xref: /linux-6.15/include/linux/perf_event.h (revision 586bc5cc)
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 /*
22  * User-space ABI bits:
23  */
24 
25 /*
26  * attr.type
27  */
28 enum perf_type_id {
29 	PERF_TYPE_HARDWARE			= 0,
30 	PERF_TYPE_SOFTWARE			= 1,
31 	PERF_TYPE_TRACEPOINT			= 2,
32 	PERF_TYPE_HW_CACHE			= 3,
33 	PERF_TYPE_RAW				= 4,
34 	PERF_TYPE_BREAKPOINT			= 5,
35 
36 	PERF_TYPE_MAX,				/* non-ABI */
37 };
38 
39 /*
40  * Generalized performance event event_id types, used by the
41  * attr.event_id parameter of the sys_perf_event_open()
42  * syscall:
43  */
44 enum perf_hw_id {
45 	/*
46 	 * Common hardware events, generalized by the kernel:
47 	 */
48 	PERF_COUNT_HW_CPU_CYCLES		= 0,
49 	PERF_COUNT_HW_INSTRUCTIONS		= 1,
50 	PERF_COUNT_HW_CACHE_REFERENCES		= 2,
51 	PERF_COUNT_HW_CACHE_MISSES		= 3,
52 	PERF_COUNT_HW_BRANCH_INSTRUCTIONS	= 4,
53 	PERF_COUNT_HW_BRANCH_MISSES		= 5,
54 	PERF_COUNT_HW_BUS_CYCLES		= 6,
55 
56 	PERF_COUNT_HW_MAX,			/* non-ABI */
57 };
58 
59 /*
60  * Generalized hardware cache events:
61  *
62  *       { L1-D, L1-I, LLC, ITLB, DTLB, BPU } x
63  *       { read, write, prefetch } x
64  *       { accesses, misses }
65  */
66 enum perf_hw_cache_id {
67 	PERF_COUNT_HW_CACHE_L1D			= 0,
68 	PERF_COUNT_HW_CACHE_L1I			= 1,
69 	PERF_COUNT_HW_CACHE_LL			= 2,
70 	PERF_COUNT_HW_CACHE_DTLB		= 3,
71 	PERF_COUNT_HW_CACHE_ITLB		= 4,
72 	PERF_COUNT_HW_CACHE_BPU			= 5,
73 
74 	PERF_COUNT_HW_CACHE_MAX,		/* non-ABI */
75 };
76 
77 enum perf_hw_cache_op_id {
78 	PERF_COUNT_HW_CACHE_OP_READ		= 0,
79 	PERF_COUNT_HW_CACHE_OP_WRITE		= 1,
80 	PERF_COUNT_HW_CACHE_OP_PREFETCH		= 2,
81 
82 	PERF_COUNT_HW_CACHE_OP_MAX,		/* non-ABI */
83 };
84 
85 enum perf_hw_cache_op_result_id {
86 	PERF_COUNT_HW_CACHE_RESULT_ACCESS	= 0,
87 	PERF_COUNT_HW_CACHE_RESULT_MISS		= 1,
88 
89 	PERF_COUNT_HW_CACHE_RESULT_MAX,		/* non-ABI */
90 };
91 
92 /*
93  * Special "software" events provided by the kernel, even if the hardware
94  * does not support performance events. These events measure various
95  * physical and sw events of the kernel (and allow the profiling of them as
96  * well):
97  */
98 enum perf_sw_ids {
99 	PERF_COUNT_SW_CPU_CLOCK			= 0,
100 	PERF_COUNT_SW_TASK_CLOCK		= 1,
101 	PERF_COUNT_SW_PAGE_FAULTS		= 2,
102 	PERF_COUNT_SW_CONTEXT_SWITCHES		= 3,
103 	PERF_COUNT_SW_CPU_MIGRATIONS		= 4,
104 	PERF_COUNT_SW_PAGE_FAULTS_MIN		= 5,
105 	PERF_COUNT_SW_PAGE_FAULTS_MAJ		= 6,
106 	PERF_COUNT_SW_ALIGNMENT_FAULTS		= 7,
107 	PERF_COUNT_SW_EMULATION_FAULTS		= 8,
108 
109 	PERF_COUNT_SW_MAX,			/* non-ABI */
110 };
111 
112 /*
113  * Bits that can be set in attr.sample_type to request information
114  * in the overflow packets.
115  */
116 enum perf_event_sample_format {
117 	PERF_SAMPLE_IP				= 1U << 0,
118 	PERF_SAMPLE_TID				= 1U << 1,
119 	PERF_SAMPLE_TIME			= 1U << 2,
120 	PERF_SAMPLE_ADDR			= 1U << 3,
121 	PERF_SAMPLE_READ			= 1U << 4,
122 	PERF_SAMPLE_CALLCHAIN			= 1U << 5,
123 	PERF_SAMPLE_ID				= 1U << 6,
124 	PERF_SAMPLE_CPU				= 1U << 7,
125 	PERF_SAMPLE_PERIOD			= 1U << 8,
126 	PERF_SAMPLE_STREAM_ID			= 1U << 9,
127 	PERF_SAMPLE_RAW				= 1U << 10,
128 
129 	PERF_SAMPLE_MAX = 1U << 11,		/* non-ABI */
130 };
131 
132 /*
133  * The format of the data returned by read() on a perf event fd,
134  * as specified by attr.read_format:
135  *
136  * struct read_format {
137  *	{ u64		value;
138  *	  { u64		time_enabled; } && PERF_FORMAT_ENABLED
139  *	  { u64		time_running; } && PERF_FORMAT_RUNNING
140  *	  { u64		id;           } && PERF_FORMAT_ID
141  *	} && !PERF_FORMAT_GROUP
142  *
143  *	{ u64		nr;
144  *	  { u64		time_enabled; } && PERF_FORMAT_ENABLED
145  *	  { u64		time_running; } && PERF_FORMAT_RUNNING
146  *	  { u64		value;
147  *	    { u64	id;           } && PERF_FORMAT_ID
148  *	  }		cntr[nr];
149  *	} && PERF_FORMAT_GROUP
150  * };
151  */
152 enum perf_event_read_format {
153 	PERF_FORMAT_TOTAL_TIME_ENABLED		= 1U << 0,
154 	PERF_FORMAT_TOTAL_TIME_RUNNING		= 1U << 1,
155 	PERF_FORMAT_ID				= 1U << 2,
156 	PERF_FORMAT_GROUP			= 1U << 3,
157 
158 	PERF_FORMAT_MAX = 1U << 4,		/* non-ABI */
159 };
160 
161 #define PERF_ATTR_SIZE_VER0	64	/* sizeof first published struct */
162 
163 /*
164  * Hardware event_id to monitor via a performance monitoring event:
165  */
166 struct perf_event_attr {
167 
168 	/*
169 	 * Major type: hardware/software/tracepoint/etc.
170 	 */
171 	__u32			type;
172 
173 	/*
174 	 * Size of the attr structure, for fwd/bwd compat.
175 	 */
176 	__u32			size;
177 
178 	/*
179 	 * Type specific configuration information.
180 	 */
181 	__u64			config;
182 
183 	union {
184 		__u64		sample_period;
185 		__u64		sample_freq;
186 	};
187 
188 	__u64			sample_type;
189 	__u64			read_format;
190 
191 	__u64			disabled       :  1, /* off by default        */
192 				inherit	       :  1, /* children inherit it   */
193 				pinned	       :  1, /* must always be on PMU */
194 				exclusive      :  1, /* only group on PMU     */
195 				exclude_user   :  1, /* don't count user      */
196 				exclude_kernel :  1, /* ditto kernel          */
197 				exclude_hv     :  1, /* ditto hypervisor      */
198 				exclude_idle   :  1, /* don't count when idle */
199 				mmap           :  1, /* include mmap data     */
200 				comm	       :  1, /* include comm data     */
201 				freq           :  1, /* use freq, not period  */
202 				inherit_stat   :  1, /* per task counts       */
203 				enable_on_exec :  1, /* next exec enables     */
204 				task           :  1, /* trace fork/exit       */
205 				watermark      :  1, /* wakeup_watermark      */
206 
207 				__reserved_1   : 49;
208 
209 	union {
210 		__u32		wakeup_events;	  /* wakeup every n events */
211 		__u32		wakeup_watermark; /* bytes before wakeup   */
212 	};
213 
214 	struct { /* Hardware breakpoint info */
215 		__u64		bp_addr;
216 		__u32		bp_type;
217 		__u32		bp_len;
218 		__u64		__bp_reserved_1;
219 		__u64		__bp_reserved_2;
220 	};
221 
222 	__u32			__reserved_2;
223 
224 	__u64			__reserved_3;
225 };
226 
227 /*
228  * Ioctls that can be done on a perf event fd:
229  */
230 #define PERF_EVENT_IOC_ENABLE		_IO ('$', 0)
231 #define PERF_EVENT_IOC_DISABLE		_IO ('$', 1)
232 #define PERF_EVENT_IOC_REFRESH		_IO ('$', 2)
233 #define PERF_EVENT_IOC_RESET		_IO ('$', 3)
234 #define PERF_EVENT_IOC_PERIOD		_IOW('$', 4, __u64)
235 #define PERF_EVENT_IOC_SET_OUTPUT	_IO ('$', 5)
236 #define PERF_EVENT_IOC_SET_FILTER	_IOW('$', 6, char *)
237 
238 enum perf_event_ioc_flags {
239 	PERF_IOC_FLAG_GROUP		= 1U << 0,
240 };
241 
242 /*
243  * Structure of the page that can be mapped via mmap
244  */
245 struct perf_event_mmap_page {
246 	__u32	version;		/* version number of this structure */
247 	__u32	compat_version;		/* lowest version this is compat with */
248 
249 	/*
250 	 * Bits needed to read the hw events in user-space.
251 	 *
252 	 *   u32 seq;
253 	 *   s64 count;
254 	 *
255 	 *   do {
256 	 *     seq = pc->lock;
257 	 *
258 	 *     barrier()
259 	 *     if (pc->index) {
260 	 *       count = pmc_read(pc->index - 1);
261 	 *       count += pc->offset;
262 	 *     } else
263 	 *       goto regular_read;
264 	 *
265 	 *     barrier();
266 	 *   } while (pc->lock != seq);
267 	 *
268 	 * NOTE: for obvious reason this only works on self-monitoring
269 	 *       processes.
270 	 */
271 	__u32	lock;			/* seqlock for synchronization */
272 	__u32	index;			/* hardware event identifier */
273 	__s64	offset;			/* add to hardware event value */
274 	__u64	time_enabled;		/* time event active */
275 	__u64	time_running;		/* time event on cpu */
276 
277 		/*
278 		 * Hole for extension of the self monitor capabilities
279 		 */
280 
281 	__u64	__reserved[123];	/* align to 1k */
282 
283 	/*
284 	 * Control data for the mmap() data buffer.
285 	 *
286 	 * User-space reading the @data_head value should issue an rmb(), on
287 	 * SMP capable platforms, after reading this value -- see
288 	 * perf_event_wakeup().
289 	 *
290 	 * When the mapping is PROT_WRITE the @data_tail value should be
291 	 * written by userspace to reflect the last read data. In this case
292 	 * the kernel will not over-write unread data.
293 	 */
294 	__u64   data_head;		/* head in the data section */
295 	__u64	data_tail;		/* user-space written tail */
296 };
297 
298 #define PERF_RECORD_MISC_CPUMODE_MASK		(3 << 0)
299 #define PERF_RECORD_MISC_CPUMODE_UNKNOWN		(0 << 0)
300 #define PERF_RECORD_MISC_KERNEL			(1 << 0)
301 #define PERF_RECORD_MISC_USER			(2 << 0)
302 #define PERF_RECORD_MISC_HYPERVISOR		(3 << 0)
303 
304 struct perf_event_header {
305 	__u32	type;
306 	__u16	misc;
307 	__u16	size;
308 };
309 
310 enum perf_event_type {
311 
312 	/*
313 	 * The MMAP events record the PROT_EXEC mappings so that we can
314 	 * correlate userspace IPs to code. They have the following structure:
315 	 *
316 	 * struct {
317 	 *	struct perf_event_header	header;
318 	 *
319 	 *	u32				pid, tid;
320 	 *	u64				addr;
321 	 *	u64				len;
322 	 *	u64				pgoff;
323 	 *	char				filename[];
324 	 * };
325 	 */
326 	PERF_RECORD_MMAP			= 1,
327 
328 	/*
329 	 * struct {
330 	 *	struct perf_event_header	header;
331 	 *	u64				id;
332 	 *	u64				lost;
333 	 * };
334 	 */
335 	PERF_RECORD_LOST			= 2,
336 
337 	/*
338 	 * struct {
339 	 *	struct perf_event_header	header;
340 	 *
341 	 *	u32				pid, tid;
342 	 *	char				comm[];
343 	 * };
344 	 */
345 	PERF_RECORD_COMM			= 3,
346 
347 	/*
348 	 * struct {
349 	 *	struct perf_event_header	header;
350 	 *	u32				pid, ppid;
351 	 *	u32				tid, ptid;
352 	 *	u64				time;
353 	 * };
354 	 */
355 	PERF_RECORD_EXIT			= 4,
356 
357 	/*
358 	 * struct {
359 	 *	struct perf_event_header	header;
360 	 *	u64				time;
361 	 *	u64				id;
362 	 *	u64				stream_id;
363 	 * };
364 	 */
365 	PERF_RECORD_THROTTLE		= 5,
366 	PERF_RECORD_UNTHROTTLE		= 6,
367 
368 	/*
369 	 * struct {
370 	 *	struct perf_event_header	header;
371 	 *	u32				pid, ppid;
372 	 *	u32				tid, ptid;
373 	 *	u64				time;
374 	 * };
375 	 */
376 	PERF_RECORD_FORK			= 7,
377 
378 	/*
379 	 * struct {
380 	 * 	struct perf_event_header	header;
381 	 * 	u32				pid, tid;
382 	 *
383 	 * 	struct read_format		values;
384 	 * };
385 	 */
386 	PERF_RECORD_READ			= 8,
387 
388 	/*
389 	 * struct {
390 	 *	struct perf_event_header	header;
391 	 *
392 	 *	{ u64			ip;	  } && PERF_SAMPLE_IP
393 	 *	{ u32			pid, tid; } && PERF_SAMPLE_TID
394 	 *	{ u64			time;     } && PERF_SAMPLE_TIME
395 	 *	{ u64			addr;     } && PERF_SAMPLE_ADDR
396 	 *	{ u64			id;	  } && PERF_SAMPLE_ID
397 	 *	{ u64			stream_id;} && PERF_SAMPLE_STREAM_ID
398 	 *	{ u32			cpu, res; } && PERF_SAMPLE_CPU
399 	 *	{ u64			period;   } && PERF_SAMPLE_PERIOD
400 	 *
401 	 *	{ struct read_format	values;	  } && PERF_SAMPLE_READ
402 	 *
403 	 *	{ u64			nr,
404 	 *	  u64			ips[nr];  } && PERF_SAMPLE_CALLCHAIN
405 	 *
406 	 *	#
407 	 *	# The RAW record below is opaque data wrt the ABI
408 	 *	#
409 	 *	# That is, the ABI doesn't make any promises wrt to
410 	 *	# the stability of its content, it may vary depending
411 	 *	# on event, hardware, kernel version and phase of
412 	 *	# the moon.
413 	 *	#
414 	 *	# In other words, PERF_SAMPLE_RAW contents are not an ABI.
415 	 *	#
416 	 *
417 	 *	{ u32			size;
418 	 *	  char                  data[size];}&& PERF_SAMPLE_RAW
419 	 * };
420 	 */
421 	PERF_RECORD_SAMPLE		= 9,
422 
423 	PERF_RECORD_MAX,			/* non-ABI */
424 };
425 
426 enum perf_callchain_context {
427 	PERF_CONTEXT_HV			= (__u64)-32,
428 	PERF_CONTEXT_KERNEL		= (__u64)-128,
429 	PERF_CONTEXT_USER		= (__u64)-512,
430 
431 	PERF_CONTEXT_GUEST		= (__u64)-2048,
432 	PERF_CONTEXT_GUEST_KERNEL	= (__u64)-2176,
433 	PERF_CONTEXT_GUEST_USER		= (__u64)-2560,
434 
435 	PERF_CONTEXT_MAX		= (__u64)-4095,
436 };
437 
438 #define PERF_FLAG_FD_NO_GROUP	(1U << 0)
439 #define PERF_FLAG_FD_OUTPUT	(1U << 1)
440 
441 #ifdef __KERNEL__
442 /*
443  * Kernel-internal data types and definitions:
444  */
445 
446 #ifdef CONFIG_PERF_EVENTS
447 # include <asm/perf_event.h>
448 #endif
449 
450 #ifdef CONFIG_HAVE_HW_BREAKPOINT
451 #include <asm/hw_breakpoint.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 struct perf_sample_data;
569 
570 typedef void (*perf_overflow_handler_t)(struct perf_event *, int,
571 					struct perf_sample_data *,
572 					struct pt_regs *regs);
573 
574 /**
575  * struct perf_event - performance event kernel representation:
576  */
577 struct perf_event {
578 #ifdef CONFIG_PERF_EVENTS
579 	struct list_head		group_entry;
580 	struct list_head		event_entry;
581 	struct list_head		sibling_list;
582 	int				nr_siblings;
583 	struct perf_event		*group_leader;
584 	struct perf_event		*output;
585 	const struct pmu		*pmu;
586 
587 	enum perf_event_active_state	state;
588 	atomic64_t			count;
589 
590 	/*
591 	 * These are the total time in nanoseconds that the event
592 	 * has been enabled (i.e. eligible to run, and the task has
593 	 * been scheduled in, if this is a per-task event)
594 	 * and running (scheduled onto the CPU), respectively.
595 	 *
596 	 * They are computed from tstamp_enabled, tstamp_running and
597 	 * tstamp_stopped when the event is in INACTIVE or ACTIVE state.
598 	 */
599 	u64				total_time_enabled;
600 	u64				total_time_running;
601 
602 	/*
603 	 * These are timestamps used for computing total_time_enabled
604 	 * and total_time_running when the event is in INACTIVE or
605 	 * ACTIVE state, measured in nanoseconds from an arbitrary point
606 	 * in time.
607 	 * tstamp_enabled: the notional time when the event was enabled
608 	 * tstamp_running: the notional time when the event was scheduled on
609 	 * tstamp_stopped: in INACTIVE state, the notional time when the
610 	 *	event was scheduled off.
611 	 */
612 	u64				tstamp_enabled;
613 	u64				tstamp_running;
614 	u64				tstamp_stopped;
615 
616 	struct perf_event_attr		attr;
617 	struct hw_perf_event		hw;
618 
619 	struct perf_event_context	*ctx;
620 	struct file			*filp;
621 
622 	/*
623 	 * These accumulate total time (in nanoseconds) that children
624 	 * events have been enabled and running, respectively.
625 	 */
626 	atomic64_t			child_total_time_enabled;
627 	atomic64_t			child_total_time_running;
628 
629 	/*
630 	 * Protect attach/detach and child_list:
631 	 */
632 	struct mutex			child_mutex;
633 	struct list_head		child_list;
634 	struct perf_event		*parent;
635 
636 	int				oncpu;
637 	int				cpu;
638 
639 	struct list_head		owner_entry;
640 	struct task_struct		*owner;
641 
642 	/* mmap bits */
643 	struct mutex			mmap_mutex;
644 	atomic_t			mmap_count;
645 	struct perf_mmap_data		*data;
646 
647 	/* poll related */
648 	wait_queue_head_t		waitq;
649 	struct fasync_struct		*fasync;
650 
651 	/* delayed work for NMIs and such */
652 	int				pending_wakeup;
653 	int				pending_kill;
654 	int				pending_disable;
655 	struct perf_pending_entry	pending;
656 
657 	atomic_t			event_limit;
658 
659 	void (*destroy)(struct perf_event *);
660 	struct rcu_head			rcu_head;
661 
662 	struct pid_namespace		*ns;
663 	u64				id;
664 
665 	perf_overflow_handler_t		overflow_handler;
666 
667 #ifdef CONFIG_EVENT_PROFILE
668 	struct event_filter		*filter;
669 #endif
670 
671 #endif /* CONFIG_PERF_EVENTS */
672 };
673 
674 /**
675  * struct perf_event_context - event context structure
676  *
677  * Used as a container for task events and CPU events as well:
678  */
679 struct perf_event_context {
680 	/*
681 	 * Protect the states of the events in the list,
682 	 * nr_active, and the list:
683 	 */
684 	spinlock_t			lock;
685 	/*
686 	 * Protect the list of events.  Locking either mutex or lock
687 	 * is sufficient to ensure the list doesn't change; to change
688 	 * the list you need to lock both the mutex and the spinlock.
689 	 */
690 	struct mutex			mutex;
691 
692 	struct list_head		group_list;
693 	struct list_head		event_list;
694 	int				nr_events;
695 	int				nr_active;
696 	int				is_active;
697 	int				nr_stat;
698 	atomic_t			refcount;
699 	struct task_struct		*task;
700 
701 	/*
702 	 * Context clock, runs when context enabled.
703 	 */
704 	u64				time;
705 	u64				timestamp;
706 
707 	/*
708 	 * These fields let us detect when two contexts have both
709 	 * been cloned (inherited) from a common ancestor.
710 	 */
711 	struct perf_event_context	*parent_ctx;
712 	u64				parent_gen;
713 	u64				generation;
714 	int				pin_count;
715 	struct rcu_head			rcu_head;
716 };
717 
718 /**
719  * struct perf_event_cpu_context - per cpu event context structure
720  */
721 struct perf_cpu_context {
722 	struct perf_event_context	ctx;
723 	struct perf_event_context	*task_ctx;
724 	int				active_oncpu;
725 	int				max_pertask;
726 	int				exclusive;
727 
728 	/*
729 	 * Recursion avoidance:
730 	 *
731 	 * task, softirq, irq, nmi context
732 	 */
733 	int				recursion[4];
734 };
735 
736 struct perf_output_handle {
737 	struct perf_event		*event;
738 	struct perf_mmap_data		*data;
739 	unsigned long			head;
740 	unsigned long			offset;
741 	int				nmi;
742 	int				sample;
743 	int				locked;
744 };
745 
746 #ifdef CONFIG_PERF_EVENTS
747 
748 /*
749  * Set by architecture code:
750  */
751 extern int perf_max_events;
752 
753 extern const struct pmu *hw_perf_event_init(struct perf_event *event);
754 
755 extern void perf_event_task_sched_in(struct task_struct *task, int cpu);
756 extern void perf_event_task_sched_out(struct task_struct *task,
757 					struct task_struct *next, int cpu);
758 extern void perf_event_task_tick(struct task_struct *task, int cpu);
759 extern int perf_event_init_task(struct task_struct *child);
760 extern void perf_event_exit_task(struct task_struct *child);
761 extern void perf_event_free_task(struct task_struct *task);
762 extern void set_perf_event_pending(void);
763 extern void perf_event_do_pending(void);
764 extern void perf_event_print_debug(void);
765 extern void __perf_disable(void);
766 extern bool __perf_enable(void);
767 extern void perf_disable(void);
768 extern void perf_enable(void);
769 extern int perf_event_task_disable(void);
770 extern int perf_event_task_enable(void);
771 extern int hw_perf_group_sched_in(struct perf_event *group_leader,
772 	       struct perf_cpu_context *cpuctx,
773 	       struct perf_event_context *ctx, int cpu);
774 extern void perf_event_update_userpage(struct perf_event *event);
775 extern int perf_event_release_kernel(struct perf_event *event);
776 extern struct perf_event *
777 perf_event_create_kernel_counter(struct perf_event_attr *attr,
778 				int cpu,
779 				pid_t pid,
780 				perf_overflow_handler_t callback);
781 extern u64 perf_event_read_value(struct perf_event *event,
782 				 u64 *enabled, u64 *running);
783 
784 struct perf_sample_data {
785 	u64				type;
786 
787 	u64				ip;
788 	struct {
789 		u32	pid;
790 		u32	tid;
791 	}				tid_entry;
792 	u64				time;
793 	u64				addr;
794 	u64				id;
795 	u64				stream_id;
796 	struct {
797 		u32	cpu;
798 		u32	reserved;
799 	}				cpu_entry;
800 	u64				period;
801 	struct perf_callchain_entry	*callchain;
802 	struct perf_raw_record		*raw;
803 };
804 
805 extern void perf_output_sample(struct perf_output_handle *handle,
806 			       struct perf_event_header *header,
807 			       struct perf_sample_data *data,
808 			       struct perf_event *event);
809 extern void perf_prepare_sample(struct perf_event_header *header,
810 				struct perf_sample_data *data,
811 				struct perf_event *event,
812 				struct pt_regs *regs);
813 
814 extern int perf_event_overflow(struct perf_event *event, int nmi,
815 				 struct perf_sample_data *data,
816 				 struct pt_regs *regs);
817 
818 /*
819  * Return 1 for a software event, 0 for a hardware event
820  */
821 static inline int is_software_event(struct perf_event *event)
822 {
823 	return (event->attr.type != PERF_TYPE_RAW) &&
824 		(event->attr.type != PERF_TYPE_HARDWARE) &&
825 		(event->attr.type != PERF_TYPE_HW_CACHE);
826 }
827 
828 extern atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX];
829 
830 extern void __perf_sw_event(u32, u64, int, struct pt_regs *, u64);
831 
832 static inline void
833 perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr)
834 {
835 	if (atomic_read(&perf_swevent_enabled[event_id]))
836 		__perf_sw_event(event_id, nr, nmi, regs, addr);
837 }
838 
839 extern void __perf_event_mmap(struct vm_area_struct *vma);
840 
841 static inline void perf_event_mmap(struct vm_area_struct *vma)
842 {
843 	if (vma->vm_flags & VM_EXEC)
844 		__perf_event_mmap(vma);
845 }
846 
847 extern void perf_event_comm(struct task_struct *tsk);
848 extern void perf_event_fork(struct task_struct *tsk);
849 
850 extern struct perf_callchain_entry *perf_callchain(struct pt_regs *regs);
851 
852 extern int sysctl_perf_event_paranoid;
853 extern int sysctl_perf_event_mlock;
854 extern int sysctl_perf_event_sample_rate;
855 
856 extern void perf_event_init(void);
857 extern void perf_tp_event(int event_id, u64 addr, u64 count,
858 				 void *record, int entry_size);
859 extern void perf_bp_event(struct perf_event *event, void *data);
860 
861 #ifndef perf_misc_flags
862 #define perf_misc_flags(regs)	(user_mode(regs) ? PERF_RECORD_MISC_USER : \
863 				 PERF_RECORD_MISC_KERNEL)
864 #define perf_instruction_pointer(regs)	instruction_pointer(regs)
865 #endif
866 
867 extern int perf_output_begin(struct perf_output_handle *handle,
868 			     struct perf_event *event, unsigned int size,
869 			     int nmi, int sample);
870 extern void perf_output_end(struct perf_output_handle *handle);
871 extern void perf_output_copy(struct perf_output_handle *handle,
872 			     const void *buf, unsigned int len);
873 extern int perf_swevent_get_recursion_context(void);
874 extern void perf_swevent_put_recursion_context(int rctx);
875 extern void perf_event_enable(struct perf_event *event);
876 extern void perf_event_disable(struct perf_event *event);
877 #else
878 static inline void
879 perf_event_task_sched_in(struct task_struct *task, int cpu)		{ }
880 static inline void
881 perf_event_task_sched_out(struct task_struct *task,
882 			    struct task_struct *next, int cpu)		{ }
883 static inline void
884 perf_event_task_tick(struct task_struct *task, int cpu)			{ }
885 static inline int perf_event_init_task(struct task_struct *child)	{ return 0; }
886 static inline void perf_event_exit_task(struct task_struct *child)	{ }
887 static inline void perf_event_free_task(struct task_struct *task)	{ }
888 static inline void perf_event_do_pending(void)				{ }
889 static inline void perf_event_print_debug(void)				{ }
890 static inline void perf_disable(void)					{ }
891 static inline void perf_enable(void)					{ }
892 static inline int perf_event_task_disable(void)				{ return -EINVAL; }
893 static inline int perf_event_task_enable(void)				{ return -EINVAL; }
894 
895 static inline void
896 perf_sw_event(u32 event_id, u64 nr, int nmi,
897 		     struct pt_regs *regs, u64 addr)			{ }
898 static inline void
899 perf_bp_event(struct perf_event *event, void *data)		{ }
900 
901 static inline void perf_event_mmap(struct vm_area_struct *vma)		{ }
902 static inline void perf_event_comm(struct task_struct *tsk)		{ }
903 static inline void perf_event_fork(struct task_struct *tsk)		{ }
904 static inline void perf_event_init(void)				{ }
905 static inline int  perf_swevent_get_recursion_context(void)  { return -1; }
906 static inline void perf_swevent_put_recursion_context(int rctx)		{ }
907 static inline void perf_event_enable(struct perf_event *event)		{ }
908 static inline void perf_event_disable(struct perf_event *event)		{ }
909 #endif
910 
911 #define perf_output_put(handle, x) \
912 	perf_output_copy((handle), &(x), sizeof(x))
913 
914 #endif /* __KERNEL__ */
915 #endif /* _LINUX_PERF_EVENT_H */
916