xref: /linux-6.15/include/linux/perf_event.h (revision 266fe2f2)
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 	__u32			__reserved_2;
215 
216 	__u64			bp_addr;
217 	__u32			bp_type;
218 	__u32			bp_len;
219 };
220 
221 /*
222  * Ioctls that can be done on a perf event fd:
223  */
224 #define PERF_EVENT_IOC_ENABLE		_IO ('$', 0)
225 #define PERF_EVENT_IOC_DISABLE		_IO ('$', 1)
226 #define PERF_EVENT_IOC_REFRESH		_IO ('$', 2)
227 #define PERF_EVENT_IOC_RESET		_IO ('$', 3)
228 #define PERF_EVENT_IOC_PERIOD		_IOW('$', 4, __u64)
229 #define PERF_EVENT_IOC_SET_OUTPUT	_IO ('$', 5)
230 #define PERF_EVENT_IOC_SET_FILTER	_IOW('$', 6, char *)
231 
232 enum perf_event_ioc_flags {
233 	PERF_IOC_FLAG_GROUP		= 1U << 0,
234 };
235 
236 /*
237  * Structure of the page that can be mapped via mmap
238  */
239 struct perf_event_mmap_page {
240 	__u32	version;		/* version number of this structure */
241 	__u32	compat_version;		/* lowest version this is compat with */
242 
243 	/*
244 	 * Bits needed to read the hw events in user-space.
245 	 *
246 	 *   u32 seq;
247 	 *   s64 count;
248 	 *
249 	 *   do {
250 	 *     seq = pc->lock;
251 	 *
252 	 *     barrier()
253 	 *     if (pc->index) {
254 	 *       count = pmc_read(pc->index - 1);
255 	 *       count += pc->offset;
256 	 *     } else
257 	 *       goto regular_read;
258 	 *
259 	 *     barrier();
260 	 *   } while (pc->lock != seq);
261 	 *
262 	 * NOTE: for obvious reason this only works on self-monitoring
263 	 *       processes.
264 	 */
265 	__u32	lock;			/* seqlock for synchronization */
266 	__u32	index;			/* hardware event identifier */
267 	__s64	offset;			/* add to hardware event value */
268 	__u64	time_enabled;		/* time event active */
269 	__u64	time_running;		/* time event on cpu */
270 
271 		/*
272 		 * Hole for extension of the self monitor capabilities
273 		 */
274 
275 	__u64	__reserved[123];	/* align to 1k */
276 
277 	/*
278 	 * Control data for the mmap() data buffer.
279 	 *
280 	 * User-space reading the @data_head value should issue an rmb(), on
281 	 * SMP capable platforms, after reading this value -- see
282 	 * perf_event_wakeup().
283 	 *
284 	 * When the mapping is PROT_WRITE the @data_tail value should be
285 	 * written by userspace to reflect the last read data. In this case
286 	 * the kernel will not over-write unread data.
287 	 */
288 	__u64   data_head;		/* head in the data section */
289 	__u64	data_tail;		/* user-space written tail */
290 };
291 
292 #define PERF_RECORD_MISC_CPUMODE_MASK		(3 << 0)
293 #define PERF_RECORD_MISC_CPUMODE_UNKNOWN	(0 << 0)
294 #define PERF_RECORD_MISC_KERNEL			(1 << 0)
295 #define PERF_RECORD_MISC_USER			(2 << 0)
296 #define PERF_RECORD_MISC_HYPERVISOR		(3 << 0)
297 
298 struct perf_event_header {
299 	__u32	type;
300 	__u16	misc;
301 	__u16	size;
302 };
303 
304 enum perf_event_type {
305 
306 	/*
307 	 * The MMAP events record the PROT_EXEC mappings so that we can
308 	 * correlate userspace IPs to code. They have the following structure:
309 	 *
310 	 * struct {
311 	 *	struct perf_event_header	header;
312 	 *
313 	 *	u32				pid, tid;
314 	 *	u64				addr;
315 	 *	u64				len;
316 	 *	u64				pgoff;
317 	 *	char				filename[];
318 	 * };
319 	 */
320 	PERF_RECORD_MMAP			= 1,
321 
322 	/*
323 	 * struct {
324 	 *	struct perf_event_header	header;
325 	 *	u64				id;
326 	 *	u64				lost;
327 	 * };
328 	 */
329 	PERF_RECORD_LOST			= 2,
330 
331 	/*
332 	 * struct {
333 	 *	struct perf_event_header	header;
334 	 *
335 	 *	u32				pid, tid;
336 	 *	char				comm[];
337 	 * };
338 	 */
339 	PERF_RECORD_COMM			= 3,
340 
341 	/*
342 	 * struct {
343 	 *	struct perf_event_header	header;
344 	 *	u32				pid, ppid;
345 	 *	u32				tid, ptid;
346 	 *	u64				time;
347 	 * };
348 	 */
349 	PERF_RECORD_EXIT			= 4,
350 
351 	/*
352 	 * struct {
353 	 *	struct perf_event_header	header;
354 	 *	u64				time;
355 	 *	u64				id;
356 	 *	u64				stream_id;
357 	 * };
358 	 */
359 	PERF_RECORD_THROTTLE			= 5,
360 	PERF_RECORD_UNTHROTTLE			= 6,
361 
362 	/*
363 	 * struct {
364 	 *	struct perf_event_header	header;
365 	 *	u32				pid, ppid;
366 	 *	u32				tid, ptid;
367 	 *	u64				time;
368 	 * };
369 	 */
370 	PERF_RECORD_FORK			= 7,
371 
372 	/*
373 	 * struct {
374 	 *	struct perf_event_header	header;
375 	 *	u32				pid, tid;
376 	 *
377 	 *	struct read_format		values;
378 	 * };
379 	 */
380 	PERF_RECORD_READ			= 8,
381 
382 	/*
383 	 * struct {
384 	 *	struct perf_event_header	header;
385 	 *
386 	 *	{ u64			ip;	  } && PERF_SAMPLE_IP
387 	 *	{ u32			pid, tid; } && PERF_SAMPLE_TID
388 	 *	{ u64			time;     } && PERF_SAMPLE_TIME
389 	 *	{ u64			addr;     } && PERF_SAMPLE_ADDR
390 	 *	{ u64			id;	  } && PERF_SAMPLE_ID
391 	 *	{ u64			stream_id;} && PERF_SAMPLE_STREAM_ID
392 	 *	{ u32			cpu, res; } && PERF_SAMPLE_CPU
393 	 *	{ u64			period;   } && PERF_SAMPLE_PERIOD
394 	 *
395 	 *	{ struct read_format	values;	  } && PERF_SAMPLE_READ
396 	 *
397 	 *	{ u64			nr,
398 	 *	  u64			ips[nr];  } && PERF_SAMPLE_CALLCHAIN
399 	 *
400 	 *	#
401 	 *	# The RAW record below is opaque data wrt the ABI
402 	 *	#
403 	 *	# That is, the ABI doesn't make any promises wrt to
404 	 *	# the stability of its content, it may vary depending
405 	 *	# on event, hardware, kernel version and phase of
406 	 *	# the moon.
407 	 *	#
408 	 *	# In other words, PERF_SAMPLE_RAW contents are not an ABI.
409 	 *	#
410 	 *
411 	 *	{ u32			size;
412 	 *	  char                  data[size];}&& PERF_SAMPLE_RAW
413 	 * };
414 	 */
415 	PERF_RECORD_SAMPLE			= 9,
416 
417 	PERF_RECORD_MAX,			/* non-ABI */
418 };
419 
420 enum perf_callchain_context {
421 	PERF_CONTEXT_HV			= (__u64)-32,
422 	PERF_CONTEXT_KERNEL		= (__u64)-128,
423 	PERF_CONTEXT_USER		= (__u64)-512,
424 
425 	PERF_CONTEXT_GUEST		= (__u64)-2048,
426 	PERF_CONTEXT_GUEST_KERNEL	= (__u64)-2176,
427 	PERF_CONTEXT_GUEST_USER		= (__u64)-2560,
428 
429 	PERF_CONTEXT_MAX		= (__u64)-4095,
430 };
431 
432 #define PERF_FLAG_FD_NO_GROUP	(1U << 0)
433 #define PERF_FLAG_FD_OUTPUT	(1U << 1)
434 
435 #ifdef __KERNEL__
436 /*
437  * Kernel-internal data types and definitions:
438  */
439 
440 #ifdef CONFIG_PERF_EVENTS
441 # include <asm/perf_event.h>
442 #endif
443 
444 #ifdef CONFIG_HAVE_HW_BREAKPOINT
445 #include <asm/hw_breakpoint.h>
446 #endif
447 
448 #include <linux/list.h>
449 #include <linux/mutex.h>
450 #include <linux/rculist.h>
451 #include <linux/rcupdate.h>
452 #include <linux/spinlock.h>
453 #include <linux/hrtimer.h>
454 #include <linux/fs.h>
455 #include <linux/pid_namespace.h>
456 #include <linux/workqueue.h>
457 #include <asm/atomic.h>
458 
459 #define PERF_MAX_STACK_DEPTH		255
460 
461 struct perf_callchain_entry {
462 	__u64				nr;
463 	__u64				ip[PERF_MAX_STACK_DEPTH];
464 };
465 
466 struct perf_raw_record {
467 	u32				size;
468 	void				*data;
469 };
470 
471 struct task_struct;
472 
473 /**
474  * struct hw_perf_event - performance event hardware details:
475  */
476 struct hw_perf_event {
477 #ifdef CONFIG_PERF_EVENTS
478 	union {
479 		struct { /* hardware */
480 			u64		config;
481 			u64		last_tag;
482 			unsigned long	config_base;
483 			unsigned long	event_base;
484 			int		idx;
485 			int		last_cpu;
486 		};
487 		struct { /* software */
488 			s64		remaining;
489 			struct hrtimer	hrtimer;
490 		};
491 #ifdef CONFIG_HAVE_HW_BREAKPOINT
492 		union { /* breakpoint */
493 			struct arch_hw_breakpoint	info;
494 		};
495 #endif
496 	};
497 	atomic64_t			prev_count;
498 	u64				sample_period;
499 	u64				last_period;
500 	atomic64_t			period_left;
501 	u64				interrupts;
502 
503 	u64				freq_time_stamp;
504 	u64				freq_count_stamp;
505 #endif
506 };
507 
508 struct perf_event;
509 
510 /**
511  * struct pmu - generic performance monitoring unit
512  */
513 struct pmu {
514 	int (*enable)			(struct perf_event *event);
515 	void (*disable)			(struct perf_event *event);
516 	void (*read)			(struct perf_event *event);
517 	void (*unthrottle)		(struct perf_event *event);
518 };
519 
520 /**
521  * enum perf_event_active_state - the states of a event
522  */
523 enum perf_event_active_state {
524 	PERF_EVENT_STATE_ERROR		= -2,
525 	PERF_EVENT_STATE_OFF		= -1,
526 	PERF_EVENT_STATE_INACTIVE	=  0,
527 	PERF_EVENT_STATE_ACTIVE		=  1,
528 };
529 
530 struct file;
531 
532 struct perf_mmap_data {
533 	struct rcu_head			rcu_head;
534 #ifdef CONFIG_PERF_USE_VMALLOC
535 	struct work_struct		work;
536 #endif
537 	int				data_order;
538 	int				nr_pages;	/* nr of data pages  */
539 	int				writable;	/* are we writable   */
540 	int				nr_locked;	/* nr pages mlocked  */
541 
542 	atomic_t			poll;		/* POLL_ for wakeups */
543 	atomic_t			events;		/* event_id limit       */
544 
545 	atomic_long_t			head;		/* write position    */
546 	atomic_long_t			done_head;	/* completed head    */
547 
548 	atomic_t			lock;		/* concurrent writes */
549 	atomic_t			wakeup;		/* needs a wakeup    */
550 	atomic_t			lost;		/* nr records lost   */
551 
552 	long				watermark;	/* wakeup watermark  */
553 
554 	struct perf_event_mmap_page	*user_page;
555 	void				*data_pages[0];
556 };
557 
558 struct perf_pending_entry {
559 	struct perf_pending_entry *next;
560 	void (*func)(struct perf_pending_entry *);
561 };
562 
563 struct perf_sample_data;
564 
565 typedef void (*perf_overflow_handler_t)(struct perf_event *, int,
566 					struct perf_sample_data *,
567 					struct pt_regs *regs);
568 
569 enum perf_group_flag {
570 	PERF_GROUP_SOFTWARE = 0x1,
571 };
572 
573 /**
574  * struct perf_event - performance event kernel representation:
575  */
576 struct perf_event {
577 #ifdef CONFIG_PERF_EVENTS
578 	struct list_head		group_entry;
579 	struct list_head		event_entry;
580 	struct list_head		sibling_list;
581 	int				nr_siblings;
582 	int				group_flags;
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_TRACING
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 	raw_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		pinned_groups;
693 	struct list_head		flexible_groups;
694 	struct list_head		event_list;
695 	int				nr_events;
696 	int				nr_active;
697 	int				is_active;
698 	int				nr_stat;
699 	atomic_t			refcount;
700 	struct task_struct		*task;
701 
702 	/*
703 	 * Context clock, runs when context enabled.
704 	 */
705 	u64				time;
706 	u64				timestamp;
707 
708 	/*
709 	 * These fields let us detect when two contexts have both
710 	 * been cloned (inherited) from a common ancestor.
711 	 */
712 	struct perf_event_context	*parent_ctx;
713 	u64				parent_gen;
714 	u64				generation;
715 	int				pin_count;
716 	struct rcu_head			rcu_head;
717 };
718 
719 /**
720  * struct perf_event_cpu_context - per cpu event context structure
721  */
722 struct perf_cpu_context {
723 	struct perf_event_context	ctx;
724 	struct perf_event_context	*task_ctx;
725 	int				active_oncpu;
726 	int				max_pertask;
727 	int				exclusive;
728 
729 	/*
730 	 * Recursion avoidance:
731 	 *
732 	 * task, softirq, irq, nmi context
733 	 */
734 	int				recursion[4];
735 };
736 
737 struct perf_output_handle {
738 	struct perf_event		*event;
739 	struct perf_mmap_data		*data;
740 	unsigned long			head;
741 	unsigned long			offset;
742 	int				nmi;
743 	int				sample;
744 	int				locked;
745 };
746 
747 #ifdef CONFIG_PERF_EVENTS
748 
749 /*
750  * Set by architecture code:
751  */
752 extern int perf_max_events;
753 
754 extern const struct pmu *hw_perf_event_init(struct perf_event *event);
755 
756 extern void perf_event_task_sched_in(struct task_struct *task);
757 extern void perf_event_task_sched_out(struct task_struct *task, struct task_struct *next);
758 extern void perf_event_task_tick(struct task_struct *task);
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 	switch (event->attr.type) {
824 	case PERF_TYPE_SOFTWARE:
825 	case PERF_TYPE_TRACEPOINT:
826 	/* for now the breakpoint stuff also works as software event */
827 	case PERF_TYPE_BREAKPOINT:
828 		return 1;
829 	}
830 	return 0;
831 }
832 
833 extern atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX];
834 
835 extern void __perf_sw_event(u32, u64, int, struct pt_regs *, u64);
836 
837 static inline void
838 perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr)
839 {
840 	if (atomic_read(&perf_swevent_enabled[event_id]))
841 		__perf_sw_event(event_id, nr, nmi, regs, addr);
842 }
843 
844 extern void __perf_event_mmap(struct vm_area_struct *vma);
845 
846 static inline void perf_event_mmap(struct vm_area_struct *vma)
847 {
848 	if (vma->vm_flags & VM_EXEC)
849 		__perf_event_mmap(vma);
850 }
851 
852 extern void perf_event_comm(struct task_struct *tsk);
853 extern void perf_event_fork(struct task_struct *tsk);
854 
855 extern struct perf_callchain_entry *perf_callchain(struct pt_regs *regs);
856 
857 extern int sysctl_perf_event_paranoid;
858 extern int sysctl_perf_event_mlock;
859 extern int sysctl_perf_event_sample_rate;
860 
861 extern void perf_event_init(void);
862 extern void perf_tp_event(int event_id, u64 addr, u64 count, 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 extern void perf_event_enable(struct perf_event *event);
880 extern void perf_event_disable(struct perf_event *event);
881 #else
882 static inline void
883 perf_event_task_sched_in(struct task_struct *task)			{ }
884 static inline void
885 perf_event_task_sched_out(struct task_struct *task,
886 			    struct task_struct *next)			{ }
887 static inline void
888 perf_event_task_tick(struct task_struct *task)				{ }
889 static inline int perf_event_init_task(struct task_struct *child)	{ return 0; }
890 static inline void perf_event_exit_task(struct task_struct *child)	{ }
891 static inline void perf_event_free_task(struct task_struct *task)	{ }
892 static inline void perf_event_do_pending(void)				{ }
893 static inline void perf_event_print_debug(void)				{ }
894 static inline void perf_disable(void)					{ }
895 static inline void perf_enable(void)					{ }
896 static inline int perf_event_task_disable(void)				{ return -EINVAL; }
897 static inline int perf_event_task_enable(void)				{ return -EINVAL; }
898 
899 static inline void
900 perf_sw_event(u32 event_id, u64 nr, int nmi,
901 		     struct pt_regs *regs, u64 addr)			{ }
902 static inline void
903 perf_bp_event(struct perf_event *event, void *data)			{ }
904 
905 static inline void perf_event_mmap(struct vm_area_struct *vma)		{ }
906 static inline void perf_event_comm(struct task_struct *tsk)		{ }
907 static inline void perf_event_fork(struct task_struct *tsk)		{ }
908 static inline void perf_event_init(void)				{ }
909 static inline int  perf_swevent_get_recursion_context(void)		{ return -1; }
910 static inline void perf_swevent_put_recursion_context(int rctx)		{ }
911 static inline void perf_event_enable(struct perf_event *event)		{ }
912 static inline void perf_event_disable(struct perf_event *event)		{ }
913 #endif
914 
915 #define perf_output_put(handle, x) \
916 	perf_output_copy((handle), &(x), sizeof(x))
917 
918 #endif /* __KERNEL__ */
919 #endif /* _LINUX_PERF_EVENT_H */
920