xref: /linux-6.15/include/linux/perf_event.h (revision bcefe12e)
1 /*
2  * Performance events:
3  *
4  *    Copyright (C) 2008-2009, Thomas Gleixner <[email protected]>
5  *    Copyright (C) 2008-2009, Red Hat, Inc., Ingo Molnar
6  *    Copyright (C) 2008-2009, Red Hat, Inc., Peter Zijlstra
7  *
8  * Data type definitions, declarations, prototypes.
9  *
10  *    Started by: Thomas Gleixner and Ingo Molnar
11  *
12  * For licencing details see kernel-base/COPYING
13  */
14 #ifndef _LINUX_PERF_EVENT_H
15 #define _LINUX_PERF_EVENT_H
16 
17 #include <linux/types.h>
18 #include <linux/ioctl.h>
19 #include <asm/byteorder.h>
20 
21 #ifdef CONFIG_HAVE_HW_BREAKPOINT
22 #include <asm/hw_breakpoint.h>
23 #endif
24 
25 /*
26  * User-space ABI bits:
27  */
28 
29 /*
30  * attr.type
31  */
32 enum perf_type_id {
33 	PERF_TYPE_HARDWARE			= 0,
34 	PERF_TYPE_SOFTWARE			= 1,
35 	PERF_TYPE_TRACEPOINT			= 2,
36 	PERF_TYPE_HW_CACHE			= 3,
37 	PERF_TYPE_RAW				= 4,
38 	PERF_TYPE_BREAKPOINT			= 5,
39 
40 	PERF_TYPE_MAX,				/* non-ABI */
41 };
42 
43 /*
44  * Generalized performance event event_id types, used by the
45  * attr.event_id parameter of the sys_perf_event_open()
46  * syscall:
47  */
48 enum perf_hw_id {
49 	/*
50 	 * Common hardware events, generalized by the kernel:
51 	 */
52 	PERF_COUNT_HW_CPU_CYCLES		= 0,
53 	PERF_COUNT_HW_INSTRUCTIONS		= 1,
54 	PERF_COUNT_HW_CACHE_REFERENCES		= 2,
55 	PERF_COUNT_HW_CACHE_MISSES		= 3,
56 	PERF_COUNT_HW_BRANCH_INSTRUCTIONS	= 4,
57 	PERF_COUNT_HW_BRANCH_MISSES		= 5,
58 	PERF_COUNT_HW_BUS_CYCLES		= 6,
59 
60 	PERF_COUNT_HW_MAX,			/* non-ABI */
61 };
62 
63 /*
64  * Generalized hardware cache events:
65  *
66  *       { L1-D, L1-I, LLC, ITLB, DTLB, BPU } x
67  *       { read, write, prefetch } x
68  *       { accesses, misses }
69  */
70 enum perf_hw_cache_id {
71 	PERF_COUNT_HW_CACHE_L1D			= 0,
72 	PERF_COUNT_HW_CACHE_L1I			= 1,
73 	PERF_COUNT_HW_CACHE_LL			= 2,
74 	PERF_COUNT_HW_CACHE_DTLB		= 3,
75 	PERF_COUNT_HW_CACHE_ITLB		= 4,
76 	PERF_COUNT_HW_CACHE_BPU			= 5,
77 
78 	PERF_COUNT_HW_CACHE_MAX,		/* non-ABI */
79 };
80 
81 enum perf_hw_cache_op_id {
82 	PERF_COUNT_HW_CACHE_OP_READ		= 0,
83 	PERF_COUNT_HW_CACHE_OP_WRITE		= 1,
84 	PERF_COUNT_HW_CACHE_OP_PREFETCH		= 2,
85 
86 	PERF_COUNT_HW_CACHE_OP_MAX,		/* non-ABI */
87 };
88 
89 enum perf_hw_cache_op_result_id {
90 	PERF_COUNT_HW_CACHE_RESULT_ACCESS	= 0,
91 	PERF_COUNT_HW_CACHE_RESULT_MISS		= 1,
92 
93 	PERF_COUNT_HW_CACHE_RESULT_MAX,		/* non-ABI */
94 };
95 
96 /*
97  * Special "software" events provided by the kernel, even if the hardware
98  * does not support performance events. These events measure various
99  * physical and sw events of the kernel (and allow the profiling of them as
100  * well):
101  */
102 enum perf_sw_ids {
103 	PERF_COUNT_SW_CPU_CLOCK			= 0,
104 	PERF_COUNT_SW_TASK_CLOCK		= 1,
105 	PERF_COUNT_SW_PAGE_FAULTS		= 2,
106 	PERF_COUNT_SW_CONTEXT_SWITCHES		= 3,
107 	PERF_COUNT_SW_CPU_MIGRATIONS		= 4,
108 	PERF_COUNT_SW_PAGE_FAULTS_MIN		= 5,
109 	PERF_COUNT_SW_PAGE_FAULTS_MAJ		= 6,
110 	PERF_COUNT_SW_ALIGNMENT_FAULTS		= 7,
111 	PERF_COUNT_SW_EMULATION_FAULTS		= 8,
112 
113 	PERF_COUNT_SW_MAX,			/* non-ABI */
114 };
115 
116 /*
117  * Bits that can be set in attr.sample_type to request information
118  * in the overflow packets.
119  */
120 enum perf_event_sample_format {
121 	PERF_SAMPLE_IP				= 1U << 0,
122 	PERF_SAMPLE_TID				= 1U << 1,
123 	PERF_SAMPLE_TIME			= 1U << 2,
124 	PERF_SAMPLE_ADDR			= 1U << 3,
125 	PERF_SAMPLE_READ			= 1U << 4,
126 	PERF_SAMPLE_CALLCHAIN			= 1U << 5,
127 	PERF_SAMPLE_ID				= 1U << 6,
128 	PERF_SAMPLE_CPU				= 1U << 7,
129 	PERF_SAMPLE_PERIOD			= 1U << 8,
130 	PERF_SAMPLE_STREAM_ID			= 1U << 9,
131 	PERF_SAMPLE_RAW				= 1U << 10,
132 
133 	PERF_SAMPLE_MAX = 1U << 11,		/* non-ABI */
134 };
135 
136 /*
137  * The format of the data returned by read() on a perf event fd,
138  * as specified by attr.read_format:
139  *
140  * struct read_format {
141  *	{ u64		value;
142  *	  { u64		time_enabled; } && PERF_FORMAT_ENABLED
143  *	  { u64		time_running; } && PERF_FORMAT_RUNNING
144  *	  { u64		id;           } && PERF_FORMAT_ID
145  *	} && !PERF_FORMAT_GROUP
146  *
147  *	{ u64		nr;
148  *	  { u64		time_enabled; } && PERF_FORMAT_ENABLED
149  *	  { u64		time_running; } && PERF_FORMAT_RUNNING
150  *	  { u64		value;
151  *	    { u64	id;           } && PERF_FORMAT_ID
152  *	  }		cntr[nr];
153  *	} && PERF_FORMAT_GROUP
154  * };
155  */
156 enum perf_event_read_format {
157 	PERF_FORMAT_TOTAL_TIME_ENABLED		= 1U << 0,
158 	PERF_FORMAT_TOTAL_TIME_RUNNING		= 1U << 1,
159 	PERF_FORMAT_ID				= 1U << 2,
160 	PERF_FORMAT_GROUP			= 1U << 3,
161 
162 	PERF_FORMAT_MAX = 1U << 4,		/* non-ABI */
163 };
164 
165 #define PERF_ATTR_SIZE_VER0	64	/* sizeof first published struct */
166 
167 /*
168  * Hardware event_id to monitor via a performance monitoring event:
169  */
170 struct perf_event_attr {
171 
172 	/*
173 	 * Major type: hardware/software/tracepoint/etc.
174 	 */
175 	__u32			type;
176 
177 	/*
178 	 * Size of the attr structure, for fwd/bwd compat.
179 	 */
180 	__u32			size;
181 
182 	/*
183 	 * Type specific configuration information.
184 	 */
185 	__u64			config;
186 
187 	union {
188 		__u64		sample_period;
189 		__u64		sample_freq;
190 	};
191 
192 	__u64			sample_type;
193 	__u64			read_format;
194 
195 	__u64			disabled       :  1, /* off by default        */
196 				inherit	       :  1, /* children inherit it   */
197 				pinned	       :  1, /* must always be on PMU */
198 				exclusive      :  1, /* only group on PMU     */
199 				exclude_user   :  1, /* don't count user      */
200 				exclude_kernel :  1, /* ditto kernel          */
201 				exclude_hv     :  1, /* ditto hypervisor      */
202 				exclude_idle   :  1, /* don't count when idle */
203 				mmap           :  1, /* include mmap data     */
204 				comm	       :  1, /* include comm data     */
205 				freq           :  1, /* use freq, not period  */
206 				inherit_stat   :  1, /* per task counts       */
207 				enable_on_exec :  1, /* next exec enables     */
208 				task           :  1, /* trace fork/exit       */
209 				watermark      :  1, /* wakeup_watermark      */
210 
211 				__reserved_1   : 49;
212 
213 	union {
214 		__u32		wakeup_events;	  /* wakeup every n events */
215 		__u32		wakeup_watermark; /* bytes before wakeup   */
216 	};
217 
218 	union {
219 		struct { /* Hardware breakpoint info */
220 			__u64		bp_addr;
221 			__u32		bp_type;
222 			__u32		bp_len;
223 		};
224 	};
225 
226 	__u32			__reserved_2;
227 
228 	__u64			__reserved_3;
229 };
230 
231 /*
232  * Ioctls that can be done on a perf event fd:
233  */
234 #define PERF_EVENT_IOC_ENABLE		_IO ('$', 0)
235 #define PERF_EVENT_IOC_DISABLE		_IO ('$', 1)
236 #define PERF_EVENT_IOC_REFRESH		_IO ('$', 2)
237 #define PERF_EVENT_IOC_RESET		_IO ('$', 3)
238 #define PERF_EVENT_IOC_PERIOD		_IOW('$', 4, __u64)
239 #define PERF_EVENT_IOC_SET_OUTPUT	_IO ('$', 5)
240 #define PERF_EVENT_IOC_SET_FILTER	_IOW('$', 6, char *)
241 
242 enum perf_event_ioc_flags {
243 	PERF_IOC_FLAG_GROUP		= 1U << 0,
244 };
245 
246 /*
247  * Structure of the page that can be mapped via mmap
248  */
249 struct perf_event_mmap_page {
250 	__u32	version;		/* version number of this structure */
251 	__u32	compat_version;		/* lowest version this is compat with */
252 
253 	/*
254 	 * Bits needed to read the hw events in user-space.
255 	 *
256 	 *   u32 seq;
257 	 *   s64 count;
258 	 *
259 	 *   do {
260 	 *     seq = pc->lock;
261 	 *
262 	 *     barrier()
263 	 *     if (pc->index) {
264 	 *       count = pmc_read(pc->index - 1);
265 	 *       count += pc->offset;
266 	 *     } else
267 	 *       goto regular_read;
268 	 *
269 	 *     barrier();
270 	 *   } while (pc->lock != seq);
271 	 *
272 	 * NOTE: for obvious reason this only works on self-monitoring
273 	 *       processes.
274 	 */
275 	__u32	lock;			/* seqlock for synchronization */
276 	__u32	index;			/* hardware event identifier */
277 	__s64	offset;			/* add to hardware event value */
278 	__u64	time_enabled;		/* time event active */
279 	__u64	time_running;		/* time event on cpu */
280 
281 		/*
282 		 * Hole for extension of the self monitor capabilities
283 		 */
284 
285 	__u64	__reserved[123];	/* align to 1k */
286 
287 	/*
288 	 * Control data for the mmap() data buffer.
289 	 *
290 	 * User-space reading the @data_head value should issue an rmb(), on
291 	 * SMP capable platforms, after reading this value -- see
292 	 * perf_event_wakeup().
293 	 *
294 	 * When the mapping is PROT_WRITE the @data_tail value should be
295 	 * written by userspace to reflect the last read data. In this case
296 	 * the kernel will not over-write unread data.
297 	 */
298 	__u64   data_head;		/* head in the data section */
299 	__u64	data_tail;		/* user-space written tail */
300 };
301 
302 #define PERF_RECORD_MISC_CPUMODE_MASK		(3 << 0)
303 #define PERF_RECORD_MISC_CPUMODE_UNKNOWN		(0 << 0)
304 #define PERF_RECORD_MISC_KERNEL			(1 << 0)
305 #define PERF_RECORD_MISC_USER			(2 << 0)
306 #define PERF_RECORD_MISC_HYPERVISOR		(3 << 0)
307 
308 struct perf_event_header {
309 	__u32	type;
310 	__u16	misc;
311 	__u16	size;
312 };
313 
314 enum perf_event_type {
315 
316 	/*
317 	 * The MMAP events record the PROT_EXEC mappings so that we can
318 	 * correlate userspace IPs to code. They have the following structure:
319 	 *
320 	 * struct {
321 	 *	struct perf_event_header	header;
322 	 *
323 	 *	u32				pid, tid;
324 	 *	u64				addr;
325 	 *	u64				len;
326 	 *	u64				pgoff;
327 	 *	char				filename[];
328 	 * };
329 	 */
330 	PERF_RECORD_MMAP			= 1,
331 
332 	/*
333 	 * struct {
334 	 *	struct perf_event_header	header;
335 	 *	u64				id;
336 	 *	u64				lost;
337 	 * };
338 	 */
339 	PERF_RECORD_LOST			= 2,
340 
341 	/*
342 	 * struct {
343 	 *	struct perf_event_header	header;
344 	 *
345 	 *	u32				pid, tid;
346 	 *	char				comm[];
347 	 * };
348 	 */
349 	PERF_RECORD_COMM			= 3,
350 
351 	/*
352 	 * struct {
353 	 *	struct perf_event_header	header;
354 	 *	u32				pid, ppid;
355 	 *	u32				tid, ptid;
356 	 *	u64				time;
357 	 * };
358 	 */
359 	PERF_RECORD_EXIT			= 4,
360 
361 	/*
362 	 * struct {
363 	 *	struct perf_event_header	header;
364 	 *	u64				time;
365 	 *	u64				id;
366 	 *	u64				stream_id;
367 	 * };
368 	 */
369 	PERF_RECORD_THROTTLE		= 5,
370 	PERF_RECORD_UNTHROTTLE		= 6,
371 
372 	/*
373 	 * struct {
374 	 *	struct perf_event_header	header;
375 	 *	u32				pid, ppid;
376 	 *	u32				tid, ptid;
377 	 *	u64				time;
378 	 * };
379 	 */
380 	PERF_RECORD_FORK			= 7,
381 
382 	/*
383 	 * struct {
384 	 * 	struct perf_event_header	header;
385 	 * 	u32				pid, tid;
386 	 *
387 	 * 	struct read_format		values;
388 	 * };
389 	 */
390 	PERF_RECORD_READ			= 8,
391 
392 	/*
393 	 * struct {
394 	 *	struct perf_event_header	header;
395 	 *
396 	 *	{ u64			ip;	  } && PERF_SAMPLE_IP
397 	 *	{ u32			pid, tid; } && PERF_SAMPLE_TID
398 	 *	{ u64			time;     } && PERF_SAMPLE_TIME
399 	 *	{ u64			addr;     } && PERF_SAMPLE_ADDR
400 	 *	{ u64			id;	  } && PERF_SAMPLE_ID
401 	 *	{ u64			stream_id;} && PERF_SAMPLE_STREAM_ID
402 	 *	{ u32			cpu, res; } && PERF_SAMPLE_CPU
403 	 *	{ u64			period;   } && PERF_SAMPLE_PERIOD
404 	 *
405 	 *	{ struct read_format	values;	  } && PERF_SAMPLE_READ
406 	 *
407 	 *	{ u64			nr,
408 	 *	  u64			ips[nr];  } && PERF_SAMPLE_CALLCHAIN
409 	 *
410 	 *	#
411 	 *	# The RAW record below is opaque data wrt the ABI
412 	 *	#
413 	 *	# That is, the ABI doesn't make any promises wrt to
414 	 *	# the stability of its content, it may vary depending
415 	 *	# on event, hardware, kernel version and phase of
416 	 *	# the moon.
417 	 *	#
418 	 *	# In other words, PERF_SAMPLE_RAW contents are not an ABI.
419 	 *	#
420 	 *
421 	 *	{ u32			size;
422 	 *	  char                  data[size];}&& PERF_SAMPLE_RAW
423 	 * };
424 	 */
425 	PERF_RECORD_SAMPLE		= 9,
426 
427 	PERF_RECORD_MAX,			/* non-ABI */
428 };
429 
430 enum perf_callchain_context {
431 	PERF_CONTEXT_HV			= (__u64)-32,
432 	PERF_CONTEXT_KERNEL		= (__u64)-128,
433 	PERF_CONTEXT_USER		= (__u64)-512,
434 
435 	PERF_CONTEXT_GUEST		= (__u64)-2048,
436 	PERF_CONTEXT_GUEST_KERNEL	= (__u64)-2176,
437 	PERF_CONTEXT_GUEST_USER		= (__u64)-2560,
438 
439 	PERF_CONTEXT_MAX		= (__u64)-4095,
440 };
441 
442 #define PERF_FLAG_FD_NO_GROUP	(1U << 0)
443 #define PERF_FLAG_FD_OUTPUT	(1U << 1)
444 
445 #ifdef __KERNEL__
446 /*
447  * Kernel-internal data types and definitions:
448  */
449 
450 #ifdef CONFIG_PERF_EVENTS
451 # include <asm/perf_event.h>
452 #endif
453 
454 #include <linux/list.h>
455 #include <linux/mutex.h>
456 #include <linux/rculist.h>
457 #include <linux/rcupdate.h>
458 #include <linux/spinlock.h>
459 #include <linux/hrtimer.h>
460 #include <linux/fs.h>
461 #include <linux/pid_namespace.h>
462 #include <linux/workqueue.h>
463 #include <asm/atomic.h>
464 
465 #define PERF_MAX_STACK_DEPTH		255
466 
467 struct perf_callchain_entry {
468 	__u64				nr;
469 	__u64				ip[PERF_MAX_STACK_DEPTH];
470 };
471 
472 struct perf_raw_record {
473 	u32				size;
474 	void				*data;
475 };
476 
477 struct task_struct;
478 
479 /**
480  * struct hw_perf_event - performance event hardware details:
481  */
482 struct hw_perf_event {
483 #ifdef CONFIG_PERF_EVENTS
484 	union {
485 		struct { /* hardware */
486 			u64		config;
487 			unsigned long	config_base;
488 			unsigned long	event_base;
489 			int		idx;
490 		};
491 		struct { /* software */
492 			s64		remaining;
493 			struct hrtimer	hrtimer;
494 		};
495 #ifdef CONFIG_HAVE_HW_BREAKPOINT
496 		union { /* breakpoint */
497 			struct arch_hw_breakpoint	info;
498 		};
499 #endif
500 	};
501 	atomic64_t			prev_count;
502 	u64				sample_period;
503 	u64				last_period;
504 	atomic64_t			period_left;
505 	u64				interrupts;
506 
507 	u64				freq_count;
508 	u64				freq_interrupts;
509 	u64				freq_stamp;
510 #endif
511 };
512 
513 struct perf_event;
514 
515 /**
516  * struct pmu - generic performance monitoring unit
517  */
518 struct pmu {
519 	int (*enable)			(struct perf_event *event);
520 	void (*disable)			(struct perf_event *event);
521 	void (*read)			(struct perf_event *event);
522 	void (*unthrottle)		(struct perf_event *event);
523 };
524 
525 /**
526  * enum perf_event_active_state - the states of a event
527  */
528 enum perf_event_active_state {
529 	PERF_EVENT_STATE_ERROR		= -2,
530 	PERF_EVENT_STATE_OFF		= -1,
531 	PERF_EVENT_STATE_INACTIVE	=  0,
532 	PERF_EVENT_STATE_ACTIVE		=  1,
533 };
534 
535 struct file;
536 
537 struct perf_mmap_data {
538 	struct rcu_head			rcu_head;
539 #ifdef CONFIG_PERF_USE_VMALLOC
540 	struct work_struct		work;
541 #endif
542 	int				data_order;
543 	int				nr_pages;	/* nr of data pages  */
544 	int				writable;	/* are we writable   */
545 	int				nr_locked;	/* nr pages mlocked  */
546 
547 	atomic_t			poll;		/* POLL_ for wakeups */
548 	atomic_t			events;		/* event_id limit       */
549 
550 	atomic_long_t			head;		/* write position    */
551 	atomic_long_t			done_head;	/* completed head    */
552 
553 	atomic_t			lock;		/* concurrent writes */
554 	atomic_t			wakeup;		/* needs a wakeup    */
555 	atomic_t			lost;		/* nr records lost   */
556 
557 	long				watermark;	/* wakeup watermark  */
558 
559 	struct perf_event_mmap_page	*user_page;
560 	void				*data_pages[0];
561 };
562 
563 struct perf_pending_entry {
564 	struct perf_pending_entry *next;
565 	void (*func)(struct perf_pending_entry *);
566 };
567 
568 typedef void (*perf_callback_t)(struct perf_event *, void *);
569 
570 struct perf_sample_data;
571 
572 /**
573  * struct perf_event - performance event kernel representation:
574  */
575 struct perf_event {
576 #ifdef CONFIG_PERF_EVENTS
577 	struct list_head		group_entry;
578 	struct list_head		event_entry;
579 	struct list_head		sibling_list;
580 	int				nr_siblings;
581 	struct perf_event		*group_leader;
582 	struct perf_event		*output;
583 	const struct pmu		*pmu;
584 
585 	enum perf_event_active_state	state;
586 	atomic64_t			count;
587 
588 	/*
589 	 * These are the total time in nanoseconds that the event
590 	 * has been enabled (i.e. eligible to run, and the task has
591 	 * been scheduled in, if this is a per-task event)
592 	 * and running (scheduled onto the CPU), respectively.
593 	 *
594 	 * They are computed from tstamp_enabled, tstamp_running and
595 	 * tstamp_stopped when the event is in INACTIVE or ACTIVE state.
596 	 */
597 	u64				total_time_enabled;
598 	u64				total_time_running;
599 
600 	/*
601 	 * These are timestamps used for computing total_time_enabled
602 	 * and total_time_running when the event is in INACTIVE or
603 	 * ACTIVE state, measured in nanoseconds from an arbitrary point
604 	 * in time.
605 	 * tstamp_enabled: the notional time when the event was enabled
606 	 * tstamp_running: the notional time when the event was scheduled on
607 	 * tstamp_stopped: in INACTIVE state, the notional time when the
608 	 *	event was scheduled off.
609 	 */
610 	u64				tstamp_enabled;
611 	u64				tstamp_running;
612 	u64				tstamp_stopped;
613 
614 	struct perf_event_attr		attr;
615 	struct hw_perf_event		hw;
616 
617 	struct perf_event_context	*ctx;
618 	struct file			*filp;
619 
620 	/*
621 	 * These accumulate total time (in nanoseconds) that children
622 	 * events have been enabled and running, respectively.
623 	 */
624 	atomic64_t			child_total_time_enabled;
625 	atomic64_t			child_total_time_running;
626 
627 	/*
628 	 * Protect attach/detach and child_list:
629 	 */
630 	struct mutex			child_mutex;
631 	struct list_head		child_list;
632 	struct perf_event		*parent;
633 
634 	int				oncpu;
635 	int				cpu;
636 
637 	struct list_head		owner_entry;
638 	struct task_struct		*owner;
639 
640 	/* mmap bits */
641 	struct mutex			mmap_mutex;
642 	atomic_t			mmap_count;
643 	struct perf_mmap_data		*data;
644 
645 	/* poll related */
646 	wait_queue_head_t		waitq;
647 	struct fasync_struct		*fasync;
648 
649 	/* delayed work for NMIs and such */
650 	int				pending_wakeup;
651 	int				pending_kill;
652 	int				pending_disable;
653 	struct perf_pending_entry	pending;
654 
655 	atomic_t			event_limit;
656 
657 	void (*destroy)(struct perf_event *);
658 	struct rcu_head			rcu_head;
659 
660 	struct pid_namespace		*ns;
661 	u64				id;
662 
663 	void (*overflow_handler)(struct perf_event *event,
664 			int nmi, struct perf_sample_data *data,
665 			struct pt_regs *regs);
666 
667 #ifdef CONFIG_EVENT_PROFILE
668 	struct event_filter		*filter;
669 #endif
670 
671 	perf_callback_t			callback;
672 
673 	perf_callback_t			event_callback;
674 
675 #endif /* CONFIG_PERF_EVENTS */
676 };
677 
678 /**
679  * struct perf_event_context - event context structure
680  *
681  * Used as a container for task events and CPU events as well:
682  */
683 struct perf_event_context {
684 	/*
685 	 * Protect the states of the events in the list,
686 	 * nr_active, and the list:
687 	 */
688 	spinlock_t			lock;
689 	/*
690 	 * Protect the list of events.  Locking either mutex or lock
691 	 * is sufficient to ensure the list doesn't change; to change
692 	 * the list you need to lock both the mutex and the spinlock.
693 	 */
694 	struct mutex			mutex;
695 
696 	struct list_head		group_list;
697 	struct list_head		event_list;
698 	int				nr_events;
699 	int				nr_active;
700 	int				is_active;
701 	int				nr_stat;
702 	atomic_t			refcount;
703 	struct task_struct		*task;
704 
705 	/*
706 	 * Context clock, runs when context enabled.
707 	 */
708 	u64				time;
709 	u64				timestamp;
710 
711 	/*
712 	 * These fields let us detect when two contexts have both
713 	 * been cloned (inherited) from a common ancestor.
714 	 */
715 	struct perf_event_context	*parent_ctx;
716 	u64				parent_gen;
717 	u64				generation;
718 	int				pin_count;
719 	struct rcu_head			rcu_head;
720 };
721 
722 /**
723  * struct perf_event_cpu_context - per cpu event context structure
724  */
725 struct perf_cpu_context {
726 	struct perf_event_context	ctx;
727 	struct perf_event_context	*task_ctx;
728 	int				active_oncpu;
729 	int				max_pertask;
730 	int				exclusive;
731 
732 	/*
733 	 * Recursion avoidance:
734 	 *
735 	 * task, softirq, irq, nmi context
736 	 */
737 	int				recursion[4];
738 };
739 
740 struct perf_output_handle {
741 	struct perf_event		*event;
742 	struct perf_mmap_data		*data;
743 	unsigned long			head;
744 	unsigned long			offset;
745 	int				nmi;
746 	int				sample;
747 	int				locked;
748 };
749 
750 #ifdef CONFIG_PERF_EVENTS
751 
752 /*
753  * Set by architecture code:
754  */
755 extern int perf_max_events;
756 
757 extern const struct pmu *hw_perf_event_init(struct perf_event *event);
758 
759 extern void perf_event_task_sched_in(struct task_struct *task, int cpu);
760 extern void perf_event_task_sched_out(struct task_struct *task,
761 					struct task_struct *next, int cpu);
762 extern void perf_event_task_tick(struct task_struct *task, int cpu);
763 extern int perf_event_init_task(struct task_struct *child);
764 extern void perf_event_exit_task(struct task_struct *child);
765 extern void perf_event_free_task(struct task_struct *task);
766 extern void set_perf_event_pending(void);
767 extern void perf_event_do_pending(void);
768 extern void perf_event_print_debug(void);
769 extern void __perf_disable(void);
770 extern bool __perf_enable(void);
771 extern void perf_disable(void);
772 extern void perf_enable(void);
773 extern int perf_event_task_disable(void);
774 extern int perf_event_task_enable(void);
775 extern int hw_perf_group_sched_in(struct perf_event *group_leader,
776 	       struct perf_cpu_context *cpuctx,
777 	       struct perf_event_context *ctx, int cpu);
778 extern void perf_event_update_userpage(struct perf_event *event);
779 extern int perf_event_release_kernel(struct perf_event *event);
780 extern struct perf_event *
781 perf_event_create_kernel_counter(struct perf_event_attr *attr,
782 				int cpu,
783 				pid_t pid,
784 				perf_callback_t callback);
785 extern u64 perf_event_read_value(struct perf_event *event,
786 				 u64 *enabled, u64 *running);
787 
788 struct perf_sample_data {
789 	u64				type;
790 
791 	u64				ip;
792 	struct {
793 		u32	pid;
794 		u32	tid;
795 	}				tid_entry;
796 	u64				time;
797 	u64				addr;
798 	u64				id;
799 	u64				stream_id;
800 	struct {
801 		u32	cpu;
802 		u32	reserved;
803 	}				cpu_entry;
804 	u64				period;
805 	struct perf_callchain_entry	*callchain;
806 	struct perf_raw_record		*raw;
807 };
808 
809 extern void perf_output_sample(struct perf_output_handle *handle,
810 			       struct perf_event_header *header,
811 			       struct perf_sample_data *data,
812 			       struct perf_event *event);
813 extern void perf_prepare_sample(struct perf_event_header *header,
814 				struct perf_sample_data *data,
815 				struct perf_event *event,
816 				struct pt_regs *regs);
817 
818 extern int perf_event_overflow(struct perf_event *event, int nmi,
819 				 struct perf_sample_data *data,
820 				 struct pt_regs *regs);
821 
822 /*
823  * Return 1 for a software event, 0 for a hardware event
824  */
825 static inline int is_software_event(struct perf_event *event)
826 {
827 	return (event->attr.type != PERF_TYPE_RAW) &&
828 		(event->attr.type != PERF_TYPE_HARDWARE) &&
829 		(event->attr.type != PERF_TYPE_HW_CACHE);
830 }
831 
832 extern atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX];
833 
834 extern void __perf_sw_event(u32, u64, int, struct pt_regs *, u64);
835 
836 static inline void
837 perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr)
838 {
839 	if (atomic_read(&perf_swevent_enabled[event_id]))
840 		__perf_sw_event(event_id, nr, nmi, regs, addr);
841 }
842 
843 extern void __perf_event_mmap(struct vm_area_struct *vma);
844 
845 static inline void perf_event_mmap(struct vm_area_struct *vma)
846 {
847 	if (vma->vm_flags & VM_EXEC)
848 		__perf_event_mmap(vma);
849 }
850 
851 extern void perf_event_comm(struct task_struct *tsk);
852 extern void perf_event_fork(struct task_struct *tsk);
853 
854 extern struct perf_callchain_entry *perf_callchain(struct pt_regs *regs);
855 
856 extern int sysctl_perf_event_paranoid;
857 extern int sysctl_perf_event_mlock;
858 extern int sysctl_perf_event_sample_rate;
859 
860 extern void perf_event_init(void);
861 extern void perf_tp_event(int event_id, u64 addr, u64 count,
862 				 void *record, int entry_size);
863 extern void perf_bp_event(struct perf_event *event, void *data);
864 
865 #ifndef perf_misc_flags
866 #define perf_misc_flags(regs)	(user_mode(regs) ? PERF_RECORD_MISC_USER : \
867 				 PERF_RECORD_MISC_KERNEL)
868 #define perf_instruction_pointer(regs)	instruction_pointer(regs)
869 #endif
870 
871 extern int perf_output_begin(struct perf_output_handle *handle,
872 			     struct perf_event *event, unsigned int size,
873 			     int nmi, int sample);
874 extern void perf_output_end(struct perf_output_handle *handle);
875 extern void perf_output_copy(struct perf_output_handle *handle,
876 			     const void *buf, unsigned int len);
877 extern int perf_swevent_get_recursion_context(void);
878 extern void perf_swevent_put_recursion_context(int rctx);
879 #else
880 static inline void
881 perf_event_task_sched_in(struct task_struct *task, int cpu)		{ }
882 static inline void
883 perf_event_task_sched_out(struct task_struct *task,
884 			    struct task_struct *next, int cpu)		{ }
885 static inline void
886 perf_event_task_tick(struct task_struct *task, int cpu)			{ }
887 static inline int perf_event_init_task(struct task_struct *child)	{ return 0; }
888 static inline void perf_event_exit_task(struct task_struct *child)	{ }
889 static inline void perf_event_free_task(struct task_struct *task)	{ }
890 static inline void perf_event_do_pending(void)				{ }
891 static inline void perf_event_print_debug(void)				{ }
892 static inline void perf_disable(void)					{ }
893 static inline void perf_enable(void)					{ }
894 static inline int perf_event_task_disable(void)				{ return -EINVAL; }
895 static inline int perf_event_task_enable(void)				{ return -EINVAL; }
896 
897 static inline void
898 perf_sw_event(u32 event_id, u64 nr, int nmi,
899 		     struct pt_regs *regs, u64 addr)			{ }
900 static inline void
901 perf_bp_event(struct perf_event *event, void *data)		{ }
902 
903 static inline void perf_event_mmap(struct vm_area_struct *vma)		{ }
904 static inline void perf_event_comm(struct task_struct *tsk)		{ }
905 static inline void perf_event_fork(struct task_struct *tsk)		{ }
906 static inline void perf_event_init(void)				{ }
907 static inline int  perf_swevent_get_recursion_context(void)  { return -1; }
908 static inline void perf_swevent_put_recursion_context(int rctx)		{ }
909 
910 #endif
911 
912 #define perf_output_put(handle, x) \
913 	perf_output_copy((handle), &(x), sizeof(x))
914 
915 #endif /* __KERNEL__ */
916 #endif /* _LINUX_PERF_EVENT_H */
917