perf_event.h 23.8 KB
Newer Older
T
Thomas Gleixner 已提交
1
/*
I
Ingo Molnar 已提交
2
 * Performance events:
T
Thomas Gleixner 已提交
3
 *
I
Ingo Molnar 已提交
4
 *    Copyright (C) 2008-2009, Thomas Gleixner <tglx@linutronix.de>
5 6
 *    Copyright (C) 2008-2011, Red Hat, Inc., Ingo Molnar
 *    Copyright (C) 2008-2011, Red Hat, Inc., Peter Zijlstra
T
Thomas Gleixner 已提交
7
 *
I
Ingo Molnar 已提交
8
 * Data type definitions, declarations, prototypes.
T
Thomas Gleixner 已提交
9
 *
I
Ingo Molnar 已提交
10
 *    Started by: Thomas Gleixner and Ingo Molnar
T
Thomas Gleixner 已提交
11
 *
I
Ingo Molnar 已提交
12
 * For licencing details see kernel-base/COPYING
T
Thomas Gleixner 已提交
13
 */
14 15
#ifndef _LINUX_PERF_EVENT_H
#define _LINUX_PERF_EVENT_H
T
Thomas Gleixner 已提交
16

17
#include <uapi/linux/perf_event.h>
T
Thomas Gleixner 已提交
18

I
Ingo Molnar 已提交
19
/*
20
 * Kernel-internal data types and definitions:
I
Ingo Molnar 已提交
21 22
 */

23 24
#ifdef CONFIG_PERF_EVENTS
# include <asm/perf_event.h>
25
# include <asm/local64.h>
26 27
#endif

28
struct perf_guest_info_callbacks {
29 30 31
	int				(*is_in_guest)(void);
	int				(*is_user_mode)(void);
	unsigned long			(*get_guest_ip)(void);
32 33
};

34 35 36 37
#ifdef CONFIG_HAVE_HW_BREAKPOINT
#include <asm/hw_breakpoint.h>
#endif

38 39 40 41 42
#include <linux/list.h>
#include <linux/mutex.h>
#include <linux/rculist.h>
#include <linux/rcupdate.h>
#include <linux/spinlock.h>
43
#include <linux/hrtimer.h>
P
Peter Zijlstra 已提交
44
#include <linux/fs.h>
45
#include <linux/pid_namespace.h>
46
#include <linux/workqueue.h>
47
#include <linux/ftrace.h>
48
#include <linux/cpu.h>
49
#include <linux/irq_work.h>
50
#include <linux/static_key.h>
A
Arun Sharma 已提交
51
#include <linux/atomic.h>
52
#include <linux/sysfs.h>
53
#include <linux/perf_regs.h>
54
#include <asm/local.h>
55

56 57 58 59 60
struct perf_callchain_entry {
	__u64				nr;
	__u64				ip[PERF_MAX_STACK_DEPTH];
};

61 62 63
struct perf_raw_record {
	u32				size;
	void				*data;
64 65
};

66 67 68 69 70 71 72 73 74 75
/*
 * single taken branch record layout:
 *
 *      from: source instruction (may not always be a branch insn)
 *        to: branch target
 *   mispred: branch target was mispredicted
 * predicted: branch target was predicted
 *
 * support for mispred, predicted is optional. In case it
 * is not supported mispred = predicted = 0.
76 77 78
 *
 *     in_tx: running in a hardware transaction
 *     abort: aborting a hardware transaction
79
 */
80
struct perf_branch_entry {
81 82 83 84
	__u64	from;
	__u64	to;
	__u64	mispred:1,  /* target mispredicted */
		predicted:1,/* target predicted */
85 86 87
		in_tx:1,    /* in transaction */
		abort:1,    /* transaction abort */
		reserved:60;
88 89
};

90 91 92 93 94 95 96 97 98
/*
 * branch stack layout:
 *  nr: number of taken branches stored in entries[]
 *
 * Note that nr can vary from sample to sample
 * branches (to, from) are stored from most recent
 * to least recent, i.e., entries[0] contains the most
 * recent branch.
 */
99 100 101 102 103
struct perf_branch_stack {
	__u64				nr;
	struct perf_branch_entry	entries[0];
};

104 105 106 107 108
struct perf_regs_user {
	__u64		abi;
	struct pt_regs	*regs;
};

109 110
struct task_struct;

111 112 113 114 115 116 117 118 119 120
/*
 * extra PMU register associated with an event
 */
struct hw_perf_event_extra {
	u64		config;	/* register value */
	unsigned int	reg;	/* register address or index */
	int		alloc;	/* extra register already allocated */
	int		idx;	/* index in shared_regs->regs[] */
};

121 122
struct event_constraint;

T
Thomas Gleixner 已提交
123
/**
124
 * struct hw_perf_event - performance event hardware details:
T
Thomas Gleixner 已提交
125
 */
126 127
struct hw_perf_event {
#ifdef CONFIG_PERF_EVENTS
128 129
	union {
		struct { /* hardware */
I
Ingo Molnar 已提交
130
			u64		config;
131
			u64		last_tag;
I
Ingo Molnar 已提交
132
			unsigned long	config_base;
133
			unsigned long	event_base;
134
			int		event_base_rdpmc;
I
Ingo Molnar 已提交
135
			int		idx;
136
			int		last_cpu;
137
			int		flags;
138

139
			struct hw_perf_event_extra extra_reg;
140
			struct hw_perf_event_extra branch_reg;
141 142

			struct event_constraint *constraint;
143
		};
144
		struct { /* software */
I
Ingo Molnar 已提交
145
			struct hrtimer	hrtimer;
146
		};
147 148 149 150 151
		struct { /* tracepoint */
			struct task_struct	*tp_target;
			/* for tp_event->class */
			struct list_head	tp_list;
		};
152
#ifdef CONFIG_HAVE_HW_BREAKPOINT
153
		struct { /* breakpoint */
154 155 156 157 158 159
			/*
			 * Crufty hack to avoid the chicken and egg
			 * problem hw_breakpoint has with context
			 * creation and event initalization.
			 */
			struct task_struct		*bp_target;
160 161
			struct arch_hw_breakpoint	info;
			struct list_head		bp_list;
162
		};
163
#endif
164
	};
P
Peter Zijlstra 已提交
165
	int				state;
166
	local64_t			prev_count;
167
	u64				sample_period;
168
	u64				last_period;
169
	local64_t			period_left;
170
	u64                             interrupts_seq;
171
	u64				interrupts;
172

173 174
	u64				freq_time_stamp;
	u64				freq_count_stamp;
175
#endif
T
Thomas Gleixner 已提交
176 177
};

P
Peter Zijlstra 已提交
178 179 180 181 182 183 184
/*
 * hw_perf_event::state flags
 */
#define PERF_HES_STOPPED	0x01 /* the counter is stopped */
#define PERF_HES_UPTODATE	0x02 /* event->count up-to-date */
#define PERF_HES_ARCH		0x04

185
struct perf_event;
I
Ingo Molnar 已提交
186

187 188 189 190
/*
 * Common implementation detail of pmu::{start,commit,cancel}_txn
 */
#define PERF_EVENT_TXN 0x1
191

I
Ingo Molnar 已提交
192
/**
193
 * struct pmu - generic performance monitoring unit
I
Ingo Molnar 已提交
194
 */
195
struct pmu {
196 197
	struct list_head		entry;

P
Peter Zijlstra 已提交
198
	struct device			*dev;
199
	const struct attribute_group	**attr_groups;
200
	const char			*name;
P
Peter Zijlstra 已提交
201 202
	int				type;

P
Peter Zijlstra 已提交
203 204
	int * __percpu			pmu_disable_count;
	struct perf_cpu_context * __percpu pmu_cpu_context;
P
Peter Zijlstra 已提交
205
	int				task_ctx_nr;
206
	int				hrtimer_interval_ms;
207 208

	/*
P
Peter Zijlstra 已提交
209 210
	 * Fully disable/enable this PMU, can be used to protect from the PMI
	 * as well as for lazy/batch writing of the MSRs.
211
	 */
P
Peter Zijlstra 已提交
212 213
	void (*pmu_enable)		(struct pmu *pmu); /* optional */
	void (*pmu_disable)		(struct pmu *pmu); /* optional */
214

215
	/*
P
Peter Zijlstra 已提交
216
	 * Try and initialize the event for this PMU.
217
	 * Should return -ENOENT when the @event doesn't match this PMU.
218
	 */
219 220
	int (*event_init)		(struct perf_event *event);

P
Peter Zijlstra 已提交
221 222 223 224
#define PERF_EF_START	0x01		/* start the counter when adding    */
#define PERF_EF_RELOAD	0x02		/* reload the counter when starting */
#define PERF_EF_UPDATE	0x04		/* update the counter when stopping */

225
	/*
P
Peter Zijlstra 已提交
226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242
	 * Adds/Removes a counter to/from the PMU, can be done inside
	 * a transaction, see the ->*_txn() methods.
	 */
	int  (*add)			(struct perf_event *event, int flags);
	void (*del)			(struct perf_event *event, int flags);

	/*
	 * Starts/Stops a counter present on the PMU. The PMI handler
	 * should stop the counter when perf_event_overflow() returns
	 * !0. ->start() will be used to continue.
	 */
	void (*start)			(struct perf_event *event, int flags);
	void (*stop)			(struct perf_event *event, int flags);

	/*
	 * Updates the counter value of the event.
	 */
243
	void (*read)			(struct perf_event *event);
244 245

	/*
246 247 248
	 * Group events scheduling is treated as a transaction, add
	 * group events as a whole and perform one schedulability test.
	 * If the test fails, roll back the whole group
P
Peter Zijlstra 已提交
249 250
	 *
	 * Start the transaction, after this ->add() doesn't need to
251
	 * do schedulability tests.
252
	 */
253
	void (*start_txn)		(struct pmu *pmu); /* optional */
254
	/*
P
Peter Zijlstra 已提交
255
	 * If ->start_txn() disabled the ->add() schedulability test
256 257 258 259
	 * then ->commit_txn() is required to perform one. On success
	 * the transaction is closed. On error the transaction is kept
	 * open until ->cancel_txn() is called.
	 */
260
	int  (*commit_txn)		(struct pmu *pmu); /* optional */
261
	/*
P
Peter Zijlstra 已提交
262
	 * Will cancel the transaction, assumes ->del() is called
L
Lucas De Marchi 已提交
263
	 * for each successful ->add() during the transaction.
264
	 */
265
	void (*cancel_txn)		(struct pmu *pmu); /* optional */
266 267 268 269 270 271

	/*
	 * Will return the value for perf_event_mmap_page::index for this event,
	 * if no implementation is provided it will default to: event->hw.idx + 1.
	 */
	int (*event_idx)		(struct perf_event *event); /*optional */
272 273 274 275 276

	/*
	 * flush branch stack on context-switches (needed in cpu-wide mode)
	 */
	void (*flush_branch_stack)	(void);
I
Ingo Molnar 已提交
277 278
};

279
/**
280
 * enum perf_event_active_state - the states of a event
281
 */
282
enum perf_event_active_state {
I
Ingo Molnar 已提交
283
	PERF_EVENT_STATE_ERROR		= -2,
284 285
	PERF_EVENT_STATE_OFF		= -1,
	PERF_EVENT_STATE_INACTIVE	=  0,
I
Ingo Molnar 已提交
286
	PERF_EVENT_STATE_ACTIVE		=  1,
287 288
};

289
struct file;
290 291
struct perf_sample_data;

292
typedef void (*perf_overflow_handler_t)(struct perf_event *,
293 294 295
					struct perf_sample_data *,
					struct pt_regs *regs);

296
enum perf_group_flag {
297
	PERF_GROUP_SOFTWARE		= 0x1,
298 299
};

300 301
#define SWEVENT_HLIST_BITS		8
#define SWEVENT_HLIST_SIZE		(1 << SWEVENT_HLIST_BITS)
302 303

struct swevent_hlist {
304 305
	struct hlist_head		heads[SWEVENT_HLIST_SIZE];
	struct rcu_head			rcu_head;
306 307
};

308 309
#define PERF_ATTACH_CONTEXT	0x01
#define PERF_ATTACH_GROUP	0x02
310
#define PERF_ATTACH_TASK	0x04
311

312
struct perf_cgroup;
313 314
struct ring_buffer;

T
Thomas Gleixner 已提交
315
/**
316
 * struct perf_event - performance event kernel representation:
T
Thomas Gleixner 已提交
317
 */
318 319
struct perf_event {
#ifdef CONFIG_PERF_EVENTS
320
	struct list_head		group_entry;
P
Peter Zijlstra 已提交
321
	struct list_head		event_entry;
322
	struct list_head		sibling_list;
323
	struct hlist_node		hlist_entry;
I
Ingo Molnar 已提交
324
	int				nr_siblings;
325
	int				group_flags;
326
	struct perf_event		*group_leader;
P
Peter Zijlstra 已提交
327
	struct pmu			*pmu;
328

329
	enum perf_event_active_state	state;
330
	unsigned int			attach_state;
331
	local64_t			count;
332
	atomic64_t			child_count;
333

334
	/*
335
	 * These are the total time in nanoseconds that the event
336
	 * has been enabled (i.e. eligible to run, and the task has
337
	 * been scheduled in, if this is a per-task event)
338 339 340
	 * and running (scheduled onto the CPU), respectively.
	 *
	 * They are computed from tstamp_enabled, tstamp_running and
341
	 * tstamp_stopped when the event is in INACTIVE or ACTIVE state.
342 343 344 345 346 347
	 */
	u64				total_time_enabled;
	u64				total_time_running;

	/*
	 * These are timestamps used for computing total_time_enabled
348
	 * and total_time_running when the event is in INACTIVE or
349 350
	 * ACTIVE state, measured in nanoseconds from an arbitrary point
	 * in time.
351 352
	 * tstamp_enabled: the notional time when the event was enabled
	 * tstamp_running: the notional time when the event was scheduled on
353
	 * tstamp_stopped: in INACTIVE state, the notional time when the
354
	 *	event was scheduled off.
355 356 357 358 359
	 */
	u64				tstamp_enabled;
	u64				tstamp_running;
	u64				tstamp_stopped;

360 361 362 363 364 365 366 367 368 369
	/*
	 * timestamp shadows the actual context timing but it can
	 * be safely used in NMI interrupt context. It reflects the
	 * context time as it was when the event was last scheduled in.
	 *
	 * ctx_time already accounts for ctx->timestamp. Therefore to
	 * compute ctx_time for a sample, simply add perf_clock().
	 */
	u64				shadow_ctx_time;

370
	struct perf_event_attr		attr;
371
	u16				header_size;
372
	u16				id_header_size;
373
	u16				read_size;
374
	struct hw_perf_event		hw;
T
Thomas Gleixner 已提交
375

376
	struct perf_event_context	*ctx;
377
	atomic_long_t			refcount;
T
Thomas Gleixner 已提交
378

379 380
	/*
	 * These accumulate total time (in nanoseconds) that children
381
	 * events have been enabled and running, respectively.
382 383 384 385
	 */
	atomic64_t			child_total_time_enabled;
	atomic64_t			child_total_time_running;

T
Thomas Gleixner 已提交
386
	/*
387
	 * Protect attach/detach and child_list:
T
Thomas Gleixner 已提交
388
	 */
389 390
	struct mutex			child_mutex;
	struct list_head		child_list;
391
	struct perf_event		*parent;
T
Thomas Gleixner 已提交
392 393 394 395

	int				oncpu;
	int				cpu;

396 397 398
	struct list_head		owner_entry;
	struct task_struct		*owner;

399 400 401
	/* mmap bits */
	struct mutex			mmap_mutex;
	atomic_t			mmap_count;
P
Peter Zijlstra 已提交
402

403
	struct ring_buffer		*rb;
404
	struct list_head		rb_entry;
405

406
	/* poll related */
T
Thomas Gleixner 已提交
407
	wait_queue_head_t		waitq;
P
Peter Zijlstra 已提交
408
	struct fasync_struct		*fasync;
409 410 411

	/* delayed work for NMIs and such */
	int				pending_wakeup;
412
	int				pending_kill;
413
	int				pending_disable;
414
	struct irq_work			pending;
P
Peter Zijlstra 已提交
415

416 417
	atomic_t			event_limit;

418
	void (*destroy)(struct perf_event *);
P
Peter Zijlstra 已提交
419
	struct rcu_head			rcu_head;
420 421

	struct pid_namespace		*ns;
422
	u64				id;
L
Li Zefan 已提交
423

424
	perf_overflow_handler_t		overflow_handler;
425
	void				*overflow_handler_context;
426

427
#ifdef CONFIG_EVENT_TRACING
428
	struct ftrace_event_call	*tp_event;
L
Li Zefan 已提交
429
	struct event_filter		*filter;
430 431 432
#ifdef CONFIG_FUNCTION_TRACER
	struct ftrace_ops               ftrace_ops;
#endif
433
#endif
L
Li Zefan 已提交
434

S
Stephane Eranian 已提交
435 436 437 438 439
#ifdef CONFIG_CGROUP_PERF
	struct perf_cgroup		*cgrp; /* cgroup event is attach to */
	int				cgrp_defer_enabled;
#endif

L
Li Zefan 已提交
440
#endif /* CONFIG_PERF_EVENTS */
T
Thomas Gleixner 已提交
441 442
};

443 444 445 446 447
enum perf_event_context_type {
	task_context,
	cpu_context,
};

T
Thomas Gleixner 已提交
448
/**
449
 * struct perf_event_context - event context structure
T
Thomas Gleixner 已提交
450
 *
451
 * Used as a container for task events and CPU events as well:
T
Thomas Gleixner 已提交
452
 */
453
struct perf_event_context {
P
Peter Zijlstra 已提交
454
	struct pmu			*pmu;
455
	enum perf_event_context_type	type;
T
Thomas Gleixner 已提交
456
	/*
457
	 * Protect the states of the events in the list,
458
	 * nr_active, and the list:
T
Thomas Gleixner 已提交
459
	 */
460
	raw_spinlock_t			lock;
461
	/*
462
	 * Protect the list of events.  Locking either mutex or lock
463 464 465
	 * is sufficient to ensure the list doesn't change; to change
	 * the list you need to lock both the mutex and the spinlock.
	 */
I
Ingo Molnar 已提交
466
	struct mutex			mutex;
467

468 469
	struct list_head		pinned_groups;
	struct list_head		flexible_groups;
I
Ingo Molnar 已提交
470
	struct list_head		event_list;
471
	int				nr_events;
I
Ingo Molnar 已提交
472 473
	int				nr_active;
	int				is_active;
474
	int				nr_stat;
475
	int				nr_freq;
476
	int				rotate_disable;
I
Ingo Molnar 已提交
477 478
	atomic_t			refcount;
	struct task_struct		*task;
479 480

	/*
481
	 * Context clock, runs when context enabled.
482
	 */
I
Ingo Molnar 已提交
483 484
	u64				time;
	u64				timestamp;
485 486 487 488 489

	/*
	 * These fields let us detect when two contexts have both
	 * been cloned (inherited) from a common ancestor.
	 */
490
	struct perf_event_context	*parent_ctx;
I
Ingo Molnar 已提交
491 492 493
	u64				parent_gen;
	u64				generation;
	int				pin_count;
494 495
	int				nr_cgroups;	 /* cgroup evts */
	int				nr_branch_stack; /* branch_stack evt */
496
	struct rcu_head			rcu_head;
T
Thomas Gleixner 已提交
497 498
};

499 500
/*
 * Number of contexts where an event can trigger:
501
 *	task, softirq, hardirq, nmi.
502 503 504
 */
#define PERF_NR_CONTEXTS	4

T
Thomas Gleixner 已提交
505
/**
506
 * struct perf_event_cpu_context - per cpu event context structure
T
Thomas Gleixner 已提交
507 508
 */
struct perf_cpu_context {
509 510
	struct perf_event_context	ctx;
	struct perf_event_context	*task_ctx;
T
Thomas Gleixner 已提交
511
	int				active_oncpu;
512
	int				exclusive;
513 514
	struct hrtimer			hrtimer;
	ktime_t				hrtimer_interval;
515
	struct list_head		rotation_list;
516
	struct pmu			*unique_pmu;
S
Stephane Eranian 已提交
517
	struct perf_cgroup		*cgrp;
T
Thomas Gleixner 已提交
518 519
};

520
struct perf_output_handle {
I
Ingo Molnar 已提交
521
	struct perf_event		*event;
522
	struct ring_buffer		*rb;
523
	unsigned long			wakeup;
524 525 526
	unsigned long			size;
	void				*addr;
	int				page;
527 528
};

529
#ifdef CONFIG_PERF_EVENTS
530

531
extern int perf_pmu_register(struct pmu *pmu, const char *name, int type);
532
extern void perf_pmu_unregister(struct pmu *pmu);
I
Ingo Molnar 已提交
533

534
extern int perf_num_counters(void);
535
extern const char *perf_pmu_name(void);
536 537 538 539
extern void __perf_event_task_sched_in(struct task_struct *prev,
				       struct task_struct *task);
extern void __perf_event_task_sched_out(struct task_struct *prev,
					struct task_struct *next);
540 541 542
extern int perf_event_init_task(struct task_struct *child);
extern void perf_event_exit_task(struct task_struct *child);
extern void perf_event_free_task(struct task_struct *task);
543
extern void perf_event_delayed_put(struct task_struct *task);
544
extern void perf_event_print_debug(void);
P
Peter Zijlstra 已提交
545 546
extern void perf_pmu_disable(struct pmu *pmu);
extern void perf_pmu_enable(struct pmu *pmu);
547 548
extern int perf_event_task_disable(void);
extern int perf_event_task_enable(void);
549
extern int perf_event_refresh(struct perf_event *event, int refresh);
550
extern void perf_event_update_userpage(struct perf_event *event);
551 552 553 554
extern int perf_event_release_kernel(struct perf_event *event);
extern struct perf_event *
perf_event_create_kernel_counter(struct perf_event_attr *attr,
				int cpu,
M
Matt Helsley 已提交
555
				struct task_struct *task,
556 557
				perf_overflow_handler_t callback,
				void *context);
558 559
extern void perf_pmu_migrate_context(struct pmu *pmu,
				int src_cpu, int dst_cpu);
560 561
extern u64 perf_event_read_value(struct perf_event *event,
				 u64 *enabled, u64 *running);
562

563

564
struct perf_sample_data {
565 566 567 568 569 570 571 572
	u64				type;

	u64				ip;
	struct {
		u32	pid;
		u32	tid;
	}				tid_entry;
	u64				time;
I
Ingo Molnar 已提交
573
	u64				addr;
574 575 576 577 578 579
	u64				id;
	u64				stream_id;
	struct {
		u32	cpu;
		u32	reserved;
	}				cpu_entry;
I
Ingo Molnar 已提交
580
	u64				period;
581
	union  perf_mem_data_src	data_src;
582
	struct perf_callchain_entry	*callchain;
583
	struct perf_raw_record		*raw;
584
	struct perf_branch_stack	*br_stack;
585
	struct perf_regs_user		regs_user;
586
	u64				stack_user_size;
A
Andi Kleen 已提交
587
	u64				weight;
588 589
};

590 591
static inline void perf_sample_data_init(struct perf_sample_data *data,
					 u64 addr, u64 period)
592
{
593
	/* remaining struct members initialized in perf_prepare_sample() */
594 595
	data->addr = addr;
	data->raw  = NULL;
596
	data->br_stack = NULL;
597 598 599
	data->period = period;
	data->regs_user.abi = PERF_SAMPLE_REGS_ABI_NONE;
	data->regs_user.regs = NULL;
600
	data->stack_user_size = 0;
A
Andi Kleen 已提交
601
	data->weight = 0;
602
	data->data_src.val = 0;
603 604
}

605 606 607
extern void perf_output_sample(struct perf_output_handle *handle,
			       struct perf_event_header *header,
			       struct perf_sample_data *data,
608
			       struct perf_event *event);
609 610
extern void perf_prepare_sample(struct perf_event_header *header,
				struct perf_sample_data *data,
611
				struct perf_event *event,
612 613
				struct pt_regs *regs);

614
extern int perf_event_overflow(struct perf_event *event,
615 616
				 struct perf_sample_data *data,
				 struct pt_regs *regs);
617

618 619 620 621 622
static inline bool is_sampling_event(struct perf_event *event)
{
	return event->attr.sample_period != 0;
}

623
/*
624
 * Return 1 for a software event, 0 for a hardware event
625
 */
626
static inline int is_software_event(struct perf_event *event)
627
{
628
	return event->pmu->task_ctx_nr == perf_sw_context;
629 630
}

631
extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX];
632

633
extern void __perf_sw_event(u32, u64, struct pt_regs *, u64);
634

635
#ifndef perf_arch_fetch_caller_regs
636
static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { }
637
#endif
638 639 640 641 642 643 644 645 646

/*
 * Take a snapshot of the regs. Skip ip and frame pointer to
 * the nth caller. We only need a few of the regs:
 * - ip for PERF_SAMPLE_IP
 * - cs for user_mode() tests
 * - bp for callchains
 * - eflags, for future purposes, just in case
 */
647
static inline void perf_fetch_caller_regs(struct pt_regs *regs)
648 649 650
{
	memset(regs, 0, sizeof(*regs));

651
	perf_arch_fetch_caller_regs(regs, CALLER_ADDR0);
652 653
}

P
Peter Zijlstra 已提交
654
static __always_inline void
655
perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
656
{
P
Peter Zijlstra 已提交
657 658
	struct pt_regs hot_regs;

659
	if (static_key_false(&perf_swevent_enabled[event_id])) {
660 661 662 663
		if (!regs) {
			perf_fetch_caller_regs(&hot_regs);
			regs = &hot_regs;
		}
664
		__perf_sw_event(event_id, nr, regs, addr);
665 666 667
	}
}

668
extern struct static_key_deferred perf_sched_events;
669

670
static inline void perf_event_task_sched_in(struct task_struct *prev,
671
					    struct task_struct *task)
672 673 674 675 676 677 678
{
	if (static_key_false(&perf_sched_events.key))
		__perf_event_task_sched_in(prev, task);
}

static inline void perf_event_task_sched_out(struct task_struct *prev,
					     struct task_struct *next)
679
{
680
	perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, NULL, 0);
681

682
	if (static_key_false(&perf_sched_events.key))
683
		__perf_event_task_sched_out(prev, next);
684 685
}

686
extern void perf_event_mmap(struct vm_area_struct *vma);
687
extern struct perf_guest_info_callbacks *perf_guest_cbs;
688 689
extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks);
extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks);
690

691 692
extern void perf_event_comm(struct task_struct *tsk);
extern void perf_event_fork(struct task_struct *tsk);
693

694 695 696
/* Callchains */
DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry);

697 698
extern void perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs);
extern void perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs);
699

700
static inline void perf_callchain_store(struct perf_callchain_entry *entry, u64 ip)
701 702 703 704
{
	if (entry->nr < PERF_MAX_STACK_DEPTH)
		entry->ip[entry->nr++] = ip;
}
705

706 707 708
extern int sysctl_perf_event_paranoid;
extern int sysctl_perf_event_mlock;
extern int sysctl_perf_event_sample_rate;
709 710 711
extern int sysctl_perf_cpu_time_max_percent;

extern void perf_sample_event_took(u64 sample_len_ns);
712

P
Peter Zijlstra 已提交
713 714 715
extern int perf_proc_update_handler(struct ctl_table *table, int write,
		void __user *buffer, size_t *lenp,
		loff_t *ppos);
716 717 718 719
extern int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write,
		void __user *buffer, size_t *lenp,
		loff_t *ppos);

P
Peter Zijlstra 已提交
720

721 722 723 724 725 726 727 728 729 730 731 732 733 734 735
static inline bool perf_paranoid_tracepoint_raw(void)
{
	return sysctl_perf_event_paranoid > -1;
}

static inline bool perf_paranoid_cpu(void)
{
	return sysctl_perf_event_paranoid > 0;
}

static inline bool perf_paranoid_kernel(void)
{
	return sysctl_perf_event_paranoid > 1;
}

736
extern void perf_event_init(void);
737 738
extern void perf_tp_event(u64 addr, u64 count, void *record,
			  int entry_size, struct pt_regs *regs,
739 740
			  struct hlist_head *head, int rctx,
			  struct task_struct *task);
741
extern void perf_bp_event(struct perf_event *event, void *data);
742

743
#ifndef perf_misc_flags
744 745 746
# define perf_misc_flags(regs) \
		(user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL)
# define perf_instruction_pointer(regs)	instruction_pointer(regs)
747 748
#endif

749 750 751 752 753
static inline bool has_branch_stack(struct perf_event *event)
{
	return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK;
}

754
extern int perf_output_begin(struct perf_output_handle *handle,
755
			     struct perf_event *event, unsigned int size);
756
extern void perf_output_end(struct perf_output_handle *handle);
757
extern unsigned int perf_output_copy(struct perf_output_handle *handle,
758
			     const void *buf, unsigned int len);
759 760
extern unsigned int perf_output_skip(struct perf_output_handle *handle,
				     unsigned int len);
761 762
extern int perf_swevent_get_recursion_context(void);
extern void perf_swevent_put_recursion_context(int rctx);
763
extern u64 perf_swevent_set_period(struct perf_event *event);
764 765
extern void perf_event_enable(struct perf_event *event);
extern void perf_event_disable(struct perf_event *event);
766
extern int __perf_event_disable(void *info);
767
extern void perf_event_task_tick(void);
T
Thomas Gleixner 已提交
768 769
#else
static inline void
770 771 772 773 774
perf_event_task_sched_in(struct task_struct *prev,
			 struct task_struct *task)			{ }
static inline void
perf_event_task_sched_out(struct task_struct *prev,
			  struct task_struct *next)			{ }
775 776 777
static inline int perf_event_init_task(struct task_struct *child)	{ return 0; }
static inline void perf_event_exit_task(struct task_struct *child)	{ }
static inline void perf_event_free_task(struct task_struct *task)	{ }
778
static inline void perf_event_delayed_put(struct task_struct *task)	{ }
I
Ingo Molnar 已提交
779 780 781
static inline void perf_event_print_debug(void)				{ }
static inline int perf_event_task_disable(void)				{ return -EINVAL; }
static inline int perf_event_task_enable(void)				{ return -EINVAL; }
782 783 784 785
static inline int perf_event_refresh(struct perf_event *event, int refresh)
{
	return -EINVAL;
}
786

787
static inline void
788
perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)	{ }
789
static inline void
790
perf_bp_event(struct perf_event *event, void *data)			{ }
791

792
static inline int perf_register_guest_info_callbacks
793
(struct perf_guest_info_callbacks *callbacks)				{ return 0; }
794
static inline int perf_unregister_guest_info_callbacks
795
(struct perf_guest_info_callbacks *callbacks)				{ return 0; }
796

I
Ingo Molnar 已提交
797
static inline void perf_event_mmap(struct vm_area_struct *vma)		{ }
798 799 800
static inline void perf_event_comm(struct task_struct *tsk)		{ }
static inline void perf_event_fork(struct task_struct *tsk)		{ }
static inline void perf_event_init(void)				{ }
801
static inline int  perf_swevent_get_recursion_context(void)		{ return -1; }
802
static inline void perf_swevent_put_recursion_context(int rctx)		{ }
803
static inline u64 perf_swevent_set_period(struct perf_event *event)	{ return 0; }
804 805
static inline void perf_event_enable(struct perf_event *event)		{ }
static inline void perf_event_disable(struct perf_event *event)		{ }
806
static inline int __perf_event_disable(void *info)			{ return -1; }
807
static inline void perf_event_task_tick(void)				{ }
T
Thomas Gleixner 已提交
808 809
#endif

810 811 812 813 814 815
#if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_NO_HZ_FULL)
extern bool perf_event_can_stop_tick(void);
#else
static inline bool perf_event_can_stop_tick(void)			{ return true; }
#endif

816 817 818
#if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
extern void perf_restore_debug_store(void);
#else
819
static inline void perf_restore_debug_store(void)			{ }
T
Thomas Gleixner 已提交
820 821
#endif

822
#define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x))
823

824
/*
825
 * This has to have a higher priority than migration_notifier in sched/core.c.
826
 */
827 828 829 830
#define perf_cpu_notifier(fn)						\
do {									\
	static struct notifier_block fn##_nb __cpuinitdata =		\
		{ .notifier_call = fn, .priority = CPU_PRI_PERF };	\
831
	unsigned long cpu = smp_processor_id();				\
832
	unsigned long flags;						\
833
	fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE,			\
834
		(void *)(unsigned long)cpu);				\
835
	local_irq_save(flags);						\
836
	fn(&fn##_nb, (unsigned long)CPU_STARTING,			\
837
		(void *)(unsigned long)cpu);				\
838
	local_irq_restore(flags);					\
839
	fn(&fn##_nb, (unsigned long)CPU_ONLINE,				\
840
		(void *)(unsigned long)cpu);				\
841
	register_cpu_notifier(&fn##_nb);				\
842 843
} while (0)

844

845 846 847
struct perf_pmu_events_attr {
	struct device_attribute attr;
	u64 id;
848
	const char *event_str;
849 850 851 852 853 854 855 856
};

#define PMU_EVENT_ATTR(_name, _var, _id, _show)				\
static struct perf_pmu_events_attr _var = {				\
	.attr = __ATTR(_name, 0444, _show, NULL),			\
	.id   =  _id,							\
};

857 858 859 860 861 862 863 864 865 866 867 868
#define PMU_FORMAT_ATTR(_name, _format)					\
static ssize_t								\
_name##_show(struct device *dev,					\
			       struct device_attribute *attr,		\
			       char *page)				\
{									\
	BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE);			\
	return sprintf(page, _format "\n");				\
}									\
									\
static struct device_attribute format_attr_##_name = __ATTR_RO(_name)

869
#endif /* _LINUX_PERF_EVENT_H */