perf_event.h 24.7 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

169 170 171 172 173
/**
 * pmu::capabilities flags
 */
#define PERF_PMU_CAP_NO_INTERRUPT		0x01

I
Ingo Molnar 已提交
174
/**
175
 * struct pmu - generic performance monitoring unit
I
Ingo Molnar 已提交
176
 */
177
struct pmu {
178 179
	struct list_head		entry;

180
	struct module			*module;
P
Peter Zijlstra 已提交
181
	struct device			*dev;
182
	const struct attribute_group	**attr_groups;
183
	const char			*name;
P
Peter Zijlstra 已提交
184 185
	int				type;

186 187 188 189 190
	/*
	 * various common per-pmu feature flags
	 */
	int				capabilities;

P
Peter Zijlstra 已提交
191 192
	int * __percpu			pmu_disable_count;
	struct perf_cpu_context * __percpu pmu_cpu_context;
P
Peter Zijlstra 已提交
193
	int				task_ctx_nr;
194
	int				hrtimer_interval_ms;
195 196

	/*
P
Peter Zijlstra 已提交
197 198
	 * Fully disable/enable this PMU, can be used to protect from the PMI
	 * as well as for lazy/batch writing of the MSRs.
199
	 */
P
Peter Zijlstra 已提交
200 201
	void (*pmu_enable)		(struct pmu *pmu); /* optional */
	void (*pmu_disable)		(struct pmu *pmu); /* optional */
202

203
	/*
P
Peter Zijlstra 已提交
204
	 * Try and initialize the event for this PMU.
205
	 * Should return -ENOENT when the @event doesn't match this PMU.
206
	 */
207 208
	int (*event_init)		(struct perf_event *event);

P
Peter Zijlstra 已提交
209 210 211 212
#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 */

213
	/*
P
Peter Zijlstra 已提交
214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230
	 * 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.
	 */
231
	void (*read)			(struct perf_event *event);
232 233

	/*
234 235 236
	 * 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
237 238
	 *
	 * Start the transaction, after this ->add() doesn't need to
239
	 * do schedulability tests.
240
	 */
241
	void (*start_txn)		(struct pmu *pmu); /* optional */
242
	/*
P
Peter Zijlstra 已提交
243
	 * If ->start_txn() disabled the ->add() schedulability test
244 245 246 247
	 * 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.
	 */
248
	int  (*commit_txn)		(struct pmu *pmu); /* optional */
249
	/*
P
Peter Zijlstra 已提交
250
	 * Will cancel the transaction, assumes ->del() is called
L
Lucas De Marchi 已提交
251
	 * for each successful ->add() during the transaction.
252
	 */
253
	void (*cancel_txn)		(struct pmu *pmu); /* optional */
254 255 256 257 258 259

	/*
	 * 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 */
260 261 262 263 264

	/*
	 * flush branch stack on context-switches (needed in cpu-wide mode)
	 */
	void (*flush_branch_stack)	(void);
I
Ingo Molnar 已提交
265 266
};

267
/**
268
 * enum perf_event_active_state - the states of a event
269
 */
270
enum perf_event_active_state {
I
Ingo Molnar 已提交
271
	PERF_EVENT_STATE_ERROR		= -2,
272 273
	PERF_EVENT_STATE_OFF		= -1,
	PERF_EVENT_STATE_INACTIVE	=  0,
I
Ingo Molnar 已提交
274
	PERF_EVENT_STATE_ACTIVE		=  1,
275 276
};

277
struct file;
278 279
struct perf_sample_data;

280
typedef void (*perf_overflow_handler_t)(struct perf_event *,
281 282 283
					struct perf_sample_data *,
					struct pt_regs *regs);

284
enum perf_group_flag {
285
	PERF_GROUP_SOFTWARE		= 0x1,
286 287
};

288 289
#define SWEVENT_HLIST_BITS		8
#define SWEVENT_HLIST_SIZE		(1 << SWEVENT_HLIST_BITS)
290 291

struct swevent_hlist {
292 293
	struct hlist_head		heads[SWEVENT_HLIST_SIZE];
	struct rcu_head			rcu_head;
294 295
};

296 297
#define PERF_ATTACH_CONTEXT	0x01
#define PERF_ATTACH_GROUP	0x02
298
#define PERF_ATTACH_TASK	0x04
299

300
struct perf_cgroup;
301 302
struct ring_buffer;

T
Thomas Gleixner 已提交
303
/**
304
 * struct perf_event - performance event kernel representation:
T
Thomas Gleixner 已提交
305
 */
306 307
struct perf_event {
#ifdef CONFIG_PERF_EVENTS
308 309 310 311 312
	/*
	 * entry onto perf_event_context::event_list;
	 *   modifications require ctx->lock
	 *   RCU safe iterations.
	 */
P
Peter Zijlstra 已提交
313
	struct list_head		event_entry;
314 315 316 317 318 319 320 321 322 323

	/*
	 * 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;
324
	struct list_head		sibling_list;
325 326 327 328 329 330 331 332

	/*
	 * 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;

333 334
	struct hlist_node		hlist_entry;
	struct list_head		active_entry;
I
Ingo Molnar 已提交
335
	int				nr_siblings;
336
	int				group_flags;
337
	struct perf_event		*group_leader;
P
Peter Zijlstra 已提交
338
	struct pmu			*pmu;
339

340
	enum perf_event_active_state	state;
341
	unsigned int			attach_state;
342
	local64_t			count;
343
	atomic64_t			child_count;
344

345
	/*
346
	 * These are the total time in nanoseconds that the event
347
	 * has been enabled (i.e. eligible to run, and the task has
348
	 * been scheduled in, if this is a per-task event)
349 350 351
	 * and running (scheduled onto the CPU), respectively.
	 *
	 * They are computed from tstamp_enabled, tstamp_running and
352
	 * tstamp_stopped when the event is in INACTIVE or ACTIVE state.
353 354 355 356 357 358
	 */
	u64				total_time_enabled;
	u64				total_time_running;

	/*
	 * These are timestamps used for computing total_time_enabled
359
	 * and total_time_running when the event is in INACTIVE or
360 361
	 * ACTIVE state, measured in nanoseconds from an arbitrary point
	 * in time.
362 363
	 * tstamp_enabled: the notional time when the event was enabled
	 * tstamp_running: the notional time when the event was scheduled on
364
	 * tstamp_stopped: in INACTIVE state, the notional time when the
365
	 *	event was scheduled off.
366 367 368 369 370
	 */
	u64				tstamp_enabled;
	u64				tstamp_running;
	u64				tstamp_stopped;

371 372 373 374 375 376 377 378 379 380
	/*
	 * 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;

381
	struct perf_event_attr		attr;
382
	u16				header_size;
383
	u16				id_header_size;
384
	u16				read_size;
385
	struct hw_perf_event		hw;
T
Thomas Gleixner 已提交
386

387
	struct perf_event_context	*ctx;
388
	atomic_long_t			refcount;
T
Thomas Gleixner 已提交
389

390 391
	/*
	 * These accumulate total time (in nanoseconds) that children
392
	 * events have been enabled and running, respectively.
393 394 395 396
	 */
	atomic64_t			child_total_time_enabled;
	atomic64_t			child_total_time_running;

T
Thomas Gleixner 已提交
397
	/*
398
	 * Protect attach/detach and child_list:
T
Thomas Gleixner 已提交
399
	 */
400 401
	struct mutex			child_mutex;
	struct list_head		child_list;
402
	struct perf_event		*parent;
T
Thomas Gleixner 已提交
403 404 405 406

	int				oncpu;
	int				cpu;

407 408 409
	struct list_head		owner_entry;
	struct task_struct		*owner;

410 411 412
	/* mmap bits */
	struct mutex			mmap_mutex;
	atomic_t			mmap_count;
P
Peter Zijlstra 已提交
413

414
	struct ring_buffer		*rb;
415
	struct list_head		rb_entry;
416 417
	unsigned long			rcu_batches;
	int				rcu_pending;
418

419
	/* poll related */
T
Thomas Gleixner 已提交
420
	wait_queue_head_t		waitq;
P
Peter Zijlstra 已提交
421
	struct fasync_struct		*fasync;
422 423 424

	/* delayed work for NMIs and such */
	int				pending_wakeup;
425
	int				pending_kill;
426
	int				pending_disable;
427
	struct irq_work			pending;
P
Peter Zijlstra 已提交
428

429 430
	atomic_t			event_limit;

431
	void (*destroy)(struct perf_event *);
P
Peter Zijlstra 已提交
432
	struct rcu_head			rcu_head;
433 434

	struct pid_namespace		*ns;
435
	u64				id;
L
Li Zefan 已提交
436

437
	perf_overflow_handler_t		overflow_handler;
438
	void				*overflow_handler_context;
439

440
#ifdef CONFIG_EVENT_TRACING
441
	struct ftrace_event_call	*tp_event;
L
Li Zefan 已提交
442
	struct event_filter		*filter;
443 444 445
#ifdef CONFIG_FUNCTION_TRACER
	struct ftrace_ops               ftrace_ops;
#endif
446
#endif
L
Li Zefan 已提交
447

S
Stephane Eranian 已提交
448 449 450 451 452
#ifdef CONFIG_CGROUP_PERF
	struct perf_cgroup		*cgrp; /* cgroup event is attach to */
	int				cgrp_defer_enabled;
#endif

L
Li Zefan 已提交
453
#endif /* CONFIG_PERF_EVENTS */
T
Thomas Gleixner 已提交
454 455
};

456 457 458 459 460
enum perf_event_context_type {
	task_context,
	cpu_context,
};

T
Thomas Gleixner 已提交
461
/**
462
 * struct perf_event_context - event context structure
T
Thomas Gleixner 已提交
463
 *
464
 * Used as a container for task events and CPU events as well:
T
Thomas Gleixner 已提交
465
 */
466
struct perf_event_context {
P
Peter Zijlstra 已提交
467
	struct pmu			*pmu;
468
	enum perf_event_context_type	type;
T
Thomas Gleixner 已提交
469
	/*
470
	 * Protect the states of the events in the list,
471
	 * nr_active, and the list:
T
Thomas Gleixner 已提交
472
	 */
473
	raw_spinlock_t			lock;
474
	/*
475
	 * Protect the list of events.  Locking either mutex or lock
476 477 478
	 * 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
479
	struct mutex			mutex;
480

481 482
	struct list_head		pinned_groups;
	struct list_head		flexible_groups;
I
Ingo Molnar 已提交
483
	struct list_head		event_list;
484
	int				nr_events;
I
Ingo Molnar 已提交
485 486
	int				nr_active;
	int				is_active;
487
	int				nr_stat;
488
	int				nr_freq;
489
	int				rotate_disable;
I
Ingo Molnar 已提交
490 491
	atomic_t			refcount;
	struct task_struct		*task;
492 493

	/*
494
	 * Context clock, runs when context enabled.
495
	 */
I
Ingo Molnar 已提交
496 497
	u64				time;
	u64				timestamp;
498 499 500 501 502

	/*
	 * These fields let us detect when two contexts have both
	 * been cloned (inherited) from a common ancestor.
	 */
503
	struct perf_event_context	*parent_ctx;
I
Ingo Molnar 已提交
504 505 506
	u64				parent_gen;
	u64				generation;
	int				pin_count;
507 508
	int				nr_cgroups;	 /* cgroup evts */
	int				nr_branch_stack; /* branch_stack evt */
509
	struct rcu_head			rcu_head;
T
Thomas Gleixner 已提交
510 511
};

512 513
/*
 * Number of contexts where an event can trigger:
514
 *	task, softirq, hardirq, nmi.
515 516 517
 */
#define PERF_NR_CONTEXTS	4

T
Thomas Gleixner 已提交
518
/**
519
 * struct perf_event_cpu_context - per cpu event context structure
T
Thomas Gleixner 已提交
520 521
 */
struct perf_cpu_context {
522 523
	struct perf_event_context	ctx;
	struct perf_event_context	*task_ctx;
T
Thomas Gleixner 已提交
524
	int				active_oncpu;
525
	int				exclusive;
526 527
	struct hrtimer			hrtimer;
	ktime_t				hrtimer_interval;
528
	struct list_head		rotation_list;
529
	struct pmu			*unique_pmu;
S
Stephane Eranian 已提交
530
	struct perf_cgroup		*cgrp;
T
Thomas Gleixner 已提交
531 532
};

533
struct perf_output_handle {
I
Ingo Molnar 已提交
534
	struct perf_event		*event;
535
	struct ring_buffer		*rb;
536
	unsigned long			wakeup;
537 538 539
	unsigned long			size;
	void				*addr;
	int				page;
540 541
};

542
#ifdef CONFIG_PERF_EVENTS
543

544
extern int perf_pmu_register(struct pmu *pmu, const char *name, int type);
545
extern void perf_pmu_unregister(struct pmu *pmu);
I
Ingo Molnar 已提交
546

547
extern int perf_num_counters(void);
548
extern const char *perf_pmu_name(void);
549 550 551 552
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);
553 554 555
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);
556
extern void perf_event_delayed_put(struct task_struct *task);
557
extern void perf_event_print_debug(void);
P
Peter Zijlstra 已提交
558 559
extern void perf_pmu_disable(struct pmu *pmu);
extern void perf_pmu_enable(struct pmu *pmu);
560 561
extern int perf_event_task_disable(void);
extern int perf_event_task_enable(void);
562
extern int perf_event_refresh(struct perf_event *event, int refresh);
563
extern void perf_event_update_userpage(struct perf_event *event);
564 565 566 567
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
568
				struct task_struct *task,
569 570
				perf_overflow_handler_t callback,
				void *context);
571 572
extern void perf_pmu_migrate_context(struct pmu *pmu,
				int src_cpu, int dst_cpu);
573 574
extern u64 perf_event_read_value(struct perf_event *event,
				 u64 *enabled, u64 *running);
575

576

577
struct perf_sample_data {
578 579 580 581 582 583 584 585
	u64				type;

	u64				ip;
	struct {
		u32	pid;
		u32	tid;
	}				tid_entry;
	u64				time;
I
Ingo Molnar 已提交
586
	u64				addr;
587 588 589 590 591 592
	u64				id;
	u64				stream_id;
	struct {
		u32	cpu;
		u32	reserved;
	}				cpu_entry;
I
Ingo Molnar 已提交
593
	u64				period;
594
	union  perf_mem_data_src	data_src;
595
	struct perf_callchain_entry	*callchain;
596
	struct perf_raw_record		*raw;
597
	struct perf_branch_stack	*br_stack;
598
	struct perf_regs_user		regs_user;
599
	u64				stack_user_size;
A
Andi Kleen 已提交
600
	u64				weight;
A
Andi Kleen 已提交
601 602 603 604
	/*
	 * Transaction flags for abort events:
	 */
	u64				txn;
605 606
};

607 608
static inline void perf_sample_data_init(struct perf_sample_data *data,
					 u64 addr, u64 period)
609
{
610
	/* remaining struct members initialized in perf_prepare_sample() */
611 612
	data->addr = addr;
	data->raw  = NULL;
613
	data->br_stack = NULL;
614 615 616
	data->period = period;
	data->regs_user.abi = PERF_SAMPLE_REGS_ABI_NONE;
	data->regs_user.regs = NULL;
617
	data->stack_user_size = 0;
A
Andi Kleen 已提交
618
	data->weight = 0;
619
	data->data_src.val = 0;
A
Andi Kleen 已提交
620
	data->txn = 0;
621 622
}

623 624 625
extern void perf_output_sample(struct perf_output_handle *handle,
			       struct perf_event_header *header,
			       struct perf_sample_data *data,
626
			       struct perf_event *event);
627 628
extern void perf_prepare_sample(struct perf_event_header *header,
				struct perf_sample_data *data,
629
				struct perf_event *event,
630 631
				struct pt_regs *regs);

632
extern int perf_event_overflow(struct perf_event *event,
633 634
				 struct perf_sample_data *data,
				 struct pt_regs *regs);
635

636 637 638 639 640
static inline bool is_sampling_event(struct perf_event *event)
{
	return event->attr.sample_period != 0;
}

641
/*
642
 * Return 1 for a software event, 0 for a hardware event
643
 */
644
static inline int is_software_event(struct perf_event *event)
645
{
646
	return event->pmu->task_ctx_nr == perf_sw_context;
647 648
}

649
extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX];
650

651
extern void __perf_sw_event(u32, u64, struct pt_regs *, u64);
652

653
#ifndef perf_arch_fetch_caller_regs
654
static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { }
655
#endif
656 657 658 659 660 661 662 663 664

/*
 * 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
 */
665
static inline void perf_fetch_caller_regs(struct pt_regs *regs)
666 667 668
{
	memset(regs, 0, sizeof(*regs));

669
	perf_arch_fetch_caller_regs(regs, CALLER_ADDR0);
670 671
}

P
Peter Zijlstra 已提交
672
static __always_inline void
673
perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
674
{
P
Peter Zijlstra 已提交
675 676
	struct pt_regs hot_regs;

677
	if (static_key_false(&perf_swevent_enabled[event_id])) {
678 679 680 681
		if (!regs) {
			perf_fetch_caller_regs(&hot_regs);
			regs = &hot_regs;
		}
682
		__perf_sw_event(event_id, nr, regs, addr);
683 684 685
	}
}

686
extern struct static_key_deferred perf_sched_events;
687

688
static inline void perf_event_task_sched_in(struct task_struct *prev,
689
					    struct task_struct *task)
690 691 692 693 694 695 696
{
	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)
697
{
698
	perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, NULL, 0);
699

700
	if (static_key_false(&perf_sched_events.key))
701
		__perf_event_task_sched_out(prev, next);
702 703
}

704
extern void perf_event_mmap(struct vm_area_struct *vma);
705
extern struct perf_guest_info_callbacks *perf_guest_cbs;
706 707
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);
708

709
extern void perf_event_exec(void);
710
extern void perf_event_comm(struct task_struct *tsk, bool exec);
711
extern void perf_event_fork(struct task_struct *tsk);
712

713 714 715
/* Callchains */
DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry);

716 717
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);
718

719
static inline void perf_callchain_store(struct perf_callchain_entry *entry, u64 ip)
720 721 722 723
{
	if (entry->nr < PERF_MAX_STACK_DEPTH)
		entry->ip[entry->nr++] = ip;
}
724

725 726 727
extern int sysctl_perf_event_paranoid;
extern int sysctl_perf_event_mlock;
extern int sysctl_perf_event_sample_rate;
728 729 730
extern int sysctl_perf_cpu_time_max_percent;

extern void perf_sample_event_took(u64 sample_len_ns);
731

P
Peter Zijlstra 已提交
732 733 734
extern int perf_proc_update_handler(struct ctl_table *table, int write,
		void __user *buffer, size_t *lenp,
		loff_t *ppos);
735 736 737 738
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
739

740 741 742 743 744 745 746 747 748 749 750 751 752 753 754
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;
}

755
extern void perf_event_init(void);
756 757
extern void perf_tp_event(u64 addr, u64 count, void *record,
			  int entry_size, struct pt_regs *regs,
758 759
			  struct hlist_head *head, int rctx,
			  struct task_struct *task);
760
extern void perf_bp_event(struct perf_event *event, void *data);
761

762
#ifndef perf_misc_flags
763 764 765
# define perf_misc_flags(regs) \
		(user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL)
# define perf_instruction_pointer(regs)	instruction_pointer(regs)
766 767
#endif

768 769 770 771 772
static inline bool has_branch_stack(struct perf_event *event)
{
	return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK;
}

773
extern int perf_output_begin(struct perf_output_handle *handle,
774
			     struct perf_event *event, unsigned int size);
775
extern void perf_output_end(struct perf_output_handle *handle);
776
extern unsigned int perf_output_copy(struct perf_output_handle *handle,
777
			     const void *buf, unsigned int len);
778 779
extern unsigned int perf_output_skip(struct perf_output_handle *handle,
				     unsigned int len);
780 781
extern int perf_swevent_get_recursion_context(void);
extern void perf_swevent_put_recursion_context(int rctx);
782
extern u64 perf_swevent_set_period(struct perf_event *event);
783 784
extern void perf_event_enable(struct perf_event *event);
extern void perf_event_disable(struct perf_event *event);
785
extern int __perf_event_disable(void *info);
786
extern void perf_event_task_tick(void);
787
#else /* !CONFIG_PERF_EVENTS: */
T
Thomas Gleixner 已提交
788
static inline void
789 790 791 792 793
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)			{ }
794 795 796
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)	{ }
797
static inline void perf_event_delayed_put(struct task_struct *task)	{ }
I
Ingo Molnar 已提交
798 799 800
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; }
801 802 803 804
static inline int perf_event_refresh(struct perf_event *event, int refresh)
{
	return -EINVAL;
}
805

806
static inline void
807
perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)	{ }
808
static inline void
809
perf_bp_event(struct perf_event *event, void *data)			{ }
810

811
static inline int perf_register_guest_info_callbacks
812
(struct perf_guest_info_callbacks *callbacks)				{ return 0; }
813
static inline int perf_unregister_guest_info_callbacks
814
(struct perf_guest_info_callbacks *callbacks)				{ return 0; }
815

I
Ingo Molnar 已提交
816
static inline void perf_event_mmap(struct vm_area_struct *vma)		{ }
817
static inline void perf_event_exec(void)				{ }
818
static inline void perf_event_comm(struct task_struct *tsk, bool exec)	{ }
819 820
static inline void perf_event_fork(struct task_struct *tsk)		{ }
static inline void perf_event_init(void)				{ }
821
static inline int  perf_swevent_get_recursion_context(void)		{ return -1; }
822
static inline void perf_swevent_put_recursion_context(int rctx)		{ }
823
static inline u64 perf_swevent_set_period(struct perf_event *event)	{ return 0; }
824 825
static inline void perf_event_enable(struct perf_event *event)		{ }
static inline void perf_event_disable(struct perf_event *event)		{ }
826
static inline int __perf_event_disable(void *info)			{ return -1; }
827
static inline void perf_event_task_tick(void)				{ }
T
Thomas Gleixner 已提交
828 829
#endif

830 831 832 833 834 835
#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

836 837 838
#if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
extern void perf_restore_debug_store(void);
#else
839
static inline void perf_restore_debug_store(void)			{ }
T
Thomas Gleixner 已提交
840 841
#endif

842
#define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x))
843

844
/*
845
 * This has to have a higher priority than migration_notifier in sched/core.c.
846
 */
847 848
#define perf_cpu_notifier(fn)						\
do {									\
849
	static struct notifier_block fn##_nb =				\
850
		{ .notifier_call = fn, .priority = CPU_PRI_PERF };	\
851
	unsigned long cpu = smp_processor_id();				\
852
	unsigned long flags;						\
853 854
									\
	cpu_notifier_register_begin();					\
855
	fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE,			\
856
		(void *)(unsigned long)cpu);				\
857
	local_irq_save(flags);						\
858
	fn(&fn##_nb, (unsigned long)CPU_STARTING,			\
859
		(void *)(unsigned long)cpu);				\
860
	local_irq_restore(flags);					\
861
	fn(&fn##_nb, (unsigned long)CPU_ONLINE,				\
862
		(void *)(unsigned long)cpu);				\
863 864
	__register_cpu_notifier(&fn##_nb);				\
	cpu_notifier_register_done();					\
865 866
} while (0)

867 868 869 870 871 872 873 874 875 876 877
/*
 * 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)
878

879 880 881
struct perf_pmu_events_attr {
	struct device_attribute attr;
	u64 id;
882
	const char *event_str;
883 884 885 886 887 888 889 890
};

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

891 892 893 894 895 896 897 898 899 900 901 902
#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)

903
#endif /* _LINUX_PERF_EVENT_H */