perf_event.h 24.0 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>
51
#include <linux/jump_label_ratelimit.h>
A
Arun Sharma 已提交
52
#include <linux/atomic.h>
53
#include <linux/sysfs.h>
54
#include <linux/perf_regs.h>
55
#include <asm/local.h>
56

57 58 59 60 61
struct perf_callchain_entry {
	__u64				nr;
	__u64				ip[PERF_MAX_STACK_DEPTH];
};

62 63 64
struct perf_raw_record {
	u32				size;
	void				*data;
65 66
};

67 68 69 70 71 72 73 74 75
/*
 * 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.
 */
76 77 78 79 80
struct perf_branch_stack {
	__u64				nr;
	struct perf_branch_entry	entries[0];
};

81 82 83 84 85
struct perf_regs_user {
	__u64		abi;
	struct pt_regs	*regs;
};

86 87
struct task_struct;

88 89 90 91 92 93 94 95 96 97
/*
 * 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[] */
};

98 99
struct event_constraint;

T
Thomas Gleixner 已提交
100
/**
101
 * struct hw_perf_event - performance event hardware details:
T
Thomas Gleixner 已提交
102
 */
103 104
struct hw_perf_event {
#ifdef CONFIG_PERF_EVENTS
105 106
	union {
		struct { /* hardware */
I
Ingo Molnar 已提交
107
			u64		config;
108
			u64		last_tag;
I
Ingo Molnar 已提交
109
			unsigned long	config_base;
110
			unsigned long	event_base;
111
			int		event_base_rdpmc;
I
Ingo Molnar 已提交
112
			int		idx;
113
			int		last_cpu;
114
			int		flags;
115

116
			struct hw_perf_event_extra extra_reg;
117
			struct hw_perf_event_extra branch_reg;
118 119

			struct event_constraint *constraint;
120
		};
121
		struct { /* software */
I
Ingo Molnar 已提交
122
			struct hrtimer	hrtimer;
123
		};
124 125 126 127 128
		struct { /* tracepoint */
			struct task_struct	*tp_target;
			/* for tp_event->class */
			struct list_head	tp_list;
		};
129
#ifdef CONFIG_HAVE_HW_BREAKPOINT
130
		struct { /* breakpoint */
131 132 133 134 135 136
			/*
			 * Crufty hack to avoid the chicken and egg
			 * problem hw_breakpoint has with context
			 * creation and event initalization.
			 */
			struct task_struct		*bp_target;
137 138
			struct arch_hw_breakpoint	info;
			struct list_head		bp_list;
139
		};
140
#endif
141
	};
P
Peter Zijlstra 已提交
142
	int				state;
143
	local64_t			prev_count;
144
	u64				sample_period;
145
	u64				last_period;
146
	local64_t			period_left;
147
	u64                             interrupts_seq;
148
	u64				interrupts;
149

150 151
	u64				freq_time_stamp;
	u64				freq_count_stamp;
152
#endif
T
Thomas Gleixner 已提交
153 154
};

P
Peter Zijlstra 已提交
155 156 157 158 159 160 161
/*
 * 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

162
struct perf_event;
I
Ingo Molnar 已提交
163

164 165 166 167
/*
 * Common implementation detail of pmu::{start,commit,cancel}_txn
 */
#define PERF_EVENT_TXN 0x1
168

I
Ingo Molnar 已提交
169
/**
170
 * struct pmu - generic performance monitoring unit
I
Ingo Molnar 已提交
171
 */
172
struct pmu {
173 174
	struct list_head		entry;

P
Peter Zijlstra 已提交
175
	struct device			*dev;
176
	const struct attribute_group	**attr_groups;
177
	const char			*name;
P
Peter Zijlstra 已提交
178 179
	int				type;

P
Peter Zijlstra 已提交
180 181
	int * __percpu			pmu_disable_count;
	struct perf_cpu_context * __percpu pmu_cpu_context;
P
Peter Zijlstra 已提交
182
	int				task_ctx_nr;
183
	int				hrtimer_interval_ms;
184 185

	/*
P
Peter Zijlstra 已提交
186 187
	 * Fully disable/enable this PMU, can be used to protect from the PMI
	 * as well as for lazy/batch writing of the MSRs.
188
	 */
P
Peter Zijlstra 已提交
189 190
	void (*pmu_enable)		(struct pmu *pmu); /* optional */
	void (*pmu_disable)		(struct pmu *pmu); /* optional */
191

192
	/*
P
Peter Zijlstra 已提交
193
	 * Try and initialize the event for this PMU.
194
	 * Should return -ENOENT when the @event doesn't match this PMU.
195
	 */
196 197
	int (*event_init)		(struct perf_event *event);

P
Peter Zijlstra 已提交
198 199 200 201
#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 */

202
	/*
P
Peter Zijlstra 已提交
203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219
	 * 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.
	 */
220
	void (*read)			(struct perf_event *event);
221 222

	/*
223 224 225
	 * 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
226 227
	 *
	 * Start the transaction, after this ->add() doesn't need to
228
	 * do schedulability tests.
229
	 */
230
	void (*start_txn)		(struct pmu *pmu); /* optional */
231
	/*
P
Peter Zijlstra 已提交
232
	 * If ->start_txn() disabled the ->add() schedulability test
233 234 235 236
	 * 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.
	 */
237
	int  (*commit_txn)		(struct pmu *pmu); /* optional */
238
	/*
P
Peter Zijlstra 已提交
239
	 * Will cancel the transaction, assumes ->del() is called
L
Lucas De Marchi 已提交
240
	 * for each successful ->add() during the transaction.
241
	 */
242
	void (*cancel_txn)		(struct pmu *pmu); /* optional */
243 244 245 246 247 248

	/*
	 * 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 */
249 250 251 252 253

	/*
	 * flush branch stack on context-switches (needed in cpu-wide mode)
	 */
	void (*flush_branch_stack)	(void);
I
Ingo Molnar 已提交
254 255
};

256
/**
257
 * enum perf_event_active_state - the states of a event
258
 */
259
enum perf_event_active_state {
I
Ingo Molnar 已提交
260
	PERF_EVENT_STATE_ERROR		= -2,
261 262
	PERF_EVENT_STATE_OFF		= -1,
	PERF_EVENT_STATE_INACTIVE	=  0,
I
Ingo Molnar 已提交
263
	PERF_EVENT_STATE_ACTIVE		=  1,
264 265
};

266
struct file;
267 268
struct perf_sample_data;

269
typedef void (*perf_overflow_handler_t)(struct perf_event *,
270 271 272
					struct perf_sample_data *,
					struct pt_regs *regs);

273
enum perf_group_flag {
274
	PERF_GROUP_SOFTWARE		= 0x1,
275 276
};

277 278
#define SWEVENT_HLIST_BITS		8
#define SWEVENT_HLIST_SIZE		(1 << SWEVENT_HLIST_BITS)
279 280

struct swevent_hlist {
281 282
	struct hlist_head		heads[SWEVENT_HLIST_SIZE];
	struct rcu_head			rcu_head;
283 284
};

285 286
#define PERF_ATTACH_CONTEXT	0x01
#define PERF_ATTACH_GROUP	0x02
287
#define PERF_ATTACH_TASK	0x04
288

289
struct perf_cgroup;
290 291
struct ring_buffer;

T
Thomas Gleixner 已提交
292
/**
293
 * struct perf_event - performance event kernel representation:
T
Thomas Gleixner 已提交
294
 */
295 296
struct perf_event {
#ifdef CONFIG_PERF_EVENTS
297 298 299 300 301
	/*
	 * entry onto perf_event_context::event_list;
	 *   modifications require ctx->lock
	 *   RCU safe iterations.
	 */
P
Peter Zijlstra 已提交
302
	struct list_head		event_entry;
303 304 305 306 307 308 309 310 311 312

	/*
	 * XXX: group_entry and sibling_list should be mutually exclusive;
	 * either you're a sibling on a group, or you're the group leader.
	 * Rework the code to always use the same list element.
	 *
	 * Locked for modification by both ctx->mutex and ctx->lock; holding
	 * either sufficies for read.
	 */
	struct list_head		group_entry;
313
	struct list_head		sibling_list;
314 315 316 317 318 319 320 321

	/*
	 * We need storage to track the entries in perf_pmu_migrate_context; we
	 * cannot use the event_entry because of RCU and we want to keep the
	 * group in tact which avoids us using the other two entries.
	 */
	struct list_head		migrate_entry;

322 323 324 325
	union {
		struct hlist_node	hlist_entry;
		struct list_head	active_entry;
	};
I
Ingo Molnar 已提交
326
	int				nr_siblings;
327
	int				group_flags;
328
	struct perf_event		*group_leader;
P
Peter Zijlstra 已提交
329
	struct pmu			*pmu;
330

331
	enum perf_event_active_state	state;
332
	unsigned int			attach_state;
333
	local64_t			count;
334
	atomic64_t			child_count;
335

336
	/*
337
	 * These are the total time in nanoseconds that the event
338
	 * has been enabled (i.e. eligible to run, and the task has
339
	 * been scheduled in, if this is a per-task event)
340 341 342
	 * and running (scheduled onto the CPU), respectively.
	 *
	 * They are computed from tstamp_enabled, tstamp_running and
343
	 * tstamp_stopped when the event is in INACTIVE or ACTIVE state.
344 345 346 347 348 349
	 */
	u64				total_time_enabled;
	u64				total_time_running;

	/*
	 * These are timestamps used for computing total_time_enabled
350
	 * and total_time_running when the event is in INACTIVE or
351 352
	 * ACTIVE state, measured in nanoseconds from an arbitrary point
	 * in time.
353 354
	 * tstamp_enabled: the notional time when the event was enabled
	 * tstamp_running: the notional time when the event was scheduled on
355
	 * tstamp_stopped: in INACTIVE state, the notional time when the
356
	 *	event was scheduled off.
357 358 359 360 361
	 */
	u64				tstamp_enabled;
	u64				tstamp_running;
	u64				tstamp_stopped;

362 363 364 365 366 367 368 369 370 371
	/*
	 * 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;

372
	struct perf_event_attr		attr;
373
	u16				header_size;
374
	u16				id_header_size;
375
	u16				read_size;
376
	struct hw_perf_event		hw;
T
Thomas Gleixner 已提交
377

378
	struct perf_event_context	*ctx;
379
	atomic_long_t			refcount;
T
Thomas Gleixner 已提交
380

381 382
	/*
	 * These accumulate total time (in nanoseconds) that children
383
	 * events have been enabled and running, respectively.
384 385 386 387
	 */
	atomic64_t			child_total_time_enabled;
	atomic64_t			child_total_time_running;

T
Thomas Gleixner 已提交
388
	/*
389
	 * Protect attach/detach and child_list:
T
Thomas Gleixner 已提交
390
	 */
391 392
	struct mutex			child_mutex;
	struct list_head		child_list;
393
	struct perf_event		*parent;
T
Thomas Gleixner 已提交
394 395 396 397

	int				oncpu;
	int				cpu;

398 399 400
	struct list_head		owner_entry;
	struct task_struct		*owner;

401 402 403
	/* mmap bits */
	struct mutex			mmap_mutex;
	atomic_t			mmap_count;
P
Peter Zijlstra 已提交
404

405
	struct ring_buffer		*rb;
406
	struct list_head		rb_entry;
407

408
	/* poll related */
T
Thomas Gleixner 已提交
409
	wait_queue_head_t		waitq;
P
Peter Zijlstra 已提交
410
	struct fasync_struct		*fasync;
411 412 413

	/* delayed work for NMIs and such */
	int				pending_wakeup;
414
	int				pending_kill;
415
	int				pending_disable;
416
	struct irq_work			pending;
P
Peter Zijlstra 已提交
417

418 419
	atomic_t			event_limit;

420
	void (*destroy)(struct perf_event *);
P
Peter Zijlstra 已提交
421
	struct rcu_head			rcu_head;
422 423

	struct pid_namespace		*ns;
424
	u64				id;
L
Li Zefan 已提交
425

426
	perf_overflow_handler_t		overflow_handler;
427
	void				*overflow_handler_context;
428

429
#ifdef CONFIG_EVENT_TRACING
430
	struct ftrace_event_call	*tp_event;
L
Li Zefan 已提交
431
	struct event_filter		*filter;
432 433 434
#ifdef CONFIG_FUNCTION_TRACER
	struct ftrace_ops               ftrace_ops;
#endif
435
#endif
L
Li Zefan 已提交
436

S
Stephane Eranian 已提交
437 438 439 440 441
#ifdef CONFIG_CGROUP_PERF
	struct perf_cgroup		*cgrp; /* cgroup event is attach to */
	int				cgrp_defer_enabled;
#endif

L
Li Zefan 已提交
442
#endif /* CONFIG_PERF_EVENTS */
T
Thomas Gleixner 已提交
443 444
};

445 446 447 448 449
enum perf_event_context_type {
	task_context,
	cpu_context,
};

T
Thomas Gleixner 已提交
450
/**
451
 * struct perf_event_context - event context structure
T
Thomas Gleixner 已提交
452
 *
453
 * Used as a container for task events and CPU events as well:
T
Thomas Gleixner 已提交
454
 */
455
struct perf_event_context {
P
Peter Zijlstra 已提交
456
	struct pmu			*pmu;
457
	enum perf_event_context_type	type;
T
Thomas Gleixner 已提交
458
	/*
459
	 * Protect the states of the events in the list,
460
	 * nr_active, and the list:
T
Thomas Gleixner 已提交
461
	 */
462
	raw_spinlock_t			lock;
463
	/*
464
	 * Protect the list of events.  Locking either mutex or lock
465 466 467
	 * 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
468
	struct mutex			mutex;
469

470 471
	struct list_head		pinned_groups;
	struct list_head		flexible_groups;
I
Ingo Molnar 已提交
472
	struct list_head		event_list;
473
	int				nr_events;
I
Ingo Molnar 已提交
474 475
	int				nr_active;
	int				is_active;
476
	int				nr_stat;
477
	int				nr_freq;
478
	int				rotate_disable;
I
Ingo Molnar 已提交
479 480
	atomic_t			refcount;
	struct task_struct		*task;
481 482

	/*
483
	 * Context clock, runs when context enabled.
484
	 */
I
Ingo Molnar 已提交
485 486
	u64				time;
	u64				timestamp;
487 488 489 490 491

	/*
	 * These fields let us detect when two contexts have both
	 * been cloned (inherited) from a common ancestor.
	 */
492
	struct perf_event_context	*parent_ctx;
I
Ingo Molnar 已提交
493 494 495
	u64				parent_gen;
	u64				generation;
	int				pin_count;
496 497
	int				nr_cgroups;	 /* cgroup evts */
	int				nr_branch_stack; /* branch_stack evt */
498
	struct rcu_head			rcu_head;
T
Thomas Gleixner 已提交
499 500
};

501 502
/*
 * Number of contexts where an event can trigger:
503
 *	task, softirq, hardirq, nmi.
504 505 506
 */
#define PERF_NR_CONTEXTS	4

T
Thomas Gleixner 已提交
507
/**
508
 * struct perf_event_cpu_context - per cpu event context structure
T
Thomas Gleixner 已提交
509 510
 */
struct perf_cpu_context {
511 512
	struct perf_event_context	ctx;
	struct perf_event_context	*task_ctx;
T
Thomas Gleixner 已提交
513
	int				active_oncpu;
514
	int				exclusive;
515 516
	struct hrtimer			hrtimer;
	ktime_t				hrtimer_interval;
517
	struct list_head		rotation_list;
518
	struct pmu			*unique_pmu;
S
Stephane Eranian 已提交
519
	struct perf_cgroup		*cgrp;
T
Thomas Gleixner 已提交
520 521
};

522
struct perf_output_handle {
I
Ingo Molnar 已提交
523
	struct perf_event		*event;
524
	struct ring_buffer		*rb;
525
	unsigned long			wakeup;
526 527 528
	unsigned long			size;
	void				*addr;
	int				page;
529 530
};

531
#ifdef CONFIG_PERF_EVENTS
532

533
extern int perf_pmu_register(struct pmu *pmu, const char *name, int type);
534
extern void perf_pmu_unregister(struct pmu *pmu);
I
Ingo Molnar 已提交
535

536
extern int perf_num_counters(void);
537
extern const char *perf_pmu_name(void);
538 539 540 541
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);
542 543 544
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);
545
extern void perf_event_delayed_put(struct task_struct *task);
546
extern void perf_event_print_debug(void);
P
Peter Zijlstra 已提交
547 548
extern void perf_pmu_disable(struct pmu *pmu);
extern void perf_pmu_enable(struct pmu *pmu);
549 550
extern int perf_event_task_disable(void);
extern int perf_event_task_enable(void);
551
extern int perf_event_refresh(struct perf_event *event, int refresh);
552
extern void perf_event_update_userpage(struct perf_event *event);
553 554 555 556
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
557
				struct task_struct *task,
558 559
				perf_overflow_handler_t callback,
				void *context);
560 561
extern void perf_pmu_migrate_context(struct pmu *pmu,
				int src_cpu, int dst_cpu);
562 563
extern u64 perf_event_read_value(struct perf_event *event,
				 u64 *enabled, u64 *running);
564

565

566
struct perf_sample_data {
567 568 569 570 571 572 573 574
	u64				type;

	u64				ip;
	struct {
		u32	pid;
		u32	tid;
	}				tid_entry;
	u64				time;
I
Ingo Molnar 已提交
575
	u64				addr;
576 577 578 579 580 581
	u64				id;
	u64				stream_id;
	struct {
		u32	cpu;
		u32	reserved;
	}				cpu_entry;
I
Ingo Molnar 已提交
582
	u64				period;
583
	union  perf_mem_data_src	data_src;
584
	struct perf_callchain_entry	*callchain;
585
	struct perf_raw_record		*raw;
586
	struct perf_branch_stack	*br_stack;
587
	struct perf_regs_user		regs_user;
588
	u64				stack_user_size;
A
Andi Kleen 已提交
589
	u64				weight;
A
Andi Kleen 已提交
590 591 592 593
	/*
	 * Transaction flags for abort events:
	 */
	u64				txn;
594 595
};

596 597
static inline void perf_sample_data_init(struct perf_sample_data *data,
					 u64 addr, u64 period)
598
{
599
	/* remaining struct members initialized in perf_prepare_sample() */
600 601
	data->addr = addr;
	data->raw  = NULL;
602
	data->br_stack = NULL;
603 604 605
	data->period = period;
	data->regs_user.abi = PERF_SAMPLE_REGS_ABI_NONE;
	data->regs_user.regs = NULL;
606
	data->stack_user_size = 0;
A
Andi Kleen 已提交
607
	data->weight = 0;
608
	data->data_src.val = 0;
A
Andi Kleen 已提交
609
	data->txn = 0;
610 611
}

612 613 614
extern void perf_output_sample(struct perf_output_handle *handle,
			       struct perf_event_header *header,
			       struct perf_sample_data *data,
615
			       struct perf_event *event);
616 617
extern void perf_prepare_sample(struct perf_event_header *header,
				struct perf_sample_data *data,
618
				struct perf_event *event,
619 620
				struct pt_regs *regs);

621
extern int perf_event_overflow(struct perf_event *event,
622 623
				 struct perf_sample_data *data,
				 struct pt_regs *regs);
624

625 626 627 628 629
static inline bool is_sampling_event(struct perf_event *event)
{
	return event->attr.sample_period != 0;
}

630
/*
631
 * Return 1 for a software event, 0 for a hardware event
632
 */
633
static inline int is_software_event(struct perf_event *event)
634
{
635
	return event->pmu->task_ctx_nr == perf_sw_context;
636 637
}

638
extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX];
639

640
extern void __perf_sw_event(u32, u64, struct pt_regs *, u64);
641

642
#ifndef perf_arch_fetch_caller_regs
643
static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { }
644
#endif
645 646 647 648 649 650 651 652 653

/*
 * 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
 */
654
static inline void perf_fetch_caller_regs(struct pt_regs *regs)
655 656 657
{
	memset(regs, 0, sizeof(*regs));

658
	perf_arch_fetch_caller_regs(regs, CALLER_ADDR0);
659 660
}

P
Peter Zijlstra 已提交
661
static __always_inline void
662
perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
663
{
P
Peter Zijlstra 已提交
664 665
	struct pt_regs hot_regs;

666
	if (static_key_false(&perf_swevent_enabled[event_id])) {
667 668 669 670
		if (!regs) {
			perf_fetch_caller_regs(&hot_regs);
			regs = &hot_regs;
		}
671
		__perf_sw_event(event_id, nr, regs, addr);
672 673 674
	}
}

675
extern struct static_key_deferred perf_sched_events;
676

677
static inline void perf_event_task_sched_in(struct task_struct *prev,
678
					    struct task_struct *task)
679 680 681 682 683 684 685
{
	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)
686
{
687
	perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, NULL, 0);
688

689
	if (static_key_false(&perf_sched_events.key))
690
		__perf_event_task_sched_out(prev, next);
691 692
}

693
extern void perf_event_mmap(struct vm_area_struct *vma);
694
extern struct perf_guest_info_callbacks *perf_guest_cbs;
695 696
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);
697

698 699
extern void perf_event_comm(struct task_struct *tsk);
extern void perf_event_fork(struct task_struct *tsk);
700

701 702 703
/* Callchains */
DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry);

704 705
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);
706

707
static inline void perf_callchain_store(struct perf_callchain_entry *entry, u64 ip)
708 709 710 711
{
	if (entry->nr < PERF_MAX_STACK_DEPTH)
		entry->ip[entry->nr++] = ip;
}
712

713 714 715
extern int sysctl_perf_event_paranoid;
extern int sysctl_perf_event_mlock;
extern int sysctl_perf_event_sample_rate;
716 717 718
extern int sysctl_perf_cpu_time_max_percent;

extern void perf_sample_event_took(u64 sample_len_ns);
719

P
Peter Zijlstra 已提交
720 721 722
extern int perf_proc_update_handler(struct ctl_table *table, int write,
		void __user *buffer, size_t *lenp,
		loff_t *ppos);
723 724 725 726
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
727

728 729 730 731 732 733 734 735 736 737 738 739 740 741 742
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;
}

743
extern void perf_event_init(void);
744 745
extern void perf_tp_event(u64 addr, u64 count, void *record,
			  int entry_size, struct pt_regs *regs,
746 747
			  struct hlist_head *head, int rctx,
			  struct task_struct *task);
748
extern void perf_bp_event(struct perf_event *event, void *data);
749

750
#ifndef perf_misc_flags
751 752 753
# define perf_misc_flags(regs) \
		(user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL)
# define perf_instruction_pointer(regs)	instruction_pointer(regs)
754 755
#endif

756 757 758 759 760
static inline bool has_branch_stack(struct perf_event *event)
{
	return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK;
}

761
extern int perf_output_begin(struct perf_output_handle *handle,
762
			     struct perf_event *event, unsigned int size);
763
extern void perf_output_end(struct perf_output_handle *handle);
764
extern unsigned int perf_output_copy(struct perf_output_handle *handle,
765
			     const void *buf, unsigned int len);
766 767
extern unsigned int perf_output_skip(struct perf_output_handle *handle,
				     unsigned int len);
768 769
extern int perf_swevent_get_recursion_context(void);
extern void perf_swevent_put_recursion_context(int rctx);
770
extern u64 perf_swevent_set_period(struct perf_event *event);
771 772
extern void perf_event_enable(struct perf_event *event);
extern void perf_event_disable(struct perf_event *event);
773
extern int __perf_event_disable(void *info);
774
extern void perf_event_task_tick(void);
T
Thomas Gleixner 已提交
775 776
#else
static inline void
777 778 779 780 781
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)			{ }
782 783 784
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)	{ }
785
static inline void perf_event_delayed_put(struct task_struct *task)	{ }
I
Ingo Molnar 已提交
786 787 788
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; }
789 790 791 792
static inline int perf_event_refresh(struct perf_event *event, int refresh)
{
	return -EINVAL;
}
793

794
static inline void
795
perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)	{ }
796
static inline void
797
perf_bp_event(struct perf_event *event, void *data)			{ }
798

799
static inline int perf_register_guest_info_callbacks
800
(struct perf_guest_info_callbacks *callbacks)				{ return 0; }
801
static inline int perf_unregister_guest_info_callbacks
802
(struct perf_guest_info_callbacks *callbacks)				{ return 0; }
803

I
Ingo Molnar 已提交
804
static inline void perf_event_mmap(struct vm_area_struct *vma)		{ }
805 806 807
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)				{ }
808
static inline int  perf_swevent_get_recursion_context(void)		{ return -1; }
809
static inline void perf_swevent_put_recursion_context(int rctx)		{ }
810
static inline u64 perf_swevent_set_period(struct perf_event *event)	{ return 0; }
811 812
static inline void perf_event_enable(struct perf_event *event)		{ }
static inline void perf_event_disable(struct perf_event *event)		{ }
813
static inline int __perf_event_disable(void *info)			{ return -1; }
814
static inline void perf_event_task_tick(void)				{ }
T
Thomas Gleixner 已提交
815 816
#endif

817 818 819 820 821 822
#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

823 824 825
#if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
extern void perf_restore_debug_store(void);
#else
826
static inline void perf_restore_debug_store(void)			{ }
T
Thomas Gleixner 已提交
827 828
#endif

829
#define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x))
830

831
/*
832
 * This has to have a higher priority than migration_notifier in sched/core.c.
833
 */
834 835
#define perf_cpu_notifier(fn)						\
do {									\
836
	static struct notifier_block fn##_nb =				\
837
		{ .notifier_call = fn, .priority = CPU_PRI_PERF };	\
838
	unsigned long cpu = smp_processor_id();				\
839
	unsigned long flags;						\
840
	fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE,			\
841
		(void *)(unsigned long)cpu);				\
842
	local_irq_save(flags);						\
843
	fn(&fn##_nb, (unsigned long)CPU_STARTING,			\
844
		(void *)(unsigned long)cpu);				\
845
	local_irq_restore(flags);					\
846
	fn(&fn##_nb, (unsigned long)CPU_ONLINE,				\
847
		(void *)(unsigned long)cpu);				\
848
	register_cpu_notifier(&fn##_nb);				\
849 850
} while (0)

851

852 853 854
struct perf_pmu_events_attr {
	struct device_attribute attr;
	u64 id;
855
	const char *event_str;
856 857 858 859 860 861 862 863
};

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

864 865 866 867 868 869 870 871 872 873 874 875
#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)

876
#endif /* _LINUX_PERF_EVENT_H */