kernel/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/uaccess.h>
  10 #include <linux/ftrace.h>
  11 #include <linux/slab.h>
  12 #include <linux/fs.h>
  13
  14 #include "trace.h"
  15 #include "trace_output.h"
  16
  17 static bool kill_ftrace_graph;
  18
  19 /**
  20  * ftrace_graph_is_dead - returns true if ftrace_graph_stop() was called
  21  *
  22  * ftrace_graph_stop() is called when a severe error is detected in
  23  * the function graph tracing. This function is called by the critical
  24  * paths of function graph to keep those paths from doing any more harm.
  25  */
  26 bool ftrace_graph_is_dead(void)
  27 {
  28         return kill_ftrace_graph;
  29 }
  30
  31 /**
  32  * ftrace_graph_stop - set to permanently disable function graph tracincg
  33  *
  34  * In case of an error int function graph tracing, this is called
  35  * to try to keep function graph tracing from causing any more harm.
  36  * Usually this is pretty severe and this is called to try to at least
  37  * get a warning out to the user.
  38  */
  39 void ftrace_graph_stop(void)
  40 {
  41         kill_ftrace_graph = true;
  42 }
  43
  44 /* When set, irq functions will be ignored */
  45 static int ftrace_graph_skip_irqs;
  46
  47 struct fgraph_cpu_data {
  48         pid_t           last_pid;
  49         int             depth;
  50         int             depth_irq;
  51         int             ignore;
  52         unsigned long   enter_funcs[FTRACE_RETFUNC_DEPTH];
  53 };
  54
  55 struct fgraph_data {
  56         struct fgraph_cpu_data __percpu *cpu_data;
  57
  58         /* Place to preserve last processed entry. */
  59         struct ftrace_graph_ent_entry   ent;
  60         struct ftrace_graph_ret_entry   ret;
  61         int                             failed;
  62         int                             cpu;
  63 };
  64
  65 #define TRACE_GRAPH_INDENT      2
  66
  67 static unsigned int max_depth;
  68
  69 static struct tracer_opt trace_opts[] = {
  70         /* Display overruns? (for self-debug purpose) */
  71         { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
  72         /* Display CPU ? */
  73         { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
  74         /* Display Overhead ? */
  75         { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
  76         /* Display proc name/pid */
  77         { TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
  78         /* Display duration of execution */
  79         { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
  80         /* Display absolute time of an entry */
  81         { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
  82         /* Display interrupts */
  83         { TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
  84         /* Display function name after trailing } */
  85         { TRACER_OPT(funcgraph-tail, TRACE_GRAPH_PRINT_TAIL) },
  86         /* Include sleep time (scheduled out) between entry and return */
  87         { TRACER_OPT(sleep-time, TRACE_GRAPH_SLEEP_TIME) },
  88         /* Include time within nested functions */
  89         { TRACER_OPT(graph-time, TRACE_GRAPH_GRAPH_TIME) },
  90         { } /* Empty entry */
  91 };
  92
  93 static struct tracer_flags tracer_flags = {
  94         /* Don't display overruns, proc, or tail by default */
  95         .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
  96                TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS |
  97                TRACE_GRAPH_SLEEP_TIME | TRACE_GRAPH_GRAPH_TIME,
  98         .opts = trace_opts
  99 };
 100
 101 static struct trace_array *graph_array;
 102
 103 /*
 104  * DURATION column is being also used to display IRQ signs,
 105  * following values are used by print_graph_irq and others
 106  * to fill in space into DURATION column.
 107  */
 108 enum {
 109         FLAGS_FILL_FULL  = 1 << TRACE_GRAPH_PRINT_FILL_SHIFT,
 110         FLAGS_FILL_START = 2 << TRACE_GRAPH_PRINT_FILL_SHIFT,
 111         FLAGS_FILL_END   = 3 << TRACE_GRAPH_PRINT_FILL_SHIFT,
 112 };
 113
 114 static void
 115 print_graph_duration(struct trace_array *tr, unsigned long long duration,
 116                      struct trace_seq *s, u32 flags);
 117
 118 /* Add a function return address to the trace stack on thread info.*/
 119 int
 120 ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
 121                          unsigned long frame_pointer)
 122 {
 123         unsigned long long calltime;
 124         int index;
 125
 126         if (unlikely(ftrace_graph_is_dead()))
 127                 return -EBUSY;
 128
 129         if (!current->ret_stack)
 130                 return -EBUSY;
 131
 132         /*
 133          * We must make sure the ret_stack is tested before we read
 134          * anything else.
 135          */
 136         smp_rmb();
 137
 138         /* The return trace stack is full */
 139         if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
 140                 atomic_inc(&current->trace_overrun);
 141                 return -EBUSY;
 142         }
 143
 144         /*
 145          * The curr_ret_stack is an index to ftrace return stack of
 146          * current task.  Its value should be in [0, FTRACE_RETFUNC_
 147          * DEPTH) when the function graph tracer is used.  To support
 148          * filtering out specific functions, it makes the index
 149          * negative by subtracting huge value (FTRACE_NOTRACE_DEPTH)
 150          * so when it sees a negative index the ftrace will ignore
 151          * the record.  And the index gets recovered when returning
 152          * from the filtered function by adding the FTRACE_NOTRACE_
 153          * DEPTH and then it'll continue to record functions normally.
 154          *
 155          * The curr_ret_stack is initialized to -1 and get increased
 156          * in this function.  So it can be less than -1 only if it was
 157          * filtered out via ftrace_graph_notrace_addr() which can be
 158          * set from set_graph_notrace file in tracefs by user.
 159          */
 160         if (current->curr_ret_stack < -1)
 161                 return -EBUSY;
 162
 163         calltime = trace_clock_local();
 164
 165         index = ++current->curr_ret_stack;
 166         if (ftrace_graph_notrace_addr(func))
 167                 current->curr_ret_stack -= FTRACE_NOTRACE_DEPTH;
 168         barrier();
 169         current->ret_stack[index].ret = ret;
 170         current->ret_stack[index].func = func;
 171         current->ret_stack[index].calltime = calltime;
 172         current->ret_stack[index].subtime = 0;
 173         current->ret_stack[index].fp = frame_pointer;
 174         *depth = current->curr_ret_stack;
 175
 176         return 0;
 177 }
 178
 179 /* Retrieve a function return address to the trace stack on thread info.*/
 180 static void
 181 ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
 182                         unsigned long frame_pointer)
 183 {
 184         int index;
 185
 186         index = current->curr_ret_stack;
 187
 188         /*
 189          * A negative index here means that it's just returned from a
 190          * notrace'd function.  Recover index to get an original
 191          * return address.  See ftrace_push_return_trace().
 192          *
 193          * TODO: Need to check whether the stack gets corrupted.
 194          */
 195         if (index < 0)
 196                 index += FTRACE_NOTRACE_DEPTH;
 197
 198         if (unlikely(index < 0 || index >= FTRACE_RETFUNC_DEPTH)) {
 199                 ftrace_graph_stop();
 200                 WARN_ON(1);
 201                 /* Might as well panic, otherwise we have no where to go */
 202                 *ret = (unsigned long)panic;
 203                 return;
 204         }
 205
 206 #if defined(CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST) && !defined(CC_USING_FENTRY)
 207         /*
 208          * The arch may choose to record the frame pointer used
 209          * and check it here to make sure that it is what we expect it
 210          * to be. If gcc does not set the place holder of the return
 211          * address in the frame pointer, and does a copy instead, then
 212          * the function graph trace will fail. This test detects this
 213          * case.
 214          *
 215          * Currently, x86_32 with optimize for size (-Os) makes the latest
 216          * gcc do the above.
 217          *
 218          * Note, -mfentry does not use frame pointers, and this test
 219          *  is not needed if CC_USING_FENTRY is set.
 220          */
 221         if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
 222                 ftrace_graph_stop();
 223                 WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
 224                      "  from func %ps return to %lx\n",
 225                      current->ret_stack[index].fp,
 226                      frame_pointer,
 227                      (void *)current->ret_stack[index].func,
 228                      current->ret_stack[index].ret);
 229                 *ret = (unsigned long)panic;
 230                 return;
 231         }
 232 #endif
 233
 234         *ret = current->ret_stack[index].ret;
 235         trace->func = current->ret_stack[index].func;
 236         trace->calltime = current->ret_stack[index].calltime;
 237         trace->overrun = atomic_read(&current->trace_overrun);
 238         trace->depth = index;
 239 }
 240
 241 /*
 242  * Send the trace to the ring-buffer.
 243  * @return the original return address.
 244  */
 245 unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
 246 {
 247         struct ftrace_graph_ret trace;
 248         unsigned long ret;
 249
 250         ftrace_pop_return_trace(&trace, &ret, frame_pointer);
 251         trace.rettime = trace_clock_local();
 252         barrier();
 253         current->curr_ret_stack--;
 254         /*
 255          * The curr_ret_stack can be less than -1 only if it was
 256          * filtered out and it's about to return from the function.
 257          * Recover the index and continue to trace normal functions.
 258          */
 259         if (current->curr_ret_stack < -1) {
 260                 current->curr_ret_stack += FTRACE_NOTRACE_DEPTH;
 261                 return ret;
 262         }
 263
 264         /*
 265          * The trace should run after decrementing the ret counter
 266          * in case an interrupt were to come in. We don't want to
 267          * lose the interrupt if max_depth is set.
 268          */
 269         ftrace_graph_return(&trace);
 270
 271         if (unlikely(!ret)) {
 272                 ftrace_graph_stop();
 273                 WARN_ON(1);
 274                 /* Might as well panic. What else to do? */
 275                 ret = (unsigned long)panic;
 276         }
 277
 278         return ret;
 279 }
 280
 281 int __trace_graph_entry(struct trace_array *tr,
 282                                 struct ftrace_graph_ent *trace,
 283                                 unsigned long flags,
 284                                 int pc)
 285 {
 286         struct trace_event_call *call = &event_funcgraph_entry;
 287         struct ring_buffer_event *event;
 288         struct ring_buffer *buffer = tr->trace_buffer.buffer;
 289         struct ftrace_graph_ent_entry *entry;
 290
 291         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
 292                                           sizeof(*entry), flags, pc);
 293         if (!event)
 294                 return 0;
 295         entry   = ring_buffer_event_data(event);
 296         entry->graph_ent                        = *trace;
 297         if (!call_filter_check_discard(call, entry, buffer, event))
 298                 __buffer_unlock_commit(buffer, event);
 299
 300         return 1;
 301 }
 302
 303 static inline int ftrace_graph_ignore_irqs(void)
 304 {
 305         if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
 306                 return 0;
 307
 308         return in_irq();
 309 }
 310
 311 int trace_graph_entry(struct ftrace_graph_ent *trace)
 312 {
 313         struct trace_array *tr = graph_array;
 314         struct trace_array_cpu *data;
 315         unsigned long flags;
 316         long disabled;
 317         int ret;
 318         int cpu;
 319         int pc;
 320
 321         if (!ftrace_trace_task(current))
 322                 return 0;
 323
 324         /* trace it when it is-nested-in or is a function enabled. */
 325         if ((!(trace->depth || ftrace_graph_addr(trace->func)) ||
 326              ftrace_graph_ignore_irqs()) || (trace->depth < 0) ||
 327             (max_depth && trace->depth >= max_depth))
 328                 return 0;
 329
 330         /*
 331          * Do not trace a function if it's filtered by set_graph_notrace.
 332          * Make the index of ret stack negative to indicate that it should
 333          * ignore further functions.  But it needs its own ret stack entry
 334          * to recover the original index in order to continue tracing after
 335          * returning from the function.
 336          */
 337         if (ftrace_graph_notrace_addr(trace->func))
 338                 return 1;
 339
 340         local_irq_save(flags);
 341         cpu = raw_smp_processor_id();
 342         data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 343         disabled = atomic_inc_return(&data->disabled);
 344         if (likely(disabled == 1)) {
 345                 pc = preempt_count();
 346                 ret = __trace_graph_entry(tr, trace, flags, pc);
 347         } else {
 348                 ret = 0;
 349         }
 350
 351         atomic_dec(&data->disabled);
 352         local_irq_restore(flags);
 353
 354         return ret;
 355 }
 356
 357 static int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
 358 {
 359         if (tracing_thresh)
 360                 return 1;
 361         else
 362                 return trace_graph_entry(trace);
 363 }
 364
 365 static void
 366 __trace_graph_function(struct trace_array *tr,
 367                 unsigned long ip, unsigned long flags, int pc)
 368 {
 369         u64 time = trace_clock_local();
 370         struct ftrace_graph_ent ent = {
 371                 .func  = ip,
 372                 .depth = 0,
 373         };
 374         struct ftrace_graph_ret ret = {
 375                 .func     = ip,
 376                 .depth    = 0,
 377                 .calltime = time,
 378                 .rettime  = time,
 379         };
 380
 381         __trace_graph_entry(tr, &ent, flags, pc);
 382         __trace_graph_return(tr, &ret, flags, pc);
 383 }
 384
 385 void
 386 trace_graph_function(struct trace_array *tr,
 387                 unsigned long ip, unsigned long parent_ip,
 388                 unsigned long flags, int pc)
 389 {
 390         __trace_graph_function(tr, ip, flags, pc);
 391 }
 392
 393 void __trace_graph_return(struct trace_array *tr,
 394                                 struct ftrace_graph_ret *trace,
 395                                 unsigned long flags,
 396                                 int pc)
 397 {
 398         struct trace_event_call *call = &event_funcgraph_exit;
 399         struct ring_buffer_event *event;
 400         struct ring_buffer *buffer = tr->trace_buffer.buffer;
 401         struct ftrace_graph_ret_entry *entry;
 402
 403         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
 404                                           sizeof(*entry), flags, pc);
 405         if (!event)
 406                 return;
 407         entry   = ring_buffer_event_data(event);
 408         entry->ret                              = *trace;
 409         if (!call_filter_check_discard(call, entry, buffer, event))
 410                 __buffer_unlock_commit(buffer, event);
 411 }
 412
 413 void trace_graph_return(struct ftrace_graph_ret *trace)
 414 {
 415         struct trace_array *tr = graph_array;
 416         struct trace_array_cpu *data;
 417         unsigned long flags;
 418         long disabled;
 419         int cpu;
 420         int pc;
 421
 422         local_irq_save(flags);
 423         cpu = raw_smp_processor_id();
 424         data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 425         disabled = atomic_inc_return(&data->disabled);
 426         if (likely(disabled == 1)) {
 427                 pc = preempt_count();
 428                 __trace_graph_return(tr, trace, flags, pc);
 429         }
 430         atomic_dec(&data->disabled);
 431         local_irq_restore(flags);
 432 }
 433
 434 void set_graph_array(struct trace_array *tr)
 435 {
 436         graph_array = tr;
 437
 438         /* Make graph_array visible before we start tracing */
 439
 440         smp_mb();
 441 }
 442
 443 static void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
 444 {
 445         if (tracing_thresh &&
 446             (trace->rettime - trace->calltime < tracing_thresh))
 447                 return;
 448         else
 449                 trace_graph_return(trace);
 450 }
 451
 452 static int graph_trace_init(struct trace_array *tr)
 453 {
 454         int ret;
 455
 456         set_graph_array(tr);
 457         if (tracing_thresh)
 458                 ret = register_ftrace_graph(&trace_graph_thresh_return,
 459                                             &trace_graph_thresh_entry);
 460         else
 461                 ret = register_ftrace_graph(&trace_graph_return,
 462                                             &trace_graph_entry);
 463         if (ret)
 464                 return ret;
 465         tracing_start_cmdline_record();
 466
 467         return 0;
 468 }
 469
 470 static void graph_trace_reset(struct trace_array *tr)
 471 {
 472         tracing_stop_cmdline_record();
 473         unregister_ftrace_graph();
 474 }
 475
 476 static int graph_trace_update_thresh(struct trace_array *tr)
 477 {
 478         graph_trace_reset(tr);
 479         return graph_trace_init(tr);
 480 }
 481
 482 static int max_bytes_for_cpu;
 483
 484 static void print_graph_cpu(struct trace_seq *s, int cpu)
 485 {
 486         /*
 487          * Start with a space character - to make it stand out
 488          * to the right a bit when trace output is pasted into
 489          * email:
 490          */
 491         trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
 492 }
 493
 494 #define TRACE_GRAPH_PROCINFO_LENGTH     14
 495
 496 static void print_graph_proc(struct trace_seq *s, pid_t pid)
 497 {
 498         char comm[TASK_COMM_LEN];
 499         /* sign + log10(MAX_INT) + '\0' */
 500         char pid_str[11];
 501         int spaces = 0;
 502         int len;
 503         int i;
 504
 505         trace_find_cmdline(pid, comm);
 506         comm[7] = '\0';
 507         sprintf(pid_str, "%d", pid);
 508
 509         /* 1 stands for the "-" character */
 510         len = strlen(comm) + strlen(pid_str) + 1;
 511
 512         if (len < TRACE_GRAPH_PROCINFO_LENGTH)
 513                 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
 514
 515         /* First spaces to align center */
 516         for (i = 0; i < spaces / 2; i++)
 517                 trace_seq_putc(s, ' ');
 518
 519         trace_seq_printf(s, "%s-%s", comm, pid_str);
 520
 521         /* Last spaces to align center */
 522         for (i = 0; i < spaces - (spaces / 2); i++)
 523                 trace_seq_putc(s, ' ');
 524 }
 525
 526
 527 static void print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
 528 {
 529         trace_seq_putc(s, ' ');
 530         trace_print_lat_fmt(s, entry);
 531 }
 532
 533 /* If the pid changed since the last trace, output this event */
 534 static void
 535 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
 536 {
 537         pid_t prev_pid;
 538         pid_t *last_pid;
 539
 540         if (!data)
 541                 return;
 542
 543         last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
 544
 545         if (*last_pid == pid)
 546                 return;
 547
 548         prev_pid = *last_pid;
 549         *last_pid = pid;
 550
 551         if (prev_pid == -1)
 552                 return;
 553 /*
 554  * Context-switch trace line:
 555
 556  ------------------------------------------
 557  | 1)  migration/0--1  =>  sshd-1755
 558  ------------------------------------------
 559
 560  */
 561         trace_seq_puts(s, " ------------------------------------------\n");
 562         print_graph_cpu(s, cpu);
 563         print_graph_proc(s, prev_pid);
 564         trace_seq_puts(s, " => ");
 565         print_graph_proc(s, pid);
 566         trace_seq_puts(s, "\n ------------------------------------------\n\n");
 567 }
 568
 569 static struct ftrace_graph_ret_entry *
 570 get_return_for_leaf(struct trace_iterator *iter,
 571                 struct ftrace_graph_ent_entry *curr)
 572 {
 573         struct fgraph_data *data = iter->private;
 574         struct ring_buffer_iter *ring_iter = NULL;
 575         struct ring_buffer_event *event;
 576         struct ftrace_graph_ret_entry *next;
 577
 578         /*
 579          * If the previous output failed to write to the seq buffer,
 580          * then we just reuse the data from before.
 581          */
 582         if (data && data->failed) {
 583                 curr = &data->ent;
 584                 next = &data->ret;
 585         } else {
 586
 587                 ring_iter = trace_buffer_iter(iter, iter->cpu);
 588
 589                 /* First peek to compare current entry and the next one */
 590                 if (ring_iter)
 591                         event = ring_buffer_iter_peek(ring_iter, NULL);
 592                 else {
 593                         /*
 594                          * We need to consume the current entry to see
 595                          * the next one.
 596                          */
 597                         ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu,
 598                                             NULL, NULL);
 599                         event = ring_buffer_peek(iter->trace_buffer->buffer, iter->cpu,
 600                                                  NULL, NULL);
 601                 }
 602
 603                 if (!event)
 604                         return NULL;
 605
 606                 next = ring_buffer_event_data(event);
 607
 608                 if (data) {
 609                         /*
 610                          * Save current and next entries for later reference
 611                          * if the output fails.
 612                          */
 613                         data->ent = *curr;
 614                         /*
 615                          * If the next event is not a return type, then
 616                          * we only care about what type it is. Otherwise we can
 617                          * safely copy the entire event.
 618                          */
 619                         if (next->ent.type == TRACE_GRAPH_RET)
 620                                 data->ret = *next;
 621                         else
 622                                 data->ret.ent.type = next->ent.type;
 623                 }
 624         }
 625
 626         if (next->ent.type != TRACE_GRAPH_RET)
 627                 return NULL;
 628
 629         if (curr->ent.pid != next->ent.pid ||
 630                         curr->graph_ent.func != next->ret.func)
 631                 return NULL;
 632
 633         /* this is a leaf, now advance the iterator */
 634         if (ring_iter)
 635                 ring_buffer_read(ring_iter, NULL);
 636
 637         return next;
 638 }
 639
 640 static void print_graph_abs_time(u64 t, struct trace_seq *s)
 641 {
 642         unsigned long usecs_rem;
 643
 644         usecs_rem = do_div(t, NSEC_PER_SEC);
 645         usecs_rem /= 1000;
 646
 647         trace_seq_printf(s, "%5lu.%06lu |  ",
 648                          (unsigned long)t, usecs_rem);
 649 }
 650
 651 static void
 652 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
 653                 enum trace_type type, int cpu, pid_t pid, u32 flags)
 654 {
 655         struct trace_array *tr = iter->tr;
 656         struct trace_seq *s = &iter->seq;
 657         struct trace_entry *ent = iter->ent;
 658
 659         if (addr < (unsigned long)__irqentry_text_start ||
 660                 addr >= (unsigned long)__irqentry_text_end)
 661                 return;
 662
 663         if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
 664                 /* Absolute time */
 665                 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
 666                         print_graph_abs_time(iter->ts, s);
 667
 668                 /* Cpu */
 669                 if (flags & TRACE_GRAPH_PRINT_CPU)
 670                         print_graph_cpu(s, cpu);
 671
 672                 /* Proc */
 673                 if (flags & TRACE_GRAPH_PRINT_PROC) {
 674                         print_graph_proc(s, pid);
 675                         trace_seq_puts(s, " | ");
 676                 }
 677
 678                 /* Latency format */
 679                 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
 680                         print_graph_lat_fmt(s, ent);
 681         }
 682
 683         /* No overhead */
 684         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_START);
 685
 686         if (type == TRACE_GRAPH_ENT)
 687                 trace_seq_puts(s, "==========>");
 688         else
 689                 trace_seq_puts(s, "<==========");
 690
 691         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_END);
 692         trace_seq_putc(s, '\n');
 693 }
 694
 695 void
 696 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
 697 {
 698         unsigned long nsecs_rem = do_div(duration, 1000);
 699         /* log10(ULONG_MAX) + '\0' */
 700         char usecs_str[21];
 701         char nsecs_str[5];
 702         int len;
 703         int i;
 704
 705         sprintf(usecs_str, "%lu", (unsigned long) duration);
 706
 707         /* Print msecs */
 708         trace_seq_printf(s, "%s", usecs_str);
 709
 710         len = strlen(usecs_str);
 711
 712         /* Print nsecs (we don't want to exceed 7 numbers) */
 713         if (len < 7) {
 714                 size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
 715
 716                 snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
 717                 trace_seq_printf(s, ".%s", nsecs_str);
 718                 len += strlen(nsecs_str) + 1;
 719         }
 720
 721         trace_seq_puts(s, " us ");
 722
 723         /* Print remaining spaces to fit the row's width */
 724         for (i = len; i < 8; i++)
 725                 trace_seq_putc(s, ' ');
 726 }
 727
 728 static void
 729 print_graph_duration(struct trace_array *tr, unsigned long long duration,
 730                      struct trace_seq *s, u32 flags)
 731 {
 732         if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
 733             !(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
 734                 return;
 735
 736         /* No real adata, just filling the column with spaces */
 737         switch (flags & TRACE_GRAPH_PRINT_FILL_MASK) {
 738         case FLAGS_FILL_FULL:
 739                 trace_seq_puts(s, "              |  ");
 740                 return;
 741         case FLAGS_FILL_START:
 742                 trace_seq_puts(s, "  ");
 743                 return;
 744         case FLAGS_FILL_END:
 745                 trace_seq_puts(s, " |");
 746                 return;
 747         }
 748
 749         /* Signal a overhead of time execution to the output */
 750         if (flags & TRACE_GRAPH_PRINT_OVERHEAD)
 751                 trace_seq_printf(s, "%c ", trace_find_mark(duration));
 752         else
 753                 trace_seq_puts(s, "  ");
 754
 755         trace_print_graph_duration(duration, s);
 756         trace_seq_puts(s, "|  ");
 757 }
 758
 759 /* Case of a leaf function on its call entry */
 760 static enum print_line_t
 761 print_graph_entry_leaf(struct trace_iterator *iter,
 762                 struct ftrace_graph_ent_entry *entry,
 763                 struct ftrace_graph_ret_entry *ret_entry,
 764                 struct trace_seq *s, u32 flags)
 765 {
 766         struct fgraph_data *data = iter->private;
 767         struct trace_array *tr = iter->tr;
 768         struct ftrace_graph_ret *graph_ret;
 769         struct ftrace_graph_ent *call;
 770         unsigned long long duration;
 771         int i;
 772
 773         graph_ret = &ret_entry->ret;
 774         call = &entry->graph_ent;
 775         duration = graph_ret->rettime - graph_ret->calltime;
 776
 777         if (data) {
 778                 struct fgraph_cpu_data *cpu_data;
 779                 int cpu = iter->cpu;
 780
 781                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 782
 783                 /* If a graph tracer ignored set_graph_notrace */
 784                 if (call->depth < -1)
 785                         call->depth += FTRACE_NOTRACE_DEPTH;
 786
 787                 /*
 788                  * Comments display at + 1 to depth. Since
 789                  * this is a leaf function, keep the comments
 790                  * equal to this depth.
 791                  */
 792                 cpu_data->depth = call->depth - 1;
 793
 794                 /* No need to keep this function around for this depth */
 795                 if (call->depth < FTRACE_RETFUNC_DEPTH &&
 796                     !WARN_ON_ONCE(call->depth < 0))
 797                         cpu_data->enter_funcs[call->depth] = 0;
 798         }
 799
 800         /* Overhead and duration */
 801         print_graph_duration(tr, duration, s, flags);
 802
 803         /* Function */
 804         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
 805                 trace_seq_putc(s, ' ');
 806
 807         trace_seq_printf(s, "%ps();\n", (void *)call->func);
 808
 809         return trace_handle_return(s);
 810 }
 811
 812 static enum print_line_t
 813 print_graph_entry_nested(struct trace_iterator *iter,
 814                          struct ftrace_graph_ent_entry *entry,
 815                          struct trace_seq *s, int cpu, u32 flags)
 816 {
 817         struct ftrace_graph_ent *call = &entry->graph_ent;
 818         struct fgraph_data *data = iter->private;
 819         struct trace_array *tr = iter->tr;
 820         int i;
 821
 822         if (data) {
 823                 struct fgraph_cpu_data *cpu_data;
 824                 int cpu = iter->cpu;
 825
 826                 /* If a graph tracer ignored set_graph_notrace */
 827                 if (call->depth < -1)
 828                         call->depth += FTRACE_NOTRACE_DEPTH;
 829
 830                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 831                 cpu_data->depth = call->depth;
 832
 833                 /* Save this function pointer to see if the exit matches */
 834                 if (call->depth < FTRACE_RETFUNC_DEPTH &&
 835                     !WARN_ON_ONCE(call->depth < 0))
 836                         cpu_data->enter_funcs[call->depth] = call->func;
 837         }
 838
 839         /* No time */
 840         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
 841
 842         /* Function */
 843         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
 844                 trace_seq_putc(s, ' ');
 845
 846         trace_seq_printf(s, "%ps() {\n", (void *)call->func);
 847
 848         if (trace_seq_has_overflowed(s))
 849                 return TRACE_TYPE_PARTIAL_LINE;
 850
 851         /*
 852          * we already consumed the current entry to check the next one
 853          * and see if this is a leaf.
 854          */
 855         return TRACE_TYPE_NO_CONSUME;
 856 }
 857
 858 static void
 859 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
 860                      int type, unsigned long addr, u32 flags)
 861 {
 862         struct fgraph_data *data = iter->private;
 863         struct trace_entry *ent = iter->ent;
 864         struct trace_array *tr = iter->tr;
 865         int cpu = iter->cpu;
 866
 867         /* Pid */
 868         verif_pid(s, ent->pid, cpu, data);
 869
 870         if (type)
 871                 /* Interrupt */
 872                 print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
 873
 874         if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
 875                 return;
 876
 877         /* Absolute time */
 878         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
 879                 print_graph_abs_time(iter->ts, s);
 880
 881         /* Cpu */
 882         if (flags & TRACE_GRAPH_PRINT_CPU)
 883                 print_graph_cpu(s, cpu);
 884
 885         /* Proc */
 886         if (flags & TRACE_GRAPH_PRINT_PROC) {
 887                 print_graph_proc(s, ent->pid);
 888                 trace_seq_puts(s, " | ");
 889         }
 890
 891         /* Latency format */
 892         if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
 893                 print_graph_lat_fmt(s, ent);
 894
 895         return;
 896 }
 897
 898 /*
 899  * Entry check for irq code
 900  *
 901  * returns 1 if
 902  *  - we are inside irq code
 903  *  - we just entered irq code
 904  *
 905  * retunns 0 if
 906  *  - funcgraph-interrupts option is set
 907  *  - we are not inside irq code
 908  */
 909 static int
 910 check_irq_entry(struct trace_iterator *iter, u32 flags,
 911                 unsigned long addr, int depth)
 912 {
 913         int cpu = iter->cpu;
 914         int *depth_irq;
 915         struct fgraph_data *data = iter->private;
 916
 917         /*
 918          * If we are either displaying irqs, or we got called as
 919          * a graph event and private data does not exist,
 920          * then we bypass the irq check.
 921          */
 922         if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
 923             (!data))
 924                 return 0;
 925
 926         depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
 927
 928         /*
 929          * We are inside the irq code
 930          */
 931         if (*depth_irq >= 0)
 932                 return 1;
 933
 934         if ((addr < (unsigned long)__irqentry_text_start) ||
 935             (addr >= (unsigned long)__irqentry_text_end))
 936                 return 0;
 937
 938         /*
 939          * We are entering irq code.
 940          */
 941         *depth_irq = depth;
 942         return 1;
 943 }
 944
 945 /*
 946  * Return check for irq code
 947  *
 948  * returns 1 if
 949  *  - we are inside irq code
 950  *  - we just left irq code
 951  *
 952  * returns 0 if
 953  *  - funcgraph-interrupts option is set
 954  *  - we are not inside irq code
 955  */
 956 static int
 957 check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
 958 {
 959         int cpu = iter->cpu;
 960         int *depth_irq;
 961         struct fgraph_data *data = iter->private;
 962
 963         /*
 964          * If we are either displaying irqs, or we got called as
 965          * a graph event and private data does not exist,
 966          * then we bypass the irq check.
 967          */
 968         if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
 969             (!data))
 970                 return 0;
 971
 972         depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
 973
 974         /*
 975          * We are not inside the irq code.
 976          */
 977         if (*depth_irq == -1)
 978                 return 0;
 979
 980         /*
 981          * We are inside the irq code, and this is returning entry.
 982          * Let's not trace it and clear the entry depth, since
 983          * we are out of irq code.
 984          *
 985          * This condition ensures that we 'leave the irq code' once
 986          * we are out of the entry depth. Thus protecting us from
 987          * the RETURN entry loss.
 988          */
 989         if (*depth_irq >= depth) {
 990                 *depth_irq = -1;
 991                 return 1;
 992         }
 993
 994         /*
 995          * We are inside the irq code, and this is not the entry.
 996          */
 997         return 1;
 998 }
 999
1000 static enum print_line_t
1001 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
1002                         struct trace_iterator *iter, u32 flags)
1003 {
1004         struct fgraph_data *data = iter->private;
1005         struct ftrace_graph_ent *call = &field->graph_ent;
1006         struct ftrace_graph_ret_entry *leaf_ret;
1007         static enum print_line_t ret;
1008         int cpu = iter->cpu;
1009
1010         if (check_irq_entry(iter, flags, call->func, call->depth))
1011                 return TRACE_TYPE_HANDLED;
1012
1013         print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags);
1014
1015         leaf_ret = get_return_for_leaf(iter, field);
1016         if (leaf_ret)
1017                 ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
1018         else
1019                 ret = print_graph_entry_nested(iter, field, s, cpu, flags);
1020
1021         if (data) {
1022                 /*
1023                  * If we failed to write our output, then we need to make
1024                  * note of it. Because we already consumed our entry.
1025                  */
1026                 if (s->full) {
1027                         data->failed = 1;
1028                         data->cpu = cpu;
1029                 } else
1030                         data->failed = 0;
1031         }
1032
1033         return ret;
1034 }
1035
1036 static enum print_line_t
1037 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
1038                    struct trace_entry *ent, struct trace_iterator *iter,
1039                    u32 flags)
1040 {
1041         unsigned long long duration = trace->rettime - trace->calltime;
1042         struct fgraph_data *data = iter->private;
1043         struct trace_array *tr = iter->tr;
1044         pid_t pid = ent->pid;
1045         int cpu = iter->cpu;
1046         int func_match = 1;
1047         int i;
1048
1049         if (check_irq_return(iter, flags, trace->depth))
1050                 return TRACE_TYPE_HANDLED;
1051
1052         if (data) {
1053                 struct fgraph_cpu_data *cpu_data;
1054                 int cpu = iter->cpu;
1055
1056                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
1057
1058                 /*
1059                  * Comments display at + 1 to depth. This is the
1060                  * return from a function, we now want the comments
1061                  * to display at the same level of the bracket.
1062                  */
1063                 cpu_data->depth = trace->depth - 1;
1064
1065                 if (trace->depth < FTRACE_RETFUNC_DEPTH &&
1066                     !WARN_ON_ONCE(trace->depth < 0)) {
1067                         if (cpu_data->enter_funcs[trace->depth] != trace->func)
1068                                 func_match = 0;
1069                         cpu_data->enter_funcs[trace->depth] = 0;
1070                 }
1071         }
1072
1073         print_graph_prologue(iter, s, 0, 0, flags);
1074
1075         /* Overhead and duration */
1076         print_graph_duration(tr, duration, s, flags);
1077
1078         /* Closing brace */
1079         for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++)
1080                 trace_seq_putc(s, ' ');
1081
1082         /*
1083          * If the return function does not have a matching entry,
1084          * then the entry was lost. Instead of just printing
1085          * the '}' and letting the user guess what function this
1086          * belongs to, write out the function name. Always do
1087          * that if the funcgraph-tail option is enabled.
1088          */
1089         if (func_match && !(flags & TRACE_GRAPH_PRINT_TAIL))
1090                 trace_seq_puts(s, "}\n");
1091         else
1092                 trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
1093
1094         /* Overrun */
1095         if (flags & TRACE_GRAPH_PRINT_OVERRUN)
1096                 trace_seq_printf(s, " (Overruns: %lu)\n",
1097                                  trace->overrun);
1098
1099         print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1100                         cpu, pid, flags);
1101
1102         return trace_handle_return(s);
1103 }
1104
1105 static enum print_line_t
1106 print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1107                     struct trace_iterator *iter, u32 flags)
1108 {
1109         struct trace_array *tr = iter->tr;
1110         unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
1111         struct fgraph_data *data = iter->private;
1112         struct trace_event *event;
1113         int depth = 0;
1114         int ret;
1115         int i;
1116
1117         if (data)
1118                 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1119
1120         print_graph_prologue(iter, s, 0, 0, flags);
1121
1122         /* No time */
1123         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
1124
1125         /* Indentation */
1126         if (depth > 0)
1127                 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++)
1128                         trace_seq_putc(s, ' ');
1129
1130         /* The comment */
1131         trace_seq_puts(s, "/* ");
1132
1133         switch (iter->ent->type) {
1134         case TRACE_BPRINT:
1135                 ret = trace_print_bprintk_msg_only(iter);
1136                 if (ret != TRACE_TYPE_HANDLED)
1137                         return ret;
1138                 break;
1139         case TRACE_PRINT:
1140                 ret = trace_print_printk_msg_only(iter);
1141                 if (ret != TRACE_TYPE_HANDLED)
1142                         return ret;
1143                 break;
1144         default:
1145                 event = ftrace_find_event(ent->type);
1146                 if (!event)
1147                         return TRACE_TYPE_UNHANDLED;
1148
1149                 ret = event->funcs->trace(iter, sym_flags, event);
1150                 if (ret != TRACE_TYPE_HANDLED)
1151                         return ret;
1152         }
1153
1154         if (trace_seq_has_overflowed(s))
1155                 goto out;
1156
1157         /* Strip ending newline */
1158         if (s->buffer[s->seq.len - 1] == '\n') {
1159                 s->buffer[s->seq.len - 1] = '\0';
1160                 s->seq.len--;
1161         }
1162
1163         trace_seq_puts(s, " */\n");
1164  out:
1165         return trace_handle_return(s);
1166 }
1167
1168
1169 enum print_line_t
1170 print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1171 {
1172         struct ftrace_graph_ent_entry *field;
1173         struct fgraph_data *data = iter->private;
1174         struct trace_entry *entry = iter->ent;
1175         struct trace_seq *s = &iter->seq;
1176         int cpu = iter->cpu;
1177         int ret;
1178
1179         if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1180                 per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1181                 return TRACE_TYPE_HANDLED;
1182         }
1183
1184         /*
1185          * If the last output failed, there's a possibility we need
1186          * to print out the missing entry which would never go out.
1187          */
1188         if (data && data->failed) {
1189                 field = &data->ent;
1190                 iter->cpu = data->cpu;
1191                 ret = print_graph_entry(field, s, iter, flags);
1192                 if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1193                         per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1194                         ret = TRACE_TYPE_NO_CONSUME;
1195                 }
1196                 iter->cpu = cpu;
1197                 return ret;
1198         }
1199
1200         switch (entry->type) {
1201         case TRACE_GRAPH_ENT: {
1202                 /*
1203                  * print_graph_entry() may consume the current event,
1204                  * thus @field may become invalid, so we need to save it.
1205                  * sizeof(struct ftrace_graph_ent_entry) is very small,
1206                  * it can be safely saved at the stack.
1207                  */
1208                 struct ftrace_graph_ent_entry saved;
1209                 trace_assign_type(field, entry);
1210                 saved = *field;
1211                 return print_graph_entry(&saved, s, iter, flags);
1212         }
1213         case TRACE_GRAPH_RET: {
1214                 struct ftrace_graph_ret_entry *field;
1215                 trace_assign_type(field, entry);
1216                 return print_graph_return(&field->ret, s, entry, iter, flags);
1217         }
1218         case TRACE_STACK:
1219         case TRACE_FN:
1220                 /* dont trace stack and functions as comments */
1221                 return TRACE_TYPE_UNHANDLED;
1222
1223         default:
1224                 return print_graph_comment(s, entry, iter, flags);
1225         }
1226
1227         return TRACE_TYPE_HANDLED;
1228 }
1229
1230 static enum print_line_t
1231 print_graph_function(struct trace_iterator *iter)
1232 {
1233         return print_graph_function_flags(iter, tracer_flags.val);
1234 }
1235
1236 static enum print_line_t
1237 print_graph_function_event(struct trace_iterator *iter, int flags,
1238                            struct trace_event *event)
1239 {
1240         return print_graph_function(iter);
1241 }
1242
1243 static void print_lat_header(struct seq_file *s, u32 flags)
1244 {
1245         static const char spaces[] = "                " /* 16 spaces */
1246                 "    "                                  /* 4 spaces */
1247                 "                 ";                    /* 17 spaces */
1248         int size = 0;
1249
1250         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1251                 size += 16;
1252         if (flags & TRACE_GRAPH_PRINT_CPU)
1253                 size += 4;
1254         if (flags & TRACE_GRAPH_PRINT_PROC)
1255                 size += 17;
1256
1257         seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1258         seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1259         seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1260         seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1261         seq_printf(s, "#%.*s||| /                      \n", size, spaces);
1262 }
1263
1264 static void __print_graph_headers_flags(struct trace_array *tr,
1265                                         struct seq_file *s, u32 flags)
1266 {
1267         int lat = tr->trace_flags & TRACE_ITER_LATENCY_FMT;
1268
1269         if (lat)
1270                 print_lat_header(s, flags);
1271
1272         /* 1st line */
1273         seq_putc(s, '#');
1274         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1275                 seq_puts(s, "     TIME       ");
1276         if (flags & TRACE_GRAPH_PRINT_CPU)
1277                 seq_puts(s, " CPU");
1278         if (flags & TRACE_GRAPH_PRINT_PROC)
1279                 seq_puts(s, "  TASK/PID       ");
1280         if (lat)
1281                 seq_puts(s, "||||");
1282         if (flags & TRACE_GRAPH_PRINT_DURATION)
1283                 seq_puts(s, "  DURATION   ");
1284         seq_puts(s, "               FUNCTION CALLS\n");
1285
1286         /* 2nd line */
1287         seq_putc(s, '#');
1288         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1289                 seq_puts(s, "      |         ");
1290         if (flags & TRACE_GRAPH_PRINT_CPU)
1291                 seq_puts(s, " |  ");
1292         if (flags & TRACE_GRAPH_PRINT_PROC)
1293                 seq_puts(s, "   |    |        ");
1294         if (lat)
1295                 seq_puts(s, "||||");
1296         if (flags & TRACE_GRAPH_PRINT_DURATION)
1297                 seq_puts(s, "   |   |      ");
1298         seq_puts(s, "               |   |   |   |\n");
1299 }
1300
1301 static void print_graph_headers(struct seq_file *s)
1302 {
1303         print_graph_headers_flags(s, tracer_flags.val);
1304 }
1305
1306 void print_graph_headers_flags(struct seq_file *s, u32 flags)
1307 {
1308         struct trace_iterator *iter = s->private;
1309         struct trace_array *tr = iter->tr;
1310
1311         if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
1312                 return;
1313
1314         if (tr->trace_flags & TRACE_ITER_LATENCY_FMT) {
1315                 /* print nothing if the buffers are empty */
1316                 if (trace_empty(iter))
1317                         return;
1318
1319                 print_trace_header(s, iter);
1320         }
1321
1322         __print_graph_headers_flags(tr, s, flags);
1323 }
1324
1325 void graph_trace_open(struct trace_iterator *iter)
1326 {
1327         /* pid and depth on the last trace processed */
1328         struct fgraph_data *data;
1329         gfp_t gfpflags;
1330         int cpu;
1331
1332         iter->private = NULL;
1333
1334         /* We can be called in atomic context via ftrace_dump() */
1335         gfpflags = (in_atomic() || irqs_disabled()) ? GFP_ATOMIC : GFP_KERNEL;
1336
1337         data = kzalloc(sizeof(*data), gfpflags);
1338         if (!data)
1339                 goto out_err;
1340
1341         data->cpu_data = alloc_percpu_gfp(struct fgraph_cpu_data, gfpflags);
1342         if (!data->cpu_data)
1343                 goto out_err_free;
1344
1345         for_each_possible_cpu(cpu) {
1346                 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1347                 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1348                 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1349                 int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1350
1351                 *pid = -1;
1352                 *depth = 0;
1353                 *ignore = 0;
1354                 *depth_irq = -1;
1355         }
1356
1357         iter->private = data;
1358
1359         return;
1360
1361  out_err_free:
1362         kfree(data);
1363  out_err:
1364         pr_warning("function graph tracer: not enough memory\n");
1365 }
1366
1367 void graph_trace_close(struct trace_iterator *iter)
1368 {
1369         struct fgraph_data *data = iter->private;
1370
1371         if (data) {
1372                 free_percpu(data->cpu_data);
1373                 kfree(data);
1374         }
1375 }
1376
1377 static int
1378 func_graph_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1379 {
1380         if (bit == TRACE_GRAPH_PRINT_IRQS)
1381                 ftrace_graph_skip_irqs = !set;
1382
1383         if (bit == TRACE_GRAPH_SLEEP_TIME)
1384                 ftrace_graph_sleep_time_control(set);
1385
1386         if (bit == TRACE_GRAPH_GRAPH_TIME)
1387                 ftrace_graph_graph_time_control(set);
1388
1389         return 0;
1390 }
1391
1392 static struct trace_event_functions graph_functions = {
1393         .trace          = print_graph_function_event,
1394 };
1395
1396 static struct trace_event graph_trace_entry_event = {
1397         .type           = TRACE_GRAPH_ENT,
1398         .funcs          = &graph_functions,
1399 };
1400
1401 static struct trace_event graph_trace_ret_event = {
1402         .type           = TRACE_GRAPH_RET,
1403         .funcs          = &graph_functions
1404 };
1405
1406 static struct tracer graph_trace __tracer_data = {
1407         .name           = "function_graph",
1408         .update_thresh  = graph_trace_update_thresh,
1409         .open           = graph_trace_open,
1410         .pipe_open      = graph_trace_open,
1411         .close          = graph_trace_close,
1412         .pipe_close     = graph_trace_close,
1413         .init           = graph_trace_init,
1414         .reset          = graph_trace_reset,
1415         .print_line     = print_graph_function,
1416         .print_header   = print_graph_headers,
1417         .flags          = &tracer_flags,
1418         .set_flag       = func_graph_set_flag,
1419 #ifdef CONFIG_FTRACE_SELFTEST
1420         .selftest       = trace_selftest_startup_function_graph,
1421 #endif
1422 };
1423
1424
1425 static ssize_t
1426 graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1427                   loff_t *ppos)
1428 {
1429         unsigned long val;
1430         int ret;
1431
1432         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1433         if (ret)
1434                 return ret;
1435
1436         max_depth = val;
1437
1438         *ppos += cnt;
1439
1440         return cnt;
1441 }
1442
1443 static ssize_t
1444 graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1445                  loff_t *ppos)
1446 {
1447         char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1448         int n;
1449
1450         n = sprintf(buf, "%d\n", max_depth);
1451
1452         return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1453 }
1454
1455 static const struct file_operations graph_depth_fops = {
1456         .open           = tracing_open_generic,
1457         .write          = graph_depth_write,
1458         .read           = graph_depth_read,
1459         .llseek         = generic_file_llseek,
1460 };
1461
1462 static __init int init_graph_tracefs(void)
1463 {
1464         struct dentry *d_tracer;
1465
1466         d_tracer = tracing_init_dentry();
1467         if (IS_ERR(d_tracer))
1468                 return 0;
1469
1470         trace_create_file("max_graph_depth", 0644, d_tracer,
1471                           NULL, &graph_depth_fops);
1472
1473         return 0;
1474 }
1475 fs_initcall(init_graph_tracefs);
1476
1477 static __init int init_graph_trace(void)
1478 {
1479         max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1480
1481         if (!register_trace_event(&graph_trace_entry_event)) {
1482                 pr_warning("Warning: could not register graph trace events\n");
1483                 return 1;
1484         }
1485
1486         if (!register_trace_event(&graph_trace_ret_event)) {
1487                 pr_warning("Warning: could not register graph trace events\n");
1488                 return 1;
1489         }
1490
1491         return register_tracer(&graph_trace);
1492 }
1493
1494 core_initcall(init_graph_trace);