perf_event.h 24.8 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>
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 <linux/workqueue.h>
56
#include <asm/local.h>
57

58 59 60 61 62
struct perf_callchain_entry {
	__u64				nr;
	__u64				ip[PERF_MAX_STACK_DEPTH];
};

63 64 65
struct perf_raw_record {
	u32				size;
	void				*data;
66 67
};

68 69 70 71 72 73 74 75 76
/*
 * 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.
 */
77 78 79 80 81
struct perf_branch_stack {
	__u64				nr;
	struct perf_branch_entry	entries[0];
};

82 83 84 85 86
struct perf_regs_user {
	__u64		abi;
	struct pt_regs	*regs;
};

87 88
struct task_struct;

89 90 91 92 93 94 95 96 97 98
/*
 * 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[] */
};

99 100
struct event_constraint;

T
Thomas Gleixner 已提交
101
/**
102
 * struct hw_perf_event - performance event hardware details:
T
Thomas Gleixner 已提交
103
 */
104 105
struct hw_perf_event {
#ifdef CONFIG_PERF_EVENTS
106 107
	union {
		struct { /* hardware */
I
Ingo Molnar 已提交
108
			u64		config;
109
			u64		last_tag;
I
Ingo Molnar 已提交
110
			unsigned long	config_base;
111
			unsigned long	event_base;
112
			int		event_base_rdpmc;
I
Ingo Molnar 已提交
113
			int		idx;
114
			int		last_cpu;
115
			int		flags;
116

117
			struct hw_perf_event_extra extra_reg;
118
			struct hw_perf_event_extra branch_reg;
119 120

			struct event_constraint *constraint;
121
		};
122
		struct { /* software */
I
Ingo Molnar 已提交
123
			struct hrtimer	hrtimer;
124
		};
125 126 127 128 129
		struct { /* tracepoint */
			struct task_struct	*tp_target;
			/* for tp_event->class */
			struct list_head	tp_list;
		};
130
#ifdef CONFIG_HAVE_HW_BREAKPOINT
131
		struct { /* breakpoint */
132 133 134 135 136 137
			/*
			 * Crufty hack to avoid the chicken and egg
			 * problem hw_breakpoint has with context
			 * creation and event initalization.
			 */
			struct task_struct		*bp_target;
138 139
			struct arch_hw_breakpoint	info;
			struct list_head		bp_list;
140
		};
141
#endif
142
	};
P
Peter Zijlstra 已提交
143
	int				state;
144
	local64_t			prev_count;
145
	u64				sample_period;
146
	u64				last_period;
147
	local64_t			period_left;
148
	u64                             interrupts_seq;
149
	u64				interrupts;
150

151 152
	u64				freq_time_stamp;
	u64				freq_count_stamp;
153
#endif
T
Thomas Gleixner 已提交
154 155
};

P
Peter Zijlstra 已提交
156 157 158 159 160 161 162
/*
 * 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

163
struct perf_event;
I
Ingo Molnar 已提交
164

165 166 167 168
/*
 * Common implementation detail of pmu::{start,commit,cancel}_txn
 */
#define PERF_EVENT_TXN 0x1
169

170 171 172 173 174
/**
 * pmu::capabilities flags
 */
#define PERF_PMU_CAP_NO_INTERRUPT		0x01

I
Ingo Molnar 已提交
175
/**
176
 * struct pmu - generic performance monitoring unit
I
Ingo Molnar 已提交
177
 */
178
struct pmu {
179 180
	struct list_head		entry;

181
	struct module			*module;
P
Peter Zijlstra 已提交
182
	struct device			*dev;
183
	const struct attribute_group	**attr_groups;
184
	const char			*name;
P
Peter Zijlstra 已提交
185 186
	int				type;

187 188 189 190 191
	/*
	 * various common per-pmu feature flags
	 */
	int				capabilities;

P
Peter Zijlstra 已提交
192 193
	int * __percpu			pmu_disable_count;
	struct perf_cpu_context * __percpu pmu_cpu_context;
P
Peter Zijlstra 已提交
194
	int				task_ctx_nr;
195
	int				hrtimer_interval_ms;
196 197

	/*
P
Peter Zijlstra 已提交
198 199
	 * Fully disable/enable this PMU, can be used to protect from the PMI
	 * as well as for lazy/batch writing of the MSRs.
200
	 */
P
Peter Zijlstra 已提交
201 202
	void (*pmu_enable)		(struct pmu *pmu); /* optional */
	void (*pmu_disable)		(struct pmu *pmu); /* optional */
203

204
	/*
P
Peter Zijlstra 已提交
205
	 * Try and initialize the event for this PMU.
206
	 * Should return -ENOENT when the @event doesn't match this PMU.
207
	 */
208 209
	int (*event_init)		(struct perf_event *event);

P
Peter Zijlstra 已提交
210 211 212 213
#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 */

214
	/*
P
Peter Zijlstra 已提交
215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231
	 * 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.
	 */
232
	void (*read)			(struct perf_event *event);
233 234

	/*
235 236 237
	 * 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
238 239
	 *
	 * Start the transaction, after this ->add() doesn't need to
240
	 * do schedulability tests.
241
	 */
242
	void (*start_txn)		(struct pmu *pmu); /* optional */
243
	/*
P
Peter Zijlstra 已提交
244
	 * If ->start_txn() disabled the ->add() schedulability test
245 246 247 248
	 * 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.
	 */
249
	int  (*commit_txn)		(struct pmu *pmu); /* optional */
250
	/*
P
Peter Zijlstra 已提交
251
	 * Will cancel the transaction, assumes ->del() is called
L
Lucas De Marchi 已提交
252
	 * for each successful ->add() during the transaction.
253
	 */
254
	void (*cancel_txn)		(struct pmu *pmu); /* optional */
255 256 257 258 259 260

	/*
	 * 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 */
261 262 263 264 265

	/*
	 * flush branch stack on context-switches (needed in cpu-wide mode)
	 */
	void (*flush_branch_stack)	(void);
I
Ingo Molnar 已提交
266 267
};

268
/**
269
 * enum perf_event_active_state - the states of a event
270
 */
271
enum perf_event_active_state {
I
Ingo Molnar 已提交
272
	PERF_EVENT_STATE_ERROR		= -2,
273 274
	PERF_EVENT_STATE_OFF		= -1,
	PERF_EVENT_STATE_INACTIVE	=  0,
I
Ingo Molnar 已提交
275
	PERF_EVENT_STATE_ACTIVE		=  1,
276 277
};

278
struct file;
279 280
struct perf_sample_data;

281
typedef void (*perf_overflow_handler_t)(struct perf_event *,
282 283 284
					struct perf_sample_data *,
					struct pt_regs *regs);

285
enum perf_group_flag {
286
	PERF_GROUP_SOFTWARE		= 0x1,
287 288
};

289 290
#define SWEVENT_HLIST_BITS		8
#define SWEVENT_HLIST_SIZE		(1 << SWEVENT_HLIST_BITS)
291 292

struct swevent_hlist {
293 294
	struct hlist_head		heads[SWEVENT_HLIST_SIZE];
	struct rcu_head			rcu_head;
295 296
};

297 298
#define PERF_ATTACH_CONTEXT	0x01
#define PERF_ATTACH_GROUP	0x02
299
#define PERF_ATTACH_TASK	0x04
300

301
struct perf_cgroup;
302 303
struct ring_buffer;

T
Thomas Gleixner 已提交
304
/**
305
 * struct perf_event - performance event kernel representation:
T
Thomas Gleixner 已提交
306
 */
307 308
struct perf_event {
#ifdef CONFIG_PERF_EVENTS
309 310 311 312 313
	/*
	 * entry onto perf_event_context::event_list;
	 *   modifications require ctx->lock
	 *   RCU safe iterations.
	 */
P
Peter Zijlstra 已提交
314
	struct list_head		event_entry;
315 316 317 318 319 320 321 322 323 324

	/*
	 * 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;
325
	struct list_head		sibling_list;
326 327 328 329 330 331 332 333

	/*
	 * 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;

334 335
	struct hlist_node		hlist_entry;
	struct list_head		active_entry;
I
Ingo Molnar 已提交
336
	int				nr_siblings;
337
	int				group_flags;
338
	struct perf_event		*group_leader;
P
Peter Zijlstra 已提交
339
	struct pmu			*pmu;
340

341
	enum perf_event_active_state	state;
342
	unsigned int			attach_state;
343
	local64_t			count;
344
	atomic64_t			child_count;
345

346
	/*
347
	 * These are the total time in nanoseconds that the event
348
	 * has been enabled (i.e. eligible to run, and the task has
349
	 * been scheduled in, if this is a per-task event)
350 351 352
	 * and running (scheduled onto the CPU), respectively.
	 *
	 * They are computed from tstamp_enabled, tstamp_running and
353
	 * tstamp_stopped when the event is in INACTIVE or ACTIVE state.
354 355 356 357 358 359
	 */
	u64				total_time_enabled;
	u64				total_time_running;

	/*
	 * These are timestamps used for computing total_time_enabled
360
	 * and total_time_running when the event is in INACTIVE or
361 362
	 * ACTIVE state, measured in nanoseconds from an arbitrary point
	 * in time.
363 364
	 * tstamp_enabled: the notional time when the event was enabled
	 * tstamp_running: the notional time when the event was scheduled on
365
	 * tstamp_stopped: in INACTIVE state, the notional time when the
366
	 *	event was scheduled off.
367 368 369 370 371
	 */
	u64				tstamp_enabled;
	u64				tstamp_running;
	u64				tstamp_stopped;

372 373 374 375 376 377 378 379 380 381
	/*
	 * 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;

382
	struct perf_event_attr		attr;
383
	u16				header_size;
384
	u16				id_header_size;
385
	u16				read_size;
386
	struct hw_perf_event		hw;
T
Thomas Gleixner 已提交
387

388
	struct perf_event_context	*ctx;
389
	atomic_long_t			refcount;
T
Thomas Gleixner 已提交
390

391 392
	/*
	 * These accumulate total time (in nanoseconds) that children
393
	 * events have been enabled and running, respectively.
394 395 396 397
	 */
	atomic64_t			child_total_time_enabled;
	atomic64_t			child_total_time_running;

T
Thomas Gleixner 已提交
398
	/*
399
	 * Protect attach/detach and child_list:
T
Thomas Gleixner 已提交
400
	 */
401 402
	struct mutex			child_mutex;
	struct list_head		child_list;
403
	struct perf_event		*parent;
T
Thomas Gleixner 已提交
404 405 406 407

	int				oncpu;
	int				cpu;

408 409 410
	struct list_head		owner_entry;
	struct task_struct		*owner;

411 412 413
	/* mmap bits */
	struct mutex			mmap_mutex;
	atomic_t			mmap_count;
P
Peter Zijlstra 已提交
414

415
	struct ring_buffer		*rb;
416
	struct list_head		rb_entry;
417 418
	unsigned long			rcu_batches;
	int				rcu_pending;
419

420
	/* poll related */
T
Thomas Gleixner 已提交
421
	wait_queue_head_t		waitq;
P
Peter Zijlstra 已提交
422
	struct fasync_struct		*fasync;
423 424 425

	/* delayed work for NMIs and such */
	int				pending_wakeup;
426
	int				pending_kill;
427
	int				pending_disable;
428
	struct irq_work			pending;
P
Peter Zijlstra 已提交
429

430 431
	atomic_t			event_limit;

432
	void (*destroy)(struct perf_event *);
P
Peter Zijlstra 已提交
433
	struct rcu_head			rcu_head;
434 435

	struct pid_namespace		*ns;
436
	u64				id;
L
Li Zefan 已提交
437

438
	perf_overflow_handler_t		overflow_handler;
439
	void				*overflow_handler_context;
440

441
#ifdef CONFIG_EVENT_TRACING
442
	struct ftrace_event_call	*tp_event;
L
Li Zefan 已提交
443
	struct event_filter		*filter;
444 445 446
#ifdef CONFIG_FUNCTION_TRACER
	struct ftrace_ops               ftrace_ops;
#endif
447
#endif
L
Li Zefan 已提交
448

S
Stephane Eranian 已提交
449 450 451 452 453
#ifdef CONFIG_CGROUP_PERF
	struct perf_cgroup		*cgrp; /* cgroup event is attach to */
	int				cgrp_defer_enabled;
#endif

L
Li Zefan 已提交
454
#endif /* CONFIG_PERF_EVENTS */
T
Thomas Gleixner 已提交
455 456
};

457 458 459 460 461
enum perf_event_context_type {
	task_context,
	cpu_context,
};

T
Thomas Gleixner 已提交
462
/**
463
 * struct perf_event_context - event context structure
T
Thomas Gleixner 已提交
464
 *
465
 * Used as a container for task events and CPU events as well:
T
Thomas Gleixner 已提交
466
 */
467
struct perf_event_context {
P
Peter Zijlstra 已提交
468
	struct pmu			*pmu;
469
	enum perf_event_context_type	type;
T
Thomas Gleixner 已提交
470
	/*
471
	 * Protect the states of the events in the list,
472
	 * nr_active, and the list:
T
Thomas Gleixner 已提交
473
	 */
474
	raw_spinlock_t			lock;
475
	/*
476
	 * Protect the list of events.  Locking either mutex or lock
477 478 479
	 * 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
480
	struct mutex			mutex;
481

482 483
	struct list_head		pinned_groups;
	struct list_head		flexible_groups;
I
Ingo Molnar 已提交
484
	struct list_head		event_list;
485
	int				nr_events;
I
Ingo Molnar 已提交
486 487
	int				nr_active;
	int				is_active;
488
	int				nr_stat;
489
	int				nr_freq;
490
	int				rotate_disable;
I
Ingo Molnar 已提交
491 492
	atomic_t			refcount;
	struct task_struct		*task;
493 494

	/*
495
	 * Context clock, runs when context enabled.
496
	 */
I
Ingo Molnar 已提交
497 498
	u64				time;
	u64				timestamp;
499 500 501 502 503

	/*
	 * These fields let us detect when two contexts have both
	 * been cloned (inherited) from a common ancestor.
	 */
504
	struct perf_event_context	*parent_ctx;
I
Ingo Molnar 已提交
505 506 507
	u64				parent_gen;
	u64				generation;
	int				pin_count;
508 509
	int				nr_cgroups;	 /* cgroup evts */
	int				nr_branch_stack; /* branch_stack evt */
510
	struct rcu_head			rcu_head;
511 512 513

	struct delayed_work		orphans_remove;
	bool				orphans_remove_sched;
T
Thomas Gleixner 已提交
514 515
};

516 517
/*
 * Number of contexts where an event can trigger:
518
 *	task, softirq, hardirq, nmi.
519 520 521
 */
#define PERF_NR_CONTEXTS	4

T
Thomas Gleixner 已提交
522
/**
523
 * struct perf_event_cpu_context - per cpu event context structure
T
Thomas Gleixner 已提交
524 525
 */
struct perf_cpu_context {
526 527
	struct perf_event_context	ctx;
	struct perf_event_context	*task_ctx;
T
Thomas Gleixner 已提交
528
	int				active_oncpu;
529
	int				exclusive;
530 531
	struct hrtimer			hrtimer;
	ktime_t				hrtimer_interval;
532
	struct list_head		rotation_list;
533
	struct pmu			*unique_pmu;
S
Stephane Eranian 已提交
534
	struct perf_cgroup		*cgrp;
T
Thomas Gleixner 已提交
535 536
};

537
struct perf_output_handle {
I
Ingo Molnar 已提交
538
	struct perf_event		*event;
539
	struct ring_buffer		*rb;
540
	unsigned long			wakeup;
541 542 543
	unsigned long			size;
	void				*addr;
	int				page;
544 545
};

546
#ifdef CONFIG_PERF_EVENTS
547

548
extern int perf_pmu_register(struct pmu *pmu, const char *name, int type);
549
extern void perf_pmu_unregister(struct pmu *pmu);
I
Ingo Molnar 已提交
550

551
extern int perf_num_counters(void);
552
extern const char *perf_pmu_name(void);
553 554 555 556
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);
557 558 559
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);
560
extern void perf_event_delayed_put(struct task_struct *task);
561
extern void perf_event_print_debug(void);
P
Peter Zijlstra 已提交
562 563
extern void perf_pmu_disable(struct pmu *pmu);
extern void perf_pmu_enable(struct pmu *pmu);
564 565
extern int perf_event_task_disable(void);
extern int perf_event_task_enable(void);
566
extern int perf_event_refresh(struct perf_event *event, int refresh);
567
extern void perf_event_update_userpage(struct perf_event *event);
568 569 570 571
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
572
				struct task_struct *task,
573 574
				perf_overflow_handler_t callback,
				void *context);
575 576
extern void perf_pmu_migrate_context(struct pmu *pmu,
				int src_cpu, int dst_cpu);
577 578
extern u64 perf_event_read_value(struct perf_event *event,
				 u64 *enabled, u64 *running);
579

580

581
struct perf_sample_data {
582 583 584 585 586 587 588 589
	u64				type;

	u64				ip;
	struct {
		u32	pid;
		u32	tid;
	}				tid_entry;
	u64				time;
I
Ingo Molnar 已提交
590
	u64				addr;
591 592 593 594 595 596
	u64				id;
	u64				stream_id;
	struct {
		u32	cpu;
		u32	reserved;
	}				cpu_entry;
I
Ingo Molnar 已提交
597
	u64				period;
598
	union  perf_mem_data_src	data_src;
599
	struct perf_callchain_entry	*callchain;
600
	struct perf_raw_record		*raw;
601
	struct perf_branch_stack	*br_stack;
602
	struct perf_regs_user		regs_user;
603
	u64				stack_user_size;
A
Andi Kleen 已提交
604
	u64				weight;
A
Andi Kleen 已提交
605 606 607 608
	/*
	 * Transaction flags for abort events:
	 */
	u64				txn;
609 610
};

611 612
static inline void perf_sample_data_init(struct perf_sample_data *data,
					 u64 addr, u64 period)
613
{
614
	/* remaining struct members initialized in perf_prepare_sample() */
615 616
	data->addr = addr;
	data->raw  = NULL;
617
	data->br_stack = NULL;
618 619 620
	data->period = period;
	data->regs_user.abi = PERF_SAMPLE_REGS_ABI_NONE;
	data->regs_user.regs = NULL;
621
	data->stack_user_size = 0;
A
Andi Kleen 已提交
622
	data->weight = 0;
623
	data->data_src.val = 0;
A
Andi Kleen 已提交
624
	data->txn = 0;
625 626
}

627 628 629
extern void perf_output_sample(struct perf_output_handle *handle,
			       struct perf_event_header *header,
			       struct perf_sample_data *data,
630
			       struct perf_event *event);
631 632
extern void perf_prepare_sample(struct perf_event_header *header,
				struct perf_sample_data *data,
633
				struct perf_event *event,
634 635
				struct pt_regs *regs);

636
extern int perf_event_overflow(struct perf_event *event,
637 638
				 struct perf_sample_data *data,
				 struct pt_regs *regs);
639

640 641 642 643 644
static inline bool is_sampling_event(struct perf_event *event)
{
	return event->attr.sample_period != 0;
}

645
/*
646
 * Return 1 for a software event, 0 for a hardware event
647
 */
648
static inline int is_software_event(struct perf_event *event)
649
{
650
	return event->pmu->task_ctx_nr == perf_sw_context;
651 652
}

653
extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX];
654

655
extern void __perf_sw_event(u32, u64, struct pt_regs *, u64);
656

657
#ifndef perf_arch_fetch_caller_regs
658
static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { }
659
#endif
660 661 662 663 664 665 666 667 668

/*
 * 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
 */
669
static inline void perf_fetch_caller_regs(struct pt_regs *regs)
670 671 672
{
	memset(regs, 0, sizeof(*regs));

673
	perf_arch_fetch_caller_regs(regs, CALLER_ADDR0);
674 675
}

P
Peter Zijlstra 已提交
676
static __always_inline void
677
perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
678
{
P
Peter Zijlstra 已提交
679 680
	struct pt_regs hot_regs;

681
	if (static_key_false(&perf_swevent_enabled[event_id])) {
682 683 684 685
		if (!regs) {
			perf_fetch_caller_regs(&hot_regs);
			regs = &hot_regs;
		}
686
		__perf_sw_event(event_id, nr, regs, addr);
687 688 689
	}
}

690
extern struct static_key_deferred perf_sched_events;
691

692
static inline void perf_event_task_sched_in(struct task_struct *prev,
693
					    struct task_struct *task)
694 695 696 697 698 699 700
{
	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)
701
{
702
	perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, NULL, 0);
703

704
	if (static_key_false(&perf_sched_events.key))
705
		__perf_event_task_sched_out(prev, next);
706 707
}

708
extern void perf_event_mmap(struct vm_area_struct *vma);
709
extern struct perf_guest_info_callbacks *perf_guest_cbs;
710 711
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);
712

713
extern void perf_event_exec(void);
714
extern void perf_event_comm(struct task_struct *tsk, bool exec);
715
extern void perf_event_fork(struct task_struct *tsk);
716

717 718 719
/* Callchains */
DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry);

720 721
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);
722

723
static inline void perf_callchain_store(struct perf_callchain_entry *entry, u64 ip)
724 725 726 727
{
	if (entry->nr < PERF_MAX_STACK_DEPTH)
		entry->ip[entry->nr++] = ip;
}
728

729 730 731
extern int sysctl_perf_event_paranoid;
extern int sysctl_perf_event_mlock;
extern int sysctl_perf_event_sample_rate;
732 733 734
extern int sysctl_perf_cpu_time_max_percent;

extern void perf_sample_event_took(u64 sample_len_ns);
735

P
Peter Zijlstra 已提交
736 737 738
extern int perf_proc_update_handler(struct ctl_table *table, int write,
		void __user *buffer, size_t *lenp,
		loff_t *ppos);
739 740 741 742
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
743

744 745 746 747 748 749 750 751 752 753 754 755 756 757 758
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;
}

759
extern void perf_event_init(void);
760 761
extern void perf_tp_event(u64 addr, u64 count, void *record,
			  int entry_size, struct pt_regs *regs,
762 763
			  struct hlist_head *head, int rctx,
			  struct task_struct *task);
764
extern void perf_bp_event(struct perf_event *event, void *data);
765

766
#ifndef perf_misc_flags
767 768 769
# define perf_misc_flags(regs) \
		(user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL)
# define perf_instruction_pointer(regs)	instruction_pointer(regs)
770 771
#endif

772 773 774 775 776
static inline bool has_branch_stack(struct perf_event *event)
{
	return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK;
}

777
extern int perf_output_begin(struct perf_output_handle *handle,
778
			     struct perf_event *event, unsigned int size);
779
extern void perf_output_end(struct perf_output_handle *handle);
780
extern unsigned int perf_output_copy(struct perf_output_handle *handle,
781
			     const void *buf, unsigned int len);
782 783
extern unsigned int perf_output_skip(struct perf_output_handle *handle,
				     unsigned int len);
784 785
extern int perf_swevent_get_recursion_context(void);
extern void perf_swevent_put_recursion_context(int rctx);
786
extern u64 perf_swevent_set_period(struct perf_event *event);
787 788
extern void perf_event_enable(struct perf_event *event);
extern void perf_event_disable(struct perf_event *event);
789
extern int __perf_event_disable(void *info);
790
extern void perf_event_task_tick(void);
791
#else /* !CONFIG_PERF_EVENTS: */
T
Thomas Gleixner 已提交
792
static inline void
793 794 795 796 797
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)			{ }
798 799 800
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)	{ }
801
static inline void perf_event_delayed_put(struct task_struct *task)	{ }
I
Ingo Molnar 已提交
802 803 804
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; }
805 806 807 808
static inline int perf_event_refresh(struct perf_event *event, int refresh)
{
	return -EINVAL;
}
809

810
static inline void
811
perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)	{ }
812
static inline void
813
perf_bp_event(struct perf_event *event, void *data)			{ }
814

815
static inline int perf_register_guest_info_callbacks
816
(struct perf_guest_info_callbacks *callbacks)				{ return 0; }
817
static inline int perf_unregister_guest_info_callbacks
818
(struct perf_guest_info_callbacks *callbacks)				{ return 0; }
819

I
Ingo Molnar 已提交
820
static inline void perf_event_mmap(struct vm_area_struct *vma)		{ }
821
static inline void perf_event_exec(void)				{ }
822
static inline void perf_event_comm(struct task_struct *tsk, bool exec)	{ }
823 824
static inline void perf_event_fork(struct task_struct *tsk)		{ }
static inline void perf_event_init(void)				{ }
825
static inline int  perf_swevent_get_recursion_context(void)		{ return -1; }
826
static inline void perf_swevent_put_recursion_context(int rctx)		{ }
827
static inline u64 perf_swevent_set_period(struct perf_event *event)	{ return 0; }
828 829
static inline void perf_event_enable(struct perf_event *event)		{ }
static inline void perf_event_disable(struct perf_event *event)		{ }
830
static inline int __perf_event_disable(void *info)			{ return -1; }
831
static inline void perf_event_task_tick(void)				{ }
T
Thomas Gleixner 已提交
832 833
#endif

834 835 836 837 838 839
#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

840 841 842
#if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
extern void perf_restore_debug_store(void);
#else
843
static inline void perf_restore_debug_store(void)			{ }
T
Thomas Gleixner 已提交
844 845
#endif

846
#define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x))
847

848
/*
849
 * This has to have a higher priority than migration_notifier in sched/core.c.
850
 */
851 852
#define perf_cpu_notifier(fn)						\
do {									\
853
	static struct notifier_block fn##_nb =				\
854
		{ .notifier_call = fn, .priority = CPU_PRI_PERF };	\
855
	unsigned long cpu = smp_processor_id();				\
856
	unsigned long flags;						\
857 858
									\
	cpu_notifier_register_begin();					\
859
	fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE,			\
860
		(void *)(unsigned long)cpu);				\
861
	local_irq_save(flags);						\
862
	fn(&fn##_nb, (unsigned long)CPU_STARTING,			\
863
		(void *)(unsigned long)cpu);				\
864
	local_irq_restore(flags);					\
865
	fn(&fn##_nb, (unsigned long)CPU_ONLINE,				\
866
		(void *)(unsigned long)cpu);				\
867 868
	__register_cpu_notifier(&fn##_nb);				\
	cpu_notifier_register_done();					\
869 870
} while (0)

871 872 873 874 875 876 877 878 879 880 881
/*
 * Bare-bones version of perf_cpu_notifier(), which doesn't invoke the
 * callback for already online CPUs.
 */
#define __perf_cpu_notifier(fn)						\
do {									\
	static struct notifier_block fn##_nb =				\
		{ .notifier_call = fn, .priority = CPU_PRI_PERF };	\
									\
	__register_cpu_notifier(&fn##_nb);				\
} while (0)
882

883 884 885
struct perf_pmu_events_attr {
	struct device_attribute attr;
	u64 id;
886
	const char *event_str;
887 888 889 890 891 892 893 894
};

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

895 896 897 898 899 900 901 902 903 904 905 906
#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)

907
#endif /* _LINUX_PERF_EVENT_H */