kernel/trace/trace_functions_graph.c

   1 /*
   2  *
   3  * Function graph tracer.
   4  * Copyright (c) 2008-2009 Frederic Weisbecker <fweisbec@gmail.com>
   5  * Mostly borrowed from function tracer which
   6  * is Copyright (c) Steven Rostedt <srostedt@redhat.com>
   7  *
   8  */
   9 #include <linux/debugfs.h>
  10 #include <linux/uaccess.h>
  11 #include <linux/ftrace.h>
  12 #include <linux/slab.h>
  13 #include <linux/fs.h>
  14
  15 #include "trace.h"
  16 #include "trace_output.h"
  17
  18 /* When set, irq functions will be ignored */
  19 static int ftrace_graph_skip_irqs;
  20
  21 struct fgraph_cpu_data {
  22         pid_t           last_pid;
  23         int             depth;
  24         int             depth_irq;
  25         int             ignore;
  26         unsigned long   enter_funcs[FTRACE_RETFUNC_DEPTH];
  27 };
  28
  29 struct fgraph_data {
  30         struct fgraph_cpu_data __percpu *cpu_data;
  31
  32         /* Place to preserve last processed entry. */
  33         struct ftrace_graph_ent_entry   ent;
  34         struct ftrace_graph_ret_entry   ret;
  35         int                             failed;
  36         int                             cpu;
  37 };
  38
  39 #define TRACE_GRAPH_INDENT      2
  40
  41 static unsigned int max_depth;
  42
  43 static struct tracer_opt trace_opts[] = {
  44         /* Display overruns? (for self-debug purpose) */
  45         { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
  46         /* Display CPU ? */
  47         { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
  48         /* Display Overhead ? */
  49         { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
  50         /* Display proc name/pid */
  51         { TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
  52         /* Display duration of execution */
  53         { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
  54         /* Display absolute time of an entry */
  55         { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
  56         /* Display interrupts */
  57         { TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
  58         /* Display function name after trailing } */
  59         { TRACER_OPT(funcgraph-tail, TRACE_GRAPH_PRINT_TAIL) },
  60         { } /* Empty entry */
  61 };
  62
  63 static struct tracer_flags tracer_flags = {
  64         /* Don't display overruns, proc, or tail by default */
  65         .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
  66                TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS,
  67         .opts = trace_opts
  68 };
  69
  70 static struct trace_array *graph_array;
  71
  72 /*
  73  * DURATION column is being also used to display IRQ signs,
  74  * following values are used by print_graph_irq and others
  75  * to fill in space into DURATION column.
  76  */
  77 enum {
  78         FLAGS_FILL_FULL  = 1 << TRACE_GRAPH_PRINT_FILL_SHIFT,
  79         FLAGS_FILL_START = 2 << TRACE_GRAPH_PRINT_FILL_SHIFT,
  80         FLAGS_FILL_END   = 3 << TRACE_GRAPH_PRINT_FILL_SHIFT,
  81 };
  82
  83 static enum print_line_t
  84 print_graph_duration(unsigned long long duration, struct trace_seq *s,
  85                      u32 flags);
  86
  87 /* Add a function return address to the trace stack on thread info.*/
  88 int
  89 ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
  90                          unsigned long frame_pointer)
  91 {
  92         unsigned long long calltime;
  93         int index;
  94
  95         if (!current->ret_stack)
  96                 return -EBUSY;
  97
  98         /*
  99          * We must make sure the ret_stack is tested before we read
 100          * anything else.
 101          */
 102         smp_rmb();
 103
 104         /* The return trace stack is full */
 105         if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
 106                 atomic_inc(&current->trace_overrun);
 107                 return -EBUSY;
 108         }
 109
 110         /*
 111          * The curr_ret_stack is an index to ftrace return stack of
 112          * current task.  Its value should be in [0, FTRACE_RETFUNC_
 113          * DEPTH) when the function graph tracer is used.  To support
 114          * filtering out specific functions, it makes the index
 115          * negative by subtracting huge value (FTRACE_NOTRACE_DEPTH)
 116          * so when it sees a negative index the ftrace will ignore
 117          * the record.  And the index gets recovered when returning
 118          * from the filtered function by adding the FTRACE_NOTRACE_
 119          * DEPTH and then it'll continue to record functions normally.
 120          *
 121          * The curr_ret_stack is initialized to -1 and get increased
 122          * in this function.  So it can be less than -1 only if it was
 123          * filtered out via ftrace_graph_notrace_addr() which can be
 124          * set from set_graph_notrace file in debugfs by user.
 125          */
 126         if (current->curr_ret_stack < -1)
 127                 return -EBUSY;
 128
 129         calltime = trace_clock_local();
 130
 131         index = ++current->curr_ret_stack;
 132         if (ftrace_graph_notrace_addr(func))
 133                 current->curr_ret_stack -= FTRACE_NOTRACE_DEPTH;
 134         barrier();
 135         current->ret_stack[index].ret = ret;
 136         current->ret_stack[index].func = func;
 137         current->ret_stack[index].calltime = calltime;
 138         current->ret_stack[index].subtime = 0;
 139         current->ret_stack[index].fp = frame_pointer;
 140         *depth = current->curr_ret_stack;
 141
 142         return 0;
 143 }
 144
 145 /* Retrieve a function return address to the trace stack on thread info.*/
 146 static void
 147 ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
 148                         unsigned long frame_pointer)
 149 {
 150         int index;
 151
 152         index = current->curr_ret_stack;
 153
 154         /*
 155          * A negative index here means that it's just returned from a
 156          * notrace'd function.  Recover index to get an original
 157          * return address.  See ftrace_push_return_trace().
 158          *
 159          * TODO: Need to check whether the stack gets corrupted.
 160          */
 161         if (index < 0)
 162                 index += FTRACE_NOTRACE_DEPTH;
 163
 164         if (unlikely(index < 0 || index >= FTRACE_RETFUNC_DEPTH)) {
 165                 ftrace_graph_stop();
 166                 WARN_ON(1);
 167                 /* Might as well panic, otherwise we have no where to go */
 168                 *ret = (unsigned long)panic;
 169                 return;
 170         }
 171
 172 #if defined(CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST) && !defined(CC_USING_FENTRY)
 173         /*
 174          * The arch may choose to record the frame pointer used
 175          * and check it here to make sure that it is what we expect it
 176          * to be. If gcc does not set the place holder of the return
 177          * address in the frame pointer, and does a copy instead, then
 178          * the function graph trace will fail. This test detects this
 179          * case.
 180          *
 181          * Currently, x86_32 with optimize for size (-Os) makes the latest
 182          * gcc do the above.
 183          *
 184          * Note, -mfentry does not use frame pointers, and this test
 185          *  is not needed if CC_USING_FENTRY is set.
 186          */
 187         if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
 188                 ftrace_graph_stop();
 189                 WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
 190                      "  from func %ps return to %lx\n",
 191                      current->ret_stack[index].fp,
 192                      frame_pointer,
 193                      (void *)current->ret_stack[index].func,
 194                      current->ret_stack[index].ret);
 195                 *ret = (unsigned long)panic;
 196                 return;
 197         }
 198 #endif
 199
 200         *ret = current->ret_stack[index].ret;
 201         trace->func = current->ret_stack[index].func;
 202         trace->calltime = current->ret_stack[index].calltime;
 203         trace->overrun = atomic_read(&current->trace_overrun);
 204         trace->depth = index;
 205 }
 206
 207 /*
 208  * Send the trace to the ring-buffer.
 209  * @return the original return address.
 210  */
 211 unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
 212 {
 213         struct ftrace_graph_ret trace;
 214         unsigned long ret;
 215
 216         ftrace_pop_return_trace(&trace, &ret, frame_pointer);
 217         trace.rettime = trace_clock_local();
 218         barrier();
 219         current->curr_ret_stack--;
 220         /*
 221          * The curr_ret_stack can be less than -1 only if it was
 222          * filtered out and it's about to return from the function.
 223          * Recover the index and continue to trace normal functions.
 224          */
 225         if (current->curr_ret_stack < -1) {
 226                 current->curr_ret_stack += FTRACE_NOTRACE_DEPTH;
 227                 return ret;
 228         }
 229
 230         /*
 231          * The trace should run after decrementing the ret counter
 232          * in case an interrupt were to come in. We don't want to
 233          * lose the interrupt if max_depth is set.
 234          */
 235         ftrace_graph_return(&trace);
 236
 237         if (unlikely(!ret)) {
 238                 ftrace_graph_stop();
 239                 WARN_ON(1);
 240                 /* Might as well panic. What else to do? */
 241                 ret = (unsigned long)panic;
 242         }
 243
 244         return ret;
 245 }
 246
 247 int __trace_graph_entry(struct trace_array *tr,
 248                                 struct ftrace_graph_ent *trace,
 249                                 unsigned long flags,
 250                                 int pc)
 251 {
 252         struct ftrace_event_call *call = &event_funcgraph_entry;
 253         struct ring_buffer_event *event;
 254         struct ring_buffer *buffer = tr->trace_buffer.buffer;
 255         struct ftrace_graph_ent_entry *entry;
 256
 257         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
 258                 return 0;
 259
 260         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
 261                                           sizeof(*entry), flags, pc);
 262         if (!event)
 263                 return 0;
 264         entry   = ring_buffer_event_data(event);
 265         entry->graph_ent                        = *trace;
 266         if (!call_filter_check_discard(call, entry, buffer, event))
 267                 __buffer_unlock_commit(buffer, event);
 268
 269         return 1;
 270 }
 271
 272 static inline int ftrace_graph_ignore_irqs(void)
 273 {
 274         if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
 275                 return 0;
 276
 277         return in_irq();
 278 }
 279
 280 int trace_graph_entry(struct ftrace_graph_ent *trace)
 281 {
 282         struct trace_array *tr = graph_array;
 283         struct trace_array_cpu *data;
 284         unsigned long flags;
 285         long disabled;
 286         int ret;
 287         int cpu;
 288         int pc;
 289
 290         if (!ftrace_trace_task(current))
 291                 return 0;
 292
 293         /* trace it when it is-nested-in or is a function enabled. */
 294         if ((!(trace->depth || ftrace_graph_addr(trace->func)) ||
 295              ftrace_graph_ignore_irqs()) || (trace->depth < 0) ||
 296             (max_depth && trace->depth >= max_depth))
 297                 return 0;
 298
 299         /*
 300          * Do not trace a function if it's filtered by set_graph_notrace.
 301          * Make the index of ret stack negative to indicate that it should
 302          * ignore further functions.  But it needs its own ret stack entry
 303          * to recover the original index in order to continue tracing after
 304          * returning from the function.
 305          */
 306         if (ftrace_graph_notrace_addr(trace->func))
 307                 return 1;
 308
 309         local_irq_save(flags);
 310         cpu = raw_smp_processor_id();
 311         data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 312         disabled = atomic_inc_return(&data->disabled);
 313         if (likely(disabled == 1)) {
 314                 pc = preempt_count();
 315                 ret = __trace_graph_entry(tr, trace, flags, pc);
 316         } else {
 317                 ret = 0;
 318         }
 319
 320         atomic_dec(&data->disabled);
 321         local_irq_restore(flags);
 322
 323         return ret;
 324 }
 325
 326 int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
 327 {
 328         if (tracing_thresh)
 329                 return 1;
 330         else
 331                 return trace_graph_entry(trace);
 332 }
 333
 334 static void
 335 __trace_graph_function(struct trace_array *tr,
 336                 unsigned long ip, unsigned long flags, int pc)
 337 {
 338         u64 time = trace_clock_local();
 339         struct ftrace_graph_ent ent = {
 340                 .func  = ip,
 341                 .depth = 0,
 342         };
 343         struct ftrace_graph_ret ret = {
 344                 .func     = ip,
 345                 .depth    = 0,
 346                 .calltime = time,
 347                 .rettime  = time,
 348         };
 349
 350         __trace_graph_entry(tr, &ent, flags, pc);
 351         __trace_graph_return(tr, &ret, flags, pc);
 352 }
 353
 354 void
 355 trace_graph_function(struct trace_array *tr,
 356                 unsigned long ip, unsigned long parent_ip,
 357                 unsigned long flags, int pc)
 358 {
 359         __trace_graph_function(tr, ip, flags, pc);
 360 }
 361
 362 void __trace_graph_return(struct trace_array *tr,
 363                                 struct ftrace_graph_ret *trace,
 364                                 unsigned long flags,
 365                                 int pc)
 366 {
 367         struct ftrace_event_call *call = &event_funcgraph_exit;
 368         struct ring_buffer_event *event;
 369         struct ring_buffer *buffer = tr->trace_buffer.buffer;
 370         struct ftrace_graph_ret_entry *entry;
 371
 372         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
 373                 return;
 374
 375         event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
 376                                           sizeof(*entry), flags, pc);
 377         if (!event)
 378                 return;
 379         entry   = ring_buffer_event_data(event);
 380         entry->ret                              = *trace;
 381         if (!call_filter_check_discard(call, entry, buffer, event))
 382                 __buffer_unlock_commit(buffer, event);
 383 }
 384
 385 void trace_graph_return(struct ftrace_graph_ret *trace)
 386 {
 387         struct trace_array *tr = graph_array;
 388         struct trace_array_cpu *data;
 389         unsigned long flags;
 390         long disabled;
 391         int cpu;
 392         int pc;
 393
 394         local_irq_save(flags);
 395         cpu = raw_smp_processor_id();
 396         data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 397         disabled = atomic_inc_return(&data->disabled);
 398         if (likely(disabled == 1)) {
 399                 pc = preempt_count();
 400                 __trace_graph_return(tr, trace, flags, pc);
 401         }
 402         atomic_dec(&data->disabled);
 403         local_irq_restore(flags);
 404 }
 405
 406 void set_graph_array(struct trace_array *tr)
 407 {
 408         graph_array = tr;
 409
 410         /* Make graph_array visible before we start tracing */
 411
 412         smp_mb();
 413 }
 414
 415 void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
 416 {
 417         if (tracing_thresh &&
 418             (trace->rettime - trace->calltime < tracing_thresh))
 419                 return;
 420         else
 421                 trace_graph_return(trace);
 422 }
 423
 424 static int graph_trace_init(struct trace_array *tr)
 425 {
 426         int ret;
 427
 428         set_graph_array(tr);
 429         if (tracing_thresh)
 430                 ret = register_ftrace_graph(&trace_graph_thresh_return,
 431                                             &trace_graph_thresh_entry);
 432         else
 433                 ret = register_ftrace_graph(&trace_graph_return,
 434                                             &trace_graph_entry);
 435         if (ret)
 436                 return ret;
 437         tracing_start_cmdline_record();
 438
 439         return 0;
 440 }
 441
 442 static void graph_trace_reset(struct trace_array *tr)
 443 {
 444         tracing_stop_cmdline_record();
 445         unregister_ftrace_graph();
 446 }
 447
 448 static int max_bytes_for_cpu;
 449
 450 static enum print_line_t
 451 print_graph_cpu(struct trace_seq *s, int cpu)
 452 {
 453         int ret;
 454
 455         /*
 456          * Start with a space character - to make it stand out
 457          * to the right a bit when trace output is pasted into
 458          * email:
 459          */
 460         ret = trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
 461         if (!ret)
 462                 return TRACE_TYPE_PARTIAL_LINE;
 463
 464         return TRACE_TYPE_HANDLED;
 465 }
 466
 467 #define TRACE_GRAPH_PROCINFO_LENGTH     14
 468
 469 static enum print_line_t
 470 print_graph_proc(struct trace_seq *s, pid_t pid)
 471 {
 472         char comm[TASK_COMM_LEN];
 473         /* sign + log10(MAX_INT) + '\0' */
 474         char pid_str[11];
 475         int spaces = 0;
 476         int ret;
 477         int len;
 478         int i;
 479
 480         trace_find_cmdline(pid, comm);
 481         comm[7] = '\0';
 482         sprintf(pid_str, "%d", pid);
 483
 484         /* 1 stands for the "-" character */
 485         len = strlen(comm) + strlen(pid_str) + 1;
 486
 487         if (len < TRACE_GRAPH_PROCINFO_LENGTH)
 488                 spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
 489
 490         /* First spaces to align center */
 491         for (i = 0; i < spaces / 2; i++) {
 492                 ret = trace_seq_putc(s, ' ');
 493                 if (!ret)
 494                         return TRACE_TYPE_PARTIAL_LINE;
 495         }
 496
 497         ret = trace_seq_printf(s, "%s-%s", comm, pid_str);
 498         if (!ret)
 499                 return TRACE_TYPE_PARTIAL_LINE;
 500
 501         /* Last spaces to align center */
 502         for (i = 0; i < spaces - (spaces / 2); i++) {
 503                 ret = trace_seq_putc(s, ' ');
 504                 if (!ret)
 505                         return TRACE_TYPE_PARTIAL_LINE;
 506         }
 507         return TRACE_TYPE_HANDLED;
 508 }
 509
 510
 511 static enum print_line_t
 512 print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
 513 {
 514         if (!trace_seq_putc(s, ' '))
 515                 return 0;
 516
 517         return trace_print_lat_fmt(s, entry);
 518 }
 519
 520 /* If the pid changed since the last trace, output this event */
 521 static enum print_line_t
 522 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
 523 {
 524         pid_t prev_pid;
 525         pid_t *last_pid;
 526         int ret;
 527
 528         if (!data)
 529                 return TRACE_TYPE_HANDLED;
 530
 531         last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
 532
 533         if (*last_pid == pid)
 534                 return TRACE_TYPE_HANDLED;
 535
 536         prev_pid = *last_pid;
 537         *last_pid = pid;
 538
 539         if (prev_pid == -1)
 540                 return TRACE_TYPE_HANDLED;
 541 /*
 542  * Context-switch trace line:
 543
 544  ------------------------------------------
 545  | 1)  migration/0--1  =>  sshd-1755
 546  ------------------------------------------
 547
 548  */
 549         ret = trace_seq_puts(s,
 550                 " ------------------------------------------\n");
 551         if (!ret)
 552                 return TRACE_TYPE_PARTIAL_LINE;
 553
 554         ret = print_graph_cpu(s, cpu);
 555         if (ret == TRACE_TYPE_PARTIAL_LINE)
 556                 return TRACE_TYPE_PARTIAL_LINE;
 557
 558         ret = print_graph_proc(s, prev_pid);
 559         if (ret == TRACE_TYPE_PARTIAL_LINE)
 560                 return TRACE_TYPE_PARTIAL_LINE;
 561
 562         ret = trace_seq_puts(s, " => ");
 563         if (!ret)
 564                 return TRACE_TYPE_PARTIAL_LINE;
 565
 566         ret = print_graph_proc(s, pid);
 567         if (ret == TRACE_TYPE_PARTIAL_LINE)
 568                 return TRACE_TYPE_PARTIAL_LINE;
 569
 570         ret = trace_seq_puts(s,
 571                 "\n ------------------------------------------\n\n");
 572         if (!ret)
 573                 return TRACE_TYPE_PARTIAL_LINE;
 574
 575         return TRACE_TYPE_HANDLED;
 576 }
 577
 578 static struct ftrace_graph_ret_entry *
 579 get_return_for_leaf(struct trace_iterator *iter,
 580                 struct ftrace_graph_ent_entry *curr)
 581 {
 582         struct fgraph_data *data = iter->private;
 583         struct ring_buffer_iter *ring_iter = NULL;
 584         struct ring_buffer_event *event;
 585         struct ftrace_graph_ret_entry *next;
 586
 587         /*
 588          * If the previous output failed to write to the seq buffer,
 589          * then we just reuse the data from before.
 590          */
 591         if (data && data->failed) {
 592                 curr = &data->ent;
 593                 next = &data->ret;
 594         } else {
 595
 596                 ring_iter = trace_buffer_iter(iter, iter->cpu);
 597
 598                 /* First peek to compare current entry and the next one */
 599                 if (ring_iter)
 600                         event = ring_buffer_iter_peek(ring_iter, NULL);
 601                 else {
 602                         /*
 603                          * We need to consume the current entry to see
 604                          * the next one.
 605                          */
 606                         ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu,
 607                                             NULL, NULL);
 608                         event = ring_buffer_peek(iter->trace_buffer->buffer, iter->cpu,
 609                                                  NULL, NULL);
 610                 }
 611
 612                 if (!event)
 613                         return NULL;
 614
 615                 next = ring_buffer_event_data(event);
 616
 617                 if (data) {
 618                         /*
 619                          * Save current and next entries for later reference
 620                          * if the output fails.
 621                          */
 622                         data->ent = *curr;
 623                         /*
 624                          * If the next event is not a return type, then
 625                          * we only care about what type it is. Otherwise we can
 626                          * safely copy the entire event.
 627                          */
 628                         if (next->ent.type == TRACE_GRAPH_RET)
 629                                 data->ret = *next;
 630                         else
 631                                 data->ret.ent.type = next->ent.type;
 632                 }
 633         }
 634
 635         if (next->ent.type != TRACE_GRAPH_RET)
 636                 return NULL;
 637
 638         if (curr->ent.pid != next->ent.pid ||
 639                         curr->graph_ent.func != next->ret.func)
 640                 return NULL;
 641
 642         /* this is a leaf, now advance the iterator */
 643         if (ring_iter)
 644                 ring_buffer_read(ring_iter, NULL);
 645
 646         return next;
 647 }
 648
 649 static int print_graph_abs_time(u64 t, struct trace_seq *s)
 650 {
 651         unsigned long usecs_rem;
 652
 653         usecs_rem = do_div(t, NSEC_PER_SEC);
 654         usecs_rem /= 1000;
 655
 656         return trace_seq_printf(s, "%5lu.%06lu |  ",
 657                         (unsigned long)t, usecs_rem);
 658 }
 659
 660 static enum print_line_t
 661 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
 662                 enum trace_type type, int cpu, pid_t pid, u32 flags)
 663 {
 664         int ret;
 665         struct trace_seq *s = &iter->seq;
 666
 667         if (addr < (unsigned long)__irqentry_text_start ||
 668                 addr >= (unsigned long)__irqentry_text_end)
 669                 return TRACE_TYPE_UNHANDLED;
 670
 671         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
 672                 /* Absolute time */
 673                 if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
 674                         ret = print_graph_abs_time(iter->ts, s);
 675                         if (!ret)
 676                                 return TRACE_TYPE_PARTIAL_LINE;
 677                 }
 678
 679                 /* Cpu */
 680                 if (flags & TRACE_GRAPH_PRINT_CPU) {
 681                         ret = print_graph_cpu(s, cpu);
 682                         if (ret == TRACE_TYPE_PARTIAL_LINE)
 683                                 return TRACE_TYPE_PARTIAL_LINE;
 684                 }
 685
 686                 /* Proc */
 687                 if (flags & TRACE_GRAPH_PRINT_PROC) {
 688                         ret = print_graph_proc(s, pid);
 689                         if (ret == TRACE_TYPE_PARTIAL_LINE)
 690                                 return TRACE_TYPE_PARTIAL_LINE;
 691                         ret = trace_seq_puts(s, " | ");
 692                         if (!ret)
 693                                 return TRACE_TYPE_PARTIAL_LINE;
 694                 }
 695         }
 696
 697         /* No overhead */
 698         ret = print_graph_duration(0, s, flags | FLAGS_FILL_START);
 699         if (ret != TRACE_TYPE_HANDLED)
 700                 return ret;
 701
 702         if (type == TRACE_GRAPH_ENT)
 703                 ret = trace_seq_puts(s, "==========>");
 704         else
 705                 ret = trace_seq_puts(s, "<==========");
 706
 707         if (!ret)
 708                 return TRACE_TYPE_PARTIAL_LINE;
 709
 710         ret = print_graph_duration(0, s, flags | FLAGS_FILL_END);
 711         if (ret != TRACE_TYPE_HANDLED)
 712                 return ret;
 713
 714         ret = trace_seq_putc(s, '\n');
 715
 716         if (!ret)
 717                 return TRACE_TYPE_PARTIAL_LINE;
 718         return TRACE_TYPE_HANDLED;
 719 }
 720
 721 enum print_line_t
 722 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
 723 {
 724         unsigned long nsecs_rem = do_div(duration, 1000);
 725         /* log10(ULONG_MAX) + '\0' */
 726         char msecs_str[21];
 727         char nsecs_str[5];
 728         int ret, len;
 729         int i;
 730
 731         sprintf(msecs_str, "%lu", (unsigned long) duration);
 732
 733         /* Print msecs */
 734         ret = trace_seq_printf(s, "%s", msecs_str);
 735         if (!ret)
 736                 return TRACE_TYPE_PARTIAL_LINE;
 737
 738         len = strlen(msecs_str);
 739
 740         /* Print nsecs (we don't want to exceed 7 numbers) */
 741         if (len < 7) {
 742                 size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
 743
 744                 snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
 745                 ret = trace_seq_printf(s, ".%s", nsecs_str);
 746                 if (!ret)
 747                         return TRACE_TYPE_PARTIAL_LINE;
 748                 len += strlen(nsecs_str);
 749         }
 750
 751         ret = trace_seq_puts(s, " us ");
 752         if (!ret)
 753                 return TRACE_TYPE_PARTIAL_LINE;
 754
 755         /* Print remaining spaces to fit the row's width */
 756         for (i = len; i < 7; i++) {
 757                 ret = trace_seq_putc(s, ' ');
 758                 if (!ret)
 759                         return TRACE_TYPE_PARTIAL_LINE;
 760         }
 761         return TRACE_TYPE_HANDLED;
 762 }
 763
 764 static enum print_line_t
 765 print_graph_duration(unsigned long long duration, struct trace_seq *s,
 766                      u32 flags)
 767 {
 768         int ret = -1;
 769
 770         if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
 771             !(trace_flags & TRACE_ITER_CONTEXT_INFO))
 772                         return TRACE_TYPE_HANDLED;
 773
 774         /* No real adata, just filling the column with spaces */
 775         switch (flags & TRACE_GRAPH_PRINT_FILL_MASK) {
 776         case FLAGS_FILL_FULL:
 777                 ret = trace_seq_puts(s, "              |  ");
 778                 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 779         case FLAGS_FILL_START:
 780                 ret = trace_seq_puts(s, "  ");
 781                 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 782         case FLAGS_FILL_END:
 783                 ret = trace_seq_puts(s, " |");
 784                 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 785         }
 786
 787         /* Signal a overhead of time execution to the output */
 788         if (flags & TRACE_GRAPH_PRINT_OVERHEAD) {
 789                 /* Duration exceeded 100 msecs */
 790                 if (duration > 100000ULL)
 791                         ret = trace_seq_puts(s, "! ");
 792                 /* Duration exceeded 10 msecs */
 793                 else if (duration > 10000ULL)
 794                         ret = trace_seq_puts(s, "+ ");
 795         }
 796
 797         /*
 798          * The -1 means we either did not exceed the duration tresholds
 799          * or we dont want to print out the overhead. Either way we need
 800          * to fill out the space.
 801          */
 802         if (ret == -1)
 803                 ret = trace_seq_puts(s, "  ");
 804
 805         /* Catching here any failure happenned above */
 806         if (!ret)
 807                 return TRACE_TYPE_PARTIAL_LINE;
 808
 809         ret = trace_print_graph_duration(duration, s);
 810         if (ret != TRACE_TYPE_HANDLED)
 811                 return ret;
 812
 813         ret = trace_seq_puts(s, "|  ");
 814         if (!ret)
 815                 return TRACE_TYPE_PARTIAL_LINE;
 816
 817         return TRACE_TYPE_HANDLED;
 818 }
 819
 820 /* Case of a leaf function on its call entry */
 821 static enum print_line_t
 822 print_graph_entry_leaf(struct trace_iterator *iter,
 823                 struct ftrace_graph_ent_entry *entry,
 824                 struct ftrace_graph_ret_entry *ret_entry,
 825                 struct trace_seq *s, u32 flags)
 826 {
 827         struct fgraph_data *data = iter->private;
 828         struct ftrace_graph_ret *graph_ret;
 829         struct ftrace_graph_ent *call;
 830         unsigned long long duration;
 831         int ret;
 832         int i;
 833
 834         graph_ret = &ret_entry->ret;
 835         call = &entry->graph_ent;
 836         duration = graph_ret->rettime - graph_ret->calltime;
 837
 838         if (data) {
 839                 struct fgraph_cpu_data *cpu_data;
 840                 int cpu = iter->cpu;
 841
 842                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 843
 844                 /*
 845                  * Comments display at + 1 to depth. Since
 846                  * this is a leaf function, keep the comments
 847                  * equal to this depth.
 848                  */
 849                 cpu_data->depth = call->depth - 1;
 850
 851                 /* No need to keep this function around for this depth */
 852                 if (call->depth < FTRACE_RETFUNC_DEPTH)
 853                         cpu_data->enter_funcs[call->depth] = 0;
 854         }
 855
 856         /* Overhead and duration */
 857         ret = print_graph_duration(duration, s, flags);
 858         if (ret == TRACE_TYPE_PARTIAL_LINE)
 859                 return TRACE_TYPE_PARTIAL_LINE;
 860
 861         /* Function */
 862         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 863                 ret = trace_seq_putc(s, ' ');
 864                 if (!ret)
 865                         return TRACE_TYPE_PARTIAL_LINE;
 866         }
 867
 868         ret = trace_seq_printf(s, "%ps();\n", (void *)call->func);
 869         if (!ret)
 870                 return TRACE_TYPE_PARTIAL_LINE;
 871
 872         return TRACE_TYPE_HANDLED;
 873 }
 874
 875 static enum print_line_t
 876 print_graph_entry_nested(struct trace_iterator *iter,
 877                          struct ftrace_graph_ent_entry *entry,
 878                          struct trace_seq *s, int cpu, u32 flags)
 879 {
 880         struct ftrace_graph_ent *call = &entry->graph_ent;
 881         struct fgraph_data *data = iter->private;
 882         int ret;
 883         int i;
 884
 885         if (data) {
 886                 struct fgraph_cpu_data *cpu_data;
 887                 int cpu = iter->cpu;
 888
 889                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 890                 cpu_data->depth = call->depth;
 891
 892                 /* Save this function pointer to see if the exit matches */
 893                 if (call->depth < FTRACE_RETFUNC_DEPTH)
 894                         cpu_data->enter_funcs[call->depth] = call->func;
 895         }
 896
 897         /* No time */
 898         ret = print_graph_duration(0, s, flags | FLAGS_FILL_FULL);
 899         if (ret != TRACE_TYPE_HANDLED)
 900                 return ret;
 901
 902         /* Function */
 903         for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 904                 ret = trace_seq_putc(s, ' ');
 905                 if (!ret)
 906                         return TRACE_TYPE_PARTIAL_LINE;
 907         }
 908
 909         ret = trace_seq_printf(s, "%ps() {\n", (void *)call->func);
 910         if (!ret)
 911                 return TRACE_TYPE_PARTIAL_LINE;
 912
 913         /*
 914          * we already consumed the current entry to check the next one
 915          * and see if this is a leaf.
 916          */
 917         return TRACE_TYPE_NO_CONSUME;
 918 }
 919
 920 static enum print_line_t
 921 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
 922                      int type, unsigned long addr, u32 flags)
 923 {
 924         struct fgraph_data *data = iter->private;
 925         struct trace_entry *ent = iter->ent;
 926         int cpu = iter->cpu;
 927         int ret;
 928
 929         /* Pid */
 930         if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
 931                 return TRACE_TYPE_PARTIAL_LINE;
 932
 933         if (type) {
 934                 /* Interrupt */
 935                 ret = print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
 936                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 937                         return TRACE_TYPE_PARTIAL_LINE;
 938         }
 939
 940         if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
 941                 return 0;
 942
 943         /* Absolute time */
 944         if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
 945                 ret = print_graph_abs_time(iter->ts, s);
 946                 if (!ret)
 947                         return TRACE_TYPE_PARTIAL_LINE;
 948         }
 949
 950         /* Cpu */
 951         if (flags & TRACE_GRAPH_PRINT_CPU) {
 952                 ret = print_graph_cpu(s, cpu);
 953                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 954                         return TRACE_TYPE_PARTIAL_LINE;
 955         }
 956
 957         /* Proc */
 958         if (flags & TRACE_GRAPH_PRINT_PROC) {
 959                 ret = print_graph_proc(s, ent->pid);
 960                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 961                         return TRACE_TYPE_PARTIAL_LINE;
 962
 963                 ret = trace_seq_puts(s, " | ");
 964                 if (!ret)
 965                         return TRACE_TYPE_PARTIAL_LINE;
 966         }
 967
 968         /* Latency format */
 969         if (trace_flags & TRACE_ITER_LATENCY_FMT) {
 970                 ret = print_graph_lat_fmt(s, ent);
 971                 if (ret == TRACE_TYPE_PARTIAL_LINE)
 972                         return TRACE_TYPE_PARTIAL_LINE;
 973         }
 974
 975         return 0;
 976 }
 977
 978 /*
 979  * Entry check for irq code
 980  *
 981  * returns 1 if
 982  *  - we are inside irq code
 983  *  - we just entered irq code
 984  *
 985  * retunns 0 if
 986  *  - funcgraph-interrupts option is set
 987  *  - we are not inside irq code
 988  */
 989 static int
 990 check_irq_entry(struct trace_iterator *iter, u32 flags,
 991                 unsigned long addr, int depth)
 992 {
 993         int cpu = iter->cpu;
 994         int *depth_irq;
 995         struct fgraph_data *data = iter->private;
 996
 997         /*
 998          * If we are either displaying irqs, or we got called as
 999          * a graph event and private data does not exist,
1000          * then we bypass the irq check.
1001          */
1002         if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
1003             (!data))
1004                 return 0;
1005
1006         depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1007
1008         /*
1009          * We are inside the irq code
1010          */
1011         if (*depth_irq >= 0)
1012                 return 1;
1013
1014         if ((addr < (unsigned long)__irqentry_text_start) ||
1015             (addr >= (unsigned long)__irqentry_text_end))
1016                 return 0;
1017
1018         /*
1019          * We are entering irq code.
1020          */
1021         *depth_irq = depth;
1022         return 1;
1023 }
1024
1025 /*
1026  * Return check for irq code
1027  *
1028  * returns 1 if
1029  *  - we are inside irq code
1030  *  - we just left irq code
1031  *
1032  * returns 0 if
1033  *  - funcgraph-interrupts option is set
1034  *  - we are not inside irq code
1035  */
1036 static int
1037 check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
1038 {
1039         int cpu = iter->cpu;
1040         int *depth_irq;
1041         struct fgraph_data *data = iter->private;
1042
1043         /*
1044          * If we are either displaying irqs, or we got called as
1045          * a graph event and private data does not exist,
1046          * then we bypass the irq check.
1047          */
1048         if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
1049             (!data))
1050                 return 0;
1051
1052         depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1053
1054         /*
1055          * We are not inside the irq code.
1056          */
1057         if (*depth_irq == -1)
1058                 return 0;
1059
1060         /*
1061          * We are inside the irq code, and this is returning entry.
1062          * Let's not trace it and clear the entry depth, since
1063          * we are out of irq code.
1064          *
1065          * This condition ensures that we 'leave the irq code' once
1066          * we are out of the entry depth. Thus protecting us from
1067          * the RETURN entry loss.
1068          */
1069         if (*depth_irq >= depth) {
1070                 *depth_irq = -1;
1071                 return 1;
1072         }
1073
1074         /*
1075          * We are inside the irq code, and this is not the entry.
1076          */
1077         return 1;
1078 }
1079
1080 static enum print_line_t
1081 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
1082                         struct trace_iterator *iter, u32 flags)
1083 {
1084         struct fgraph_data *data = iter->private;
1085         struct ftrace_graph_ent *call = &field->graph_ent;
1086         struct ftrace_graph_ret_entry *leaf_ret;
1087         static enum print_line_t ret;
1088         int cpu = iter->cpu;
1089
1090         if (check_irq_entry(iter, flags, call->func, call->depth))
1091                 return TRACE_TYPE_HANDLED;
1092
1093         if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags))
1094                 return TRACE_TYPE_PARTIAL_LINE;
1095
1096         leaf_ret = get_return_for_leaf(iter, field);
1097         if (leaf_ret)
1098                 ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
1099         else
1100                 ret = print_graph_entry_nested(iter, field, s, cpu, flags);
1101
1102         if (data) {
1103                 /*
1104                  * If we failed to write our output, then we need to make
1105                  * note of it. Because we already consumed our entry.
1106                  */
1107                 if (s->full) {
1108                         data->failed = 1;
1109                         data->cpu = cpu;
1110                 } else
1111                         data->failed = 0;
1112         }
1113
1114         return ret;
1115 }
1116
1117 static enum print_line_t
1118 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
1119                    struct trace_entry *ent, struct trace_iterator *iter,
1120                    u32 flags)
1121 {
1122         unsigned long long duration = trace->rettime - trace->calltime;
1123         struct fgraph_data *data = iter->private;
1124         pid_t pid = ent->pid;
1125         int cpu = iter->cpu;
1126         int func_match = 1;
1127         int ret;
1128         int i;
1129
1130         if (check_irq_return(iter, flags, trace->depth))
1131                 return TRACE_TYPE_HANDLED;
1132
1133         if (data) {
1134                 struct fgraph_cpu_data *cpu_data;
1135                 int cpu = iter->cpu;
1136
1137                 cpu_data = per_cpu_ptr(data->cpu_data, cpu);
1138
1139                 /*
1140                  * Comments display at + 1 to depth. This is the
1141                  * return from a function, we now want the comments
1142                  * to display at the same level of the bracket.
1143                  */
1144                 cpu_data->depth = trace->depth - 1;
1145
1146                 if (trace->depth < FTRACE_RETFUNC_DEPTH) {
1147                         if (cpu_data->enter_funcs[trace->depth] != trace->func)
1148                                 func_match = 0;
1149                         cpu_data->enter_funcs[trace->depth] = 0;
1150                 }
1151         }
1152
1153         if (print_graph_prologue(iter, s, 0, 0, flags))
1154                 return TRACE_TYPE_PARTIAL_LINE;
1155
1156         /* Overhead and duration */
1157         ret = print_graph_duration(duration, s, flags);
1158         if (ret == TRACE_TYPE_PARTIAL_LINE)
1159                 return TRACE_TYPE_PARTIAL_LINE;
1160
1161         /* Closing brace */
1162         for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
1163                 ret = trace_seq_putc(s, ' ');
1164                 if (!ret)
1165                         return TRACE_TYPE_PARTIAL_LINE;
1166         }
1167
1168         /*
1169          * If the return function does not have a matching entry,
1170          * then the entry was lost. Instead of just printing
1171          * the '}' and letting the user guess what function this
1172          * belongs to, write out the function name. Always do
1173          * that if the funcgraph-tail option is enabled.
1174          */
1175         if (func_match && !(flags & TRACE_GRAPH_PRINT_TAIL)) {
1176                 ret = trace_seq_puts(s, "}\n");
1177                 if (!ret)
1178                         return TRACE_TYPE_PARTIAL_LINE;
1179         } else {
1180                 ret = trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
1181                 if (!ret)
1182                         return TRACE_TYPE_PARTIAL_LINE;
1183         }
1184
1185         /* Overrun */
1186         if (flags & TRACE_GRAPH_PRINT_OVERRUN) {
1187                 ret = trace_seq_printf(s, " (Overruns: %lu)\n",
1188                                         trace->overrun);
1189                 if (!ret)
1190                         return TRACE_TYPE_PARTIAL_LINE;
1191         }
1192
1193         ret = print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1194                               cpu, pid, flags);
1195         if (ret == TRACE_TYPE_PARTIAL_LINE)
1196                 return TRACE_TYPE_PARTIAL_LINE;
1197
1198         return TRACE_TYPE_HANDLED;
1199 }
1200
1201 static enum print_line_t
1202 print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1203                     struct trace_iterator *iter, u32 flags)
1204 {
1205         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1206         struct fgraph_data *data = iter->private;
1207         struct trace_event *event;
1208         int depth = 0;
1209         int ret;
1210         int i;
1211
1212         if (data)
1213                 depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1214
1215         if (print_graph_prologue(iter, s, 0, 0, flags))
1216                 return TRACE_TYPE_PARTIAL_LINE;
1217
1218         /* No time */
1219         ret = print_graph_duration(0, s, flags | FLAGS_FILL_FULL);
1220         if (ret != TRACE_TYPE_HANDLED)
1221                 return ret;
1222
1223         /* Indentation */
1224         if (depth > 0)
1225                 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++) {
1226                         ret = trace_seq_putc(s, ' ');
1227                         if (!ret)
1228                                 return TRACE_TYPE_PARTIAL_LINE;
1229                 }
1230
1231         /* The comment */
1232         ret = trace_seq_puts(s, "/* ");
1233         if (!ret)
1234                 return TRACE_TYPE_PARTIAL_LINE;
1235
1236         switch (iter->ent->type) {
1237         case TRACE_BPRINT:
1238                 ret = trace_print_bprintk_msg_only(iter);
1239                 if (ret != TRACE_TYPE_HANDLED)
1240                         return ret;
1241                 break;
1242         case TRACE_PRINT:
1243                 ret = trace_print_printk_msg_only(iter);
1244                 if (ret != TRACE_TYPE_HANDLED)
1245                         return ret;
1246                 break;
1247         default:
1248                 event = ftrace_find_event(ent->type);
1249                 if (!event)
1250                         return TRACE_TYPE_UNHANDLED;
1251
1252                 ret = event->funcs->trace(iter, sym_flags, event);
1253                 if (ret != TRACE_TYPE_HANDLED)
1254                         return ret;
1255         }
1256
1257         /* Strip ending newline */
1258         if (s->buffer[s->len - 1] == '\n') {
1259                 s->buffer[s->len - 1] = '\0';
1260                 s->len--;
1261         }
1262
1263         ret = trace_seq_puts(s, " */\n");
1264         if (!ret)
1265                 return TRACE_TYPE_PARTIAL_LINE;
1266
1267         return TRACE_TYPE_HANDLED;
1268 }
1269
1270
1271 enum print_line_t
1272 print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1273 {
1274         struct ftrace_graph_ent_entry *field;
1275         struct fgraph_data *data = iter->private;
1276         struct trace_entry *entry = iter->ent;
1277         struct trace_seq *s = &iter->seq;
1278         int cpu = iter->cpu;
1279         int ret;
1280
1281         if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1282                 per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1283                 return TRACE_TYPE_HANDLED;
1284         }
1285
1286         /*
1287          * If the last output failed, there's a possibility we need
1288          * to print out the missing entry which would never go out.
1289          */
1290         if (data && data->failed) {
1291                 field = &data->ent;
1292                 iter->cpu = data->cpu;
1293                 ret = print_graph_entry(field, s, iter, flags);
1294                 if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1295                         per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1296                         ret = TRACE_TYPE_NO_CONSUME;
1297                 }
1298                 iter->cpu = cpu;
1299                 return ret;
1300         }
1301
1302         switch (entry->type) {
1303         case TRACE_GRAPH_ENT: {
1304                 /*
1305                  * print_graph_entry() may consume the current event,
1306                  * thus @field may become invalid, so we need to save it.
1307                  * sizeof(struct ftrace_graph_ent_entry) is very small,
1308                  * it can be safely saved at the stack.
1309                  */
1310                 struct ftrace_graph_ent_entry saved;
1311                 trace_assign_type(field, entry);
1312                 saved = *field;
1313                 return print_graph_entry(&saved, s, iter, flags);
1314         }
1315         case TRACE_GRAPH_RET: {
1316                 struct ftrace_graph_ret_entry *field;
1317                 trace_assign_type(field, entry);
1318                 return print_graph_return(&field->ret, s, entry, iter, flags);
1319         }
1320         case TRACE_STACK:
1321         case TRACE_FN:
1322                 /* dont trace stack and functions as comments */
1323                 return TRACE_TYPE_UNHANDLED;
1324
1325         default:
1326                 return print_graph_comment(s, entry, iter, flags);
1327         }
1328
1329         return TRACE_TYPE_HANDLED;
1330 }
1331
1332 static enum print_line_t
1333 print_graph_function(struct trace_iterator *iter)
1334 {
1335         return print_graph_function_flags(iter, tracer_flags.val);
1336 }
1337
1338 static enum print_line_t
1339 print_graph_function_event(struct trace_iterator *iter, int flags,
1340                            struct trace_event *event)
1341 {
1342         return print_graph_function(iter);
1343 }
1344
1345 static void print_lat_header(struct seq_file *s, u32 flags)
1346 {
1347         static const char spaces[] = "                " /* 16 spaces */
1348                 "    "                                  /* 4 spaces */
1349                 "                 ";                    /* 17 spaces */
1350         int size = 0;
1351
1352         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1353                 size += 16;
1354         if (flags & TRACE_GRAPH_PRINT_CPU)
1355                 size += 4;
1356         if (flags & TRACE_GRAPH_PRINT_PROC)
1357                 size += 17;
1358
1359         seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1360         seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1361         seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1362         seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1363         seq_printf(s, "#%.*s||| /                      \n", size, spaces);
1364 }
1365
1366 static void __print_graph_headers_flags(struct seq_file *s, u32 flags)
1367 {
1368         int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
1369
1370         if (lat)
1371                 print_lat_header(s, flags);
1372
1373         /* 1st line */
1374         seq_printf(s, "#");
1375         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1376                 seq_printf(s, "     TIME       ");
1377         if (flags & TRACE_GRAPH_PRINT_CPU)
1378                 seq_printf(s, " CPU");
1379         if (flags & TRACE_GRAPH_PRINT_PROC)
1380                 seq_printf(s, "  TASK/PID       ");
1381         if (lat)
1382                 seq_printf(s, "||||");
1383         if (flags & TRACE_GRAPH_PRINT_DURATION)
1384                 seq_printf(s, "  DURATION   ");
1385         seq_printf(s, "               FUNCTION CALLS\n");
1386
1387         /* 2nd line */
1388         seq_printf(s, "#");
1389         if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1390                 seq_printf(s, "      |         ");
1391         if (flags & TRACE_GRAPH_PRINT_CPU)
1392                 seq_printf(s, " |  ");
1393         if (flags & TRACE_GRAPH_PRINT_PROC)
1394                 seq_printf(s, "   |    |        ");
1395         if (lat)
1396                 seq_printf(s, "||||");
1397         if (flags & TRACE_GRAPH_PRINT_DURATION)
1398                 seq_printf(s, "   |   |      ");
1399         seq_printf(s, "               |   |   |   |\n");
1400 }
1401
1402 void print_graph_headers(struct seq_file *s)
1403 {
1404         print_graph_headers_flags(s, tracer_flags.val);
1405 }
1406
1407 void print_graph_headers_flags(struct seq_file *s, u32 flags)
1408 {
1409         struct trace_iterator *iter = s->private;
1410
1411         if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
1412                 return;
1413
1414         if (trace_flags & TRACE_ITER_LATENCY_FMT) {
1415                 /* print nothing if the buffers are empty */
1416                 if (trace_empty(iter))
1417                         return;
1418
1419                 print_trace_header(s, iter);
1420         }
1421
1422         __print_graph_headers_flags(s, flags);
1423 }
1424
1425 void graph_trace_open(struct trace_iterator *iter)
1426 {
1427         /* pid and depth on the last trace processed */
1428         struct fgraph_data *data;
1429         int cpu;
1430
1431         iter->private = NULL;
1432
1433         data = kzalloc(sizeof(*data), GFP_KERNEL);
1434         if (!data)
1435                 goto out_err;
1436
1437         data->cpu_data = alloc_percpu(struct fgraph_cpu_data);
1438         if (!data->cpu_data)
1439                 goto out_err_free;
1440
1441         for_each_possible_cpu(cpu) {
1442                 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1443                 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1444                 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1445                 int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1446
1447                 *pid = -1;
1448                 *depth = 0;
1449                 *ignore = 0;
1450                 *depth_irq = -1;
1451         }
1452
1453         iter->private = data;
1454
1455         return;
1456
1457  out_err_free:
1458         kfree(data);
1459  out_err:
1460         pr_warning("function graph tracer: not enough memory\n");
1461 }
1462
1463 void graph_trace_close(struct trace_iterator *iter)
1464 {
1465         struct fgraph_data *data = iter->private;
1466
1467         if (data) {
1468                 free_percpu(data->cpu_data);
1469                 kfree(data);
1470         }
1471 }
1472
1473 static int
1474 func_graph_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1475 {
1476         if (bit == TRACE_GRAPH_PRINT_IRQS)
1477                 ftrace_graph_skip_irqs = !set;
1478
1479         return 0;
1480 }
1481
1482 static struct trace_event_functions graph_functions = {
1483         .trace          = print_graph_function_event,
1484 };
1485
1486 static struct trace_event graph_trace_entry_event = {
1487         .type           = TRACE_GRAPH_ENT,
1488         .funcs          = &graph_functions,
1489 };
1490
1491 static struct trace_event graph_trace_ret_event = {
1492         .type           = TRACE_GRAPH_RET,
1493         .funcs          = &graph_functions
1494 };
1495
1496 static struct tracer graph_trace __tracer_data = {
1497         .name           = "function_graph",
1498         .open           = graph_trace_open,
1499         .pipe_open      = graph_trace_open,
1500         .close          = graph_trace_close,
1501         .pipe_close     = graph_trace_close,
1502         .init           = graph_trace_init,
1503         .reset          = graph_trace_reset,
1504         .print_line     = print_graph_function,
1505         .print_header   = print_graph_headers,
1506         .flags          = &tracer_flags,
1507         .set_flag       = func_graph_set_flag,
1508 #ifdef CONFIG_FTRACE_SELFTEST
1509         .selftest       = trace_selftest_startup_function_graph,
1510 #endif
1511 };
1512
1513
1514 static ssize_t
1515 graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1516                   loff_t *ppos)
1517 {
1518         unsigned long val;
1519         int ret;
1520
1521         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1522         if (ret)
1523                 return ret;
1524
1525         max_depth = val;
1526
1527         *ppos += cnt;
1528
1529         return cnt;
1530 }
1531
1532 static ssize_t
1533 graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1534                  loff_t *ppos)
1535 {
1536         char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1537         int n;
1538
1539         n = sprintf(buf, "%d\n", max_depth);
1540
1541         return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1542 }
1543
1544 static const struct file_operations graph_depth_fops = {
1545         .open           = tracing_open_generic,
1546         .write          = graph_depth_write,
1547         .read           = graph_depth_read,
1548         .llseek         = generic_file_llseek,
1549 };
1550
1551 static __init int init_graph_debugfs(void)
1552 {
1553         struct dentry *d_tracer;
1554
1555         d_tracer = tracing_init_dentry();
1556         if (!d_tracer)
1557                 return 0;
1558
1559         trace_create_file("max_graph_depth", 0644, d_tracer,
1560                           NULL, &graph_depth_fops);
1561
1562         return 0;
1563 }
1564 fs_initcall(init_graph_debugfs);
1565
1566 static __init int init_graph_trace(void)
1567 {
1568         max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1569
1570         if (!register_ftrace_event(&graph_trace_entry_event)) {
1571                 pr_warning("Warning: could not register graph trace events\n");
1572                 return 1;
1573         }
1574
1575         if (!register_ftrace_event(&graph_trace_ret_event)) {
1576                 pr_warning("Warning: could not register graph trace events\n");
1577                 return 1;
1578         }
1579
1580         return register_tracer(&graph_trace);
1581 }
1582
1583 core_initcall(init_graph_trace);