Linux Audio

Check our new training course

Loading...
v4.17
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *
   4 * Function graph tracer.
   5 * Copyright (c) 2008-2009 Frederic Weisbecker <fweisbec@gmail.com>
   6 * Mostly borrowed from function tracer which
   7 * is Copyright (c) Steven Rostedt <srostedt@redhat.com>
   8 *
   9 */
  10#include <linux/uaccess.h>
  11#include <linux/ftrace.h>
  12#include <linux/interrupt.h>
  13#include <linux/slab.h>
  14#include <linux/fs.h>
  15
  16#include "trace.h"
  17#include "trace_output.h"
  18
  19static bool kill_ftrace_graph;
  20
  21/**
  22 * ftrace_graph_is_dead - returns true if ftrace_graph_stop() was called
  23 *
  24 * ftrace_graph_stop() is called when a severe error is detected in
  25 * the function graph tracing. This function is called by the critical
  26 * paths of function graph to keep those paths from doing any more harm.
  27 */
  28bool ftrace_graph_is_dead(void)
  29{
  30	return kill_ftrace_graph;
  31}
  32
  33/**
  34 * ftrace_graph_stop - set to permanently disable function graph tracincg
  35 *
  36 * In case of an error int function graph tracing, this is called
  37 * to try to keep function graph tracing from causing any more harm.
  38 * Usually this is pretty severe and this is called to try to at least
  39 * get a warning out to the user.
  40 */
  41void ftrace_graph_stop(void)
  42{
  43	kill_ftrace_graph = true;
  44}
  45
  46/* When set, irq functions will be ignored */
  47static int ftrace_graph_skip_irqs;
  48
  49struct fgraph_cpu_data {
  50	pid_t		last_pid;
  51	int		depth;
  52	int		depth_irq;
  53	int		ignore;
  54	unsigned long	enter_funcs[FTRACE_RETFUNC_DEPTH];
  55};
  56
  57struct fgraph_data {
  58	struct fgraph_cpu_data __percpu *cpu_data;
  59
  60	/* Place to preserve last processed entry. */
  61	struct ftrace_graph_ent_entry	ent;
  62	struct ftrace_graph_ret_entry	ret;
  63	int				failed;
  64	int				cpu;
  65};
  66
  67#define TRACE_GRAPH_INDENT	2
  68
  69unsigned int fgraph_max_depth;
  70
  71static struct tracer_opt trace_opts[] = {
  72	/* Display overruns? (for self-debug purpose) */
  73	{ TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
  74	/* Display CPU ? */
  75	{ TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
  76	/* Display Overhead ? */
  77	{ TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
  78	/* Display proc name/pid */
  79	{ TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
  80	/* Display duration of execution */
  81	{ TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
  82	/* Display absolute time of an entry */
  83	{ TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
  84	/* Display interrupts */
  85	{ TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
  86	/* Display function name after trailing } */
  87	{ TRACER_OPT(funcgraph-tail, TRACE_GRAPH_PRINT_TAIL) },
  88	/* Include sleep time (scheduled out) between entry and return */
  89	{ TRACER_OPT(sleep-time, TRACE_GRAPH_SLEEP_TIME) },
  90	/* Include time within nested functions */
  91	{ TRACER_OPT(graph-time, TRACE_GRAPH_GRAPH_TIME) },
  92	{ } /* Empty entry */
  93};
  94
  95static struct tracer_flags tracer_flags = {
  96	/* Don't display overruns, proc, or tail by default */
  97	.val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
  98	       TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS |
  99	       TRACE_GRAPH_SLEEP_TIME | TRACE_GRAPH_GRAPH_TIME,
 100	.opts = trace_opts
 101};
 102
 103static struct trace_array *graph_array;
 104
 105/*
 106 * DURATION column is being also used to display IRQ signs,
 107 * following values are used by print_graph_irq and others
 108 * to fill in space into DURATION column.
 109 */
 110enum {
 111	FLAGS_FILL_FULL  = 1 << TRACE_GRAPH_PRINT_FILL_SHIFT,
 112	FLAGS_FILL_START = 2 << TRACE_GRAPH_PRINT_FILL_SHIFT,
 113	FLAGS_FILL_END   = 3 << TRACE_GRAPH_PRINT_FILL_SHIFT,
 114};
 115
 116static void
 117print_graph_duration(struct trace_array *tr, unsigned long long duration,
 118		     struct trace_seq *s, u32 flags);
 119
 120/* Add a function return address to the trace stack on thread info.*/
 121int
 122ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
 123			 unsigned long frame_pointer, unsigned long *retp)
 124{
 125	unsigned long long calltime;
 126	int index;
 127
 128	if (unlikely(ftrace_graph_is_dead()))
 129		return -EBUSY;
 130
 131	if (!current->ret_stack)
 132		return -EBUSY;
 133
 134	/*
 135	 * We must make sure the ret_stack is tested before we read
 136	 * anything else.
 137	 */
 138	smp_rmb();
 139
 140	/* The return trace stack is full */
 141	if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
 142		atomic_inc(&current->trace_overrun);
 143		return -EBUSY;
 144	}
 145
 146	/*
 147	 * The curr_ret_stack is an index to ftrace return stack of
 148	 * current task.  Its value should be in [0, FTRACE_RETFUNC_
 149	 * DEPTH) when the function graph tracer is used.  To support
 150	 * filtering out specific functions, it makes the index
 151	 * negative by subtracting huge value (FTRACE_NOTRACE_DEPTH)
 152	 * so when it sees a negative index the ftrace will ignore
 153	 * the record.  And the index gets recovered when returning
 154	 * from the filtered function by adding the FTRACE_NOTRACE_
 155	 * DEPTH and then it'll continue to record functions normally.
 156	 *
 157	 * The curr_ret_stack is initialized to -1 and get increased
 158	 * in this function.  So it can be less than -1 only if it was
 159	 * filtered out via ftrace_graph_notrace_addr() which can be
 160	 * set from set_graph_notrace file in tracefs by user.
 161	 */
 162	if (current->curr_ret_stack < -1)
 163		return -EBUSY;
 164
 165	calltime = trace_clock_local();
 166
 167	index = ++current->curr_ret_stack;
 168	if (ftrace_graph_notrace_addr(func))
 169		current->curr_ret_stack -= FTRACE_NOTRACE_DEPTH;
 170	barrier();
 171	current->ret_stack[index].ret = ret;
 172	current->ret_stack[index].func = func;
 173	current->ret_stack[index].calltime = calltime;
 174#ifdef HAVE_FUNCTION_GRAPH_FP_TEST
 175	current->ret_stack[index].fp = frame_pointer;
 176#endif
 177#ifdef HAVE_FUNCTION_GRAPH_RET_ADDR_PTR
 178	current->ret_stack[index].retp = retp;
 179#endif
 180	*depth = current->curr_ret_stack;
 181
 182	return 0;
 183}
 184
 185/* Retrieve a function return address to the trace stack on thread info.*/
 186static void
 187ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
 188			unsigned long frame_pointer)
 189{
 190	int index;
 191
 192	index = current->curr_ret_stack;
 193
 194	/*
 195	 * A negative index here means that it's just returned from a
 196	 * notrace'd function.  Recover index to get an original
 197	 * return address.  See ftrace_push_return_trace().
 198	 *
 199	 * TODO: Need to check whether the stack gets corrupted.
 200	 */
 201	if (index < 0)
 202		index += FTRACE_NOTRACE_DEPTH;
 203
 204	if (unlikely(index < 0 || index >= FTRACE_RETFUNC_DEPTH)) {
 205		ftrace_graph_stop();
 206		WARN_ON(1);
 207		/* Might as well panic, otherwise we have no where to go */
 208		*ret = (unsigned long)panic;
 209		return;
 210	}
 211
 212#ifdef HAVE_FUNCTION_GRAPH_FP_TEST
 213	/*
 214	 * The arch may choose to record the frame pointer used
 215	 * and check it here to make sure that it is what we expect it
 216	 * to be. If gcc does not set the place holder of the return
 217	 * address in the frame pointer, and does a copy instead, then
 218	 * the function graph trace will fail. This test detects this
 219	 * case.
 220	 *
 221	 * Currently, x86_32 with optimize for size (-Os) makes the latest
 222	 * gcc do the above.
 223	 *
 224	 * Note, -mfentry does not use frame pointers, and this test
 225	 *  is not needed if CC_USING_FENTRY is set.
 226	 */
 227	if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
 228		ftrace_graph_stop();
 229		WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
 230		     "  from func %ps return to %lx\n",
 231		     current->ret_stack[index].fp,
 232		     frame_pointer,
 233		     (void *)current->ret_stack[index].func,
 234		     current->ret_stack[index].ret);
 235		*ret = (unsigned long)panic;
 236		return;
 237	}
 238#endif
 239
 240	*ret = current->ret_stack[index].ret;
 241	trace->func = current->ret_stack[index].func;
 242	trace->calltime = current->ret_stack[index].calltime;
 243	trace->overrun = atomic_read(&current->trace_overrun);
 244	trace->depth = index;
 245}
 246
 247/*
 248 * Send the trace to the ring-buffer.
 249 * @return the original return address.
 250 */
 251unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
 252{
 253	struct ftrace_graph_ret trace;
 254	unsigned long ret;
 255
 256	ftrace_pop_return_trace(&trace, &ret, frame_pointer);
 257	trace.rettime = trace_clock_local();
 258	barrier();
 259	current->curr_ret_stack--;
 260	/*
 261	 * The curr_ret_stack can be less than -1 only if it was
 262	 * filtered out and it's about to return from the function.
 263	 * Recover the index and continue to trace normal functions.
 264	 */
 265	if (current->curr_ret_stack < -1) {
 266		current->curr_ret_stack += FTRACE_NOTRACE_DEPTH;
 267		return ret;
 268	}
 269
 270	/*
 271	 * The trace should run after decrementing the ret counter
 272	 * in case an interrupt were to come in. We don't want to
 273	 * lose the interrupt if max_depth is set.
 274	 */
 275	ftrace_graph_return(&trace);
 276
 277	if (unlikely(!ret)) {
 278		ftrace_graph_stop();
 279		WARN_ON(1);
 280		/* Might as well panic. What else to do? */
 281		ret = (unsigned long)panic;
 282	}
 283
 284	return ret;
 285}
 286
 287/**
 288 * ftrace_graph_ret_addr - convert a potentially modified stack return address
 289 *			   to its original value
 290 *
 291 * This function can be called by stack unwinding code to convert a found stack
 292 * return address ('ret') to its original value, in case the function graph
 293 * tracer has modified it to be 'return_to_handler'.  If the address hasn't
 294 * been modified, the unchanged value of 'ret' is returned.
 295 *
 296 * 'idx' is a state variable which should be initialized by the caller to zero
 297 * before the first call.
 298 *
 299 * 'retp' is a pointer to the return address on the stack.  It's ignored if
 300 * the arch doesn't have HAVE_FUNCTION_GRAPH_RET_ADDR_PTR defined.
 301 */
 302#ifdef HAVE_FUNCTION_GRAPH_RET_ADDR_PTR
 303unsigned long ftrace_graph_ret_addr(struct task_struct *task, int *idx,
 304				    unsigned long ret, unsigned long *retp)
 305{
 306	int index = task->curr_ret_stack;
 307	int i;
 308
 309	if (ret != (unsigned long)return_to_handler)
 310		return ret;
 311
 312	if (index < -1)
 313		index += FTRACE_NOTRACE_DEPTH;
 314
 315	if (index < 0)
 316		return ret;
 317
 318	for (i = 0; i <= index; i++)
 319		if (task->ret_stack[i].retp == retp)
 320			return task->ret_stack[i].ret;
 321
 322	return ret;
 323}
 324#else /* !HAVE_FUNCTION_GRAPH_RET_ADDR_PTR */
 325unsigned long ftrace_graph_ret_addr(struct task_struct *task, int *idx,
 326				    unsigned long ret, unsigned long *retp)
 327{
 328	int task_idx;
 329
 330	if (ret != (unsigned long)return_to_handler)
 331		return ret;
 332
 333	task_idx = task->curr_ret_stack;
 334
 335	if (!task->ret_stack || task_idx < *idx)
 336		return ret;
 337
 338	task_idx -= *idx;
 339	(*idx)++;
 340
 341	return task->ret_stack[task_idx].ret;
 342}
 343#endif /* HAVE_FUNCTION_GRAPH_RET_ADDR_PTR */
 344
 345int __trace_graph_entry(struct trace_array *tr,
 346				struct ftrace_graph_ent *trace,
 347				unsigned long flags,
 348				int pc)
 349{
 350	struct trace_event_call *call = &event_funcgraph_entry;
 351	struct ring_buffer_event *event;
 352	struct ring_buffer *buffer = tr->trace_buffer.buffer;
 353	struct ftrace_graph_ent_entry *entry;
 354
 355	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
 356					  sizeof(*entry), flags, pc);
 357	if (!event)
 358		return 0;
 359	entry	= ring_buffer_event_data(event);
 360	entry->graph_ent			= *trace;
 361	if (!call_filter_check_discard(call, entry, buffer, event))
 362		trace_buffer_unlock_commit_nostack(buffer, event);
 363
 364	return 1;
 365}
 366
 367static inline int ftrace_graph_ignore_irqs(void)
 368{
 369	if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
 370		return 0;
 371
 372	return in_irq();
 373}
 374
 375int trace_graph_entry(struct ftrace_graph_ent *trace)
 376{
 377	struct trace_array *tr = graph_array;
 378	struct trace_array_cpu *data;
 379	unsigned long flags;
 380	long disabled;
 381	int ret;
 382	int cpu;
 383	int pc;
 384
 385	if (!ftrace_trace_task(tr))
 386		return 0;
 387
 388	if (ftrace_graph_ignore_func(trace))
 389		return 0;
 390
 391	if (ftrace_graph_ignore_irqs())
 
 
 
 392		return 0;
 393
 394	/*
 395	 * Do not trace a function if it's filtered by set_graph_notrace.
 396	 * Make the index of ret stack negative to indicate that it should
 397	 * ignore further functions.  But it needs its own ret stack entry
 398	 * to recover the original index in order to continue tracing after
 399	 * returning from the function.
 400	 */
 401	if (ftrace_graph_notrace_addr(trace->func))
 402		return 1;
 403
 404	/*
 405	 * Stop here if tracing_threshold is set. We only write function return
 406	 * events to the ring buffer.
 407	 */
 408	if (tracing_thresh)
 409		return 1;
 410
 411	local_irq_save(flags);
 412	cpu = raw_smp_processor_id();
 413	data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 414	disabled = atomic_inc_return(&data->disabled);
 415	if (likely(disabled == 1)) {
 416		pc = preempt_count();
 417		ret = __trace_graph_entry(tr, trace, flags, pc);
 418	} else {
 419		ret = 0;
 420	}
 421
 422	atomic_dec(&data->disabled);
 423	local_irq_restore(flags);
 424
 425	return ret;
 426}
 427
 
 
 
 
 
 
 
 
 428static void
 429__trace_graph_function(struct trace_array *tr,
 430		unsigned long ip, unsigned long flags, int pc)
 431{
 432	u64 time = trace_clock_local();
 433	struct ftrace_graph_ent ent = {
 434		.func  = ip,
 435		.depth = 0,
 436	};
 437	struct ftrace_graph_ret ret = {
 438		.func     = ip,
 439		.depth    = 0,
 440		.calltime = time,
 441		.rettime  = time,
 442	};
 443
 444	__trace_graph_entry(tr, &ent, flags, pc);
 445	__trace_graph_return(tr, &ret, flags, pc);
 446}
 447
 448void
 449trace_graph_function(struct trace_array *tr,
 450		unsigned long ip, unsigned long parent_ip,
 451		unsigned long flags, int pc)
 452{
 453	__trace_graph_function(tr, ip, flags, pc);
 454}
 455
 456void __trace_graph_return(struct trace_array *tr,
 457				struct ftrace_graph_ret *trace,
 458				unsigned long flags,
 459				int pc)
 460{
 461	struct trace_event_call *call = &event_funcgraph_exit;
 462	struct ring_buffer_event *event;
 463	struct ring_buffer *buffer = tr->trace_buffer.buffer;
 464	struct ftrace_graph_ret_entry *entry;
 465
 466	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
 467					  sizeof(*entry), flags, pc);
 468	if (!event)
 469		return;
 470	entry	= ring_buffer_event_data(event);
 471	entry->ret				= *trace;
 472	if (!call_filter_check_discard(call, entry, buffer, event))
 473		trace_buffer_unlock_commit_nostack(buffer, event);
 474}
 475
 476void trace_graph_return(struct ftrace_graph_ret *trace)
 477{
 478	struct trace_array *tr = graph_array;
 479	struct trace_array_cpu *data;
 480	unsigned long flags;
 481	long disabled;
 482	int cpu;
 483	int pc;
 484
 485	local_irq_save(flags);
 486	cpu = raw_smp_processor_id();
 487	data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 488	disabled = atomic_inc_return(&data->disabled);
 489	if (likely(disabled == 1)) {
 490		pc = preempt_count();
 491		__trace_graph_return(tr, trace, flags, pc);
 492	}
 493	atomic_dec(&data->disabled);
 494	local_irq_restore(flags);
 495}
 496
 497void set_graph_array(struct trace_array *tr)
 498{
 499	graph_array = tr;
 500
 501	/* Make graph_array visible before we start tracing */
 502
 503	smp_mb();
 504}
 505
 506static void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
 507{
 508	if (tracing_thresh &&
 509	    (trace->rettime - trace->calltime < tracing_thresh))
 510		return;
 511	else
 512		trace_graph_return(trace);
 513}
 514
 515static int graph_trace_init(struct trace_array *tr)
 516{
 517	int ret;
 518
 519	set_graph_array(tr);
 520	if (tracing_thresh)
 521		ret = register_ftrace_graph(&trace_graph_thresh_return,
 522					    &trace_graph_entry);
 523	else
 524		ret = register_ftrace_graph(&trace_graph_return,
 525					    &trace_graph_entry);
 526	if (ret)
 527		return ret;
 528	tracing_start_cmdline_record();
 529
 530	return 0;
 531}
 532
 533static void graph_trace_reset(struct trace_array *tr)
 534{
 535	tracing_stop_cmdline_record();
 536	unregister_ftrace_graph();
 537}
 538
 539static int graph_trace_update_thresh(struct trace_array *tr)
 540{
 541	graph_trace_reset(tr);
 542	return graph_trace_init(tr);
 543}
 544
 545static int max_bytes_for_cpu;
 546
 547static void print_graph_cpu(struct trace_seq *s, int cpu)
 548{
 549	/*
 550	 * Start with a space character - to make it stand out
 551	 * to the right a bit when trace output is pasted into
 552	 * email:
 553	 */
 554	trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
 555}
 556
 557#define TRACE_GRAPH_PROCINFO_LENGTH	14
 558
 559static void print_graph_proc(struct trace_seq *s, pid_t pid)
 560{
 561	char comm[TASK_COMM_LEN];
 562	/* sign + log10(MAX_INT) + '\0' */
 563	char pid_str[11];
 564	int spaces = 0;
 565	int len;
 566	int i;
 567
 568	trace_find_cmdline(pid, comm);
 569	comm[7] = '\0';
 570	sprintf(pid_str, "%d", pid);
 571
 572	/* 1 stands for the "-" character */
 573	len = strlen(comm) + strlen(pid_str) + 1;
 574
 575	if (len < TRACE_GRAPH_PROCINFO_LENGTH)
 576		spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
 577
 578	/* First spaces to align center */
 579	for (i = 0; i < spaces / 2; i++)
 580		trace_seq_putc(s, ' ');
 581
 582	trace_seq_printf(s, "%s-%s", comm, pid_str);
 583
 584	/* Last spaces to align center */
 585	for (i = 0; i < spaces - (spaces / 2); i++)
 586		trace_seq_putc(s, ' ');
 587}
 588
 589
 590static void print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
 591{
 592	trace_seq_putc(s, ' ');
 593	trace_print_lat_fmt(s, entry);
 594}
 595
 596/* If the pid changed since the last trace, output this event */
 597static void
 598verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
 599{
 600	pid_t prev_pid;
 601	pid_t *last_pid;
 602
 603	if (!data)
 604		return;
 605
 606	last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
 607
 608	if (*last_pid == pid)
 609		return;
 610
 611	prev_pid = *last_pid;
 612	*last_pid = pid;
 613
 614	if (prev_pid == -1)
 615		return;
 616/*
 617 * Context-switch trace line:
 618
 619 ------------------------------------------
 620 | 1)  migration/0--1  =>  sshd-1755
 621 ------------------------------------------
 622
 623 */
 624	trace_seq_puts(s, " ------------------------------------------\n");
 625	print_graph_cpu(s, cpu);
 626	print_graph_proc(s, prev_pid);
 627	trace_seq_puts(s, " => ");
 628	print_graph_proc(s, pid);
 629	trace_seq_puts(s, "\n ------------------------------------------\n\n");
 630}
 631
 632static struct ftrace_graph_ret_entry *
 633get_return_for_leaf(struct trace_iterator *iter,
 634		struct ftrace_graph_ent_entry *curr)
 635{
 636	struct fgraph_data *data = iter->private;
 637	struct ring_buffer_iter *ring_iter = NULL;
 638	struct ring_buffer_event *event;
 639	struct ftrace_graph_ret_entry *next;
 640
 641	/*
 642	 * If the previous output failed to write to the seq buffer,
 643	 * then we just reuse the data from before.
 644	 */
 645	if (data && data->failed) {
 646		curr = &data->ent;
 647		next = &data->ret;
 648	} else {
 649
 650		ring_iter = trace_buffer_iter(iter, iter->cpu);
 651
 652		/* First peek to compare current entry and the next one */
 653		if (ring_iter)
 654			event = ring_buffer_iter_peek(ring_iter, NULL);
 655		else {
 656			/*
 657			 * We need to consume the current entry to see
 658			 * the next one.
 659			 */
 660			ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu,
 661					    NULL, NULL);
 662			event = ring_buffer_peek(iter->trace_buffer->buffer, iter->cpu,
 663						 NULL, NULL);
 664		}
 665
 666		if (!event)
 667			return NULL;
 668
 669		next = ring_buffer_event_data(event);
 670
 671		if (data) {
 672			/*
 673			 * Save current and next entries for later reference
 674			 * if the output fails.
 675			 */
 676			data->ent = *curr;
 677			/*
 678			 * If the next event is not a return type, then
 679			 * we only care about what type it is. Otherwise we can
 680			 * safely copy the entire event.
 681			 */
 682			if (next->ent.type == TRACE_GRAPH_RET)
 683				data->ret = *next;
 684			else
 685				data->ret.ent.type = next->ent.type;
 686		}
 687	}
 688
 689	if (next->ent.type != TRACE_GRAPH_RET)
 690		return NULL;
 691
 692	if (curr->ent.pid != next->ent.pid ||
 693			curr->graph_ent.func != next->ret.func)
 694		return NULL;
 695
 696	/* this is a leaf, now advance the iterator */
 697	if (ring_iter)
 698		ring_buffer_read(ring_iter, NULL);
 699
 700	return next;
 701}
 702
 703static void print_graph_abs_time(u64 t, struct trace_seq *s)
 704{
 705	unsigned long usecs_rem;
 706
 707	usecs_rem = do_div(t, NSEC_PER_SEC);
 708	usecs_rem /= 1000;
 709
 710	trace_seq_printf(s, "%5lu.%06lu |  ",
 711			 (unsigned long)t, usecs_rem);
 712}
 713
 714static void
 715print_graph_irq(struct trace_iterator *iter, unsigned long addr,
 716		enum trace_type type, int cpu, pid_t pid, u32 flags)
 717{
 718	struct trace_array *tr = iter->tr;
 719	struct trace_seq *s = &iter->seq;
 720	struct trace_entry *ent = iter->ent;
 721
 722	if (addr < (unsigned long)__irqentry_text_start ||
 723		addr >= (unsigned long)__irqentry_text_end)
 724		return;
 725
 726	if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
 727		/* Absolute time */
 728		if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
 729			print_graph_abs_time(iter->ts, s);
 730
 731		/* Cpu */
 732		if (flags & TRACE_GRAPH_PRINT_CPU)
 733			print_graph_cpu(s, cpu);
 734
 735		/* Proc */
 736		if (flags & TRACE_GRAPH_PRINT_PROC) {
 737			print_graph_proc(s, pid);
 738			trace_seq_puts(s, " | ");
 739		}
 740
 741		/* Latency format */
 742		if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
 743			print_graph_lat_fmt(s, ent);
 744	}
 745
 746	/* No overhead */
 747	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_START);
 748
 749	if (type == TRACE_GRAPH_ENT)
 750		trace_seq_puts(s, "==========>");
 751	else
 752		trace_seq_puts(s, "<==========");
 753
 754	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_END);
 755	trace_seq_putc(s, '\n');
 756}
 757
 758void
 759trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
 760{
 761	unsigned long nsecs_rem = do_div(duration, 1000);
 762	/* log10(ULONG_MAX) + '\0' */
 763	char usecs_str[21];
 764	char nsecs_str[5];
 765	int len;
 766	int i;
 767
 768	sprintf(usecs_str, "%lu", (unsigned long) duration);
 769
 770	/* Print msecs */
 771	trace_seq_printf(s, "%s", usecs_str);
 772
 773	len = strlen(usecs_str);
 774
 775	/* Print nsecs (we don't want to exceed 7 numbers) */
 776	if (len < 7) {
 777		size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
 778
 779		snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
 780		trace_seq_printf(s, ".%s", nsecs_str);
 781		len += strlen(nsecs_str) + 1;
 782	}
 783
 784	trace_seq_puts(s, " us ");
 785
 786	/* Print remaining spaces to fit the row's width */
 787	for (i = len; i < 8; i++)
 788		trace_seq_putc(s, ' ');
 789}
 790
 791static void
 792print_graph_duration(struct trace_array *tr, unsigned long long duration,
 793		     struct trace_seq *s, u32 flags)
 794{
 795	if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
 796	    !(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
 797		return;
 798
 799	/* No real adata, just filling the column with spaces */
 800	switch (flags & TRACE_GRAPH_PRINT_FILL_MASK) {
 801	case FLAGS_FILL_FULL:
 802		trace_seq_puts(s, "              |  ");
 803		return;
 804	case FLAGS_FILL_START:
 805		trace_seq_puts(s, "  ");
 806		return;
 807	case FLAGS_FILL_END:
 808		trace_seq_puts(s, " |");
 809		return;
 810	}
 811
 812	/* Signal a overhead of time execution to the output */
 813	if (flags & TRACE_GRAPH_PRINT_OVERHEAD)
 814		trace_seq_printf(s, "%c ", trace_find_mark(duration));
 815	else
 816		trace_seq_puts(s, "  ");
 817
 818	trace_print_graph_duration(duration, s);
 819	trace_seq_puts(s, "|  ");
 820}
 821
 822/* Case of a leaf function on its call entry */
 823static enum print_line_t
 824print_graph_entry_leaf(struct trace_iterator *iter,
 825		struct ftrace_graph_ent_entry *entry,
 826		struct ftrace_graph_ret_entry *ret_entry,
 827		struct trace_seq *s, u32 flags)
 828{
 829	struct fgraph_data *data = iter->private;
 830	struct trace_array *tr = iter->tr;
 831	struct ftrace_graph_ret *graph_ret;
 832	struct ftrace_graph_ent *call;
 833	unsigned long long duration;
 834	int i;
 835
 836	graph_ret = &ret_entry->ret;
 837	call = &entry->graph_ent;
 838	duration = graph_ret->rettime - graph_ret->calltime;
 839
 840	if (data) {
 841		struct fgraph_cpu_data *cpu_data;
 842		int cpu = iter->cpu;
 843
 844		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 845
 846		/* If a graph tracer ignored set_graph_notrace */
 847		if (call->depth < -1)
 848			call->depth += FTRACE_NOTRACE_DEPTH;
 849
 850		/*
 851		 * Comments display at + 1 to depth. Since
 852		 * this is a leaf function, keep the comments
 853		 * equal to this depth.
 854		 */
 855		cpu_data->depth = call->depth - 1;
 856
 857		/* No need to keep this function around for this depth */
 858		if (call->depth < FTRACE_RETFUNC_DEPTH &&
 859		    !WARN_ON_ONCE(call->depth < 0))
 860			cpu_data->enter_funcs[call->depth] = 0;
 861	}
 862
 863	/* Overhead and duration */
 864	print_graph_duration(tr, duration, s, flags);
 865
 866	/* Function */
 867	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
 868		trace_seq_putc(s, ' ');
 869
 870	trace_seq_printf(s, "%ps();\n", (void *)call->func);
 871
 872	return trace_handle_return(s);
 873}
 874
 875static enum print_line_t
 876print_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	struct trace_array *tr = iter->tr;
 883	int i;
 884
 885	if (data) {
 886		struct fgraph_cpu_data *cpu_data;
 887		int cpu = iter->cpu;
 888
 889		/* If a graph tracer ignored set_graph_notrace */
 890		if (call->depth < -1)
 891			call->depth += FTRACE_NOTRACE_DEPTH;
 892
 893		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 894		cpu_data->depth = call->depth;
 895
 896		/* Save this function pointer to see if the exit matches */
 897		if (call->depth < FTRACE_RETFUNC_DEPTH &&
 898		    !WARN_ON_ONCE(call->depth < 0))
 899			cpu_data->enter_funcs[call->depth] = call->func;
 900	}
 901
 902	/* No time */
 903	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
 904
 905	/* Function */
 906	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
 907		trace_seq_putc(s, ' ');
 908
 909	trace_seq_printf(s, "%ps() {\n", (void *)call->func);
 910
 911	if (trace_seq_has_overflowed(s))
 912		return TRACE_TYPE_PARTIAL_LINE;
 913
 914	/*
 915	 * we already consumed the current entry to check the next one
 916	 * and see if this is a leaf.
 917	 */
 918	return TRACE_TYPE_NO_CONSUME;
 919}
 920
 921static void
 922print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
 923		     int type, unsigned long addr, u32 flags)
 924{
 925	struct fgraph_data *data = iter->private;
 926	struct trace_entry *ent = iter->ent;
 927	struct trace_array *tr = iter->tr;
 928	int cpu = iter->cpu;
 929
 930	/* Pid */
 931	verif_pid(s, ent->pid, cpu, data);
 932
 933	if (type)
 934		/* Interrupt */
 935		print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
 936
 937	if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
 938		return;
 939
 940	/* Absolute time */
 941	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
 942		print_graph_abs_time(iter->ts, s);
 943
 944	/* Cpu */
 945	if (flags & TRACE_GRAPH_PRINT_CPU)
 946		print_graph_cpu(s, cpu);
 947
 948	/* Proc */
 949	if (flags & TRACE_GRAPH_PRINT_PROC) {
 950		print_graph_proc(s, ent->pid);
 951		trace_seq_puts(s, " | ");
 952	}
 953
 954	/* Latency format */
 955	if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
 956		print_graph_lat_fmt(s, ent);
 957
 958	return;
 959}
 960
 961/*
 962 * Entry check for irq code
 963 *
 964 * returns 1 if
 965 *  - we are inside irq code
 966 *  - we just entered irq code
 967 *
 968 * retunns 0 if
 969 *  - funcgraph-interrupts option is set
 970 *  - we are not inside irq code
 971 */
 972static int
 973check_irq_entry(struct trace_iterator *iter, u32 flags,
 974		unsigned long addr, int depth)
 975{
 976	int cpu = iter->cpu;
 977	int *depth_irq;
 978	struct fgraph_data *data = iter->private;
 979
 980	/*
 981	 * If we are either displaying irqs, or we got called as
 982	 * a graph event and private data does not exist,
 983	 * then we bypass the irq check.
 984	 */
 985	if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
 986	    (!data))
 987		return 0;
 988
 989	depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
 990
 991	/*
 992	 * We are inside the irq code
 993	 */
 994	if (*depth_irq >= 0)
 995		return 1;
 996
 997	if ((addr < (unsigned long)__irqentry_text_start) ||
 998	    (addr >= (unsigned long)__irqentry_text_end))
 999		return 0;
1000
1001	/*
1002	 * We are entering irq code.
1003	 */
1004	*depth_irq = depth;
1005	return 1;
1006}
1007
1008/*
1009 * Return check for irq code
1010 *
1011 * returns 1 if
1012 *  - we are inside irq code
1013 *  - we just left irq code
1014 *
1015 * returns 0 if
1016 *  - funcgraph-interrupts option is set
1017 *  - we are not inside irq code
1018 */
1019static int
1020check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
1021{
1022	int cpu = iter->cpu;
1023	int *depth_irq;
1024	struct fgraph_data *data = iter->private;
1025
1026	/*
1027	 * If we are either displaying irqs, or we got called as
1028	 * a graph event and private data does not exist,
1029	 * then we bypass the irq check.
1030	 */
1031	if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
1032	    (!data))
1033		return 0;
1034
1035	depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1036
1037	/*
1038	 * We are not inside the irq code.
1039	 */
1040	if (*depth_irq == -1)
1041		return 0;
1042
1043	/*
1044	 * We are inside the irq code, and this is returning entry.
1045	 * Let's not trace it and clear the entry depth, since
1046	 * we are out of irq code.
1047	 *
1048	 * This condition ensures that we 'leave the irq code' once
1049	 * we are out of the entry depth. Thus protecting us from
1050	 * the RETURN entry loss.
1051	 */
1052	if (*depth_irq >= depth) {
1053		*depth_irq = -1;
1054		return 1;
1055	}
1056
1057	/*
1058	 * We are inside the irq code, and this is not the entry.
1059	 */
1060	return 1;
1061}
1062
1063static enum print_line_t
1064print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
1065			struct trace_iterator *iter, u32 flags)
1066{
1067	struct fgraph_data *data = iter->private;
1068	struct ftrace_graph_ent *call = &field->graph_ent;
1069	struct ftrace_graph_ret_entry *leaf_ret;
1070	static enum print_line_t ret;
1071	int cpu = iter->cpu;
1072
1073	if (check_irq_entry(iter, flags, call->func, call->depth))
1074		return TRACE_TYPE_HANDLED;
1075
1076	print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags);
1077
1078	leaf_ret = get_return_for_leaf(iter, field);
1079	if (leaf_ret)
1080		ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
1081	else
1082		ret = print_graph_entry_nested(iter, field, s, cpu, flags);
1083
1084	if (data) {
1085		/*
1086		 * If we failed to write our output, then we need to make
1087		 * note of it. Because we already consumed our entry.
1088		 */
1089		if (s->full) {
1090			data->failed = 1;
1091			data->cpu = cpu;
1092		} else
1093			data->failed = 0;
1094	}
1095
1096	return ret;
1097}
1098
1099static enum print_line_t
1100print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
1101		   struct trace_entry *ent, struct trace_iterator *iter,
1102		   u32 flags)
1103{
1104	unsigned long long duration = trace->rettime - trace->calltime;
1105	struct fgraph_data *data = iter->private;
1106	struct trace_array *tr = iter->tr;
1107	pid_t pid = ent->pid;
1108	int cpu = iter->cpu;
1109	int func_match = 1;
1110	int i;
1111
1112	if (check_irq_return(iter, flags, trace->depth))
1113		return TRACE_TYPE_HANDLED;
1114
1115	if (data) {
1116		struct fgraph_cpu_data *cpu_data;
1117		int cpu = iter->cpu;
1118
1119		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
1120
1121		/*
1122		 * Comments display at + 1 to depth. This is the
1123		 * return from a function, we now want the comments
1124		 * to display at the same level of the bracket.
1125		 */
1126		cpu_data->depth = trace->depth - 1;
1127
1128		if (trace->depth < FTRACE_RETFUNC_DEPTH &&
1129		    !WARN_ON_ONCE(trace->depth < 0)) {
1130			if (cpu_data->enter_funcs[trace->depth] != trace->func)
1131				func_match = 0;
1132			cpu_data->enter_funcs[trace->depth] = 0;
1133		}
1134	}
1135
1136	print_graph_prologue(iter, s, 0, 0, flags);
1137
1138	/* Overhead and duration */
1139	print_graph_duration(tr, duration, s, flags);
1140
1141	/* Closing brace */
1142	for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++)
1143		trace_seq_putc(s, ' ');
1144
1145	/*
1146	 * If the return function does not have a matching entry,
1147	 * then the entry was lost. Instead of just printing
1148	 * the '}' and letting the user guess what function this
1149	 * belongs to, write out the function name. Always do
1150	 * that if the funcgraph-tail option is enabled.
1151	 */
1152	if (func_match && !(flags & TRACE_GRAPH_PRINT_TAIL))
1153		trace_seq_puts(s, "}\n");
1154	else
1155		trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
1156
1157	/* Overrun */
1158	if (flags & TRACE_GRAPH_PRINT_OVERRUN)
1159		trace_seq_printf(s, " (Overruns: %lu)\n",
1160				 trace->overrun);
1161
1162	print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1163			cpu, pid, flags);
1164
1165	return trace_handle_return(s);
1166}
1167
1168static enum print_line_t
1169print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1170		    struct trace_iterator *iter, u32 flags)
1171{
1172	struct trace_array *tr = iter->tr;
1173	unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
1174	struct fgraph_data *data = iter->private;
1175	struct trace_event *event;
1176	int depth = 0;
1177	int ret;
1178	int i;
1179
1180	if (data)
1181		depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1182
1183	print_graph_prologue(iter, s, 0, 0, flags);
1184
1185	/* No time */
1186	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
1187
1188	/* Indentation */
1189	if (depth > 0)
1190		for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++)
1191			trace_seq_putc(s, ' ');
1192
1193	/* The comment */
1194	trace_seq_puts(s, "/* ");
1195
1196	switch (iter->ent->type) {
1197	case TRACE_BPUTS:
1198		ret = trace_print_bputs_msg_only(iter);
1199		if (ret != TRACE_TYPE_HANDLED)
1200			return ret;
1201		break;
1202	case TRACE_BPRINT:
1203		ret = trace_print_bprintk_msg_only(iter);
1204		if (ret != TRACE_TYPE_HANDLED)
1205			return ret;
1206		break;
1207	case TRACE_PRINT:
1208		ret = trace_print_printk_msg_only(iter);
1209		if (ret != TRACE_TYPE_HANDLED)
1210			return ret;
1211		break;
1212	default:
1213		event = ftrace_find_event(ent->type);
1214		if (!event)
1215			return TRACE_TYPE_UNHANDLED;
1216
1217		ret = event->funcs->trace(iter, sym_flags, event);
1218		if (ret != TRACE_TYPE_HANDLED)
1219			return ret;
1220	}
1221
1222	if (trace_seq_has_overflowed(s))
1223		goto out;
1224
1225	/* Strip ending newline */
1226	if (s->buffer[s->seq.len - 1] == '\n') {
1227		s->buffer[s->seq.len - 1] = '\0';
1228		s->seq.len--;
1229	}
1230
1231	trace_seq_puts(s, " */\n");
1232 out:
1233	return trace_handle_return(s);
1234}
1235
1236
1237enum print_line_t
1238print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1239{
1240	struct ftrace_graph_ent_entry *field;
1241	struct fgraph_data *data = iter->private;
1242	struct trace_entry *entry = iter->ent;
1243	struct trace_seq *s = &iter->seq;
1244	int cpu = iter->cpu;
1245	int ret;
1246
1247	if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1248		per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1249		return TRACE_TYPE_HANDLED;
1250	}
1251
1252	/*
1253	 * If the last output failed, there's a possibility we need
1254	 * to print out the missing entry which would never go out.
1255	 */
1256	if (data && data->failed) {
1257		field = &data->ent;
1258		iter->cpu = data->cpu;
1259		ret = print_graph_entry(field, s, iter, flags);
1260		if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1261			per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1262			ret = TRACE_TYPE_NO_CONSUME;
1263		}
1264		iter->cpu = cpu;
1265		return ret;
1266	}
1267
1268	switch (entry->type) {
1269	case TRACE_GRAPH_ENT: {
1270		/*
1271		 * print_graph_entry() may consume the current event,
1272		 * thus @field may become invalid, so we need to save it.
1273		 * sizeof(struct ftrace_graph_ent_entry) is very small,
1274		 * it can be safely saved at the stack.
1275		 */
1276		struct ftrace_graph_ent_entry saved;
1277		trace_assign_type(field, entry);
1278		saved = *field;
1279		return print_graph_entry(&saved, s, iter, flags);
1280	}
1281	case TRACE_GRAPH_RET: {
1282		struct ftrace_graph_ret_entry *field;
1283		trace_assign_type(field, entry);
1284		return print_graph_return(&field->ret, s, entry, iter, flags);
1285	}
1286	case TRACE_STACK:
1287	case TRACE_FN:
1288		/* dont trace stack and functions as comments */
1289		return TRACE_TYPE_UNHANDLED;
1290
1291	default:
1292		return print_graph_comment(s, entry, iter, flags);
1293	}
1294
1295	return TRACE_TYPE_HANDLED;
1296}
1297
1298static enum print_line_t
1299print_graph_function(struct trace_iterator *iter)
1300{
1301	return print_graph_function_flags(iter, tracer_flags.val);
1302}
1303
1304static enum print_line_t
1305print_graph_function_event(struct trace_iterator *iter, int flags,
1306			   struct trace_event *event)
1307{
1308	return print_graph_function(iter);
1309}
1310
1311static void print_lat_header(struct seq_file *s, u32 flags)
1312{
1313	static const char spaces[] = "                "	/* 16 spaces */
1314		"    "					/* 4 spaces */
1315		"                 ";			/* 17 spaces */
1316	int size = 0;
1317
1318	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1319		size += 16;
1320	if (flags & TRACE_GRAPH_PRINT_CPU)
1321		size += 4;
1322	if (flags & TRACE_GRAPH_PRINT_PROC)
1323		size += 17;
1324
1325	seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1326	seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1327	seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1328	seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1329	seq_printf(s, "#%.*s||| /                      \n", size, spaces);
1330}
1331
1332static void __print_graph_headers_flags(struct trace_array *tr,
1333					struct seq_file *s, u32 flags)
1334{
1335	int lat = tr->trace_flags & TRACE_ITER_LATENCY_FMT;
1336
1337	if (lat)
1338		print_lat_header(s, flags);
1339
1340	/* 1st line */
1341	seq_putc(s, '#');
1342	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1343		seq_puts(s, "     TIME       ");
1344	if (flags & TRACE_GRAPH_PRINT_CPU)
1345		seq_puts(s, " CPU");
1346	if (flags & TRACE_GRAPH_PRINT_PROC)
1347		seq_puts(s, "  TASK/PID       ");
1348	if (lat)
1349		seq_puts(s, "||||");
1350	if (flags & TRACE_GRAPH_PRINT_DURATION)
1351		seq_puts(s, "  DURATION   ");
1352	seq_puts(s, "               FUNCTION CALLS\n");
1353
1354	/* 2nd line */
1355	seq_putc(s, '#');
1356	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1357		seq_puts(s, "      |         ");
1358	if (flags & TRACE_GRAPH_PRINT_CPU)
1359		seq_puts(s, " |  ");
1360	if (flags & TRACE_GRAPH_PRINT_PROC)
1361		seq_puts(s, "   |    |        ");
1362	if (lat)
1363		seq_puts(s, "||||");
1364	if (flags & TRACE_GRAPH_PRINT_DURATION)
1365		seq_puts(s, "   |   |      ");
1366	seq_puts(s, "               |   |   |   |\n");
1367}
1368
1369static void print_graph_headers(struct seq_file *s)
1370{
1371	print_graph_headers_flags(s, tracer_flags.val);
1372}
1373
1374void print_graph_headers_flags(struct seq_file *s, u32 flags)
1375{
1376	struct trace_iterator *iter = s->private;
1377	struct trace_array *tr = iter->tr;
1378
1379	if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
1380		return;
1381
1382	if (tr->trace_flags & TRACE_ITER_LATENCY_FMT) {
1383		/* print nothing if the buffers are empty */
1384		if (trace_empty(iter))
1385			return;
1386
1387		print_trace_header(s, iter);
1388	}
1389
1390	__print_graph_headers_flags(tr, s, flags);
1391}
1392
1393void graph_trace_open(struct trace_iterator *iter)
1394{
1395	/* pid and depth on the last trace processed */
1396	struct fgraph_data *data;
1397	gfp_t gfpflags;
1398	int cpu;
1399
1400	iter->private = NULL;
1401
1402	/* We can be called in atomic context via ftrace_dump() */
1403	gfpflags = (in_atomic() || irqs_disabled()) ? GFP_ATOMIC : GFP_KERNEL;
1404
1405	data = kzalloc(sizeof(*data), gfpflags);
1406	if (!data)
1407		goto out_err;
1408
1409	data->cpu_data = alloc_percpu_gfp(struct fgraph_cpu_data, gfpflags);
1410	if (!data->cpu_data)
1411		goto out_err_free;
1412
1413	for_each_possible_cpu(cpu) {
1414		pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1415		int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1416		int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1417		int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1418
1419		*pid = -1;
1420		*depth = 0;
1421		*ignore = 0;
1422		*depth_irq = -1;
1423	}
1424
1425	iter->private = data;
1426
1427	return;
1428
1429 out_err_free:
1430	kfree(data);
1431 out_err:
1432	pr_warn("function graph tracer: not enough memory\n");
1433}
1434
1435void graph_trace_close(struct trace_iterator *iter)
1436{
1437	struct fgraph_data *data = iter->private;
1438
1439	if (data) {
1440		free_percpu(data->cpu_data);
1441		kfree(data);
1442	}
1443}
1444
1445static int
1446func_graph_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1447{
1448	if (bit == TRACE_GRAPH_PRINT_IRQS)
1449		ftrace_graph_skip_irqs = !set;
1450
1451	if (bit == TRACE_GRAPH_SLEEP_TIME)
1452		ftrace_graph_sleep_time_control(set);
1453
1454	if (bit == TRACE_GRAPH_GRAPH_TIME)
1455		ftrace_graph_graph_time_control(set);
1456
1457	return 0;
1458}
1459
1460static struct trace_event_functions graph_functions = {
1461	.trace		= print_graph_function_event,
1462};
1463
1464static struct trace_event graph_trace_entry_event = {
1465	.type		= TRACE_GRAPH_ENT,
1466	.funcs		= &graph_functions,
1467};
1468
1469static struct trace_event graph_trace_ret_event = {
1470	.type		= TRACE_GRAPH_RET,
1471	.funcs		= &graph_functions
1472};
1473
1474static struct tracer graph_trace __tracer_data = {
1475	.name		= "function_graph",
1476	.update_thresh	= graph_trace_update_thresh,
1477	.open		= graph_trace_open,
1478	.pipe_open	= graph_trace_open,
1479	.close		= graph_trace_close,
1480	.pipe_close	= graph_trace_close,
1481	.init		= graph_trace_init,
1482	.reset		= graph_trace_reset,
1483	.print_line	= print_graph_function,
1484	.print_header	= print_graph_headers,
1485	.flags		= &tracer_flags,
1486	.set_flag	= func_graph_set_flag,
1487#ifdef CONFIG_FTRACE_SELFTEST
1488	.selftest	= trace_selftest_startup_function_graph,
1489#endif
1490};
1491
1492
1493static ssize_t
1494graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1495		  loff_t *ppos)
1496{
1497	unsigned long val;
1498	int ret;
1499
1500	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1501	if (ret)
1502		return ret;
1503
1504	fgraph_max_depth = val;
1505
1506	*ppos += cnt;
1507
1508	return cnt;
1509}
1510
1511static ssize_t
1512graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1513		 loff_t *ppos)
1514{
1515	char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1516	int n;
1517
1518	n = sprintf(buf, "%d\n", fgraph_max_depth);
1519
1520	return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1521}
1522
1523static const struct file_operations graph_depth_fops = {
1524	.open		= tracing_open_generic,
1525	.write		= graph_depth_write,
1526	.read		= graph_depth_read,
1527	.llseek		= generic_file_llseek,
1528};
1529
1530static __init int init_graph_tracefs(void)
1531{
1532	struct dentry *d_tracer;
1533
1534	d_tracer = tracing_init_dentry();
1535	if (IS_ERR(d_tracer))
1536		return 0;
1537
1538	trace_create_file("max_graph_depth", 0644, d_tracer,
1539			  NULL, &graph_depth_fops);
1540
1541	return 0;
1542}
1543fs_initcall(init_graph_tracefs);
1544
1545static __init int init_graph_trace(void)
1546{
1547	max_bytes_for_cpu = snprintf(NULL, 0, "%u", nr_cpu_ids - 1);
1548
1549	if (!register_trace_event(&graph_trace_entry_event)) {
1550		pr_warn("Warning: could not register graph trace events\n");
1551		return 1;
1552	}
1553
1554	if (!register_trace_event(&graph_trace_ret_event)) {
1555		pr_warn("Warning: could not register graph trace events\n");
1556		return 1;
1557	}
1558
1559	return register_tracer(&graph_trace);
1560}
1561
1562core_initcall(init_graph_trace);
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);