LINE;
 586         }
 587
 588         /* Proc */
 589         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) {
 590                 ret = print_graph_proc(s, pid);
 591                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 592                         return TRACE_TYPE_PARTIAL_LINE;
 593                 ret = trace_seq_printf(s, " | ");
 594                 if (!ret)
 595                         return TRACE_TYPE_PARTIAL_LINE;
 596         }
 597
 598         /* No overhead */
 599         ret = print_graph_overhead(-1, s);
 600         if (!ret)
 601                 return TRACE_TYPE_PARTIAL_LINE;
 602
 603         if (type == TRACE_GRAPH_ENT)
 604                 ret = trace_seq_printf(s, "==========>");
 605         else
 606                 ret = trace_seq_printf(s, "<==========");
 607
 608         if (!ret)
 609                 return TRACE_TYPE_PARTIAL_LINE;
 610
 611         /* Don't close the duration column if haven't one */
 612         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
 613                 trace_seq_printf(s, " |");
 614         ret = trace_seq_printf(s, "\n");
 615
 616         if (!ret)
 617                 return TRACE_TYPE_PARTIAL_LINE;
 618         return TRACE_TYPE_HANDLED;
 619 }
 620
 621 enum print_line_t
 622 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
 623 {
 624         unsigned long nsecs_rem = do_div(duration, 1000);
 625         /* log10(ULONG_MAX) + '\0' */
 626         char msecs_str[21];
 627         char nsecs_str[5];
 628         int ret, len;
 629         int i;
 630
 631         sprintf(msecs_str, "%lu", (unsigned long) duration);
 632
 633         /* Print msecs */
 634         ret = trace_seq_printf(s, "%s", msecs_str);
 635         if (!ret)
 636                 return TRACE_TYPE_PARTIAL_LINE;
 637
 638         len = strlen(msecs_str);
 639
 640         /* Print nsecs (we don't want to exceed 7 numbers) */
 641         if (len < 7) {
 642                 snprintf(nsecs_str, 8 - len, "%03lu", nsecs_rem);
 643                 ret = trace_seq_printf(s, ".%s", nsecs_str);
 644                 if (!ret)
 645                         return TRACE_TYPE_PARTIAL_LINE;
 646                 len += strlen(nsecs_str);
 647         }
 648
 649         ret = trace_seq_printf(s, " us ");
 650         if (!ret)
 651                 return TRACE_TYPE_PARTIAL_LINE;
 652
 653         /* Print remaining spaces to fit the row's width */
 654         for (i = len; i < 7; i++) {
 655                 ret = trace_seq_printf(s, " ");
 656                 if (!ret)
 657                         return TRACE_TYPE_PARTIAL_LINE;
 658         }
 659         return TRACE_TYPE_HANDLED;
 660 }
 661
 662 static enum print_line_t
 663 print_graph_duration(unsigned long long duration, struct trace_seq *s)
 664 {
 665         int ret;
 666
 667         ret = trace_print_graph_duration(duration, s);
 668         if (ret != TRACE_TYPE_HANDLED)
 669                 return ret;
 670
 671         ret = trace_seq_printf(s, "|  ");
 672         if (!ret)
 673                 return TRACE_TYPE_PARTIAL_LINE;
 674
 675         return TRACE_TYPE_HANDLED;
 676 }
 677
 678 /* Case of a leaf function on its call entry */
 679 static enum print_line_t
 680 print_graph_entry_leaf(struct trace_iterator *iter,
 681                 struct ftrace_graph_ent_entry *entry,
 682                 struct ftrace_graph_ret_entry *ret_entry, struct trace_seq *s)
 683 {
 684         struct fgraph_data *data = iter->private;
 685         struct ftrace_graph_ret *graph_ret;
 686         struct ftrace_graph_ent *call;
 687         unsigned long long duration;
 688         int ret;
 689         int i;
 690
 691         graph_ret = &ret_entry->ret;
 692         call = &entry->graph_ent;
 693         duration = graph_ret->rettime - graph_ret->calltime;
 694
 695         if (data) {
 696                 struct fgraph_cpu_data *cpu_data;
 697                 int cpu = iter->cpu;
 698
 699                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 700
 701                 /*
 702                  * Comments display at + 1 to depth. Since
 703                  * this is a leaf function, keep the comments
 704                  * equal to this depth.
 705                  */
 706                 cpu_data->depth = call->depth - 1;
 707
 708                 /* No need to keep this function around for this depth */
 709                 if (call->depth < FTRACE_RETFUNC_DEPTH)
 710                         cpu_data->enter_funcs[call->depth] = 0;
 711         }
 712
 713         /* Overhead */
 714         ret = print_graph_overhead(duration, s);
 715         if (!ret)
 716                 return TRACE_TYPE_PARTIAL_LINE;
 717
 718         /* Duration */
 719         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 720                 ret = print_graph_duration(duration, s);
 721                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 722                         return TRACE_TYPE_PARTIAL_LINE;
 723         }
 724
 725         /* Function */
 726         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 727                 ret = trace_seq_printf(s, " ");
 728                 if (!ret)
 729                         return TRACE_TYPE_PARTIAL_LINE;
 730         }
 731
 732         ret = trace_seq_printf(s, "%ps();\n", (void *)call->func);
 733         if (!ret)
 734                 return TRACE_TYPE_PARTIAL_LINE;
 735
 736         return TRACE_TYPE_HANDLED;
 737 }
 738
 739 static enum print_line_t
 740 print_graph_entry_nested(struct trace_iterator *iter,
 741                          struct ftrace_graph_ent_entry *entry,
 742                          struct trace_seq *s, int cpu)
 743 {
 744         struct ftrace_graph_ent *call = &entry->graph_ent;
 745         struct fgraph_data *data = iter->private;
 746         int ret;
 747         int i;
 748
 749         if (data) {
 750                 struct fgraph_cpu_data *cpu_data;
 751                 int cpu = iter->cpu;
 752
 753                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 754                 cpu_data->depth = call->depth;
 755
 756                 /* Save this function pointer to see if the exit matches */
 757                 if (call->depth < FTRACE_RETFUNC_DEPTH)
 758                         cpu_data->enter_funcs[call->depth] = call->func;
 759         }
 760
 761         /* No overhead */
 762         ret = print_graph_overhead(-1, s);
 763         if (!ret)
 764                 return TRACE_TYPE_PARTIAL_LINE;
 765
 766         /* No time */
 767         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 768                 ret = trace_seq_printf(s, "            |  ");
 769                 if (!ret)
 770                         return TRACE_TYPE_PARTIAL_LINE;
 771         }
 772
 773         /* Function */
 774         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 775                 ret = trace_seq_printf(s, " ");
 776                 if (!ret)
 777                         return TRACE_TYPE_PARTIAL_LINE;
 778         }
 779
 780         ret = trace_seq_printf(s, "%ps() {\n", (void *)call->func);
 781         if (!ret)
 782                 return TRACE_TYPE_PARTIAL_LINE;
 783
 784         /*
 785          * we already consumed the current entry to check the next one
 786          * and see if this is a leaf.
 787          */
 788         return TRACE_TYPE_NO_CONSUME;
 789 }
 790
 791 static enum print_line_t
 792 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
 793                      int type, unsigned long addr)
 794 {
 795         struct fgraph_data *data = iter->private;
 796         struct trace_entry *ent = iter->ent;
 797         int cpu = iter->cpu;
 798         int ret;
 799
 800         /* Pid */
 801         if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
 802                 return TRACE_TYPE_PARTIAL_LINE;
 803
 804         if (type) {
 805                 /* Interrupt */
 806                 ret = print_graph_irq(iter, addr, type, cpu, ent->pid);
 807                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 808                         return TRACE_TYPE_PARTIAL_LINE;
 809         }
 810
 811         /* Absolute time */
 812         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME) {
 813                 ret = print_graph_abs_time(iter->ts, s);
 814                 if (!ret)
 815                         return TRACE_TYPE_PARTIAL_LINE;
 816         }
 817
 818         /* Cpu */
 819         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
 820                 ret = print_graph_cpu(s, cpu);
 821                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 822                         return TRACE_TYPE_PARTIAL_LINE;
 823         }
 824
 825         /* Proc */
 826         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) {
 827                 ret = print_graph_proc(s, ent->pid);
 828                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 829                         return TRACE_TYPE_PARTIAL_LINE;
 830
 831                 ret = trace_seq_printf(s, " | ");
 832                 if (!ret)
 833                         return TRACE_TYPE_PARTIAL_LINE;
 834         }
 835
 836         /* Latency format */
 837         if (trace_flags & TRACE_ITER_LATENCY_FMT) {
 838                 ret = print_graph_lat_fmt(s, ent);
 839                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 840                         return TRACE_TYPE_PARTIAL_LINE;
 841         }
 842
 843         return 0;
 844 }
 845
 846 static enum print_line_t
 847 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
 848                         struct trace_iterator *iter)
 849 {
 850         struct fgraph_data *data = iter->private;
 851         struct ftrace_graph_ent *call = &field->graph_ent;
 852         struct ftrace_graph_ret_entry *leaf_ret;
 853         static enum print_line_t ret;
 854         int cpu = iter->cpu;
 855
 856         if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func))
 857                 return TRACE_TYPE_PARTIAL_LINE;
 858
 859         leaf_ret = get_return_for_leaf(iter, field);
 860         if (leaf_ret)
 861                 ret = print_graph_entry_leaf(iter, field, leaf_ret, s);
 862         else
 863                 ret = print_graph_entry_nested(iter, field, s, cpu);
 864
 865         if (data) {
 866                 /*
 867                  * If we failed to write our output, then we need to make
 868                  * note of it. Because we already consumed our entry.
 869                  */
 870                 if (s->full) {
 871                         data->failed = 1;
 872                         data->cpu = cpu;
 873                 } else
 874                         data->failed = 0;
 875         }
 876
 877         return ret;
 878 }
 879
 880 static enum print_line_t
 881 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
 882                    struct trace_entry *ent, struct trace_iterator *iter)
 883 {
 884         unsigned long long duration = trace->rettime - trace->calltime;
 885         struct fgraph_data *data = iter->private;
 886         pid_t pid = ent->pid;
 887         int cpu = iter->cpu;
 888         int func_match = 1;
 889         int ret;
 890         int i;
 891
 892         if (data) {
 893                 struct fgraph_cpu_data *cpu_data;
 894                 int cpu = iter->cpu;
 895
 896                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 897
 898                 /*
 899                  * Comments display at + 1 to depth. This is the
 900                  * return from a function, we now want the comments
 901                  * to display at the same level of the bracket.
 902                  */
 903                 cpu_data->depth = trace->depth - 1;
 904
 905                 if (trace->depth < FTRACE_RETFUNC_DEPTH) {
 906                         if (cpu_data->enter_funcs[trace->depth] != trace->func)
 907                                 func_match = 0;
 908                         cpu_data->enter_funcs[trace->depth] = 0;
 909                 }
 910         }
 911
 912         if (print_graph_prologue(iter, s, 0, 0))
 913                 return TRACE_TYPE_PARTIAL_LINE;
 914
 915         /* Overhead */
 916         ret = print_graph_overhead(duration, s);
 917         if (!ret)
 918                 return TRACE_TYPE_PARTIAL_LINE;
 919
 920         /* Duration */
 921         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 922                 ret = print_graph_duration(duration, s);
 923                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 924                         return TRACE_TYPE_PARTIAL_LINE;
 925         }
 926
 927         /* Closing brace */
 928         for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
 929                 ret = trace_seq_printf(s, " ");
 930                 if (!ret)
 931                         return TRACE_TYPE_PARTIAL_LINE;
 932         }
 933
 934         /*
 935          * If the return function does not have a matching entry,
 936          * then the entry was lost. Instead of just printing
 937          * the '}' and letting the user guess what function this
 938          * belongs to, write out the function name.
 939          */
 940         if (func_match) {
 941                 ret = trace_seq_printf(s, "}\n");
 942                 if (!ret)
 943                         return TRACE_TYPE_PARTIAL_LINE;
 944         } else {
 945                 ret = trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
 946                 if (!ret)
 947                         return TRACE_TYPE_PARTIAL_LINE;
 948         }
 949
 950         /* Overrun */
 951         if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERRUN) {
 952                 ret = trace_seq_printf(s, " (Overruns: %lu)\n",
 953                                         trace->overrun);
 954                 if (!ret)
 955                         return TRACE_TYPE_PARTIAL_LINE;
 956         }
 957
 958         ret = print_graph_irq(iter, trace->func, TRACE_GRAPH_RET, cpu, pid);
 959         if (ret == TRACE_TYPE_PARTIAL_LINE)
 960                 return TRACE_TYPE_PARTIAL_LINE;
 961
 962         return TRACE_TYPE_HANDLED;
 963 }
 964
 965 static enum print_line_t
 966 print_graph_comment(struct trace_seq *s,  struct trace_entry *ent,
 967                     struct trace_iterator *iter)
 968 {
 969         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 970         struct fgraph_data *data = iter->private;
 971         struct trace_event *event;
 972         int depth = 0;
 973         int ret;
 974         int i;
 975
 976         if (data)
 977                 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
 978
 979         if (print_graph_prologue(iter, s, 0, 0))
 980                 return TRACE_TYPE_PARTIAL_LINE;
 981
 982         /* No overhead */
 983         ret = print_graph_overhead(-1, s);
 984         if (!ret)
 985                 return TRACE_TYPE_PARTIAL_LINE;
 986
 987         /* No time */
 988         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
 989                 ret = trace_seq_printf(s, "            |  ");
 990                 if (!ret)
 991                         return TRACE_TYPE_PARTIAL_LINE;
 992         }
 993
 994         /* Indentation */
 995         if (depth > 0)
 996                 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++) {
 997                         ret = trace_seq_printf(s, " ");
 998                         if (!ret)
 999                                 return TRACE_TYPE_PARTIAL_LINE;
1000                 }
1001
1002         /* The comment */
1003         ret = trace_seq_printf(s, "/* ");
1004         if (!ret)
1005                 return TRACE_TYPE_PARTIAL_LINE;
1006
1007         switch (iter->ent->type) {
1008         case TRACE_BPRINT:
1009                 ret = trace_print_bprintk_msg_only(iter);
1010                 if (ret != TRACE_TYPE_HANDLED)
1011                         return ret;
1012                 break;
1013         case TRACE_PRINT:
1014                 ret = trace_print_printk_msg_only(iter);
1015                 if (ret != TRACE_TYPE_HANDLED)
1016                         return ret;
1017                 break;
1018         default:
1019                 event = ftrace_find_event(ent->type);
1020                 if (!event)
1021                         return TRACE_TYPE_UNHANDLED;
1022
1023                 ret = event->trace(iter, sym_flags);
1024                 if (ret != TRACE_TYPE_HANDLED)
1025                         return ret;
1026         }
1027
1028         /* Strip ending newline */
1029         if (s->buffer[s->len - 1] == '\n') {
1030                 s->buffer[s->len - 1] = '\0';
1031                 s->len--;
1032         }
1033
1034         ret = trace_seq_printf(s, " */\n");
1035         if (!ret)
1036                 return TRACE_TYPE_PARTIAL_LINE;
1037
1038         return TRACE_TYPE_HANDLED;
1039 }
1040
1041
1042 enum print_line_t
1043 print_graph_function(struct trace_iterator *iter)
1044 {
1045         struct ftrace_graph_ent_entry *field;
1046         struct fgraph_data *data = iter->private;
1047         struct trace_entry *entry = iter->ent;
1048         struct trace_seq *s = &iter->seq;
1049         int cpu = iter->cpu;
1050         int ret;
1051
1052         if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1053                 per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1054                 return TRACE_TYPE_HANDLED;
1055         }
1056
1057         /*
1058          * If the last output failed, there's a possibility we need
1059          * to print out the missing entry which would never go out.
1060          */
1061         if (data && data->failed) {
1062                 field = &data->ent;
1063                 iter->cpu = data->cpu;
1064                 ret = print_graph_entry(field, s, iter);
1065                 if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1066                         per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1067                         ret = TRACE_TYPE_NO_CONSUME;
1068                 }
1069                 iter->cpu = cpu;
1070                 return ret;
1071         }
1072
1073         switch (entry->type) {
1074         case TRACE_GRAPH_ENT: {
1075                 /*
1076                  * print_graph_entry() may consume the current event,
1077                  * thus @field may become invalid, so we need to save it.
1078                  * sizeof(struct ftrace_graph_ent_entry) is very small,
1079                  * it can be safely saved at the stack.
1080                  */
1081                 struct ftrace_graph_ent_entry saved;
1082                 trace_assign_type(field, entry);
1083                 saved = *field;
1084                 return print_graph_entry(&saved, s, iter);
1085         }
1086         case TRACE_GRAPH_RET: {
1087                 struct ftrace_graph_ret_entry *field;
1088                 trace_assign_type(field, entry);
1089                 return print_graph_return(&field->ret, s, entry, iter);
1090         }
1091         default:
1092                 return print_graph_comment(s, entry, iter);
1093         }
1094
1095         return TRACE_TYPE_HANDLED;
1096 }
1097
1098 static void print_lat_header(struct seq_file *s)
1099 {
1100         static const char spaces[] = "                " /* 16 spaces */
1101                 "    "                                  /* 4 spaces */
1102                 "                 ";                    /* 17 spaces */
1103         int size = 0;
1104
1105         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
1106                 size += 16;
1107         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
1108                 size += 4;
1109         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
1110                 size += 17;
1111
1112         seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1113         seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1114         seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1115         seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1116         seq_printf(s, "#%.*s||| / _-=> lock-depth      \n", size, spaces);
1117         seq_printf(s, "#%.*s|||| /                     \n", size, spaces);
1118 }
1119
1120 static void print_graph_headers(struct seq_file *s)
1121 {
1122         int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
1123
1124         if (lat)
1125                 print_lat_header(s);
1126
1127         /* 1st line */
1128         seq_printf(s, "#");
1129         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
1130                 seq_printf(s, "     TIME       ");
1131         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
1132                 seq_printf(s, " CPU");
1133         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
1134                 seq_printf(s, "  TASK/PID       ");
1135         if (lat)
1136                 seq_printf(s, "|||||");
1137         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
1138                 seq_printf(s, "  DURATION   ");
1139         seq_printf(s, "               FUNCTION CALLS\n");
1140
1141         /* 2nd line */
1142         seq_printf(s, "#");
1143         if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
1144                 seq_printf(s, "      |         ");
1145         if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
1146                 seq_printf(s, " |  ");
1147         if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
1148                 seq_printf(s, "   |    |        ");
1149         if (lat)
1150                 seq_printf(s, "|||||");
1151         if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
1152                 seq_printf(s, "   |   |      ");
1153         seq_printf(s, "               |   |   |   |\n");
1154 }
1155
1156 static void graph_trace_open(struct trace_iterator *iter)
1157 {
1158         /* pid and depth on the last trace processed */
1159         struct fgraph_data *data;
1160         int cpu;
1161
1162         iter->private = NULL;
1163
1164         data = kzalloc(sizeof(*data), GFP_KERNEL);
1165         if (!data)
1166                 goto out_err;
1167
1168         data->cpu_data = alloc_percpu(struct fgraph_cpu_data);
1169         if (!data->cpu_data)
1170                 goto out_err_free;
1171
1172         for_each_possible_cpu(cpu) {
1173                 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1174                 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1175                 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1176                 *pid = -1;
1177                 *depth = 0;
1178                 *ignore = 0;
1179         }
1180
1181         iter->private = data;
1182
1183         return;
1184
1185  out_err_free:
1186         kfree(data);
1187  out_err:
1188         pr_warning("function graph tracer: not enough memory\n");
1189 }
1190
1191 static void graph_trace_close(struct trace_iterator *iter)
1192 {
1193         struct fgraph_data *data = iter->private;
1194
1195         if (data) {
1196                 free_percpu(data->cpu_data);
1197                 kfree(data);
1198         }
1199 }
1200
1201 static struct tracer graph_trace __read_mostly = {
1202         .name           = "function_graph",
1203         .open           = graph_trace_open,
1204         .pipe_open      = graph_trace_open,
1205         .close          = graph_trace_close,
1206         .pipe_close     = graph_trace_close,
1207         .wait_pipe      = poll_wait_pipe,
1208         .init           = graph_trace_init,
1209         .reset          = graph_trace_reset,
1210         .print_line     = print_graph_function,
1211         .print_header   = print_graph_headers,
1212         .flags          = &tracer_flags,
1213 #ifdef CONFIG_FTRACE_SELFTEST
1214         .selftest       = trace_selftest_startup_function_graph,
1215 #endif
1216 };
1217
1218 static __init int init_graph_trace(void)
1219 {
1220         max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1221
1222         return register_tracer(&graph_trace);
1223 }
1224
1225 device_initcall(init_graph_trace);