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