Linux Audio

Check our new training course

Loading...
v3.15
   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 unsigned int max_depth;
  51
  52static struct tracer_opt trace_opts[] = {
  53	/* Display overruns? (for self-debug purpose) */
  54	{ TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
  55	/* Display CPU ? */
  56	{ TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
  57	/* Display Overhead ? */
  58	{ TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
  59	/* Display proc name/pid */
  60	{ TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
  61	/* Display duration of execution */
  62	{ TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
  63	/* Display absolute time of an entry */
  64	{ TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
  65	/* Display interrupts */
  66	{ TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
 
 
 
 
 
 
  67	{ } /* Empty entry */
  68};
  69
  70static struct tracer_flags tracer_flags = {
  71	/* Don't display overruns and proc by default */
  72	.val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
  73	       TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS,
 
  74	.opts = trace_opts
  75};
  76
  77static struct trace_array *graph_array;
  78
  79/*
  80 * DURATION column is being also used to display IRQ signs,
  81 * following values are used by print_graph_irq and others
  82 * to fill in space into DURATION column.
  83 */
  84enum {
  85	FLAGS_FILL_FULL  = 1 << TRACE_GRAPH_PRINT_FILL_SHIFT,
  86	FLAGS_FILL_START = 2 << TRACE_GRAPH_PRINT_FILL_SHIFT,
  87	FLAGS_FILL_END   = 3 << TRACE_GRAPH_PRINT_FILL_SHIFT,
  88};
  89
  90static enum print_line_t
  91print_graph_duration(unsigned long long duration, struct trace_seq *s,
  92		     u32 flags);
  93
  94/* Add a function return address to the trace stack on thread info.*/
  95int
  96ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
  97			 unsigned long frame_pointer)
  98{
  99	unsigned long long calltime;
 100	int index;
 101
 
 
 
 102	if (!current->ret_stack)
 103		return -EBUSY;
 104
 105	/*
 106	 * We must make sure the ret_stack is tested before we read
 107	 * anything else.
 108	 */
 109	smp_rmb();
 110
 111	/* The return trace stack is full */
 112	if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
 113		atomic_inc(&current->trace_overrun);
 114		return -EBUSY;
 115	}
 116
 117	/*
 118	 * The curr_ret_stack is an index to ftrace return stack of
 119	 * current task.  Its value should be in [0, FTRACE_RETFUNC_
 120	 * DEPTH) when the function graph tracer is used.  To support
 121	 * filtering out specific functions, it makes the index
 122	 * negative by subtracting huge value (FTRACE_NOTRACE_DEPTH)
 123	 * so when it sees a negative index the ftrace will ignore
 124	 * the record.  And the index gets recovered when returning
 125	 * from the filtered function by adding the FTRACE_NOTRACE_
 126	 * DEPTH and then it'll continue to record functions normally.
 127	 *
 128	 * The curr_ret_stack is initialized to -1 and get increased
 129	 * in this function.  So it can be less than -1 only if it was
 130	 * filtered out via ftrace_graph_notrace_addr() which can be
 131	 * set from set_graph_notrace file in debugfs by user.
 132	 */
 133	if (current->curr_ret_stack < -1)
 134		return -EBUSY;
 135
 136	calltime = trace_clock_local();
 137
 138	index = ++current->curr_ret_stack;
 139	if (ftrace_graph_notrace_addr(func))
 140		current->curr_ret_stack -= FTRACE_NOTRACE_DEPTH;
 141	barrier();
 142	current->ret_stack[index].ret = ret;
 143	current->ret_stack[index].func = func;
 144	current->ret_stack[index].calltime = calltime;
 145	current->ret_stack[index].subtime = 0;
 146	current->ret_stack[index].fp = frame_pointer;
 147	*depth = current->curr_ret_stack;
 148
 149	return 0;
 150}
 151
 152/* Retrieve a function return address to the trace stack on thread info.*/
 153static void
 154ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
 155			unsigned long frame_pointer)
 156{
 157	int index;
 158
 159	index = current->curr_ret_stack;
 160
 161	/*
 162	 * A negative index here means that it's just returned from a
 163	 * notrace'd function.  Recover index to get an original
 164	 * return address.  See ftrace_push_return_trace().
 165	 *
 166	 * TODO: Need to check whether the stack gets corrupted.
 167	 */
 168	if (index < 0)
 169		index += FTRACE_NOTRACE_DEPTH;
 170
 171	if (unlikely(index < 0 || index >= FTRACE_RETFUNC_DEPTH)) {
 172		ftrace_graph_stop();
 173		WARN_ON(1);
 174		/* Might as well panic, otherwise we have no where to go */
 175		*ret = (unsigned long)panic;
 176		return;
 177	}
 178
 179#if defined(CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST) && !defined(CC_USING_FENTRY)
 180	/*
 181	 * The arch may choose to record the frame pointer used
 182	 * and check it here to make sure that it is what we expect it
 183	 * to be. If gcc does not set the place holder of the return
 184	 * address in the frame pointer, and does a copy instead, then
 185	 * the function graph trace will fail. This test detects this
 186	 * case.
 187	 *
 188	 * Currently, x86_32 with optimize for size (-Os) makes the latest
 189	 * gcc do the above.
 190	 *
 191	 * Note, -mfentry does not use frame pointers, and this test
 192	 *  is not needed if CC_USING_FENTRY is set.
 193	 */
 194	if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
 195		ftrace_graph_stop();
 196		WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
 197		     "  from func %ps return to %lx\n",
 198		     current->ret_stack[index].fp,
 199		     frame_pointer,
 200		     (void *)current->ret_stack[index].func,
 201		     current->ret_stack[index].ret);
 202		*ret = (unsigned long)panic;
 203		return;
 204	}
 205#endif
 206
 207	*ret = current->ret_stack[index].ret;
 208	trace->func = current->ret_stack[index].func;
 209	trace->calltime = current->ret_stack[index].calltime;
 210	trace->overrun = atomic_read(&current->trace_overrun);
 211	trace->depth = index;
 212}
 213
 214/*
 215 * Send the trace to the ring-buffer.
 216 * @return the original return address.
 217 */
 218unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
 219{
 220	struct ftrace_graph_ret trace;
 221	unsigned long ret;
 222
 223	ftrace_pop_return_trace(&trace, &ret, frame_pointer);
 224	trace.rettime = trace_clock_local();
 225	barrier();
 226	current->curr_ret_stack--;
 227	/*
 228	 * The curr_ret_stack can be less than -1 only if it was
 229	 * filtered out and it's about to return from the function.
 230	 * Recover the index and continue to trace normal functions.
 231	 */
 232	if (current->curr_ret_stack < -1) {
 233		current->curr_ret_stack += FTRACE_NOTRACE_DEPTH;
 234		return ret;
 235	}
 236
 237	/*
 238	 * The trace should run after decrementing the ret counter
 239	 * in case an interrupt were to come in. We don't want to
 240	 * lose the interrupt if max_depth is set.
 241	 */
 242	ftrace_graph_return(&trace);
 243
 244	if (unlikely(!ret)) {
 245		ftrace_graph_stop();
 246		WARN_ON(1);
 247		/* Might as well panic. What else to do? */
 248		ret = (unsigned long)panic;
 249	}
 250
 251	return ret;
 252}
 253
 254int __trace_graph_entry(struct trace_array *tr,
 255				struct ftrace_graph_ent *trace,
 256				unsigned long flags,
 257				int pc)
 258{
 259	struct ftrace_event_call *call = &event_funcgraph_entry;
 260	struct ring_buffer_event *event;
 261	struct ring_buffer *buffer = tr->trace_buffer.buffer;
 262	struct ftrace_graph_ent_entry *entry;
 263
 264	if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
 265		return 0;
 266
 267	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
 268					  sizeof(*entry), flags, pc);
 269	if (!event)
 270		return 0;
 271	entry	= ring_buffer_event_data(event);
 272	entry->graph_ent			= *trace;
 273	if (!call_filter_check_discard(call, entry, buffer, event))
 274		__buffer_unlock_commit(buffer, event);
 275
 276	return 1;
 277}
 278
 279static inline int ftrace_graph_ignore_irqs(void)
 280{
 281	if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
 282		return 0;
 283
 284	return in_irq();
 285}
 286
 287int trace_graph_entry(struct ftrace_graph_ent *trace)
 288{
 289	struct trace_array *tr = graph_array;
 290	struct trace_array_cpu *data;
 291	unsigned long flags;
 292	long disabled;
 293	int ret;
 294	int cpu;
 295	int pc;
 296
 297	if (!ftrace_trace_task(current))
 298		return 0;
 299
 300	/* trace it when it is-nested-in or is a function enabled. */
 301	if ((!(trace->depth || ftrace_graph_addr(trace->func)) ||
 302	     ftrace_graph_ignore_irqs()) || (trace->depth < 0) ||
 303	    (max_depth && trace->depth >= max_depth))
 304		return 0;
 305
 306	/*
 307	 * Do not trace a function if it's filtered by set_graph_notrace.
 308	 * Make the index of ret stack negative to indicate that it should
 309	 * ignore further functions.  But it needs its own ret stack entry
 310	 * to recover the original index in order to continue tracing after
 311	 * returning from the function.
 312	 */
 313	if (ftrace_graph_notrace_addr(trace->func))
 314		return 1;
 315
 316	local_irq_save(flags);
 317	cpu = raw_smp_processor_id();
 318	data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 319	disabled = atomic_inc_return(&data->disabled);
 320	if (likely(disabled == 1)) {
 321		pc = preempt_count();
 322		ret = __trace_graph_entry(tr, trace, flags, pc);
 323	} else {
 324		ret = 0;
 325	}
 326
 327	atomic_dec(&data->disabled);
 328	local_irq_restore(flags);
 329
 330	return ret;
 331}
 332
 333int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
 334{
 335	if (tracing_thresh)
 336		return 1;
 337	else
 338		return trace_graph_entry(trace);
 339}
 340
 341static void
 342__trace_graph_function(struct trace_array *tr,
 343		unsigned long ip, unsigned long flags, int pc)
 344{
 345	u64 time = trace_clock_local();
 346	struct ftrace_graph_ent ent = {
 347		.func  = ip,
 348		.depth = 0,
 349	};
 350	struct ftrace_graph_ret ret = {
 351		.func     = ip,
 352		.depth    = 0,
 353		.calltime = time,
 354		.rettime  = time,
 355	};
 356
 357	__trace_graph_entry(tr, &ent, flags, pc);
 358	__trace_graph_return(tr, &ret, flags, pc);
 359}
 360
 361void
 362trace_graph_function(struct trace_array *tr,
 363		unsigned long ip, unsigned long parent_ip,
 364		unsigned long flags, int pc)
 365{
 366	__trace_graph_function(tr, ip, flags, pc);
 367}
 368
 369void __trace_graph_return(struct trace_array *tr,
 370				struct ftrace_graph_ret *trace,
 371				unsigned long flags,
 372				int pc)
 373{
 374	struct ftrace_event_call *call = &event_funcgraph_exit;
 375	struct ring_buffer_event *event;
 376	struct ring_buffer *buffer = tr->trace_buffer.buffer;
 377	struct ftrace_graph_ret_entry *entry;
 378
 379	if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
 380		return;
 381
 382	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
 383					  sizeof(*entry), flags, pc);
 384	if (!event)
 385		return;
 386	entry	= ring_buffer_event_data(event);
 387	entry->ret				= *trace;
 388	if (!call_filter_check_discard(call, entry, buffer, event))
 389		__buffer_unlock_commit(buffer, event);
 390}
 391
 392void trace_graph_return(struct ftrace_graph_ret *trace)
 393{
 394	struct trace_array *tr = graph_array;
 395	struct trace_array_cpu *data;
 396	unsigned long flags;
 397	long disabled;
 398	int cpu;
 399	int pc;
 400
 401	local_irq_save(flags);
 402	cpu = raw_smp_processor_id();
 403	data = per_cpu_ptr(tr->trace_buffer.data, cpu);
 404	disabled = atomic_inc_return(&data->disabled);
 405	if (likely(disabled == 1)) {
 406		pc = preempt_count();
 407		__trace_graph_return(tr, trace, flags, pc);
 408	}
 409	atomic_dec(&data->disabled);
 410	local_irq_restore(flags);
 411}
 412
 413void set_graph_array(struct trace_array *tr)
 414{
 415	graph_array = tr;
 416
 417	/* Make graph_array visible before we start tracing */
 418
 419	smp_mb();
 420}
 421
 422void trace_graph_thresh_return(struct ftrace_graph_ret *trace)
 423{
 424	if (tracing_thresh &&
 425	    (trace->rettime - trace->calltime < tracing_thresh))
 426		return;
 427	else
 428		trace_graph_return(trace);
 429}
 430
 431static int graph_trace_init(struct trace_array *tr)
 432{
 433	int ret;
 434
 435	set_graph_array(tr);
 436	if (tracing_thresh)
 437		ret = register_ftrace_graph(&trace_graph_thresh_return,
 438					    &trace_graph_thresh_entry);
 439	else
 440		ret = register_ftrace_graph(&trace_graph_return,
 441					    &trace_graph_entry);
 442	if (ret)
 443		return ret;
 444	tracing_start_cmdline_record();
 445
 446	return 0;
 447}
 448
 449static void graph_trace_reset(struct trace_array *tr)
 450{
 451	tracing_stop_cmdline_record();
 452	unregister_ftrace_graph();
 453}
 454
 
 
 
 
 
 
 455static int max_bytes_for_cpu;
 456
 457static enum print_line_t
 458print_graph_cpu(struct trace_seq *s, int cpu)
 459{
 460	int ret;
 461
 462	/*
 463	 * Start with a space character - to make it stand out
 464	 * to the right a bit when trace output is pasted into
 465	 * email:
 466	 */
 467	ret = trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
 468	if (!ret)
 469		return TRACE_TYPE_PARTIAL_LINE;
 470
 471	return TRACE_TYPE_HANDLED;
 472}
 473
 474#define TRACE_GRAPH_PROCINFO_LENGTH	14
 475
 476static enum print_line_t
 477print_graph_proc(struct trace_seq *s, pid_t pid)
 478{
 479	char comm[TASK_COMM_LEN];
 480	/* sign + log10(MAX_INT) + '\0' */
 481	char pid_str[11];
 482	int spaces = 0;
 483	int ret;
 484	int len;
 485	int i;
 486
 487	trace_find_cmdline(pid, comm);
 488	comm[7] = '\0';
 489	sprintf(pid_str, "%d", pid);
 490
 491	/* 1 stands for the "-" character */
 492	len = strlen(comm) + strlen(pid_str) + 1;
 493
 494	if (len < TRACE_GRAPH_PROCINFO_LENGTH)
 495		spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
 496
 497	/* First spaces to align center */
 498	for (i = 0; i < spaces / 2; i++) {
 499		ret = trace_seq_putc(s, ' ');
 500		if (!ret)
 501			return TRACE_TYPE_PARTIAL_LINE;
 502	}
 503
 504	ret = trace_seq_printf(s, "%s-%s", comm, pid_str);
 505	if (!ret)
 506		return TRACE_TYPE_PARTIAL_LINE;
 507
 508	/* Last spaces to align center */
 509	for (i = 0; i < spaces - (spaces / 2); i++) {
 510		ret = trace_seq_putc(s, ' ');
 511		if (!ret)
 512			return TRACE_TYPE_PARTIAL_LINE;
 513	}
 514	return TRACE_TYPE_HANDLED;
 515}
 516
 517
 518static enum print_line_t
 519print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
 520{
 521	if (!trace_seq_putc(s, ' '))
 522		return 0;
 523
 524	return trace_print_lat_fmt(s, entry);
 525}
 526
 527/* If the pid changed since the last trace, output this event */
 528static enum print_line_t
 529verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
 530{
 531	pid_t prev_pid;
 532	pid_t *last_pid;
 533	int ret;
 534
 535	if (!data)
 536		return TRACE_TYPE_HANDLED;
 537
 538	last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
 539
 540	if (*last_pid == pid)
 541		return TRACE_TYPE_HANDLED;
 542
 543	prev_pid = *last_pid;
 544	*last_pid = pid;
 545
 546	if (prev_pid == -1)
 547		return TRACE_TYPE_HANDLED;
 548/*
 549 * Context-switch trace line:
 550
 551 ------------------------------------------
 552 | 1)  migration/0--1  =>  sshd-1755
 553 ------------------------------------------
 554
 555 */
 556	ret = trace_seq_puts(s,
 557		" ------------------------------------------\n");
 558	if (!ret)
 559		return TRACE_TYPE_PARTIAL_LINE;
 560
 561	ret = print_graph_cpu(s, cpu);
 562	if (ret == TRACE_TYPE_PARTIAL_LINE)
 563		return TRACE_TYPE_PARTIAL_LINE;
 564
 565	ret = print_graph_proc(s, prev_pid);
 566	if (ret == TRACE_TYPE_PARTIAL_LINE)
 567		return TRACE_TYPE_PARTIAL_LINE;
 568
 569	ret = trace_seq_puts(s, " => ");
 570	if (!ret)
 571		return TRACE_TYPE_PARTIAL_LINE;
 572
 573	ret = print_graph_proc(s, pid);
 574	if (ret == TRACE_TYPE_PARTIAL_LINE)
 575		return TRACE_TYPE_PARTIAL_LINE;
 576
 577	ret = trace_seq_puts(s,
 578		"\n ------------------------------------------\n\n");
 579	if (!ret)
 580		return TRACE_TYPE_PARTIAL_LINE;
 581
 582	return TRACE_TYPE_HANDLED;
 583}
 584
 585static struct ftrace_graph_ret_entry *
 586get_return_for_leaf(struct trace_iterator *iter,
 587		struct ftrace_graph_ent_entry *curr)
 588{
 589	struct fgraph_data *data = iter->private;
 590	struct ring_buffer_iter *ring_iter = NULL;
 591	struct ring_buffer_event *event;
 592	struct ftrace_graph_ret_entry *next;
 593
 594	/*
 595	 * If the previous output failed to write to the seq buffer,
 596	 * then we just reuse the data from before.
 597	 */
 598	if (data && data->failed) {
 599		curr = &data->ent;
 600		next = &data->ret;
 601	} else {
 602
 603		ring_iter = trace_buffer_iter(iter, iter->cpu);
 604
 605		/* First peek to compare current entry and the next one */
 606		if (ring_iter)
 607			event = ring_buffer_iter_peek(ring_iter, NULL);
 608		else {
 609			/*
 610			 * We need to consume the current entry to see
 611			 * the next one.
 612			 */
 613			ring_buffer_consume(iter->trace_buffer->buffer, iter->cpu,
 614					    NULL, NULL);
 615			event = ring_buffer_peek(iter->trace_buffer->buffer, iter->cpu,
 616						 NULL, NULL);
 617		}
 618
 619		if (!event)
 620			return NULL;
 621
 622		next = ring_buffer_event_data(event);
 623
 624		if (data) {
 625			/*
 626			 * Save current and next entries for later reference
 627			 * if the output fails.
 628			 */
 629			data->ent = *curr;
 630			/*
 631			 * If the next event is not a return type, then
 632			 * we only care about what type it is. Otherwise we can
 633			 * safely copy the entire event.
 634			 */
 635			if (next->ent.type == TRACE_GRAPH_RET)
 636				data->ret = *next;
 637			else
 638				data->ret.ent.type = next->ent.type;
 639		}
 640	}
 641
 642	if (next->ent.type != TRACE_GRAPH_RET)
 643		return NULL;
 644
 645	if (curr->ent.pid != next->ent.pid ||
 646			curr->graph_ent.func != next->ret.func)
 647		return NULL;
 648
 649	/* this is a leaf, now advance the iterator */
 650	if (ring_iter)
 651		ring_buffer_read(ring_iter, NULL);
 652
 653	return next;
 654}
 655
 656static int print_graph_abs_time(u64 t, struct trace_seq *s)
 657{
 658	unsigned long usecs_rem;
 659
 660	usecs_rem = do_div(t, NSEC_PER_SEC);
 661	usecs_rem /= 1000;
 662
 663	return trace_seq_printf(s, "%5lu.%06lu |  ",
 664			(unsigned long)t, usecs_rem);
 665}
 666
 667static enum print_line_t
 668print_graph_irq(struct trace_iterator *iter, unsigned long addr,
 669		enum trace_type type, int cpu, pid_t pid, u32 flags)
 670{
 671	int ret;
 672	struct trace_seq *s = &iter->seq;
 
 673
 674	if (addr < (unsigned long)__irqentry_text_start ||
 675		addr >= (unsigned long)__irqentry_text_end)
 676		return TRACE_TYPE_UNHANDLED;
 677
 678	if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
 679		/* Absolute time */
 680		if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
 681			ret = print_graph_abs_time(iter->ts, s);
 682			if (!ret)
 683				return TRACE_TYPE_PARTIAL_LINE;
 684		}
 685
 686		/* Cpu */
 687		if (flags & TRACE_GRAPH_PRINT_CPU) {
 688			ret = print_graph_cpu(s, cpu);
 689			if (ret == TRACE_TYPE_PARTIAL_LINE)
 690				return TRACE_TYPE_PARTIAL_LINE;
 691		}
 692
 693		/* Proc */
 694		if (flags & TRACE_GRAPH_PRINT_PROC) {
 695			ret = print_graph_proc(s, pid);
 696			if (ret == TRACE_TYPE_PARTIAL_LINE)
 697				return TRACE_TYPE_PARTIAL_LINE;
 698			ret = trace_seq_puts(s, " | ");
 699			if (!ret)
 700				return TRACE_TYPE_PARTIAL_LINE;
 701		}
 
 
 
 
 702	}
 703
 704	/* No overhead */
 705	ret = print_graph_duration(0, s, flags | FLAGS_FILL_START);
 706	if (ret != TRACE_TYPE_HANDLED)
 707		return ret;
 708
 709	if (type == TRACE_GRAPH_ENT)
 710		ret = trace_seq_puts(s, "==========>");
 711	else
 712		ret = trace_seq_puts(s, "<==========");
 713
 714	if (!ret)
 715		return TRACE_TYPE_PARTIAL_LINE;
 716
 717	ret = print_graph_duration(0, s, flags | FLAGS_FILL_END);
 718	if (ret != TRACE_TYPE_HANDLED)
 719		return ret;
 720
 721	ret = trace_seq_putc(s, '\n');
 722
 723	if (!ret)
 724		return TRACE_TYPE_PARTIAL_LINE;
 725	return TRACE_TYPE_HANDLED;
 726}
 727
 728enum print_line_t
 729trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
 730{
 731	unsigned long nsecs_rem = do_div(duration, 1000);
 732	/* log10(ULONG_MAX) + '\0' */
 733	char msecs_str[21];
 734	char nsecs_str[5];
 735	int ret, len;
 736	int i;
 737
 738	sprintf(msecs_str, "%lu", (unsigned long) duration);
 739
 740	/* Print msecs */
 741	ret = trace_seq_printf(s, "%s", msecs_str);
 742	if (!ret)
 743		return TRACE_TYPE_PARTIAL_LINE;
 744
 745	len = strlen(msecs_str);
 746
 747	/* Print nsecs (we don't want to exceed 7 numbers) */
 748	if (len < 7) {
 749		size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
 750
 751		snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
 752		ret = trace_seq_printf(s, ".%s", nsecs_str);
 753		if (!ret)
 754			return TRACE_TYPE_PARTIAL_LINE;
 755		len += strlen(nsecs_str);
 756	}
 757
 758	ret = trace_seq_puts(s, " us ");
 759	if (!ret)
 760		return TRACE_TYPE_PARTIAL_LINE;
 761
 762	/* Print remaining spaces to fit the row's width */
 763	for (i = len; i < 7; i++) {
 764		ret = trace_seq_putc(s, ' ');
 765		if (!ret)
 766			return TRACE_TYPE_PARTIAL_LINE;
 767	}
 768	return TRACE_TYPE_HANDLED;
 769}
 770
 771static enum print_line_t
 772print_graph_duration(unsigned long long duration, struct trace_seq *s,
 773		     u32 flags)
 774{
 775	int ret = -1;
 776
 777	if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
 778	    !(trace_flags & TRACE_ITER_CONTEXT_INFO))
 779			return TRACE_TYPE_HANDLED;
 780
 781	/* No real adata, just filling the column with spaces */
 782	switch (flags & TRACE_GRAPH_PRINT_FILL_MASK) {
 783	case FLAGS_FILL_FULL:
 784		ret = trace_seq_puts(s, "              |  ");
 785		return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 786	case FLAGS_FILL_START:
 787		ret = trace_seq_puts(s, "  ");
 788		return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 789	case FLAGS_FILL_END:
 790		ret = trace_seq_puts(s, " |");
 791		return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
 792	}
 793
 794	/* Signal a overhead of time execution to the output */
 795	if (flags & TRACE_GRAPH_PRINT_OVERHEAD) {
 796		/* Duration exceeded 100 msecs */
 797		if (duration > 100000ULL)
 798			ret = trace_seq_puts(s, "! ");
 799		/* Duration exceeded 10 msecs */
 800		else if (duration > 10000ULL)
 801			ret = trace_seq_puts(s, "+ ");
 802	}
 803
 804	/*
 805	 * The -1 means we either did not exceed the duration tresholds
 806	 * or we dont want to print out the overhead. Either way we need
 807	 * to fill out the space.
 808	 */
 809	if (ret == -1)
 810		ret = trace_seq_puts(s, "  ");
 811
 812	/* Catching here any failure happenned above */
 813	if (!ret)
 814		return TRACE_TYPE_PARTIAL_LINE;
 815
 816	ret = trace_print_graph_duration(duration, s);
 817	if (ret != TRACE_TYPE_HANDLED)
 818		return ret;
 819
 820	ret = trace_seq_puts(s, "|  ");
 821	if (!ret)
 822		return TRACE_TYPE_PARTIAL_LINE;
 823
 824	return TRACE_TYPE_HANDLED;
 
 825}
 826
 827/* Case of a leaf function on its call entry */
 828static enum print_line_t
 829print_graph_entry_leaf(struct trace_iterator *iter,
 830		struct ftrace_graph_ent_entry *entry,
 831		struct ftrace_graph_ret_entry *ret_entry,
 832		struct trace_seq *s, u32 flags)
 833{
 834	struct fgraph_data *data = iter->private;
 
 835	struct ftrace_graph_ret *graph_ret;
 836	struct ftrace_graph_ent *call;
 837	unsigned long long duration;
 838	int ret;
 839	int i;
 840
 841	graph_ret = &ret_entry->ret;
 842	call = &entry->graph_ent;
 843	duration = graph_ret->rettime - graph_ret->calltime;
 844
 845	if (data) {
 846		struct fgraph_cpu_data *cpu_data;
 847		int cpu = iter->cpu;
 848
 849		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 850
 851		/*
 852		 * Comments display at + 1 to depth. Since
 853		 * this is a leaf function, keep the comments
 854		 * equal to this depth.
 855		 */
 856		cpu_data->depth = call->depth - 1;
 857
 858		/* No need to keep this function around for this depth */
 859		if (call->depth < FTRACE_RETFUNC_DEPTH)
 860			cpu_data->enter_funcs[call->depth] = 0;
 861	}
 862
 863	/* Overhead and duration */
 864	ret = print_graph_duration(duration, s, flags);
 865	if (ret == TRACE_TYPE_PARTIAL_LINE)
 866		return TRACE_TYPE_PARTIAL_LINE;
 867
 868	/* Function */
 869	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 870		ret = trace_seq_putc(s, ' ');
 871		if (!ret)
 872			return TRACE_TYPE_PARTIAL_LINE;
 873	}
 874
 875	ret = trace_seq_printf(s, "%ps();\n", (void *)call->func);
 876	if (!ret)
 877		return TRACE_TYPE_PARTIAL_LINE;
 878
 879	return TRACE_TYPE_HANDLED;
 880}
 881
 882static enum print_line_t
 883print_graph_entry_nested(struct trace_iterator *iter,
 884			 struct ftrace_graph_ent_entry *entry,
 885			 struct trace_seq *s, int cpu, u32 flags)
 886{
 887	struct ftrace_graph_ent *call = &entry->graph_ent;
 888	struct fgraph_data *data = iter->private;
 889	int ret;
 890	int i;
 891
 892	if (data) {
 893		struct fgraph_cpu_data *cpu_data;
 894		int cpu = iter->cpu;
 895
 896		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
 897		cpu_data->depth = call->depth;
 898
 899		/* Save this function pointer to see if the exit matches */
 900		if (call->depth < FTRACE_RETFUNC_DEPTH)
 901			cpu_data->enter_funcs[call->depth] = call->func;
 902	}
 903
 904	/* No time */
 905	ret = print_graph_duration(0, s, flags | FLAGS_FILL_FULL);
 906	if (ret != TRACE_TYPE_HANDLED)
 907		return ret;
 908
 909	/* Function */
 910	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
 911		ret = trace_seq_putc(s, ' ');
 912		if (!ret)
 913			return TRACE_TYPE_PARTIAL_LINE;
 914	}
 915
 916	ret = trace_seq_printf(s, "%ps() {\n", (void *)call->func);
 917	if (!ret)
 
 918		return TRACE_TYPE_PARTIAL_LINE;
 919
 920	/*
 921	 * we already consumed the current entry to check the next one
 922	 * and see if this is a leaf.
 923	 */
 924	return TRACE_TYPE_NO_CONSUME;
 925}
 926
 927static enum print_line_t
 928print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
 929		     int type, unsigned long addr, u32 flags)
 930{
 931	struct fgraph_data *data = iter->private;
 932	struct trace_entry *ent = iter->ent;
 
 933	int cpu = iter->cpu;
 934	int ret;
 935
 936	/* Pid */
 937	if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
 938		return TRACE_TYPE_PARTIAL_LINE;
 939
 940	if (type) {
 941		/* Interrupt */
 942		ret = print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
 943		if (ret == TRACE_TYPE_PARTIAL_LINE)
 944			return TRACE_TYPE_PARTIAL_LINE;
 945	}
 946
 947	if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
 948		return 0;
 949
 950	/* Absolute time */
 951	if (flags & TRACE_GRAPH_PRINT_ABS_TIME) {
 952		ret = print_graph_abs_time(iter->ts, s);
 953		if (!ret)
 954			return TRACE_TYPE_PARTIAL_LINE;
 955	}
 956
 957	/* Cpu */
 958	if (flags & TRACE_GRAPH_PRINT_CPU) {
 959		ret = print_graph_cpu(s, cpu);
 960		if (ret == TRACE_TYPE_PARTIAL_LINE)
 961			return TRACE_TYPE_PARTIAL_LINE;
 962	}
 963
 964	/* Proc */
 965	if (flags & TRACE_GRAPH_PRINT_PROC) {
 966		ret = print_graph_proc(s, ent->pid);
 967		if (ret == TRACE_TYPE_PARTIAL_LINE)
 968			return TRACE_TYPE_PARTIAL_LINE;
 969
 970		ret = trace_seq_puts(s, " | ");
 971		if (!ret)
 972			return TRACE_TYPE_PARTIAL_LINE;
 973	}
 974
 975	/* Latency format */
 976	if (trace_flags & TRACE_ITER_LATENCY_FMT) {
 977		ret = print_graph_lat_fmt(s, ent);
 978		if (ret == TRACE_TYPE_PARTIAL_LINE)
 979			return TRACE_TYPE_PARTIAL_LINE;
 980	}
 981
 982	return 0;
 983}
 984
 985/*
 986 * Entry check for irq code
 987 *
 988 * returns 1 if
 989 *  - we are inside irq code
 990 *  - we just entered irq code
 991 *
 992 * retunns 0 if
 993 *  - funcgraph-interrupts option is set
 994 *  - we are not inside irq code
 995 */
 996static int
 997check_irq_entry(struct trace_iterator *iter, u32 flags,
 998		unsigned long addr, int depth)
 999{
1000	int cpu = iter->cpu;
1001	int *depth_irq;
1002	struct fgraph_data *data = iter->private;
1003
1004	/*
1005	 * If we are either displaying irqs, or we got called as
1006	 * a graph event and private data does not exist,
1007	 * then we bypass the irq check.
1008	 */
1009	if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
1010	    (!data))
1011		return 0;
1012
1013	depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1014
1015	/*
1016	 * We are inside the irq code
1017	 */
1018	if (*depth_irq >= 0)
1019		return 1;
1020
1021	if ((addr < (unsigned long)__irqentry_text_start) ||
1022	    (addr >= (unsigned long)__irqentry_text_end))
1023		return 0;
1024
1025	/*
1026	 * We are entering irq code.
1027	 */
1028	*depth_irq = depth;
1029	return 1;
1030}
1031
1032/*
1033 * Return check for irq code
1034 *
1035 * returns 1 if
1036 *  - we are inside irq code
1037 *  - we just left irq code
1038 *
1039 * returns 0 if
1040 *  - funcgraph-interrupts option is set
1041 *  - we are not inside irq code
1042 */
1043static int
1044check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
1045{
1046	int cpu = iter->cpu;
1047	int *depth_irq;
1048	struct fgraph_data *data = iter->private;
1049
1050	/*
1051	 * If we are either displaying irqs, or we got called as
1052	 * a graph event and private data does not exist,
1053	 * then we bypass the irq check.
1054	 */
1055	if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
1056	    (!data))
1057		return 0;
1058
1059	depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1060
1061	/*
1062	 * We are not inside the irq code.
1063	 */
1064	if (*depth_irq == -1)
1065		return 0;
1066
1067	/*
1068	 * We are inside the irq code, and this is returning entry.
1069	 * Let's not trace it and clear the entry depth, since
1070	 * we are out of irq code.
1071	 *
1072	 * This condition ensures that we 'leave the irq code' once
1073	 * we are out of the entry depth. Thus protecting us from
1074	 * the RETURN entry loss.
1075	 */
1076	if (*depth_irq >= depth) {
1077		*depth_irq = -1;
1078		return 1;
1079	}
1080
1081	/*
1082	 * We are inside the irq code, and this is not the entry.
1083	 */
1084	return 1;
1085}
1086
1087static enum print_line_t
1088print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
1089			struct trace_iterator *iter, u32 flags)
1090{
1091	struct fgraph_data *data = iter->private;
1092	struct ftrace_graph_ent *call = &field->graph_ent;
1093	struct ftrace_graph_ret_entry *leaf_ret;
1094	static enum print_line_t ret;
1095	int cpu = iter->cpu;
1096
1097	if (check_irq_entry(iter, flags, call->func, call->depth))
1098		return TRACE_TYPE_HANDLED;
1099
1100	if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags))
1101		return TRACE_TYPE_PARTIAL_LINE;
1102
1103	leaf_ret = get_return_for_leaf(iter, field);
1104	if (leaf_ret)
1105		ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
1106	else
1107		ret = print_graph_entry_nested(iter, field, s, cpu, flags);
1108
1109	if (data) {
1110		/*
1111		 * If we failed to write our output, then we need to make
1112		 * note of it. Because we already consumed our entry.
1113		 */
1114		if (s->full) {
1115			data->failed = 1;
1116			data->cpu = cpu;
1117		} else
1118			data->failed = 0;
1119	}
1120
1121	return ret;
1122}
1123
1124static enum print_line_t
1125print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
1126		   struct trace_entry *ent, struct trace_iterator *iter,
1127		   u32 flags)
1128{
1129	unsigned long long duration = trace->rettime - trace->calltime;
1130	struct fgraph_data *data = iter->private;
 
1131	pid_t pid = ent->pid;
1132	int cpu = iter->cpu;
1133	int func_match = 1;
1134	int ret;
1135	int i;
1136
1137	if (check_irq_return(iter, flags, trace->depth))
1138		return TRACE_TYPE_HANDLED;
1139
1140	if (data) {
1141		struct fgraph_cpu_data *cpu_data;
1142		int cpu = iter->cpu;
1143
1144		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
1145
1146		/*
1147		 * Comments display at + 1 to depth. This is the
1148		 * return from a function, we now want the comments
1149		 * to display at the same level of the bracket.
1150		 */
1151		cpu_data->depth = trace->depth - 1;
1152
1153		if (trace->depth < FTRACE_RETFUNC_DEPTH) {
1154			if (cpu_data->enter_funcs[trace->depth] != trace->func)
1155				func_match = 0;
1156			cpu_data->enter_funcs[trace->depth] = 0;
1157		}
1158	}
1159
1160	if (print_graph_prologue(iter, s, 0, 0, flags))
1161		return TRACE_TYPE_PARTIAL_LINE;
1162
1163	/* Overhead and duration */
1164	ret = print_graph_duration(duration, s, flags);
1165	if (ret == TRACE_TYPE_PARTIAL_LINE)
1166		return TRACE_TYPE_PARTIAL_LINE;
1167
1168	/* Closing brace */
1169	for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
1170		ret = trace_seq_putc(s, ' ');
1171		if (!ret)
1172			return TRACE_TYPE_PARTIAL_LINE;
1173	}
1174
1175	/*
1176	 * If the return function does not have a matching entry,
1177	 * then the entry was lost. Instead of just printing
1178	 * the '}' and letting the user guess what function this
1179	 * belongs to, write out the function name.
 
1180	 */
1181	if (func_match) {
1182		ret = trace_seq_puts(s, "}\n");
1183		if (!ret)
1184			return TRACE_TYPE_PARTIAL_LINE;
1185	} else {
1186		ret = trace_seq_printf(s, "} /* %ps */\n", (void *)trace->func);
1187		if (!ret)
1188			return TRACE_TYPE_PARTIAL_LINE;
1189	}
1190
1191	/* Overrun */
1192	if (flags & TRACE_GRAPH_PRINT_OVERRUN) {
1193		ret = trace_seq_printf(s, " (Overruns: %lu)\n",
1194					trace->overrun);
1195		if (!ret)
1196			return TRACE_TYPE_PARTIAL_LINE;
1197	}
1198
1199	ret = print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1200			      cpu, pid, flags);
1201	if (ret == TRACE_TYPE_PARTIAL_LINE)
1202		return TRACE_TYPE_PARTIAL_LINE;
1203
1204	return TRACE_TYPE_HANDLED;
1205}
1206
1207static enum print_line_t
1208print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1209		    struct trace_iterator *iter, u32 flags)
1210{
1211	unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
 
1212	struct fgraph_data *data = iter->private;
1213	struct trace_event *event;
1214	int depth = 0;
1215	int ret;
1216	int i;
1217
1218	if (data)
1219		depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1220
1221	if (print_graph_prologue(iter, s, 0, 0, flags))
1222		return TRACE_TYPE_PARTIAL_LINE;
1223
1224	/* No time */
1225	ret = print_graph_duration(0, s, flags | FLAGS_FILL_FULL);
1226	if (ret != TRACE_TYPE_HANDLED)
1227		return ret;
1228
1229	/* Indentation */
1230	if (depth > 0)
1231		for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++) {
1232			ret = trace_seq_putc(s, ' ');
1233			if (!ret)
1234				return TRACE_TYPE_PARTIAL_LINE;
1235		}
1236
1237	/* The comment */
1238	ret = trace_seq_puts(s, "/* ");
1239	if (!ret)
1240		return TRACE_TYPE_PARTIAL_LINE;
1241
1242	switch (iter->ent->type) {
1243	case TRACE_BPRINT:
1244		ret = trace_print_bprintk_msg_only(iter);
1245		if (ret != TRACE_TYPE_HANDLED)
1246			return ret;
1247		break;
1248	case TRACE_PRINT:
1249		ret = trace_print_printk_msg_only(iter);
1250		if (ret != TRACE_TYPE_HANDLED)
1251			return ret;
1252		break;
1253	default:
1254		event = ftrace_find_event(ent->type);
1255		if (!event)
1256			return TRACE_TYPE_UNHANDLED;
1257
1258		ret = event->funcs->trace(iter, sym_flags, event);
1259		if (ret != TRACE_TYPE_HANDLED)
1260			return ret;
1261	}
1262
 
 
 
1263	/* Strip ending newline */
1264	if (s->buffer[s->len - 1] == '\n') {
1265		s->buffer[s->len - 1] = '\0';
1266		s->len--;
1267	}
1268
1269	ret = trace_seq_puts(s, " */\n");
1270	if (!ret)
1271		return TRACE_TYPE_PARTIAL_LINE;
1272
1273	return TRACE_TYPE_HANDLED;
1274}
1275
1276
1277enum print_line_t
1278print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1279{
1280	struct ftrace_graph_ent_entry *field;
1281	struct fgraph_data *data = iter->private;
1282	struct trace_entry *entry = iter->ent;
1283	struct trace_seq *s = &iter->seq;
1284	int cpu = iter->cpu;
1285	int ret;
1286
1287	if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1288		per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1289		return TRACE_TYPE_HANDLED;
1290	}
1291
1292	/*
1293	 * If the last output failed, there's a possibility we need
1294	 * to print out the missing entry which would never go out.
1295	 */
1296	if (data && data->failed) {
1297		field = &data->ent;
1298		iter->cpu = data->cpu;
1299		ret = print_graph_entry(field, s, iter, flags);
1300		if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1301			per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1302			ret = TRACE_TYPE_NO_CONSUME;
1303		}
1304		iter->cpu = cpu;
1305		return ret;
1306	}
1307
1308	switch (entry->type) {
1309	case TRACE_GRAPH_ENT: {
1310		/*
1311		 * print_graph_entry() may consume the current event,
1312		 * thus @field may become invalid, so we need to save it.
1313		 * sizeof(struct ftrace_graph_ent_entry) is very small,
1314		 * it can be safely saved at the stack.
1315		 */
1316		struct ftrace_graph_ent_entry saved;
1317		trace_assign_type(field, entry);
1318		saved = *field;
1319		return print_graph_entry(&saved, s, iter, flags);
1320	}
1321	case TRACE_GRAPH_RET: {
1322		struct ftrace_graph_ret_entry *field;
1323		trace_assign_type(field, entry);
1324		return print_graph_return(&field->ret, s, entry, iter, flags);
1325	}
1326	case TRACE_STACK:
1327	case TRACE_FN:
1328		/* dont trace stack and functions as comments */
1329		return TRACE_TYPE_UNHANDLED;
1330
1331	default:
1332		return print_graph_comment(s, entry, iter, flags);
1333	}
1334
1335	return TRACE_TYPE_HANDLED;
1336}
1337
1338static enum print_line_t
1339print_graph_function(struct trace_iterator *iter)
1340{
1341	return print_graph_function_flags(iter, tracer_flags.val);
1342}
1343
1344static enum print_line_t
1345print_graph_function_event(struct trace_iterator *iter, int flags,
1346			   struct trace_event *event)
1347{
1348	return print_graph_function(iter);
1349}
1350
1351static void print_lat_header(struct seq_file *s, u32 flags)
1352{
1353	static const char spaces[] = "                "	/* 16 spaces */
1354		"    "					/* 4 spaces */
1355		"                 ";			/* 17 spaces */
1356	int size = 0;
1357
1358	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1359		size += 16;
1360	if (flags & TRACE_GRAPH_PRINT_CPU)
1361		size += 4;
1362	if (flags & TRACE_GRAPH_PRINT_PROC)
1363		size += 17;
1364
1365	seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1366	seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1367	seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1368	seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1369	seq_printf(s, "#%.*s||| /                      \n", size, spaces);
1370}
1371
1372static void __print_graph_headers_flags(struct seq_file *s, u32 flags)
 
1373{
1374	int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
1375
1376	if (lat)
1377		print_lat_header(s, flags);
1378
1379	/* 1st line */
1380	seq_printf(s, "#");
1381	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1382		seq_printf(s, "     TIME       ");
1383	if (flags & TRACE_GRAPH_PRINT_CPU)
1384		seq_printf(s, " CPU");
1385	if (flags & TRACE_GRAPH_PRINT_PROC)
1386		seq_printf(s, "  TASK/PID       ");
1387	if (lat)
1388		seq_printf(s, "||||");
1389	if (flags & TRACE_GRAPH_PRINT_DURATION)
1390		seq_printf(s, "  DURATION   ");
1391	seq_printf(s, "               FUNCTION CALLS\n");
1392
1393	/* 2nd line */
1394	seq_printf(s, "#");
1395	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1396		seq_printf(s, "      |         ");
1397	if (flags & TRACE_GRAPH_PRINT_CPU)
1398		seq_printf(s, " |  ");
1399	if (flags & TRACE_GRAPH_PRINT_PROC)
1400		seq_printf(s, "   |    |        ");
1401	if (lat)
1402		seq_printf(s, "||||");
1403	if (flags & TRACE_GRAPH_PRINT_DURATION)
1404		seq_printf(s, "   |   |      ");
1405	seq_printf(s, "               |   |   |   |\n");
1406}
1407
1408void print_graph_headers(struct seq_file *s)
1409{
1410	print_graph_headers_flags(s, tracer_flags.val);
1411}
1412
1413void print_graph_headers_flags(struct seq_file *s, u32 flags)
1414{
1415	struct trace_iterator *iter = s->private;
 
1416
1417	if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
1418		return;
1419
1420	if (trace_flags & TRACE_ITER_LATENCY_FMT) {
1421		/* print nothing if the buffers are empty */
1422		if (trace_empty(iter))
1423			return;
1424
1425		print_trace_header(s, iter);
1426	}
1427
1428	__print_graph_headers_flags(s, flags);
1429}
1430
1431void graph_trace_open(struct trace_iterator *iter)
1432{
1433	/* pid and depth on the last trace processed */
1434	struct fgraph_data *data;
 
1435	int cpu;
1436
1437	iter->private = NULL;
1438
1439	data = kzalloc(sizeof(*data), GFP_KERNEL);
 
 
 
1440	if (!data)
1441		goto out_err;
1442
1443	data->cpu_data = alloc_percpu(struct fgraph_cpu_data);
1444	if (!data->cpu_data)
1445		goto out_err_free;
1446
1447	for_each_possible_cpu(cpu) {
1448		pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1449		int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1450		int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1451		int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1452
1453		*pid = -1;
1454		*depth = 0;
1455		*ignore = 0;
1456		*depth_irq = -1;
1457	}
1458
1459	iter->private = data;
1460
1461	return;
1462
1463 out_err_free:
1464	kfree(data);
1465 out_err:
1466	pr_warning("function graph tracer: not enough memory\n");
1467}
1468
1469void graph_trace_close(struct trace_iterator *iter)
1470{
1471	struct fgraph_data *data = iter->private;
1472
1473	if (data) {
1474		free_percpu(data->cpu_data);
1475		kfree(data);
1476	}
1477}
1478
1479static int
1480func_graph_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1481{
1482	if (bit == TRACE_GRAPH_PRINT_IRQS)
1483		ftrace_graph_skip_irqs = !set;
1484
 
 
 
 
 
 
1485	return 0;
1486}
1487
1488static struct trace_event_functions graph_functions = {
1489	.trace		= print_graph_function_event,
1490};
1491
1492static struct trace_event graph_trace_entry_event = {
1493	.type		= TRACE_GRAPH_ENT,
1494	.funcs		= &graph_functions,
1495};
1496
1497static struct trace_event graph_trace_ret_event = {
1498	.type		= TRACE_GRAPH_RET,
1499	.funcs		= &graph_functions
1500};
1501
1502static struct tracer graph_trace __tracer_data = {
1503	.name		= "function_graph",
 
1504	.open		= graph_trace_open,
1505	.pipe_open	= graph_trace_open,
1506	.close		= graph_trace_close,
1507	.pipe_close	= graph_trace_close,
1508	.wait_pipe	= poll_wait_pipe,
1509	.init		= graph_trace_init,
1510	.reset		= graph_trace_reset,
1511	.print_line	= print_graph_function,
1512	.print_header	= print_graph_headers,
1513	.flags		= &tracer_flags,
1514	.set_flag	= func_graph_set_flag,
1515#ifdef CONFIG_FTRACE_SELFTEST
1516	.selftest	= trace_selftest_startup_function_graph,
1517#endif
1518};
1519
1520
1521static ssize_t
1522graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1523		  loff_t *ppos)
1524{
1525	unsigned long val;
1526	int ret;
1527
1528	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1529	if (ret)
1530		return ret;
1531
1532	max_depth = val;
1533
1534	*ppos += cnt;
1535
1536	return cnt;
1537}
1538
1539static ssize_t
1540graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1541		 loff_t *ppos)
1542{
1543	char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1544	int n;
1545
1546	n = sprintf(buf, "%d\n", max_depth);
1547
1548	return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1549}
1550
1551static const struct file_operations graph_depth_fops = {
1552	.open		= tracing_open_generic,
1553	.write		= graph_depth_write,
1554	.read		= graph_depth_read,
1555	.llseek		= generic_file_llseek,
1556};
1557
1558static __init int init_graph_debugfs(void)
1559{
1560	struct dentry *d_tracer;
1561
1562	d_tracer = tracing_init_dentry();
1563	if (!d_tracer)
1564		return 0;
1565
1566	trace_create_file("max_graph_depth", 0644, d_tracer,
1567			  NULL, &graph_depth_fops);
1568
1569	return 0;
1570}
1571fs_initcall(init_graph_debugfs);
1572
1573static __init int init_graph_trace(void)
1574{
1575	max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1576
1577	if (!register_ftrace_event(&graph_trace_entry_event)) {
1578		pr_warning("Warning: could not register graph trace events\n");
1579		return 1;
1580	}
1581
1582	if (!register_ftrace_event(&graph_trace_ret_event)) {
1583		pr_warning("Warning: could not register graph trace events\n");
1584		return 1;
1585	}
1586
1587	return register_tracer(&graph_trace);
1588}
1589
1590core_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);