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         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
 292                 return 0;
 293
 294         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
 295                                           sizeof(*entry), flags, pc);
 296         if (!event)
 297                 return 0;
 298         entry   = ring_buffer_event_data(event);
 299         entry->graph_ent                        = *trace;
 300         if (!call_filter_check_discard(call, entry, buffer, event))
 301                 __buffer_unlock_commit(buffer, event);
 302
 303         return 1;
 304 }
 305
 306 static inline int ftrace_graph_ignore_irqs(void)
 307 {
 308         if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
 309                 return 0;
 310
 311         return in_irq();
 312 }
 313
 314 int trace_graph_entry(struct ftrace_graph_ent *trace)
 315 {
 316         struct trace_array *tr = graph_array;
 317         struct trace_array_cpu *data;
 318         unsigned long flags;
 319         long disabled;
 320         int ret;
 321         int cpu;
 322         int pc;
 323
 324         if (!ftrace_trace_task(current))
 325                 return 0;
 326
 327         /* trace it when it is-nested-in or is a function enabled. */
 328         if ((!(trace->depth || ftrace_graph_addr(trace->func)) ||
 329              ftrace_graph_ignore_irqs()) || (trace->depth < 0) ||
 330             (max_depth && trace->depth >= max_depth))
 331                 return 0;
 332
 333         /*
 334          * Do not trace a function if it's filtered by set_graph_notrace.
 335          * Make the index of ret stack negative to indicate that it should
 336          * ignore further functions.  But it needs its own ret stack entry
 337          * to recover the original index in order to continue tracing after
 338          * returning from the function.
 339          */
 340         if (ftrace_graph_notrace_addr(trace->func))
 341                 return 1;
 342
 343         local_irq_save(flags);
 344         cpu = raw_smp_processor_id();
 345         data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 346         disabled = atomic_inc_return(&data->disabled);
 347         if (likely(disabled == 1)) {
 348                 pc = preempt_count();
 349                 ret = __trace_graph_entry(tr, trace, flags, pc);
 350         } else {
 351                 ret = 0;
 352         }
 353
 354         atomic_dec(&data->disabled);
 355         local_irq_restore(flags);
 356
 357         return ret;
 358 }
 359
 360 static int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
 361 {
 362         if (tracing_thresh)
 363                 return 1;
 364         else
 365                 return trace_graph_entry(trace);
 366 }
 367
 368 static void
 369 __trace_graph_function(struct trace_array *tr,
 370                 unsigned long ip, unsigned long flags, int pc)
 371 {
 372         u64 time = trace_clock_local();
 373         struct ftrace_graph_ent ent = {
 374                 .func  = ip,
 375                 .depth = 0,
 376         };
 377         struct ftrace_graph_ret ret = {
 378                 .func     = ip,
 379                 .depth    = 0,
 380                 .calltime = time,
 381                 .rettime  = time,
 382         };
 383
 384         __trace_graph_entry(tr, &ent, flags, pc);
 385         __trace_graph_return(tr, &ret, flags, pc);
 386 }
 387
 388 void
 389 trace_graph_function(struct trace_array *tr,
 390                 unsigned long ip, unsigned long parent_ip,
 391                 unsigned long flags, int pc)
 392 {
 393         __trace_graph_function(tr, ip, flags, pc);
 394 }
 395
 396 void __trace_graph_return(struct trace_array *tr,
 397                                 struct ftrace_graph_ret *trace,
 398                                 unsigned long flags,
 399                                 int pc)
 400 {
 401         struct trace_event_call *call = &event_funcgraph_exit;
 402         struct ring_buffer_event *event;
 403         struct ring_buffer *buffer = tr->trace_buffer.buffer;
 404         struct ftrace_graph_ret_entry *entry;
 405
 406         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
 407                 return;
 408
 409         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
 410                                           sizeof(*entry), flags, pc);
 411         if (!event)
 412                 return;
 413         entry   = ring_buffer_event_data(event);
 414         entry->ret                              = *trace;
 415         if (!call_filter_check_discard(call, entry, buffer, event))
 416                 __buffer_unlock_commit(buffer, event);
 417 }
 418
 419 void trace_graph_return(struct ftrace_graph_ret *trace)
 420 {
 421         struct trace_array *tr = graph_array;
 422         struct trace_array_cpu *data;
 423         unsigned long flags;
 424         long disabled;
 425         int cpu;
 426         int pc;
 427
 428         local_irq_save(flags);
 429         cpu = raw_smp_processor_id();
 430         data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 431         disabled = atomic_inc_return(&data->disabled);
 432         if (likely(disabled == 1)) {
 433                 pc = preempt_count();
 434                 __trace_graph_return(tr, trace, flags, pc);
 435         }
 436         atomic_dec(&data->disabled);
 437         local_irq_restore(flags);
 438 }
 439
 440 void set_graph_array(struct trace_array *tr)
 441 {
 442         graph_array = tr;
 443
 444         /* Make graph_array visible before we start tracing */
 445
 446         smp_mb();
 447 }
 448
 449 static void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
 450 {
 451         if (tracing_thresh &&
 452             (trace->rettime - trace->calltime < tracing_thresh))
 453                 return;
 454         else
 455                 trace_graph_return(trace);
 456 }
 457
 458 static int graph_trace_init(struct trace_array *tr)
 459 {
 460         int ret;
 461
 462         set_graph_array(tr);
 463         if (tracing_thresh)
 464                 ret = register_ftrace_graph(&trace_graph_thresh_return,
 465                                             &trace_graph_thresh_entry);
 466         else
 467                 ret = register_ftrace_graph(&trace_graph_return,
 468                                             &trace_graph_entry);
 469         if (ret)
 470                 return ret;
 471         tracing_start_cmdline_record();
 472
 473         return 0;
 474 }
 475
 476 static void graph_trace_reset(struct trace_array *tr)
 477 {
 478         tracing_stop_cmdline_record();
 479         unregister_ftrace_graph();
 480 }
 481
 482 static int graph_trace_update_thresh(struct trace_array *tr)
 483 {
 484         graph_trace_reset(tr);
 485         return graph_trace_init(tr);
 486 }
 487
 488 static int max_bytes_for_cpu;
 489
 490 static void print_graph_cpu(struct trace_seq *s, int cpu)
 491 {
 492         /*
 493          * Start with a space character - to make it stand out
 494          * to the right a bit when trace output is pasted into
 495          * email:
 496          */
 497         trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
 498 }
 499
 500 #define TRACE_GRAPH_PROCINFO_LENGTH     14
 501
 502 static void print_graph_proc(struct trace_seq *s, pid_t pid)
 503 {
 504         char comm[TASK_COMM_LEN];
 505         /* sign + log10(MAX_INT) + '\0' */
 506         char pid_str[11];
 507         int spaces = 0;
 508         int len;
 509         int i;
 510
 511         trace_find_cmdline(pid, comm);
 512         comm[7] = '\0';
 513         sprintf(pid_str, "%d", pid);
 514
 515         /* 1 stands for the "-" character */
 516         len = strlen(comm) + strlen(pid_str) + 1;
 517
 518         if (len < TRACE_GRAPH_PROCINFO_LENGTH)
 519                 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
 520
 521         /* First spaces to align center */
 522         for (i = 0; i < spaces / 2; i++)
 523                 trace_seq_putc(s, ' ');
 524
 525         trace_seq_printf(s, "%s-%s", comm, pid_str);
 526
 527         /* Last spaces to align center */
 528         for (i = 0; i < spaces - (spaces / 2); i++)
 529                 trace_seq_putc(s, ' ');
 530 }
 531
 532
 533 static void print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
 534 {
 535         trace_seq_putc(s, ' ');
 536         trace_print_lat_fmt(s, entry);
 537 }
 538
 539 /* If the pid changed since the last trace, output this event */
 540 static void
 541 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
 542 {
 543         pid_t prev_pid;
 544         pid_t *last_pid;
 545
 546         if (!data)
 547                 return;
 548
 549         last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
 550
 551         if (*last_pid == pid)
 552                 return;
 553
 554         prev_pid = *last_pid;
 555         *last_pid = pid;
 556
 557         if (prev_pid == -1)
 558                 return;
 559 /*
 560  * Context-switch trace line:
 561
 562  ------------------------------------------
 563  | 1)  migration/0--1  =>  sshd-1755
 564  ------------------------------------------
 565
 566  */
 567         trace_seq_puts(s, " ------------------------------------------\n");
 568         print_graph_cpu(s, cpu);
 569         print_graph_proc(s, prev_pid);
 570         trace_seq_puts(s, " => ");
 571         print_graph_proc(s, pid);
 572         trace_seq_puts(s, "\n ------------------------------------------\n\n");
 573 }
 574
 575 static struct ftrace_graph_ret_entry *
 576 get_return_for_leaf(struct trace_iterator *iter,
 577                 struct ftrace_graph_ent_entry *curr)
 578 {
 579         struct fgraph_data *data = iter->private;
 580         struct ring_buffer_iter *ring_iter = NULL;
 581         struct ring_buffer_event *event;
 582         struct ftrace_graph_ret_entry *next;
 583
 584         /*
 585          * If the previous output failed to write to the seq buffer,
 586          * then we just reuse the data from before.
 587          */
 588         if (data && data->failed) {
 589                 curr = &data->ent;
 590                 next = &data->ret;
 591         } else {
 592
 593                 ring_iter = trace_buffer_iter(iter, iter->cpu);
 594
 595                 /* First peek to compare current entry and the next one */
 596                 if (ring_iter)
 597                         event = ring_buffer_iter_peek(ring_iter, NULL);
 598                 else {
 599                         /*
 600                          * We need to consume the current entry to see
 601                          * the next one.
 602                          */
 603                         ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu,
 604                                             NULL, NULL);
 605                         event = ring_buffer_peek(iter->trace_buffer->buffer, iter->cpu,
 606                                                  NULL, NULL);
 607                 }
 608
 609                 if (!event)
 610                         return NULL;
 611
 612                 next = ring_buffer_event_data(event);
 613
 614                 if (data) {
 615                         /*
 616                          * Save current and next entries for later reference
 617                          * if the output fails.
 618                          */
 619                         data->ent = *curr;
 620                         /*
 621                          * If the next event is not a return type, then
 622                          * we only care about what type it is. Otherwise we can
 623                          * safely copy the entire event.
 624                          */
 625                         if (next->ent.type == TRACE_GRAPH_RET)
 626                                 data->ret = *next;
 627                         else
 628                                 data->ret.ent.type = next->ent.type;
 629                 }
 630         }
 631
 632         if (next->ent.type != TRACE_GRAPH_RET)
 633                 return NULL;
 634
 635         if (curr->ent.pid != next->ent.pid ||
 636                         curr->graph_ent.func != next->ret.func)
 637                 return NULL;
 638
 639         /* this is a leaf, now advance the iterator */
 640         if (ring_iter)
 641                 ring_buffer_read(ring_iter, NULL);
 642
 643         return next;
 644 }
 645
 646 static void print_graph_abs_time(u64 t, struct trace_seq *s)
 647 {
 648         unsigned long usecs_rem;
 649
 650         usecs_rem = do_div(t, NSEC_PER_SEC);
 651         usecs_rem /= 1000;
 652
 653         trace_seq_printf(s, "%5lu.%06lu |  ",
 654                          (unsigned long)t, usecs_rem);
 655 }
 656
 657 static void
 658 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
 659                 enum trace_type type, int cpu, pid_t pid, u32 flags)
 660 {
 661         struct trace_array *tr = iter->tr;
 662         struct trace_seq *s = &iter->seq;
 663         struct trace_entry *ent = iter->ent;
 664
 665         if (addr < (unsigned long)__irqentry_text_start ||
 666                 addr >= (unsigned long)__irqentry_text_end)
 667                 return;
 668
 669         if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
 670                 /* Absolute time */
 671                 if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
 672                         print_graph_abs_time(iter->ts, s);
 673
 674                 /* Cpu */
 675                 if (flags & TRACE_GRAPH_PRINT_CPU)
 676                         print_graph_cpu(s, cpu);
 677
 678                 /* Proc */
 679                 if (flags & TRACE_GRAPH_PRINT_PROC) {
 680                         print_graph_proc(s, pid);
 681                         trace_seq_puts(s, " | ");
 682                 }
 683
 684                 /* Latency format */
 685                 if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
 686                         print_graph_lat_fmt(s, ent);
 687         }
 688
 689         /* No overhead */
 690         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_START);
 691
 692         if (type == TRACE_GRAPH_ENT)
 693                 trace_seq_puts(s, "==========>");
 694         else
 695                 trace_seq_puts(s, "<==========");
 696
 697         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_END);
 698         trace_seq_putc(s, '\n');
 699 }
 700
 701 void
 702 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
 703 {
 704         unsigned long nsecs_rem = do_div(duration, 1000);
 705         /* log10(ULONG_MAX) + '\0' */
 706         char usecs_str[21];
 707         char nsecs_str[5];
 708         int len;
 709         int i;
 710
 711         sprintf(usecs_str, "%lu", (unsigned long) duration);
 712
 713         /* Print msecs */
 714         trace_seq_printf(s, "%s", usecs_str);
 715
 716         len = strlen(usecs_str);
 717
 718         /* Print nsecs (we don't want to exceed 7 numbers) */
 719         if (len < 7) {
 720                 size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
 721
 722                 snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
 723                 trace_seq_printf(s, ".%s", nsecs_str);
 724                 len += strlen(nsecs_str) + 1;
 725         }
 726
 727         trace_seq_puts(s, " us ");
 728
 729         /* Print remaining spaces to fit the row's width */
 730         for (i = len; i < 8; i++)
 731                 trace_seq_putc(s, ' ');
 732 }
 733
 734 static void
 735 print_graph_duration(struct trace_array *tr, unsigned long long duration,
 736                      struct trace_seq *s, u32 flags)
 737 {
 738         if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
 739             !(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
 740                 return;
 741
 742         /* No real adata, just filling the column with spaces */
 743         switch (flags & TRACE_GRAPH_PRINT_FILL_MASK) {
 744         case FLAGS_FILL_FULL:
 745                 trace_seq_puts(s, "              |  ");
 746                 return;
 747         case FLAGS_FILL_START:
 748                 trace_seq_puts(s, "  ");
 749                 return;
 750         case FLAGS_FILL_END:
 751                 trace_seq_puts(s, " |");
 752                 return;
 753         }
 754
 755         /* Signal a overhead of time execution to the output */
 756         if (flags & TRACE_GRAPH_PRINT_OVERHEAD)
 757                 trace_seq_printf(s, "%c ", trace_find_mark(duration));
 758         else
 759                 trace_seq_puts(s, "  ");
 760
 761         trace_print_graph_duration(duration, s);
 762         trace_seq_puts(s, "|  ");
 763 }
 764
 765 /* Case of a leaf function on its call entry */
 766 static enum print_line_t
 767 print_graph_entry_leaf(struct trace_iterator *iter,
 768                 struct ftrace_graph_ent_entry *entry,
 769                 struct ftrace_graph_ret_entry *ret_entry,
 770                 struct trace_seq *s, u32 flags)
 771 {
 772         struct fgraph_data *data = iter->private;
 773         struct trace_array *tr = iter->tr;
 774         struct ftrace_graph_ret *graph_ret;
 775         struct ftrace_graph_ent *call;
 776         unsigned long long duration;
 777         int i;
 778
 779         graph_ret = &ret_entry->ret;
 780         call = &entry->graph_ent;
 781         duration = graph_ret->rettime - graph_ret->calltime;
 782
 783         if (data) {
 784                 struct fgraph_cpu_data *cpu_data;
 785                 int cpu = iter->cpu;
 786
 787                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 788
 789                 /*
 790                  * Comments display at + 1 to depth. Since
 791                  * this is a leaf function, keep the comments
 792                  * equal to this depth.
 793                  */
 794                 cpu_data->depth = call->depth - 1;
 795
 796                 /* No need to keep this function around for this depth */
 797                 if (call->depth < FTRACE_RETFUNC_DEPTH)
 798                         cpu_data->enter_funcs[call->depth] = 0;
 799         }
 800
 801         /* Overhead and duration */
 802         print_graph_duration(tr, duration, s, flags);
 803
 804         /* Function */
 805         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
 806                 trace_seq_putc(s, ' ');
 807
 808         trace_seq_printf(s, "%ps();\n", (void *)call->func);
 809
 810         return trace_handle_return(s);
 811 }
 812
 813 static enum print_line_t
 814 print_graph_entry_nested(struct trace_iterator *iter,
 815                          struct ftrace_graph_ent_entry *entry,
 816                          struct trace_seq *s, int cpu, u32 flags)
 817 {
 818         struct ftrace_graph_ent *call = &entry->graph_ent;
 819         struct fgraph_data *data = iter->private;
 820         struct trace_array *tr = iter->tr;
 821         int i;
 822
 823         if (data) {
 824                 struct fgraph_cpu_data *cpu_data;
 825                 int cpu = iter->cpu;
 826
 827                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 828                 cpu_data->depth = call->depth;
 829
 830                 /* Save this function pointer to see if the exit matches */
 831                 if (call->depth < FTRACE_RETFUNC_DEPTH)
 832                         cpu_data->enter_funcs[call->depth] = call->func;
 833         }
 834
 835         /* No time */
 836         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
 837
 838         /* Function */
 839         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
 840                 trace_seq_putc(s, ' ');
 841
 842         trace_seq_printf(s, "%ps() {\n", (void *)call->func);
 843
 844         if (trace_seq_has_overflowed(s))
 845                 return TRACE_TYPE_PARTIAL_LINE;
 846
 847         /*
 848          * we already consumed the current entry to check the next one
 849          * and see if this is a leaf.
 850          */
 851         return TRACE_TYPE_NO_CONSUME;
 852 }
 853
 854 static void
 855 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
 856                      int type, unsigned long addr, u32 flags)
 857 {
 858         struct fgraph_data *data = iter->private;
 859         struct trace_entry *ent = iter->ent;
 860         struct trace_array *tr = iter->tr;
 861         int cpu = iter->cpu;
 862
 863         /* Pid */
 864         verif_pid(s, ent->pid, cpu, data);
 865
 866         if (type)
 867                 /* Interrupt */
 868                 print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
 869
 870         if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
 871                 return;
 872
 873         /* Absolute time */
 874         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
 875                 print_graph_abs_time(iter->ts, s);
 876
 877         /* Cpu */
 878         if (flags & TRACE_GRAPH_PRINT_CPU)
 879                 print_graph_cpu(s, cpu);
 880
 881         /* Proc */
 882         if (flags & TRACE_GRAPH_PRINT_PROC) {
 883                 print_graph_proc(s, ent->pid);
 884                 trace_seq_puts(s, " | ");
 885         }
 886
 887         /* Latency format */
 888         if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
 889                 print_graph_lat_fmt(s, ent);
 890
 891         return;
 892 }
 893
 894 /*
 895  * Entry check for irq code
 896  *
 897  * returns 1 if
 898  *  - we are inside irq code
 899  *  - we just entered irq code
 900  *
 901  * retunns 0 if
 902  *  - funcgraph-interrupts option is set
 903  *  - we are not inside irq code
 904  */
 905 static int
 906 check_irq_entry(struct trace_iterator *iter, u32 flags,
 907                 unsigned long addr, int depth)
 908 {
 909         int cpu = iter->cpu;
 910         int *depth_irq;
 911         struct fgraph_data *data = iter->private;
 912
 913         /*
 914          * If we are either displaying irqs, or we got called as
 915          * a graph event and private data does not exist,
 916          * then we bypass the irq check.
 917          */
 918         if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
 919             (!data))
 920                 return 0;
 921
 922         depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
 923
 924         /*
 925          * We are inside the irq code
 926          */
 927         if (*depth_irq >= 0)
 928                 return 1;
 929
 930         if ((addr < (unsigned long)__irqentry_text_start) ||
 931             (addr >= (unsigned long)__irqentry_text_end))
 932                 return 0;
 933
 934         /*
 935          * We are entering irq code.
 936          */
 937         *depth_irq = depth;
 938         return 1;
 939 }
 940
 941 /*
 942  * Return check for irq code
 943  *
 944  * returns 1 if
 945  *  - we are inside irq code
 946  *  - we just left irq code
 947  *
 948  * returns 0 if
 949  *  - funcgraph-interrupts option is set
 950  *  - we are not inside irq code
 951  */
 952 static int
 953 check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
 954 {
 955         int cpu = iter->cpu;
 956         int *depth_irq;
 957         struct fgraph_data *data = iter->private;
 958
 959         /*
 960          * If we are either displaying irqs, or we got called as
 961          * a graph event and private data does not exist,
 962          * then we bypass the irq check.
 963          */
 964         if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
 965             (!data))
 966                 return 0;
 967
 968         depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
 969
 970         /*
 971          * We are not inside the irq code.
 972          */
 973         if (*depth_irq == -1)
 974                 return 0;
 975
 976         /*
 977          * We are inside the irq code, and this is returning entry.
 978          * Let's not trace it and clear the entry depth, since
 979          * we are out of irq code.
 980          *
 981          * This condition ensures that we 'leave the irq code' once
 982          * we are out of the entry depth. Thus protecting us from
 983          * the RETURN entry loss.
 984          */
 985         if (*depth_irq >= depth) {
 986                 *depth_irq = -1;
 987                 return 1;
 988         }
 989
 990         /*
 991          * We are inside the irq code, and this is not the entry.
 992          */
 993         return 1;
 994 }
 995
 996 static enum print_line_t
 997 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
 998                         struct trace_iterator *iter, u32 flags)
 999 {
1000         struct fgraph_data *data = iter->private;
1001         struct ftrace_graph_ent *call = &field->graph_ent;
1002         struct ftrace_graph_ret_entry *leaf_ret;
1003         static enum print_line_t ret;
1004         int cpu = iter->cpu;
1005
1006         if (check_irq_entry(iter, flags, call->func, call->depth))
1007                 return TRACE_TYPE_HANDLED;
1008
1009         print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags);
1010
1011         leaf_ret = get_return_for_leaf(iter, field);
1012         if (leaf_ret)
1013                 ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
1014         else
1015                 ret = print_graph_entry_nested(iter, field, s, cpu, flags);
1016
1017         if (data) {
1018                 /*
1019                  * If we failed to write our output, then we need to make
1020                  * note of it. Because we already consumed our entry.
1021                  */
1022                 if (s->full) {
1023                         data->failed = 1;
1024                         data->cpu = cpu;
1025                 } else
1026                         data->failed = 0;
1027         }
1028
1029         return ret;
1030 }
1031
1032 static enum print_line_t
1033 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
1034                    struct trace_entry *ent, struct trace_iterator *iter,
1035                    u32 flags)
1036 {
1037         unsigned long long duration = trace->rettime - trace->calltime;
1038         struct fgraph_data *data = iter->private;
1039         struct trace_array *tr = iter->tr;
1040         pid_t pid = ent->pid;
1041         int cpu = iter->cpu;
1042         int func_match = 1;
1043         int i;
1044
1045         if (check_irq_return(iter, flags, trace->depth))
1046                 return TRACE_TYPE_HANDLED;
1047
1048         if (data) {
1049                 struct fgraph_cpu_data *cpu_data;
1050                 int cpu = iter->cpu;
1051
1052                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
1053
1054                 /*
1055                  * Comments display at + 1 to depth. This is the
1056                  * return from a function, we now want the comments
1057                  * to display at the same level of the bracket.
1058                  */
1059                 cpu_data->depth = trace->depth - 1;
1060
1061                 if (trace->depth < FTRACE_RETFUNC_DEPTH) {
1062                         if (cpu_data->enter_funcs[trace->depth] != trace->func)
1063                                 func_match = 0;
1064                         cpu_data->enter_funcs[trace->depth] = 0;
1065                 }
1066         }
1067
1068         print_graph_prologue(iter, s, 0, 0, flags);
1069
1070         /* Overhead and duration */
1071         print_graph_duration(tr, duration, s, flags);
1072
1073         /* Closing brace */
1074         for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++)
1075                 trace_seq_putc(s, ' ');
1076
1077         /*
1078          * If the return function does not have a matching entry,
1079          * then the entry was lost. Instead of just printing
1080          * the '}' and letting the user guess what function this
1081          * belongs to, write out the function name. Always do
1082          * that if the funcgraph-tail option is enabled.
1083          */
1084         if (func_match && !(flags & TRACE_GRAPH_PRINT_TAIL))
1085                 trace_seq_puts(s, "}\n");
1086         else
1087                 trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
1088
1089         /* Overrun */
1090         if (flags & TRACE_GRAPH_PRINT_OVERRUN)
1091                 trace_seq_printf(s, " (Overruns: %lu)\n",
1092                                  trace->overrun);
1093
1094         print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1095                         cpu, pid, flags);
1096
1097         return trace_handle_return(s);
1098 }
1099
1100 static enum print_line_t
1101 print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1102                     struct trace_iterator *iter, u32 flags)
1103 {
1104         struct trace_array *tr = iter->tr;
1105         unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
1106         struct fgraph_data *data = iter->private;
1107         struct trace_event *event;
1108         int depth = 0;
1109         int ret;
1110         int i;
1111
1112         if (data)
1113                 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1114
1115         print_graph_prologue(iter, s, 0, 0, flags);
1116
1117         /* No time */
1118         print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
1119
1120         /* Indentation */
1121         if (depth > 0)
1122                 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++)
1123                         trace_seq_putc(s, ' ');
1124
1125         /* The comment */
1126         trace_seq_puts(s, "/* ");
1127
1128         switch (iter->ent->type) {
1129         case TRACE_BPRINT:
1130                 ret = trace_print_bprintk_msg_only(iter);
1131                 if (ret != TRACE_TYPE_HANDLED)
1132                         return ret;
1133                 break;
1134         case TRACE_PRINT:
1135                 ret = trace_print_printk_msg_only(iter);
1136                 if (ret != TRACE_TYPE_HANDLED)
1137                         return ret;
1138                 break;
1139         default:
1140                 event = ftrace_find_event(ent->type);
1141                 if (!event)
1142                         return TRACE_TYPE_UNHANDLED;
1143
1144                 ret = event->funcs->trace(iter, sym_flags, event);
1145                 if (ret != TRACE_TYPE_HANDLED)
1146                         return ret;
1147         }
1148
1149         if (trace_seq_has_overflowed(s))
1150                 goto out;
1151
1152         /* Strip ending newline */
1153         if (s->buffer[s->seq.len - 1] == '\n') {
1154                 s->buffer[s->seq.len - 1] = '\0';
1155                 s->seq.len--;
1156         }
1157
1158         trace_seq_puts(s, " */\n");
1159  out:
1160         return trace_handle_return(s);
1161 }
1162
1163
1164 enum print_line_t
1165 print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1166 {
1167         struct ftrace_graph_ent_entry *field;
1168         struct fgraph_data *data = iter->private;
1169         struct trace_entry *entry = iter->ent;
1170         struct trace_seq *s = &iter->seq;
1171         int cpu = iter->cpu;
1172         int ret;
1173
1174         if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1175                 per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1176                 return TRACE_TYPE_HANDLED;
1177         }
1178
1179         /*
1180          * If the last output failed, there's a possibility we need
1181          * to print out the missing entry which would never go out.
1182          */
1183         if (data && data->failed) {
1184                 field = &data->ent;
1185                 iter->cpu = data->cpu;
1186                 ret = print_graph_entry(field, s, iter, flags);
1187                 if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1188                         per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1189                         ret = TRACE_TYPE_NO_CONSUME;
1190                 }
1191                 iter->cpu = cpu;
1192                 return ret;
1193         }
1194
1195         switch (entry->type) {
1196         case TRACE_GRAPH_ENT: {
1197                 /*
1198                  * print_graph_entry() may consume the current event,
1199                  * thus @field may become invalid, so we need to save it.
1200                  * sizeof(struct ftrace_graph_ent_entry) is very small,
1201                  * it can be safely saved at the stack.
1202                  */
1203                 struct ftrace_graph_ent_entry saved;
1204                 trace_assign_type(field, entry);
1205                 saved = *field;
1206                 return print_graph_entry(&saved, s, iter, flags);
1207         }
1208         case TRACE_GRAPH_RET: {
1209                 struct ftrace_graph_ret_entry *field;
1210                 trace_assign_type(field, entry);
1211                 return print_graph_return(&field->ret, s, entry, iter, flags);
1212         }
1213         case TRACE_STACK:
1214         case TRACE_FN:
1215                 /* dont trace stack and functions as comments */
1216                 return TRACE_TYPE_UNHANDLED;
1217
1218         default:
1219                 return print_graph_comment(s, entry, iter, flags);
1220         }
1221
1222         return TRACE_TYPE_HANDLED;
1223 }
1224
1225 static enum print_line_t
1226 print_graph_function(struct trace_iterator *iter)
1227 {
1228         return print_graph_function_flags(iter, tracer_flags.val);
1229 }
1230
1231 static enum print_line_t
1232 print_graph_function_event(struct trace_iterator *iter, int flags,
1233                            struct trace_event *event)
1234 {
1235         return print_graph_function(iter);
1236 }
1237
1238 static void print_lat_header(struct seq_file *s, u32 flags)
1239 {
1240         static const char spaces[] = "                " /* 16 spaces */
1241                 "    "                                  /* 4 spaces */
1242                 "                 ";                    /* 17 spaces */
1243         int size = 0;
1244
1245         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1246                 size += 16;
1247         if (flags & TRACE_GRAPH_PRINT_CPU)
1248                 size += 4;
1249         if (flags & TRACE_GRAPH_PRINT_PROC)
1250                 size += 17;
1251
1252         seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1253         seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1254         seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1255         seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1256         seq_printf(s, "#%.*s||| /                      \n", size, spaces);
1257 }
1258
1259 static void __print_graph_headers_flags(struct trace_array *tr,
1260                                         struct seq_file *s, u32 flags)
1261 {
1262         int lat = tr->trace_flags & TRACE_ITER_LATENCY_FMT;
1263
1264         if (lat)
1265                 print_lat_header(s, flags);
1266
1267         /* 1st line */
1268         seq_putc(s, '#');
1269         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1270                 seq_puts(s, "     TIME       ");
1271         if (flags & TRACE_GRAPH_PRINT_CPU)
1272                 seq_puts(s, " CPU");
1273         if (flags & TRACE_GRAPH_PRINT_PROC)
1274                 seq_puts(s, "  TASK/PID       ");
1275         if (lat)
1276                 seq_puts(s, "||||");
1277         if (flags & TRACE_GRAPH_PRINT_DURATION)
1278                 seq_puts(s, "  DURATION   ");
1279         seq_puts(s, "               FUNCTION CALLS\n");
1280
1281         /* 2nd line */
1282         seq_putc(s, '#');
1283         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1284                 seq_puts(s, "      |         ");
1285         if (flags & TRACE_GRAPH_PRINT_CPU)
1286                 seq_puts(s, " |  ");
1287         if (flags & TRACE_GRAPH_PRINT_PROC)
1288                 seq_puts(s, "   |    |        ");
1289         if (lat)
1290                 seq_puts(s, "||||");
1291         if (flags & TRACE_GRAPH_PRINT_DURATION)
1292                 seq_puts(s, "   |   |      ");
1293         seq_puts(s, "               |   |   |   |\n");
1294 }
1295
1296 static void print_graph_headers(struct seq_file *s)
1297 {
1298         print_graph_headers_flags(s, tracer_flags.val);
1299 }
1300
1301 void print_graph_headers_flags(struct seq_file *s, u32 flags)
1302 {
1303         struct trace_iterator *iter = s->private;
1304         struct trace_array *tr = iter->tr;
1305
1306         if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
1307                 return;
1308
1309         if (tr->trace_flags & TRACE_ITER_LATENCY_FMT) {
1310                 /* print nothing if the buffers are empty */
1311                 if (trace_empty(iter))
1312                         return;
1313
1314                 print_trace_header(s, iter);
1315         }
1316
1317         __print_graph_headers_flags(tr, s, flags);
1318 }
1319
1320 void graph_trace_open(struct trace_iterator *iter)
1321 {
1322         /* pid and depth on the last trace processed */
1323         struct fgraph_data *data;
1324         gfp_t gfpflags;
1325         int cpu;
1326
1327         iter->private = NULL;
1328
1329         /* We can be called in atomic context via ftrace_dump() */
1330         gfpflags = (in_atomic() || irqs_disabled()) ? GFP_ATOMIC : GFP_KERNEL;
1331
1332         data = kzalloc(sizeof(*data), gfpflags);
1333         if (!data)
1334                 goto out_err;
1335
1336         data->cpu_data = alloc_percpu_gfp(struct fgraph_cpu_data, gfpflags);
1337         if (!data->cpu_data)
1338                 goto out_err_free;
1339
1340         for_each_possible_cpu(cpu) {
1341                 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1342                 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1343                 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1344                 int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1345
1346                 *pid = -1;
1347                 *depth = 0;
1348                 *ignore = 0;
1349                 *depth_irq = -1;
1350         }
1351
1352         iter->private = data;
1353
1354         return;
1355
1356  out_err_free:
1357         kfree(data);
1358  out_err:
1359         pr_warning("function graph tracer: not enough memory\n");
1360 }
1361
1362 void graph_trace_close(struct trace_iterator *iter)
1363 {
1364         struct fgraph_data *data = iter->private;
1365
1366         if (data) {
1367                 free_percpu(data->cpu_data);
1368                 kfree(data);
1369         }
1370 }
1371
1372 static int
1373 func_graph_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1374 {
1375         if (bit == TRACE_GRAPH_PRINT_IRQS)
1376                 ftrace_graph_skip_irqs = !set;
1377
1378         if (bit == TRACE_GRAPH_SLEEP_TIME)
1379                 ftrace_graph_sleep_time_control(set);
1380
1381         if (bit == TRACE_GRAPH_GRAPH_TIME)
1382                 ftrace_graph_graph_time_control(set);
1383
1384         return 0;
1385 }
1386
1387 static struct trace_event_functions graph_functions = {
1388         .trace          = print_graph_function_event,
1389 };
1390
1391 static struct trace_event graph_trace_entry_event = {
1392         .type           = TRACE_GRAPH_ENT,
1393         .funcs          = &graph_functions,
1394 };
1395
1396 static struct trace_event graph_trace_ret_event = {
1397         .type           = TRACE_GRAPH_RET,
1398         .funcs          = &graph_functions
1399 };
1400
1401 static struct tracer graph_trace __tracer_data = {
1402         .name           = "function_graph",
1403         .update_thresh  = graph_trace_update_thresh,
1404         .open           = graph_trace_open,
1405         .pipe_open      = graph_trace_open,
1406         .close          = graph_trace_close,
1407         .pipe_close     = graph_trace_close,
1408         .init           = graph_trace_init,
1409         .reset          = graph_trace_reset,
1410         .print_line     = print_graph_function,
1411         .print_header   = print_graph_headers,
1412         .flags          = &tracer_flags,
1413         .set_flag       = func_graph_set_flag,
1414 #ifdef CONFIG_FTRACE_SELFTEST
1415         .selftest       = trace_selftest_startup_function_graph,
1416 #endif
1417 };
1418
1419
1420 static ssize_t
1421 graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1422                   loff_t *ppos)
1423 {
1424         unsigned long val;
1425         int ret;
1426
1427         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1428         if (ret)
1429                 return ret;
1430
1431         max_depth = val;
1432
1433         *ppos += cnt;
1434
1435         return cnt;
1436 }
1437
1438 static ssize_t
1439 graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1440                  loff_t *ppos)
1441 {
1442         char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1443         int n;
1444
1445         n = sprintf(buf, "%d\n", max_depth);
1446
1447         return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1448 }
1449
1450 static const struct file_operations graph_depth_fops = {
1451         .open           = tracing_open_generic,
1452         .write          = graph_depth_write,
1453         .read           = graph_depth_read,
1454         .llseek         = generic_file_llseek,
1455 };
1456
1457 static __init int init_graph_tracefs(void)
1458 {
1459         struct dentry *d_tracer;
1460
1461         d_tracer = tracing_init_dentry();
1462         if (IS_ERR(d_tracer))
1463                 return 0;
1464
1465         trace_create_file("max_graph_depth", 0644, d_tracer,
1466                           NULL, &graph_depth_fops);
1467
1468         return 0;
1469 }
1470 fs_initcall(init_graph_tracefs);
1471
1472 static __init int init_graph_trace(void)
1473 {
1474         max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1475
1476         if (!register_trace_event(&graph_trace_entry_event)) {
1477                 pr_warning("Warning: could not register graph trace events\n");
1478                 return 1;
1479         }
1480
1481         if (!register_trace_event(&graph_trace_ret_event)) {
1482                 pr_warning("Warning: could not register graph trace events\n");
1483                 return 1;
1484         }
1485
1486         return register_tracer(&graph_trace);
1487 }
1488
1489 core_initcall(init_graph_trace);