1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Infrastructure for profiling code inserted by 'gcc -pg'.
   4 *
   5 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
   6 * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
   7 *
   8 * Originally ported from the -rt patch by:
   9 *   Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
  10 *
  11 * Based on code in the latency_tracer, that is:
  12 *
  13 *  Copyright (C) 2004-2006 Ingo Molnar
  14 *  Copyright (C) 2004 Nadia Yvette Chambers
  15 */
  16
  17#include <linux/stop_machine.h>
  18#include <linux/clocksource.h>
  19#include <linux/sched/task.h>
  20#include <linux/kallsyms.h>
  21#include <linux/security.h>
  22#include <linux/seq_file.h>
  23#include <linux/tracefs.h>
  24#include <linux/hardirq.h>
  25#include <linux/kthread.h>
  26#include <linux/uaccess.h>
  27#include <linux/bsearch.h>
  28#include <linux/module.h>
  29#include <linux/ftrace.h>
  30#include <linux/sysctl.h>
  31#include <linux/slab.h>
  32#include <linux/ctype.h>
  33#include <linux/sort.h>
  34#include <linux/list.h>
  35#include <linux/hash.h>
  36#include <linux/rcupdate.h>
  37#include <linux/kprobes.h>
  38
  39#include <trace/events/sched.h>
  40
  41#include <asm/sections.h>
  42#include <asm/setup.h>
  43
  44#include "ftrace_internal.h"
  45#include "trace_output.h"
  46#include "trace_stat.h"
  47
  48#define FTRACE_WARN_ON(cond)			\
  49	({					\
  50		int ___r = cond;		\
  51		if (WARN_ON(___r))		\
  52			ftrace_kill();		\
  53		___r;				\
  54	})
  55
  56#define FTRACE_WARN_ON_ONCE(cond)		\
  57	({					\
  58		int ___r = cond;		\
  59		if (WARN_ON_ONCE(___r))		\
  60			ftrace_kill();		\
  61		___r;				\
  62	})
  63
  64/* hash bits for specific function selection */
  65#define FTRACE_HASH_DEFAULT_BITS 10
  66#define FTRACE_HASH_MAX_BITS 12
  67
  68#ifdef CONFIG_DYNAMIC_FTRACE
  69#define INIT_OPS_HASH(opsname)	\
  70	.func_hash		= &opsname.local_hash,			\
  71	.local_hash.regex_lock	= __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
  72#else
  73#define INIT_OPS_HASH(opsname)
  74#endif
  75
  76enum {
  77	FTRACE_MODIFY_ENABLE_FL		= (1 << 0),
  78	FTRACE_MODIFY_MAY_SLEEP_FL	= (1 << 1),
  79};
  80
  81struct ftrace_ops ftrace_list_end __read_mostly = {
  82	.func		= ftrace_stub,
  83	.flags		= FTRACE_OPS_FL_STUB,
  84	INIT_OPS_HASH(ftrace_list_end)
  85};
  86
  87/* ftrace_enabled is a method to turn ftrace on or off */
  88int ftrace_enabled __read_mostly;
  89static int last_ftrace_enabled;
  90
  91/* Current function tracing op */
  92struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
  93/* What to set function_trace_op to */
  94static struct ftrace_ops *set_function_trace_op;
  95
  96static bool ftrace_pids_enabled(struct ftrace_ops *ops)
  97{
  98	struct trace_array *tr;
  99
 100	if (!(ops->flags & FTRACE_OPS_FL_PID) || !ops->private)
 101		return false;
 102
 103	tr = ops->private;
 104
 105	return tr->function_pids != NULL || tr->function_no_pids != NULL;
 106}
 107
 108static void ftrace_update_trampoline(struct ftrace_ops *ops);
 109
 110/*
 111 * ftrace_disabled is set when an anomaly is discovered.
 112 * ftrace_disabled is much stronger than ftrace_enabled.
 113 */
 114static int ftrace_disabled __read_mostly;
 115
 116DEFINE_MUTEX(ftrace_lock);
 117
 118struct ftrace_ops __rcu *ftrace_ops_list __read_mostly = &ftrace_list_end;
 119ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
 120struct ftrace_ops global_ops;
 121
 122#if ARCH_SUPPORTS_FTRACE_OPS
 123static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
 124				 struct ftrace_ops *op, struct ftrace_regs *fregs);
 125#else
 126/* See comment below, where ftrace_ops_list_func is defined */
 127static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
 128#define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
 129#endif
 130
 131static inline void ftrace_ops_init(struct ftrace_ops *ops)
 132{
 133#ifdef CONFIG_DYNAMIC_FTRACE
 134	if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
 135		mutex_init(&ops->local_hash.regex_lock);
 136		ops->func_hash = &ops->local_hash;
 137		ops->flags |= FTRACE_OPS_FL_INITIALIZED;
 138	}
 139#endif
 140}
 141
 142static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
 143			    struct ftrace_ops *op, struct ftrace_regs *fregs)
 144{
 145	struct trace_array *tr = op->private;
 146	int pid;
 147
 148	if (tr) {
 149		pid = this_cpu_read(tr->array_buffer.data->ftrace_ignore_pid);
 150		if (pid == FTRACE_PID_IGNORE)
 151			return;
 152		if (pid != FTRACE_PID_TRACE &&
 153		    pid != current->pid)
 154			return;
 155	}
 156
 157	op->saved_func(ip, parent_ip, op, fregs);
 158}
 159
 160static void ftrace_sync_ipi(void *data)
 161{
 162	/* Probably not needed, but do it anyway */
 163	smp_rmb();
 164}
 165
 166static ftrace_func_t ftrace_ops_get_list_func(struct ftrace_ops *ops)
 167{
 168	/*
 169	 * If this is a dynamic, RCU, or per CPU ops, or we force list func,
 170	 * then it needs to call the list anyway.
 171	 */
 172	if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_RCU) ||
 173	    FTRACE_FORCE_LIST_FUNC)
 174		return ftrace_ops_list_func;
 175
 176	return ftrace_ops_get_func(ops);
 177}
 178
 179static void update_ftrace_function(void)
 180{
 181	ftrace_func_t func;
 182
 183	/*
 184	 * Prepare the ftrace_ops that the arch callback will use.
 185	 * If there's only one ftrace_ops registered, the ftrace_ops_list
 186	 * will point to the ops we want.
 187	 */
 188	set_function_trace_op = rcu_dereference_protected(ftrace_ops_list,
 189						lockdep_is_held(&ftrace_lock));
 190
 191	/* If there's no ftrace_ops registered, just call the stub function */
 192	if (set_function_trace_op == &ftrace_list_end) {
 193		func = ftrace_stub;
 194
 195	/*
 196	 * If we are at the end of the list and this ops is
 197	 * recursion safe and not dynamic and the arch supports passing ops,
 198	 * then have the mcount trampoline call the function directly.
 199	 */
 200	} else if (rcu_dereference_protected(ftrace_ops_list->next,
 201			lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
 202		func = ftrace_ops_get_list_func(ftrace_ops_list);
 203
 204	} else {
 205		/* Just use the default ftrace_ops */
 206		set_function_trace_op = &ftrace_list_end;
 207		func = ftrace_ops_list_func;
 208	}
 209
 210	update_function_graph_func();
 211
 212	/* If there's no change, then do nothing more here */
 213	if (ftrace_trace_function == func)
 214		return;
 215
 216	/*
 217	 * If we are using the list function, it doesn't care
 218	 * about the function_trace_ops.
 219	 */
 220	if (func == ftrace_ops_list_func) {
 221		ftrace_trace_function = func;
 222		/*
 223		 * Don't even bother setting function_trace_ops,
 224		 * it would be racy to do so anyway.
 225		 */
 226		return;
 227	}
 228
 229#ifndef CONFIG_DYNAMIC_FTRACE
 230	/*
 231	 * For static tracing, we need to be a bit more careful.
 232	 * The function change takes affect immediately. Thus,
 233	 * we need to coordinate the setting of the function_trace_ops
 234	 * with the setting of the ftrace_trace_function.
 235	 *
 236	 * Set the function to the list ops, which will call the
 237	 * function we want, albeit indirectly, but it handles the
 238	 * ftrace_ops and doesn't depend on function_trace_op.
 239	 */
 240	ftrace_trace_function = ftrace_ops_list_func;
 241	/*
 242	 * Make sure all CPUs see this. Yes this is slow, but static
 243	 * tracing is slow and nasty to have enabled.
 244	 */
 245	synchronize_rcu_tasks_rude();
 246	/* Now all cpus are using the list ops. */
 247	function_trace_op = set_function_trace_op;
 248	/* Make sure the function_trace_op is visible on all CPUs */
 249	smp_wmb();
 250	/* Nasty way to force a rmb on all cpus */
 251	smp_call_function(ftrace_sync_ipi, NULL, 1);
 252	/* OK, we are all set to update the ftrace_trace_function now! */
 253#endif /* !CONFIG_DYNAMIC_FTRACE */
 254
 255	ftrace_trace_function = func;
 256}
 257
 258static void add_ftrace_ops(struct ftrace_ops __rcu **list,
 259			   struct ftrace_ops *ops)
 260{
 261	rcu_assign_pointer(ops->next, *list);
 262
 263	/*
 264	 * We are entering ops into the list but another
 265	 * CPU might be walking that list. We need to make sure
 266	 * the ops->next pointer is valid before another CPU sees
 267	 * the ops pointer included into the list.
 268	 */
 269	rcu_assign_pointer(*list, ops);
 270}
 271
 272static int remove_ftrace_ops(struct ftrace_ops __rcu **list,
 273			     struct ftrace_ops *ops)
 274{
 275	struct ftrace_ops **p;
 276
 277	/*
 278	 * If we are removing the last function, then simply point
 279	 * to the ftrace_stub.
 280	 */
 281	if (rcu_dereference_protected(*list,
 282			lockdep_is_held(&ftrace_lock)) == ops &&
 283	    rcu_dereference_protected(ops->next,
 284			lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
 285		*list = &ftrace_list_end;
 286		return 0;
 287	}
 288
 289	for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
 290		if (*p == ops)
 291			break;
 292
 293	if (*p != ops)
 294		return -1;
 295
 296	*p = (*p)->next;
 297	return 0;
 298}
 299
 300static void ftrace_update_trampoline(struct ftrace_ops *ops);
 301
 302int __register_ftrace_function(struct ftrace_ops *ops)
 303{
 304	if (ops->flags & FTRACE_OPS_FL_DELETED)
 305		return -EINVAL;
 306
 307	if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
 308		return -EBUSY;
 309
 310#ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
 311	/*
 312	 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
 313	 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
 314	 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
 315	 */
 316	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
 317	    !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
 318		return -EINVAL;
 319
 320	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
 321		ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
 322#endif
 323	if (!ftrace_enabled && (ops->flags & FTRACE_OPS_FL_PERMANENT))
 324		return -EBUSY;
 325
 326	if (!core_kernel_data((unsigned long)ops))
 327		ops->flags |= FTRACE_OPS_FL_DYNAMIC;
 328
 329	add_ftrace_ops(&ftrace_ops_list, ops);
 330
 331	/* Always save the function, and reset at unregistering */
 332	ops->saved_func = ops->func;
 333
 334	if (ftrace_pids_enabled(ops))
 335		ops->func = ftrace_pid_func;
 336
 337	ftrace_update_trampoline(ops);
 338
 339	if (ftrace_enabled)
 340		update_ftrace_function();
 341
 342	return 0;
 343}
 344
 345int __unregister_ftrace_function(struct ftrace_ops *ops)
 346{
 347	int ret;
 348
 349	if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
 350		return -EBUSY;
 351
 352	ret = remove_ftrace_ops(&ftrace_ops_list, ops);
 353
 354	if (ret < 0)
 355		return ret;
 356
 357	if (ftrace_enabled)
 358		update_ftrace_function();
 359
 360	ops->func = ops->saved_func;
 361
 362	return 0;
 363}
 364
 365static void ftrace_update_pid_func(void)
 366{
 367	struct ftrace_ops *op;
 368
 369	/* Only do something if we are tracing something */
 370	if (ftrace_trace_function == ftrace_stub)
 371		return;
 372
 373	do_for_each_ftrace_op(op, ftrace_ops_list) {
 374		if (op->flags & FTRACE_OPS_FL_PID) {
 375			op->func = ftrace_pids_enabled(op) ?
 376				ftrace_pid_func : op->saved_func;
 377			ftrace_update_trampoline(op);
 378		}
 379	} while_for_each_ftrace_op(op);
 380
 381	update_ftrace_function();
 382}
 383
 384#ifdef CONFIG_FUNCTION_PROFILER
 385struct ftrace_profile {
 386	struct hlist_node		node;
 387	unsigned long			ip;
 388	unsigned long			counter;
 389#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 390	unsigned long long		time;
 391	unsigned long long		time_squared;
 392#endif
 393};
 394
 395struct ftrace_profile_page {
 396	struct ftrace_profile_page	*next;
 397	unsigned long			index;
 398	struct ftrace_profile		records[];
 399};
 400
 401struct ftrace_profile_stat {
 402	atomic_t			disabled;
 403	struct hlist_head		*hash;
 404	struct ftrace_profile_page	*pages;
 405	struct ftrace_profile_page	*start;
 406	struct tracer_stat		stat;
 407};
 408
 409#define PROFILE_RECORDS_SIZE						\
 410	(PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
 411
 412#define PROFILES_PER_PAGE					\
 413	(PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
 414
 415static int ftrace_profile_enabled __read_mostly;
 416
 417/* ftrace_profile_lock - synchronize the enable and disable of the profiler */
 418static DEFINE_MUTEX(ftrace_profile_lock);
 419
 420static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
 421
 422#define FTRACE_PROFILE_HASH_BITS 10
 423#define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
 424
 425static void *
 426function_stat_next(void *v, int idx)
 427{
 428	struct ftrace_profile *rec = v;
 429	struct ftrace_profile_page *pg;
 430
 431	pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
 432
 433 again:
 434	if (idx != 0)
 435		rec++;
 436
 437	if ((void *)rec >= (void *)&pg->records[pg->index]) {
 438		pg = pg->next;
 439		if (!pg)
 440			return NULL;
 441		rec = &pg->records[0];
 442		if (!rec->counter)
 443			goto again;
 444	}
 445
 446	return rec;
 447}
 448
 449static void *function_stat_start(struct tracer_stat *trace)
 450{
 451	struct ftrace_profile_stat *stat =
 452		container_of(trace, struct ftrace_profile_stat, stat);
 453
 454	if (!stat || !stat->start)
 455		return NULL;
 456
 457	return function_stat_next(&stat->start->records[0], 0);
 458}
 459
 460#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 461/* function graph compares on total time */
 462static int function_stat_cmp(const void *p1, const void *p2)
 463{
 464	const struct ftrace_profile *a = p1;
 465	const struct ftrace_profile *b = p2;
 466
 467	if (a->time < b->time)
 468		return -1;
 469	if (a->time > b->time)
 470		return 1;
 471	else
 472		return 0;
 473}
 474#else
 475/* not function graph compares against hits */
 476static int function_stat_cmp(const void *p1, const void *p2)
 477{
 478	const struct ftrace_profile *a = p1;
 479	const struct ftrace_profile *b = p2;
 480
 481	if (a->counter < b->counter)
 482		return -1;
 483	if (a->counter > b->counter)
 484		return 1;
 485	else
 486		return 0;
 487}
 488#endif
 489
 490static int function_stat_headers(struct seq_file *m)
 491{
 492#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 493	seq_puts(m, "  Function                               "
 494		 "Hit    Time            Avg             s^2\n"
 495		    "  --------                               "
 496		 "---    ----            ---             ---\n");
 497#else
 498	seq_puts(m, "  Function                               Hit\n"
 499		    "  --------                               ---\n");
 500#endif
 501	return 0;
 502}
 503
 504static int function_stat_show(struct seq_file *m, void *v)
 505{
 506	struct ftrace_profile *rec = v;
 507	char str[KSYM_SYMBOL_LEN];
 508	int ret = 0;
 509#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 510	static struct trace_seq s;
 511	unsigned long long avg;
 512	unsigned long long stddev;
 513#endif
 514	mutex_lock(&ftrace_profile_lock);
 515
 516	/* we raced with function_profile_reset() */
 517	if (unlikely(rec->counter == 0)) {
 518		ret = -EBUSY;
 519		goto out;
 520	}
 521
 522#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 523	avg = div64_ul(rec->time, rec->counter);
 524	if (tracing_thresh && (avg < tracing_thresh))
 525		goto out;
 526#endif
 527
 528	kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
 529	seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
 530
 531#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 532	seq_puts(m, "    ");
 533
 534	/* Sample standard deviation (s^2) */
 535	if (rec->counter <= 1)
 536		stddev = 0;
 537	else {
 538		/*
 539		 * Apply Welford's method:
 540		 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
 541		 */
 542		stddev = rec->counter * rec->time_squared -
 543			 rec->time * rec->time;
 544
 545		/*
 546		 * Divide only 1000 for ns^2 -> us^2 conversion.
 547		 * trace_print_graph_duration will divide 1000 again.
 548		 */
 549		stddev = div64_ul(stddev,
 550				  rec->counter * (rec->counter - 1) * 1000);
 551	}
 552
 553	trace_seq_init(&s);
 554	trace_print_graph_duration(rec->time, &s);
 555	trace_seq_puts(&s, "    ");
 556	trace_print_graph_duration(avg, &s);
 557	trace_seq_puts(&s, "    ");
 558	trace_print_graph_duration(stddev, &s);
 559	trace_print_seq(m, &s);
 560#endif
 561	seq_putc(m, '\n');
 562out:
 563	mutex_unlock(&ftrace_profile_lock);
 564
 565	return ret;
 566}
 567
 568static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
 569{
 570	struct ftrace_profile_page *pg;
 571
 572	pg = stat->pages = stat->start;
 573
 574	while (pg) {
 575		memset(pg->records, 0, PROFILE_RECORDS_SIZE);
 576		pg->index = 0;
 577		pg = pg->next;
 578	}
 579
 580	memset(stat->hash, 0,
 581	       FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
 582}
 583
 584int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
 585{
 586	struct ftrace_profile_page *pg;
 587	int functions;
 588	int pages;
 589	int i;
 590
 591	/* If we already allocated, do nothing */
 592	if (stat->pages)
 593		return 0;
 594
 595	stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
 596	if (!stat->pages)
 597		return -ENOMEM;
 598
 599#ifdef CONFIG_DYNAMIC_FTRACE
 600	functions = ftrace_update_tot_cnt;
 601#else
 602	/*
 603	 * We do not know the number of functions that exist because
 604	 * dynamic tracing is what counts them. With past experience
 605	 * we have around 20K functions. That should be more than enough.
 606	 * It is highly unlikely we will execute every function in
 607	 * the kernel.
 608	 */
 609	functions = 20000;
 610#endif
 611
 612	pg = stat->start = stat->pages;
 613
 614	pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
 615
 616	for (i = 1; i < pages; i++) {
 617		pg->next = (void *)get_zeroed_page(GFP_KERNEL);
 618		if (!pg->next)
 619			goto out_free;
 620		pg = pg->next;
 621	}
 622
 623	return 0;
 624
 625 out_free:
 626	pg = stat->start;
 627	while (pg) {
 628		unsigned long tmp = (unsigned long)pg;
 629
 630		pg = pg->next;
 631		free_page(tmp);
 632	}
 633
 634	stat->pages = NULL;
 635	stat->start = NULL;
 636
 637	return -ENOMEM;
 638}
 639
 640static int ftrace_profile_init_cpu(int cpu)
 641{
 642	struct ftrace_profile_stat *stat;
 643	int size;
 644
 645	stat = &per_cpu(ftrace_profile_stats, cpu);
 646
 647	if (stat->hash) {
 648		/* If the profile is already created, simply reset it */
 649		ftrace_profile_reset(stat);
 650		return 0;
 651	}
 652
 653	/*
 654	 * We are profiling all functions, but usually only a few thousand
 655	 * functions are hit. We'll make a hash of 1024 items.
 656	 */
 657	size = FTRACE_PROFILE_HASH_SIZE;
 658
 659	stat->hash = kcalloc(size, sizeof(struct hlist_head), GFP_KERNEL);
 660
 661	if (!stat->hash)
 662		return -ENOMEM;
 663
 664	/* Preallocate the function profiling pages */
 665	if (ftrace_profile_pages_init(stat) < 0) {
 666		kfree(stat->hash);
 667		stat->hash = NULL;
 668		return -ENOMEM;
 669	}
 670
 671	return 0;
 672}
 673
 674static int ftrace_profile_init(void)
 675{
 676	int cpu;
 677	int ret = 0;
 678
 679	for_each_possible_cpu(cpu) {
 680		ret = ftrace_profile_init_cpu(cpu);
 681		if (ret)
 682			break;
 683	}
 684
 685	return ret;
 686}
 687
 688/* interrupts must be disabled */
 689static struct ftrace_profile *
 690ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
 691{
 692	struct ftrace_profile *rec;
 693	struct hlist_head *hhd;
 694	unsigned long key;
 695
 696	key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
 697	hhd = &stat->hash[key];
 698
 699	if (hlist_empty(hhd))
 700		return NULL;
 701
 702	hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
 703		if (rec->ip == ip)
 704			return rec;
 705	}
 706
 707	return NULL;
 708}
 709
 710static void ftrace_add_profile(struct ftrace_profile_stat *stat,
 711			       struct ftrace_profile *rec)
 712{
 713	unsigned long key;
 714
 715	key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
 716	hlist_add_head_rcu(&rec->node, &stat->hash[key]);
 717}
 718
 719/*
 720 * The memory is already allocated, this simply finds a new record to use.
 721 */
 722static struct ftrace_profile *
 723ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
 724{
 725	struct ftrace_profile *rec = NULL;
 726
 727	/* prevent recursion (from NMIs) */
 728	if (atomic_inc_return(&stat->disabled) != 1)
 729		goto out;
 730
 731	/*
 732	 * Try to find the function again since an NMI
 733	 * could have added it
 734	 */
 735	rec = ftrace_find_profiled_func(stat, ip);
 736	if (rec)
 737		goto out;
 738
 739	if (stat->pages->index == PROFILES_PER_PAGE) {
 740		if (!stat->pages->next)
 741			goto out;
 742		stat->pages = stat->pages->next;
 743	}
 744
 745	rec = &stat->pages->records[stat->pages->index++];
 746	rec->ip = ip;
 747	ftrace_add_profile(stat, rec);
 748
 749 out:
 750	atomic_dec(&stat->disabled);
 751
 752	return rec;
 753}
 754
 755static void
 756function_profile_call(unsigned long ip, unsigned long parent_ip,
 757		      struct ftrace_ops *ops, struct ftrace_regs *fregs)
 758{
 759	struct ftrace_profile_stat *stat;
 760	struct ftrace_profile *rec;
 761	unsigned long flags;
 762
 763	if (!ftrace_profile_enabled)
 764		return;
 765
 766	local_irq_save(flags);
 767
 768	stat = this_cpu_ptr(&ftrace_profile_stats);
 769	if (!stat->hash || !ftrace_profile_enabled)
 770		goto out;
 771
 772	rec = ftrace_find_profiled_func(stat, ip);
 773	if (!rec) {
 774		rec = ftrace_profile_alloc(stat, ip);
 775		if (!rec)
 776			goto out;
 777	}
 778
 779	rec->counter++;
 780 out:
 781	local_irq_restore(flags);
 782}
 783
 784#ifdef CONFIG_FUNCTION_GRAPH_TRACER
 785static bool fgraph_graph_time = true;
 786
 787void ftrace_graph_graph_time_control(bool enable)
 788{
 789	fgraph_graph_time = enable;
 790}
 791
 792static int profile_graph_entry(struct ftrace_graph_ent *trace)
 793{
 794	struct ftrace_ret_stack *ret_stack;
 795
 796	function_profile_call(trace->func, 0, NULL, NULL);
 797
 798	/* If function graph is shutting down, ret_stack can be NULL */
 799	if (!current->ret_stack)
 800		return 0;
 801
 802	ret_stack = ftrace_graph_get_ret_stack(current, 0);
 803	if (ret_stack)
 804		ret_stack->subtime = 0;
 805
 806	return 1;
 807}
 808
 809static void profile_graph_return(struct ftrace_graph_ret *trace)
 810{
 811	struct ftrace_ret_stack *ret_stack;
 812	struct ftrace_profile_stat *stat;
 813	unsigned long long calltime;
 814	struct ftrace_profile *rec;
 815	unsigned long flags;
 816
 817	local_irq_save(flags);
 818	stat = this_cpu_ptr(&ftrace_profile_stats);
 819	if (!stat->hash || !ftrace_profile_enabled)
 820		goto out;
 821
 822	/* If the calltime was zero'd ignore it */
 823	if (!trace->calltime)
 824		goto out;
 825
 826	calltime = trace->rettime - trace->calltime;
 827
 828	if (!fgraph_graph_time) {
 829
 830		/* Append this call time to the parent time to subtract */
 831		ret_stack = ftrace_graph_get_ret_stack(current, 1);
 832		if (ret_stack)
 833			ret_stack->subtime += calltime;
 834
 835		ret_stack = ftrace_graph_get_ret_stack(current, 0);
 836		if (ret_stack && ret_stack->subtime < calltime)
 837			calltime -= ret_stack->subtime;
 838		else
 839			calltime = 0;
 840	}
 841
 842	rec = ftrace_find_profiled_func(stat, trace->func);
 843	if (rec) {
 844		rec->time += calltime;
 845		rec->time_squared += calltime * calltime;
 846	}
 847
 848 out:
 849	local_irq_restore(flags);
 850}
 851
 852static struct fgraph_ops fprofiler_ops = {
 853	.entryfunc = &profile_graph_entry,
 854	.retfunc = &profile_graph_return,
 855};
 856
 857static int register_ftrace_profiler(void)
 858{
 859	return register_ftrace_graph(&fprofiler_ops);
 860}
 861
 862static void unregister_ftrace_profiler(void)
 863{
 864	unregister_ftrace_graph(&fprofiler_ops);
 865}
 866#else
 867static struct ftrace_ops ftrace_profile_ops __read_mostly = {
 868	.func		= function_profile_call,
 869	.flags		= FTRACE_OPS_FL_INITIALIZED,
 870	INIT_OPS_HASH(ftrace_profile_ops)
 871};
 872
 873static int register_ftrace_profiler(void)
 874{
 875	return register_ftrace_function(&ftrace_profile_ops);
 876}
 877
 878static void unregister_ftrace_profiler(void)
 879{
 880	unregister_ftrace_function(&ftrace_profile_ops);
 881}
 882#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
 883
 884static ssize_t
 885ftrace_profile_write(struct file *filp, const char __user *ubuf,
 886		     size_t cnt, loff_t *ppos)
 887{
 888	unsigned long val;
 889	int ret;
 890
 891	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
 892	if (ret)
 893		return ret;
 894
 895	val = !!val;
 896
 897	mutex_lock(&ftrace_profile_lock);
 898	if (ftrace_profile_enabled ^ val) {
 899		if (val) {
 900			ret = ftrace_profile_init();
 901			if (ret < 0) {
 902				cnt = ret;
 903				goto out;
 904			}
 905
 906			ret = register_ftrace_profiler();
 907			if (ret < 0) {
 908				cnt = ret;
 909				goto out;
 910			}
 911			ftrace_profile_enabled = 1;
 912		} else {
 913			ftrace_profile_enabled = 0;
 914			/*
 915			 * unregister_ftrace_profiler calls stop_machine
 916			 * so this acts like an synchronize_rcu.
 917			 */
 918			unregister_ftrace_profiler();
 919		}
 920	}
 921 out:
 922	mutex_unlock(&ftrace_profile_lock);
 923
 924	*ppos += cnt;
 925
 926	return cnt;
 927}
 928
 929static ssize_t
 930ftrace_profile_read(struct file *filp, char __user *ubuf,
 931		     size_t cnt, loff_t *ppos)
 932{
 933	char buf[64];		/* big enough to hold a number */
 934	int r;
 935
 936	r = sprintf(buf, "%u\n", ftrace_profile_enabled);
 937	return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
 938}
 939
 940static const struct file_operations ftrace_profile_fops = {
 941	.open		= tracing_open_generic,
 942	.read		= ftrace_profile_read,
 943	.write		= ftrace_profile_write,
 944	.llseek		= default_llseek,
 945};
 946
 947/* used to initialize the real stat files */
 948static struct tracer_stat function_stats __initdata = {
 949	.name		= "functions",
 950	.stat_start	= function_stat_start,
 951	.stat_next	= function_stat_next,
 952	.stat_cmp	= function_stat_cmp,
 953	.stat_headers	= function_stat_headers,
 954	.stat_show	= function_stat_show
 955};
 956
 957static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
 958{
 959	struct ftrace_profile_stat *stat;
 960	struct dentry *entry;
 961	char *name;
 962	int ret;
 963	int cpu;
 964
 965	for_each_possible_cpu(cpu) {
 966		stat = &per_cpu(ftrace_profile_stats, cpu);
 967
 968		name = kasprintf(GFP_KERNEL, "function%d", cpu);
 969		if (!name) {
 970			/*
 971			 * The files created are permanent, if something happens
 972			 * we still do not free memory.
 973			 */
 974			WARN(1,
 975			     "Could not allocate stat file for cpu %d\n",
 976			     cpu);
 977			return;
 978		}
 979		stat->stat = function_stats;
 980		stat->stat.name = name;
 981		ret = register_stat_tracer(&stat->stat);
 982		if (ret) {
 983			WARN(1,
 984			     "Could not register function stat for cpu %d\n",
 985			     cpu);
 986			kfree(name);
 987			return;
 988		}
 989	}
 990
 991	entry = tracefs_create_file("function_profile_enabled", 0644,
 992				    d_tracer, NULL, &ftrace_profile_fops);
 993	if (!entry)
 994		pr_warn("Could not create tracefs 'function_profile_enabled' entry\n");
 995}
 996
 997#else /* CONFIG_FUNCTION_PROFILER */
 998static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
 999{
1000}
1001#endif /* CONFIG_FUNCTION_PROFILER */
1002
1003#ifdef CONFIG_DYNAMIC_FTRACE
1004
1005static struct ftrace_ops *removed_ops;
1006
1007/*
1008 * Set when doing a global update, like enabling all recs or disabling them.
1009 * It is not set when just updating a single ftrace_ops.
1010 */
1011static bool update_all_ops;
1012
1013#ifndef CONFIG_FTRACE_MCOUNT_RECORD
1014# error Dynamic ftrace depends on MCOUNT_RECORD
1015#endif
1016
1017struct ftrace_func_probe {
1018	struct ftrace_probe_ops	*probe_ops;
1019	struct ftrace_ops	ops;
1020	struct trace_array	*tr;
1021	struct list_head	list;
1022	void			*data;
1023	int			ref;
1024};
1025
1026/*
1027 * We make these constant because no one should touch them,
1028 * but they are used as the default "empty hash", to avoid allocating
1029 * it all the time. These are in a read only section such that if
1030 * anyone does try to modify it, it will cause an exception.
1031 */
1032static const struct hlist_head empty_buckets[1];
1033static const struct ftrace_hash empty_hash = {
1034	.buckets = (struct hlist_head *)empty_buckets,
1035};
1036#define EMPTY_HASH	((struct ftrace_hash *)&empty_hash)
1037
1038struct ftrace_ops global_ops = {
1039	.func				= ftrace_stub,
1040	.local_hash.notrace_hash	= EMPTY_HASH,
1041	.local_hash.filter_hash		= EMPTY_HASH,
1042	INIT_OPS_HASH(global_ops)
1043	.flags				= FTRACE_OPS_FL_INITIALIZED |
 
1044					  FTRACE_OPS_FL_PID,
1045};
1046
1047/*
1048 * Used by the stack unwinder to know about dynamic ftrace trampolines.
1049 */
1050struct ftrace_ops *ftrace_ops_trampoline(unsigned long addr)
1051{
1052	struct ftrace_ops *op = NULL;
1053
1054	/*
1055	 * Some of the ops may be dynamically allocated,
1056	 * they are freed after a synchronize_rcu().
1057	 */
1058	preempt_disable_notrace();
1059
1060	do_for_each_ftrace_op(op, ftrace_ops_list) {
1061		/*
1062		 * This is to check for dynamically allocated trampolines.
1063		 * Trampolines that are in kernel text will have
1064		 * core_kernel_text() return true.
1065		 */
1066		if (op->trampoline && op->trampoline_size)
1067			if (addr >= op->trampoline &&
1068			    addr < op->trampoline + op->trampoline_size) {
1069				preempt_enable_notrace();
1070				return op;
1071			}
1072	} while_for_each_ftrace_op(op);
1073	preempt_enable_notrace();
1074
1075	return NULL;
1076}
1077
1078/*
1079 * This is used by __kernel_text_address() to return true if the
1080 * address is on a dynamically allocated trampoline that would
1081 * not return true for either core_kernel_text() or
1082 * is_module_text_address().
1083 */
1084bool is_ftrace_trampoline(unsigned long addr)
1085{
1086	return ftrace_ops_trampoline(addr) != NULL;
1087}
1088
1089struct ftrace_page {
1090	struct ftrace_page	*next;
1091	struct dyn_ftrace	*records;
1092	int			index;
1093	int			order;
1094};
1095
1096#define ENTRY_SIZE sizeof(struct dyn_ftrace)
1097#define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1098
1099static struct ftrace_page	*ftrace_pages_start;
1100static struct ftrace_page	*ftrace_pages;
1101
1102static __always_inline unsigned long
1103ftrace_hash_key(struct ftrace_hash *hash, unsigned long ip)
1104{
1105	if (hash->size_bits > 0)
1106		return hash_long(ip, hash->size_bits);
1107
1108	return 0;
1109}
1110
1111/* Only use this function if ftrace_hash_empty() has already been tested */
1112static __always_inline struct ftrace_func_entry *
1113__ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1114{
1115	unsigned long key;
1116	struct ftrace_func_entry *entry;
1117	struct hlist_head *hhd;
1118
1119	key = ftrace_hash_key(hash, ip);
1120	hhd = &hash->buckets[key];
1121
1122	hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1123		if (entry->ip == ip)
1124			return entry;
1125	}
1126	return NULL;
1127}
1128
1129/**
1130 * ftrace_lookup_ip - Test to see if an ip exists in an ftrace_hash
1131 * @hash: The hash to look at
1132 * @ip: The instruction pointer to test
1133 *
1134 * Search a given @hash to see if a given instruction pointer (@ip)
1135 * exists in it.
1136 *
1137 * Returns the entry that holds the @ip if found. NULL otherwise.
1138 */
1139struct ftrace_func_entry *
1140ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1141{
1142	if (ftrace_hash_empty(hash))
1143		return NULL;
1144
1145	return __ftrace_lookup_ip(hash, ip);
1146}
1147
1148static void __add_hash_entry(struct ftrace_hash *hash,
1149			     struct ftrace_func_entry *entry)
1150{
1151	struct hlist_head *hhd;
1152	unsigned long key;
1153
1154	key = ftrace_hash_key(hash, entry->ip);
1155	hhd = &hash->buckets[key];
1156	hlist_add_head(&entry->hlist, hhd);
1157	hash->count++;
1158}
1159
1160static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1161{
1162	struct ftrace_func_entry *entry;
1163
1164	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1165	if (!entry)
1166		return -ENOMEM;
1167
1168	entry->ip = ip;
1169	__add_hash_entry(hash, entry);
1170
1171	return 0;
1172}
1173
1174static void
1175free_hash_entry(struct ftrace_hash *hash,
1176		  struct ftrace_func_entry *entry)
1177{
1178	hlist_del(&entry->hlist);
1179	kfree(entry);
1180	hash->count--;
1181}
1182
1183static void
1184remove_hash_entry(struct ftrace_hash *hash,
1185		  struct ftrace_func_entry *entry)
1186{
1187	hlist_del_rcu(&entry->hlist);
1188	hash->count--;
1189}
1190
1191static void ftrace_hash_clear(struct ftrace_hash *hash)
1192{
1193	struct hlist_head *hhd;
1194	struct hlist_node *tn;
1195	struct ftrace_func_entry *entry;
1196	int size = 1 << hash->size_bits;
1197	int i;
1198
1199	if (!hash->count)
1200		return;
1201
1202	for (i = 0; i < size; i++) {
1203		hhd = &hash->buckets[i];
1204		hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1205			free_hash_entry(hash, entry);
1206	}
1207	FTRACE_WARN_ON(hash->count);
1208}
1209
1210static void free_ftrace_mod(struct ftrace_mod_load *ftrace_mod)
1211{
1212	list_del(&ftrace_mod->list);
1213	kfree(ftrace_mod->module);
1214	kfree(ftrace_mod->func);
1215	kfree(ftrace_mod);
1216}
1217
1218static void clear_ftrace_mod_list(struct list_head *head)
1219{
1220	struct ftrace_mod_load *p, *n;
1221
1222	/* stack tracer isn't supported yet */
1223	if (!head)
1224		return;
1225
1226	mutex_lock(&ftrace_lock);
1227	list_for_each_entry_safe(p, n, head, list)
1228		free_ftrace_mod(p);
1229	mutex_unlock(&ftrace_lock);
1230}
1231
1232static void free_ftrace_hash(struct ftrace_hash *hash)
1233{
1234	if (!hash || hash == EMPTY_HASH)
1235		return;
1236	ftrace_hash_clear(hash);
1237	kfree(hash->buckets);
1238	kfree(hash);
1239}
1240
1241static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1242{
1243	struct ftrace_hash *hash;
1244
1245	hash = container_of(rcu, struct ftrace_hash, rcu);
1246	free_ftrace_hash(hash);
1247}
1248
1249static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1250{
1251	if (!hash || hash == EMPTY_HASH)
1252		return;
1253	call_rcu(&hash->rcu, __free_ftrace_hash_rcu);
1254}
1255
1256void ftrace_free_filter(struct ftrace_ops *ops)
1257{
1258	ftrace_ops_init(ops);
1259	free_ftrace_hash(ops->func_hash->filter_hash);
1260	free_ftrace_hash(ops->func_hash->notrace_hash);
1261}
1262
1263static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1264{
1265	struct ftrace_hash *hash;
1266	int size;
1267
1268	hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1269	if (!hash)
1270		return NULL;
1271
1272	size = 1 << size_bits;
1273	hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1274
1275	if (!hash->buckets) {
1276		kfree(hash);
1277		return NULL;
1278	}
1279
1280	hash->size_bits = size_bits;
1281
1282	return hash;
1283}
1284
1285
1286static int ftrace_add_mod(struct trace_array *tr,
1287			  const char *func, const char *module,
1288			  int enable)
1289{
1290	struct ftrace_mod_load *ftrace_mod;
1291	struct list_head *mod_head = enable ? &tr->mod_trace : &tr->mod_notrace;
1292
1293	ftrace_mod = kzalloc(sizeof(*ftrace_mod), GFP_KERNEL);
1294	if (!ftrace_mod)
1295		return -ENOMEM;
1296
1297	ftrace_mod->func = kstrdup(func, GFP_KERNEL);
1298	ftrace_mod->module = kstrdup(module, GFP_KERNEL);
1299	ftrace_mod->enable = enable;
1300
1301	if (!ftrace_mod->func || !ftrace_mod->module)
1302		goto out_free;
1303
1304	list_add(&ftrace_mod->list, mod_head);
1305
1306	return 0;
1307
1308 out_free:
1309	free_ftrace_mod(ftrace_mod);
1310
1311	return -ENOMEM;
1312}
1313
1314static struct ftrace_hash *
1315alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1316{
1317	struct ftrace_func_entry *entry;
1318	struct ftrace_hash *new_hash;
1319	int size;
1320	int ret;
1321	int i;
1322
1323	new_hash = alloc_ftrace_hash(size_bits);
1324	if (!new_hash)
1325		return NULL;
1326
1327	if (hash)
1328		new_hash->flags = hash->flags;
1329
1330	/* Empty hash? */
1331	if (ftrace_hash_empty(hash))
1332		return new_hash;
1333
1334	size = 1 << hash->size_bits;
1335	for (i = 0; i < size; i++) {
1336		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1337			ret = add_hash_entry(new_hash, entry->ip);
1338			if (ret < 0)
1339				goto free_hash;
1340		}
1341	}
1342
1343	FTRACE_WARN_ON(new_hash->count != hash->count);
1344
1345	return new_hash;
1346
1347 free_hash:
1348	free_ftrace_hash(new_hash);
1349	return NULL;
1350}
1351
1352static void
1353ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1354static void
1355ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1356
1357static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1358				       struct ftrace_hash *new_hash);
1359
1360static struct ftrace_hash *dup_hash(struct ftrace_hash *src, int size)
1361{
1362	struct ftrace_func_entry *entry;
1363	struct ftrace_hash *new_hash;
1364	struct hlist_head *hhd;
1365	struct hlist_node *tn;
1366	int bits = 0;
1367	int i;
1368
1369	/*
1370	 * Use around half the size (max bit of it), but
1371	 * a minimum of 2 is fine (as size of 0 or 1 both give 1 for bits).
1372	 */
1373	bits = fls(size / 2);
 
1374
1375	/* Don't allocate too much */
1376	if (bits > FTRACE_HASH_MAX_BITS)
1377		bits = FTRACE_HASH_MAX_BITS;
1378
1379	new_hash = alloc_ftrace_hash(bits);
1380	if (!new_hash)
1381		return NULL;
1382
1383	new_hash->flags = src->flags;
1384
1385	size = 1 << src->size_bits;
1386	for (i = 0; i < size; i++) {
1387		hhd = &src->buckets[i];
1388		hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1389			remove_hash_entry(src, entry);
1390			__add_hash_entry(new_hash, entry);
1391		}
1392	}
1393	return new_hash;
1394}
1395
1396static struct ftrace_hash *
1397__ftrace_hash_move(struct ftrace_hash *src)
1398{
1399	int size = src->count;
1400
1401	/*
1402	 * If the new source is empty, just return the empty_hash.
1403	 */
1404	if (ftrace_hash_empty(src))
1405		return EMPTY_HASH;
1406
1407	return dup_hash(src, size);
1408}
1409
1410static int
1411ftrace_hash_move(struct ftrace_ops *ops, int enable,
1412		 struct ftrace_hash **dst, struct ftrace_hash *src)
1413{
1414	struct ftrace_hash *new_hash;
1415	int ret;
1416
1417	/* Reject setting notrace hash on IPMODIFY ftrace_ops */
1418	if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1419		return -EINVAL;
1420
1421	new_hash = __ftrace_hash_move(src);
1422	if (!new_hash)
1423		return -ENOMEM;
1424
1425	/* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1426	if (enable) {
1427		/* IPMODIFY should be updated only when filter_hash updating */
1428		ret = ftrace_hash_ipmodify_update(ops, new_hash);
1429		if (ret < 0) {
1430			free_ftrace_hash(new_hash);
1431			return ret;
1432		}
1433	}
1434
1435	/*
1436	 * Remove the current set, update the hash and add
1437	 * them back.
1438	 */
1439	ftrace_hash_rec_disable_modify(ops, enable);
1440
1441	rcu_assign_pointer(*dst, new_hash);
1442
1443	ftrace_hash_rec_enable_modify(ops, enable);
1444
1445	return 0;
1446}
1447
1448static bool hash_contains_ip(unsigned long ip,
1449			     struct ftrace_ops_hash *hash)
1450{
1451	/*
1452	 * The function record is a match if it exists in the filter
1453	 * hash and not in the notrace hash. Note, an empty hash is
1454	 * considered a match for the filter hash, but an empty
1455	 * notrace hash is considered not in the notrace hash.
1456	 */
1457	return (ftrace_hash_empty(hash->filter_hash) ||
1458		__ftrace_lookup_ip(hash->filter_hash, ip)) &&
1459		(ftrace_hash_empty(hash->notrace_hash) ||
1460		 !__ftrace_lookup_ip(hash->notrace_hash, ip));
1461}
1462
1463/*
1464 * Test the hashes for this ops to see if we want to call
1465 * the ops->func or not.
1466 *
1467 * It's a match if the ip is in the ops->filter_hash or
1468 * the filter_hash does not exist or is empty,
1469 *  AND
1470 * the ip is not in the ops->notrace_hash.
1471 *
1472 * This needs to be called with preemption disabled as
1473 * the hashes are freed with call_rcu().
1474 */
1475int
1476ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1477{
1478	struct ftrace_ops_hash hash;
1479	int ret;
1480
1481#ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1482	/*
1483	 * There's a small race when adding ops that the ftrace handler
1484	 * that wants regs, may be called without them. We can not
1485	 * allow that handler to be called if regs is NULL.
1486	 */
1487	if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1488		return 0;
1489#endif
1490
1491	rcu_assign_pointer(hash.filter_hash, ops->func_hash->filter_hash);
1492	rcu_assign_pointer(hash.notrace_hash, ops->func_hash->notrace_hash);
1493
1494	if (hash_contains_ip(ip, &hash))
1495		ret = 1;
1496	else
1497		ret = 0;
1498
1499	return ret;
1500}
1501
1502/*
1503 * This is a double for. Do not use 'break' to break out of the loop,
1504 * you must use a goto.
1505 */
1506#define do_for_each_ftrace_rec(pg, rec)					\
1507	for (pg = ftrace_pages_start; pg; pg = pg->next) {		\
1508		int _____i;						\
1509		for (_____i = 0; _____i < pg->index; _____i++) {	\
1510			rec = &pg->records[_____i];
1511
1512#define while_for_each_ftrace_rec()		\
1513		}				\
1514	}
1515
1516
1517static int ftrace_cmp_recs(const void *a, const void *b)
1518{
1519	const struct dyn_ftrace *key = a;
1520	const struct dyn_ftrace *rec = b;
1521
1522	if (key->flags < rec->ip)
1523		return -1;
1524	if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1525		return 1;
1526	return 0;
1527}
1528
1529static struct dyn_ftrace *lookup_rec(unsigned long start, unsigned long end)
1530{
1531	struct ftrace_page *pg;
1532	struct dyn_ftrace *rec = NULL;
1533	struct dyn_ftrace key;
1534
1535	key.ip = start;
1536	key.flags = end;	/* overload flags, as it is unsigned long */
1537
1538	for (pg = ftrace_pages_start; pg; pg = pg->next) {
1539		if (end < pg->records[0].ip ||
1540		    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1541			continue;
1542		rec = bsearch(&key, pg->records, pg->index,
1543			      sizeof(struct dyn_ftrace),
1544			      ftrace_cmp_recs);
1545		if (rec)
1546			break;
1547	}
1548	return rec;
1549}
1550
1551/**
1552 * ftrace_location_range - return the first address of a traced location
1553 *	if it touches the given ip range
1554 * @start: start of range to search.
1555 * @end: end of range to search (inclusive). @end points to the last byte
1556 *	to check.
1557 *
1558 * Returns rec->ip if the related ftrace location is a least partly within
1559 * the given address range. That is, the first address of the instruction
1560 * that is either a NOP or call to the function tracer. It checks the ftrace
1561 * internal tables to determine if the address belongs or not.
1562 */
1563unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1564{
1565	struct dyn_ftrace *rec;
1566
1567	rec = lookup_rec(start, end);
1568	if (rec)
1569		return rec->ip;
1570
1571	return 0;
1572}
1573
1574/**
1575 * ftrace_location - return true if the ip giving is a traced location
1576 * @ip: the instruction pointer to check
1577 *
1578 * Returns rec->ip if @ip given is a pointer to a ftrace location.
1579 * That is, the instruction that is either a NOP or call to
1580 * the function tracer. It checks the ftrace internal tables to
1581 * determine if the address belongs or not.
1582 */
1583unsigned long ftrace_location(unsigned long ip)
1584{
1585	return ftrace_location_range(ip, ip);
1586}
1587
1588/**
1589 * ftrace_text_reserved - return true if range contains an ftrace location
1590 * @start: start of range to search
1591 * @end: end of range to search (inclusive). @end points to the last byte to check.
1592 *
1593 * Returns 1 if @start and @end contains a ftrace location.
1594 * That is, the instruction that is either a NOP or call to
1595 * the function tracer. It checks the ftrace internal tables to
1596 * determine if the address belongs or not.
1597 */
1598int ftrace_text_reserved(const void *start, const void *end)
1599{
1600	unsigned long ret;
1601
1602	ret = ftrace_location_range((unsigned long)start,
1603				    (unsigned long)end);
1604
1605	return (int)!!ret;
1606}
1607
1608/* Test if ops registered to this rec needs regs */
1609static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1610{
1611	struct ftrace_ops *ops;
1612	bool keep_regs = false;
1613
1614	for (ops = ftrace_ops_list;
1615	     ops != &ftrace_list_end; ops = ops->next) {
1616		/* pass rec in as regs to have non-NULL val */
1617		if (ftrace_ops_test(ops, rec->ip, rec)) {
1618			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1619				keep_regs = true;
1620				break;
1621			}
1622		}
1623	}
1624
1625	return  keep_regs;
1626}
1627
1628static struct ftrace_ops *
1629ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
1630static struct ftrace_ops *
1631ftrace_find_tramp_ops_any_other(struct dyn_ftrace *rec, struct ftrace_ops *op_exclude);
1632static struct ftrace_ops *
1633ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
1634
1635static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1636				     int filter_hash,
1637				     bool inc)
1638{
1639	struct ftrace_hash *hash;
1640	struct ftrace_hash *other_hash;
1641	struct ftrace_page *pg;
1642	struct dyn_ftrace *rec;
1643	bool update = false;
1644	int count = 0;
1645	int all = false;
1646
1647	/* Only update if the ops has been registered */
1648	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1649		return false;
1650
1651	/*
1652	 * In the filter_hash case:
1653	 *   If the count is zero, we update all records.
1654	 *   Otherwise we just update the items in the hash.
1655	 *
1656	 * In the notrace_hash case:
1657	 *   We enable the update in the hash.
1658	 *   As disabling notrace means enabling the tracing,
1659	 *   and enabling notrace means disabling, the inc variable
1660	 *   gets inversed.
1661	 */
1662	if (filter_hash) {
1663		hash = ops->func_hash->filter_hash;
1664		other_hash = ops->func_hash->notrace_hash;
1665		if (ftrace_hash_empty(hash))
1666			all = true;
1667	} else {
1668		inc = !inc;
1669		hash = ops->func_hash->notrace_hash;
1670		other_hash = ops->func_hash->filter_hash;
1671		/*
1672		 * If the notrace hash has no items,
1673		 * then there's nothing to do.
1674		 */
1675		if (ftrace_hash_empty(hash))
1676			return false;
1677	}
1678
1679	do_for_each_ftrace_rec(pg, rec) {
1680		int in_other_hash = 0;
1681		int in_hash = 0;
1682		int match = 0;
1683
1684		if (rec->flags & FTRACE_FL_DISABLED)
1685			continue;
1686
1687		if (all) {
1688			/*
1689			 * Only the filter_hash affects all records.
1690			 * Update if the record is not in the notrace hash.
1691			 */
1692			if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1693				match = 1;
1694		} else {
1695			in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1696			in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1697
1698			/*
1699			 * If filter_hash is set, we want to match all functions
1700			 * that are in the hash but not in the other hash.
1701			 *
1702			 * If filter_hash is not set, then we are decrementing.
1703			 * That means we match anything that is in the hash
1704			 * and also in the other_hash. That is, we need to turn
1705			 * off functions in the other hash because they are disabled
1706			 * by this hash.
1707			 */
1708			if (filter_hash && in_hash && !in_other_hash)
1709				match = 1;
1710			else if (!filter_hash && in_hash &&
1711				 (in_other_hash || ftrace_hash_empty(other_hash)))
1712				match = 1;
1713		}
1714		if (!match)
1715			continue;
1716
1717		if (inc) {
1718			rec->flags++;
1719			if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1720				return false;
1721
1722			if (ops->flags & FTRACE_OPS_FL_DIRECT)
1723				rec->flags |= FTRACE_FL_DIRECT;
1724
1725			/*
1726			 * If there's only a single callback registered to a
1727			 * function, and the ops has a trampoline registered
1728			 * for it, then we can call it directly.
1729			 */
1730			if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1731				rec->flags |= FTRACE_FL_TRAMP;
1732			else
1733				/*
1734				 * If we are adding another function callback
1735				 * to this function, and the previous had a
1736				 * custom trampoline in use, then we need to go
1737				 * back to the default trampoline.
1738				 */
1739				rec->flags &= ~FTRACE_FL_TRAMP;
1740
1741			/*
1742			 * If any ops wants regs saved for this function
1743			 * then all ops will get saved regs.
1744			 */
1745			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1746				rec->flags |= FTRACE_FL_REGS;
1747		} else {
1748			if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1749				return false;
1750			rec->flags--;
1751
1752			/*
1753			 * Only the internal direct_ops should have the
1754			 * DIRECT flag set. Thus, if it is removing a
1755			 * function, then that function should no longer
1756			 * be direct.
1757			 */
1758			if (ops->flags & FTRACE_OPS_FL_DIRECT)
1759				rec->flags &= ~FTRACE_FL_DIRECT;
1760
1761			/*
1762			 * If the rec had REGS enabled and the ops that is
1763			 * being removed had REGS set, then see if there is
1764			 * still any ops for this record that wants regs.
1765			 * If not, we can stop recording them.
1766			 */
1767			if (ftrace_rec_count(rec) > 0 &&
1768			    rec->flags & FTRACE_FL_REGS &&
1769			    ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1770				if (!test_rec_ops_needs_regs(rec))
1771					rec->flags &= ~FTRACE_FL_REGS;
1772			}
1773
1774			/*
1775			 * The TRAMP needs to be set only if rec count
1776			 * is decremented to one, and the ops that is
1777			 * left has a trampoline. As TRAMP can only be
1778			 * enabled if there is only a single ops attached
1779			 * to it.
1780			 */
1781			if (ftrace_rec_count(rec) == 1 &&
1782			    ftrace_find_tramp_ops_any_other(rec, ops))
1783				rec->flags |= FTRACE_FL_TRAMP;
1784			else
1785				rec->flags &= ~FTRACE_FL_TRAMP;
1786
1787			/*
1788			 * flags will be cleared in ftrace_check_record()
1789			 * if rec count is zero.
1790			 */
1791		}
1792		count++;
1793
1794		/* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1795		update |= ftrace_test_record(rec, true) != FTRACE_UPDATE_IGNORE;
1796
1797		/* Shortcut, if we handled all records, we are done. */
1798		if (!all && count == hash->count)
1799			return update;
1800	} while_for_each_ftrace_rec();
1801
1802	return update;
1803}
1804
1805static bool ftrace_hash_rec_disable(struct ftrace_ops *ops,
1806				    int filter_hash)
1807{
1808	return __ftrace_hash_rec_update(ops, filter_hash, 0);
1809}
1810
1811static bool ftrace_hash_rec_enable(struct ftrace_ops *ops,
1812				   int filter_hash)
1813{
1814	return __ftrace_hash_rec_update(ops, filter_hash, 1);
1815}
1816
1817static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops,
1818					  int filter_hash, int inc)
1819{
1820	struct ftrace_ops *op;
1821
1822	__ftrace_hash_rec_update(ops, filter_hash, inc);
1823
1824	if (ops->func_hash != &global_ops.local_hash)
1825		return;
1826
1827	/*
1828	 * If the ops shares the global_ops hash, then we need to update
1829	 * all ops that are enabled and use this hash.
1830	 */
1831	do_for_each_ftrace_op(op, ftrace_ops_list) {
1832		/* Already done */
1833		if (op == ops)
1834			continue;
1835		if (op->func_hash == &global_ops.local_hash)
1836			__ftrace_hash_rec_update(op, filter_hash, inc);
1837	} while_for_each_ftrace_op(op);
1838}
1839
1840static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops,
1841					   int filter_hash)
1842{
1843	ftrace_hash_rec_update_modify(ops, filter_hash, 0);
1844}
1845
1846static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops,
1847					  int filter_hash)
1848{
1849	ftrace_hash_rec_update_modify(ops, filter_hash, 1);
1850}
1851
1852/*
1853 * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1854 * or no-needed to update, -EBUSY if it detects a conflict of the flag
1855 * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1856 * Note that old_hash and new_hash has below meanings
1857 *  - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1858 *  - If the hash is EMPTY_HASH, it hits nothing
1859 *  - Anything else hits the recs which match the hash entries.
1860 */
1861static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1862					 struct ftrace_hash *old_hash,
1863					 struct ftrace_hash *new_hash)
1864{
1865	struct ftrace_page *pg;
1866	struct dyn_ftrace *rec, *end = NULL;
1867	int in_old, in_new;
1868
1869	/* Only update if the ops has been registered */
1870	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1871		return 0;
1872
1873	if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
1874		return 0;
1875
1876	/*
1877	 * Since the IPMODIFY is a very address sensitive action, we do not
1878	 * allow ftrace_ops to set all functions to new hash.
1879	 */
1880	if (!new_hash || !old_hash)
1881		return -EINVAL;
1882
1883	/* Update rec->flags */
1884	do_for_each_ftrace_rec(pg, rec) {
1885
1886		if (rec->flags & FTRACE_FL_DISABLED)
1887			continue;
1888
1889		/* We need to update only differences of filter_hash */
1890		in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1891		in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1892		if (in_old == in_new)
1893			continue;
1894
1895		if (in_new) {
1896			/* New entries must ensure no others are using it */
1897			if (rec->flags & FTRACE_FL_IPMODIFY)
1898				goto rollback;
1899			rec->flags |= FTRACE_FL_IPMODIFY;
1900		} else /* Removed entry */
1901			rec->flags &= ~FTRACE_FL_IPMODIFY;
1902	} while_for_each_ftrace_rec();
1903
1904	return 0;
1905
1906rollback:
1907	end = rec;
1908
1909	/* Roll back what we did above */
1910	do_for_each_ftrace_rec(pg, rec) {
1911
1912		if (rec->flags & FTRACE_FL_DISABLED)
1913			continue;
1914
1915		if (rec == end)
1916			goto err_out;
1917
1918		in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1919		in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1920		if (in_old == in_new)
1921			continue;
1922
1923		if (in_new)
1924			rec->flags &= ~FTRACE_FL_IPMODIFY;
1925		else
1926			rec->flags |= FTRACE_FL_IPMODIFY;
1927	} while_for_each_ftrace_rec();
1928
1929err_out:
1930	return -EBUSY;
1931}
1932
1933static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
1934{
1935	struct ftrace_hash *hash = ops->func_hash->filter_hash;
1936
1937	if (ftrace_hash_empty(hash))
1938		hash = NULL;
1939
1940	return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
1941}
1942
1943/* Disabling always succeeds */
1944static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
1945{
1946	struct ftrace_hash *hash = ops->func_hash->filter_hash;
1947
1948	if (ftrace_hash_empty(hash))
1949		hash = NULL;
1950
1951	__ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
1952}
1953
1954static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1955				       struct ftrace_hash *new_hash)
1956{
1957	struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
1958
1959	if (ftrace_hash_empty(old_hash))
1960		old_hash = NULL;
1961
1962	if (ftrace_hash_empty(new_hash))
1963		new_hash = NULL;
1964
1965	return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
1966}
1967
1968static void print_ip_ins(const char *fmt, const unsigned char *p)
1969{
1970	char ins[MCOUNT_INSN_SIZE];
1971	int i;
1972
1973	if (copy_from_kernel_nofault(ins, p, MCOUNT_INSN_SIZE)) {
1974		printk(KERN_CONT "%s[FAULT] %px\n", fmt, p);
1975		return;
1976	}
1977
1978	printk(KERN_CONT "%s", fmt);
1979
1980	for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1981		printk(KERN_CONT "%s%02x", i ? ":" : "", ins[i]);
1982}
1983
1984enum ftrace_bug_type ftrace_bug_type;
1985const void *ftrace_expected;
1986
1987static void print_bug_type(void)
1988{
1989	switch (ftrace_bug_type) {
1990	case FTRACE_BUG_UNKNOWN:
1991		break;
1992	case FTRACE_BUG_INIT:
1993		pr_info("Initializing ftrace call sites\n");
1994		break;
1995	case FTRACE_BUG_NOP:
1996		pr_info("Setting ftrace call site to NOP\n");
1997		break;
1998	case FTRACE_BUG_CALL:
1999		pr_info("Setting ftrace call site to call ftrace function\n");
2000		break;
2001	case FTRACE_BUG_UPDATE:
2002		pr_info("Updating ftrace call site to call a different ftrace function\n");
2003		break;
2004	}
2005}
2006
2007/**
2008 * ftrace_bug - report and shutdown function tracer
2009 * @failed: The failed type (EFAULT, EINVAL, EPERM)
2010 * @rec: The record that failed
2011 *
2012 * The arch code that enables or disables the function tracing
2013 * can call ftrace_bug() when it has detected a problem in
2014 * modifying the code. @failed should be one of either:
2015 * EFAULT - if the problem happens on reading the @ip address
2016 * EINVAL - if what is read at @ip is not what was expected
2017 * EPERM - if the problem happens on writing to the @ip address
2018 */
2019void ftrace_bug(int failed, struct dyn_ftrace *rec)
2020{
2021	unsigned long ip = rec ? rec->ip : 0;
2022
2023	pr_info("------------[ ftrace bug ]------------\n");
2024
2025	switch (failed) {
2026	case -EFAULT:
2027		pr_info("ftrace faulted on modifying ");
2028		print_ip_sym(KERN_INFO, ip);
2029		break;
2030	case -EINVAL:
2031		pr_info("ftrace failed to modify ");
2032		print_ip_sym(KERN_INFO, ip);
2033		print_ip_ins(" actual:   ", (unsigned char *)ip);
2034		pr_cont("\n");
2035		if (ftrace_expected) {
2036			print_ip_ins(" expected: ", ftrace_expected);
2037			pr_cont("\n");
2038		}
2039		break;
2040	case -EPERM:
2041		pr_info("ftrace faulted on writing ");
2042		print_ip_sym(KERN_INFO, ip);
2043		break;
2044	default:
2045		pr_info("ftrace faulted on unknown error ");
2046		print_ip_sym(KERN_INFO, ip);
2047	}
2048	print_bug_type();
2049	if (rec) {
2050		struct ftrace_ops *ops = NULL;
2051
2052		pr_info("ftrace record flags: %lx\n", rec->flags);
2053		pr_cont(" (%ld)%s", ftrace_rec_count(rec),
2054			rec->flags & FTRACE_FL_REGS ? " R" : "  ");
2055		if (rec->flags & FTRACE_FL_TRAMP_EN) {
2056			ops = ftrace_find_tramp_ops_any(rec);
2057			if (ops) {
2058				do {
2059					pr_cont("\ttramp: %pS (%pS)",
2060						(void *)ops->trampoline,
2061						(void *)ops->func);
2062					ops = ftrace_find_tramp_ops_next(rec, ops);
2063				} while (ops);
2064			} else
2065				pr_cont("\ttramp: ERROR!");
2066
2067		}
2068		ip = ftrace_get_addr_curr(rec);
2069		pr_cont("\n expected tramp: %lx\n", ip);
2070	}
2071
2072	FTRACE_WARN_ON_ONCE(1);
2073}
2074
2075static int ftrace_check_record(struct dyn_ftrace *rec, bool enable, bool update)
2076{
2077	unsigned long flag = 0UL;
2078
2079	ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2080
2081	if (rec->flags & FTRACE_FL_DISABLED)
2082		return FTRACE_UPDATE_IGNORE;
2083
2084	/*
2085	 * If we are updating calls:
2086	 *
2087	 *   If the record has a ref count, then we need to enable it
2088	 *   because someone is using it.
2089	 *
2090	 *   Otherwise we make sure its disabled.
2091	 *
2092	 * If we are disabling calls, then disable all records that
2093	 * are enabled.
2094	 */
2095	if (enable && ftrace_rec_count(rec))
2096		flag = FTRACE_FL_ENABLED;
2097
2098	/*
2099	 * If enabling and the REGS flag does not match the REGS_EN, or
2100	 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2101	 * this record. Set flags to fail the compare against ENABLED.
2102	 * Same for direct calls.
2103	 */
2104	if (flag) {
2105		if (!(rec->flags & FTRACE_FL_REGS) !=
2106		    !(rec->flags & FTRACE_FL_REGS_EN))
2107			flag |= FTRACE_FL_REGS;
2108
2109		if (!(rec->flags & FTRACE_FL_TRAMP) !=
2110		    !(rec->flags & FTRACE_FL_TRAMP_EN))
2111			flag |= FTRACE_FL_TRAMP;
2112
2113		/*
2114		 * Direct calls are special, as count matters.
2115		 * We must test the record for direct, if the
2116		 * DIRECT and DIRECT_EN do not match, but only
2117		 * if the count is 1. That's because, if the
2118		 * count is something other than one, we do not
2119		 * want the direct enabled (it will be done via the
2120		 * direct helper). But if DIRECT_EN is set, and
2121		 * the count is not one, we need to clear it.
2122		 */
2123		if (ftrace_rec_count(rec) == 1) {
2124			if (!(rec->flags & FTRACE_FL_DIRECT) !=
2125			    !(rec->flags & FTRACE_FL_DIRECT_EN))
2126				flag |= FTRACE_FL_DIRECT;
2127		} else if (rec->flags & FTRACE_FL_DIRECT_EN) {
2128			flag |= FTRACE_FL_DIRECT;
2129		}
2130	}
2131
2132	/* If the state of this record hasn't changed, then do nothing */
2133	if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2134		return FTRACE_UPDATE_IGNORE;
2135
2136	if (flag) {
2137		/* Save off if rec is being enabled (for return value) */
2138		flag ^= rec->flags & FTRACE_FL_ENABLED;
2139
2140		if (update) {
2141			rec->flags |= FTRACE_FL_ENABLED;
2142			if (flag & FTRACE_FL_REGS) {
2143				if (rec->flags & FTRACE_FL_REGS)
2144					rec->flags |= FTRACE_FL_REGS_EN;
2145				else
2146					rec->flags &= ~FTRACE_FL_REGS_EN;
2147			}
2148			if (flag & FTRACE_FL_TRAMP) {
2149				if (rec->flags & FTRACE_FL_TRAMP)
2150					rec->flags |= FTRACE_FL_TRAMP_EN;
2151				else
2152					rec->flags &= ~FTRACE_FL_TRAMP_EN;
2153			}
2154
2155			if (flag & FTRACE_FL_DIRECT) {
2156				/*
2157				 * If there's only one user (direct_ops helper)
2158				 * then we can call the direct function
2159				 * directly (no ftrace trampoline).
2160				 */
2161				if (ftrace_rec_count(rec) == 1) {
2162					if (rec->flags & FTRACE_FL_DIRECT)
2163						rec->flags |= FTRACE_FL_DIRECT_EN;
2164					else
2165						rec->flags &= ~FTRACE_FL_DIRECT_EN;
2166				} else {
2167					/*
2168					 * Can only call directly if there's
2169					 * only one callback to the function.
2170					 */
2171					rec->flags &= ~FTRACE_FL_DIRECT_EN;
2172				}
2173			}
2174		}
2175
2176		/*
2177		 * If this record is being updated from a nop, then
2178		 *   return UPDATE_MAKE_CALL.
2179		 * Otherwise,
2180		 *   return UPDATE_MODIFY_CALL to tell the caller to convert
2181		 *   from the save regs, to a non-save regs function or
2182		 *   vice versa, or from a trampoline call.
2183		 */
2184		if (flag & FTRACE_FL_ENABLED) {
2185			ftrace_bug_type = FTRACE_BUG_CALL;
2186			return FTRACE_UPDATE_MAKE_CALL;
2187		}
2188
2189		ftrace_bug_type = FTRACE_BUG_UPDATE;
2190		return FTRACE_UPDATE_MODIFY_CALL;
2191	}
2192
2193	if (update) {
2194		/* If there's no more users, clear all flags */
2195		if (!ftrace_rec_count(rec))
2196			rec->flags = 0;
2197		else
2198			/*
2199			 * Just disable the record, but keep the ops TRAMP
2200			 * and REGS states. The _EN flags must be disabled though.
2201			 */
2202			rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2203					FTRACE_FL_REGS_EN | FTRACE_FL_DIRECT_EN);
2204	}
2205
2206	ftrace_bug_type = FTRACE_BUG_NOP;
2207	return FTRACE_UPDATE_MAKE_NOP;
2208}
2209
2210/**
2211 * ftrace_update_record, set a record that now is tracing or not
2212 * @rec: the record to update
2213 * @enable: set to true if the record is tracing, false to force disable
2214 *
2215 * The records that represent all functions that can be traced need
2216 * to be updated when tracing has been enabled.
2217 */
2218int ftrace_update_record(struct dyn_ftrace *rec, bool enable)
2219{
2220	return ftrace_check_record(rec, enable, true);
2221}
2222
2223/**
2224 * ftrace_test_record, check if the record has been enabled or not
2225 * @rec: the record to test
2226 * @enable: set to true to check if enabled, false if it is disabled
2227 *
2228 * The arch code may need to test if a record is already set to
2229 * tracing to determine how to modify the function code that it
2230 * represents.
2231 */
2232int ftrace_test_record(struct dyn_ftrace *rec, bool enable)
2233{
2234	return ftrace_check_record(rec, enable, false);
2235}
2236
2237static struct ftrace_ops *
2238ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2239{
2240	struct ftrace_ops *op;
2241	unsigned long ip = rec->ip;
2242
2243	do_for_each_ftrace_op(op, ftrace_ops_list) {
2244
2245		if (!op->trampoline)
2246			continue;
2247
2248		if (hash_contains_ip(ip, op->func_hash))
2249			return op;
2250	} while_for_each_ftrace_op(op);
2251
2252	return NULL;
2253}
2254
2255static struct ftrace_ops *
2256ftrace_find_tramp_ops_any_other(struct dyn_ftrace *rec, struct ftrace_ops *op_exclude)
2257{
2258	struct ftrace_ops *op;
2259	unsigned long ip = rec->ip;
2260
2261	do_for_each_ftrace_op(op, ftrace_ops_list) {
2262
2263		if (op == op_exclude || !op->trampoline)
2264			continue;
2265
2266		if (hash_contains_ip(ip, op->func_hash))
2267			return op;
2268	} while_for_each_ftrace_op(op);
2269
2270	return NULL;
2271}
2272
2273static struct ftrace_ops *
2274ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2275			   struct ftrace_ops *op)
2276{
2277	unsigned long ip = rec->ip;
2278
2279	while_for_each_ftrace_op(op) {
2280
2281		if (!op->trampoline)
2282			continue;
2283
2284		if (hash_contains_ip(ip, op->func_hash))
2285			return op;
2286	}
2287
2288	return NULL;
2289}
2290
2291static struct ftrace_ops *
2292ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2293{
2294	struct ftrace_ops *op;
2295	unsigned long ip = rec->ip;
2296
2297	/*
2298	 * Need to check removed ops first.
2299	 * If they are being removed, and this rec has a tramp,
2300	 * and this rec is in the ops list, then it would be the
2301	 * one with the tramp.
2302	 */
2303	if (removed_ops) {
2304		if (hash_contains_ip(ip, &removed_ops->old_hash))
2305			return removed_ops;
2306	}
2307
2308	/*
2309	 * Need to find the current trampoline for a rec.
2310	 * Now, a trampoline is only attached to a rec if there
2311	 * was a single 'ops' attached to it. But this can be called
2312	 * when we are adding another op to the rec or removing the
2313	 * current one. Thus, if the op is being added, we can
2314	 * ignore it because it hasn't attached itself to the rec
2315	 * yet.
2316	 *
2317	 * If an ops is being modified (hooking to different functions)
2318	 * then we don't care about the new functions that are being
2319	 * added, just the old ones (that are probably being removed).
2320	 *
2321	 * If we are adding an ops to a function that already is using
2322	 * a trampoline, it needs to be removed (trampolines are only
2323	 * for single ops connected), then an ops that is not being
2324	 * modified also needs to be checked.
2325	 */
2326	do_for_each_ftrace_op(op, ftrace_ops_list) {
2327
2328		if (!op->trampoline)
2329			continue;
2330
2331		/*
2332		 * If the ops is being added, it hasn't gotten to
2333		 * the point to be removed from this tree yet.
2334		 */
2335		if (op->flags & FTRACE_OPS_FL_ADDING)
2336			continue;
2337
2338
2339		/*
2340		 * If the ops is being modified and is in the old
2341		 * hash, then it is probably being removed from this
2342		 * function.
2343		 */
2344		if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2345		    hash_contains_ip(ip, &op->old_hash))
2346			return op;
2347		/*
2348		 * If the ops is not being added or modified, and it's
2349		 * in its normal filter hash, then this must be the one
2350		 * we want!
2351		 */
2352		if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2353		    hash_contains_ip(ip, op->func_hash))
2354			return op;
2355
2356	} while_for_each_ftrace_op(op);
2357
2358	return NULL;
2359}
2360
2361static struct ftrace_ops *
2362ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2363{
2364	struct ftrace_ops *op;
2365	unsigned long ip = rec->ip;
2366
2367	do_for_each_ftrace_op(op, ftrace_ops_list) {
2368		/* pass rec in as regs to have non-NULL val */
2369		if (hash_contains_ip(ip, op->func_hash))
2370			return op;
2371	} while_for_each_ftrace_op(op);
2372
2373	return NULL;
2374}
2375
2376#ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
2377/* Protected by rcu_tasks for reading, and direct_mutex for writing */
2378static struct ftrace_hash *direct_functions = EMPTY_HASH;
2379static DEFINE_MUTEX(direct_mutex);
2380int ftrace_direct_func_count;
2381
2382/*
2383 * Search the direct_functions hash to see if the given instruction pointer
2384 * has a direct caller attached to it.
2385 */
2386unsigned long ftrace_find_rec_direct(unsigned long ip)
2387{
2388	struct ftrace_func_entry *entry;
2389
2390	entry = __ftrace_lookup_ip(direct_functions, ip);
2391	if (!entry)
2392		return 0;
2393
2394	return entry->direct;
2395}
2396
2397static void call_direct_funcs(unsigned long ip, unsigned long pip,
2398			      struct ftrace_ops *ops, struct ftrace_regs *fregs)
2399{
2400	struct pt_regs *regs = ftrace_get_regs(fregs);
2401	unsigned long addr;
2402
2403	addr = ftrace_find_rec_direct(ip);
2404	if (!addr)
2405		return;
2406
2407	arch_ftrace_set_direct_caller(regs, addr);
2408}
2409
2410struct ftrace_ops direct_ops = {
2411	.func		= call_direct_funcs,
2412	.flags		= FTRACE_OPS_FL_IPMODIFY
2413			  | FTRACE_OPS_FL_DIRECT | FTRACE_OPS_FL_SAVE_REGS
2414			  | FTRACE_OPS_FL_PERMANENT,
2415	/*
2416	 * By declaring the main trampoline as this trampoline
2417	 * it will never have one allocated for it. Allocated
2418	 * trampolines should not call direct functions.
2419	 * The direct_ops should only be called by the builtin
2420	 * ftrace_regs_caller trampoline.
2421	 */
2422	.trampoline	= FTRACE_REGS_ADDR,
2423};
2424#endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
2425
2426/**
2427 * ftrace_get_addr_new - Get the call address to set to
2428 * @rec:  The ftrace record descriptor
2429 *
2430 * If the record has the FTRACE_FL_REGS set, that means that it
2431 * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2432 * is not set, then it wants to convert to the normal callback.
2433 *
2434 * Returns the address of the trampoline to set to
2435 */
2436unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2437{
2438	struct ftrace_ops *ops;
2439	unsigned long addr;
2440
2441	if ((rec->flags & FTRACE_FL_DIRECT) &&
2442	    (ftrace_rec_count(rec) == 1)) {
2443		addr = ftrace_find_rec_direct(rec->ip);
2444		if (addr)
2445			return addr;
2446		WARN_ON_ONCE(1);
2447	}
2448
2449	/* Trampolines take precedence over regs */
2450	if (rec->flags & FTRACE_FL_TRAMP) {
2451		ops = ftrace_find_tramp_ops_new(rec);
2452		if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2453			pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2454				(void *)rec->ip, (void *)rec->ip, rec->flags);
2455			/* Ftrace is shutting down, return anything */
2456			return (unsigned long)FTRACE_ADDR;
2457		}
2458		return ops->trampoline;
2459	}
2460
2461	if (rec->flags & FTRACE_FL_REGS)
2462		return (unsigned long)FTRACE_REGS_ADDR;
2463	else
2464		return (unsigned long)FTRACE_ADDR;
2465}
2466
2467/**
2468 * ftrace_get_addr_curr - Get the call address that is already there
2469 * @rec:  The ftrace record descriptor
2470 *
2471 * The FTRACE_FL_REGS_EN is set when the record already points to
2472 * a function that saves all the regs. Basically the '_EN' version
2473 * represents the current state of the function.
2474 *
2475 * Returns the address of the trampoline that is currently being called
2476 */
2477unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2478{
2479	struct ftrace_ops *ops;
2480	unsigned long addr;
2481
2482	/* Direct calls take precedence over trampolines */
2483	if (rec->flags & FTRACE_FL_DIRECT_EN) {
2484		addr = ftrace_find_rec_direct(rec->ip);
2485		if (addr)
2486			return addr;
2487		WARN_ON_ONCE(1);
2488	}
2489
2490	/* Trampolines take precedence over regs */
2491	if (rec->flags & FTRACE_FL_TRAMP_EN) {
2492		ops = ftrace_find_tramp_ops_curr(rec);
2493		if (FTRACE_WARN_ON(!ops)) {
2494			pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2495				(void *)rec->ip, (void *)rec->ip);
2496			/* Ftrace is shutting down, return anything */
2497			return (unsigned long)FTRACE_ADDR;
2498		}
2499		return ops->trampoline;
2500	}
2501
2502	if (rec->flags & FTRACE_FL_REGS_EN)
2503		return (unsigned long)FTRACE_REGS_ADDR;
2504	else
2505		return (unsigned long)FTRACE_ADDR;
2506}
2507
2508static int
2509__ftrace_replace_code(struct dyn_ftrace *rec, bool enable)
2510{
2511	unsigned long ftrace_old_addr;
2512	unsigned long ftrace_addr;
2513	int ret;
2514
2515	ftrace_addr = ftrace_get_addr_new(rec);
2516
2517	/* This needs to be done before we call ftrace_update_record */
2518	ftrace_old_addr = ftrace_get_addr_curr(rec);
2519
2520	ret = ftrace_update_record(rec, enable);
2521
2522	ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2523
2524	switch (ret) {
2525	case FTRACE_UPDATE_IGNORE:
2526		return 0;
2527
2528	case FTRACE_UPDATE_MAKE_CALL:
2529		ftrace_bug_type = FTRACE_BUG_CALL;
2530		return ftrace_make_call(rec, ftrace_addr);
2531
2532	case FTRACE_UPDATE_MAKE_NOP:
2533		ftrace_bug_type = FTRACE_BUG_NOP;
2534		return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2535
2536	case FTRACE_UPDATE_MODIFY_CALL:
2537		ftrace_bug_type = FTRACE_BUG_UPDATE;
2538		return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2539	}
2540
2541	return -1; /* unknown ftrace bug */
2542}
2543
2544void __weak ftrace_replace_code(int mod_flags)
2545{
2546	struct dyn_ftrace *rec;
2547	struct ftrace_page *pg;
2548	bool enable = mod_flags & FTRACE_MODIFY_ENABLE_FL;
2549	int schedulable = mod_flags & FTRACE_MODIFY_MAY_SLEEP_FL;
2550	int failed;
2551
2552	if (unlikely(ftrace_disabled))
2553		return;
2554
2555	do_for_each_ftrace_rec(pg, rec) {
2556
2557		if (rec->flags & FTRACE_FL_DISABLED)
2558			continue;
2559
2560		failed = __ftrace_replace_code(rec, enable);
2561		if (failed) {
2562			ftrace_bug(failed, rec);
2563			/* Stop processing */
2564			return;
2565		}
2566		if (schedulable)
2567			cond_resched();
2568	} while_for_each_ftrace_rec();
2569}
2570
2571struct ftrace_rec_iter {
2572	struct ftrace_page	*pg;
2573	int			index;
2574};
2575
2576/**
2577 * ftrace_rec_iter_start, start up iterating over traced functions
2578 *
2579 * Returns an iterator handle that is used to iterate over all
2580 * the records that represent address locations where functions
2581 * are traced.
2582 *
2583 * May return NULL if no records are available.
2584 */
2585struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2586{
2587	/*
2588	 * We only use a single iterator.
2589	 * Protected by the ftrace_lock mutex.
2590	 */
2591	static struct ftrace_rec_iter ftrace_rec_iter;
2592	struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2593
2594	iter->pg = ftrace_pages_start;
2595	iter->index = 0;
2596
2597	/* Could have empty pages */
2598	while (iter->pg && !iter->pg->index)
2599		iter->pg = iter->pg->next;
2600
2601	if (!iter->pg)
2602		return NULL;
2603
2604	return iter;
2605}
2606
2607/**
2608 * ftrace_rec_iter_next, get the next record to process.
2609 * @iter: The handle to the iterator.
2610 *
2611 * Returns the next iterator after the given iterator @iter.
2612 */
2613struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2614{
2615	iter->index++;
2616
2617	if (iter->index >= iter->pg->index) {
2618		iter->pg = iter->pg->next;
2619		iter->index = 0;
2620
2621		/* Could have empty pages */
2622		while (iter->pg && !iter->pg->index)
2623			iter->pg = iter->pg->next;
2624	}
2625
2626	if (!iter->pg)
2627		return NULL;
2628
2629	return iter;
2630}
2631
2632/**
2633 * ftrace_rec_iter_record, get the record at the iterator location
2634 * @iter: The current iterator location
2635 *
2636 * Returns the record that the current @iter is at.
2637 */
2638struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2639{
2640	return &iter->pg->records[iter->index];
2641}
2642
2643static int
2644ftrace_nop_initialize(struct module *mod, struct dyn_ftrace *rec)
2645{
2646	int ret;
2647
2648	if (unlikely(ftrace_disabled))
2649		return 0;
2650
2651	ret = ftrace_init_nop(mod, rec);
2652	if (ret) {
2653		ftrace_bug_type = FTRACE_BUG_INIT;
2654		ftrace_bug(ret, rec);
2655		return 0;
2656	}
2657	return 1;
2658}
2659
2660/*
2661 * archs can override this function if they must do something
2662 * before the modifying code is performed.
2663 */
2664int __weak ftrace_arch_code_modify_prepare(void)
2665{
2666	return 0;
2667}
2668
2669/*
2670 * archs can override this function if they must do something
2671 * after the modifying code is performed.
2672 */
2673int __weak ftrace_arch_code_modify_post_process(void)
2674{
2675	return 0;
2676}
2677
2678void ftrace_modify_all_code(int command)
2679{
2680	int update = command & FTRACE_UPDATE_TRACE_FUNC;
2681	int mod_flags = 0;
2682	int err = 0;
2683
2684	if (command & FTRACE_MAY_SLEEP)
2685		mod_flags = FTRACE_MODIFY_MAY_SLEEP_FL;
2686
2687	/*
2688	 * If the ftrace_caller calls a ftrace_ops func directly,
2689	 * we need to make sure that it only traces functions it
2690	 * expects to trace. When doing the switch of functions,
2691	 * we need to update to the ftrace_ops_list_func first
2692	 * before the transition between old and new calls are set,
2693	 * as the ftrace_ops_list_func will check the ops hashes
2694	 * to make sure the ops are having the right functions
2695	 * traced.
2696	 */
2697	if (update) {
2698		err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2699		if (FTRACE_WARN_ON(err))
2700			return;
2701	}
2702
2703	if (command & FTRACE_UPDATE_CALLS)
2704		ftrace_replace_code(mod_flags | FTRACE_MODIFY_ENABLE_FL);
2705	else if (command & FTRACE_DISABLE_CALLS)
2706		ftrace_replace_code(mod_flags);
2707
2708	if (update && ftrace_trace_function != ftrace_ops_list_func) {
2709		function_trace_op = set_function_trace_op;
2710		smp_wmb();
2711		/* If irqs are disabled, we are in stop machine */
2712		if (!irqs_disabled())
2713			smp_call_function(ftrace_sync_ipi, NULL, 1);
2714		err = ftrace_update_ftrace_func(ftrace_trace_function);
2715		if (FTRACE_WARN_ON(err))
2716			return;
2717	}
2718
2719	if (command & FTRACE_START_FUNC_RET)
2720		err = ftrace_enable_ftrace_graph_caller();
2721	else if (command & FTRACE_STOP_FUNC_RET)
2722		err = ftrace_disable_ftrace_graph_caller();
2723	FTRACE_WARN_ON(err);
2724}
2725
2726static int __ftrace_modify_code(void *data)
2727{
2728	int *command = data;
2729
2730	ftrace_modify_all_code(*command);
2731
2732	return 0;
2733}
2734
2735/**
2736 * ftrace_run_stop_machine, go back to the stop machine method
2737 * @command: The command to tell ftrace what to do
2738 *
2739 * If an arch needs to fall back to the stop machine method, the
2740 * it can call this function.
2741 */
2742void ftrace_run_stop_machine(int command)
2743{
2744	stop_machine(__ftrace_modify_code, &command, NULL);
2745}
2746
2747/**
2748 * arch_ftrace_update_code, modify the code to trace or not trace
2749 * @command: The command that needs to be done
2750 *
2751 * Archs can override this function if it does not need to
2752 * run stop_machine() to modify code.
2753 */
2754void __weak arch_ftrace_update_code(int command)
2755{
2756	ftrace_run_stop_machine(command);
2757}
2758
2759static void ftrace_run_update_code(int command)
2760{
2761	int ret;
2762
2763	ret = ftrace_arch_code_modify_prepare();
2764	FTRACE_WARN_ON(ret);
2765	if (ret)
2766		return;
2767
2768	/*
2769	 * By default we use stop_machine() to modify the code.
2770	 * But archs can do what ever they want as long as it
2771	 * is safe. The stop_machine() is the safest, but also
2772	 * produces the most overhead.
2773	 */
2774	arch_ftrace_update_code(command);
2775
2776	ret = ftrace_arch_code_modify_post_process();
2777	FTRACE_WARN_ON(ret);
2778}
2779
2780static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2781				   struct ftrace_ops_hash *old_hash)
2782{
2783	ops->flags |= FTRACE_OPS_FL_MODIFYING;
2784	ops->old_hash.filter_hash = old_hash->filter_hash;
2785	ops->old_hash.notrace_hash = old_hash->notrace_hash;
2786	ftrace_run_update_code(command);
2787	ops->old_hash.filter_hash = NULL;
2788	ops->old_hash.notrace_hash = NULL;
2789	ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2790}
2791
2792static ftrace_func_t saved_ftrace_func;
2793static int ftrace_start_up;
2794
2795void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2796{
2797}
2798
2799/* List of trace_ops that have allocated trampolines */
2800static LIST_HEAD(ftrace_ops_trampoline_list);
2801
2802static void ftrace_add_trampoline_to_kallsyms(struct ftrace_ops *ops)
2803{
2804	lockdep_assert_held(&ftrace_lock);
2805	list_add_rcu(&ops->list, &ftrace_ops_trampoline_list);
2806}
2807
2808static void ftrace_remove_trampoline_from_kallsyms(struct ftrace_ops *ops)
2809{
2810	lockdep_assert_held(&ftrace_lock);
2811	list_del_rcu(&ops->list);
2812	synchronize_rcu();
2813}
2814
2815/*
2816 * "__builtin__ftrace" is used as a module name in /proc/kallsyms for symbols
2817 * for pages allocated for ftrace purposes, even though "__builtin__ftrace" is
2818 * not a module.
2819 */
2820#define FTRACE_TRAMPOLINE_MOD "__builtin__ftrace"
2821#define FTRACE_TRAMPOLINE_SYM "ftrace_trampoline"
2822
2823static void ftrace_trampoline_free(struct ftrace_ops *ops)
2824{
2825	if (ops && (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP) &&
2826	    ops->trampoline) {
2827		/*
2828		 * Record the text poke event before the ksymbol unregister
2829		 * event.
2830		 */
2831		perf_event_text_poke((void *)ops->trampoline,
2832				     (void *)ops->trampoline,
2833				     ops->trampoline_size, NULL, 0);
2834		perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
2835				   ops->trampoline, ops->trampoline_size,
2836				   true, FTRACE_TRAMPOLINE_SYM);
2837		/* Remove from kallsyms after the perf events */
2838		ftrace_remove_trampoline_from_kallsyms(ops);
2839	}
2840
2841	arch_ftrace_trampoline_free(ops);
2842}
2843
2844static void ftrace_startup_enable(int command)
2845{
2846	if (saved_ftrace_func != ftrace_trace_function) {
2847		saved_ftrace_func = ftrace_trace_function;
2848		command |= FTRACE_UPDATE_TRACE_FUNC;
2849	}
2850
2851	if (!command || !ftrace_enabled)
2852		return;
2853
2854	ftrace_run_update_code(command);
2855}
2856
2857static void ftrace_startup_all(int command)
2858{
2859	update_all_ops = true;
2860	ftrace_startup_enable(command);
2861	update_all_ops = false;
2862}
2863
2864int ftrace_startup(struct ftrace_ops *ops, int command)
2865{
2866	int ret;
2867
2868	if (unlikely(ftrace_disabled))
2869		return -ENODEV;
2870
2871	ret = __register_ftrace_function(ops);
2872	if (ret)
2873		return ret;
2874
2875	ftrace_start_up++;
2876
2877	/*
2878	 * Note that ftrace probes uses this to start up
2879	 * and modify functions it will probe. But we still
2880	 * set the ADDING flag for modification, as probes
2881	 * do not have trampolines. If they add them in the
2882	 * future, then the probes will need to distinguish
2883	 * between adding and updating probes.
2884	 */
2885	ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
2886
2887	ret = ftrace_hash_ipmodify_enable(ops);
2888	if (ret < 0) {
2889		/* Rollback registration process */
2890		__unregister_ftrace_function(ops);
2891		ftrace_start_up--;
2892		ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2893		if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
2894			ftrace_trampoline_free(ops);
2895		return ret;
2896	}
2897
2898	if (ftrace_hash_rec_enable(ops, 1))
2899		command |= FTRACE_UPDATE_CALLS;
2900
2901	ftrace_startup_enable(command);
2902
2903	ops->flags &= ~FTRACE_OPS_FL_ADDING;
2904
2905	return 0;
2906}
2907
2908int ftrace_shutdown(struct ftrace_ops *ops, int command)
2909{
2910	int ret;
2911
2912	if (unlikely(ftrace_disabled))
2913		return -ENODEV;
2914
2915	ret = __unregister_ftrace_function(ops);
2916	if (ret)
2917		return ret;
2918
2919	ftrace_start_up--;
2920	/*
2921	 * Just warn in case of unbalance, no need to kill ftrace, it's not
2922	 * critical but the ftrace_call callers may be never nopped again after
2923	 * further ftrace uses.
2924	 */
2925	WARN_ON_ONCE(ftrace_start_up < 0);
2926
2927	/* Disabling ipmodify never fails */
2928	ftrace_hash_ipmodify_disable(ops);
2929
2930	if (ftrace_hash_rec_disable(ops, 1))
2931		command |= FTRACE_UPDATE_CALLS;
2932
2933	ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2934
2935	if (saved_ftrace_func != ftrace_trace_function) {
2936		saved_ftrace_func = ftrace_trace_function;
2937		command |= FTRACE_UPDATE_TRACE_FUNC;
2938	}
2939
2940	if (!command || !ftrace_enabled) {
2941		/*
2942		 * If these are dynamic or per_cpu ops, they still
2943		 * need their data freed. Since, function tracing is
2944		 * not currently active, we can just free them
2945		 * without synchronizing all CPUs.
2946		 */
2947		if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
2948			goto free_ops;
2949
2950		return 0;
2951	}
2952
2953	/*
2954	 * If the ops uses a trampoline, then it needs to be
2955	 * tested first on update.
2956	 */
2957	ops->flags |= FTRACE_OPS_FL_REMOVING;
2958	removed_ops = ops;
2959
2960	/* The trampoline logic checks the old hashes */
2961	ops->old_hash.filter_hash = ops->func_hash->filter_hash;
2962	ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
2963
2964	ftrace_run_update_code(command);
2965
2966	/*
2967	 * If there's no more ops registered with ftrace, run a
2968	 * sanity check to make sure all rec flags are cleared.
2969	 */
2970	if (rcu_dereference_protected(ftrace_ops_list,
2971			lockdep_is_held(&ftrace_lock)) == &ftrace_list_end) {
2972		struct ftrace_page *pg;
2973		struct dyn_ftrace *rec;
2974
2975		do_for_each_ftrace_rec(pg, rec) {
2976			if (FTRACE_WARN_ON_ONCE(rec->flags & ~FTRACE_FL_DISABLED))
2977				pr_warn("  %pS flags:%lx\n",
2978					(void *)rec->ip, rec->flags);
2979		} while_for_each_ftrace_rec();
2980	}
2981
2982	ops->old_hash.filter_hash = NULL;
2983	ops->old_hash.notrace_hash = NULL;
2984
2985	removed_ops = NULL;
2986	ops->flags &= ~FTRACE_OPS_FL_REMOVING;
2987
2988	/*
2989	 * Dynamic ops may be freed, we must make sure that all
2990	 * callers are done before leaving this function.
2991	 * The same goes for freeing the per_cpu data of the per_cpu
2992	 * ops.
2993	 */
2994	if (ops->flags & FTRACE_OPS_FL_DYNAMIC) {
2995		/*
2996		 * We need to do a hard force of sched synchronization.
2997		 * This is because we use preempt_disable() to do RCU, but
2998		 * the function tracers can be called where RCU is not watching
2999		 * (like before user_exit()). We can not rely on the RCU
3000		 * infrastructure to do the synchronization, thus we must do it
3001		 * ourselves.
3002		 */
3003		synchronize_rcu_tasks_rude();
3004
3005		/*
3006		 * When the kernel is preemptive, tasks can be preempted
3007		 * while on a ftrace trampoline. Just scheduling a task on
3008		 * a CPU is not good enough to flush them. Calling
3009		 * synchronize_rcu_tasks() will wait for those tasks to
3010		 * execute and either schedule voluntarily or enter user space.
3011		 */
3012		if (IS_ENABLED(CONFIG_PREEMPTION))
3013			synchronize_rcu_tasks();
3014
3015 free_ops:
3016		ftrace_trampoline_free(ops);
3017	}
3018
3019	return 0;
3020}
3021
3022static void ftrace_startup_sysctl(void)
3023{
3024	int command;
3025
3026	if (unlikely(ftrace_disabled))
3027		return;
3028
3029	/* Force update next time */
3030	saved_ftrace_func = NULL;
3031	/* ftrace_start_up is true if we want ftrace running */
3032	if (ftrace_start_up) {
3033		command = FTRACE_UPDATE_CALLS;
3034		if (ftrace_graph_active)
3035			command |= FTRACE_START_FUNC_RET;
3036		ftrace_startup_enable(command);
3037	}
3038}
3039
3040static void ftrace_shutdown_sysctl(void)
3041{
3042	int command;
3043
3044	if (unlikely(ftrace_disabled))
3045		return;
3046
3047	/* ftrace_start_up is true if ftrace is running */
3048	if (ftrace_start_up) {
3049		command = FTRACE_DISABLE_CALLS;
3050		if (ftrace_graph_active)
3051			command |= FTRACE_STOP_FUNC_RET;
3052		ftrace_run_update_code(command);
3053	}
3054}
3055
3056static u64		ftrace_update_time;
3057unsigned long		ftrace_update_tot_cnt;
3058unsigned long		ftrace_number_of_pages;
3059unsigned long		ftrace_number_of_groups;
3060
3061static inline int ops_traces_mod(struct ftrace_ops *ops)
3062{
3063	/*
3064	 * Filter_hash being empty will default to trace module.
3065	 * But notrace hash requires a test of individual module functions.
3066	 */
3067	return ftrace_hash_empty(ops->func_hash->filter_hash) &&
3068		ftrace_hash_empty(ops->func_hash->notrace_hash);
3069}
3070
3071/*
3072 * Check if the current ops references the record.
3073 *
3074 * If the ops traces all functions, then it was already accounted for.
3075 * If the ops does not trace the current record function, skip it.
3076 * If the ops ignores the function via notrace filter, skip it.
3077 */
3078static inline bool
3079ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3080{
3081	/* If ops isn't enabled, ignore it */
3082	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
3083		return false;
3084
3085	/* If ops traces all then it includes this function */
3086	if (ops_traces_mod(ops))
3087		return true;
3088
3089	/* The function must be in the filter */
3090	if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
3091	    !__ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))
3092		return false;
3093
3094	/* If in notrace hash, we ignore it too */
3095	if (ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
3096		return false;
3097
3098	return true;
3099}
3100
3101static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
3102{
3103	struct ftrace_page *pg;
3104	struct dyn_ftrace *p;
3105	u64 start, stop;
3106	unsigned long update_cnt = 0;
3107	unsigned long rec_flags = 0;
3108	int i;
3109
3110	start = ftrace_now(raw_smp_processor_id());
3111
3112	/*
3113	 * When a module is loaded, this function is called to convert
3114	 * the calls to mcount in its text to nops, and also to create
3115	 * an entry in the ftrace data. Now, if ftrace is activated
3116	 * after this call, but before the module sets its text to
3117	 * read-only, the modification of enabling ftrace can fail if
3118	 * the read-only is done while ftrace is converting the calls.
3119	 * To prevent this, the module's records are set as disabled
3120	 * and will be enabled after the call to set the module's text
3121	 * to read-only.
3122	 */
3123	if (mod)
3124		rec_flags |= FTRACE_FL_DISABLED;
3125
3126	for (pg = new_pgs; pg; pg = pg->next) {
3127
3128		for (i = 0; i < pg->index; i++) {
3129
3130			/* If something went wrong, bail without enabling anything */
3131			if (unlikely(ftrace_disabled))
3132				return -1;
3133
3134			p = &pg->records[i];
3135			p->flags = rec_flags;
3136
3137			/*
3138			 * Do the initial record conversion from mcount jump
3139			 * to the NOP instructions.
3140			 */
3141			if (!__is_defined(CC_USING_NOP_MCOUNT) &&
3142			    !ftrace_nop_initialize(mod, p))
3143				break;
3144
3145			update_cnt++;
3146		}
3147	}
3148
3149	stop = ftrace_now(raw_smp_processor_id());
3150	ftrace_update_time = stop - start;
3151	ftrace_update_tot_cnt += update_cnt;
3152
3153	return 0;
3154}
3155
3156static int ftrace_allocate_records(struct ftrace_page *pg, int count)
3157{
3158	int order;
3159	int pages;
3160	int cnt;
3161
3162	if (WARN_ON(!count))
3163		return -EINVAL;
3164
3165	/* We want to fill as much as possible, with no empty pages */
3166	pages = DIV_ROUND_UP(count, ENTRIES_PER_PAGE);
3167	order = fls(pages) - 1;
 
 
 
 
 
3168
3169 again:
3170	pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
3171
3172	if (!pg->records) {
3173		/* if we can't allocate this size, try something smaller */
3174		if (!order)
3175			return -ENOMEM;
3176		order >>= 1;
3177		goto again;
3178	}
3179
3180	ftrace_number_of_pages += 1 << order;
3181	ftrace_number_of_groups++;
3182
3183	cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
3184	pg->order = order;
3185
3186	if (cnt > count)
3187		cnt = count;
3188
3189	return cnt;
3190}
3191
3192static struct ftrace_page *
3193ftrace_allocate_pages(unsigned long num_to_init)
3194{
3195	struct ftrace_page *start_pg;
3196	struct ftrace_page *pg;
 
3197	int cnt;
3198
3199	if (!num_to_init)
3200		return NULL;
3201
3202	start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
3203	if (!pg)
3204		return NULL;
3205
3206	/*
3207	 * Try to allocate as much as possible in one continues
3208	 * location that fills in all of the space. We want to
3209	 * waste as little space as possible.
3210	 */
3211	for (;;) {
3212		cnt = ftrace_allocate_records(pg, num_to_init);
3213		if (cnt < 0)
3214			goto free_pages;
3215
3216		num_to_init -= cnt;
3217		if (!num_to_init)
3218			break;
3219
3220		pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3221		if (!pg->next)
3222			goto free_pages;
3223
3224		pg = pg->next;
3225	}
3226
3227	return start_pg;
3228
3229 free_pages:
3230	pg = start_pg;
3231	while (pg) {
3232		if (pg->records) {
3233			free_pages((unsigned long)pg->records, pg->order);
3234			ftrace_number_of_pages -= 1 << pg->order;
3235		}
3236		start_pg = pg->next;
3237		kfree(pg);
3238		pg = start_pg;
 
3239		ftrace_number_of_groups--;
3240	}
3241	pr_info("ftrace: FAILED to allocate memory for functions\n");
3242	return NULL;
3243}
3244
3245#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3246
3247struct ftrace_iterator {
3248	loff_t				pos;
3249	loff_t				func_pos;
3250	loff_t				mod_pos;
3251	struct ftrace_page		*pg;
3252	struct dyn_ftrace		*func;
3253	struct ftrace_func_probe	*probe;
3254	struct ftrace_func_entry	*probe_entry;
3255	struct trace_parser		parser;
3256	struct ftrace_hash		*hash;
3257	struct ftrace_ops		*ops;
3258	struct trace_array		*tr;
3259	struct list_head		*mod_list;
3260	int				pidx;
3261	int				idx;
3262	unsigned			flags;
3263};
3264
3265static void *
3266t_probe_next(struct seq_file *m, loff_t *pos)
3267{
3268	struct ftrace_iterator *iter = m->private;
3269	struct trace_array *tr = iter->ops->private;
3270	struct list_head *func_probes;
3271	struct ftrace_hash *hash;
3272	struct list_head *next;
3273	struct hlist_node *hnd = NULL;
3274	struct hlist_head *hhd;
3275	int size;
3276
3277	(*pos)++;
3278	iter->pos = *pos;
3279
3280	if (!tr)
3281		return NULL;
3282
3283	func_probes = &tr->func_probes;
3284	if (list_empty(func_probes))
3285		return NULL;
3286
3287	if (!iter->probe) {
3288		next = func_probes->next;
3289		iter->probe = list_entry(next, struct ftrace_func_probe, list);
3290	}
3291
3292	if (iter->probe_entry)
3293		hnd = &iter->probe_entry->hlist;
3294
3295	hash = iter->probe->ops.func_hash->filter_hash;
3296
3297	/*
3298	 * A probe being registered may temporarily have an empty hash
3299	 * and it's at the end of the func_probes list.
3300	 */
3301	if (!hash || hash == EMPTY_HASH)
3302		return NULL;
3303
3304	size = 1 << hash->size_bits;
3305
3306 retry:
3307	if (iter->pidx >= size) {
3308		if (iter->probe->list.next == func_probes)
3309			return NULL;
3310		next = iter->probe->list.next;
3311		iter->probe = list_entry(next, struct ftrace_func_probe, list);
3312		hash = iter->probe->ops.func_hash->filter_hash;
3313		size = 1 << hash->size_bits;
3314		iter->pidx = 0;
3315	}
3316
3317	hhd = &hash->buckets[iter->pidx];
3318
3319	if (hlist_empty(hhd)) {
3320		iter->pidx++;
3321		hnd = NULL;
3322		goto retry;
3323	}
3324
3325	if (!hnd)
3326		hnd = hhd->first;
3327	else {
3328		hnd = hnd->next;
3329		if (!hnd) {
3330			iter->pidx++;
3331			goto retry;
3332		}
3333	}
3334
3335	if (WARN_ON_ONCE(!hnd))
3336		return NULL;
3337
3338	iter->probe_entry = hlist_entry(hnd, struct ftrace_func_entry, hlist);
3339
3340	return iter;
3341}
3342
3343static void *t_probe_start(struct seq_file *m, loff_t *pos)
3344{
3345	struct ftrace_iterator *iter = m->private;
3346	void *p = NULL;
3347	loff_t l;
3348
3349	if (!(iter->flags & FTRACE_ITER_DO_PROBES))
3350		return NULL;
3351
3352	if (iter->mod_pos > *pos)
3353		return NULL;
3354
3355	iter->probe = NULL;
3356	iter->probe_entry = NULL;
3357	iter->pidx = 0;
3358	for (l = 0; l <= (*pos - iter->mod_pos); ) {
3359		p = t_probe_next(m, &l);
3360		if (!p)
3361			break;
3362	}
3363	if (!p)
3364		return NULL;
3365
3366	/* Only set this if we have an item */
3367	iter->flags |= FTRACE_ITER_PROBE;
3368
3369	return iter;
3370}
3371
3372static int
3373t_probe_show(struct seq_file *m, struct ftrace_iterator *iter)
3374{
3375	struct ftrace_func_entry *probe_entry;
3376	struct ftrace_probe_ops *probe_ops;
3377	struct ftrace_func_probe *probe;
3378
3379	probe = iter->probe;
3380	probe_entry = iter->probe_entry;
3381
3382	if (WARN_ON_ONCE(!probe || !probe_entry))
3383		return -EIO;
3384
3385	probe_ops = probe->probe_ops;
3386
3387	if (probe_ops->print)
3388		return probe_ops->print(m, probe_entry->ip, probe_ops, probe->data);
3389
3390	seq_printf(m, "%ps:%ps\n", (void *)probe_entry->ip,
3391		   (void *)probe_ops->func);
3392
3393	return 0;
3394}
3395
3396static void *
3397t_mod_next(struct seq_file *m, loff_t *pos)
3398{
3399	struct ftrace_iterator *iter = m->private;
3400	struct trace_array *tr = iter->tr;
3401
3402	(*pos)++;
3403	iter->pos = *pos;
3404
3405	iter->mod_list = iter->mod_list->next;
3406
3407	if (iter->mod_list == &tr->mod_trace ||
3408	    iter->mod_list == &tr->mod_notrace) {
3409		iter->flags &= ~FTRACE_ITER_MOD;
3410		return NULL;
3411	}
3412
3413	iter->mod_pos = *pos;
3414
3415	return iter;
3416}
3417
3418static void *t_mod_start(struct seq_file *m, loff_t *pos)
3419{
3420	struct ftrace_iterator *iter = m->private;
3421	void *p = NULL;
3422	loff_t l;
3423
3424	if (iter->func_pos > *pos)
3425		return NULL;
3426
3427	iter->mod_pos = iter->func_pos;
3428
3429	/* probes are only available if tr is set */
3430	if (!iter->tr)
3431		return NULL;
3432
3433	for (l = 0; l <= (*pos - iter->func_pos); ) {
3434		p = t_mod_next(m, &l);
3435		if (!p)
3436			break;
3437	}
3438	if (!p) {
3439		iter->flags &= ~FTRACE_ITER_MOD;
3440		return t_probe_start(m, pos);
3441	}
3442
3443	/* Only set this if we have an item */
3444	iter->flags |= FTRACE_ITER_MOD;
3445
3446	return iter;
3447}
3448
3449static int
3450t_mod_show(struct seq_file *m, struct ftrace_iterator *iter)
3451{
3452	struct ftrace_mod_load *ftrace_mod;
3453	struct trace_array *tr = iter->tr;
3454
3455	if (WARN_ON_ONCE(!iter->mod_list) ||
3456			 iter->mod_list == &tr->mod_trace ||
3457			 iter->mod_list == &tr->mod_notrace)
3458		return -EIO;
3459
3460	ftrace_mod = list_entry(iter->mod_list, struct ftrace_mod_load, list);
3461
3462	if (ftrace_mod->func)
3463		seq_printf(m, "%s", ftrace_mod->func);
3464	else
3465		seq_putc(m, '*');
3466
3467	seq_printf(m, ":mod:%s\n", ftrace_mod->module);
3468
3469	return 0;
3470}
3471
3472static void *
3473t_func_next(struct seq_file *m, loff_t *pos)
3474{
3475	struct ftrace_iterator *iter = m->private;
3476	struct dyn_ftrace *rec = NULL;
3477
3478	(*pos)++;
3479
3480 retry:
3481	if (iter->idx >= iter->pg->index) {
3482		if (iter->pg->next) {
3483			iter->pg = iter->pg->next;
3484			iter->idx = 0;
3485			goto retry;
3486		}
3487	} else {
3488		rec = &iter->pg->records[iter->idx++];
3489		if (((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3490		     !ftrace_lookup_ip(iter->hash, rec->ip)) ||
3491
3492		    ((iter->flags & FTRACE_ITER_ENABLED) &&
3493		     !(rec->flags & FTRACE_FL_ENABLED))) {
3494
3495			rec = NULL;
3496			goto retry;
3497		}
3498	}
3499
3500	if (!rec)
3501		return NULL;
3502
3503	iter->pos = iter->func_pos = *pos;
3504	iter->func = rec;
3505
3506	return iter;
3507}
3508
3509static void *
3510t_next(struct seq_file *m, void *v, loff_t *pos)
3511{
3512	struct ftrace_iterator *iter = m->private;
3513	loff_t l = *pos; /* t_probe_start() must use original pos */
3514	void *ret;
3515
3516	if (unlikely(ftrace_disabled))
3517		return NULL;
3518
3519	if (iter->flags & FTRACE_ITER_PROBE)
3520		return t_probe_next(m, pos);
3521
3522	if (iter->flags & FTRACE_ITER_MOD)
3523		return t_mod_next(m, pos);
3524
3525	if (iter->flags & FTRACE_ITER_PRINTALL) {
3526		/* next must increment pos, and t_probe_start does not */
3527		(*pos)++;
3528		return t_mod_start(m, &l);
3529	}
3530
3531	ret = t_func_next(m, pos);
3532
3533	if (!ret)
3534		return t_mod_start(m, &l);
3535
3536	return ret;
3537}
3538
3539static void reset_iter_read(struct ftrace_iterator *iter)
3540{
3541	iter->pos = 0;
3542	iter->func_pos = 0;
3543	iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_PROBE | FTRACE_ITER_MOD);
3544}
3545
3546static void *t_start(struct seq_file *m, loff_t *pos)
3547{
3548	struct ftrace_iterator *iter = m->private;
3549	void *p = NULL;
3550	loff_t l;
3551
3552	mutex_lock(&ftrace_lock);
3553
3554	if (unlikely(ftrace_disabled))
3555		return NULL;
3556
3557	/*
3558	 * If an lseek was done, then reset and start from beginning.
3559	 */
3560	if (*pos < iter->pos)
3561		reset_iter_read(iter);
3562
3563	/*
3564	 * For set_ftrace_filter reading, if we have the filter
3565	 * off, we can short cut and just print out that all
3566	 * functions are enabled.
3567	 */
3568	if ((iter->flags & (FTRACE_ITER_FILTER | FTRACE_ITER_NOTRACE)) &&
3569	    ftrace_hash_empty(iter->hash)) {
3570		iter->func_pos = 1; /* Account for the message */
3571		if (*pos > 0)
3572			return t_mod_start(m, pos);
3573		iter->flags |= FTRACE_ITER_PRINTALL;
3574		/* reset in case of seek/pread */
3575		iter->flags &= ~FTRACE_ITER_PROBE;
3576		return iter;
3577	}
3578
3579	if (iter->flags & FTRACE_ITER_MOD)
3580		return t_mod_start(m, pos);
3581
3582	/*
3583	 * Unfortunately, we need to restart at ftrace_pages_start
3584	 * every time we let go of the ftrace_mutex. This is because
3585	 * those pointers can change without the lock.
3586	 */
3587	iter->pg = ftrace_pages_start;
3588	iter->idx = 0;
3589	for (l = 0; l <= *pos; ) {
3590		p = t_func_next(m, &l);
3591		if (!p)
3592			break;
3593	}
3594
3595	if (!p)
3596		return t_mod_start(m, pos);
3597
3598	return iter;
3599}
3600
3601static void t_stop(struct seq_file *m, void *p)
3602{
3603	mutex_unlock(&ftrace_lock);
3604}
3605
3606void * __weak
3607arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3608{
3609	return NULL;
3610}
3611
3612static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
3613				struct dyn_ftrace *rec)
3614{
3615	void *ptr;
3616
3617	ptr = arch_ftrace_trampoline_func(ops, rec);
3618	if (ptr)
3619		seq_printf(m, " ->%pS", ptr);
3620}
3621
3622static int t_show(struct seq_file *m, void *v)
3623{
3624	struct ftrace_iterator *iter = m->private;
3625	struct dyn_ftrace *rec;
3626
3627	if (iter->flags & FTRACE_ITER_PROBE)
3628		return t_probe_show(m, iter);
3629
3630	if (iter->flags & FTRACE_ITER_MOD)
3631		return t_mod_show(m, iter);
3632
3633	if (iter->flags & FTRACE_ITER_PRINTALL) {
3634		if (iter->flags & FTRACE_ITER_NOTRACE)
3635			seq_puts(m, "#### no functions disabled ####\n");
3636		else
3637			seq_puts(m, "#### all functions enabled ####\n");
3638		return 0;
3639	}
3640
3641	rec = iter->func;
3642
3643	if (!rec)
3644		return 0;
3645
3646	seq_printf(m, "%ps", (void *)rec->ip);
3647	if (iter->flags & FTRACE_ITER_ENABLED) {
3648		struct ftrace_ops *ops;
3649
3650		seq_printf(m, " (%ld)%s%s%s",
3651			   ftrace_rec_count(rec),
3652			   rec->flags & FTRACE_FL_REGS ? " R" : "  ",
3653			   rec->flags & FTRACE_FL_IPMODIFY ? " I" : "  ",
3654			   rec->flags & FTRACE_FL_DIRECT ? " D" : "  ");
3655		if (rec->flags & FTRACE_FL_TRAMP_EN) {
3656			ops = ftrace_find_tramp_ops_any(rec);
3657			if (ops) {
3658				do {
3659					seq_printf(m, "\ttramp: %pS (%pS)",
3660						   (void *)ops->trampoline,
3661						   (void *)ops->func);
3662					add_trampoline_func(m, ops, rec);
3663					ops = ftrace_find_tramp_ops_next(rec, ops);
3664				} while (ops);
3665			} else
3666				seq_puts(m, "\ttramp: ERROR!");
3667		} else {
3668			add_trampoline_func(m, NULL, rec);
3669		}
3670		if (rec->flags & FTRACE_FL_DIRECT) {
3671			unsigned long direct;
3672
3673			direct = ftrace_find_rec_direct(rec->ip);
3674			if (direct)
3675				seq_printf(m, "\n\tdirect-->%pS", (void *)direct);
3676		}
3677	}
3678
3679	seq_putc(m, '\n');
3680
3681	return 0;
3682}
3683
3684static const struct seq_operations show_ftrace_seq_ops = {
3685	.start = t_start,
3686	.next = t_next,
3687	.stop = t_stop,
3688	.show = t_show,
3689};
3690
3691static int
3692ftrace_avail_open(struct inode *inode, struct file *file)
3693{
3694	struct ftrace_iterator *iter;
3695	int ret;
3696
3697	ret = security_locked_down(LOCKDOWN_TRACEFS);
3698	if (ret)
3699		return ret;
3700
3701	if (unlikely(ftrace_disabled))
3702		return -ENODEV;
3703
3704	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3705	if (!iter)
3706		return -ENOMEM;
3707
3708	iter->pg = ftrace_pages_start;
3709	iter->ops = &global_ops;
3710
3711	return 0;
3712}
3713
3714static int
3715ftrace_enabled_open(struct inode *inode, struct file *file)
3716{
3717	struct ftrace_iterator *iter;
3718
3719	/*
3720	 * This shows us what functions are currently being
3721	 * traced and by what. Not sure if we want lockdown
3722	 * to hide such critical information for an admin.
3723	 * Although, perhaps it can show information we don't
3724	 * want people to see, but if something is tracing
3725	 * something, we probably want to know about it.
3726	 */
3727
3728	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3729	if (!iter)
3730		return -ENOMEM;
3731
3732	iter->pg = ftrace_pages_start;
3733	iter->flags = FTRACE_ITER_ENABLED;
3734	iter->ops = &global_ops;
3735
3736	return 0;
3737}
3738
3739/**
3740 * ftrace_regex_open - initialize function tracer filter files
3741 * @ops: The ftrace_ops that hold the hash filters
3742 * @flag: The type of filter to process
3743 * @inode: The inode, usually passed in to your open routine
3744 * @file: The file, usually passed in to your open routine
3745 *
3746 * ftrace_regex_open() initializes the filter files for the
3747 * @ops. Depending on @flag it may process the filter hash or
3748 * the notrace hash of @ops. With this called from the open
3749 * routine, you can use ftrace_filter_write() for the write
3750 * routine if @flag has FTRACE_ITER_FILTER set, or
3751 * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
3752 * tracing_lseek() should be used as the lseek routine, and
3753 * release must call ftrace_regex_release().
3754 */
3755int
3756ftrace_regex_open(struct ftrace_ops *ops, int flag,
3757		  struct inode *inode, struct file *file)
3758{
3759	struct ftrace_iterator *iter;
3760	struct ftrace_hash *hash;
3761	struct list_head *mod_head;
3762	struct trace_array *tr = ops->private;
3763	int ret = -ENOMEM;
3764
3765	ftrace_ops_init(ops);
3766
3767	if (unlikely(ftrace_disabled))
3768		return -ENODEV;
3769
3770	if (tracing_check_open_get_tr(tr))
3771		return -ENODEV;
3772
3773	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3774	if (!iter)
3775		goto out;
3776
3777	if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX))
3778		goto out;
3779
3780	iter->ops = ops;
3781	iter->flags = flag;
3782	iter->tr = tr;
3783
3784	mutex_lock(&ops->func_hash->regex_lock);
3785
3786	if (flag & FTRACE_ITER_NOTRACE) {
3787		hash = ops->func_hash->notrace_hash;
3788		mod_head = tr ? &tr->mod_notrace : NULL;
3789	} else {
3790		hash = ops->func_hash->filter_hash;
3791		mod_head = tr ? &tr->mod_trace : NULL;
3792	}
3793
3794	iter->mod_list = mod_head;
3795
3796	if (file->f_mode & FMODE_WRITE) {
3797		const int size_bits = FTRACE_HASH_DEFAULT_BITS;
3798
3799		if (file->f_flags & O_TRUNC) {
3800			iter->hash = alloc_ftrace_hash(size_bits);
3801			clear_ftrace_mod_list(mod_head);
3802	        } else {
3803			iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
3804		}
3805
3806		if (!iter->hash) {
3807			trace_parser_put(&iter->parser);
3808			goto out_unlock;
3809		}
3810	} else
3811		iter->hash = hash;
3812
3813	ret = 0;
3814
3815	if (file->f_mode & FMODE_READ) {
3816		iter->pg = ftrace_pages_start;
3817
3818		ret = seq_open(file, &show_ftrace_seq_ops);
3819		if (!ret) {
3820			struct seq_file *m = file->private_data;
3821			m->private = iter;
3822		} else {
3823			/* Failed */
3824			free_ftrace_hash(iter->hash);
3825			trace_parser_put(&iter->parser);
3826		}
3827	} else
3828		file->private_data = iter;
3829
3830 out_unlock:
3831	mutex_unlock(&ops->func_hash->regex_lock);
3832
3833 out:
3834	if (ret) {
3835		kfree(iter);
3836		if (tr)
3837			trace_array_put(tr);
3838	}
3839
3840	return ret;
3841}
3842
3843static int
3844ftrace_filter_open(struct inode *inode, struct file *file)
3845{
3846	struct ftrace_ops *ops = inode->i_private;
3847
3848	/* Checks for tracefs lockdown */
3849	return ftrace_regex_open(ops,
3850			FTRACE_ITER_FILTER | FTRACE_ITER_DO_PROBES,
3851			inode, file);
3852}
3853
3854static int
3855ftrace_notrace_open(struct inode *inode, struct file *file)
3856{
3857	struct ftrace_ops *ops = inode->i_private;
3858
3859	/* Checks for tracefs lockdown */
3860	return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
3861				 inode, file);
3862}
3863
3864/* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
3865struct ftrace_glob {
3866	char *search;
3867	unsigned len;
3868	int type;
3869};
3870
3871/*
3872 * If symbols in an architecture don't correspond exactly to the user-visible
3873 * name of what they represent, it is possible to define this function to
3874 * perform the necessary adjustments.
3875*/
3876char * __weak arch_ftrace_match_adjust(char *str, const char *search)
3877{
3878	return str;
3879}
3880
3881static int ftrace_match(char *str, struct ftrace_glob *g)
3882{
3883	int matched = 0;
3884	int slen;
3885
3886	str = arch_ftrace_match_adjust(str, g->search);
3887
3888	switch (g->type) {
3889	case MATCH_FULL:
3890		if (strcmp(str, g->search) == 0)
3891			matched = 1;
3892		break;
3893	case MATCH_FRONT_ONLY:
3894		if (strncmp(str, g->search, g->len) == 0)
3895			matched = 1;
3896		break;
3897	case MATCH_MIDDLE_ONLY:
3898		if (strstr(str, g->search))
3899			matched = 1;
3900		break;
3901	case MATCH_END_ONLY:
3902		slen = strlen(str);
3903		if (slen >= g->len &&
3904		    memcmp(str + slen - g->len, g->search, g->len) == 0)
3905			matched = 1;
3906		break;
3907	case MATCH_GLOB:
3908		if (glob_match(g->search, str))
3909			matched = 1;
3910		break;
3911	}
3912
3913	return matched;
3914}
3915
3916static int
3917enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
3918{
3919	struct ftrace_func_entry *entry;
3920	int ret = 0;
3921
3922	entry = ftrace_lookup_ip(hash, rec->ip);
3923	if (clear_filter) {
3924		/* Do nothing if it doesn't exist */
3925		if (!entry)
3926			return 0;
3927
3928		free_hash_entry(hash, entry);
3929	} else {
3930		/* Do nothing if it exists */
3931		if (entry)
3932			return 0;
3933
3934		ret = add_hash_entry(hash, rec->ip);
3935	}
3936	return ret;
3937}
3938
3939static int
3940add_rec_by_index(struct ftrace_hash *hash, struct ftrace_glob *func_g,
3941		 int clear_filter)
3942{
3943	long index = simple_strtoul(func_g->search, NULL, 0);
3944	struct ftrace_page *pg;
3945	struct dyn_ftrace *rec;
3946
3947	/* The index starts at 1 */
3948	if (--index < 0)
3949		return 0;
3950
3951	do_for_each_ftrace_rec(pg, rec) {
3952		if (pg->index <= index) {
3953			index -= pg->index;
3954			/* this is a double loop, break goes to the next page */
3955			break;
3956		}
3957		rec = &pg->records[index];
3958		enter_record(hash, rec, clear_filter);
3959		return 1;
3960	} while_for_each_ftrace_rec();
3961	return 0;
3962}
3963
3964static int
3965ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
3966		struct ftrace_glob *mod_g, int exclude_mod)
3967{
3968	char str[KSYM_SYMBOL_LEN];
3969	char *modname;
3970
3971	kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
3972
3973	if (mod_g) {
3974		int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
3975
3976		/* blank module name to match all modules */
3977		if (!mod_g->len) {
3978			/* blank module globbing: modname xor exclude_mod */
3979			if (!exclude_mod != !modname)
3980				goto func_match;
3981			return 0;
3982		}
3983
3984		/*
3985		 * exclude_mod is set to trace everything but the given
3986		 * module. If it is set and the module matches, then
3987		 * return 0. If it is not set, and the module doesn't match
3988		 * also return 0. Otherwise, check the function to see if
3989		 * that matches.
3990		 */
3991		if (!mod_matches == !exclude_mod)
3992			return 0;
3993func_match:
3994		/* blank search means to match all funcs in the mod */
3995		if (!func_g->len)
3996			return 1;
3997	}
3998
3999	return ftrace_match(str, func_g);
4000}
4001
4002static int
4003match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
4004{
4005	struct ftrace_page *pg;
4006	struct dyn_ftrace *rec;
4007	struct ftrace_glob func_g = { .type = MATCH_FULL };
4008	struct ftrace_glob mod_g = { .type = MATCH_FULL };
4009	struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
4010	int exclude_mod = 0;
4011	int found = 0;
4012	int ret;
4013	int clear_filter = 0;
4014
4015	if (func) {
4016		func_g.type = filter_parse_regex(func, len, &func_g.search,
4017						 &clear_filter);
4018		func_g.len = strlen(func_g.search);
4019	}
4020
4021	if (mod) {
4022		mod_g.type = filter_parse_regex(mod, strlen(mod),
4023				&mod_g.search, &exclude_mod);
4024		mod_g.len = strlen(mod_g.search);
4025	}
4026
4027	mutex_lock(&ftrace_lock);
4028
4029	if (unlikely(ftrace_disabled))
4030		goto out_unlock;
4031
4032	if (func_g.type == MATCH_INDEX) {
4033		found = add_rec_by_index(hash, &func_g, clear_filter);
4034		goto out_unlock;
4035	}
4036
4037	do_for_each_ftrace_rec(pg, rec) {
4038
4039		if (rec->flags & FTRACE_FL_DISABLED)
4040			continue;
4041
4042		if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
4043			ret = enter_record(hash, rec, clear_filter);
4044			if (ret < 0) {
4045				found = ret;
4046				goto out_unlock;
4047			}
4048			found = 1;
4049		}
4050	} while_for_each_ftrace_rec();
4051 out_unlock:
4052	mutex_unlock(&ftrace_lock);
4053
4054	return found;
4055}
4056
4057static int
4058ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
4059{
4060	return match_records(hash, buff, len, NULL);
4061}
4062
4063static void ftrace_ops_update_code(struct ftrace_ops *ops,
4064				   struct ftrace_ops_hash *old_hash)
4065{
4066	struct ftrace_ops *op;
4067
4068	if (!ftrace_enabled)
4069		return;
4070
4071	if (ops->flags & FTRACE_OPS_FL_ENABLED) {
4072		ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
4073		return;
4074	}
4075
4076	/*
4077	 * If this is the shared global_ops filter, then we need to
4078	 * check if there is another ops that shares it, is enabled.
4079	 * If so, we still need to run the modify code.
4080	 */
4081	if (ops->func_hash != &global_ops.local_hash)
4082		return;
4083
4084	do_for_each_ftrace_op(op, ftrace_ops_list) {
4085		if (op->func_hash == &global_ops.local_hash &&
4086		    op->flags & FTRACE_OPS_FL_ENABLED) {
4087			ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
4088			/* Only need to do this once */
4089			return;
4090		}
4091	} while_for_each_ftrace_op(op);
4092}
4093
4094static int ftrace_hash_move_and_update_ops(struct ftrace_ops *ops,
4095					   struct ftrace_hash **orig_hash,
4096					   struct ftrace_hash *hash,
4097					   int enable)
4098{
4099	struct ftrace_ops_hash old_hash_ops;
4100	struct ftrace_hash *old_hash;
4101	int ret;
4102
4103	old_hash = *orig_hash;
4104	old_hash_ops.filter_hash = ops->func_hash->filter_hash;
4105	old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
4106	ret = ftrace_hash_move(ops, enable, orig_hash, hash);
4107	if (!ret) {
4108		ftrace_ops_update_code(ops, &old_hash_ops);
4109		free_ftrace_hash_rcu(old_hash);
4110	}
4111	return ret;
4112}
4113
4114static bool module_exists(const char *module)
4115{
4116	/* All modules have the symbol __this_module */
4117	static const char this_mod[] = "__this_module";
4118	char modname[MAX_PARAM_PREFIX_LEN + sizeof(this_mod) + 2];
4119	unsigned long val;
4120	int n;
4121
4122	n = snprintf(modname, sizeof(modname), "%s:%s", module, this_mod);
4123
4124	if (n > sizeof(modname) - 1)
4125		return false;
4126
4127	val = module_kallsyms_lookup_name(modname);
4128	return val != 0;
4129}
4130
4131static int cache_mod(struct trace_array *tr,
4132		     const char *func, char *module, int enable)
4133{
4134	struct ftrace_mod_load *ftrace_mod, *n;
4135	struct list_head *head = enable ? &tr->mod_trace : &tr->mod_notrace;
4136	int ret;
4137
4138	mutex_lock(&ftrace_lock);
4139
4140	/* We do not cache inverse filters */
4141	if (func[0] == '!') {
4142		func++;
4143		ret = -EINVAL;
4144
4145		/* Look to remove this hash */
4146		list_for_each_entry_safe(ftrace_mod, n, head, list) {
4147			if (strcmp(ftrace_mod->module, module) != 0)
4148				continue;
4149
4150			/* no func matches all */
4151			if (strcmp(func, "*") == 0 ||
4152			    (ftrace_mod->func &&
4153			     strcmp(ftrace_mod->func, func) == 0)) {
4154				ret = 0;
4155				free_ftrace_mod(ftrace_mod);
4156				continue;
4157			}
4158		}
4159		goto out;
4160	}
4161
4162	ret = -EINVAL;
4163	/* We only care about modules that have not been loaded yet */
4164	if (module_exists(module))
4165		goto out;
4166
4167	/* Save this string off, and execute it when the module is loaded */
4168	ret = ftrace_add_mod(tr, func, module, enable);
4169 out:
4170	mutex_unlock(&ftrace_lock);
4171
4172	return ret;
4173}
4174
4175static int
4176ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
4177		 int reset, int enable);
4178
4179#ifdef CONFIG_MODULES
4180static void process_mod_list(struct list_head *head, struct ftrace_ops *ops,
4181			     char *mod, bool enable)
4182{
4183	struct ftrace_mod_load *ftrace_mod, *n;
4184	struct ftrace_hash **orig_hash, *new_hash;
4185	LIST_HEAD(process_mods);
4186	char *func;
 
4187
4188	mutex_lock(&ops->func_hash->regex_lock);
4189
4190	if (enable)
4191		orig_hash = &ops->func_hash->filter_hash;
4192	else
4193		orig_hash = &ops->func_hash->notrace_hash;
4194
4195	new_hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS,
4196					      *orig_hash);
4197	if (!new_hash)
4198		goto out; /* warn? */
4199
4200	mutex_lock(&ftrace_lock);
4201
4202	list_for_each_entry_safe(ftrace_mod, n, head, list) {
4203
4204		if (strcmp(ftrace_mod->module, mod) != 0)
4205			continue;
4206
4207		if (ftrace_mod->func)
4208			func = kstrdup(ftrace_mod->func, GFP_KERNEL);
4209		else
4210			func = kstrdup("*", GFP_KERNEL);
4211
4212		if (!func) /* warn? */
4213			continue;
4214
4215		list_move(&ftrace_mod->list, &process_mods);
 
4216
4217		/* Use the newly allocated func, as it may be "*" */
4218		kfree(ftrace_mod->func);
4219		ftrace_mod->func = func;
4220	}
4221
4222	mutex_unlock(&ftrace_lock);
4223
4224	list_for_each_entry_safe(ftrace_mod, n, &process_mods, list) {
4225
4226		func = ftrace_mod->func;
4227
4228		/* Grabs ftrace_lock, which is why we have this extra step */
4229		match_records(new_hash, func, strlen(func), mod);
4230		free_ftrace_mod(ftrace_mod);
4231	}
4232
4233	if (enable && list_empty(head))
4234		new_hash->flags &= ~FTRACE_HASH_FL_MOD;
4235
4236	mutex_lock(&ftrace_lock);
4237
4238	ftrace_hash_move_and_update_ops(ops, orig_hash,
4239					      new_hash, enable);
4240	mutex_unlock(&ftrace_lock);
4241
4242 out:
4243	mutex_unlock(&ops->func_hash->regex_lock);
4244
4245	free_ftrace_hash(new_hash);
4246}
4247
4248static void process_cached_mods(const char *mod_name)
4249{
4250	struct trace_array *tr;
4251	char *mod;
4252
4253	mod = kstrdup(mod_name, GFP_KERNEL);
4254	if (!mod)
4255		return;
4256
4257	mutex_lock(&trace_types_lock);
4258	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
4259		if (!list_empty(&tr->mod_trace))
4260			process_mod_list(&tr->mod_trace, tr->ops, mod, true);
4261		if (!list_empty(&tr->mod_notrace))
4262			process_mod_list(&tr->mod_notrace, tr->ops, mod, false);
4263	}
4264	mutex_unlock(&trace_types_lock);
4265
4266	kfree(mod);
4267}
4268#endif
4269
4270/*
4271 * We register the module command as a template to show others how
4272 * to register the a command as well.
4273 */
4274
4275static int
4276ftrace_mod_callback(struct trace_array *tr, struct ftrace_hash *hash,
4277		    char *func_orig, char *cmd, char *module, int enable)
4278{
4279	char *func;
4280	int ret;
4281
4282	/* match_records() modifies func, and we need the original */
4283	func = kstrdup(func_orig, GFP_KERNEL);
4284	if (!func)
4285		return -ENOMEM;
4286
4287	/*
4288	 * cmd == 'mod' because we only registered this func
4289	 * for the 'mod' ftrace_func_command.
4290	 * But if you register one func with multiple commands,
4291	 * you can tell which command was used by the cmd
4292	 * parameter.
4293	 */
4294	ret = match_records(hash, func, strlen(func), module);
4295	kfree(func);
4296
4297	if (!ret)
4298		return cache_mod(tr, func_orig, module, enable);
4299	if (ret < 0)
4300		return ret;
4301	return 0;
4302}
4303
4304static struct ftrace_func_command ftrace_mod_cmd = {
4305	.name			= "mod",
4306	.func			= ftrace_mod_callback,
4307};
4308
4309static int __init ftrace_mod_cmd_init(void)
4310{
4311	return register_ftrace_command(&ftrace_mod_cmd);
4312}
4313core_initcall(ftrace_mod_cmd_init);
4314
4315static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
4316				      struct ftrace_ops *op, struct ftrace_regs *fregs)
4317{
4318	struct ftrace_probe_ops *probe_ops;
4319	struct ftrace_func_probe *probe;
4320
4321	probe = container_of(op, struct ftrace_func_probe, ops);
4322	probe_ops = probe->probe_ops;
4323
4324	/*
4325	 * Disable preemption for these calls to prevent a RCU grace
4326	 * period. This syncs the hash iteration and freeing of items
4327	 * on the hash. rcu_read_lock is too dangerous here.
4328	 */
4329	preempt_disable_notrace();
4330	probe_ops->func(ip, parent_ip, probe->tr, probe_ops, probe->data);
4331	preempt_enable_notrace();
4332}
4333
4334struct ftrace_func_map {
4335	struct ftrace_func_entry	entry;
4336	void				*data;
4337};
4338
4339struct ftrace_func_mapper {
4340	struct ftrace_hash		hash;
4341};
4342
4343/**
4344 * allocate_ftrace_func_mapper - allocate a new ftrace_func_mapper
4345 *
4346 * Returns a ftrace_func_mapper descriptor that can be used to map ips to data.
4347 */
4348struct ftrace_func_mapper *allocate_ftrace_func_mapper(void)
4349{
4350	struct ftrace_hash *hash;
4351
4352	/*
4353	 * The mapper is simply a ftrace_hash, but since the entries
4354	 * in the hash are not ftrace_func_entry type, we define it
4355	 * as a separate structure.
4356	 */
4357	hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4358	return (struct ftrace_func_mapper *)hash;
4359}
4360
4361/**
4362 * ftrace_func_mapper_find_ip - Find some data mapped to an ip
4363 * @mapper: The mapper that has the ip maps
4364 * @ip: the instruction pointer to find the data for
4365 *
4366 * Returns the data mapped to @ip if found otherwise NULL. The return
4367 * is actually the address of the mapper data pointer. The address is
4368 * returned for use cases where the data is no bigger than a long, and
4369 * the user can use the data pointer as its data instead of having to
4370 * allocate more memory for the reference.
4371 */
4372void **ftrace_func_mapper_find_ip(struct ftrace_func_mapper *mapper,
4373				  unsigned long ip)
4374{
4375	struct ftrace_func_entry *entry;
4376	struct ftrace_func_map *map;
4377
4378	entry = ftrace_lookup_ip(&mapper->hash, ip);
4379	if (!entry)
4380		return NULL;
4381
4382	map = (struct ftrace_func_map *)entry;
4383	return &map->data;
4384}
4385
4386/**
4387 * ftrace_func_mapper_add_ip - Map some data to an ip
4388 * @mapper: The mapper that has the ip maps
4389 * @ip: The instruction pointer address to map @data to
4390 * @data: The data to map to @ip
4391 *
4392 * Returns 0 on success otherwise an error.
4393 */
4394int ftrace_func_mapper_add_ip(struct ftrace_func_mapper *mapper,
4395			      unsigned long ip, void *data)
4396{
4397	struct ftrace_func_entry *entry;
4398	struct ftrace_func_map *map;
4399
4400	entry = ftrace_lookup_ip(&mapper->hash, ip);
4401	if (entry)
4402		return -EBUSY;
4403
4404	map = kmalloc(sizeof(*map), GFP_KERNEL);
4405	if (!map)
4406		return -ENOMEM;
4407
4408	map->entry.ip = ip;
4409	map->data = data;
4410
4411	__add_hash_entry(&mapper->hash, &map->entry);
4412
4413	return 0;
4414}
4415
4416/**
4417 * ftrace_func_mapper_remove_ip - Remove an ip from the mapping
4418 * @mapper: The mapper that has the ip maps
4419 * @ip: The instruction pointer address to remove the data from
4420 *
4421 * Returns the data if it is found, otherwise NULL.
4422 * Note, if the data pointer is used as the data itself, (see 
4423 * ftrace_func_mapper_find_ip(), then the return value may be meaningless,
4424 * if the data pointer was set to zero.
4425 */
4426void *ftrace_func_mapper_remove_ip(struct ftrace_func_mapper *mapper,
4427				   unsigned long ip)
4428{
4429	struct ftrace_func_entry *entry;
4430	struct ftrace_func_map *map;
4431	void *data;
4432
4433	entry = ftrace_lookup_ip(&mapper->hash, ip);
4434	if (!entry)
4435		return NULL;
4436
4437	map = (struct ftrace_func_map *)entry;
4438	data = map->data;
4439
4440	remove_hash_entry(&mapper->hash, entry);
4441	kfree(entry);
4442
4443	return data;
4444}
4445
4446/**
4447 * free_ftrace_func_mapper - free a mapping of ips and data
4448 * @mapper: The mapper that has the ip maps
4449 * @free_func: A function to be called on each data item.
4450 *
4451 * This is used to free the function mapper. The @free_func is optional
4452 * and can be used if the data needs to be freed as well.
4453 */
4454void free_ftrace_func_mapper(struct ftrace_func_mapper *mapper,
4455			     ftrace_mapper_func free_func)
4456{
4457	struct ftrace_func_entry *entry;
4458	struct ftrace_func_map *map;
4459	struct hlist_head *hhd;
4460	int size, i;
4461
4462	if (!mapper)
4463		return;
4464
4465	if (free_func && mapper->hash.count) {
4466		size = 1 << mapper->hash.size_bits;
4467		for (i = 0; i < size; i++) {
4468			hhd = &mapper->hash.buckets[i];
4469			hlist_for_each_entry(entry, hhd, hlist) {
4470				map = (struct ftrace_func_map *)entry;
4471				free_func(map);
4472			}
4473		}
4474	}
4475	free_ftrace_hash(&mapper->hash);
4476}
4477
4478static void release_probe(struct ftrace_func_probe *probe)
4479{
4480	struct ftrace_probe_ops *probe_ops;
4481
4482	mutex_lock(&ftrace_lock);
4483
4484	WARN_ON(probe->ref <= 0);
4485
4486	/* Subtract the ref that was used to protect this instance */
4487	probe->ref--;
4488
4489	if (!probe->ref) {
4490		probe_ops = probe->probe_ops;
4491		/*
4492		 * Sending zero as ip tells probe_ops to free
4493		 * the probe->data itself
4494		 */
4495		if (probe_ops->free)
4496			probe_ops->free(probe_ops, probe->tr, 0, probe->data);
4497		list_del(&probe->list);
4498		kfree(probe);
4499	}
4500	mutex_unlock(&ftrace_lock);
4501}
4502
4503static void acquire_probe_locked(struct ftrace_func_probe *probe)
4504{
4505	/*
4506	 * Add one ref to keep it from being freed when releasing the
4507	 * ftrace_lock mutex.
4508	 */
4509	probe->ref++;
4510}
4511
4512int
4513register_ftrace_function_probe(char *glob, struct trace_array *tr,
4514			       struct ftrace_probe_ops *probe_ops,
4515			       void *data)
4516{
4517	struct ftrace_func_entry *entry;
4518	struct ftrace_func_probe *probe;
4519	struct ftrace_hash **orig_hash;
4520	struct ftrace_hash *old_hash;
4521	struct ftrace_hash *hash;
4522	int count = 0;
4523	int size;
4524	int ret;
4525	int i;
4526
4527	if (WARN_ON(!tr))
4528		return -EINVAL;
4529
4530	/* We do not support '!' for function probes */
4531	if (WARN_ON(glob[0] == '!'))
4532		return -EINVAL;
4533
4534
4535	mutex_lock(&ftrace_lock);
4536	/* Check if the probe_ops is already registered */
4537	list_for_each_entry(probe, &tr->func_probes, list) {
4538		if (probe->probe_ops == probe_ops)
4539			break;
4540	}
4541	if (&probe->list == &tr->func_probes) {
4542		probe = kzalloc(sizeof(*probe), GFP_KERNEL);
4543		if (!probe) {
4544			mutex_unlock(&ftrace_lock);
4545			return -ENOMEM;
4546		}
4547		probe->probe_ops = probe_ops;
4548		probe->ops.func = function_trace_probe_call;
4549		probe->tr = tr;
4550		ftrace_ops_init(&probe->ops);
4551		list_add(&probe->list, &tr->func_probes);
4552	}
4553
4554	acquire_probe_locked(probe);
4555
4556	mutex_unlock(&ftrace_lock);
4557
4558	/*
4559	 * Note, there's a small window here that the func_hash->filter_hash
4560	 * may be NULL or empty. Need to be careful when reading the loop.
4561	 */
4562	mutex_lock(&probe->ops.func_hash->regex_lock);
4563
4564	orig_hash = &probe->ops.func_hash->filter_hash;
4565	old_hash = *orig_hash;
4566	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4567
4568	if (!hash) {
4569		ret = -ENOMEM;
4570		goto out;
4571	}
4572
4573	ret = ftrace_match_records(hash, glob, strlen(glob));
4574
4575	/* Nothing found? */
4576	if (!ret)
4577		ret = -EINVAL;
4578
4579	if (ret < 0)
4580		goto out;
4581
4582	size = 1 << hash->size_bits;
4583	for (i = 0; i < size; i++) {
4584		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4585			if (ftrace_lookup_ip(old_hash, entry->ip))
4586				continue;
4587			/*
4588			 * The caller might want to do something special
4589			 * for each function we find. We call the callback
4590			 * to give the caller an opportunity to do so.
4591			 */
4592			if (probe_ops->init) {
4593				ret = probe_ops->init(probe_ops, tr,
4594						      entry->ip, data,
4595						      &probe->data);
4596				if (ret < 0) {
4597					if (probe_ops->free && count)
4598						probe_ops->free(probe_ops, tr,
4599								0, probe->data);
4600					probe->data = NULL;
4601					goto out;
4602				}
4603			}
4604			count++;
4605		}
4606	}
4607
4608	mutex_lock(&ftrace_lock);
4609
4610	if (!count) {
4611		/* Nothing was added? */
4612		ret = -EINVAL;
4613		goto out_unlock;
4614	}
4615
4616	ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4617					      hash, 1);
4618	if (ret < 0)
4619		goto err_unlock;
4620
4621	/* One ref for each new function traced */
4622	probe->ref += count;
4623
4624	if (!(probe->ops.flags & FTRACE_OPS_FL_ENABLED))
4625		ret = ftrace_startup(&probe->ops, 0);
4626
4627 out_unlock:
4628	mutex_unlock(&ftrace_lock);
4629
4630	if (!ret)
4631		ret = count;
4632 out:
4633	mutex_unlock(&probe->ops.func_hash->regex_lock);
4634	free_ftrace_hash(hash);
4635
4636	release_probe(probe);
4637
4638	return ret;
4639
4640 err_unlock:
4641	if (!probe_ops->free || !count)
4642		goto out_unlock;
4643
4644	/* Failed to do the move, need to call the free functions */
4645	for (i = 0; i < size; i++) {
4646		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
4647			if (ftrace_lookup_ip(old_hash, entry->ip))
4648				continue;
4649			probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4650		}
4651	}
4652	goto out_unlock;
4653}
4654
4655int
4656unregister_ftrace_function_probe_func(char *glob, struct trace_array *tr,
4657				      struct ftrace_probe_ops *probe_ops)
4658{
4659	struct ftrace_ops_hash old_hash_ops;
4660	struct ftrace_func_entry *entry;
4661	struct ftrace_func_probe *probe;
4662	struct ftrace_glob func_g;
4663	struct ftrace_hash **orig_hash;
4664	struct ftrace_hash *old_hash;
4665	struct ftrace_hash *hash = NULL;
4666	struct hlist_node *tmp;
4667	struct hlist_head hhd;
4668	char str[KSYM_SYMBOL_LEN];
4669	int count = 0;
4670	int i, ret = -ENODEV;
4671	int size;
4672
4673	if (!glob || !strlen(glob) || !strcmp(glob, "*"))
4674		func_g.search = NULL;
4675	else {
4676		int not;
4677
4678		func_g.type = filter_parse_regex(glob, strlen(glob),
4679						 &func_g.search, &not);
4680		func_g.len = strlen(func_g.search);
4681
4682		/* we do not support '!' for function probes */
4683		if (WARN_ON(not))
4684			return -EINVAL;
4685	}
4686
4687	mutex_lock(&ftrace_lock);
4688	/* Check if the probe_ops is already registered */
4689	list_for_each_entry(probe, &tr->func_probes, list) {
4690		if (probe->probe_ops == probe_ops)
4691			break;
4692	}
4693	if (&probe->list == &tr->func_probes)
4694		goto err_unlock_ftrace;
4695
4696	ret = -EINVAL;
4697	if (!(probe->ops.flags & FTRACE_OPS_FL_INITIALIZED))
4698		goto err_unlock_ftrace;
4699
4700	acquire_probe_locked(probe);
4701
4702	mutex_unlock(&ftrace_lock);
4703
4704	mutex_lock(&probe->ops.func_hash->regex_lock);
4705
4706	orig_hash = &probe->ops.func_hash->filter_hash;
4707	old_hash = *orig_hash;
4708
4709	if (ftrace_hash_empty(old_hash))
4710		goto out_unlock;
4711
4712	old_hash_ops.filter_hash = old_hash;
4713	/* Probes only have filters */
4714	old_hash_ops.notrace_hash = NULL;
4715
4716	ret = -ENOMEM;
4717	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
4718	if (!hash)
4719		goto out_unlock;
4720
4721	INIT_HLIST_HEAD(&hhd);
4722
4723	size = 1 << hash->size_bits;
4724	for (i = 0; i < size; i++) {
4725		hlist_for_each_entry_safe(entry, tmp, &hash->buckets[i], hlist) {
4726
4727			if (func_g.search) {
4728				kallsyms_lookup(entry->ip, NULL, NULL,
4729						NULL, str);
4730				if (!ftrace_match(str, &func_g))
4731					continue;
4732			}
4733			count++;
4734			remove_hash_entry(hash, entry);
4735			hlist_add_head(&entry->hlist, &hhd);
4736		}
4737	}
4738
4739	/* Nothing found? */
4740	if (!count) {
4741		ret = -EINVAL;
4742		goto out_unlock;
4743	}
4744
4745	mutex_lock(&ftrace_lock);
4746
4747	WARN_ON(probe->ref < count);
4748
4749	probe->ref -= count;
4750
4751	if (ftrace_hash_empty(hash))
4752		ftrace_shutdown(&probe->ops, 0);
4753
4754	ret = ftrace_hash_move_and_update_ops(&probe->ops, orig_hash,
4755					      hash, 1);
4756
4757	/* still need to update the function call sites */
4758	if (ftrace_enabled && !ftrace_hash_empty(hash))
4759		ftrace_run_modify_code(&probe->ops, FTRACE_UPDATE_CALLS,
4760				       &old_hash_ops);
4761	synchronize_rcu();
4762
4763	hlist_for_each_entry_safe(entry, tmp, &hhd, hlist) {
4764		hlist_del(&entry->hlist);
4765		if (probe_ops->free)
4766			probe_ops->free(probe_ops, tr, entry->ip, probe->data);
4767		kfree(entry);
4768	}
4769	mutex_unlock(&ftrace_lock);
4770
4771 out_unlock:
4772	mutex_unlock(&probe->ops.func_hash->regex_lock);
4773	free_ftrace_hash(hash);
4774
4775	release_probe(probe);
4776
4777	return ret;
4778
4779 err_unlock_ftrace:
4780	mutex_unlock(&ftrace_lock);
4781	return ret;
4782}
4783
4784void clear_ftrace_function_probes(struct trace_array *tr)
4785{
4786	struct ftrace_func_probe *probe, *n;
4787
4788	list_for_each_entry_safe(probe, n, &tr->func_probes, list)
4789		unregister_ftrace_function_probe_func(NULL, tr, probe->probe_ops);
4790}
4791
4792static LIST_HEAD(ftrace_commands);
4793static DEFINE_MUTEX(ftrace_cmd_mutex);
4794
4795/*
4796 * Currently we only register ftrace commands from __init, so mark this
4797 * __init too.
4798 */
4799__init int register_ftrace_command(struct ftrace_func_command *cmd)
4800{
4801	struct ftrace_func_command *p;
4802	int ret = 0;
4803
4804	mutex_lock(&ftrace_cmd_mutex);
4805	list_for_each_entry(p, &ftrace_commands, list) {
4806		if (strcmp(cmd->name, p->name) == 0) {
4807			ret = -EBUSY;
4808			goto out_unlock;
4809		}
4810	}
4811	list_add(&cmd->list, &ftrace_commands);
4812 out_unlock:
4813	mutex_unlock(&ftrace_cmd_mutex);
4814
4815	return ret;
4816}
4817
4818/*
4819 * Currently we only unregister ftrace commands from __init, so mark
4820 * this __init too.
4821 */
4822__init int unregister_ftrace_command(struct ftrace_func_command *cmd)
4823{
4824	struct ftrace_func_command *p, *n;
4825	int ret = -ENODEV;
4826
4827	mutex_lock(&ftrace_cmd_mutex);
4828	list_for_each_entry_safe(p, n, &ftrace_commands, list) {
4829		if (strcmp(cmd->name, p->name) == 0) {
4830			ret = 0;
4831			list_del_init(&p->list);
4832			goto out_unlock;
4833		}
4834	}
4835 out_unlock:
4836	mutex_unlock(&ftrace_cmd_mutex);
4837
4838	return ret;
4839}
4840
4841static int ftrace_process_regex(struct ftrace_iterator *iter,
4842				char *buff, int len, int enable)
4843{
4844	struct ftrace_hash *hash = iter->hash;
4845	struct trace_array *tr = iter->ops->private;
4846	char *func, *command, *next = buff;
4847	struct ftrace_func_command *p;
4848	int ret = -EINVAL;
4849
4850	func = strsep(&next, ":");
4851
4852	if (!next) {
4853		ret = ftrace_match_records(hash, func, len);
4854		if (!ret)
4855			ret = -EINVAL;
4856		if (ret < 0)
4857			return ret;
4858		return 0;
4859	}
4860
4861	/* command found */
4862
4863	command = strsep(&next, ":");
4864
4865	mutex_lock(&ftrace_cmd_mutex);
4866	list_for_each_entry(p, &ftrace_commands, list) {
4867		if (strcmp(p->name, command) == 0) {
4868			ret = p->func(tr, hash, func, command, next, enable);
4869			goto out_unlock;
4870		}
4871	}
4872 out_unlock:
4873	mutex_unlock(&ftrace_cmd_mutex);
4874
4875	return ret;
4876}
4877
4878static ssize_t
4879ftrace_regex_write(struct file *file, const char __user *ubuf,
4880		   size_t cnt, loff_t *ppos, int enable)
4881{
4882	struct ftrace_iterator *iter;
4883	struct trace_parser *parser;
4884	ssize_t ret, read;
4885
4886	if (!cnt)
4887		return 0;
4888
4889	if (file->f_mode & FMODE_READ) {
4890		struct seq_file *m = file->private_data;
4891		iter = m->private;
4892	} else
4893		iter = file->private_data;
4894
4895	if (unlikely(ftrace_disabled))
4896		return -ENODEV;
4897
4898	/* iter->hash is a local copy, so we don't need regex_lock */
4899
4900	parser = &iter->parser;
4901	read = trace_get_user(parser, ubuf, cnt, ppos);
4902
4903	if (read >= 0 && trace_parser_loaded(parser) &&
4904	    !trace_parser_cont(parser)) {
4905		ret = ftrace_process_regex(iter, parser->buffer,
4906					   parser->idx, enable);
4907		trace_parser_clear(parser);
4908		if (ret < 0)
4909			goto out;
4910	}
4911
4912	ret = read;
4913 out:
4914	return ret;
4915}
4916
4917ssize_t
4918ftrace_filter_write(struct file *file, const char __user *ubuf,
4919		    size_t cnt, loff_t *ppos)
4920{
4921	return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
4922}
4923
4924ssize_t
4925ftrace_notrace_write(struct file *file, const char __user *ubuf,
4926		     size_t cnt, loff_t *ppos)
4927{
4928	return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
4929}
4930
4931static int
4932ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
4933{
4934	struct ftrace_func_entry *entry;
4935
4936	if (!ftrace_location(ip))
4937		return -EINVAL;
4938
4939	if (remove) {
4940		entry = ftrace_lookup_ip(hash, ip);
4941		if (!entry)
4942			return -ENOENT;
4943		free_hash_entry(hash, entry);
4944		return 0;
4945	}
4946
4947	return add_hash_entry(hash, ip);
4948}
4949
4950static int
4951ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
4952		unsigned long ip, int remove, int reset, int enable)
4953{
4954	struct ftrace_hash **orig_hash;
4955	struct ftrace_hash *hash;
4956	int ret;
4957
4958	if (unlikely(ftrace_disabled))
4959		return -ENODEV;
4960
4961	mutex_lock(&ops->func_hash->regex_lock);
4962
4963	if (enable)
4964		orig_hash = &ops->func_hash->filter_hash;
4965	else
4966		orig_hash = &ops->func_hash->notrace_hash;
4967
4968	if (reset)
4969		hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4970	else
4971		hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
4972
4973	if (!hash) {
4974		ret = -ENOMEM;
4975		goto out_regex_unlock;
4976	}
4977
4978	if (buf && !ftrace_match_records(hash, buf, len)) {
4979		ret = -EINVAL;
4980		goto out_regex_unlock;
4981	}
4982	if (ip) {
4983		ret = ftrace_match_addr(hash, ip, remove);
4984		if (ret < 0)
4985			goto out_regex_unlock;
4986	}
4987
4988	mutex_lock(&ftrace_lock);
4989	ret = ftrace_hash_move_and_update_ops(ops, orig_hash, hash, enable);
4990	mutex_unlock(&ftrace_lock);
4991
4992 out_regex_unlock:
4993	mutex_unlock(&ops->func_hash->regex_lock);
4994
4995	free_ftrace_hash(hash);
4996	return ret;
4997}
4998
4999static int
5000ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
5001		int reset, int enable)
5002{
5003	return ftrace_set_hash(ops, NULL, 0, ip, remove, reset, enable);
5004}
5005
5006#ifdef CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS
5007
5008struct ftrace_direct_func {
5009	struct list_head	next;
5010	unsigned long		addr;
5011	int			count;
5012};
5013
5014static LIST_HEAD(ftrace_direct_funcs);
5015
5016/**
5017 * ftrace_find_direct_func - test an address if it is a registered direct caller
5018 * @addr: The address of a registered direct caller
5019 *
5020 * This searches to see if a ftrace direct caller has been registered
5021 * at a specific address, and if so, it returns a descriptor for it.
5022 *
5023 * This can be used by architecture code to see if an address is
5024 * a direct caller (trampoline) attached to a fentry/mcount location.
5025 * This is useful for the function_graph tracer, as it may need to
5026 * do adjustments if it traced a location that also has a direct
5027 * trampoline attached to it.
5028 */
5029struct ftrace_direct_func *ftrace_find_direct_func(unsigned long addr)
5030{
5031	struct ftrace_direct_func *entry;
5032	bool found = false;
5033
5034	/* May be called by fgraph trampoline (protected by rcu tasks) */
5035	list_for_each_entry_rcu(entry, &ftrace_direct_funcs, next) {
5036		if (entry->addr == addr) {
5037			found = true;
5038			break;
5039		}
5040	}
5041	if (found)
5042		return entry;
5043
5044	return NULL;
5045}
5046
5047static struct ftrace_direct_func *ftrace_alloc_direct_func(unsigned long addr)
5048{
5049	struct ftrace_direct_func *direct;
5050
5051	direct = kmalloc(sizeof(*direct), GFP_KERNEL);
5052	if (!direct)
5053		return NULL;
5054	direct->addr = addr;
5055	direct->count = 0;
5056	list_add_rcu(&direct->next, &ftrace_direct_funcs);
5057	ftrace_direct_func_count++;
5058	return direct;
5059}
5060
5061/**
5062 * register_ftrace_direct - Call a custom trampoline directly
5063 * @ip: The address of the nop at the beginning of a function
5064 * @addr: The address of the trampoline to call at @ip
5065 *
5066 * This is used to connect a direct call from the nop location (@ip)
5067 * at the start of ftrace traced functions. The location that it calls
5068 * (@addr) must be able to handle a direct call, and save the parameters
5069 * of the function being traced, and restore them (or inject new ones
5070 * if needed), before returning.
5071 *
5072 * Returns:
5073 *  0 on success
5074 *  -EBUSY - Another direct function is already attached (there can be only one)
5075 *  -ENODEV - @ip does not point to a ftrace nop location (or not supported)
5076 *  -ENOMEM - There was an allocation failure.
5077 */
5078int register_ftrace_direct(unsigned long ip, unsigned long addr)
5079{
5080	struct ftrace_direct_func *direct;
5081	struct ftrace_func_entry *entry;
5082	struct ftrace_hash *free_hash = NULL;
5083	struct dyn_ftrace *rec;
5084	int ret = -EBUSY;
5085
5086	mutex_lock(&direct_mutex);
5087
5088	/* See if there's a direct function at @ip already */
5089	if (ftrace_find_rec_direct(ip))
5090		goto out_unlock;
5091
5092	ret = -ENODEV;
5093	rec = lookup_rec(ip, ip);
5094	if (!rec)
5095		goto out_unlock;
5096
5097	/*
5098	 * Check if the rec says it has a direct call but we didn't
5099	 * find one earlier?
5100	 */
5101	if (WARN_ON(rec->flags & FTRACE_FL_DIRECT))
5102		goto out_unlock;
5103
5104	/* Make sure the ip points to the exact record */
5105	if (ip != rec->ip) {
5106		ip = rec->ip;
5107		/* Need to check this ip for a direct. */
5108		if (ftrace_find_rec_direct(ip))
5109			goto out_unlock;
5110	}
5111
5112	ret = -ENOMEM;
5113	if (ftrace_hash_empty(direct_functions) ||
5114	    direct_functions->count > 2 * (1 << direct_functions->size_bits)) {
5115		struct ftrace_hash *new_hash;
5116		int size = ftrace_hash_empty(direct_functions) ? 0 :
5117			direct_functions->count + 1;
5118
5119		if (size < 32)
5120			size = 32;
5121
5122		new_hash = dup_hash(direct_functions, size);
5123		if (!new_hash)
5124			goto out_unlock;
5125
5126		free_hash = direct_functions;
5127		direct_functions = new_hash;
5128	}
5129
5130	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
5131	if (!entry)
5132		goto out_unlock;
5133
5134	direct = ftrace_find_direct_func(addr);
5135	if (!direct) {
5136		direct = ftrace_alloc_direct_func(addr);
5137		if (!direct) {
5138			kfree(entry);
5139			goto out_unlock;
5140		}
 
 
 
 
5141	}
5142
5143	entry->ip = ip;
5144	entry->direct = addr;
5145	__add_hash_entry(direct_functions, entry);
5146
5147	ret = ftrace_set_filter_ip(&direct_ops, ip, 0, 0);
5148	if (ret)
5149		remove_hash_entry(direct_functions, entry);
5150
5151	if (!ret && !(direct_ops.flags & FTRACE_OPS_FL_ENABLED)) {
5152		ret = register_ftrace_function(&direct_ops);
5153		if (ret)
5154			ftrace_set_filter_ip(&direct_ops, ip, 1, 0);
5155	}
5156
5157	if (ret) {
5158		kfree(entry);
5159		if (!direct->count) {
5160			list_del_rcu(&direct->next);
5161			synchronize_rcu_tasks();
5162			kfree(direct);
5163			if (free_hash)
5164				free_ftrace_hash(free_hash);
5165			free_hash = NULL;
5166			ftrace_direct_func_count--;
5167		}
5168	} else {
5169		direct->count++;
5170	}
5171 out_unlock:
5172	mutex_unlock(&direct_mutex);
5173
5174	if (free_hash) {
5175		synchronize_rcu_tasks();
5176		free_ftrace_hash(free_hash);
5177	}
5178
5179	return ret;
5180}
5181EXPORT_SYMBOL_GPL(register_ftrace_direct);
5182
5183static struct ftrace_func_entry *find_direct_entry(unsigned long *ip,
5184						   struct dyn_ftrace **recp)
5185{
5186	struct ftrace_func_entry *entry;
5187	struct dyn_ftrace *rec;
5188
5189	rec = lookup_rec(*ip, *ip);
5190	if (!rec)
5191		return NULL;
5192
5193	entry = __ftrace_lookup_ip(direct_functions, rec->ip);
5194	if (!entry) {
5195		WARN_ON(rec->flags & FTRACE_FL_DIRECT);
5196		return NULL;
5197	}
5198
5199	WARN_ON(!(rec->flags & FTRACE_FL_DIRECT));
5200
5201	/* Passed in ip just needs to be on the call site */
5202	*ip = rec->ip;
5203
5204	if (recp)
5205		*recp = rec;
5206
5207	return entry;
5208}
5209
5210int unregister_ftrace_direct(unsigned long ip, unsigned long addr)
5211{
5212	struct ftrace_direct_func *direct;
5213	struct ftrace_func_entry *entry;
5214	int ret = -ENODEV;
5215
5216	mutex_lock(&direct_mutex);
5217
5218	entry = find_direct_entry(&ip, NULL);
5219	if (!entry)
5220		goto out_unlock;
5221
5222	if (direct_functions->count == 1)
5223		unregister_ftrace_function(&direct_ops);
5224
5225	ret = ftrace_set_filter_ip(&direct_ops, ip, 1, 0);
5226
5227	WARN_ON(ret);
5228
5229	remove_hash_entry(direct_functions, entry);
5230
5231	direct = ftrace_find_direct_func(addr);
5232	if (!WARN_ON(!direct)) {
5233		/* This is the good path (see the ! before WARN) */
5234		direct->count--;
5235		WARN_ON(direct->count < 0);
5236		if (!direct->count) {
5237			list_del_rcu(&direct->next);
5238			synchronize_rcu_tasks();
5239			kfree(direct);
5240			kfree(entry);
5241			ftrace_direct_func_count--;
5242		}
5243	}
5244 out_unlock:
5245	mutex_unlock(&direct_mutex);
5246
5247	return ret;
5248}
5249EXPORT_SYMBOL_GPL(unregister_ftrace_direct);
5250
5251static struct ftrace_ops stub_ops = {
5252	.func		= ftrace_stub,
5253};
5254
5255/**
5256 * ftrace_modify_direct_caller - modify ftrace nop directly
5257 * @entry: The ftrace hash entry of the direct helper for @rec
5258 * @rec: The record representing the function site to patch
5259 * @old_addr: The location that the site at @rec->ip currently calls
5260 * @new_addr: The location that the site at @rec->ip should call
5261 *
5262 * An architecture may overwrite this function to optimize the
5263 * changing of the direct callback on an ftrace nop location.
5264 * This is called with the ftrace_lock mutex held, and no other
5265 * ftrace callbacks are on the associated record (@rec). Thus,
5266 * it is safe to modify the ftrace record, where it should be
5267 * currently calling @old_addr directly, to call @new_addr.
5268 *
5269 * Safety checks should be made to make sure that the code at
5270 * @rec->ip is currently calling @old_addr. And this must
5271 * also update entry->direct to @new_addr.
5272 */
5273int __weak ftrace_modify_direct_caller(struct ftrace_func_entry *entry,
5274				       struct dyn_ftrace *rec,
5275				       unsigned long old_addr,
5276				       unsigned long new_addr)
5277{
5278	unsigned long ip = rec->ip;
5279	int ret;
5280
5281	/*
5282	 * The ftrace_lock was used to determine if the record
5283	 * had more than one registered user to it. If it did,
5284	 * we needed to prevent that from changing to do the quick
5285	 * switch. But if it did not (only a direct caller was attached)
5286	 * then this function is called. But this function can deal
5287	 * with attached callers to the rec that we care about, and
5288	 * since this function uses standard ftrace calls that take
5289	 * the ftrace_lock mutex, we need to release it.
5290	 */
5291	mutex_unlock(&ftrace_lock);
5292
5293	/*
5294	 * By setting a stub function at the same address, we force
5295	 * the code to call the iterator and the direct_ops helper.
5296	 * This means that @ip does not call the direct call, and
5297	 * we can simply modify it.
5298	 */
5299	ret = ftrace_set_filter_ip(&stub_ops, ip, 0, 0);
5300	if (ret)
5301		goto out_lock;
5302
5303	ret = register_ftrace_function(&stub_ops);
5304	if (ret) {
5305		ftrace_set_filter_ip(&stub_ops, ip, 1, 0);
5306		goto out_lock;
5307	}
5308
5309	entry->direct = new_addr;
5310
5311	/*
5312	 * By removing the stub, we put back the direct call, calling
5313	 * the @new_addr.
5314	 */
5315	unregister_ftrace_function(&stub_ops);
5316	ftrace_set_filter_ip(&stub_ops, ip, 1, 0);
5317
5318 out_lock:
5319	mutex_lock(&ftrace_lock);
5320
5321	return ret;
5322}
5323
5324/**
5325 * modify_ftrace_direct - Modify an existing direct call to call something else
5326 * @ip: The instruction pointer to modify
5327 * @old_addr: The address that the current @ip calls directly
5328 * @new_addr: The address that the @ip should call
5329 *
5330 * This modifies a ftrace direct caller at an instruction pointer without
5331 * having to disable it first. The direct call will switch over to the
5332 * @new_addr without missing anything.
5333 *
5334 * Returns: zero on success. Non zero on error, which includes:
5335 *  -ENODEV : the @ip given has no direct caller attached
5336 *  -EINVAL : the @old_addr does not match the current direct caller
5337 */
5338int modify_ftrace_direct(unsigned long ip,
5339			 unsigned long old_addr, unsigned long new_addr)
5340{
5341	struct ftrace_direct_func *direct, *new_direct = NULL;
5342	struct ftrace_func_entry *entry;
5343	struct dyn_ftrace *rec;
5344	int ret = -ENODEV;
5345
5346	mutex_lock(&direct_mutex);
5347
5348	mutex_lock(&ftrace_lock);
5349	entry = find_direct_entry(&ip, &rec);
5350	if (!entry)
5351		goto out_unlock;
5352
5353	ret = -EINVAL;
5354	if (entry->direct != old_addr)
5355		goto out_unlock;
5356
5357	direct = ftrace_find_direct_func(old_addr);
5358	if (WARN_ON(!direct))
5359		goto out_unlock;
5360	if (direct->count > 1) {
5361		ret = -ENOMEM;
5362		new_direct = ftrace_alloc_direct_func(new_addr);
5363		if (!new_direct)
5364			goto out_unlock;
5365		direct->count--;
5366		new_direct->count++;
5367	} else {
5368		direct->addr = new_addr;
5369	}
5370
5371	/*
5372	 * If there's no other ftrace callback on the rec->ip location,
5373	 * then it can be changed directly by the architecture.
5374	 * If there is another caller, then we just need to change the
5375	 * direct caller helper to point to @new_addr.
5376	 */
5377	if (ftrace_rec_count(rec) == 1) {
5378		ret = ftrace_modify_direct_caller(entry, rec, old_addr, new_addr);
5379	} else {
5380		entry->direct = new_addr;
5381		ret = 0;
5382	}
5383
5384	if (unlikely(ret && new_direct)) {
5385		direct->count++;
5386		list_del_rcu(&new_direct->next);
5387		synchronize_rcu_tasks();
5388		kfree(new_direct);
5389		ftrace_direct_func_count--;
5390	}
5391
5392 out_unlock:
5393	mutex_unlock(&ftrace_lock);
5394	mutex_unlock(&direct_mutex);
5395	return ret;
5396}
5397EXPORT_SYMBOL_GPL(modify_ftrace_direct);
5398#endif /* CONFIG_DYNAMIC_FTRACE_WITH_DIRECT_CALLS */
5399
5400/**
5401 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
5402 * @ops - the ops to set the filter with
5403 * @ip - the address to add to or remove from the filter.
5404 * @remove - non zero to remove the ip from the filter
5405 * @reset - non zero to reset all filters before applying this filter.
5406 *
5407 * Filters denote which functions should be enabled when tracing is enabled
5408 * If @ip is NULL, it fails to update filter.
5409 */
5410int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
5411			 int remove, int reset)
5412{
5413	ftrace_ops_init(ops);
5414	return ftrace_set_addr(ops, ip, remove, reset, 1);
5415}
5416EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
5417
5418/**
5419 * ftrace_ops_set_global_filter - setup ops to use global filters
5420 * @ops - the ops which will use the global filters
5421 *
5422 * ftrace users who need global function trace filtering should call this.
5423 * It can set the global filter only if ops were not initialized before.
5424 */
5425void ftrace_ops_set_global_filter(struct ftrace_ops *ops)
5426{
5427	if (ops->flags & FTRACE_OPS_FL_INITIALIZED)
5428		return;
5429
5430	ftrace_ops_init(ops);
5431	ops->func_hash = &global_ops.local_hash;
5432}
5433EXPORT_SYMBOL_GPL(ftrace_ops_set_global_filter);
5434
5435static int
5436ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
5437		 int reset, int enable)
5438{
5439	return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
5440}
5441
5442/**
5443 * ftrace_set_filter - set a function to filter on in ftrace
5444 * @ops - the ops to set the filter with
5445 * @buf - the string that holds the function filter text.
5446 * @len - the length of the string.
5447 * @reset - non zero to reset all filters before applying this filter.
5448 *
5449 * Filters denote which functions should be enabled when tracing is enabled.
5450 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
5451 */
5452int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
5453		       int len, int reset)
5454{
5455	ftrace_ops_init(ops);
5456	return ftrace_set_regex(ops, buf, len, reset, 1);
5457}
5458EXPORT_SYMBOL_GPL(ftrace_set_filter);
5459
5460/**
5461 * ftrace_set_notrace - set a function to not trace in ftrace
5462 * @ops - the ops to set the notrace filter with
5463 * @buf - the string that holds the function notrace text.
5464 * @len - the length of the string.
5465 * @reset - non zero to reset all filters before applying this filter.
5466 *
5467 * Notrace Filters denote which functions should not be enabled when tracing
5468 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
5469 * for tracing.
5470 */
5471int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
5472			int len, int reset)
5473{
5474	ftrace_ops_init(ops);
5475	return ftrace_set_regex(ops, buf, len, reset, 0);
5476}
5477EXPORT_SYMBOL_GPL(ftrace_set_notrace);
5478/**
5479 * ftrace_set_global_filter - set a function to filter on with global tracers
5480 * @buf - the string that holds the function filter text.
5481 * @len - the length of the string.
5482 * @reset - non zero to reset all filters before applying this filter.
5483 *
5484 * Filters denote which functions should be enabled when tracing is enabled.
5485 * If @buf is NULL and reset is set, all functions will be enabled for tracing.
5486 */
5487void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
5488{
5489	ftrace_set_regex(&global_ops, buf, len, reset, 1);
5490}
5491EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
5492
5493/**
5494 * ftrace_set_global_notrace - set a function to not trace with global tracers
5495 * @buf - the string that holds the function notrace text.
5496 * @len - the length of the string.
5497 * @reset - non zero to reset all filters before applying this filter.
5498 *
5499 * Notrace Filters denote which functions should not be enabled when tracing
5500 * is enabled. If @buf is NULL and reset is set, all functions will be enabled
5501 * for tracing.
5502 */
5503void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
5504{
5505	ftrace_set_regex(&global_ops, buf, len, reset, 0);
5506}
5507EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
5508
5509/*
5510 * command line interface to allow users to set filters on boot up.
5511 */
5512#define FTRACE_FILTER_SIZE		COMMAND_LINE_SIZE
5513static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
5514static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
5515
5516/* Used by function selftest to not test if filter is set */
5517bool ftrace_filter_param __initdata;
5518
5519static int __init set_ftrace_notrace(char *str)
5520{
5521	ftrace_filter_param = true;
5522	strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
5523	return 1;
5524}
5525__setup("ftrace_notrace=", set_ftrace_notrace);
5526
5527static int __init set_ftrace_filter(char *str)
5528{
5529	ftrace_filter_param = true;
5530	strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
5531	return 1;
5532}
5533__setup("ftrace_filter=", set_ftrace_filter);
5534
5535#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5536static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
5537static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
5538static int ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer);
5539
5540static int __init set_graph_function(char *str)
5541{
5542	strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
5543	return 1;
5544}
5545__setup("ftrace_graph_filter=", set_graph_function);
5546
5547static int __init set_graph_notrace_function(char *str)
5548{
5549	strlcpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
5550	return 1;
5551}
5552__setup("ftrace_graph_notrace=", set_graph_notrace_function);
5553
5554static int __init set_graph_max_depth_function(char *str)
5555{
5556	if (!str)
5557		return 0;
5558	fgraph_max_depth = simple_strtoul(str, NULL, 0);
5559	return 1;
5560}
5561__setup("ftrace_graph_max_depth=", set_graph_max_depth_function);
5562
5563static void __init set_ftrace_early_graph(char *buf, int enable)
5564{
5565	int ret;
5566	char *func;
5567	struct ftrace_hash *hash;
5568
5569	hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
5570	if (MEM_FAIL(!hash, "Failed to allocate hash\n"))
5571		return;
5572
5573	while (buf) {
5574		func = strsep(&buf, ",");
5575		/* we allow only one expression at a time */
5576		ret = ftrace_graph_set_hash(hash, func);
5577		if (ret)
5578			printk(KERN_DEBUG "ftrace: function %s not "
5579					  "traceable\n", func);
5580	}
5581
5582	if (enable)
5583		ftrace_graph_hash = hash;
5584	else
5585		ftrace_graph_notrace_hash = hash;
5586}
5587#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5588
5589void __init
5590ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
5591{
5592	char *func;
5593
5594	ftrace_ops_init(ops);
5595
5596	while (buf) {
5597		func = strsep(&buf, ",");
5598		ftrace_set_regex(ops, func, strlen(func), 0, enable);
5599	}
5600}
5601
5602static void __init set_ftrace_early_filters(void)
5603{
5604	if (ftrace_filter_buf[0])
5605		ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
5606	if (ftrace_notrace_buf[0])
5607		ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
5608#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5609	if (ftrace_graph_buf[0])
5610		set_ftrace_early_graph(ftrace_graph_buf, 1);
5611	if (ftrace_graph_notrace_buf[0])
5612		set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
5613#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
5614}
5615
5616int ftrace_regex_release(struct inode *inode, struct file *file)
5617{
5618	struct seq_file *m = (struct seq_file *)file->private_data;
5619	struct ftrace_iterator *iter;
5620	struct ftrace_hash **orig_hash;
5621	struct trace_parser *parser;
5622	int filter_hash;
 
5623
5624	if (file->f_mode & FMODE_READ) {
5625		iter = m->private;
5626		seq_release(inode, file);
5627	} else
5628		iter = file->private_data;
5629
5630	parser = &iter->parser;
5631	if (trace_parser_loaded(parser)) {
5632		int enable = !(iter->flags & FTRACE_ITER_NOTRACE);
5633
5634		ftrace_process_regex(iter, parser->buffer,
5635				     parser->idx, enable);
5636	}
5637
5638	trace_parser_put(parser);
5639
5640	mutex_lock(&iter->ops->func_hash->regex_lock);
5641
5642	if (file->f_mode & FMODE_WRITE) {
5643		filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
5644
5645		if (filter_hash) {
5646			orig_hash = &iter->ops->func_hash->filter_hash;
5647			if (iter->tr && !list_empty(&iter->tr->mod_trace))
5648				iter->hash->flags |= FTRACE_HASH_FL_MOD;
5649		} else
5650			orig_hash = &iter->ops->func_hash->notrace_hash;
5651
5652		mutex_lock(&ftrace_lock);
5653		ftrace_hash_move_and_update_ops(iter->ops, orig_hash,
5654						      iter->hash, filter_hash);
5655		mutex_unlock(&ftrace_lock);
5656	} else {
5657		/* For read only, the hash is the ops hash */
5658		iter->hash = NULL;
5659	}
5660
5661	mutex_unlock(&iter->ops->func_hash->regex_lock);
5662	free_ftrace_hash(iter->hash);
5663	if (iter->tr)
5664		trace_array_put(iter->tr);
5665	kfree(iter);
5666
5667	return 0;
5668}
5669
5670static const struct file_operations ftrace_avail_fops = {
5671	.open = ftrace_avail_open,
5672	.read = seq_read,
5673	.llseek = seq_lseek,
5674	.release = seq_release_private,
5675};
5676
5677static const struct file_operations ftrace_enabled_fops = {
5678	.open = ftrace_enabled_open,
5679	.read = seq_read,
5680	.llseek = seq_lseek,
5681	.release = seq_release_private,
5682};
5683
5684static const struct file_operations ftrace_filter_fops = {
5685	.open = ftrace_filter_open,
5686	.read = seq_read,
5687	.write = ftrace_filter_write,
5688	.llseek = tracing_lseek,
5689	.release = ftrace_regex_release,
5690};
5691
5692static const struct file_operations ftrace_notrace_fops = {
5693	.open = ftrace_notrace_open,
5694	.read = seq_read,
5695	.write = ftrace_notrace_write,
5696	.llseek = tracing_lseek,
5697	.release = ftrace_regex_release,
5698};
5699
5700#ifdef CONFIG_FUNCTION_GRAPH_TRACER
5701
5702static DEFINE_MUTEX(graph_lock);
5703
5704struct ftrace_hash __rcu *ftrace_graph_hash = EMPTY_HASH;
5705struct ftrace_hash __rcu *ftrace_graph_notrace_hash = EMPTY_HASH;
5706
5707enum graph_filter_type {
5708	GRAPH_FILTER_NOTRACE	= 0,
5709	GRAPH_FILTER_FUNCTION,
5710};
5711
5712#define FTRACE_GRAPH_EMPTY	((void *)1)
5713
5714struct ftrace_graph_data {
5715	struct ftrace_hash		*hash;
5716	struct ftrace_func_entry	*entry;
5717	int				idx;   /* for hash table iteration */
5718	enum graph_filter_type		type;
5719	struct ftrace_hash		*new_hash;
5720	const struct seq_operations	*seq_ops;
5721	struct trace_parser		parser;
5722};
5723
5724static void *
5725__g_next(struct seq_file *m, loff_t *pos)
5726{
5727	struct ftrace_graph_data *fgd = m->private;
5728	struct ftrace_func_entry *entry = fgd->entry;
5729	struct hlist_head *head;
5730	int i, idx = fgd->idx;
5731
5732	if (*pos >= fgd->hash->count)
5733		return NULL;
5734
5735	if (entry) {
5736		hlist_for_each_entry_continue(entry, hlist) {
5737			fgd->entry = entry;
5738			return entry;
5739		}
5740
5741		idx++;
5742	}
5743
5744	for (i = idx; i < 1 << fgd->hash->size_bits; i++) {
5745		head = &fgd->hash->buckets[i];
5746		hlist_for_each_entry(entry, head, hlist) {
5747			fgd->entry = entry;
5748			fgd->idx = i;
5749			return entry;
5750		}
5751	}
5752	return NULL;
5753}
5754
5755static void *
5756g_next(struct seq_file *m, void *v, loff_t *pos)
5757{
5758	(*pos)++;
5759	return __g_next(m, pos);
5760}
5761
5762static void *g_start(struct seq_file *m, loff_t *pos)
5763{
5764	struct ftrace_graph_data *fgd = m->private;
5765
5766	mutex_lock(&graph_lock);
5767
5768	if (fgd->type == GRAPH_FILTER_FUNCTION)
5769		fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5770					lockdep_is_held(&graph_lock));
5771	else
5772		fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5773					lockdep_is_held(&graph_lock));
5774
5775	/* Nothing, tell g_show to print all functions are enabled */
5776	if (ftrace_hash_empty(fgd->hash) && !*pos)
5777		return FTRACE_GRAPH_EMPTY;
5778
5779	fgd->idx = 0;
5780	fgd->entry = NULL;
5781	return __g_next(m, pos);
5782}
5783
5784static void g_stop(struct seq_file *m, void *p)
5785{
5786	mutex_unlock(&graph_lock);
5787}
5788
5789static int g_show(struct seq_file *m, void *v)
5790{
5791	struct ftrace_func_entry *entry = v;
5792
5793	if (!entry)
5794		return 0;
5795
5796	if (entry == FTRACE_GRAPH_EMPTY) {
5797		struct ftrace_graph_data *fgd = m->private;
5798
5799		if (fgd->type == GRAPH_FILTER_FUNCTION)
5800			seq_puts(m, "#### all functions enabled ####\n");
5801		else
5802			seq_puts(m, "#### no functions disabled ####\n");
5803		return 0;
5804	}
5805
5806	seq_printf(m, "%ps\n", (void *)entry->ip);
5807
5808	return 0;
5809}
5810
5811static const struct seq_operations ftrace_graph_seq_ops = {
5812	.start = g_start,
5813	.next = g_next,
5814	.stop = g_stop,
5815	.show = g_show,
5816};
5817
5818static int
5819__ftrace_graph_open(struct inode *inode, struct file *file,
5820		    struct ftrace_graph_data *fgd)
5821{
5822	int ret;
5823	struct ftrace_hash *new_hash = NULL;
5824
5825	ret = security_locked_down(LOCKDOWN_TRACEFS);
5826	if (ret)
5827		return ret;
5828
5829	if (file->f_mode & FMODE_WRITE) {
5830		const int size_bits = FTRACE_HASH_DEFAULT_BITS;
5831
5832		if (trace_parser_get_init(&fgd->parser, FTRACE_BUFF_MAX))
5833			return -ENOMEM;
5834
5835		if (file->f_flags & O_TRUNC)
5836			new_hash = alloc_ftrace_hash(size_bits);
5837		else
5838			new_hash = alloc_and_copy_ftrace_hash(size_bits,
5839							      fgd->hash);
5840		if (!new_hash) {
5841			ret = -ENOMEM;
5842			goto out;
5843		}
5844	}
5845
5846	if (file->f_mode & FMODE_READ) {
5847		ret = seq_open(file, &ftrace_graph_seq_ops);
5848		if (!ret) {
5849			struct seq_file *m = file->private_data;
5850			m->private = fgd;
5851		} else {
5852			/* Failed */
5853			free_ftrace_hash(new_hash);
5854			new_hash = NULL;
5855		}
5856	} else
5857		file->private_data = fgd;
5858
5859out:
5860	if (ret < 0 && file->f_mode & FMODE_WRITE)
5861		trace_parser_put(&fgd->parser);
5862
5863	fgd->new_hash = new_hash;
5864
5865	/*
5866	 * All uses of fgd->hash must be taken with the graph_lock
5867	 * held. The graph_lock is going to be released, so force
5868	 * fgd->hash to be reinitialized when it is taken again.
5869	 */
5870	fgd->hash = NULL;
5871
5872	return ret;
5873}
5874
5875static int
5876ftrace_graph_open(struct inode *inode, struct file *file)
5877{
5878	struct ftrace_graph_data *fgd;
5879	int ret;
5880
5881	if (unlikely(ftrace_disabled))
5882		return -ENODEV;
5883
5884	fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5885	if (fgd == NULL)
5886		return -ENOMEM;
5887
5888	mutex_lock(&graph_lock);
5889
5890	fgd->hash = rcu_dereference_protected(ftrace_graph_hash,
5891					lockdep_is_held(&graph_lock));
5892	fgd->type = GRAPH_FILTER_FUNCTION;
5893	fgd->seq_ops = &ftrace_graph_seq_ops;
5894
5895	ret = __ftrace_graph_open(inode, file, fgd);
5896	if (ret < 0)
5897		kfree(fgd);
5898
5899	mutex_unlock(&graph_lock);
5900	return ret;
5901}
5902
5903static int
5904ftrace_graph_notrace_open(struct inode *inode, struct file *file)
5905{
5906	struct ftrace_graph_data *fgd;
5907	int ret;
5908
5909	if (unlikely(ftrace_disabled))
5910		return -ENODEV;
5911
5912	fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
5913	if (fgd == NULL)
5914		return -ENOMEM;
5915
5916	mutex_lock(&graph_lock);
5917
5918	fgd->hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5919					lockdep_is_held(&graph_lock));
5920	fgd->type = GRAPH_FILTER_NOTRACE;
5921	fgd->seq_ops = &ftrace_graph_seq_ops;
5922
5923	ret = __ftrace_graph_open(inode, file, fgd);
5924	if (ret < 0)
5925		kfree(fgd);
5926
5927	mutex_unlock(&graph_lock);
5928	return ret;
5929}
5930
5931static int
5932ftrace_graph_release(struct inode *inode, struct file *file)
5933{
5934	struct ftrace_graph_data *fgd;
5935	struct ftrace_hash *old_hash, *new_hash;
5936	struct trace_parser *parser;
5937	int ret = 0;
5938
5939	if (file->f_mode & FMODE_READ) {
5940		struct seq_file *m = file->private_data;
5941
5942		fgd = m->private;
5943		seq_release(inode, file);
5944	} else {
5945		fgd = file->private_data;
5946	}
5947
5948
5949	if (file->f_mode & FMODE_WRITE) {
5950
5951		parser = &fgd->parser;
5952
5953		if (trace_parser_loaded((parser))) {
5954			ret = ftrace_graph_set_hash(fgd->new_hash,
5955						    parser->buffer);
5956		}
5957
5958		trace_parser_put(parser);
5959
5960		new_hash = __ftrace_hash_move(fgd->new_hash);
5961		if (!new_hash) {
5962			ret = -ENOMEM;
5963			goto out;
5964		}
5965
5966		mutex_lock(&graph_lock);
5967
5968		if (fgd->type == GRAPH_FILTER_FUNCTION) {
5969			old_hash = rcu_dereference_protected(ftrace_graph_hash,
5970					lockdep_is_held(&graph_lock));
5971			rcu_assign_pointer(ftrace_graph_hash, new_hash);
5972		} else {
5973			old_hash = rcu_dereference_protected(ftrace_graph_notrace_hash,
5974					lockdep_is_held(&graph_lock));
5975			rcu_assign_pointer(ftrace_graph_notrace_hash, new_hash);
5976		}
5977
5978		mutex_unlock(&graph_lock);
5979
5980		/*
5981		 * We need to do a hard force of sched synchronization.
5982		 * This is because we use preempt_disable() to do RCU, but
5983		 * the function tracers can be called where RCU is not watching
5984		 * (like before user_exit()). We can not rely on the RCU
5985		 * infrastructure to do the synchronization, thus we must do it
5986		 * ourselves.
5987		 */
5988		if (old_hash != EMPTY_HASH)
5989			synchronize_rcu_tasks_rude();
5990
5991		free_ftrace_hash(old_hash);
5992	}
5993
5994 out:
5995	free_ftrace_hash(fgd->new_hash);
5996	kfree(fgd);
5997
5998	return ret;
5999}
6000
6001static int
6002ftrace_graph_set_hash(struct ftrace_hash *hash, char *buffer)
6003{
6004	struct ftrace_glob func_g;
6005	struct dyn_ftrace *rec;
6006	struct ftrace_page *pg;
6007	struct ftrace_func_entry *entry;
6008	int fail = 1;
6009	int not;
6010
6011	/* decode regex */
6012	func_g.type = filter_parse_regex(buffer, strlen(buffer),
6013					 &func_g.search, &not);
6014
6015	func_g.len = strlen(func_g.search);
6016
6017	mutex_lock(&ftrace_lock);
6018
6019	if (unlikely(ftrace_disabled)) {
6020		mutex_unlock(&ftrace_lock);
6021		return -ENODEV;
6022	}
6023
6024	do_for_each_ftrace_rec(pg, rec) {
6025
6026		if (rec->flags & FTRACE_FL_DISABLED)
6027			continue;
6028
6029		if (ftrace_match_record(rec, &func_g, NULL, 0)) {
6030			entry = ftrace_lookup_ip(hash, rec->ip);
6031
6032			if (!not) {
6033				fail = 0;
6034
6035				if (entry)
6036					continue;
6037				if (add_hash_entry(hash, rec->ip) < 0)
6038					goto out;
6039			} else {
6040				if (entry) {
6041					free_hash_entry(hash, entry);
6042					fail = 0;
6043				}
6044			}
6045		}
6046	} while_for_each_ftrace_rec();
6047out:
6048	mutex_unlock(&ftrace_lock);
6049
6050	if (fail)
6051		return -EINVAL;
6052
6053	return 0;
6054}
6055
6056static ssize_t
6057ftrace_graph_write(struct file *file, const char __user *ubuf,
6058		   size_t cnt, loff_t *ppos)
6059{
6060	ssize_t read, ret = 0;
6061	struct ftrace_graph_data *fgd = file->private_data;
6062	struct trace_parser *parser;
6063
6064	if (!cnt)
6065		return 0;
6066
6067	/* Read mode uses seq functions */
6068	if (file->f_mode & FMODE_READ) {
6069		struct seq_file *m = file->private_data;
6070		fgd = m->private;
6071	}
6072
6073	parser = &fgd->parser;
6074
6075	read = trace_get_user(parser, ubuf, cnt, ppos);
6076
6077	if (read >= 0 && trace_parser_loaded(parser) &&
6078	    !trace_parser_cont(parser)) {
6079
6080		ret = ftrace_graph_set_hash(fgd->new_hash,
6081					    parser->buffer);
6082		trace_parser_clear(parser);
6083	}
6084
6085	if (!ret)
6086		ret = read;
6087
6088	return ret;
6089}
6090
6091static const struct file_operations ftrace_graph_fops = {
6092	.open		= ftrace_graph_open,
6093	.read		= seq_read,
6094	.write		= ftrace_graph_write,
6095	.llseek		= tracing_lseek,
6096	.release	= ftrace_graph_release,
6097};
6098
6099static const struct file_operations ftrace_graph_notrace_fops = {
6100	.open		= ftrace_graph_notrace_open,
6101	.read		= seq_read,
6102	.write		= ftrace_graph_write,
6103	.llseek		= tracing_lseek,
6104	.release	= ftrace_graph_release,
6105};
6106#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6107
6108void ftrace_create_filter_files(struct ftrace_ops *ops,
6109				struct dentry *parent)
6110{
6111
6112	trace_create_file("set_ftrace_filter", 0644, parent,
6113			  ops, &ftrace_filter_fops);
6114
6115	trace_create_file("set_ftrace_notrace", 0644, parent,
6116			  ops, &ftrace_notrace_fops);
6117}
6118
6119/*
6120 * The name "destroy_filter_files" is really a misnomer. Although
6121 * in the future, it may actually delete the files, but this is
6122 * really intended to make sure the ops passed in are disabled
6123 * and that when this function returns, the caller is free to
6124 * free the ops.
6125 *
6126 * The "destroy" name is only to match the "create" name that this
6127 * should be paired with.
6128 */
6129void ftrace_destroy_filter_files(struct ftrace_ops *ops)
6130{
6131	mutex_lock(&ftrace_lock);
6132	if (ops->flags & FTRACE_OPS_FL_ENABLED)
6133		ftrace_shutdown(ops, 0);
6134	ops->flags |= FTRACE_OPS_FL_DELETED;
6135	ftrace_free_filter(ops);
6136	mutex_unlock(&ftrace_lock);
6137}
6138
6139static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
6140{
6141
6142	trace_create_file("available_filter_functions", 0444,
6143			d_tracer, NULL, &ftrace_avail_fops);
6144
6145	trace_create_file("enabled_functions", 0444,
6146			d_tracer, NULL, &ftrace_enabled_fops);
6147
6148	ftrace_create_filter_files(&global_ops, d_tracer);
6149
6150#ifdef CONFIG_FUNCTION_GRAPH_TRACER
6151	trace_create_file("set_graph_function", 0644, d_tracer,
6152				    NULL,
6153				    &ftrace_graph_fops);
6154	trace_create_file("set_graph_notrace", 0644, d_tracer,
6155				    NULL,
6156				    &ftrace_graph_notrace_fops);
6157#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
6158
6159	return 0;
6160}
6161
6162static int ftrace_cmp_ips(const void *a, const void *b)
6163{
6164	const unsigned long *ipa = a;
6165	const unsigned long *ipb = b;
6166
6167	if (*ipa > *ipb)
6168		return 1;
6169	if (*ipa < *ipb)
6170		return -1;
6171	return 0;
6172}
6173
6174static int ftrace_process_locs(struct module *mod,
6175			       unsigned long *start,
6176			       unsigned long *end)
6177{
6178	struct ftrace_page *start_pg;
6179	struct ftrace_page *pg;
6180	struct dyn_ftrace *rec;
6181	unsigned long count;
6182	unsigned long *p;
6183	unsigned long addr;
6184	unsigned long flags = 0; /* Shut up gcc */
6185	int ret = -ENOMEM;
6186
6187	count = end - start;
6188
6189	if (!count)
6190		return 0;
6191
6192	sort(start, count, sizeof(*start),
6193	     ftrace_cmp_ips, NULL);
6194
6195	start_pg = ftrace_allocate_pages(count);
6196	if (!start_pg)
6197		return -ENOMEM;
6198
6199	mutex_lock(&ftrace_lock);
6200
6201	/*
6202	 * Core and each module needs their own pages, as
6203	 * modules will free them when they are removed.
6204	 * Force a new page to be allocated for modules.
6205	 */
6206	if (!mod) {
6207		WARN_ON(ftrace_pages || ftrace_pages_start);
6208		/* First initialization */
6209		ftrace_pages = ftrace_pages_start = start_pg;
6210	} else {
6211		if (!ftrace_pages)
6212			goto out;
6213
6214		if (WARN_ON(ftrace_pages->next)) {
6215			/* Hmm, we have free pages? */
6216			while (ftrace_pages->next)
6217				ftrace_pages = ftrace_pages->next;
6218		}
6219
6220		ftrace_pages->next = start_pg;
6221	}
6222
6223	p = start;
6224	pg = start_pg;
6225	while (p < end) {
6226		unsigned long end_offset;
6227		addr = ftrace_call_adjust(*p++);
6228		/*
6229		 * Some architecture linkers will pad between
6230		 * the different mcount_loc sections of different
6231		 * object files to satisfy alignments.
6232		 * Skip any NULL pointers.
6233		 */
6234		if (!addr)
6235			continue;
6236
6237		end_offset = (pg->index+1) * sizeof(pg->records[0]);
6238		if (end_offset > PAGE_SIZE << pg->order) {
6239			/* We should have allocated enough */
6240			if (WARN_ON(!pg->next))
6241				break;
6242			pg = pg->next;
6243		}
6244
6245		rec = &pg->records[pg->index++];
6246		rec->ip = addr;
6247	}
6248
6249	/* We should have used all pages */
6250	WARN_ON(pg->next);
6251
6252	/* Assign the last page to ftrace_pages */
6253	ftrace_pages = pg;
6254
6255	/*
6256	 * We only need to disable interrupts on start up
6257	 * because we are modifying code that an interrupt
6258	 * may execute, and the modification is not atomic.
6259	 * But for modules, nothing runs the code we modify
6260	 * until we are finished with it, and there's no
6261	 * reason to cause large interrupt latencies while we do it.
6262	 */
6263	if (!mod)
6264		local_irq_save(flags);
6265	ftrace_update_code(mod, start_pg);
6266	if (!mod)
6267		local_irq_restore(flags);
6268	ret = 0;
6269 out:
6270	mutex_unlock(&ftrace_lock);
6271
6272	return ret;
6273}
6274
6275struct ftrace_mod_func {
6276	struct list_head	list;
6277	char			*name;
6278	unsigned long		ip;
6279	unsigned int		size;
6280};
6281
6282struct ftrace_mod_map {
6283	struct rcu_head		rcu;
6284	struct list_head	list;
6285	struct module		*mod;
6286	unsigned long		start_addr;
6287	unsigned long		end_addr;
6288	struct list_head	funcs;
6289	unsigned int		num_funcs;
6290};
6291
6292static int ftrace_get_trampoline_kallsym(unsigned int symnum,
6293					 unsigned long *value, char *type,
6294					 char *name, char *module_name,
6295					 int *exported)
6296{
6297	struct ftrace_ops *op;
6298
6299	list_for_each_entry_rcu(op, &ftrace_ops_trampoline_list, list) {
6300		if (!op->trampoline || symnum--)
6301			continue;
6302		*value = op->trampoline;
6303		*type = 't';
6304		strlcpy(name, FTRACE_TRAMPOLINE_SYM, KSYM_NAME_LEN);
6305		strlcpy(module_name, FTRACE_TRAMPOLINE_MOD, MODULE_NAME_LEN);
6306		*exported = 0;
6307		return 0;
6308	}
6309
6310	return -ERANGE;
6311}
6312
6313#ifdef CONFIG_MODULES
6314
6315#define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
6316
6317static LIST_HEAD(ftrace_mod_maps);
6318
6319static int referenced_filters(struct dyn_ftrace *rec)
6320{
6321	struct ftrace_ops *ops;
6322	int cnt = 0;
6323
6324	for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
6325		if (ops_references_rec(ops, rec)) {
6326			if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_DIRECT))
6327				continue;
6328			if (WARN_ON_ONCE(ops->flags & FTRACE_OPS_FL_IPMODIFY))
6329				continue;
6330			cnt++;
6331			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
6332				rec->flags |= FTRACE_FL_REGS;
6333			if (cnt == 1 && ops->trampoline)
6334				rec->flags |= FTRACE_FL_TRAMP;
6335			else
6336				rec->flags &= ~FTRACE_FL_TRAMP;
6337		}
6338	}
6339
6340	return cnt;
6341}
6342
6343static void
6344clear_mod_from_hash(struct ftrace_page *pg, struct ftrace_hash *hash)
6345{
6346	struct ftrace_func_entry *entry;
6347	struct dyn_ftrace *rec;
6348	int i;
6349
6350	if (ftrace_hash_empty(hash))
6351		return;
6352
6353	for (i = 0; i < pg->index; i++) {
6354		rec = &pg->records[i];
6355		entry = __ftrace_lookup_ip(hash, rec->ip);
6356		/*
6357		 * Do not allow this rec to match again.
6358		 * Yeah, it may waste some memory, but will be removed
6359		 * if/when the hash is modified again.
6360		 */
6361		if (entry)
6362			entry->ip = 0;
6363	}
6364}
6365
6366/* Clear any records from hashes */
6367static void clear_mod_from_hashes(struct ftrace_page *pg)
6368{
6369	struct trace_array *tr;
6370
6371	mutex_lock(&trace_types_lock);
6372	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6373		if (!tr->ops || !tr->ops->func_hash)
6374			continue;
6375		mutex_lock(&tr->ops->func_hash->regex_lock);
6376		clear_mod_from_hash(pg, tr->ops->func_hash->filter_hash);
6377		clear_mod_from_hash(pg, tr->ops->func_hash->notrace_hash);
6378		mutex_unlock(&tr->ops->func_hash->regex_lock);
6379	}
6380	mutex_unlock(&trace_types_lock);
6381}
6382
6383static void ftrace_free_mod_map(struct rcu_head *rcu)
6384{
6385	struct ftrace_mod_map *mod_map = container_of(rcu, struct ftrace_mod_map, rcu);
6386	struct ftrace_mod_func *mod_func;
6387	struct ftrace_mod_func *n;
6388
6389	/* All the contents of mod_map are now not visible to readers */
6390	list_for_each_entry_safe(mod_func, n, &mod_map->funcs, list) {
6391		kfree(mod_func->name);
6392		list_del(&mod_func->list);
6393		kfree(mod_func);
6394	}
6395
6396	kfree(mod_map);
6397}
6398
6399void ftrace_release_mod(struct module *mod)
6400{
6401	struct ftrace_mod_map *mod_map;
6402	struct ftrace_mod_map *n;
6403	struct dyn_ftrace *rec;
6404	struct ftrace_page **last_pg;
6405	struct ftrace_page *tmp_page = NULL;
6406	struct ftrace_page *pg;
 
6407
6408	mutex_lock(&ftrace_lock);
6409
6410	if (ftrace_disabled)
6411		goto out_unlock;
6412
6413	list_for_each_entry_safe(mod_map, n, &ftrace_mod_maps, list) {
6414		if (mod_map->mod == mod) {
6415			list_del_rcu(&mod_map->list);
6416			call_rcu(&mod_map->rcu, ftrace_free_mod_map);
6417			break;
6418		}
6419	}
6420
6421	/*
6422	 * Each module has its own ftrace_pages, remove
6423	 * them from the list.
6424	 */
6425	last_pg = &ftrace_pages_start;
6426	for (pg = ftrace_pages_start; pg; pg = *last_pg) {
6427		rec = &pg->records[0];
6428		if (within_module_core(rec->ip, mod) ||
6429		    within_module_init(rec->ip, mod)) {
6430			/*
6431			 * As core pages are first, the first
6432			 * page should never be a module page.
6433			 */
6434			if (WARN_ON(pg == ftrace_pages_start))
6435				goto out_unlock;
6436
6437			/* Check if we are deleting the last page */
6438			if (pg == ftrace_pages)
6439				ftrace_pages = next_to_ftrace_page(last_pg);
6440
6441			ftrace_update_tot_cnt -= pg->index;
6442			*last_pg = pg->next;
6443
6444			pg->next = tmp_page;
6445			tmp_page = pg;
6446		} else
6447			last_pg = &pg->next;
6448	}
6449 out_unlock:
6450	mutex_unlock(&ftrace_lock);
6451
6452	for (pg = tmp_page; pg; pg = tmp_page) {
6453
6454		/* Needs to be called outside of ftrace_lock */
6455		clear_mod_from_hashes(pg);
6456
6457		if (pg->records) {
6458			free_pages((unsigned long)pg->records, pg->order);
6459			ftrace_number_of_pages -= 1 << pg->order;
6460		}
6461		tmp_page = pg->next;
6462		kfree(pg);
 
6463		ftrace_number_of_groups--;
6464	}
6465}
6466
6467void ftrace_module_enable(struct module *mod)
6468{
6469	struct dyn_ftrace *rec;
6470	struct ftrace_page *pg;
6471
6472	mutex_lock(&ftrace_lock);
6473
6474	if (ftrace_disabled)
6475		goto out_unlock;
6476
6477	/*
6478	 * If the tracing is enabled, go ahead and enable the record.
6479	 *
6480	 * The reason not to enable the record immediately is the
6481	 * inherent check of ftrace_make_nop/ftrace_make_call for
6482	 * correct previous instructions.  Making first the NOP
6483	 * conversion puts the module to the correct state, thus
6484	 * passing the ftrace_make_call check.
6485	 *
6486	 * We also delay this to after the module code already set the
6487	 * text to read-only, as we now need to set it back to read-write
6488	 * so that we can modify the text.
6489	 */
6490	if (ftrace_start_up)
6491		ftrace_arch_code_modify_prepare();
6492
6493	do_for_each_ftrace_rec(pg, rec) {
6494		int cnt;
6495		/*
6496		 * do_for_each_ftrace_rec() is a double loop.
6497		 * module text shares the pg. If a record is
6498		 * not part of this module, then skip this pg,
6499		 * which the "break" will do.
6500		 */
6501		if (!within_module_core(rec->ip, mod) &&
6502		    !within_module_init(rec->ip, mod))
6503			break;
6504
6505		cnt = 0;
6506
6507		/*
6508		 * When adding a module, we need to check if tracers are
6509		 * currently enabled and if they are, and can trace this record,
6510		 * we need to enable the module functions as well as update the
6511		 * reference counts for those function records.
6512		 */
6513		if (ftrace_start_up)
6514			cnt += referenced_filters(rec);
6515
6516		rec->flags &= ~FTRACE_FL_DISABLED;
6517		rec->flags += cnt;
6518
6519		if (ftrace_start_up && cnt) {
6520			int failed = __ftrace_replace_code(rec, 1);
6521			if (failed) {
6522				ftrace_bug(failed, rec);
6523				goto out_loop;
6524			}
6525		}
6526
6527	} while_for_each_ftrace_rec();
6528
6529 out_loop:
6530	if (ftrace_start_up)
6531		ftrace_arch_code_modify_post_process();
6532
6533 out_unlock:
6534	mutex_unlock(&ftrace_lock);
6535
6536	process_cached_mods(mod->name);
6537}
6538
6539void ftrace_module_init(struct module *mod)
6540{
6541	if (ftrace_disabled || !mod->num_ftrace_callsites)
6542		return;
6543
6544	ftrace_process_locs(mod, mod->ftrace_callsites,
6545			    mod->ftrace_callsites + mod->num_ftrace_callsites);
6546}
6547
6548static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
6549				struct dyn_ftrace *rec)
6550{
6551	struct ftrace_mod_func *mod_func;
6552	unsigned long symsize;
6553	unsigned long offset;
6554	char str[KSYM_SYMBOL_LEN];
6555	char *modname;
6556	const char *ret;
6557
6558	ret = kallsyms_lookup(rec->ip, &symsize, &offset, &modname, str);
6559	if (!ret)
6560		return;
6561
6562	mod_func = kmalloc(sizeof(*mod_func), GFP_KERNEL);
6563	if (!mod_func)
6564		return;
6565
6566	mod_func->name = kstrdup(str, GFP_KERNEL);
6567	if (!mod_func->name) {
6568		kfree(mod_func);
6569		return;
6570	}
6571
6572	mod_func->ip = rec->ip - offset;
6573	mod_func->size = symsize;
6574
6575	mod_map->num_funcs++;
6576
6577	list_add_rcu(&mod_func->list, &mod_map->funcs);
6578}
6579
6580static struct ftrace_mod_map *
6581allocate_ftrace_mod_map(struct module *mod,
6582			unsigned long start, unsigned long end)
6583{
6584	struct ftrace_mod_map *mod_map;
6585
6586	mod_map = kmalloc(sizeof(*mod_map), GFP_KERNEL);
6587	if (!mod_map)
6588		return NULL;
6589
6590	mod_map->mod = mod;
6591	mod_map->start_addr = start;
6592	mod_map->end_addr = end;
6593	mod_map->num_funcs = 0;
6594
6595	INIT_LIST_HEAD_RCU(&mod_map->funcs);
6596
6597	list_add_rcu(&mod_map->list, &ftrace_mod_maps);
6598
6599	return mod_map;
6600}
6601
6602static const char *
6603ftrace_func_address_lookup(struct ftrace_mod_map *mod_map,
6604			   unsigned long addr, unsigned long *size,
6605			   unsigned long *off, char *sym)
6606{
6607	struct ftrace_mod_func *found_func =  NULL;
6608	struct ftrace_mod_func *mod_func;
6609
6610	list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
6611		if (addr >= mod_func->ip &&
6612		    addr < mod_func->ip + mod_func->size) {
6613			found_func = mod_func;
6614			break;
6615		}
6616	}
6617
6618	if (found_func) {
6619		if (size)
6620			*size = found_func->size;
6621		if (off)
6622			*off = addr - found_func->ip;
6623		if (sym)
6624			strlcpy(sym, found_func->name, KSYM_NAME_LEN);
6625
6626		return found_func->name;
6627	}
6628
6629	return NULL;
6630}
6631
6632const char *
6633ftrace_mod_address_lookup(unsigned long addr, unsigned long *size,
6634		   unsigned long *off, char **modname, char *sym)
6635{
6636	struct ftrace_mod_map *mod_map;
6637	const char *ret = NULL;
6638
6639	/* mod_map is freed via call_rcu() */
6640	preempt_disable();
6641	list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
6642		ret = ftrace_func_address_lookup(mod_map, addr, size, off, sym);
6643		if (ret) {
6644			if (modname)
6645				*modname = mod_map->mod->name;
6646			break;
6647		}
6648	}
6649	preempt_enable();
6650
6651	return ret;
6652}
6653
6654int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
6655			   char *type, char *name,
6656			   char *module_name, int *exported)
6657{
6658	struct ftrace_mod_map *mod_map;
6659	struct ftrace_mod_func *mod_func;
6660	int ret;
6661
6662	preempt_disable();
6663	list_for_each_entry_rcu(mod_map, &ftrace_mod_maps, list) {
6664
6665		if (symnum >= mod_map->num_funcs) {
6666			symnum -= mod_map->num_funcs;
6667			continue;
6668		}
6669
6670		list_for_each_entry_rcu(mod_func, &mod_map->funcs, list) {
6671			if (symnum > 1) {
6672				symnum--;
6673				continue;
6674			}
6675
6676			*value = mod_func->ip;
6677			*type = 'T';
6678			strlcpy(name, mod_func->name, KSYM_NAME_LEN);
6679			strlcpy(module_name, mod_map->mod->name, MODULE_NAME_LEN);
6680			*exported = 1;
6681			preempt_enable();
6682			return 0;
6683		}
6684		WARN_ON(1);
6685		break;
6686	}
6687	ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
6688					    module_name, exported);
6689	preempt_enable();
6690	return ret;
6691}
6692
6693#else
6694static void save_ftrace_mod_rec(struct ftrace_mod_map *mod_map,
6695				struct dyn_ftrace *rec) { }
6696static inline struct ftrace_mod_map *
6697allocate_ftrace_mod_map(struct module *mod,
6698			unsigned long start, unsigned long end)
6699{
6700	return NULL;
6701}
6702int ftrace_mod_get_kallsym(unsigned int symnum, unsigned long *value,
6703			   char *type, char *name, char *module_name,
6704			   int *exported)
6705{
6706	int ret;
6707
6708	preempt_disable();
6709	ret = ftrace_get_trampoline_kallsym(symnum, value, type, name,
6710					    module_name, exported);
6711	preempt_enable();
6712	return ret;
6713}
6714#endif /* CONFIG_MODULES */
6715
6716struct ftrace_init_func {
6717	struct list_head list;
6718	unsigned long ip;
6719};
6720
6721/* Clear any init ips from hashes */
6722static void
6723clear_func_from_hash(struct ftrace_init_func *func, struct ftrace_hash *hash)
6724{
6725	struct ftrace_func_entry *entry;
6726
6727	entry = ftrace_lookup_ip(hash, func->ip);
6728	/*
6729	 * Do not allow this rec to match again.
6730	 * Yeah, it may waste some memory, but will be removed
6731	 * if/when the hash is modified again.
6732	 */
6733	if (entry)
6734		entry->ip = 0;
6735}
6736
6737static void
6738clear_func_from_hashes(struct ftrace_init_func *func)
6739{
6740	struct trace_array *tr;
6741
6742	mutex_lock(&trace_types_lock);
6743	list_for_each_entry(tr, &ftrace_trace_arrays, list) {
6744		if (!tr->ops || !tr->ops->func_hash)
6745			continue;
6746		mutex_lock(&tr->ops->func_hash->regex_lock);
6747		clear_func_from_hash(func, tr->ops->func_hash->filter_hash);
6748		clear_func_from_hash(func, tr->ops->func_hash->notrace_hash);
6749		mutex_unlock(&tr->ops->func_hash->regex_lock);
6750	}
6751	mutex_unlock(&trace_types_lock);
6752}
6753
6754static void add_to_clear_hash_list(struct list_head *clear_list,
6755				   struct dyn_ftrace *rec)
6756{
6757	struct ftrace_init_func *func;
6758
6759	func = kmalloc(sizeof(*func), GFP_KERNEL);
6760	if (!func) {
6761		MEM_FAIL(1, "alloc failure, ftrace filter could be stale\n");
6762		return;
6763	}
6764
6765	func->ip = rec->ip;
6766	list_add(&func->list, clear_list);
6767}
6768
6769void ftrace_free_mem(struct module *mod, void *start_ptr, void *end_ptr)
6770{
6771	unsigned long start = (unsigned long)(start_ptr);
6772	unsigned long end = (unsigned long)(end_ptr);
6773	struct ftrace_page **last_pg = &ftrace_pages_start;
6774	struct ftrace_page *pg;
6775	struct dyn_ftrace *rec;
6776	struct dyn_ftrace key;
6777	struct ftrace_mod_map *mod_map = NULL;
6778	struct ftrace_init_func *func, *func_next;
6779	struct list_head clear_hash;
 
6780
6781	INIT_LIST_HEAD(&clear_hash);
6782
6783	key.ip = start;
6784	key.flags = end;	/* overload flags, as it is unsigned long */
6785
6786	mutex_lock(&ftrace_lock);
6787
6788	/*
6789	 * If we are freeing module init memory, then check if
6790	 * any tracer is active. If so, we need to save a mapping of
6791	 * the module functions being freed with the address.
6792	 */
6793	if (mod && ftrace_ops_list != &ftrace_list_end)
6794		mod_map = allocate_ftrace_mod_map(mod, start, end);
6795
6796	for (pg = ftrace_pages_start; pg; last_pg = &pg->next, pg = *last_pg) {
6797		if (end < pg->records[0].ip ||
6798		    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
6799			continue;
6800 again:
6801		rec = bsearch(&key, pg->records, pg->index,
6802			      sizeof(struct dyn_ftrace),
6803			      ftrace_cmp_recs);
6804		if (!rec)
6805			continue;
6806
6807		/* rec will be cleared from hashes after ftrace_lock unlock */
6808		add_to_clear_hash_list(&clear_hash, rec);
6809
6810		if (mod_map)
6811			save_ftrace_mod_rec(mod_map, rec);
6812
6813		pg->index--;
6814		ftrace_update_tot_cnt--;
6815		if (!pg->index) {
6816			*last_pg = pg->next;
6817			if (pg->records) {
6818				free_pages((unsigned long)pg->records, pg->order);
6819				ftrace_number_of_pages -= 1 << pg->order;
6820			}
6821			ftrace_number_of_groups--;
6822			kfree(pg);
6823			pg = container_of(last_pg, struct ftrace_page, next);
6824			if (!(*last_pg))
6825				ftrace_pages = pg;
6826			continue;
6827		}
6828		memmove(rec, rec + 1,
6829			(pg->index - (rec - pg->records)) * sizeof(*rec));
6830		/* More than one function may be in this block */
6831		goto again;
6832	}
6833	mutex_unlock(&ftrace_lock);
6834
6835	list_for_each_entry_safe(func, func_next, &clear_hash, list) {
6836		clear_func_from_hashes(func);
6837		kfree(func);
6838	}
6839}
6840
6841void __init ftrace_free_init_mem(void)
6842{
6843	void *start = (void *)(&__init_begin);
6844	void *end = (void *)(&__init_end);
6845
6846	ftrace_free_mem(NULL, start, end);
6847}
6848
6849void __init ftrace_init(void)
6850{
6851	extern unsigned long __start_mcount_loc[];
6852	extern unsigned long __stop_mcount_loc[];
6853	unsigned long count, flags;
6854	int ret;
6855
6856	local_irq_save(flags);
6857	ret = ftrace_dyn_arch_init();
6858	local_irq_restore(flags);
6859	if (ret)
6860		goto failed;
6861
6862	count = __stop_mcount_loc - __start_mcount_loc;
6863	if (!count) {
6864		pr_info("ftrace: No functions to be traced?\n");
6865		goto failed;
6866	}
6867
6868	pr_info("ftrace: allocating %ld entries in %ld pages\n",
6869		count, count / ENTRIES_PER_PAGE + 1);
6870
6871	last_ftrace_enabled = ftrace_enabled = 1;
6872
6873	ret = ftrace_process_locs(NULL,
6874				  __start_mcount_loc,
6875				  __stop_mcount_loc);
6876
6877	pr_info("ftrace: allocated %ld pages with %ld groups\n",
6878		ftrace_number_of_pages, ftrace_number_of_groups);
6879
6880	set_ftrace_early_filters();
6881
6882	return;
6883 failed:
6884	ftrace_disabled = 1;
6885}
6886
6887/* Do nothing if arch does not support this */
6888void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
6889{
6890}
6891
6892static void ftrace_update_trampoline(struct ftrace_ops *ops)
6893{
6894	unsigned long trampoline = ops->trampoline;
6895
6896	arch_ftrace_update_trampoline(ops);
6897	if (ops->trampoline && ops->trampoline != trampoline &&
6898	    (ops->flags & FTRACE_OPS_FL_ALLOC_TRAMP)) {
6899		/* Add to kallsyms before the perf events */
6900		ftrace_add_trampoline_to_kallsyms(ops);
6901		perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
6902				   ops->trampoline, ops->trampoline_size, false,
6903				   FTRACE_TRAMPOLINE_SYM);
6904		/*
6905		 * Record the perf text poke event after the ksymbol register
6906		 * event.
6907		 */
6908		perf_event_text_poke((void *)ops->trampoline, NULL, 0,
6909				     (void *)ops->trampoline,
6910				     ops->trampoline_size);
6911	}
6912}
6913
6914void ftrace_init_trace_array(struct trace_array *tr)
6915{
6916	INIT_LIST_HEAD(&tr->func_probes);
6917	INIT_LIST_HEAD(&tr->mod_trace);
6918	INIT_LIST_HEAD(&tr->mod_notrace);
6919}
6920#else
6921
6922struct ftrace_ops global_ops = {
6923	.func			= ftrace_stub,
6924	.flags			= FTRACE_OPS_FL_INITIALIZED |
 
6925				  FTRACE_OPS_FL_PID,
6926};
6927
6928static int __init ftrace_nodyn_init(void)
6929{
6930	ftrace_enabled = 1;
6931	return 0;
6932}
6933core_initcall(ftrace_nodyn_init);
6934
6935static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
6936static inline void ftrace_startup_enable(int command) { }
6937static inline void ftrace_startup_all(int command) { }
6938
6939# define ftrace_startup_sysctl()	do { } while (0)
6940# define ftrace_shutdown_sysctl()	do { } while (0)
6941
6942static void ftrace_update_trampoline(struct ftrace_ops *ops)
6943{
6944}
6945
6946#endif /* CONFIG_DYNAMIC_FTRACE */
6947
6948__init void ftrace_init_global_array_ops(struct trace_array *tr)
6949{
6950	tr->ops = &global_ops;
6951	tr->ops->private = tr;
6952	ftrace_init_trace_array(tr);
6953}
6954
6955void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
6956{
6957	/* If we filter on pids, update to use the pid function */
6958	if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
6959		if (WARN_ON(tr->ops->func != ftrace_stub))
6960			printk("ftrace ops had %pS for function\n",
6961			       tr->ops->func);
6962	}
6963	tr->ops->func = func;
6964	tr->ops->private = tr;
6965}
6966
6967void ftrace_reset_array_ops(struct trace_array *tr)
6968{
6969	tr->ops->func = ftrace_stub;
6970}
6971
6972static nokprobe_inline void
6973__ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
6974		       struct ftrace_ops *ignored, struct ftrace_regs *fregs)
6975{
6976	struct pt_regs *regs = ftrace_get_regs(fregs);
6977	struct ftrace_ops *op;
6978	int bit;
6979
6980	bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
6981	if (bit < 0)
6982		return;
6983
6984	/*
6985	 * Some of the ops may be dynamically allocated,
6986	 * they must be freed after a synchronize_rcu().
6987	 */
6988	preempt_disable_notrace();
6989
6990	do_for_each_ftrace_op(op, ftrace_ops_list) {
6991		/* Stub functions don't need to be called nor tested */
6992		if (op->flags & FTRACE_OPS_FL_STUB)
6993			continue;
6994		/*
6995		 * Check the following for each ops before calling their func:
6996		 *  if RCU flag is set, then rcu_is_watching() must be true
6997		 *  if PER_CPU is set, then ftrace_function_local_disable()
6998		 *                          must be false
6999		 *  Otherwise test if the ip matches the ops filter
7000		 *
7001		 * If any of the above fails then the op->func() is not executed.
7002		 */
7003		if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
7004		    ftrace_ops_test(op, ip, regs)) {
7005			if (FTRACE_WARN_ON(!op->func)) {
7006				pr_warn("op=%p %pS\n", op, op);
7007				goto out;
7008			}
7009			op->func(ip, parent_ip, op, fregs);
7010		}
7011	} while_for_each_ftrace_op(op);
7012out:
7013	preempt_enable_notrace();
7014	trace_clear_recursion(bit);
7015}
7016
7017/*
7018 * Some archs only support passing ip and parent_ip. Even though
7019 * the list function ignores the op parameter, we do not want any
7020 * C side effects, where a function is called without the caller
7021 * sending a third parameter.
7022 * Archs are to support both the regs and ftrace_ops at the same time.
7023 * If they support ftrace_ops, it is assumed they support regs.
7024 * If call backs want to use regs, they must either check for regs
7025 * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
7026 * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
7027 * An architecture can pass partial regs with ftrace_ops and still
7028 * set the ARCH_SUPPORTS_FTRACE_OPS.
7029 */
7030#if ARCH_SUPPORTS_FTRACE_OPS
7031static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
7032				 struct ftrace_ops *op, struct ftrace_regs *fregs)
7033{
7034	__ftrace_ops_list_func(ip, parent_ip, NULL, fregs);
7035}
7036NOKPROBE_SYMBOL(ftrace_ops_list_func);
7037#else
7038static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
7039{
7040	__ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
7041}
7042NOKPROBE_SYMBOL(ftrace_ops_no_ops);
7043#endif
7044
7045/*
7046 * If there's only one function registered but it does not support
7047 * recursion, needs RCU protection and/or requires per cpu handling, then
7048 * this function will be called by the mcount trampoline.
7049 */
7050static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
7051				   struct ftrace_ops *op, struct ftrace_regs *fregs)
7052{
7053	int bit;
7054
7055	bit = trace_test_and_set_recursion(ip, parent_ip, TRACE_LIST_START);
7056	if (bit < 0)
7057		return;
7058
7059	preempt_disable_notrace();
7060
7061	if (!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching())
7062		op->func(ip, parent_ip, op, fregs);
7063
7064	preempt_enable_notrace();
7065	trace_clear_recursion(bit);
7066}
7067NOKPROBE_SYMBOL(ftrace_ops_assist_func);
7068
7069/**
7070 * ftrace_ops_get_func - get the function a trampoline should call
7071 * @ops: the ops to get the function for
7072 *
7073 * Normally the mcount trampoline will call the ops->func, but there
7074 * are times that it should not. For example, if the ops does not
7075 * have its own recursion protection, then it should call the
7076 * ftrace_ops_assist_func() instead.
7077 *
7078 * Returns the function that the trampoline should call for @ops.
7079 */
7080ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
7081{
7082	/*
7083	 * If the function does not handle recursion or needs to be RCU safe,
7084	 * then we need to call the assist handler.
7085	 */
7086	if (ops->flags & (FTRACE_OPS_FL_RECURSION |
7087			  FTRACE_OPS_FL_RCU))
7088		return ftrace_ops_assist_func;
7089
7090	return ops->func;
7091}
7092
7093static void
7094ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
7095		    struct task_struct *prev, struct task_struct *next)
7096{
7097	struct trace_array *tr = data;
7098	struct trace_pid_list *pid_list;
7099	struct trace_pid_list *no_pid_list;
7100
7101	pid_list = rcu_dereference_sched(tr->function_pids);
7102	no_pid_list = rcu_dereference_sched(tr->function_no_pids);
7103
7104	if (trace_ignore_this_task(pid_list, no_pid_list, next))
7105		this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
7106			       FTRACE_PID_IGNORE);
7107	else
7108		this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
7109			       next->pid);
7110}
7111
7112static void
7113ftrace_pid_follow_sched_process_fork(void *data,
7114				     struct task_struct *self,
7115				     struct task_struct *task)
7116{
7117	struct trace_pid_list *pid_list;
7118	struct trace_array *tr = data;
7119
7120	pid_list = rcu_dereference_sched(tr->function_pids);
7121	trace_filter_add_remove_task(pid_list, self, task);
7122
7123	pid_list = rcu_dereference_sched(tr->function_no_pids);
7124	trace_filter_add_remove_task(pid_list, self, task);
7125}
7126
7127static void
7128ftrace_pid_follow_sched_process_exit(void *data, struct task_struct *task)
7129{
7130	struct trace_pid_list *pid_list;
7131	struct trace_array *tr = data;
7132
7133	pid_list = rcu_dereference_sched(tr->function_pids);
7134	trace_filter_add_remove_task(pid_list, NULL, task);
7135
7136	pid_list = rcu_dereference_sched(tr->function_no_pids);
7137	trace_filter_add_remove_task(pid_list, NULL, task);
7138}
7139
7140void ftrace_pid_follow_fork(struct trace_array *tr, bool enable)
7141{
7142	if (enable) {
7143		register_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
7144						  tr);
7145		register_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
7146						  tr);
7147	} else {
7148		unregister_trace_sched_process_fork(ftrace_pid_follow_sched_process_fork,
7149						    tr);
7150		unregister_trace_sched_process_free(ftrace_pid_follow_sched_process_exit,
7151						    tr);
7152	}
7153}
7154
7155static void clear_ftrace_pids(struct trace_array *tr, int type)
7156{
7157	struct trace_pid_list *pid_list;
7158	struct trace_pid_list *no_pid_list;
7159	int cpu;
7160
7161	pid_list = rcu_dereference_protected(tr->function_pids,
7162					     lockdep_is_held(&ftrace_lock));
7163	no_pid_list = rcu_dereference_protected(tr->function_no_pids,
7164						lockdep_is_held(&ftrace_lock));
7165
7166	/* Make sure there's something to do */
7167	if (!pid_type_enabled(type, pid_list, no_pid_list))
7168		return;
7169
7170	/* See if the pids still need to be checked after this */
7171	if (!still_need_pid_events(type, pid_list, no_pid_list)) {
7172		unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
7173		for_each_possible_cpu(cpu)
7174			per_cpu_ptr(tr->array_buffer.data, cpu)->ftrace_ignore_pid = FTRACE_PID_TRACE;
7175	}
7176
7177	if (type & TRACE_PIDS)
7178		rcu_assign_pointer(tr->function_pids, NULL);
7179
7180	if (type & TRACE_NO_PIDS)
7181		rcu_assign_pointer(tr->function_no_pids, NULL);
7182
7183	/* Wait till all users are no longer using pid filtering */
7184	synchronize_rcu();
7185
7186	if ((type & TRACE_PIDS) && pid_list)
7187		trace_free_pid_list(pid_list);
7188
7189	if ((type & TRACE_NO_PIDS) && no_pid_list)
7190		trace_free_pid_list(no_pid_list);
7191}
7192
7193void ftrace_clear_pids(struct trace_array *tr)
7194{
7195	mutex_lock(&ftrace_lock);
7196
7197	clear_ftrace_pids(tr, TRACE_PIDS | TRACE_NO_PIDS);
7198
7199	mutex_unlock(&ftrace_lock);
7200}
7201
7202static void ftrace_pid_reset(struct trace_array *tr, int type)
7203{
7204	mutex_lock(&ftrace_lock);
7205	clear_ftrace_pids(tr, type);
7206
7207	ftrace_update_pid_func();
7208	ftrace_startup_all(0);
7209
7210	mutex_unlock(&ftrace_lock);
7211}
7212
7213/* Greater than any max PID */
7214#define FTRACE_NO_PIDS		(void *)(PID_MAX_LIMIT + 1)
7215
7216static void *fpid_start(struct seq_file *m, loff_t *pos)
7217	__acquires(RCU)
7218{
7219	struct trace_pid_list *pid_list;
7220	struct trace_array *tr = m->private;
7221
7222	mutex_lock(&ftrace_lock);
7223	rcu_read_lock_sched();
7224
7225	pid_list = rcu_dereference_sched(tr->function_pids);
7226
7227	if (!pid_list)
7228		return !(*pos) ? FTRACE_NO_PIDS : NULL;
7229
7230	return trace_pid_start(pid_list, pos);
7231}
7232
7233static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
7234{
7235	struct trace_array *tr = m->private;
7236	struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
7237
7238	if (v == FTRACE_NO_PIDS) {
7239		(*pos)++;
7240		return NULL;
7241	}
7242	return trace_pid_next(pid_list, v, pos);
7243}
7244
7245static void fpid_stop(struct seq_file *m, void *p)
7246	__releases(RCU)
7247{
7248	rcu_read_unlock_sched();
7249	mutex_unlock(&ftrace_lock);
7250}
7251
7252static int fpid_show(struct seq_file *m, void *v)
7253{
7254	if (v == FTRACE_NO_PIDS) {
7255		seq_puts(m, "no pid\n");
7256		return 0;
7257	}
7258
7259	return trace_pid_show(m, v);
7260}
7261
7262static const struct seq_operations ftrace_pid_sops = {
7263	.start = fpid_start,
7264	.next = fpid_next,
7265	.stop = fpid_stop,
7266	.show = fpid_show,
7267};
7268
7269static void *fnpid_start(struct seq_file *m, loff_t *pos)
7270	__acquires(RCU)
7271{
7272	struct trace_pid_list *pid_list;
7273	struct trace_array *tr = m->private;
7274
7275	mutex_lock(&ftrace_lock);
7276	rcu_read_lock_sched();
7277
7278	pid_list = rcu_dereference_sched(tr->function_no_pids);
7279
7280	if (!pid_list)
7281		return !(*pos) ? FTRACE_NO_PIDS : NULL;
7282
7283	return trace_pid_start(pid_list, pos);
7284}
7285
7286static void *fnpid_next(struct seq_file *m, void *v, loff_t *pos)
7287{
7288	struct trace_array *tr = m->private;
7289	struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_no_pids);
7290
7291	if (v == FTRACE_NO_PIDS) {
7292		(*pos)++;
7293		return NULL;
7294	}
7295	return trace_pid_next(pid_list, v, pos);
7296}
7297
7298static const struct seq_operations ftrace_no_pid_sops = {
7299	.start = fnpid_start,
7300	.next = fnpid_next,
7301	.stop = fpid_stop,
7302	.show = fpid_show,
7303};
7304
7305static int pid_open(struct inode *inode, struct file *file, int type)
7306{
7307	const struct seq_operations *seq_ops;
7308	struct trace_array *tr = inode->i_private;
7309	struct seq_file *m;
7310	int ret = 0;
7311
7312	ret = tracing_check_open_get_tr(tr);
7313	if (ret)
7314		return ret;
7315
7316	if ((file->f_mode & FMODE_WRITE) &&
7317	    (file->f_flags & O_TRUNC))
7318		ftrace_pid_reset(tr, type);
7319
7320	switch (type) {
7321	case TRACE_PIDS:
7322		seq_ops = &ftrace_pid_sops;
7323		break;
7324	case TRACE_NO_PIDS:
7325		seq_ops = &ftrace_no_pid_sops;
7326		break;
7327	default:
7328		trace_array_put(tr);
7329		WARN_ON_ONCE(1);
7330		return -EINVAL;
7331	}
7332
7333	ret = seq_open(file, seq_ops);
7334	if (ret < 0) {
7335		trace_array_put(tr);
7336	} else {
7337		m = file->private_data;
7338		/* copy tr over to seq ops */
7339		m->private = tr;
7340	}
7341
7342	return ret;
7343}
7344
7345static int
7346ftrace_pid_open(struct inode *inode, struct file *file)
7347{
7348	return pid_open(inode, file, TRACE_PIDS);
7349}
7350
7351static int
7352ftrace_no_pid_open(struct inode *inode, struct file *file)
7353{
7354	return pid_open(inode, file, TRACE_NO_PIDS);
7355}
7356
7357static void ignore_task_cpu(void *data)
7358{
7359	struct trace_array *tr = data;
7360	struct trace_pid_list *pid_list;
7361	struct trace_pid_list *no_pid_list;
7362
7363	/*
7364	 * This function is called by on_each_cpu() while the
7365	 * event_mutex is held.
7366	 */
7367	pid_list = rcu_dereference_protected(tr->function_pids,
7368					     mutex_is_locked(&ftrace_lock));
7369	no_pid_list = rcu_dereference_protected(tr->function_no_pids,
7370						mutex_is_locked(&ftrace_lock));
7371
7372	if (trace_ignore_this_task(pid_list, no_pid_list, current))
7373		this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
7374			       FTRACE_PID_IGNORE);
7375	else
7376		this_cpu_write(tr->array_buffer.data->ftrace_ignore_pid,
7377			       current->pid);
7378}
7379
7380static ssize_t
7381pid_write(struct file *filp, const char __user *ubuf,
7382	  size_t cnt, loff_t *ppos, int type)
7383{
7384	struct seq_file *m = filp->private_data;
7385	struct trace_array *tr = m->private;
7386	struct trace_pid_list *filtered_pids;
7387	struct trace_pid_list *other_pids;
7388	struct trace_pid_list *pid_list;
7389	ssize_t ret;
7390
7391	if (!cnt)
7392		return 0;
7393
7394	mutex_lock(&ftrace_lock);
7395
7396	switch (type) {
7397	case TRACE_PIDS:
7398		filtered_pids = rcu_dereference_protected(tr->function_pids,
7399					     lockdep_is_held(&ftrace_lock));
7400		other_pids = rcu_dereference_protected(tr->function_no_pids,
7401					     lockdep_is_held(&ftrace_lock));
7402		break;
7403	case TRACE_NO_PIDS:
7404		filtered_pids = rcu_dereference_protected(tr->function_no_pids,
7405					     lockdep_is_held(&ftrace_lock));
7406		other_pids = rcu_dereference_protected(tr->function_pids,
7407					     lockdep_is_held(&ftrace_lock));
7408		break;
7409	default:
7410		ret = -EINVAL;
7411		WARN_ON_ONCE(1);
7412		goto out;
7413	}
7414
7415	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
7416	if (ret < 0)
7417		goto out;
7418
7419	switch (type) {
7420	case TRACE_PIDS:
7421		rcu_assign_pointer(tr->function_pids, pid_list);
7422		break;
7423	case TRACE_NO_PIDS:
7424		rcu_assign_pointer(tr->function_no_pids, pid_list);
7425		break;
7426	}
7427
7428
7429	if (filtered_pids) {
7430		synchronize_rcu();
7431		trace_free_pid_list(filtered_pids);
7432	} else if (pid_list && !other_pids) {
7433		/* Register a probe to set whether to ignore the tracing of a task */
7434		register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
7435	}
7436
7437	/*
7438	 * Ignoring of pids is done at task switch. But we have to
7439	 * check for those tasks that are currently running.
7440	 * Always do this in case a pid was appended or removed.
7441	 */
7442	on_each_cpu(ignore_task_cpu, tr, 1);
7443
7444	ftrace_update_pid_func();
7445	ftrace_startup_all(0);
7446 out:
7447	mutex_unlock(&ftrace_lock);
7448
7449	if (ret > 0)
7450		*ppos += ret;
7451
7452	return ret;
7453}
7454
7455static ssize_t
7456ftrace_pid_write(struct file *filp, const char __user *ubuf,
7457		 size_t cnt, loff_t *ppos)
7458{
7459	return pid_write(filp, ubuf, cnt, ppos, TRACE_PIDS);
7460}
7461
7462static ssize_t
7463ftrace_no_pid_write(struct file *filp, const char __user *ubuf,
7464		    size_t cnt, loff_t *ppos)
7465{
7466	return pid_write(filp, ubuf, cnt, ppos, TRACE_NO_PIDS);
7467}
7468
7469static int
7470ftrace_pid_release(struct inode *inode, struct file *file)
7471{
7472	struct trace_array *tr = inode->i_private;
7473
7474	trace_array_put(tr);
7475
7476	return seq_release(inode, file);
7477}
7478
7479static const struct file_operations ftrace_pid_fops = {
7480	.open		= ftrace_pid_open,
7481	.write		= ftrace_pid_write,
7482	.read		= seq_read,
7483	.llseek		= tracing_lseek,
7484	.release	= ftrace_pid_release,
7485};
7486
7487static const struct file_operations ftrace_no_pid_fops = {
7488	.open		= ftrace_no_pid_open,
7489	.write		= ftrace_no_pid_write,
7490	.read		= seq_read,
7491	.llseek		= tracing_lseek,
7492	.release	= ftrace_pid_release,
7493};
7494
7495void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
7496{
7497	trace_create_file("set_ftrace_pid", 0644, d_tracer,
7498			    tr, &ftrace_pid_fops);
7499	trace_create_file("set_ftrace_notrace_pid", 0644, d_tracer,
7500			    tr, &ftrace_no_pid_fops);
7501}
7502
7503void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
7504					 struct dentry *d_tracer)
7505{
7506	/* Only the top level directory has the dyn_tracefs and profile */
7507	WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
7508
7509	ftrace_init_dyn_tracefs(d_tracer);
7510	ftrace_profile_tracefs(d_tracer);
7511}
7512
7513/**
7514 * ftrace_kill - kill ftrace
7515 *
7516 * This function should be used by panic code. It stops ftrace
7517 * but in a not so nice way. If you need to simply kill ftrace
7518 * from a non-atomic section, use ftrace_kill.
7519 */
7520void ftrace_kill(void)
7521{
7522	ftrace_disabled = 1;
7523	ftrace_enabled = 0;
7524	ftrace_trace_function = ftrace_stub;
7525}
7526
7527/**
7528 * Test if ftrace is dead or not.
7529 */
7530int ftrace_is_dead(void)
7531{
7532	return ftrace_disabled;
7533}
7534
7535/**
7536 * register_ftrace_function - register a function for profiling
7537 * @ops - ops structure that holds the function for profiling.
7538 *
7539 * Register a function to be called by all functions in the
7540 * kernel.
7541 *
7542 * Note: @ops->func and all the functions it calls must be labeled
7543 *       with "notrace", otherwise it will go into a
7544 *       recursive loop.
7545 */
7546int register_ftrace_function(struct ftrace_ops *ops)
7547{
7548	int ret;
7549
7550	ftrace_ops_init(ops);
7551
7552	mutex_lock(&ftrace_lock);
7553
7554	ret = ftrace_startup(ops, 0);
7555
7556	mutex_unlock(&ftrace_lock);
7557
7558	return ret;
7559}
7560EXPORT_SYMBOL_GPL(register_ftrace_function);
7561
7562/**
7563 * unregister_ftrace_function - unregister a function for profiling.
7564 * @ops - ops structure that holds the function to unregister
7565 *
7566 * Unregister a function that was added to be called by ftrace profiling.
7567 */
7568int unregister_ftrace_function(struct ftrace_ops *ops)
7569{
7570	int ret;
7571
7572	mutex_lock(&ftrace_lock);
7573	ret = ftrace_shutdown(ops, 0);
7574	mutex_unlock(&ftrace_lock);
7575
7576	return ret;
7577}
7578EXPORT_SYMBOL_GPL(unregister_ftrace_function);
7579
7580static bool is_permanent_ops_registered(void)
7581{
7582	struct ftrace_ops *op;
7583
7584	do_for_each_ftrace_op(op, ftrace_ops_list) {
7585		if (op->flags & FTRACE_OPS_FL_PERMANENT)
7586			return true;
7587	} while_for_each_ftrace_op(op);
7588
7589	return false;
7590}
7591
7592int
7593ftrace_enable_sysctl(struct ctl_table *table, int write,
7594		     void *buffer, size_t *lenp, loff_t *ppos)
7595{
7596	int ret = -ENODEV;
7597
7598	mutex_lock(&ftrace_lock);
7599
7600	if (unlikely(ftrace_disabled))
7601		goto out;
7602
7603	ret = proc_dointvec(table, write, buffer, lenp, ppos);
7604
7605	if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
7606		goto out;
7607
7608	if (ftrace_enabled) {
7609
7610		/* we are starting ftrace again */
7611		if (rcu_dereference_protected(ftrace_ops_list,
7612			lockdep_is_held(&ftrace_lock)) != &ftrace_list_end)
7613			update_ftrace_function();
7614
7615		ftrace_startup_sysctl();
7616
7617	} else {
7618		if (is_permanent_ops_registered()) {
7619			ftrace_enabled = true;
7620			ret = -EBUSY;
7621			goto out;
7622		}
7623
7624		/* stopping ftrace calls (just send to ftrace_stub) */
7625		ftrace_trace_function = ftrace_stub;
7626
7627		ftrace_shutdown_sysctl();
7628	}
7629
7630	last_ftrace_enabled = !!ftrace_enabled;
7631 out:
7632	mutex_unlock(&ftrace_lock);
7633	return ret;
7634}