Linux Audio

Check our new training course

Loading...
v4.6
   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/interrupt.h>
  12#include <linux/slab.h>
  13#include <linux/fs.h>
  14
  15#include "trace.h"
  16#include "trace_output.h"
  17
  18static bool kill_ftrace_graph;
  19
  20/**
  21 * ftrace_graph_is_dead - returns true if ftrace_graph_stop() was called
  22 *
  23 * ftrace_graph_stop() is called when a severe error is detected in
  24 * the function graph tracing. This function is called by the critical
  25 * paths of function graph to keep those paths from doing any more harm.
  26 */
  27bool ftrace_graph_is_dead(void)
  28{
  29	return kill_ftrace_graph;
  30}
  31
  32/**
  33 * ftrace_graph_stop - set to permanently disable function graph tracincg
  34 *
  35 * In case of an error int function graph tracing, this is called
  36 * to try to keep function graph tracing from causing any more harm.
  37 * Usually this is pretty severe and this is called to try to at least
  38 * get a warning out to the user.
  39 */
  40void ftrace_graph_stop(void)
  41{
  42	kill_ftrace_graph = true;
  43}
  44
  45/* When set, irq functions will be ignored */
  46static int ftrace_graph_skip_irqs;
  47
  48struct fgraph_cpu_data {
  49	pid_t		last_pid;
  50	int		depth;
  51	int		depth_irq;
  52	int		ignore;
  53	unsigned long	enter_funcs[FTRACE_RETFUNC_DEPTH];
  54};
  55
  56struct fgraph_data {
  57	struct fgraph_cpu_data __percpu *cpu_data;
  58
  59	/* Place to preserve last processed entry. */
  60	struct ftrace_graph_ent_entry	ent;
  61	struct ftrace_graph_ret_entry	ret;
  62	int				failed;
  63	int				cpu;
  64};
  65
  66#define TRACE_GRAPH_INDENT	2
  67
  68static unsigned int max_depth;
 
 
 
 
 
 
 
  69
  70static struct tracer_opt trace_opts[] = {
  71	/* Display overruns? (for self-debug purpose) */
  72	{ TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
  73	/* Display CPU ? */
  74	{ TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
  75	/* Display Overhead ? */
  76	{ TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
  77	/* Display proc name/pid */
  78	{ TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
  79	/* Display duration of execution */
  80	{ TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
  81	/* Display absolute time of an entry */
  82	{ TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
  83	/* Display interrupts */
  84	{ TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
  85	/* Display function name after trailing } */
  86	{ TRACER_OPT(funcgraph-tail, TRACE_GRAPH_PRINT_TAIL) },
  87	/* Include sleep time (scheduled out) between entry and return */
  88	{ TRACER_OPT(sleep-time, TRACE_GRAPH_SLEEP_TIME) },
  89	/* Include time within nested functions */
  90	{ TRACER_OPT(graph-time, TRACE_GRAPH_GRAPH_TIME) },
  91	{ } /* Empty entry */
  92};
  93
  94static struct tracer_flags tracer_flags = {
  95	/* Don't display overruns, proc, or tail by default */
  96	.val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
  97	       TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS |
  98	       TRACE_GRAPH_SLEEP_TIME | TRACE_GRAPH_GRAPH_TIME,
  99	.opts = trace_opts
 100};
 101
 102static struct trace_array *graph_array;
 103
 104/*
 105 * DURATION column is being also used to display IRQ signs,
 106 * following values are used by print_graph_irq and others
 107 * to fill in space into DURATION column.
 108 */
 109enum {
 110	FLAGS_FILL_FULL  = 1 << TRACE_GRAPH_PRINT_FILL_SHIFT,
 111	FLAGS_FILL_START = 2 << TRACE_GRAPH_PRINT_FILL_SHIFT,
 112	FLAGS_FILL_END   = 3 << TRACE_GRAPH_PRINT_FILL_SHIFT,
 113};
 114
 115static void
 116print_graph_duration(struct trace_array *tr, unsigned long long duration,
 117		     struct trace_seq *s, u32 flags);
 118
 119/* Add a function return address to the trace stack on thread info.*/
 120int
 121ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
 122			 unsigned long frame_pointer)
 123{
 124	unsigned long long calltime;
 125	int index;
 126
 127	if (unlikely(ftrace_graph_is_dead()))
 128		return -EBUSY;
 129
 130	if (!current->ret_stack)
 131		return -EBUSY;
 132
 133	/*
 134	 * We must make sure the ret_stack is tested before we read
 135	 * anything else.
 136	 */
 137	smp_rmb();
 138
 139	/* The return trace stack is full */
 140	if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
 141		atomic_inc(&current->trace_overrun);
 142		return -EBUSY;
 143	}
 144
 145	/*
 146	 * The curr_ret_stack is an index to ftrace return stack of
 147	 * current task.  Its value should be in [0, FTRACE_RETFUNC_
 148	 * DEPTH) when the function graph tracer is used.  To support
 149	 * filtering out specific functions, it makes the index
 150	 * negative by subtracting huge value (FTRACE_NOTRACE_DEPTH)
 151	 * so when it sees a negative index the ftrace will ignore
 152	 * the record.  And the index gets recovered when returning
 153	 * from the filtered function by adding the FTRACE_NOTRACE_
 154	 * DEPTH and then it'll continue to record functions normally.
 155	 *
 156	 * The curr_ret_stack is initialized to -1 and get increased
 157	 * in this function.  So it can be less than -1 only if it was
 158	 * filtered out via ftrace_graph_notrace_addr() which can be
 159	 * set from set_graph_notrace file in tracefs by user.
 160	 */
 161	if (current->curr_ret_stack < -1)
 162		return -EBUSY;
 163
 164	calltime = trace_clock_local();
 165
 166	index = ++current->curr_ret_stack;
 167	if (ftrace_graph_notrace_addr(func))
 168		current->curr_ret_stack -= FTRACE_NOTRACE_DEPTH;
 169	barrier();
 170	current->ret_stack[index].ret = ret;
 171	current->ret_stack[index].func = func;
 172	current->ret_stack[index].calltime = calltime;
 173	current->ret_stack[index].subtime = 0;
 174	current->ret_stack[index].fp = frame_pointer;
 175	*depth = current->curr_ret_stack;
 176
 177	return 0;
 178}
 179
 180/* Retrieve a function return address to the trace stack on thread info.*/
 181static void
 182ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
 183			unsigned long frame_pointer)
 184{
 185	int index;
 186
 187	index = current->curr_ret_stack;
 188
 189	/*
 190	 * A negative index here means that it's just returned from a
 191	 * notrace'd function.  Recover index to get an original
 192	 * return address.  See ftrace_push_return_trace().
 193	 *
 194	 * TODO: Need to check whether the stack gets corrupted.
 195	 */
 196	if (index < 0)
 197		index += FTRACE_NOTRACE_DEPTH;
 198
 199	if (unlikely(index < 0 || index >= FTRACE_RETFUNC_DEPTH)) {
 200		ftrace_graph_stop();
 201		WARN_ON(1);
 202		/* Might as well panic, otherwise we have no where to go */
 203		*ret = (unsigned long)panic;
 204		return;
 205	}
 206
 207#if defined(CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST) && !defined(CC_USING_FENTRY)
 208	/*
 209	 * The arch may choose to record the frame pointer used
 210	 * and check it here to make sure that it is what we expect it
 211	 * to be. If gcc does not set the place holder of the return
 212	 * address in the frame pointer, and does a copy instead, then
 213	 * the function graph trace will fail. This test detects this
 214	 * case.
 215	 *
 216	 * Currently, x86_32 with optimize for size (-Os) makes the latest
 217	 * gcc do the above.
 218	 *
 219	 * Note, -mfentry does not use frame pointers, and this test
 220	 *  is not needed if CC_USING_FENTRY is set.
 221	 */
 222	if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
 223		ftrace_graph_stop();
 224		WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
 225		     "  from func %ps return to %lx\n",
 226		     current->ret_stack[index].fp,
 227		     frame_pointer,
 228		     (void *)current->ret_stack[index].func,
 229		     current->ret_stack[index].ret);
 230		*ret = (unsigned long)panic;
 231		return;
 232	}
 233#endif
 234
 235	*ret = current->ret_stack[index].ret;
 236	trace->func = current->ret_stack[index].func;
 237	trace->calltime = current->ret_stack[index].calltime;
 238	trace->overrun = atomic_read(&current->trace_overrun);
 239	trace->depth = index;
 240}
 241
 242/*
 243 * Send the trace to the ring-buffer.
 244 * @return the original return address.
 245 */
 246unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
 247{
 248	struct ftrace_graph_ret trace;
 249	unsigned long ret;
 250
 251	ftrace_pop_return_trace(&trace, &ret, frame_pointer);
 252	trace.rettime = trace_clock_local();
 
 253	barrier();
 254	current->curr_ret_stack--;
 255	/*
 256	 * The curr_ret_stack can be less than -1 only if it was
 257	 * filtered out and it's about to return from the function.
 258	 * Recover the index and continue to trace normal functions.
 259	 */
 260	if (current->curr_ret_stack < -1) {
 261		current->curr_ret_stack += FTRACE_NOTRACE_DEPTH;
 262		return ret;
 263	}
 264
 265	/*
 266	 * The trace should run after decrementing the ret counter
 267	 * in case an interrupt were to come in. We don't want to
 268	 * lose the interrupt if max_depth is set.
 269	 */
 270	ftrace_graph_return(&trace);
 271
 272	if (unlikely(!ret)) {
 273		ftrace_graph_stop();
 274		WARN_ON(1);
 275		/* Might as well panic. What else to do? */
 276		ret = (unsigned long)panic;
 277	}
 278
 279	return ret;
 280}
 281
 282int __trace_graph_entry(struct trace_array *tr,
 283				struct ftrace_graph_ent *trace,
 284				unsigned long flags,
 285				int pc)
 286{
 287	struct trace_event_call *call = &event_funcgraph_entry;
 288	struct ring_buffer_event *event;
 289	struct ring_buffer *buffer = tr->trace_buffer.buffer;
 290	struct ftrace_graph_ent_entry *entry;
 291
 
 
 
 292	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
 293					  sizeof(*entry), flags, pc);
 294	if (!event)
 295		return 0;
 296	entry	= ring_buffer_event_data(event);
 297	entry->graph_ent			= *trace;
 298	if (!call_filter_check_discard(call, entry, buffer, event))
 299		__buffer_unlock_commit(buffer, event);
 300
 301	return 1;
 302}
 303
 304static inline int ftrace_graph_ignore_irqs(void)
 305{
 306	if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
 307		return 0;
 308
 309	return in_irq();
 310}
 311
 312int trace_graph_entry(struct ftrace_graph_ent *trace)
 313{
 314	struct trace_array *tr = graph_array;
 315	struct trace_array_cpu *data;
 316	unsigned long flags;
 317	long disabled;
 318	int ret;
 319	int cpu;
 320	int pc;
 321
 322	if (!ftrace_trace_task(current))
 323		return 0;
 324
 325	/* trace it when it is-nested-in or is a function enabled. */
 326	if ((!(trace->depth || ftrace_graph_addr(trace->func)) ||
 327	     ftrace_graph_ignore_irqs()) || (trace->depth < 0) ||
 328	    (max_depth && trace->depth >= max_depth))
 329		return 0;
 330
 331	/*
 332	 * Do not trace a function if it's filtered by set_graph_notrace.
 333	 * Make the index of ret stack negative to indicate that it should
 334	 * ignore further functions.  But it needs its own ret stack entry
 335	 * to recover the original index in order to continue tracing after
 336	 * returning from the function.
 337	 */
 338	if (ftrace_graph_notrace_addr(trace->func))
 339		return 1;
 340
 341	local_irq_save(flags);
 342	cpu = raw_smp_processor_id();
 343	data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 344	disabled = atomic_inc_return(&data->disabled);
 345	if (likely(disabled == 1)) {
 346		pc = preempt_count();
 347		ret = __trace_graph_entry(tr, trace, flags, pc);
 348	} else {
 349		ret = 0;
 350	}
 351
 352	atomic_dec(&data->disabled);
 353	local_irq_restore(flags);
 354
 355	return ret;
 356}
 357
 358static int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
 359{
 360	if (tracing_thresh)
 361		return 1;
 362	else
 363		return trace_graph_entry(trace);
 364}
 365
 366static void
 367__trace_graph_function(struct trace_array *tr,
 368		unsigned long ip, unsigned long flags, int pc)
 369{
 370	u64 time = trace_clock_local();
 371	struct ftrace_graph_ent ent = {
 372		.func  = ip,
 373		.depth = 0,
 374	};
 375	struct ftrace_graph_ret ret = {
 376		.func     = ip,
 377		.depth    = 0,
 378		.calltime = time,
 379		.rettime  = time,
 380	};
 381
 382	__trace_graph_entry(tr, &ent, flags, pc);
 383	__trace_graph_return(tr, &ret, flags, pc);
 384}
 385
 386void
 387trace_graph_function(struct trace_array *tr,
 388		unsigned long ip, unsigned long parent_ip,
 389		unsigned long flags, int pc)
 390{
 391	__trace_graph_function(tr, ip, flags, pc);
 392}
 393
 394void __trace_graph_return(struct trace_array *tr,
 395				struct ftrace_graph_ret *trace,
 396				unsigned long flags,
 397				int pc)
 398{
 399	struct trace_event_call *call = &event_funcgraph_exit;
 400	struct ring_buffer_event *event;
 401	struct ring_buffer *buffer = tr->trace_buffer.buffer;
 402	struct ftrace_graph_ret_entry *entry;
 403
 
 
 
 404	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
 405					  sizeof(*entry), flags, pc);
 406	if (!event)
 407		return;
 408	entry	= ring_buffer_event_data(event);
 409	entry->ret				= *trace;
 410	if (!call_filter_check_discard(call, entry, buffer, event))
 411		__buffer_unlock_commit(buffer, event);
 412}
 413
 414void trace_graph_return(struct ftrace_graph_ret *trace)
 415{
 416	struct trace_array *tr = graph_array;
 417	struct trace_array_cpu *data;
 418	unsigned long flags;
 419	long disabled;
 420	int cpu;
 421	int pc;
 422
 423	local_irq_save(flags);
 424	cpu = raw_smp_processor_id();
 425	data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 426	disabled = atomic_inc_return(&data->disabled);
 427	if (likely(disabled == 1)) {
 428		pc = preempt_count();
 429		__trace_graph_return(tr, trace, flags, pc);
 430	}
 431	atomic_dec(&data->disabled);
 432	local_irq_restore(flags);
 433}
 434
 435void set_graph_array(struct trace_array *tr)
 436{
 437	graph_array = tr;
 438
 439	/* Make graph_array visible before we start tracing */
 440
 441	smp_mb();
 442}
 443
 444static void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
 445{
 446	if (tracing_thresh &&
 447	    (trace->rettime - trace->calltime < tracing_thresh))
 448		return;
 449	else
 450		trace_graph_return(trace);
 451}
 452
 453static int graph_trace_init(struct trace_array *tr)
 454{
 455	int ret;
 456
 457	set_graph_array(tr);
 458	if (tracing_thresh)
 459		ret = register_ftrace_graph(&trace_graph_thresh_return,
 460					    &trace_graph_thresh_entry);
 461	else
 462		ret = register_ftrace_graph(&trace_graph_return,
 463					    &trace_graph_entry);
 464	if (ret)
 465		return ret;
 466	tracing_start_cmdline_record();
 467
 468	return 0;
 469}
 470
 471static void graph_trace_reset(struct trace_array *tr)
 472{
 473	tracing_stop_cmdline_record();
 474	unregister_ftrace_graph();
 475}
 476
 477static int graph_trace_update_thresh(struct trace_array *tr)
 478{
 479	graph_trace_reset(tr);
 480	return graph_trace_init(tr);
 481}
 482
 483static int max_bytes_for_cpu;
 484
 485static void print_graph_cpu(struct trace_seq *s, int cpu)
 
 486{
 
 
 487	/*
 488	 * Start with a space character - to make it stand out
 489	 * to the right a bit when trace output is pasted into
 490	 * email:
 491	 */
 492	trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
 
 
 
 
 493}
 494
 495#define TRACE_GRAPH_PROCINFO_LENGTH	14
 496
 497static void print_graph_proc(struct trace_seq *s, pid_t pid)
 
 498{
 499	char comm[TASK_COMM_LEN];
 500	/* sign + log10(MAX_INT) + '\0' */
 501	char pid_str[11];
 502	int spaces = 0;
 
 503	int len;
 504	int i;
 505
 506	trace_find_cmdline(pid, comm);
 507	comm[7] = '\0';
 508	sprintf(pid_str, "%d", pid);
 509
 510	/* 1 stands for the "-" character */
 511	len = strlen(comm) + strlen(pid_str) + 1;
 512
 513	if (len < TRACE_GRAPH_PROCINFO_LENGTH)
 514		spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
 515
 516	/* First spaces to align center */
 517	for (i = 0; i < spaces / 2; i++)
 518		trace_seq_putc(s, ' ');
 
 
 
 519
 520	trace_seq_printf(s, "%s-%s", comm, pid_str);
 
 
 521
 522	/* Last spaces to align center */
 523	for (i = 0; i < spaces - (spaces / 2); i++)
 524		trace_seq_putc(s, ' ');
 
 
 
 
 525}
 526
 527
 528static void print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
 
 529{
 530	trace_seq_putc(s, ' ');
 531	trace_print_lat_fmt(s, entry);
 
 
 532}
 533
 534/* If the pid changed since the last trace, output this event */
 535static void
 536verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
 537{
 538	pid_t prev_pid;
 539	pid_t *last_pid;
 
 540
 541	if (!data)
 542		return;
 543
 544	last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
 545
 546	if (*last_pid == pid)
 547		return;
 548
 549	prev_pid = *last_pid;
 550	*last_pid = pid;
 551
 552	if (prev_pid == -1)
 553		return;
 554/*
 555 * Context-switch trace line:
 556
 557 ------------------------------------------
 558 | 1)  migration/0--1  =>  sshd-1755
 559 ------------------------------------------
 560
 561 */
 562	trace_seq_puts(s, " ------------------------------------------\n");
 563	print_graph_cpu(s, cpu);
 564	print_graph_proc(s, prev_pid);
 565	trace_seq_puts(s, " => ");
 566	print_graph_proc(s, pid);
 567	trace_seq_puts(s, "\n ------------------------------------------\n\n");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 568}
 569
 570static struct ftrace_graph_ret_entry *
 571get_return_for_leaf(struct trace_iterator *iter,
 572		struct ftrace_graph_ent_entry *curr)
 573{
 574	struct fgraph_data *data = iter->private;
 575	struct ring_buffer_iter *ring_iter = NULL;
 576	struct ring_buffer_event *event;
 577	struct ftrace_graph_ret_entry *next;
 578
 579	/*
 580	 * If the previous output failed to write to the seq buffer,
 581	 * then we just reuse the data from before.
 582	 */
 583	if (data && data->failed) {
 584		curr = &data->ent;
 585		next = &data->ret;
 586	} else {
 587
 588		ring_iter = trace_buffer_iter(iter, iter->cpu);
 589
 590		/* First peek to compare current entry and the next one */
 591		if (ring_iter)
 592			event = ring_buffer_iter_peek(ring_iter, NULL);
 593		else {
 594			/*
 595			 * We need to consume the current entry to see
 596			 * the next one.
 597			 */
 598			ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu,
 599					    NULL, NULL);
 600			event = ring_buffer_peek(iter->trace_buffer->buffer, iter->cpu,
 601						 NULL, NULL);
 602		}
 603
 604		if (!event)
 605			return NULL;
 606
 607		next = ring_buffer_event_data(event);
 608
 609		if (data) {
 610			/*
 611			 * Save current and next entries for later reference
 612			 * if the output fails.
 613			 */
 614			data->ent = *curr;
 615			/*
 616			 * If the next event is not a return type, then
 617			 * we only care about what type it is. Otherwise we can
 618			 * safely copy the entire event.
 619			 */
 620			if (next->ent.type == TRACE_GRAPH_RET)
 621				data->ret = *next;
 622			else
 623				data->ret.ent.type = next->ent.type;
 624		}
 625	}
 626
 627	if (next->ent.type != TRACE_GRAPH_RET)
 628		return NULL;
 629
 630	if (curr->ent.pid != next->ent.pid ||
 631			curr->graph_ent.func != next->ret.func)
 632		return NULL;
 633
 634	/* this is a leaf, now advance the iterator */
 635	if (ring_iter)
 636		ring_buffer_read(ring_iter, NULL);
 637
 638	return next;
 639}
 640
 641static void print_graph_abs_time(u64 t, struct trace_seq *s)
 642{
 643	unsigned long usecs_rem;
 644
 645	usecs_rem = do_div(t, NSEC_PER_SEC);
 646	usecs_rem /= 1000;
 647
 648	trace_seq_printf(s, "%5lu.%06lu |  ",
 649			 (unsigned long)t, usecs_rem);
 650}
 651
 652static void
 653print_graph_irq(struct trace_iterator *iter, unsigned long addr,
 654		enum trace_type type, int cpu, pid_t pid, u32 flags)
 655{
 656	struct trace_array *tr = iter->tr;
 657	struct trace_seq *s = &iter->seq;
 658	struct trace_entry *ent = iter->ent;
 659
 660	if (addr < (unsigned long)__irqentry_text_start ||
 661		addr >= (unsigned long)__irqentry_text_end)
 662		return;
 663
 664	if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
 665		/* Absolute time */
 666		if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
 667			print_graph_abs_time(iter->ts, s);
 
 
 
 668
 669		/* Cpu */
 670		if (flags & TRACE_GRAPH_PRINT_CPU)
 671			print_graph_cpu(s, cpu);
 
 
 
 672
 673		/* Proc */
 674		if (flags & TRACE_GRAPH_PRINT_PROC) {
 675			print_graph_proc(s, pid);
 676			trace_seq_puts(s, " | ");
 
 
 
 
 677		}
 678
 679		/* Latency format */
 680		if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
 681			print_graph_lat_fmt(s, ent);
 682	}
 683
 684	/* No overhead */
 685	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_START);
 
 
 686
 687	if (type == TRACE_GRAPH_ENT)
 688		trace_seq_puts(s, "==========>");
 689	else
 690		trace_seq_puts(s, "<==========");
 
 
 
 
 
 
 
 691
 692	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_END);
 693	trace_seq_putc(s, '\n');
 
 
 
 694}
 695
 696void
 697trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
 698{
 699	unsigned long nsecs_rem = do_div(duration, 1000);
 700	/* log10(ULONG_MAX) + '\0' */
 701	char usecs_str[21];
 702	char nsecs_str[5];
 703	int len;
 704	int i;
 705
 706	sprintf(usecs_str, "%lu", (unsigned long) duration);
 707
 708	/* Print msecs */
 709	trace_seq_printf(s, "%s", usecs_str);
 
 
 710
 711	len = strlen(usecs_str);
 712
 713	/* Print nsecs (we don't want to exceed 7 numbers) */
 714	if (len < 7) {
 715		size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
 716
 717		snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
 718		trace_seq_printf(s, ".%s", nsecs_str);
 719		len += strlen(nsecs_str) + 1;
 
 
 720	}
 721
 722	trace_seq_puts(s, " us ");
 
 
 723
 724	/* Print remaining spaces to fit the row's width */
 725	for (i = len; i < 8; i++)
 726		trace_seq_putc(s, ' ');
 
 
 
 
 727}
 728
 729static void
 730print_graph_duration(struct trace_array *tr, unsigned long long duration,
 731		     struct trace_seq *s, u32 flags)
 732{
 
 
 733	if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
 734	    !(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
 735		return;
 736
 737	/* No real adata, just filling the column with spaces */
 738	switch (flags & TRACE_GRAPH_PRINT_FILL_MASK) {
 739	case FLAGS_FILL_FULL:
 740		trace_seq_puts(s, "              |  ");
 741		return;
 742	case FLAGS_FILL_START:
 743		trace_seq_puts(s, "  ");
 744		return;
 745	case FLAGS_FILL_END:
 746		trace_seq_puts(s, " |");
 747		return;
 748	}
 749
 750	/* Signal a overhead of time execution to the output */
 751	if (flags & TRACE_GRAPH_PRINT_OVERHEAD)
 752		trace_seq_printf(s, "%c ", trace_find_mark(duration));
 753	else
 754		trace_seq_puts(s, "  ");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 755
 756	trace_print_graph_duration(duration, s);
 757	trace_seq_puts(s, "|  ");
 758}
 759
 760/* Case of a leaf function on its call entry */
 761static enum print_line_t
 762print_graph_entry_leaf(struct trace_iterator *iter,
 763		struct ftrace_graph_ent_entry *entry,
 764		struct ftrace_graph_ret_entry *ret_entry,
 765		struct trace_seq *s, u32 flags)
 766{
 767	struct fgraph_data *data = iter->private;
 768	struct trace_array *tr = iter->tr;
 769	struct ftrace_graph_ret *graph_ret;
 770	struct ftrace_graph_ent *call;
 771	unsigned long long duration;
 
 772	int i;
 773
 774	graph_ret = &ret_entry->ret;
 775	call = &entry->graph_ent;
 776	duration = graph_ret->rettime - graph_ret->calltime;
 777
 778	if (data) {
 779		struct fgraph_cpu_data *cpu_data;
 780		int cpu = iter->cpu;
 781
 782		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 783
 784		/*
 785		 * Comments display at + 1 to depth. Since
 786		 * this is a leaf function, keep the comments
 787		 * equal to this depth.
 788		 */
 789		cpu_data->depth = call->depth - 1;
 790
 791		/* No need to keep this function around for this depth */
 792		if (call->depth < FTRACE_RETFUNC_DEPTH)
 793			cpu_data->enter_funcs[call->depth] = 0;
 794	}
 795
 796	/* Overhead and duration */
 797	print_graph_duration(tr, duration, s, flags);
 
 
 798
 799	/* Function */
 800	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
 801		trace_seq_putc(s, ' ');
 
 
 
 802
 803	trace_seq_printf(s, "%ps();\n", (void *)call->func);
 
 
 804
 805	return trace_handle_return(s);
 806}
 807
 808static enum print_line_t
 809print_graph_entry_nested(struct trace_iterator *iter,
 810			 struct ftrace_graph_ent_entry *entry,
 811			 struct trace_seq *s, int cpu, u32 flags)
 812{
 813	struct ftrace_graph_ent *call = &entry->graph_ent;
 814	struct fgraph_data *data = iter->private;
 815	struct trace_array *tr = iter->tr;
 816	int i;
 817
 818	if (data) {
 819		struct fgraph_cpu_data *cpu_data;
 820		int cpu = iter->cpu;
 821
 822		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 823		cpu_data->depth = call->depth;
 824
 825		/* Save this function pointer to see if the exit matches */
 826		if (call->depth < FTRACE_RETFUNC_DEPTH)
 827			cpu_data->enter_funcs[call->depth] = call->func;
 828	}
 829
 830	/* No time */
 831	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
 
 
 832
 833	/* Function */
 834	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
 835		trace_seq_putc(s, ' ');
 836
 837	trace_seq_printf(s, "%ps() {\n", (void *)call->func);
 
 838
 839	if (trace_seq_has_overflowed(s))
 
 840		return TRACE_TYPE_PARTIAL_LINE;
 841
 842	/*
 843	 * we already consumed the current entry to check the next one
 844	 * and see if this is a leaf.
 845	 */
 846	return TRACE_TYPE_NO_CONSUME;
 847}
 848
 849static void
 850print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
 851		     int type, unsigned long addr, u32 flags)
 852{
 853	struct fgraph_data *data = iter->private;
 854	struct trace_entry *ent = iter->ent;
 855	struct trace_array *tr = iter->tr;
 856	int cpu = iter->cpu;
 
 857
 858	/* Pid */
 859	verif_pid(s, ent->pid, cpu, data);
 
 860
 861	if (type)
 862		/* Interrupt */
 863		print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
 
 
 
 864
 865	if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
 866		return;
 867
 868	/* Absolute time */
 869	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
 870		print_graph_abs_time(iter->ts, s);
 
 
 
 871
 872	/* Cpu */
 873	if (flags & TRACE_GRAPH_PRINT_CPU)
 874		print_graph_cpu(s, cpu);
 
 
 
 875
 876	/* Proc */
 877	if (flags & TRACE_GRAPH_PRINT_PROC) {
 878		print_graph_proc(s, ent->pid);
 879		trace_seq_puts(s, " | ");
 
 
 
 
 
 880	}
 881
 882	/* Latency format */
 883	if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
 884		print_graph_lat_fmt(s, ent);
 
 
 
 885
 886	return;
 887}
 888
 889/*
 890 * Entry check for irq code
 891 *
 892 * returns 1 if
 893 *  - we are inside irq code
 894 *  - we just entered irq code
 895 *
 896 * retunns 0 if
 897 *  - funcgraph-interrupts option is set
 898 *  - we are not inside irq code
 899 */
 900static int
 901check_irq_entry(struct trace_iterator *iter, u32 flags,
 902		unsigned long addr, int depth)
 903{
 904	int cpu = iter->cpu;
 905	int *depth_irq;
 906	struct fgraph_data *data = iter->private;
 907
 908	/*
 909	 * If we are either displaying irqs, or we got called as
 910	 * a graph event and private data does not exist,
 911	 * then we bypass the irq check.
 912	 */
 913	if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
 914	    (!data))
 915		return 0;
 916
 917	depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
 918
 919	/*
 920	 * We are inside the irq code
 921	 */
 922	if (*depth_irq >= 0)
 923		return 1;
 924
 925	if ((addr < (unsigned long)__irqentry_text_start) ||
 926	    (addr >= (unsigned long)__irqentry_text_end))
 927		return 0;
 928
 929	/*
 930	 * We are entering irq code.
 931	 */
 932	*depth_irq = depth;
 933	return 1;
 934}
 935
 936/*
 937 * Return check for irq code
 938 *
 939 * returns 1 if
 940 *  - we are inside irq code
 941 *  - we just left irq code
 942 *
 943 * returns 0 if
 944 *  - funcgraph-interrupts option is set
 945 *  - we are not inside irq code
 946 */
 947static int
 948check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
 949{
 950	int cpu = iter->cpu;
 951	int *depth_irq;
 952	struct fgraph_data *data = iter->private;
 953
 954	/*
 955	 * If we are either displaying irqs, or we got called as
 956	 * a graph event and private data does not exist,
 957	 * then we bypass the irq check.
 958	 */
 959	if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
 960	    (!data))
 961		return 0;
 962
 963	depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
 964
 965	/*
 966	 * We are not inside the irq code.
 967	 */
 968	if (*depth_irq == -1)
 969		return 0;
 970
 971	/*
 972	 * We are inside the irq code, and this is returning entry.
 973	 * Let's not trace it and clear the entry depth, since
 974	 * we are out of irq code.
 975	 *
 976	 * This condition ensures that we 'leave the irq code' once
 977	 * we are out of the entry depth. Thus protecting us from
 978	 * the RETURN entry loss.
 979	 */
 980	if (*depth_irq >= depth) {
 981		*depth_irq = -1;
 982		return 1;
 983	}
 984
 985	/*
 986	 * We are inside the irq code, and this is not the entry.
 987	 */
 988	return 1;
 989}
 990
 991static enum print_line_t
 992print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
 993			struct trace_iterator *iter, u32 flags)
 994{
 995	struct fgraph_data *data = iter->private;
 996	struct ftrace_graph_ent *call = &field->graph_ent;
 997	struct ftrace_graph_ret_entry *leaf_ret;
 998	static enum print_line_t ret;
 999	int cpu = iter->cpu;
1000
1001	if (check_irq_entry(iter, flags, call->func, call->depth))
1002		return TRACE_TYPE_HANDLED;
1003
1004	print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags);
 
1005
1006	leaf_ret = get_return_for_leaf(iter, field);
1007	if (leaf_ret)
1008		ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
1009	else
1010		ret = print_graph_entry_nested(iter, field, s, cpu, flags);
1011
1012	if (data) {
1013		/*
1014		 * If we failed to write our output, then we need to make
1015		 * note of it. Because we already consumed our entry.
1016		 */
1017		if (s->full) {
1018			data->failed = 1;
1019			data->cpu = cpu;
1020		} else
1021			data->failed = 0;
1022	}
1023
1024	return ret;
1025}
1026
1027static enum print_line_t
1028print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
1029		   struct trace_entry *ent, struct trace_iterator *iter,
1030		   u32 flags)
1031{
1032	unsigned long long duration = trace->rettime - trace->calltime;
1033	struct fgraph_data *data = iter->private;
1034	struct trace_array *tr = iter->tr;
1035	pid_t pid = ent->pid;
1036	int cpu = iter->cpu;
1037	int func_match = 1;
 
1038	int i;
1039
1040	if (check_irq_return(iter, flags, trace->depth))
1041		return TRACE_TYPE_HANDLED;
1042
1043	if (data) {
1044		struct fgraph_cpu_data *cpu_data;
1045		int cpu = iter->cpu;
1046
1047		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
1048
1049		/*
1050		 * Comments display at + 1 to depth. This is the
1051		 * return from a function, we now want the comments
1052		 * to display at the same level of the bracket.
1053		 */
1054		cpu_data->depth = trace->depth - 1;
1055
1056		if (trace->depth < FTRACE_RETFUNC_DEPTH) {
1057			if (cpu_data->enter_funcs[trace->depth] != trace->func)
1058				func_match = 0;
1059			cpu_data->enter_funcs[trace->depth] = 0;
1060		}
1061	}
1062
1063	print_graph_prologue(iter, s, 0, 0, flags);
 
1064
1065	/* Overhead and duration */
1066	print_graph_duration(tr, duration, s, flags);
 
 
1067
1068	/* Closing brace */
1069	for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++)
1070		trace_seq_putc(s, ' ');
 
 
 
1071
1072	/*
1073	 * If the return function does not have a matching entry,
1074	 * then the entry was lost. Instead of just printing
1075	 * the '}' and letting the user guess what function this
1076	 * belongs to, write out the function name. Always do
1077	 * that if the funcgraph-tail option is enabled.
1078	 */
1079	if (func_match && !(flags & TRACE_GRAPH_PRINT_TAIL))
1080		trace_seq_puts(s, "}\n");
1081	else
1082		trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
 
 
 
 
 
1083
1084	/* Overrun */
1085	if (flags & TRACE_GRAPH_PRINT_OVERRUN)
1086		trace_seq_printf(s, " (Overruns: %lu)\n",
1087				 trace->overrun);
 
 
 
1088
1089	print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1090			cpu, pid, flags);
 
 
1091
1092	return trace_handle_return(s);
1093}
1094
1095static enum print_line_t
1096print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1097		    struct trace_iterator *iter, u32 flags)
1098{
1099	struct trace_array *tr = iter->tr;
1100	unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
1101	struct fgraph_data *data = iter->private;
1102	struct trace_event *event;
1103	int depth = 0;
1104	int ret;
1105	int i;
1106
1107	if (data)
1108		depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1109
1110	print_graph_prologue(iter, s, 0, 0, flags);
 
1111
1112	/* No time */
1113	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
 
 
1114
1115	/* Indentation */
1116	if (depth > 0)
1117		for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++)
1118			trace_seq_putc(s, ' ');
 
 
 
1119
1120	/* The comment */
1121	trace_seq_puts(s, "/* ");
 
 
1122
1123	switch (iter->ent->type) {
1124	case TRACE_BPRINT:
1125		ret = trace_print_bprintk_msg_only(iter);
1126		if (ret != TRACE_TYPE_HANDLED)
1127			return ret;
1128		break;
1129	case TRACE_PRINT:
1130		ret = trace_print_printk_msg_only(iter);
1131		if (ret != TRACE_TYPE_HANDLED)
1132			return ret;
1133		break;
1134	default:
1135		event = ftrace_find_event(ent->type);
1136		if (!event)
1137			return TRACE_TYPE_UNHANDLED;
1138
1139		ret = event->funcs->trace(iter, sym_flags, event);
1140		if (ret != TRACE_TYPE_HANDLED)
1141			return ret;
1142	}
1143
1144	if (trace_seq_has_overflowed(s))
1145		goto out;
1146
1147	/* Strip ending newline */
1148	if (s->buffer[s->seq.len - 1] == '\n') {
1149		s->buffer[s->seq.len - 1] = '\0';
1150		s->seq.len--;
1151	}
1152
1153	trace_seq_puts(s, " */\n");
1154 out:
1155	return trace_handle_return(s);
 
 
1156}
1157
1158
1159enum print_line_t
1160print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1161{
1162	struct ftrace_graph_ent_entry *field;
1163	struct fgraph_data *data = iter->private;
1164	struct trace_entry *entry = iter->ent;
1165	struct trace_seq *s = &iter->seq;
1166	int cpu = iter->cpu;
1167	int ret;
1168
1169	if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1170		per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1171		return TRACE_TYPE_HANDLED;
1172	}
1173
1174	/*
1175	 * If the last output failed, there's a possibility we need
1176	 * to print out the missing entry which would never go out.
1177	 */
1178	if (data && data->failed) {
1179		field = &data->ent;
1180		iter->cpu = data->cpu;
1181		ret = print_graph_entry(field, s, iter, flags);
1182		if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1183			per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1184			ret = TRACE_TYPE_NO_CONSUME;
1185		}
1186		iter->cpu = cpu;
1187		return ret;
1188	}
1189
1190	switch (entry->type) {
1191	case TRACE_GRAPH_ENT: {
1192		/*
1193		 * print_graph_entry() may consume the current event,
1194		 * thus @field may become invalid, so we need to save it.
1195		 * sizeof(struct ftrace_graph_ent_entry) is very small,
1196		 * it can be safely saved at the stack.
1197		 */
1198		struct ftrace_graph_ent_entry saved;
1199		trace_assign_type(field, entry);
1200		saved = *field;
1201		return print_graph_entry(&saved, s, iter, flags);
1202	}
1203	case TRACE_GRAPH_RET: {
1204		struct ftrace_graph_ret_entry *field;
1205		trace_assign_type(field, entry);
1206		return print_graph_return(&field->ret, s, entry, iter, flags);
1207	}
1208	case TRACE_STACK:
1209	case TRACE_FN:
1210		/* dont trace stack and functions as comments */
1211		return TRACE_TYPE_UNHANDLED;
1212
1213	default:
1214		return print_graph_comment(s, entry, iter, flags);
1215	}
1216
1217	return TRACE_TYPE_HANDLED;
1218}
1219
1220static enum print_line_t
1221print_graph_function(struct trace_iterator *iter)
1222{
1223	return print_graph_function_flags(iter, tracer_flags.val);
1224}
1225
1226static enum print_line_t
1227print_graph_function_event(struct trace_iterator *iter, int flags,
1228			   struct trace_event *event)
1229{
1230	return print_graph_function(iter);
1231}
1232
1233static void print_lat_header(struct seq_file *s, u32 flags)
1234{
1235	static const char spaces[] = "                "	/* 16 spaces */
1236		"    "					/* 4 spaces */
1237		"                 ";			/* 17 spaces */
1238	int size = 0;
1239
1240	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1241		size += 16;
1242	if (flags & TRACE_GRAPH_PRINT_CPU)
1243		size += 4;
1244	if (flags & TRACE_GRAPH_PRINT_PROC)
1245		size += 17;
1246
1247	seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1248	seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1249	seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1250	seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1251	seq_printf(s, "#%.*s||| /                      \n", size, spaces);
1252}
1253
1254static void __print_graph_headers_flags(struct trace_array *tr,
1255					struct seq_file *s, u32 flags)
1256{
1257	int lat = tr->trace_flags & TRACE_ITER_LATENCY_FMT;
1258
1259	if (lat)
1260		print_lat_header(s, flags);
1261
1262	/* 1st line */
1263	seq_putc(s, '#');
1264	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1265		seq_puts(s, "     TIME       ");
1266	if (flags & TRACE_GRAPH_PRINT_CPU)
1267		seq_puts(s, " CPU");
1268	if (flags & TRACE_GRAPH_PRINT_PROC)
1269		seq_puts(s, "  TASK/PID       ");
1270	if (lat)
1271		seq_puts(s, "||||");
1272	if (flags & TRACE_GRAPH_PRINT_DURATION)
1273		seq_puts(s, "  DURATION   ");
1274	seq_puts(s, "               FUNCTION CALLS\n");
1275
1276	/* 2nd line */
1277	seq_putc(s, '#');
1278	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1279		seq_puts(s, "      |         ");
1280	if (flags & TRACE_GRAPH_PRINT_CPU)
1281		seq_puts(s, " |  ");
1282	if (flags & TRACE_GRAPH_PRINT_PROC)
1283		seq_puts(s, "   |    |        ");
1284	if (lat)
1285		seq_puts(s, "||||");
1286	if (flags & TRACE_GRAPH_PRINT_DURATION)
1287		seq_puts(s, "   |   |      ");
1288	seq_puts(s, "               |   |   |   |\n");
1289}
1290
1291static void print_graph_headers(struct seq_file *s)
1292{
1293	print_graph_headers_flags(s, tracer_flags.val);
1294}
1295
1296void print_graph_headers_flags(struct seq_file *s, u32 flags)
1297{
1298	struct trace_iterator *iter = s->private;
1299	struct trace_array *tr = iter->tr;
1300
1301	if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
1302		return;
1303
1304	if (tr->trace_flags & TRACE_ITER_LATENCY_FMT) {
1305		/* print nothing if the buffers are empty */
1306		if (trace_empty(iter))
1307			return;
1308
1309		print_trace_header(s, iter);
1310	}
1311
1312	__print_graph_headers_flags(tr, s, flags);
1313}
1314
1315void graph_trace_open(struct trace_iterator *iter)
1316{
1317	/* pid and depth on the last trace processed */
1318	struct fgraph_data *data;
1319	gfp_t gfpflags;
1320	int cpu;
1321
1322	iter->private = NULL;
1323
1324	/* We can be called in atomic context via ftrace_dump() */
1325	gfpflags = (in_atomic() || irqs_disabled()) ? GFP_ATOMIC : GFP_KERNEL;
1326
1327	data = kzalloc(sizeof(*data), gfpflags);
1328	if (!data)
1329		goto out_err;
1330
1331	data->cpu_data = alloc_percpu_gfp(struct fgraph_cpu_data, gfpflags);
1332	if (!data->cpu_data)
1333		goto out_err_free;
1334
1335	for_each_possible_cpu(cpu) {
1336		pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1337		int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1338		int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1339		int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1340
1341		*pid = -1;
1342		*depth = 0;
1343		*ignore = 0;
1344		*depth_irq = -1;
1345	}
1346
1347	iter->private = data;
1348
1349	return;
1350
1351 out_err_free:
1352	kfree(data);
1353 out_err:
1354	pr_warn("function graph tracer: not enough memory\n");
1355}
1356
1357void graph_trace_close(struct trace_iterator *iter)
1358{
1359	struct fgraph_data *data = iter->private;
1360
1361	if (data) {
1362		free_percpu(data->cpu_data);
1363		kfree(data);
1364	}
1365}
1366
1367static int
1368func_graph_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1369{
1370	if (bit == TRACE_GRAPH_PRINT_IRQS)
1371		ftrace_graph_skip_irqs = !set;
1372
1373	if (bit == TRACE_GRAPH_SLEEP_TIME)
1374		ftrace_graph_sleep_time_control(set);
1375
1376	if (bit == TRACE_GRAPH_GRAPH_TIME)
1377		ftrace_graph_graph_time_control(set);
1378
1379	return 0;
1380}
1381
1382static struct trace_event_functions graph_functions = {
1383	.trace		= print_graph_function_event,
1384};
1385
1386static struct trace_event graph_trace_entry_event = {
1387	.type		= TRACE_GRAPH_ENT,
1388	.funcs		= &graph_functions,
1389};
1390
1391static struct trace_event graph_trace_ret_event = {
1392	.type		= TRACE_GRAPH_RET,
1393	.funcs		= &graph_functions
1394};
1395
1396static struct tracer graph_trace __tracer_data = {
1397	.name		= "function_graph",
1398	.update_thresh	= graph_trace_update_thresh,
1399	.open		= graph_trace_open,
1400	.pipe_open	= graph_trace_open,
1401	.close		= graph_trace_close,
1402	.pipe_close	= graph_trace_close,
 
1403	.init		= graph_trace_init,
1404	.reset		= graph_trace_reset,
1405	.print_line	= print_graph_function,
1406	.print_header	= print_graph_headers,
1407	.flags		= &tracer_flags,
1408	.set_flag	= func_graph_set_flag,
1409#ifdef CONFIG_FTRACE_SELFTEST
1410	.selftest	= trace_selftest_startup_function_graph,
1411#endif
1412};
1413
1414
1415static ssize_t
1416graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1417		  loff_t *ppos)
1418{
1419	unsigned long val;
1420	int ret;
1421
1422	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1423	if (ret)
1424		return ret;
1425
1426	max_depth = val;
1427
1428	*ppos += cnt;
1429
1430	return cnt;
1431}
1432
1433static ssize_t
1434graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1435		 loff_t *ppos)
1436{
1437	char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1438	int n;
1439
1440	n = sprintf(buf, "%d\n", max_depth);
1441
1442	return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1443}
1444
1445static const struct file_operations graph_depth_fops = {
1446	.open		= tracing_open_generic,
1447	.write		= graph_depth_write,
1448	.read		= graph_depth_read,
1449	.llseek		= generic_file_llseek,
1450};
1451
1452static __init int init_graph_tracefs(void)
1453{
1454	struct dentry *d_tracer;
1455
1456	d_tracer = tracing_init_dentry();
1457	if (IS_ERR(d_tracer))
1458		return 0;
1459
1460	trace_create_file("max_graph_depth", 0644, d_tracer,
1461			  NULL, &graph_depth_fops);
1462
1463	return 0;
1464}
1465fs_initcall(init_graph_tracefs);
1466
1467static __init int init_graph_trace(void)
1468{
1469	max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1470
1471	if (!register_trace_event(&graph_trace_entry_event)) {
1472		pr_warn("Warning: could not register graph trace events\n");
1473		return 1;
1474	}
1475
1476	if (!register_trace_event(&graph_trace_ret_event)) {
1477		pr_warn("Warning: could not register graph trace events\n");
1478		return 1;
1479	}
1480
1481	return register_tracer(&graph_trace);
1482}
1483
1484core_initcall(init_graph_trace);
v3.1
   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 */
  19static int ftrace_graph_skip_irqs;
  20
  21struct 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
  29struct 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/* Flag options */
  42#define TRACE_GRAPH_PRINT_OVERRUN	0x1
  43#define TRACE_GRAPH_PRINT_CPU		0x2
  44#define TRACE_GRAPH_PRINT_OVERHEAD	0x4
  45#define TRACE_GRAPH_PRINT_PROC		0x8
  46#define TRACE_GRAPH_PRINT_DURATION	0x10
  47#define TRACE_GRAPH_PRINT_ABS_TIME	0x20
  48#define TRACE_GRAPH_PRINT_IRQS		0x40
  49
  50static struct tracer_opt trace_opts[] = {
  51	/* Display overruns? (for self-debug purpose) */
  52	{ TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
  53	/* Display CPU ? */
  54	{ TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
  55	/* Display Overhead ? */
  56	{ TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
  57	/* Display proc name/pid */
  58	{ TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
  59	/* Display duration of execution */
  60	{ TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
  61	/* Display absolute time of an entry */
  62	{ TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
  63	/* Display interrupts */
  64	{ TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
 
 
 
 
 
 
  65	{ } /* Empty entry */
  66};
  67
  68static struct tracer_flags tracer_flags = {
  69	/* Don't display overruns and proc by default */
  70	.val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
  71	       TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS,
 
  72	.opts = trace_opts
  73};
  74
  75static struct trace_array *graph_array;
  76
  77/*
  78 * DURATION column is being also used to display IRQ signs,
  79 * following values are used by print_graph_irq and others
  80 * to fill in space into DURATION column.
  81 */
  82enum {
  83	DURATION_FILL_FULL  = -1,
  84	DURATION_FILL_START = -2,
  85	DURATION_FILL_END   = -3,
  86};
  87
  88static enum print_line_t
  89print_graph_duration(unsigned long long duration, struct trace_seq *s,
  90		     u32 flags);
  91
  92/* Add a function return address to the trace stack on thread info.*/
  93int
  94ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
  95			 unsigned long frame_pointer)
  96{
  97	unsigned long long calltime;
  98	int index;
  99
 
 
 
 100	if (!current->ret_stack)
 101		return -EBUSY;
 102
 103	/*
 104	 * We must make sure the ret_stack is tested before we read
 105	 * anything else.
 106	 */
 107	smp_rmb();
 108
 109	/* The return trace stack is full */
 110	if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
 111		atomic_inc(&current->trace_overrun);
 112		return -EBUSY;
 113	}
 114
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 115	calltime = trace_clock_local();
 116
 117	index = ++current->curr_ret_stack;
 
 
 118	barrier();
 119	current->ret_stack[index].ret = ret;
 120	current->ret_stack[index].func = func;
 121	current->ret_stack[index].calltime = calltime;
 122	current->ret_stack[index].subtime = 0;
 123	current->ret_stack[index].fp = frame_pointer;
 124	*depth = index;
 125
 126	return 0;
 127}
 128
 129/* Retrieve a function return address to the trace stack on thread info.*/
 130static void
 131ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
 132			unsigned long frame_pointer)
 133{
 134	int index;
 135
 136	index = current->curr_ret_stack;
 137
 138	if (unlikely(index < 0)) {
 
 
 
 
 
 
 
 
 
 
 139		ftrace_graph_stop();
 140		WARN_ON(1);
 141		/* Might as well panic, otherwise we have no where to go */
 142		*ret = (unsigned long)panic;
 143		return;
 144	}
 145
 146#ifdef CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST
 147	/*
 148	 * The arch may choose to record the frame pointer used
 149	 * and check it here to make sure that it is what we expect it
 150	 * to be. If gcc does not set the place holder of the return
 151	 * address in the frame pointer, and does a copy instead, then
 152	 * the function graph trace will fail. This test detects this
 153	 * case.
 154	 *
 155	 * Currently, x86_32 with optimize for size (-Os) makes the latest
 156	 * gcc do the above.
 
 
 
 157	 */
 158	if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
 159		ftrace_graph_stop();
 160		WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
 161		     "  from func %ps return to %lx\n",
 162		     current->ret_stack[index].fp,
 163		     frame_pointer,
 164		     (void *)current->ret_stack[index].func,
 165		     current->ret_stack[index].ret);
 166		*ret = (unsigned long)panic;
 167		return;
 168	}
 169#endif
 170
 171	*ret = current->ret_stack[index].ret;
 172	trace->func = current->ret_stack[index].func;
 173	trace->calltime = current->ret_stack[index].calltime;
 174	trace->overrun = atomic_read(&current->trace_overrun);
 175	trace->depth = index;
 176}
 177
 178/*
 179 * Send the trace to the ring-buffer.
 180 * @return the original return address.
 181 */
 182unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
 183{
 184	struct ftrace_graph_ret trace;
 185	unsigned long ret;
 186
 187	ftrace_pop_return_trace(&trace, &ret, frame_pointer);
 188	trace.rettime = trace_clock_local();
 189	ftrace_graph_return(&trace);
 190	barrier();
 191	current->curr_ret_stack--;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 192
 193	if (unlikely(!ret)) {
 194		ftrace_graph_stop();
 195		WARN_ON(1);
 196		/* Might as well panic. What else to do? */
 197		ret = (unsigned long)panic;
 198	}
 199
 200	return ret;
 201}
 202
 203int __trace_graph_entry(struct trace_array *tr,
 204				struct ftrace_graph_ent *trace,
 205				unsigned long flags,
 206				int pc)
 207{
 208	struct ftrace_event_call *call = &event_funcgraph_entry;
 209	struct ring_buffer_event *event;
 210	struct ring_buffer *buffer = tr->buffer;
 211	struct ftrace_graph_ent_entry *entry;
 212
 213	if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
 214		return 0;
 215
 216	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
 217					  sizeof(*entry), flags, pc);
 218	if (!event)
 219		return 0;
 220	entry	= ring_buffer_event_data(event);
 221	entry->graph_ent			= *trace;
 222	if (!filter_current_check_discard(buffer, call, entry, event))
 223		ring_buffer_unlock_commit(buffer, event);
 224
 225	return 1;
 226}
 227
 228static inline int ftrace_graph_ignore_irqs(void)
 229{
 230	if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
 231		return 0;
 232
 233	return in_irq();
 234}
 235
 236int trace_graph_entry(struct ftrace_graph_ent *trace)
 237{
 238	struct trace_array *tr = graph_array;
 239	struct trace_array_cpu *data;
 240	unsigned long flags;
 241	long disabled;
 242	int ret;
 243	int cpu;
 244	int pc;
 245
 246	if (!ftrace_trace_task(current))
 247		return 0;
 248
 249	/* trace it when it is-nested-in or is a function enabled. */
 250	if (!(trace->depth || ftrace_graph_addr(trace->func)) ||
 251	      ftrace_graph_ignore_irqs())
 
 252		return 0;
 253
 
 
 
 
 
 
 
 
 
 
 254	local_irq_save(flags);
 255	cpu = raw_smp_processor_id();
 256	data = tr->data[cpu];
 257	disabled = atomic_inc_return(&data->disabled);
 258	if (likely(disabled == 1)) {
 259		pc = preempt_count();
 260		ret = __trace_graph_entry(tr, trace, flags, pc);
 261	} else {
 262		ret = 0;
 263	}
 264
 265	atomic_dec(&data->disabled);
 266	local_irq_restore(flags);
 267
 268	return ret;
 269}
 270
 271int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
 272{
 273	if (tracing_thresh)
 274		return 1;
 275	else
 276		return trace_graph_entry(trace);
 277}
 278
 279static void
 280__trace_graph_function(struct trace_array *tr,
 281		unsigned long ip, unsigned long flags, int pc)
 282{
 283	u64 time = trace_clock_local();
 284	struct ftrace_graph_ent ent = {
 285		.func  = ip,
 286		.depth = 0,
 287	};
 288	struct ftrace_graph_ret ret = {
 289		.func     = ip,
 290		.depth    = 0,
 291		.calltime = time,
 292		.rettime  = time,
 293	};
 294
 295	__trace_graph_entry(tr, &ent, flags, pc);
 296	__trace_graph_return(tr, &ret, flags, pc);
 297}
 298
 299void
 300trace_graph_function(struct trace_array *tr,
 301		unsigned long ip, unsigned long parent_ip,
 302		unsigned long flags, int pc)
 303{
 304	__trace_graph_function(tr, ip, flags, pc);
 305}
 306
 307void __trace_graph_return(struct trace_array *tr,
 308				struct ftrace_graph_ret *trace,
 309				unsigned long flags,
 310				int pc)
 311{
 312	struct ftrace_event_call *call = &event_funcgraph_exit;
 313	struct ring_buffer_event *event;
 314	struct ring_buffer *buffer = tr->buffer;
 315	struct ftrace_graph_ret_entry *entry;
 316
 317	if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
 318		return;
 319
 320	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
 321					  sizeof(*entry), flags, pc);
 322	if (!event)
 323		return;
 324	entry	= ring_buffer_event_data(event);
 325	entry->ret				= *trace;
 326	if (!filter_current_check_discard(buffer, call, entry, event))
 327		ring_buffer_unlock_commit(buffer, event);
 328}
 329
 330void trace_graph_return(struct ftrace_graph_ret *trace)
 331{
 332	struct trace_array *tr = graph_array;
 333	struct trace_array_cpu *data;
 334	unsigned long flags;
 335	long disabled;
 336	int cpu;
 337	int pc;
 338
 339	local_irq_save(flags);
 340	cpu = raw_smp_processor_id();
 341	data = tr->data[cpu];
 342	disabled = atomic_inc_return(&data->disabled);
 343	if (likely(disabled == 1)) {
 344		pc = preempt_count();
 345		__trace_graph_return(tr, trace, flags, pc);
 346	}
 347	atomic_dec(&data->disabled);
 348	local_irq_restore(flags);
 349}
 350
 351void set_graph_array(struct trace_array *tr)
 352{
 353	graph_array = tr;
 354
 355	/* Make graph_array visible before we start tracing */
 356
 357	smp_mb();
 358}
 359
 360void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
 361{
 362	if (tracing_thresh &&
 363	    (trace->rettime - trace->calltime < tracing_thresh))
 364		return;
 365	else
 366		trace_graph_return(trace);
 367}
 368
 369static int graph_trace_init(struct trace_array *tr)
 370{
 371	int ret;
 372
 373	set_graph_array(tr);
 374	if (tracing_thresh)
 375		ret = register_ftrace_graph(&trace_graph_thresh_return,
 376					    &trace_graph_thresh_entry);
 377	else
 378		ret = register_ftrace_graph(&trace_graph_return,
 379					    &trace_graph_entry);
 380	if (ret)
 381		return ret;
 382	tracing_start_cmdline_record();
 383
 384	return 0;
 385}
 386
 387static void graph_trace_reset(struct trace_array *tr)
 388{
 389	tracing_stop_cmdline_record();
 390	unregister_ftrace_graph();
 391}
 392
 
 
 
 
 
 
 393static int max_bytes_for_cpu;
 394
 395static enum print_line_t
 396print_graph_cpu(struct trace_seq *s, int cpu)
 397{
 398	int ret;
 399
 400	/*
 401	 * Start with a space character - to make it stand out
 402	 * to the right a bit when trace output is pasted into
 403	 * email:
 404	 */
 405	ret = trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
 406	if (!ret)
 407		return TRACE_TYPE_PARTIAL_LINE;
 408
 409	return TRACE_TYPE_HANDLED;
 410}
 411
 412#define TRACE_GRAPH_PROCINFO_LENGTH	14
 413
 414static enum print_line_t
 415print_graph_proc(struct trace_seq *s, pid_t pid)
 416{
 417	char comm[TASK_COMM_LEN];
 418	/* sign + log10(MAX_INT) + '\0' */
 419	char pid_str[11];
 420	int spaces = 0;
 421	int ret;
 422	int len;
 423	int i;
 424
 425	trace_find_cmdline(pid, comm);
 426	comm[7] = '\0';
 427	sprintf(pid_str, "%d", pid);
 428
 429	/* 1 stands for the "-" character */
 430	len = strlen(comm) + strlen(pid_str) + 1;
 431
 432	if (len < TRACE_GRAPH_PROCINFO_LENGTH)
 433		spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
 434
 435	/* First spaces to align center */
 436	for (i = 0; i < spaces / 2; i++) {
 437		ret = trace_seq_printf(s, " ");
 438		if (!ret)
 439			return TRACE_TYPE_PARTIAL_LINE;
 440	}
 441
 442	ret = trace_seq_printf(s, "%s-%s", comm, pid_str);
 443	if (!ret)
 444		return TRACE_TYPE_PARTIAL_LINE;
 445
 446	/* Last spaces to align center */
 447	for (i = 0; i < spaces - (spaces / 2); i++) {
 448		ret = trace_seq_printf(s, " ");
 449		if (!ret)
 450			return TRACE_TYPE_PARTIAL_LINE;
 451	}
 452	return TRACE_TYPE_HANDLED;
 453}
 454
 455
 456static enum print_line_t
 457print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
 458{
 459	if (!trace_seq_putc(s, ' '))
 460		return 0;
 461
 462	return trace_print_lat_fmt(s, entry);
 463}
 464
 465/* If the pid changed since the last trace, output this event */
 466static enum print_line_t
 467verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
 468{
 469	pid_t prev_pid;
 470	pid_t *last_pid;
 471	int ret;
 472
 473	if (!data)
 474		return TRACE_TYPE_HANDLED;
 475
 476	last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
 477
 478	if (*last_pid == pid)
 479		return TRACE_TYPE_HANDLED;
 480
 481	prev_pid = *last_pid;
 482	*last_pid = pid;
 483
 484	if (prev_pid == -1)
 485		return TRACE_TYPE_HANDLED;
 486/*
 487 * Context-switch trace line:
 488
 489 ------------------------------------------
 490 | 1)  migration/0--1  =>  sshd-1755
 491 ------------------------------------------
 492
 493 */
 494	ret = trace_seq_printf(s,
 495		" ------------------------------------------\n");
 496	if (!ret)
 497		return TRACE_TYPE_PARTIAL_LINE;
 498
 499	ret = print_graph_cpu(s, cpu);
 500	if (ret == TRACE_TYPE_PARTIAL_LINE)
 501		return TRACE_TYPE_PARTIAL_LINE;
 502
 503	ret = print_graph_proc(s, prev_pid);
 504	if (ret == TRACE_TYPE_PARTIAL_LINE)
 505		return TRACE_TYPE_PARTIAL_LINE;
 506
 507	ret = trace_seq_printf(s, " => ");
 508	if (!ret)
 509		return TRACE_TYPE_PARTIAL_LINE;
 510
 511	ret = print_graph_proc(s, pid);
 512	if (ret == TRACE_TYPE_PARTIAL_LINE)
 513		return TRACE_TYPE_PARTIAL_LINE;
 514
 515	ret = trace_seq_printf(s,
 516		"\n ------------------------------------------\n\n");
 517	if (!ret)
 518		return TRACE_TYPE_PARTIAL_LINE;
 519
 520	return TRACE_TYPE_HANDLED;
 521}
 522
 523static struct ftrace_graph_ret_entry *
 524get_return_for_leaf(struct trace_iterator *iter,
 525		struct ftrace_graph_ent_entry *curr)
 526{
 527	struct fgraph_data *data = iter->private;
 528	struct ring_buffer_iter *ring_iter = NULL;
 529	struct ring_buffer_event *event;
 530	struct ftrace_graph_ret_entry *next;
 531
 532	/*
 533	 * If the previous output failed to write to the seq buffer,
 534	 * then we just reuse the data from before.
 535	 */
 536	if (data && data->failed) {
 537		curr = &data->ent;
 538		next = &data->ret;
 539	} else {
 540
 541		ring_iter = iter->buffer_iter[iter->cpu];
 542
 543		/* First peek to compare current entry and the next one */
 544		if (ring_iter)
 545			event = ring_buffer_iter_peek(ring_iter, NULL);
 546		else {
 547			/*
 548			 * We need to consume the current entry to see
 549			 * the next one.
 550			 */
 551			ring_buffer_consume(iter->tr->buffer, iter->cpu,
 552					    NULL, NULL);
 553			event = ring_buffer_peek(iter->tr->buffer, iter->cpu,
 554						 NULL, NULL);
 555		}
 556
 557		if (!event)
 558			return NULL;
 559
 560		next = ring_buffer_event_data(event);
 561
 562		if (data) {
 563			/*
 564			 * Save current and next entries for later reference
 565			 * if the output fails.
 566			 */
 567			data->ent = *curr;
 568			/*
 569			 * If the next event is not a return type, then
 570			 * we only care about what type it is. Otherwise we can
 571			 * safely copy the entire event.
 572			 */
 573			if (next->ent.type == TRACE_GRAPH_RET)
 574				data->ret = *next;
 575			else
 576				data->ret.ent.type = next->ent.type;
 577		}
 578	}
 579
 580	if (next->ent.type != TRACE_GRAPH_RET)
 581		return NULL;
 582
 583	if (curr->ent.pid != next->ent.pid ||
 584			curr->graph_ent.func != next->ret.func)
 585		return NULL;
 586
 587	/* this is a leaf, now advance the iterator */
 588	if (ring_iter)
 589		ring_buffer_read(ring_iter, NULL);
 590
 591	return next;
 592}
 593
 594static int print_graph_abs_time(u64 t, struct trace_seq *s)
 595{
 596	unsigned long usecs_rem;
 597
 598	usecs_rem = do_div(t, NSEC_PER_SEC);
 599	usecs_rem /= 1000;
 600
 601	return trace_seq_printf(s, "%5lu.%06lu |  ",
 602			(unsigned long)t, usecs_rem);
 603}
 604
 605static enum print_line_t
 606print_graph_irq(struct trace_iterator *iter, unsigned long addr,
 607		enum trace_type type, int cpu, pid_t pid, u32 flags)
 608{
 609	int ret;
 610	struct trace_seq *s = &iter->seq;
 
 611
 612	if (addr < (unsigned long)__irqentry_text_start ||
 613		addr >= (unsigned long)__irqentry_text_end)
 614		return TRACE_TYPE_UNHANDLED;
 615
 616	if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
 617		/* Absolute time */
 618		if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
 619			ret = print_graph_abs_time(iter->ts, s);
 620			if (!ret)
 621				return TRACE_TYPE_PARTIAL_LINE;
 622		}
 623
 624		/* Cpu */
 625		if (flags & TRACE_GRAPH_PRINT_CPU) {
 626			ret = print_graph_cpu(s, cpu);
 627			if (ret == TRACE_TYPE_PARTIAL_LINE)
 628				return TRACE_TYPE_PARTIAL_LINE;
 629		}
 630
 631		/* Proc */
 632		if (flags & TRACE_GRAPH_PRINT_PROC) {
 633			ret = print_graph_proc(s, pid);
 634			if (ret == TRACE_TYPE_PARTIAL_LINE)
 635				return TRACE_TYPE_PARTIAL_LINE;
 636			ret = trace_seq_printf(s, " | ");
 637			if (!ret)
 638				return TRACE_TYPE_PARTIAL_LINE;
 639		}
 
 
 
 
 640	}
 641
 642	/* No overhead */
 643	ret = print_graph_duration(DURATION_FILL_START, s, flags);
 644	if (ret != TRACE_TYPE_HANDLED)
 645		return ret;
 646
 647	if (type == TRACE_GRAPH_ENT)
 648		ret = trace_seq_printf(s, "==========>");
 649	else
 650		ret = trace_seq_printf(s, "<==========");
 651
 652	if (!ret)
 653		return TRACE_TYPE_PARTIAL_LINE;
 654
 655	ret = print_graph_duration(DURATION_FILL_END, s, flags);
 656	if (ret != TRACE_TYPE_HANDLED)
 657		return ret;
 658
 659	ret = trace_seq_printf(s, "\n");
 660
 661	if (!ret)
 662		return TRACE_TYPE_PARTIAL_LINE;
 663	return TRACE_TYPE_HANDLED;
 664}
 665
 666enum print_line_t
 667trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
 668{
 669	unsigned long nsecs_rem = do_div(duration, 1000);
 670	/* log10(ULONG_MAX) + '\0' */
 671	char msecs_str[21];
 672	char nsecs_str[5];
 673	int ret, len;
 674	int i;
 675
 676	sprintf(msecs_str, "%lu", (unsigned long) duration);
 677
 678	/* Print msecs */
 679	ret = trace_seq_printf(s, "%s", msecs_str);
 680	if (!ret)
 681		return TRACE_TYPE_PARTIAL_LINE;
 682
 683	len = strlen(msecs_str);
 684
 685	/* Print nsecs (we don't want to exceed 7 numbers) */
 686	if (len < 7) {
 687		size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
 688
 689		snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
 690		ret = trace_seq_printf(s, ".%s", nsecs_str);
 691		if (!ret)
 692			return TRACE_TYPE_PARTIAL_LINE;
 693		len += strlen(nsecs_str);
 694	}
 695
 696	ret = trace_seq_printf(s, " us ");
 697	if (!ret)
 698		return TRACE_TYPE_PARTIAL_LINE;
 699
 700	/* Print remaining spaces to fit the row's width */
 701	for (i = len; i < 7; i++) {
 702		ret = trace_seq_printf(s, " ");
 703		if (!ret)
 704			return TRACE_TYPE_PARTIAL_LINE;
 705	}
 706	return TRACE_TYPE_HANDLED;
 707}
 708
 709static enum print_line_t
 710print_graph_duration(unsigned long long duration, struct trace_seq *s,
 711		     u32 flags)
 712{
 713	int ret = -1;
 714
 715	if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
 716	    !(trace_flags & TRACE_ITER_CONTEXT_INFO))
 717			return TRACE_TYPE_HANDLED;
 718
 719	/* No real adata, just filling the column with spaces */
 720	switch (duration) {
 721	case DURATION_FILL_FULL:
 722		ret = trace_seq_printf(s, "              |  ");
 723		return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 724	case DURATION_FILL_START:
 725		ret = trace_seq_printf(s, "  ");
 726		return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 727	case DURATION_FILL_END:
 728		ret = trace_seq_printf(s, " |");
 729		return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 730	}
 731
 732	/* Signal a overhead of time execution to the output */
 733	if (flags & TRACE_GRAPH_PRINT_OVERHEAD) {
 734		/* Duration exceeded 100 msecs */
 735		if (duration > 100000ULL)
 736			ret = trace_seq_printf(s, "! ");
 737		/* Duration exceeded 10 msecs */
 738		else if (duration > 10000ULL)
 739			ret = trace_seq_printf(s, "+ ");
 740	}
 741
 742	/*
 743	 * The -1 means we either did not exceed the duration tresholds
 744	 * or we dont want to print out the overhead. Either way we need
 745	 * to fill out the space.
 746	 */
 747	if (ret == -1)
 748		ret = trace_seq_printf(s, "  ");
 749
 750	/* Catching here any failure happenned above */
 751	if (!ret)
 752		return TRACE_TYPE_PARTIAL_LINE;
 753
 754	ret = trace_print_graph_duration(duration, s);
 755	if (ret != TRACE_TYPE_HANDLED)
 756		return ret;
 757
 758	ret = trace_seq_printf(s, "|  ");
 759	if (!ret)
 760		return TRACE_TYPE_PARTIAL_LINE;
 761
 762	return TRACE_TYPE_HANDLED;
 
 763}
 764
 765/* Case of a leaf function on its call entry */
 766static enum print_line_t
 767print_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 ftrace_graph_ret *graph_ret;
 774	struct ftrace_graph_ent *call;
 775	unsigned long long duration;
 776	int ret;
 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	ret = print_graph_duration(duration, s, flags);
 803	if (ret == TRACE_TYPE_PARTIAL_LINE)
 804		return TRACE_TYPE_PARTIAL_LINE;
 805
 806	/* Function */
 807	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 808		ret = trace_seq_printf(s, " ");
 809		if (!ret)
 810			return TRACE_TYPE_PARTIAL_LINE;
 811	}
 812
 813	ret = trace_seq_printf(s, "%ps();\n", (void *)call->func);
 814	if (!ret)
 815		return TRACE_TYPE_PARTIAL_LINE;
 816
 817	return TRACE_TYPE_HANDLED;
 818}
 819
 820static enum print_line_t
 821print_graph_entry_nested(struct trace_iterator *iter,
 822			 struct ftrace_graph_ent_entry *entry,
 823			 struct trace_seq *s, int cpu, u32 flags)
 824{
 825	struct ftrace_graph_ent *call = &entry->graph_ent;
 826	struct fgraph_data *data = iter->private;
 827	int ret;
 828	int i;
 829
 830	if (data) {
 831		struct fgraph_cpu_data *cpu_data;
 832		int cpu = iter->cpu;
 833
 834		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 835		cpu_data->depth = call->depth;
 836
 837		/* Save this function pointer to see if the exit matches */
 838		if (call->depth < FTRACE_RETFUNC_DEPTH)
 839			cpu_data->enter_funcs[call->depth] = call->func;
 840	}
 841
 842	/* No time */
 843	ret = print_graph_duration(DURATION_FILL_FULL, s, flags);
 844	if (ret != TRACE_TYPE_HANDLED)
 845		return ret;
 846
 847	/* Function */
 848	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 849		ret = trace_seq_printf(s, " ");
 850		if (!ret)
 851			return TRACE_TYPE_PARTIAL_LINE;
 852	}
 853
 854	ret = trace_seq_printf(s, "%ps() {\n", (void *)call->func);
 855	if (!ret)
 856		return TRACE_TYPE_PARTIAL_LINE;
 857
 858	/*
 859	 * we already consumed the current entry to check the next one
 860	 * and see if this is a leaf.
 861	 */
 862	return TRACE_TYPE_NO_CONSUME;
 863}
 864
 865static enum print_line_t
 866print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
 867		     int type, unsigned long addr, u32 flags)
 868{
 869	struct fgraph_data *data = iter->private;
 870	struct trace_entry *ent = iter->ent;
 
 871	int cpu = iter->cpu;
 872	int ret;
 873
 874	/* Pid */
 875	if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
 876		return TRACE_TYPE_PARTIAL_LINE;
 877
 878	if (type) {
 879		/* Interrupt */
 880		ret = print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
 881		if (ret == TRACE_TYPE_PARTIAL_LINE)
 882			return TRACE_TYPE_PARTIAL_LINE;
 883	}
 884
 885	if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
 886		return 0;
 887
 888	/* Absolute time */
 889	if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
 890		ret = print_graph_abs_time(iter->ts, s);
 891		if (!ret)
 892			return TRACE_TYPE_PARTIAL_LINE;
 893	}
 894
 895	/* Cpu */
 896	if (flags & TRACE_GRAPH_PRINT_CPU) {
 897		ret = print_graph_cpu(s, cpu);
 898		if (ret == TRACE_TYPE_PARTIAL_LINE)
 899			return TRACE_TYPE_PARTIAL_LINE;
 900	}
 901
 902	/* Proc */
 903	if (flags & TRACE_GRAPH_PRINT_PROC) {
 904		ret = print_graph_proc(s, ent->pid);
 905		if (ret == TRACE_TYPE_PARTIAL_LINE)
 906			return TRACE_TYPE_PARTIAL_LINE;
 907
 908		ret = trace_seq_printf(s, " | ");
 909		if (!ret)
 910			return TRACE_TYPE_PARTIAL_LINE;
 911	}
 912
 913	/* Latency format */
 914	if (trace_flags & TRACE_ITER_LATENCY_FMT) {
 915		ret = print_graph_lat_fmt(s, ent);
 916		if (ret == TRACE_TYPE_PARTIAL_LINE)
 917			return TRACE_TYPE_PARTIAL_LINE;
 918	}
 919
 920	return 0;
 921}
 922
 923/*
 924 * Entry check for irq code
 925 *
 926 * returns 1 if
 927 *  - we are inside irq code
 928 *  - we just entered irq code
 929 *
 930 * retunns 0 if
 931 *  - funcgraph-interrupts option is set
 932 *  - we are not inside irq code
 933 */
 934static int
 935check_irq_entry(struct trace_iterator *iter, u32 flags,
 936		unsigned long addr, int depth)
 937{
 938	int cpu = iter->cpu;
 939	int *depth_irq;
 940	struct fgraph_data *data = iter->private;
 941
 942	/*
 943	 * If we are either displaying irqs, or we got called as
 944	 * a graph event and private data does not exist,
 945	 * then we bypass the irq check.
 946	 */
 947	if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
 948	    (!data))
 949		return 0;
 950
 951	depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
 952
 953	/*
 954	 * We are inside the irq code
 955	 */
 956	if (*depth_irq >= 0)
 957		return 1;
 958
 959	if ((addr < (unsigned long)__irqentry_text_start) ||
 960	    (addr >= (unsigned long)__irqentry_text_end))
 961		return 0;
 962
 963	/*
 964	 * We are entering irq code.
 965	 */
 966	*depth_irq = depth;
 967	return 1;
 968}
 969
 970/*
 971 * Return check for irq code
 972 *
 973 * returns 1 if
 974 *  - we are inside irq code
 975 *  - we just left irq code
 976 *
 977 * returns 0 if
 978 *  - funcgraph-interrupts option is set
 979 *  - we are not inside irq code
 980 */
 981static int
 982check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
 983{
 984	int cpu = iter->cpu;
 985	int *depth_irq;
 986	struct fgraph_data *data = iter->private;
 987
 988	/*
 989	 * If we are either displaying irqs, or we got called as
 990	 * a graph event and private data does not exist,
 991	 * then we bypass the irq check.
 992	 */
 993	if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
 994	    (!data))
 995		return 0;
 996
 997	depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
 998
 999	/*
1000	 * We are not inside the irq code.
1001	 */
1002	if (*depth_irq == -1)
1003		return 0;
1004
1005	/*
1006	 * We are inside the irq code, and this is returning entry.
1007	 * Let's not trace it and clear the entry depth, since
1008	 * we are out of irq code.
1009	 *
1010	 * This condition ensures that we 'leave the irq code' once
1011	 * we are out of the entry depth. Thus protecting us from
1012	 * the RETURN entry loss.
1013	 */
1014	if (*depth_irq >= depth) {
1015		*depth_irq = -1;
1016		return 1;
1017	}
1018
1019	/*
1020	 * We are inside the irq code, and this is not the entry.
1021	 */
1022	return 1;
1023}
1024
1025static enum print_line_t
1026print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
1027			struct trace_iterator *iter, u32 flags)
1028{
1029	struct fgraph_data *data = iter->private;
1030	struct ftrace_graph_ent *call = &field->graph_ent;
1031	struct ftrace_graph_ret_entry *leaf_ret;
1032	static enum print_line_t ret;
1033	int cpu = iter->cpu;
1034
1035	if (check_irq_entry(iter, flags, call->func, call->depth))
1036		return TRACE_TYPE_HANDLED;
1037
1038	if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags))
1039		return TRACE_TYPE_PARTIAL_LINE;
1040
1041	leaf_ret = get_return_for_leaf(iter, field);
1042	if (leaf_ret)
1043		ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
1044	else
1045		ret = print_graph_entry_nested(iter, field, s, cpu, flags);
1046
1047	if (data) {
1048		/*
1049		 * If we failed to write our output, then we need to make
1050		 * note of it. Because we already consumed our entry.
1051		 */
1052		if (s->full) {
1053			data->failed = 1;
1054			data->cpu = cpu;
1055		} else
1056			data->failed = 0;
1057	}
1058
1059	return ret;
1060}
1061
1062static enum print_line_t
1063print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
1064		   struct trace_entry *ent, struct trace_iterator *iter,
1065		   u32 flags)
1066{
1067	unsigned long long duration = trace->rettime - trace->calltime;
1068	struct fgraph_data *data = iter->private;
 
1069	pid_t pid = ent->pid;
1070	int cpu = iter->cpu;
1071	int func_match = 1;
1072	int ret;
1073	int i;
1074
1075	if (check_irq_return(iter, flags, trace->depth))
1076		return TRACE_TYPE_HANDLED;
1077
1078	if (data) {
1079		struct fgraph_cpu_data *cpu_data;
1080		int cpu = iter->cpu;
1081
1082		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
1083
1084		/*
1085		 * Comments display at + 1 to depth. This is the
1086		 * return from a function, we now want the comments
1087		 * to display at the same level of the bracket.
1088		 */
1089		cpu_data->depth = trace->depth - 1;
1090
1091		if (trace->depth < FTRACE_RETFUNC_DEPTH) {
1092			if (cpu_data->enter_funcs[trace->depth] != trace->func)
1093				func_match = 0;
1094			cpu_data->enter_funcs[trace->depth] = 0;
1095		}
1096	}
1097
1098	if (print_graph_prologue(iter, s, 0, 0, flags))
1099		return TRACE_TYPE_PARTIAL_LINE;
1100
1101	/* Overhead and duration */
1102	ret = print_graph_duration(duration, s, flags);
1103	if (ret == TRACE_TYPE_PARTIAL_LINE)
1104		return TRACE_TYPE_PARTIAL_LINE;
1105
1106	/* Closing brace */
1107	for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
1108		ret = trace_seq_printf(s, " ");
1109		if (!ret)
1110			return TRACE_TYPE_PARTIAL_LINE;
1111	}
1112
1113	/*
1114	 * If the return function does not have a matching entry,
1115	 * then the entry was lost. Instead of just printing
1116	 * the '}' and letting the user guess what function this
1117	 * belongs to, write out the function name.
 
1118	 */
1119	if (func_match) {
1120		ret = trace_seq_printf(s, "}\n");
1121		if (!ret)
1122			return TRACE_TYPE_PARTIAL_LINE;
1123	} else {
1124		ret = trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
1125		if (!ret)
1126			return TRACE_TYPE_PARTIAL_LINE;
1127	}
1128
1129	/* Overrun */
1130	if (flags & TRACE_GRAPH_PRINT_OVERRUN) {
1131		ret = trace_seq_printf(s, " (Overruns: %lu)\n",
1132					trace->overrun);
1133		if (!ret)
1134			return TRACE_TYPE_PARTIAL_LINE;
1135	}
1136
1137	ret = print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1138			      cpu, pid, flags);
1139	if (ret == TRACE_TYPE_PARTIAL_LINE)
1140		return TRACE_TYPE_PARTIAL_LINE;
1141
1142	return TRACE_TYPE_HANDLED;
1143}
1144
1145static enum print_line_t
1146print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1147		    struct trace_iterator *iter, u32 flags)
1148{
1149	unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 
1150	struct fgraph_data *data = iter->private;
1151	struct trace_event *event;
1152	int depth = 0;
1153	int ret;
1154	int i;
1155
1156	if (data)
1157		depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1158
1159	if (print_graph_prologue(iter, s, 0, 0, flags))
1160		return TRACE_TYPE_PARTIAL_LINE;
1161
1162	/* No time */
1163	ret = print_graph_duration(DURATION_FILL_FULL, s, flags);
1164	if (ret != TRACE_TYPE_HANDLED)
1165		return ret;
1166
1167	/* Indentation */
1168	if (depth > 0)
1169		for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++) {
1170			ret = trace_seq_printf(s, " ");
1171			if (!ret)
1172				return TRACE_TYPE_PARTIAL_LINE;
1173		}
1174
1175	/* The comment */
1176	ret = trace_seq_printf(s, "/* ");
1177	if (!ret)
1178		return TRACE_TYPE_PARTIAL_LINE;
1179
1180	switch (iter->ent->type) {
1181	case TRACE_BPRINT:
1182		ret = trace_print_bprintk_msg_only(iter);
1183		if (ret != TRACE_TYPE_HANDLED)
1184			return ret;
1185		break;
1186	case TRACE_PRINT:
1187		ret = trace_print_printk_msg_only(iter);
1188		if (ret != TRACE_TYPE_HANDLED)
1189			return ret;
1190		break;
1191	default:
1192		event = ftrace_find_event(ent->type);
1193		if (!event)
1194			return TRACE_TYPE_UNHANDLED;
1195
1196		ret = event->funcs->trace(iter, sym_flags, event);
1197		if (ret != TRACE_TYPE_HANDLED)
1198			return ret;
1199	}
1200
 
 
 
1201	/* Strip ending newline */
1202	if (s->buffer[s->len - 1] == '\n') {
1203		s->buffer[s->len - 1] = '\0';
1204		s->len--;
1205	}
1206
1207	ret = trace_seq_printf(s, " */\n");
1208	if (!ret)
1209		return TRACE_TYPE_PARTIAL_LINE;
1210
1211	return TRACE_TYPE_HANDLED;
1212}
1213
1214
1215enum print_line_t
1216print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1217{
1218	struct ftrace_graph_ent_entry *field;
1219	struct fgraph_data *data = iter->private;
1220	struct trace_entry *entry = iter->ent;
1221	struct trace_seq *s = &iter->seq;
1222	int cpu = iter->cpu;
1223	int ret;
1224
1225	if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1226		per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1227		return TRACE_TYPE_HANDLED;
1228	}
1229
1230	/*
1231	 * If the last output failed, there's a possibility we need
1232	 * to print out the missing entry which would never go out.
1233	 */
1234	if (data && data->failed) {
1235		field = &data->ent;
1236		iter->cpu = data->cpu;
1237		ret = print_graph_entry(field, s, iter, flags);
1238		if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1239			per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1240			ret = TRACE_TYPE_NO_CONSUME;
1241		}
1242		iter->cpu = cpu;
1243		return ret;
1244	}
1245
1246	switch (entry->type) {
1247	case TRACE_GRAPH_ENT: {
1248		/*
1249		 * print_graph_entry() may consume the current event,
1250		 * thus @field may become invalid, so we need to save it.
1251		 * sizeof(struct ftrace_graph_ent_entry) is very small,
1252		 * it can be safely saved at the stack.
1253		 */
1254		struct ftrace_graph_ent_entry saved;
1255		trace_assign_type(field, entry);
1256		saved = *field;
1257		return print_graph_entry(&saved, s, iter, flags);
1258	}
1259	case TRACE_GRAPH_RET: {
1260		struct ftrace_graph_ret_entry *field;
1261		trace_assign_type(field, entry);
1262		return print_graph_return(&field->ret, s, entry, iter, flags);
1263	}
1264	case TRACE_STACK:
1265	case TRACE_FN:
1266		/* dont trace stack and functions as comments */
1267		return TRACE_TYPE_UNHANDLED;
1268
1269	default:
1270		return print_graph_comment(s, entry, iter, flags);
1271	}
1272
1273	return TRACE_TYPE_HANDLED;
1274}
1275
1276static enum print_line_t
1277print_graph_function(struct trace_iterator *iter)
1278{
1279	return print_graph_function_flags(iter, tracer_flags.val);
1280}
1281
1282static enum print_line_t
1283print_graph_function_event(struct trace_iterator *iter, int flags,
1284			   struct trace_event *event)
1285{
1286	return print_graph_function(iter);
1287}
1288
1289static void print_lat_header(struct seq_file *s, u32 flags)
1290{
1291	static const char spaces[] = "                "	/* 16 spaces */
1292		"    "					/* 4 spaces */
1293		"                 ";			/* 17 spaces */
1294	int size = 0;
1295
1296	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1297		size += 16;
1298	if (flags & TRACE_GRAPH_PRINT_CPU)
1299		size += 4;
1300	if (flags & TRACE_GRAPH_PRINT_PROC)
1301		size += 17;
1302
1303	seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1304	seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1305	seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1306	seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1307	seq_printf(s, "#%.*s||| /                      \n", size, spaces);
1308}
1309
1310static void __print_graph_headers_flags(struct seq_file *s, u32 flags)
 
1311{
1312	int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
1313
1314	if (lat)
1315		print_lat_header(s, flags);
1316
1317	/* 1st line */
1318	seq_printf(s, "#");
1319	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1320		seq_printf(s, "     TIME       ");
1321	if (flags & TRACE_GRAPH_PRINT_CPU)
1322		seq_printf(s, " CPU");
1323	if (flags & TRACE_GRAPH_PRINT_PROC)
1324		seq_printf(s, "  TASK/PID       ");
1325	if (lat)
1326		seq_printf(s, "||||");
1327	if (flags & TRACE_GRAPH_PRINT_DURATION)
1328		seq_printf(s, "  DURATION   ");
1329	seq_printf(s, "               FUNCTION CALLS\n");
1330
1331	/* 2nd line */
1332	seq_printf(s, "#");
1333	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1334		seq_printf(s, "      |         ");
1335	if (flags & TRACE_GRAPH_PRINT_CPU)
1336		seq_printf(s, " |  ");
1337	if (flags & TRACE_GRAPH_PRINT_PROC)
1338		seq_printf(s, "   |    |        ");
1339	if (lat)
1340		seq_printf(s, "||||");
1341	if (flags & TRACE_GRAPH_PRINT_DURATION)
1342		seq_printf(s, "   |   |      ");
1343	seq_printf(s, "               |   |   |   |\n");
1344}
1345
1346void print_graph_headers(struct seq_file *s)
1347{
1348	print_graph_headers_flags(s, tracer_flags.val);
1349}
1350
1351void print_graph_headers_flags(struct seq_file *s, u32 flags)
1352{
1353	struct trace_iterator *iter = s->private;
 
1354
1355	if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
1356		return;
1357
1358	if (trace_flags & TRACE_ITER_LATENCY_FMT) {
1359		/* print nothing if the buffers are empty */
1360		if (trace_empty(iter))
1361			return;
1362
1363		print_trace_header(s, iter);
1364	}
1365
1366	__print_graph_headers_flags(s, flags);
1367}
1368
1369void graph_trace_open(struct trace_iterator *iter)
1370{
1371	/* pid and depth on the last trace processed */
1372	struct fgraph_data *data;
 
1373	int cpu;
1374
1375	iter->private = NULL;
1376
1377	data = kzalloc(sizeof(*data), GFP_KERNEL);
 
 
 
1378	if (!data)
1379		goto out_err;
1380
1381	data->cpu_data = alloc_percpu(struct fgraph_cpu_data);
1382	if (!data->cpu_data)
1383		goto out_err_free;
1384
1385	for_each_possible_cpu(cpu) {
1386		pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1387		int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1388		int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1389		int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1390
1391		*pid = -1;
1392		*depth = 0;
1393		*ignore = 0;
1394		*depth_irq = -1;
1395	}
1396
1397	iter->private = data;
1398
1399	return;
1400
1401 out_err_free:
1402	kfree(data);
1403 out_err:
1404	pr_warning("function graph tracer: not enough memory\n");
1405}
1406
1407void graph_trace_close(struct trace_iterator *iter)
1408{
1409	struct fgraph_data *data = iter->private;
1410
1411	if (data) {
1412		free_percpu(data->cpu_data);
1413		kfree(data);
1414	}
1415}
1416
1417static int func_graph_set_flag(u32 old_flags, u32 bit, int set)
 
1418{
1419	if (bit == TRACE_GRAPH_PRINT_IRQS)
1420		ftrace_graph_skip_irqs = !set;
1421
 
 
 
 
 
 
1422	return 0;
1423}
1424
1425static struct trace_event_functions graph_functions = {
1426	.trace		= print_graph_function_event,
1427};
1428
1429static struct trace_event graph_trace_entry_event = {
1430	.type		= TRACE_GRAPH_ENT,
1431	.funcs		= &graph_functions,
1432};
1433
1434static struct trace_event graph_trace_ret_event = {
1435	.type		= TRACE_GRAPH_RET,
1436	.funcs		= &graph_functions
1437};
1438
1439static struct tracer graph_trace __read_mostly = {
1440	.name		= "function_graph",
 
1441	.open		= graph_trace_open,
1442	.pipe_open	= graph_trace_open,
1443	.close		= graph_trace_close,
1444	.pipe_close	= graph_trace_close,
1445	.wait_pipe	= poll_wait_pipe,
1446	.init		= graph_trace_init,
1447	.reset		= graph_trace_reset,
1448	.print_line	= print_graph_function,
1449	.print_header	= print_graph_headers,
1450	.flags		= &tracer_flags,
1451	.set_flag	= func_graph_set_flag,
1452#ifdef CONFIG_FTRACE_SELFTEST
1453	.selftest	= trace_selftest_startup_function_graph,
1454#endif
1455};
1456
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1457static __init int init_graph_trace(void)
1458{
1459	max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1460
1461	if (!register_ftrace_event(&graph_trace_entry_event)) {
1462		pr_warning("Warning: could not register graph trace events\n");
1463		return 1;
1464	}
1465
1466	if (!register_ftrace_event(&graph_trace_ret_event)) {
1467		pr_warning("Warning: could not register graph trace events\n");
1468		return 1;
1469	}
1470
1471	return register_tracer(&graph_trace);
1472}
1473
1474device_initcall(init_graph_trace);