• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Performance events:
3  *
4  *    Copyright (C) 2008-2009, Thomas Gleixner <tglx@linutronix.de>
5  *    Copyright (C) 2008-2011, Red Hat, Inc., Ingo Molnar
6  *    Copyright (C) 2008-2011, 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 <uapi/linux/perf_event.h>
18 #include <uapi/linux/bpf_perf_event.h>
19 
20 /*
21  * Kernel-internal data types and definitions:
22  */
23 
24 #ifdef CONFIG_PERF_EVENTS
25 # include <asm/perf_event.h>
26 # include <asm/local64.h>
27 #endif
28 
29 struct perf_guest_info_callbacks {
30 	int				(*is_in_guest)(void);
31 	int				(*is_user_mode)(void);
32 	unsigned long			(*get_guest_ip)(void);
33 	void				(*handle_intel_pt_intr)(void);
34 };
35 
36 #ifdef CONFIG_HAVE_HW_BREAKPOINT
37 #include <asm/hw_breakpoint.h>
38 #endif
39 
40 #include <linux/list.h>
41 #include <linux/mutex.h>
42 #include <linux/rculist.h>
43 #include <linux/rcupdate.h>
44 #include <linux/spinlock.h>
45 #include <linux/hrtimer.h>
46 #include <linux/fs.h>
47 #include <linux/pid_namespace.h>
48 #include <linux/workqueue.h>
49 #include <linux/ftrace.h>
50 #include <linux/cpu.h>
51 #include <linux/irq_work.h>
52 #include <linux/static_key.h>
53 #include <linux/jump_label_ratelimit.h>
54 #include <linux/atomic.h>
55 #include <linux/sysfs.h>
56 #include <linux/perf_regs.h>
57 #include <linux/cgroup.h>
58 #include <linux/refcount.h>
59 #include <linux/security.h>
60 #include <asm/local.h>
61 
62 struct perf_callchain_entry {
63 	__u64				nr;
64 	__u64				ip[]; /* /proc/sys/kernel/perf_event_max_stack */
65 };
66 
67 struct perf_callchain_entry_ctx {
68 	struct perf_callchain_entry *entry;
69 	u32			    max_stack;
70 	u32			    nr;
71 	short			    contexts;
72 	bool			    contexts_maxed;
73 };
74 
75 typedef unsigned long (*perf_copy_f)(void *dst, const void *src,
76 				     unsigned long off, unsigned long len);
77 
78 struct perf_raw_frag {
79 	union {
80 		struct perf_raw_frag	*next;
81 		unsigned long		pad;
82 	};
83 	perf_copy_f			copy;
84 	void				*data;
85 	u32				size;
86 } __packed;
87 
88 struct perf_raw_record {
89 	struct perf_raw_frag		frag;
90 	u32				size;
91 };
92 
93 /*
94  * branch stack layout:
95  *  nr: number of taken branches stored in entries[]
96  *  hw_idx: The low level index of raw branch records
97  *          for the most recent branch.
98  *          -1ULL means invalid/unknown.
99  *
100  * Note that nr can vary from sample to sample
101  * branches (to, from) are stored from most recent
102  * to least recent, i.e., entries[0] contains the most
103  * recent branch.
104  * The entries[] is an abstraction of raw branch records,
105  * which may not be stored in age order in HW, e.g. Intel LBR.
106  * The hw_idx is to expose the low level index of raw
107  * branch record for the most recent branch aka entries[0].
108  * The hw_idx index is between -1 (unknown) and max depth,
109  * which can be retrieved in /sys/devices/cpu/caps/branches.
110  * For the architectures whose raw branch records are
111  * already stored in age order, the hw_idx should be 0.
112  */
113 struct perf_branch_stack {
114 	__u64				nr;
115 	__u64				hw_idx;
116 	struct perf_branch_entry	entries[];
117 };
118 
119 struct task_struct;
120 
121 /*
122  * extra PMU register associated with an event
123  */
124 struct hw_perf_event_extra {
125 	u64		config;	/* register value */
126 	unsigned int	reg;	/* register address or index */
127 	int		alloc;	/* extra register already allocated */
128 	int		idx;	/* index in shared_regs->regs[] */
129 };
130 
131 /**
132  * struct hw_perf_event - performance event hardware details:
133  */
134 struct hw_perf_event {
135 #ifdef CONFIG_PERF_EVENTS
136 	union {
137 		struct { /* hardware */
138 			u64		config;
139 			u64		last_tag;
140 			unsigned long	config_base;
141 			unsigned long	event_base;
142 			int		event_base_rdpmc;
143 			int		idx;
144 			int		last_cpu;
145 			int		flags;
146 
147 			struct hw_perf_event_extra extra_reg;
148 			struct hw_perf_event_extra branch_reg;
149 		};
150 		struct { /* software */
151 			struct hrtimer	hrtimer;
152 		};
153 		struct { /* tracepoint */
154 			/* for tp_event->class */
155 			struct list_head	tp_list;
156 		};
157 		struct { /* amd_power */
158 			u64	pwr_acc;
159 			u64	ptsc;
160 		};
161 #ifdef CONFIG_HAVE_HW_BREAKPOINT
162 		struct { /* breakpoint */
163 			/*
164 			 * Crufty hack to avoid the chicken and egg
165 			 * problem hw_breakpoint has with context
166 			 * creation and event initalization.
167 			 */
168 			struct arch_hw_breakpoint	info;
169 			struct list_head		bp_list;
170 		};
171 #endif
172 		struct { /* amd_iommu */
173 			u8	iommu_bank;
174 			u8	iommu_cntr;
175 			u16	padding;
176 			u64	conf;
177 			u64	conf1;
178 		};
179 	};
180 	/*
181 	 * If the event is a per task event, this will point to the task in
182 	 * question. See the comment in perf_event_alloc().
183 	 */
184 	struct task_struct		*target;
185 
186 	/*
187 	 * PMU would store hardware filter configuration
188 	 * here.
189 	 */
190 	void				*addr_filters;
191 
192 	/* Last sync'ed generation of filters */
193 	unsigned long			addr_filters_gen;
194 
195 /*
196  * hw_perf_event::state flags; used to track the PERF_EF_* state.
197  */
198 #define PERF_HES_STOPPED	0x01 /* the counter is stopped */
199 #define PERF_HES_UPTODATE	0x02 /* event->count up-to-date */
200 #define PERF_HES_ARCH		0x04
201 
202 	int				state;
203 
204 	/*
205 	 * The last observed hardware counter value, updated with a
206 	 * local64_cmpxchg() such that pmu::read() can be called nested.
207 	 */
208 	local64_t			prev_count;
209 
210 	/*
211 	 * The period to start the next sample with.
212 	 */
213 	u64				sample_period;
214 
215 	union {
216 		struct { /* Sampling */
217 			/*
218 			 * The period we started this sample with.
219 			 */
220 			u64				last_period;
221 
222 			/*
223 			 * However much is left of the current period;
224 			 * note that this is a full 64bit value and
225 			 * allows for generation of periods longer
226 			 * than hardware might allow.
227 			 */
228 			local64_t			period_left;
229 		};
230 		struct { /* Topdown events counting for context switch */
231 			u64				saved_metric;
232 			u64				saved_slots;
233 		};
234 	};
235 
236 	/*
237 	 * State for throttling the event, see __perf_event_overflow() and
238 	 * perf_adjust_freq_unthr_context().
239 	 */
240 	u64                             interrupts_seq;
241 	u64				interrupts;
242 
243 	/*
244 	 * State for freq target events, see __perf_event_overflow() and
245 	 * perf_adjust_freq_unthr_context().
246 	 */
247 	u64				freq_time_stamp;
248 	u64				freq_count_stamp;
249 #endif
250 };
251 
252 struct perf_event;
253 
254 /*
255  * Common implementation detail of pmu::{start,commit,cancel}_txn
256  */
257 #define PERF_PMU_TXN_ADD  0x1		/* txn to add/schedule event on PMU */
258 #define PERF_PMU_TXN_READ 0x2		/* txn to read event group from PMU */
259 
260 /**
261  * pmu::capabilities flags
262  */
263 #define PERF_PMU_CAP_NO_INTERRUPT		0x0001
264 #define PERF_PMU_CAP_NO_NMI			0x0002
265 #define PERF_PMU_CAP_AUX_NO_SG			0x0004
266 #define PERF_PMU_CAP_EXTENDED_REGS		0x0008
267 #define PERF_PMU_CAP_EXCLUSIVE			0x0010
268 #define PERF_PMU_CAP_ITRACE			0x0020
269 #define PERF_PMU_CAP_HETEROGENEOUS_CPUS		0x0040
270 #define PERF_PMU_CAP_NO_EXCLUDE			0x0080
271 #define PERF_PMU_CAP_AUX_OUTPUT			0x0100
272 #define PERF_PMU_CAP_EXTENDED_HW_TYPE		0x0200
273 
274 struct perf_output_handle;
275 
276 /**
277  * struct pmu - generic performance monitoring unit
278  */
279 struct pmu {
280 	struct list_head		entry;
281 
282 	struct module			*module;
283 	struct device			*dev;
284 	const struct attribute_group	**attr_groups;
285 	const struct attribute_group	**attr_update;
286 	const char			*name;
287 	int				type;
288 
289 	/*
290 	 * various common per-pmu feature flags
291 	 */
292 	int				capabilities;
293 
294 	int __percpu			*pmu_disable_count;
295 	struct perf_cpu_context __percpu *pmu_cpu_context;
296 	atomic_t			exclusive_cnt; /* < 0: cpu; > 0: tsk */
297 	int				task_ctx_nr;
298 	int				hrtimer_interval_ms;
299 
300 	/* number of address filters this PMU can do */
301 	unsigned int			nr_addr_filters;
302 
303 	/*
304 	 * Fully disable/enable this PMU, can be used to protect from the PMI
305 	 * as well as for lazy/batch writing of the MSRs.
306 	 */
307 	void (*pmu_enable)		(struct pmu *pmu); /* optional */
308 	void (*pmu_disable)		(struct pmu *pmu); /* optional */
309 
310 	/*
311 	 * Try and initialize the event for this PMU.
312 	 *
313 	 * Returns:
314 	 *  -ENOENT	-- @event is not for this PMU
315 	 *
316 	 *  -ENODEV	-- @event is for this PMU but PMU not present
317 	 *  -EBUSY	-- @event is for this PMU but PMU temporarily unavailable
318 	 *  -EINVAL	-- @event is for this PMU but @event is not valid
319 	 *  -EOPNOTSUPP -- @event is for this PMU, @event is valid, but not supported
320 	 *  -EACCES	-- @event is for this PMU, @event is valid, but no privileges
321 	 *
322 	 *  0		-- @event is for this PMU and valid
323 	 *
324 	 * Other error return values are allowed.
325 	 */
326 	int (*event_init)		(struct perf_event *event);
327 
328 	/*
329 	 * Notification that the event was mapped or unmapped.  Called
330 	 * in the context of the mapping task.
331 	 */
332 	void (*event_mapped)		(struct perf_event *event, struct mm_struct *mm); /* optional */
333 	void (*event_unmapped)		(struct perf_event *event, struct mm_struct *mm); /* optional */
334 
335 	/*
336 	 * Flags for ->add()/->del()/ ->start()/->stop(). There are
337 	 * matching hw_perf_event::state flags.
338 	 */
339 #define PERF_EF_START	0x01		/* start the counter when adding    */
340 #define PERF_EF_RELOAD	0x02		/* reload the counter when starting */
341 #define PERF_EF_UPDATE	0x04		/* update the counter when stopping */
342 
343 	/*
344 	 * Adds/Removes a counter to/from the PMU, can be done inside a
345 	 * transaction, see the ->*_txn() methods.
346 	 *
347 	 * The add/del callbacks will reserve all hardware resources required
348 	 * to service the event, this includes any counter constraint
349 	 * scheduling etc.
350 	 *
351 	 * Called with IRQs disabled and the PMU disabled on the CPU the event
352 	 * is on.
353 	 *
354 	 * ->add() called without PERF_EF_START should result in the same state
355 	 *  as ->add() followed by ->stop().
356 	 *
357 	 * ->del() must always PERF_EF_UPDATE stop an event. If it calls
358 	 *  ->stop() that must deal with already being stopped without
359 	 *  PERF_EF_UPDATE.
360 	 */
361 	int  (*add)			(struct perf_event *event, int flags);
362 	void (*del)			(struct perf_event *event, int flags);
363 
364 	/*
365 	 * Starts/Stops a counter present on the PMU.
366 	 *
367 	 * The PMI handler should stop the counter when perf_event_overflow()
368 	 * returns !0. ->start() will be used to continue.
369 	 *
370 	 * Also used to change the sample period.
371 	 *
372 	 * Called with IRQs disabled and the PMU disabled on the CPU the event
373 	 * is on -- will be called from NMI context with the PMU generates
374 	 * NMIs.
375 	 *
376 	 * ->stop() with PERF_EF_UPDATE will read the counter and update
377 	 *  period/count values like ->read() would.
378 	 *
379 	 * ->start() with PERF_EF_RELOAD will reprogram the counter
380 	 *  value, must be preceded by a ->stop() with PERF_EF_UPDATE.
381 	 */
382 	void (*start)			(struct perf_event *event, int flags);
383 	void (*stop)			(struct perf_event *event, int flags);
384 
385 	/*
386 	 * Updates the counter value of the event.
387 	 *
388 	 * For sampling capable PMUs this will also update the software period
389 	 * hw_perf_event::period_left field.
390 	 */
391 	void (*read)			(struct perf_event *event);
392 
393 	/*
394 	 * Group events scheduling is treated as a transaction, add
395 	 * group events as a whole and perform one schedulability test.
396 	 * If the test fails, roll back the whole group
397 	 *
398 	 * Start the transaction, after this ->add() doesn't need to
399 	 * do schedulability tests.
400 	 *
401 	 * Optional.
402 	 */
403 	void (*start_txn)		(struct pmu *pmu, unsigned int txn_flags);
404 	/*
405 	 * If ->start_txn() disabled the ->add() schedulability test
406 	 * then ->commit_txn() is required to perform one. On success
407 	 * the transaction is closed. On error the transaction is kept
408 	 * open until ->cancel_txn() is called.
409 	 *
410 	 * Optional.
411 	 */
412 	int  (*commit_txn)		(struct pmu *pmu);
413 	/*
414 	 * Will cancel the transaction, assumes ->del() is called
415 	 * for each successful ->add() during the transaction.
416 	 *
417 	 * Optional.
418 	 */
419 	void (*cancel_txn)		(struct pmu *pmu);
420 
421 	/*
422 	 * Will return the value for perf_event_mmap_page::index for this event,
423 	 * if no implementation is provided it will default to: event->hw.idx + 1.
424 	 */
425 	int (*event_idx)		(struct perf_event *event); /*optional */
426 
427 	/*
428 	 * context-switches callback
429 	 */
430 	void (*sched_task)		(struct perf_event_context *ctx,
431 					bool sched_in);
432 
433 	/*
434 	 * Kmem cache of PMU specific data
435 	 */
436 	struct kmem_cache		*task_ctx_cache;
437 
438 	/*
439 	 * PMU specific parts of task perf event context (i.e. ctx->task_ctx_data)
440 	 * can be synchronized using this function. See Intel LBR callstack support
441 	 * implementation and Perf core context switch handling callbacks for usage
442 	 * examples.
443 	 */
444 	void (*swap_task_ctx)		(struct perf_event_context *prev,
445 					 struct perf_event_context *next);
446 					/* optional */
447 
448 	/*
449 	 * Set up pmu-private data structures for an AUX area
450 	 */
451 	void *(*setup_aux)		(struct perf_event *event, void **pages,
452 					 int nr_pages, bool overwrite);
453 					/* optional */
454 
455 	/*
456 	 * Free pmu-private AUX data structures
457 	 */
458 	void (*free_aux)		(void *aux); /* optional */
459 
460 	/*
461 	 * Take a snapshot of the AUX buffer without touching the event
462 	 * state, so that preempting ->start()/->stop() callbacks does
463 	 * not interfere with their logic. Called in PMI context.
464 	 *
465 	 * Returns the size of AUX data copied to the output handle.
466 	 *
467 	 * Optional.
468 	 */
469 	long (*snapshot_aux)		(struct perf_event *event,
470 					 struct perf_output_handle *handle,
471 					 unsigned long size);
472 
473 	/*
474 	 * Validate address range filters: make sure the HW supports the
475 	 * requested configuration and number of filters; return 0 if the
476 	 * supplied filters are valid, -errno otherwise.
477 	 *
478 	 * Runs in the context of the ioctl()ing process and is not serialized
479 	 * with the rest of the PMU callbacks.
480 	 */
481 	int (*addr_filters_validate)	(struct list_head *filters);
482 					/* optional */
483 
484 	/*
485 	 * Synchronize address range filter configuration:
486 	 * translate hw-agnostic filters into hardware configuration in
487 	 * event::hw::addr_filters.
488 	 *
489 	 * Runs as a part of filter sync sequence that is done in ->start()
490 	 * callback by calling perf_event_addr_filters_sync().
491 	 *
492 	 * May (and should) traverse event::addr_filters::list, for which its
493 	 * caller provides necessary serialization.
494 	 */
495 	void (*addr_filters_sync)	(struct perf_event *event);
496 					/* optional */
497 
498 	/*
499 	 * Check if event can be used for aux_output purposes for
500 	 * events of this PMU.
501 	 *
502 	 * Runs from perf_event_open(). Should return 0 for "no match"
503 	 * or non-zero for "match".
504 	 */
505 	int (*aux_output_match)		(struct perf_event *event);
506 					/* optional */
507 
508 	/*
509 	 * Filter events for PMU-specific reasons.
510 	 */
511 	int (*filter_match)		(struct perf_event *event); /* optional */
512 
513 	/*
514 	 * Check period value for PERF_EVENT_IOC_PERIOD ioctl.
515 	 */
516 	int (*check_period)		(struct perf_event *event, u64 value); /* optional */
517 };
518 
519 enum perf_addr_filter_action_t {
520 	PERF_ADDR_FILTER_ACTION_STOP = 0,
521 	PERF_ADDR_FILTER_ACTION_START,
522 	PERF_ADDR_FILTER_ACTION_FILTER,
523 };
524 
525 /**
526  * struct perf_addr_filter - address range filter definition
527  * @entry:	event's filter list linkage
528  * @path:	object file's path for file-based filters
529  * @offset:	filter range offset
530  * @size:	filter range size (size==0 means single address trigger)
531  * @action:	filter/start/stop
532  *
533  * This is a hardware-agnostic filter configuration as specified by the user.
534  */
535 struct perf_addr_filter {
536 	struct list_head	entry;
537 	struct path		path;
538 	unsigned long		offset;
539 	unsigned long		size;
540 	enum perf_addr_filter_action_t	action;
541 };
542 
543 /**
544  * struct perf_addr_filters_head - container for address range filters
545  * @list:	list of filters for this event
546  * @lock:	spinlock that serializes accesses to the @list and event's
547  *		(and its children's) filter generations.
548  * @nr_file_filters:	number of file-based filters
549  *
550  * A child event will use parent's @list (and therefore @lock), so they are
551  * bundled together; see perf_event_addr_filters().
552  */
553 struct perf_addr_filters_head {
554 	struct list_head	list;
555 	raw_spinlock_t		lock;
556 	unsigned int		nr_file_filters;
557 };
558 
559 struct perf_addr_filter_range {
560 	unsigned long		start;
561 	unsigned long		size;
562 };
563 
564 /**
565  * enum perf_event_state - the states of an event:
566  */
567 enum perf_event_state {
568 	PERF_EVENT_STATE_DEAD		= -4,
569 	PERF_EVENT_STATE_EXIT		= -3,
570 	PERF_EVENT_STATE_ERROR		= -2,
571 	PERF_EVENT_STATE_OFF		= -1,
572 	PERF_EVENT_STATE_INACTIVE	=  0,
573 	PERF_EVENT_STATE_ACTIVE		=  1,
574 };
575 
576 struct file;
577 struct perf_sample_data;
578 
579 typedef void (*perf_overflow_handler_t)(struct perf_event *,
580 					struct perf_sample_data *,
581 					struct pt_regs *regs);
582 
583 /*
584  * Event capabilities. For event_caps and groups caps.
585  *
586  * PERF_EV_CAP_SOFTWARE: Is a software event.
587  * PERF_EV_CAP_READ_ACTIVE_PKG: A CPU event (or cgroup event) that can be read
588  * from any CPU in the package where it is active.
589  * PERF_EV_CAP_SIBLING: An event with this flag must be a group sibling and
590  * cannot be a group leader. If an event with this flag is detached from the
591  * group it is scheduled out and moved into an unrecoverable ERROR state.
592  */
593 #define PERF_EV_CAP_SOFTWARE		BIT(0)
594 #define PERF_EV_CAP_READ_ACTIVE_PKG	BIT(1)
595 #define PERF_EV_CAP_SIBLING		BIT(2)
596 
597 #define SWEVENT_HLIST_BITS		8
598 #define SWEVENT_HLIST_SIZE		(1 << SWEVENT_HLIST_BITS)
599 
600 struct swevent_hlist {
601 	struct hlist_head		heads[SWEVENT_HLIST_SIZE];
602 	struct rcu_head			rcu_head;
603 };
604 
605 #define PERF_ATTACH_CONTEXT	0x01
606 #define PERF_ATTACH_GROUP	0x02
607 #define PERF_ATTACH_TASK	0x04
608 #define PERF_ATTACH_TASK_DATA	0x08
609 #define PERF_ATTACH_ITRACE	0x10
610 #define PERF_ATTACH_SCHED_CB	0x20
611 #define PERF_ATTACH_CHILD	0x40
612 
613 struct perf_cgroup;
614 struct perf_buffer;
615 
616 struct pmu_event_list {
617 	raw_spinlock_t		lock;
618 	struct list_head	list;
619 };
620 
621 #define for_each_sibling_event(sibling, event)			\
622 	if ((event)->group_leader == (event))			\
623 		list_for_each_entry((sibling), &(event)->sibling_list, sibling_list)
624 
625 /**
626  * struct perf_event - performance event kernel representation:
627  */
628 struct perf_event {
629 #ifdef CONFIG_PERF_EVENTS
630 	/*
631 	 * entry onto perf_event_context::event_list;
632 	 *   modifications require ctx->lock
633 	 *   RCU safe iterations.
634 	 */
635 	struct list_head		event_entry;
636 
637 	/*
638 	 * Locked for modification by both ctx->mutex and ctx->lock; holding
639 	 * either sufficies for read.
640 	 */
641 	struct list_head		sibling_list;
642 	struct list_head		active_list;
643 	/*
644 	 * Node on the pinned or flexible tree located at the event context;
645 	 */
646 	struct rb_node			group_node;
647 	u64				group_index;
648 	/*
649 	 * We need storage to track the entries in perf_pmu_migrate_context; we
650 	 * cannot use the event_entry because of RCU and we want to keep the
651 	 * group in tact which avoids us using the other two entries.
652 	 */
653 	struct list_head		migrate_entry;
654 
655 	struct hlist_node		hlist_entry;
656 	struct list_head		active_entry;
657 	int				nr_siblings;
658 
659 	/* Not serialized. Only written during event initialization. */
660 	int				event_caps;
661 	/* The cumulative AND of all event_caps for events in this group. */
662 	int				group_caps;
663 
664 #ifndef __GENKSYMS__
665 	unsigned int			group_generation;
666 #endif
667 	struct perf_event		*group_leader;
668 	struct pmu			*pmu;
669 	void				*pmu_private;
670 
671 	enum perf_event_state		state;
672 	unsigned int			attach_state;
673 	local64_t			count;
674 	atomic64_t			child_count;
675 
676 	/*
677 	 * These are the total time in nanoseconds that the event
678 	 * has been enabled (i.e. eligible to run, and the task has
679 	 * been scheduled in, if this is a per-task event)
680 	 * and running (scheduled onto the CPU), respectively.
681 	 */
682 	u64				total_time_enabled;
683 	u64				total_time_running;
684 	u64				tstamp;
685 
686 	struct perf_event_attr		attr;
687 	u16				header_size;
688 	u16				id_header_size;
689 	u16				read_size;
690 	struct hw_perf_event		hw;
691 
692 	struct perf_event_context	*ctx;
693 	atomic_long_t			refcount;
694 
695 	/*
696 	 * These accumulate total time (in nanoseconds) that children
697 	 * events have been enabled and running, respectively.
698 	 */
699 	atomic64_t			child_total_time_enabled;
700 	atomic64_t			child_total_time_running;
701 
702 	/*
703 	 * Protect attach/detach and child_list:
704 	 */
705 	struct mutex			child_mutex;
706 	struct list_head		child_list;
707 	struct perf_event		*parent;
708 
709 	int				oncpu;
710 	int				cpu;
711 
712 	struct list_head		owner_entry;
713 	struct task_struct		*owner;
714 
715 	/* mmap bits */
716 	struct mutex			mmap_mutex;
717 	atomic_t			mmap_count;
718 
719 	struct perf_buffer		*rb;
720 	struct list_head		rb_entry;
721 	unsigned long			rcu_batches;
722 	int				rcu_pending;
723 
724 	/* poll related */
725 	wait_queue_head_t		waitq;
726 	struct fasync_struct		*fasync;
727 
728 	/* delayed work for NMIs and such */
729 	int				pending_wakeup;
730 	int				pending_kill;
731 	int				pending_disable;
732 	unsigned long			pending_addr;	/* SIGTRAP */
733 	struct irq_work			pending;
734 
735 	atomic_t			event_limit;
736 
737 	/* address range filters */
738 	struct perf_addr_filters_head	addr_filters;
739 	/* vma address array for file-based filders */
740 	struct perf_addr_filter_range	*addr_filter_ranges;
741 	unsigned long			addr_filters_gen;
742 
743 	/* for aux_output events */
744 	struct perf_event		*aux_event;
745 
746 	void (*destroy)(struct perf_event *);
747 	struct rcu_head			rcu_head;
748 
749 	struct pid_namespace		*ns;
750 	u64				id;
751 
752 	u64				(*clock)(void);
753 	perf_overflow_handler_t		overflow_handler;
754 	void				*overflow_handler_context;
755 #ifdef CONFIG_BPF_SYSCALL
756 	perf_overflow_handler_t		orig_overflow_handler;
757 	struct bpf_prog			*prog;
758 	u64				bpf_cookie;
759 #endif
760 
761 #ifdef CONFIG_EVENT_TRACING
762 	struct trace_event_call		*tp_event;
763 	struct event_filter		*filter;
764 #ifdef CONFIG_FUNCTION_TRACER
765 	struct ftrace_ops               ftrace_ops;
766 #endif
767 #endif
768 
769 #ifdef CONFIG_CGROUP_PERF
770 	struct perf_cgroup		*cgrp; /* cgroup event is attach to */
771 #endif
772 
773 #ifdef CONFIG_SECURITY
774 	void *security;
775 #endif
776 	struct list_head		sb_list;
777 #endif /* CONFIG_PERF_EVENTS */
778 };
779 
780 
781 struct perf_event_groups {
782 	struct rb_root	tree;
783 	u64		index;
784 };
785 
786 /**
787  * struct perf_event_context - event context structure
788  *
789  * Used as a container for task events and CPU events as well:
790  */
791 struct perf_event_context {
792 	struct pmu			*pmu;
793 	/*
794 	 * Protect the states of the events in the list,
795 	 * nr_active, and the list:
796 	 */
797 	raw_spinlock_t			lock;
798 	/*
799 	 * Protect the list of events.  Locking either mutex or lock
800 	 * is sufficient to ensure the list doesn't change; to change
801 	 * the list you need to lock both the mutex and the spinlock.
802 	 */
803 	struct mutex			mutex;
804 
805 	struct list_head		active_ctx_list;
806 	struct perf_event_groups	pinned_groups;
807 	struct perf_event_groups	flexible_groups;
808 	struct list_head		event_list;
809 
810 	struct list_head		pinned_active;
811 	struct list_head		flexible_active;
812 
813 	int				nr_events;
814 	int				nr_active;
815 	int				is_active;
816 	int				nr_stat;
817 	int				nr_freq;
818 	int				rotate_disable;
819 	/*
820 	 * Set when nr_events != nr_active, except tolerant to events not
821 	 * necessary to be active due to scheduling constraints, such as cgroups.
822 	 */
823 	int				rotate_necessary;
824 	refcount_t			refcount;
825 	struct task_struct		*task;
826 
827 	/*
828 	 * Context clock, runs when context enabled.
829 	 */
830 	u64				time;
831 	u64				timestamp;
832 	u64				timeoffset;
833 
834 	/*
835 	 * These fields let us detect when two contexts have both
836 	 * been cloned (inherited) from a common ancestor.
837 	 */
838 	struct perf_event_context	*parent_ctx;
839 	u64				parent_gen;
840 	u64				generation;
841 	int				pin_count;
842 #ifdef CONFIG_CGROUP_PERF
843 	int				nr_cgroups;	 /* cgroup evts */
844 #endif
845 	void				*task_ctx_data; /* pmu specific data */
846 	struct rcu_head			rcu_head;
847 };
848 
849 /*
850  * Number of contexts where an event can trigger:
851  *	task, softirq, hardirq, nmi.
852  */
853 #define PERF_NR_CONTEXTS	4
854 
855 /**
856  * struct perf_event_cpu_context - per cpu event context structure
857  */
858 struct perf_cpu_context {
859 	struct perf_event_context	ctx;
860 	struct perf_event_context	*task_ctx;
861 	int				active_oncpu;
862 	int				exclusive;
863 
864 	raw_spinlock_t			hrtimer_lock;
865 	struct hrtimer			hrtimer;
866 	ktime_t				hrtimer_interval;
867 	unsigned int			hrtimer_active;
868 
869 #ifdef CONFIG_CGROUP_PERF
870 	struct perf_cgroup		*cgrp;
871 	struct list_head		cgrp_cpuctx_entry;
872 #endif
873 
874 	struct list_head		sched_cb_entry;
875 	int				sched_cb_usage;
876 
877 	int				online;
878 	/*
879 	 * Per-CPU storage for iterators used in visit_groups_merge. The default
880 	 * storage is of size 2 to hold the CPU and any CPU event iterators.
881 	 */
882 	int				heap_size;
883 	struct perf_event		**heap;
884 	struct perf_event		*heap_default[2];
885 };
886 
887 struct perf_output_handle {
888 	struct perf_event		*event;
889 	struct perf_buffer		*rb;
890 	unsigned long			wakeup;
891 	unsigned long			size;
892 	u64				aux_flags;
893 	union {
894 		void			*addr;
895 		unsigned long		head;
896 	};
897 	int				page;
898 };
899 
900 struct bpf_perf_event_data_kern {
901 	bpf_user_pt_regs_t *regs;
902 	struct perf_sample_data *data;
903 	struct perf_event *event;
904 };
905 
906 #ifdef CONFIG_CGROUP_PERF
907 
908 /*
909  * perf_cgroup_info keeps track of time_enabled for a cgroup.
910  * This is a per-cpu dynamically allocated data structure.
911  */
912 struct perf_cgroup_info {
913 	u64				time;
914 	u64				timestamp;
915 	u64				timeoffset;
916 	int				active;
917 };
918 
919 struct perf_cgroup {
920 	struct cgroup_subsys_state	css;
921 	struct perf_cgroup_info	__percpu *info;
922 };
923 
924 /*
925  * Must ensure cgroup is pinned (css_get) before calling
926  * this function. In other words, we cannot call this function
927  * if there is no cgroup event for the current CPU context.
928  */
929 static inline struct perf_cgroup *
perf_cgroup_from_task(struct task_struct * task,struct perf_event_context * ctx)930 perf_cgroup_from_task(struct task_struct *task, struct perf_event_context *ctx)
931 {
932 	return container_of(task_css_check(task, perf_event_cgrp_id,
933 					   ctx ? lockdep_is_held(&ctx->lock)
934 					       : true),
935 			    struct perf_cgroup, css);
936 }
937 #endif /* CONFIG_CGROUP_PERF */
938 
939 #ifdef CONFIG_PERF_EVENTS
940 
941 extern void *perf_aux_output_begin(struct perf_output_handle *handle,
942 				   struct perf_event *event);
943 extern void perf_aux_output_end(struct perf_output_handle *handle,
944 				unsigned long size);
945 extern int perf_aux_output_skip(struct perf_output_handle *handle,
946 				unsigned long size);
947 extern void *perf_get_aux(struct perf_output_handle *handle);
948 extern void perf_aux_output_flag(struct perf_output_handle *handle, u64 flags);
949 extern void perf_event_itrace_started(struct perf_event *event);
950 
951 extern int perf_pmu_register(struct pmu *pmu, const char *name, int type);
952 extern void perf_pmu_unregister(struct pmu *pmu);
953 
954 extern void __perf_event_task_sched_in(struct task_struct *prev,
955 				       struct task_struct *task);
956 extern void __perf_event_task_sched_out(struct task_struct *prev,
957 					struct task_struct *next);
958 extern int perf_event_init_task(struct task_struct *child, u64 clone_flags);
959 extern void perf_event_exit_task(struct task_struct *child);
960 extern void perf_event_free_task(struct task_struct *task);
961 extern void perf_event_delayed_put(struct task_struct *task);
962 extern struct file *perf_event_get(unsigned int fd);
963 extern const struct perf_event *perf_get_event(struct file *file);
964 extern const struct perf_event_attr *perf_event_attrs(struct perf_event *event);
965 extern void perf_event_print_debug(void);
966 extern void perf_pmu_disable(struct pmu *pmu);
967 extern void perf_pmu_enable(struct pmu *pmu);
968 extern void perf_sched_cb_dec(struct pmu *pmu);
969 extern void perf_sched_cb_inc(struct pmu *pmu);
970 extern int perf_event_task_disable(void);
971 extern int perf_event_task_enable(void);
972 
973 extern void perf_pmu_resched(struct pmu *pmu);
974 
975 extern int perf_event_refresh(struct perf_event *event, int refresh);
976 extern void perf_event_update_userpage(struct perf_event *event);
977 extern int perf_event_release_kernel(struct perf_event *event);
978 extern struct perf_event *
979 perf_event_create_kernel_counter(struct perf_event_attr *attr,
980 				int cpu,
981 				struct task_struct *task,
982 				perf_overflow_handler_t callback,
983 				void *context);
984 extern void perf_pmu_migrate_context(struct pmu *pmu,
985 				int src_cpu, int dst_cpu);
986 int perf_event_read_local(struct perf_event *event, u64 *value,
987 			  u64 *enabled, u64 *running);
988 extern u64 perf_event_read_value(struct perf_event *event,
989 				 u64 *enabled, u64 *running);
990 
991 
992 struct perf_sample_data {
993 	/*
994 	 * Fields set by perf_sample_data_init(), group so as to
995 	 * minimize the cachelines touched.
996 	 */
997 	u64				addr;
998 	struct perf_raw_record		*raw;
999 	struct perf_branch_stack	*br_stack;
1000 	u64				period;
1001 	union perf_sample_weight	weight;
1002 	u64				txn;
1003 	union  perf_mem_data_src	data_src;
1004 
1005 	/*
1006 	 * The other fields, optionally {set,used} by
1007 	 * perf_{prepare,output}_sample().
1008 	 */
1009 	u64				type;
1010 	u64				ip;
1011 	struct {
1012 		u32	pid;
1013 		u32	tid;
1014 	}				tid_entry;
1015 	u64				time;
1016 	u64				id;
1017 	u64				stream_id;
1018 	struct {
1019 		u32	cpu;
1020 		u32	reserved;
1021 	}				cpu_entry;
1022 	struct perf_callchain_entry	*callchain;
1023 	u64				aux_size;
1024 
1025 	struct perf_regs		regs_user;
1026 	struct perf_regs		regs_intr;
1027 	u64				stack_user_size;
1028 
1029 	u64				phys_addr;
1030 	u64				cgroup;
1031 	u64				data_page_size;
1032 	u64				code_page_size;
1033 } ____cacheline_aligned;
1034 
1035 /* default value for data source */
1036 #define PERF_MEM_NA (PERF_MEM_S(OP, NA)   |\
1037 		    PERF_MEM_S(LVL, NA)   |\
1038 		    PERF_MEM_S(SNOOP, NA) |\
1039 		    PERF_MEM_S(LOCK, NA)  |\
1040 		    PERF_MEM_S(TLB, NA))
1041 
perf_sample_data_init(struct perf_sample_data * data,u64 addr,u64 period)1042 static inline void perf_sample_data_init(struct perf_sample_data *data,
1043 					 u64 addr, u64 period)
1044 {
1045 	/* remaining struct members initialized in perf_prepare_sample() */
1046 	data->addr = addr;
1047 	data->raw  = NULL;
1048 	data->br_stack = NULL;
1049 	data->period = period;
1050 	data->weight.full = 0;
1051 	data->data_src.val = PERF_MEM_NA;
1052 	data->txn = 0;
1053 }
1054 
1055 extern void perf_output_sample(struct perf_output_handle *handle,
1056 			       struct perf_event_header *header,
1057 			       struct perf_sample_data *data,
1058 			       struct perf_event *event);
1059 extern void perf_prepare_sample(struct perf_event_header *header,
1060 				struct perf_sample_data *data,
1061 				struct perf_event *event,
1062 				struct pt_regs *regs);
1063 
1064 extern int perf_event_overflow(struct perf_event *event,
1065 				 struct perf_sample_data *data,
1066 				 struct pt_regs *regs);
1067 
1068 extern void perf_event_output_forward(struct perf_event *event,
1069 				     struct perf_sample_data *data,
1070 				     struct pt_regs *regs);
1071 extern void perf_event_output_backward(struct perf_event *event,
1072 				       struct perf_sample_data *data,
1073 				       struct pt_regs *regs);
1074 extern int perf_event_output(struct perf_event *event,
1075 			     struct perf_sample_data *data,
1076 			     struct pt_regs *regs);
1077 
1078 static inline bool
__is_default_overflow_handler(perf_overflow_handler_t overflow_handler)1079 __is_default_overflow_handler(perf_overflow_handler_t overflow_handler)
1080 {
1081 	if (likely(overflow_handler == perf_event_output_forward))
1082 		return true;
1083 	if (unlikely(overflow_handler == perf_event_output_backward))
1084 		return true;
1085 	return false;
1086 }
1087 
1088 #define is_default_overflow_handler(event) \
1089 	__is_default_overflow_handler((event)->overflow_handler)
1090 
1091 #ifdef CONFIG_BPF_SYSCALL
uses_default_overflow_handler(struct perf_event * event)1092 static inline bool uses_default_overflow_handler(struct perf_event *event)
1093 {
1094 	if (likely(is_default_overflow_handler(event)))
1095 		return true;
1096 
1097 	return __is_default_overflow_handler(event->orig_overflow_handler);
1098 }
1099 #else
1100 #define uses_default_overflow_handler(event) \
1101 	is_default_overflow_handler(event)
1102 #endif
1103 
1104 extern void
1105 perf_event_header__init_id(struct perf_event_header *header,
1106 			   struct perf_sample_data *data,
1107 			   struct perf_event *event);
1108 extern void
1109 perf_event__output_id_sample(struct perf_event *event,
1110 			     struct perf_output_handle *handle,
1111 			     struct perf_sample_data *sample);
1112 
1113 extern void
1114 perf_log_lost_samples(struct perf_event *event, u64 lost);
1115 
event_has_any_exclude_flag(struct perf_event * event)1116 static inline bool event_has_any_exclude_flag(struct perf_event *event)
1117 {
1118 	struct perf_event_attr *attr = &event->attr;
1119 
1120 	return attr->exclude_idle || attr->exclude_user ||
1121 	       attr->exclude_kernel || attr->exclude_hv ||
1122 	       attr->exclude_guest || attr->exclude_host;
1123 }
1124 
is_sampling_event(struct perf_event * event)1125 static inline bool is_sampling_event(struct perf_event *event)
1126 {
1127 	return event->attr.sample_period != 0;
1128 }
1129 
1130 /*
1131  * Return 1 for a software event, 0 for a hardware event
1132  */
is_software_event(struct perf_event * event)1133 static inline int is_software_event(struct perf_event *event)
1134 {
1135 	return event->event_caps & PERF_EV_CAP_SOFTWARE;
1136 }
1137 
1138 /*
1139  * Return 1 for event in sw context, 0 for event in hw context
1140  */
in_software_context(struct perf_event * event)1141 static inline int in_software_context(struct perf_event *event)
1142 {
1143 	return event->ctx->pmu->task_ctx_nr == perf_sw_context;
1144 }
1145 
is_exclusive_pmu(struct pmu * pmu)1146 static inline int is_exclusive_pmu(struct pmu *pmu)
1147 {
1148 	return pmu->capabilities & PERF_PMU_CAP_EXCLUSIVE;
1149 }
1150 
1151 extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX];
1152 
1153 extern void ___perf_sw_event(u32, u64, struct pt_regs *, u64);
1154 extern void __perf_sw_event(u32, u64, struct pt_regs *, u64);
1155 
1156 #ifndef perf_arch_fetch_caller_regs
perf_arch_fetch_caller_regs(struct pt_regs * regs,unsigned long ip)1157 static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { }
1158 #endif
1159 
1160 /*
1161  * When generating a perf sample in-line, instead of from an interrupt /
1162  * exception, we lack a pt_regs. This is typically used from software events
1163  * like: SW_CONTEXT_SWITCHES, SW_MIGRATIONS and the tie-in with tracepoints.
1164  *
1165  * We typically don't need a full set, but (for x86) do require:
1166  * - ip for PERF_SAMPLE_IP
1167  * - cs for user_mode() tests
1168  * - sp for PERF_SAMPLE_CALLCHAIN
1169  * - eflags for MISC bits and CALLCHAIN (see: perf_hw_regs())
1170  *
1171  * NOTE: assumes @regs is otherwise already 0 filled; this is important for
1172  * things like PERF_SAMPLE_REGS_INTR.
1173  */
perf_fetch_caller_regs(struct pt_regs * regs)1174 static inline void perf_fetch_caller_regs(struct pt_regs *regs)
1175 {
1176 	perf_arch_fetch_caller_regs(regs, CALLER_ADDR0);
1177 }
1178 
1179 static __always_inline void
perf_sw_event(u32 event_id,u64 nr,struct pt_regs * regs,u64 addr)1180 perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
1181 {
1182 	if (static_key_false(&perf_swevent_enabled[event_id]))
1183 		__perf_sw_event(event_id, nr, regs, addr);
1184 }
1185 
1186 DECLARE_PER_CPU(struct pt_regs, __perf_regs[4]);
1187 
1188 /*
1189  * 'Special' version for the scheduler, it hard assumes no recursion,
1190  * which is guaranteed by us not actually scheduling inside other swevents
1191  * because those disable preemption.
1192  */
__perf_sw_event_sched(u32 event_id,u64 nr,u64 addr)1193 static __always_inline void __perf_sw_event_sched(u32 event_id, u64 nr, u64 addr)
1194 {
1195 	struct pt_regs *regs = this_cpu_ptr(&__perf_regs[0]);
1196 
1197 	perf_fetch_caller_regs(regs);
1198 	___perf_sw_event(event_id, nr, regs, addr);
1199 }
1200 
1201 extern struct static_key_false perf_sched_events;
1202 
__perf_sw_enabled(int swevt)1203 static __always_inline bool __perf_sw_enabled(int swevt)
1204 {
1205 	return static_key_false(&perf_swevent_enabled[swevt]);
1206 }
1207 
perf_event_task_migrate(struct task_struct * task)1208 static inline void perf_event_task_migrate(struct task_struct *task)
1209 {
1210 	if (__perf_sw_enabled(PERF_COUNT_SW_CPU_MIGRATIONS))
1211 		task->sched_migrated = 1;
1212 }
1213 
perf_event_task_sched_in(struct task_struct * prev,struct task_struct * task)1214 static inline void perf_event_task_sched_in(struct task_struct *prev,
1215 					    struct task_struct *task)
1216 {
1217 	if (static_branch_unlikely(&perf_sched_events))
1218 		__perf_event_task_sched_in(prev, task);
1219 
1220 	if (__perf_sw_enabled(PERF_COUNT_SW_CPU_MIGRATIONS) &&
1221 	    task->sched_migrated) {
1222 		__perf_sw_event_sched(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 0);
1223 		task->sched_migrated = 0;
1224 	}
1225 }
1226 
perf_event_task_sched_out(struct task_struct * prev,struct task_struct * next)1227 static inline void perf_event_task_sched_out(struct task_struct *prev,
1228 					     struct task_struct *next)
1229 {
1230 	if (__perf_sw_enabled(PERF_COUNT_SW_CONTEXT_SWITCHES))
1231 		__perf_sw_event_sched(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 0);
1232 
1233 #ifdef CONFIG_CGROUP_PERF
1234 	if (__perf_sw_enabled(PERF_COUNT_SW_CGROUP_SWITCHES) &&
1235 	    perf_cgroup_from_task(prev, NULL) !=
1236 	    perf_cgroup_from_task(next, NULL))
1237 		__perf_sw_event_sched(PERF_COUNT_SW_CGROUP_SWITCHES, 1, 0);
1238 #endif
1239 
1240 	if (static_branch_unlikely(&perf_sched_events))
1241 		__perf_event_task_sched_out(prev, next);
1242 }
1243 
1244 extern void perf_event_mmap(struct vm_area_struct *vma);
1245 
1246 extern void perf_event_ksymbol(u16 ksym_type, u64 addr, u32 len,
1247 			       bool unregister, const char *sym);
1248 extern void perf_event_bpf_event(struct bpf_prog *prog,
1249 				 enum perf_bpf_event_type type,
1250 				 u16 flags);
1251 
1252 extern struct perf_guest_info_callbacks __rcu *perf_guest_cbs;
perf_get_guest_cbs(void)1253 static inline struct perf_guest_info_callbacks *perf_get_guest_cbs(void)
1254 {
1255 	/*
1256 	 * Callbacks are RCU-protected and must be READ_ONCE to avoid reloading
1257 	 * the callbacks between a !NULL check and dereferences, to ensure
1258 	 * pending stores/changes to the callback pointers are visible before a
1259 	 * non-NULL perf_guest_cbs is visible to readers, and to prevent a
1260 	 * module from unloading callbacks while readers are active.
1261 	 */
1262 	return rcu_dereference(perf_guest_cbs);
1263 }
1264 extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks);
1265 extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks);
1266 
1267 extern void perf_event_exec(void);
1268 extern void perf_event_comm(struct task_struct *tsk, bool exec);
1269 extern void perf_event_namespaces(struct task_struct *tsk);
1270 extern void perf_event_fork(struct task_struct *tsk);
1271 extern void perf_event_text_poke(const void *addr,
1272 				 const void *old_bytes, size_t old_len,
1273 				 const void *new_bytes, size_t new_len);
1274 
1275 /* Callchains */
1276 DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry);
1277 
1278 extern void perf_callchain_user(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs);
1279 extern void perf_callchain_kernel(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs);
1280 extern struct perf_callchain_entry *
1281 get_perf_callchain(struct pt_regs *regs, u32 init_nr, bool kernel, bool user,
1282 		   u32 max_stack, bool crosstask, bool add_mark);
1283 extern struct perf_callchain_entry *perf_callchain(struct perf_event *event, struct pt_regs *regs);
1284 extern int get_callchain_buffers(int max_stack);
1285 extern void put_callchain_buffers(void);
1286 extern struct perf_callchain_entry *get_callchain_entry(int *rctx);
1287 extern void put_callchain_entry(int rctx);
1288 
1289 extern int sysctl_perf_event_max_stack;
1290 extern int sysctl_perf_event_max_contexts_per_stack;
1291 
perf_callchain_store_context(struct perf_callchain_entry_ctx * ctx,u64 ip)1292 static inline int perf_callchain_store_context(struct perf_callchain_entry_ctx *ctx, u64 ip)
1293 {
1294 	if (ctx->contexts < sysctl_perf_event_max_contexts_per_stack) {
1295 		struct perf_callchain_entry *entry = ctx->entry;
1296 		entry->ip[entry->nr++] = ip;
1297 		++ctx->contexts;
1298 		return 0;
1299 	} else {
1300 		ctx->contexts_maxed = true;
1301 		return -1; /* no more room, stop walking the stack */
1302 	}
1303 }
1304 
perf_callchain_store(struct perf_callchain_entry_ctx * ctx,u64 ip)1305 static inline int perf_callchain_store(struct perf_callchain_entry_ctx *ctx, u64 ip)
1306 {
1307 	if (ctx->nr < ctx->max_stack && !ctx->contexts_maxed) {
1308 		struct perf_callchain_entry *entry = ctx->entry;
1309 		entry->ip[entry->nr++] = ip;
1310 		++ctx->nr;
1311 		return 0;
1312 	} else {
1313 		return -1; /* no more room, stop walking the stack */
1314 	}
1315 }
1316 
1317 extern int sysctl_perf_event_paranoid;
1318 extern int sysctl_perf_event_mlock;
1319 extern int sysctl_perf_event_sample_rate;
1320 extern int sysctl_perf_cpu_time_max_percent;
1321 
1322 extern void perf_sample_event_took(u64 sample_len_ns);
1323 
1324 int perf_proc_update_handler(struct ctl_table *table, int write,
1325 		void *buffer, size_t *lenp, loff_t *ppos);
1326 int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write,
1327 		void *buffer, size_t *lenp, loff_t *ppos);
1328 int perf_event_max_stack_handler(struct ctl_table *table, int write,
1329 		void *buffer, size_t *lenp, loff_t *ppos);
1330 
1331 /* Access to perf_event_open(2) syscall. */
1332 #define PERF_SECURITY_OPEN		0
1333 
1334 /* Finer grained perf_event_open(2) access control. */
1335 #define PERF_SECURITY_CPU		1
1336 #define PERF_SECURITY_KERNEL		2
1337 #define PERF_SECURITY_TRACEPOINT	3
1338 
perf_is_paranoid(void)1339 static inline int perf_is_paranoid(void)
1340 {
1341 	return sysctl_perf_event_paranoid > -1;
1342 }
1343 
perf_allow_kernel(struct perf_event_attr * attr)1344 static inline int perf_allow_kernel(struct perf_event_attr *attr)
1345 {
1346 	if (sysctl_perf_event_paranoid > 1 && !perfmon_capable())
1347 		return -EACCES;
1348 
1349 	return security_perf_event_open(attr, PERF_SECURITY_KERNEL);
1350 }
1351 
perf_allow_cpu(struct perf_event_attr * attr)1352 static inline int perf_allow_cpu(struct perf_event_attr *attr)
1353 {
1354 	if (sysctl_perf_event_paranoid > 0 && !perfmon_capable())
1355 		return -EACCES;
1356 
1357 	return security_perf_event_open(attr, PERF_SECURITY_CPU);
1358 }
1359 
perf_allow_tracepoint(struct perf_event_attr * attr)1360 static inline int perf_allow_tracepoint(struct perf_event_attr *attr)
1361 {
1362 	if (sysctl_perf_event_paranoid > -1 && !perfmon_capable())
1363 		return -EPERM;
1364 
1365 	return security_perf_event_open(attr, PERF_SECURITY_TRACEPOINT);
1366 }
1367 
1368 extern void perf_event_init(void);
1369 extern void perf_tp_event(u16 event_type, u64 count, void *record,
1370 			  int entry_size, struct pt_regs *regs,
1371 			  struct hlist_head *head, int rctx,
1372 			  struct task_struct *task);
1373 extern void perf_bp_event(struct perf_event *event, void *data);
1374 
1375 #ifndef perf_misc_flags
1376 # define perf_misc_flags(regs) \
1377 		(user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL)
1378 # define perf_instruction_pointer(regs)	instruction_pointer(regs)
1379 #endif
1380 #ifndef perf_arch_bpf_user_pt_regs
1381 # define perf_arch_bpf_user_pt_regs(regs) regs
1382 #endif
1383 
has_branch_stack(struct perf_event * event)1384 static inline bool has_branch_stack(struct perf_event *event)
1385 {
1386 	return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK;
1387 }
1388 
needs_branch_stack(struct perf_event * event)1389 static inline bool needs_branch_stack(struct perf_event *event)
1390 {
1391 	return event->attr.branch_sample_type != 0;
1392 }
1393 
has_aux(struct perf_event * event)1394 static inline bool has_aux(struct perf_event *event)
1395 {
1396 	return event->pmu->setup_aux;
1397 }
1398 
is_write_backward(struct perf_event * event)1399 static inline bool is_write_backward(struct perf_event *event)
1400 {
1401 	return !!event->attr.write_backward;
1402 }
1403 
has_addr_filter(struct perf_event * event)1404 static inline bool has_addr_filter(struct perf_event *event)
1405 {
1406 	return event->pmu->nr_addr_filters;
1407 }
1408 
1409 /*
1410  * An inherited event uses parent's filters
1411  */
1412 static inline struct perf_addr_filters_head *
perf_event_addr_filters(struct perf_event * event)1413 perf_event_addr_filters(struct perf_event *event)
1414 {
1415 	struct perf_addr_filters_head *ifh = &event->addr_filters;
1416 
1417 	if (event->parent)
1418 		ifh = &event->parent->addr_filters;
1419 
1420 	return ifh;
1421 }
1422 
1423 extern void perf_event_addr_filters_sync(struct perf_event *event);
1424 
1425 extern int perf_output_begin(struct perf_output_handle *handle,
1426 			     struct perf_sample_data *data,
1427 			     struct perf_event *event, unsigned int size);
1428 extern int perf_output_begin_forward(struct perf_output_handle *handle,
1429 				     struct perf_sample_data *data,
1430 				     struct perf_event *event,
1431 				     unsigned int size);
1432 extern int perf_output_begin_backward(struct perf_output_handle *handle,
1433 				      struct perf_sample_data *data,
1434 				      struct perf_event *event,
1435 				      unsigned int size);
1436 
1437 extern void perf_output_end(struct perf_output_handle *handle);
1438 extern unsigned int perf_output_copy(struct perf_output_handle *handle,
1439 			     const void *buf, unsigned int len);
1440 extern unsigned int perf_output_skip(struct perf_output_handle *handle,
1441 				     unsigned int len);
1442 extern long perf_output_copy_aux(struct perf_output_handle *aux_handle,
1443 				 struct perf_output_handle *handle,
1444 				 unsigned long from, unsigned long to);
1445 extern int perf_swevent_get_recursion_context(void);
1446 extern void perf_swevent_put_recursion_context(int rctx);
1447 extern u64 perf_swevent_set_period(struct perf_event *event);
1448 extern void perf_event_enable(struct perf_event *event);
1449 extern void perf_event_disable(struct perf_event *event);
1450 extern void perf_event_disable_local(struct perf_event *event);
1451 extern void perf_event_disable_inatomic(struct perf_event *event);
1452 extern void perf_event_task_tick(void);
1453 extern int perf_event_account_interrupt(struct perf_event *event);
1454 extern int perf_event_period(struct perf_event *event, u64 value);
1455 extern u64 perf_event_pause(struct perf_event *event, bool reset);
1456 #else /* !CONFIG_PERF_EVENTS: */
1457 static inline void *
perf_aux_output_begin(struct perf_output_handle * handle,struct perf_event * event)1458 perf_aux_output_begin(struct perf_output_handle *handle,
1459 		      struct perf_event *event)				{ return NULL; }
1460 static inline void
perf_aux_output_end(struct perf_output_handle * handle,unsigned long size)1461 perf_aux_output_end(struct perf_output_handle *handle, unsigned long size)
1462 									{ }
1463 static inline int
perf_aux_output_skip(struct perf_output_handle * handle,unsigned long size)1464 perf_aux_output_skip(struct perf_output_handle *handle,
1465 		     unsigned long size)				{ return -EINVAL; }
1466 static inline void *
perf_get_aux(struct perf_output_handle * handle)1467 perf_get_aux(struct perf_output_handle *handle)				{ return NULL; }
1468 static inline void
perf_event_task_migrate(struct task_struct * task)1469 perf_event_task_migrate(struct task_struct *task)			{ }
1470 static inline void
perf_event_task_sched_in(struct task_struct * prev,struct task_struct * task)1471 perf_event_task_sched_in(struct task_struct *prev,
1472 			 struct task_struct *task)			{ }
1473 static inline void
perf_event_task_sched_out(struct task_struct * prev,struct task_struct * next)1474 perf_event_task_sched_out(struct task_struct *prev,
1475 			  struct task_struct *next)			{ }
perf_event_init_task(struct task_struct * child,u64 clone_flags)1476 static inline int perf_event_init_task(struct task_struct *child,
1477 				       u64 clone_flags)			{ return 0; }
perf_event_exit_task(struct task_struct * child)1478 static inline void perf_event_exit_task(struct task_struct *child)	{ }
perf_event_free_task(struct task_struct * task)1479 static inline void perf_event_free_task(struct task_struct *task)	{ }
perf_event_delayed_put(struct task_struct * task)1480 static inline void perf_event_delayed_put(struct task_struct *task)	{ }
perf_event_get(unsigned int fd)1481 static inline struct file *perf_event_get(unsigned int fd)	{ return ERR_PTR(-EINVAL); }
perf_get_event(struct file * file)1482 static inline const struct perf_event *perf_get_event(struct file *file)
1483 {
1484 	return ERR_PTR(-EINVAL);
1485 }
perf_event_attrs(struct perf_event * event)1486 static inline const struct perf_event_attr *perf_event_attrs(struct perf_event *event)
1487 {
1488 	return ERR_PTR(-EINVAL);
1489 }
perf_event_read_local(struct perf_event * event,u64 * value,u64 * enabled,u64 * running)1490 static inline int perf_event_read_local(struct perf_event *event, u64 *value,
1491 					u64 *enabled, u64 *running)
1492 {
1493 	return -EINVAL;
1494 }
perf_event_print_debug(void)1495 static inline void perf_event_print_debug(void)				{ }
perf_event_task_disable(void)1496 static inline int perf_event_task_disable(void)				{ return -EINVAL; }
perf_event_task_enable(void)1497 static inline int perf_event_task_enable(void)				{ return -EINVAL; }
perf_event_refresh(struct perf_event * event,int refresh)1498 static inline int perf_event_refresh(struct perf_event *event, int refresh)
1499 {
1500 	return -EINVAL;
1501 }
1502 
1503 static inline void
perf_sw_event(u32 event_id,u64 nr,struct pt_regs * regs,u64 addr)1504 perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)	{ }
1505 static inline void
perf_bp_event(struct perf_event * event,void * data)1506 perf_bp_event(struct perf_event *event, void *data)			{ }
1507 
perf_register_guest_info_callbacks(struct perf_guest_info_callbacks * callbacks)1508 static inline int perf_register_guest_info_callbacks
1509 (struct perf_guest_info_callbacks *callbacks)				{ return 0; }
perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks * callbacks)1510 static inline int perf_unregister_guest_info_callbacks
1511 (struct perf_guest_info_callbacks *callbacks)				{ return 0; }
1512 
perf_event_mmap(struct vm_area_struct * vma)1513 static inline void perf_event_mmap(struct vm_area_struct *vma)		{ }
1514 
1515 typedef int (perf_ksymbol_get_name_f)(char *name, int name_len, void *data);
perf_event_ksymbol(u16 ksym_type,u64 addr,u32 len,bool unregister,const char * sym)1516 static inline void perf_event_ksymbol(u16 ksym_type, u64 addr, u32 len,
1517 				      bool unregister, const char *sym)	{ }
perf_event_bpf_event(struct bpf_prog * prog,enum perf_bpf_event_type type,u16 flags)1518 static inline void perf_event_bpf_event(struct bpf_prog *prog,
1519 					enum perf_bpf_event_type type,
1520 					u16 flags)			{ }
perf_event_exec(void)1521 static inline void perf_event_exec(void)				{ }
perf_event_comm(struct task_struct * tsk,bool exec)1522 static inline void perf_event_comm(struct task_struct *tsk, bool exec)	{ }
perf_event_namespaces(struct task_struct * tsk)1523 static inline void perf_event_namespaces(struct task_struct *tsk)	{ }
perf_event_fork(struct task_struct * tsk)1524 static inline void perf_event_fork(struct task_struct *tsk)		{ }
perf_event_text_poke(const void * addr,const void * old_bytes,size_t old_len,const void * new_bytes,size_t new_len)1525 static inline void perf_event_text_poke(const void *addr,
1526 					const void *old_bytes,
1527 					size_t old_len,
1528 					const void *new_bytes,
1529 					size_t new_len)			{ }
perf_event_init(void)1530 static inline void perf_event_init(void)				{ }
perf_swevent_get_recursion_context(void)1531 static inline int  perf_swevent_get_recursion_context(void)		{ return -1; }
perf_swevent_put_recursion_context(int rctx)1532 static inline void perf_swevent_put_recursion_context(int rctx)		{ }
perf_swevent_set_period(struct perf_event * event)1533 static inline u64 perf_swevent_set_period(struct perf_event *event)	{ return 0; }
perf_event_enable(struct perf_event * event)1534 static inline void perf_event_enable(struct perf_event *event)		{ }
perf_event_disable(struct perf_event * event)1535 static inline void perf_event_disable(struct perf_event *event)		{ }
__perf_event_disable(void * info)1536 static inline int __perf_event_disable(void *info)			{ return -1; }
perf_event_task_tick(void)1537 static inline void perf_event_task_tick(void)				{ }
perf_event_release_kernel(struct perf_event * event)1538 static inline int perf_event_release_kernel(struct perf_event *event)	{ return 0; }
perf_event_period(struct perf_event * event,u64 value)1539 static inline int perf_event_period(struct perf_event *event, u64 value)
1540 {
1541 	return -EINVAL;
1542 }
perf_event_pause(struct perf_event * event,bool reset)1543 static inline u64 perf_event_pause(struct perf_event *event, bool reset)
1544 {
1545 	return 0;
1546 }
1547 #endif
1548 
1549 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
1550 extern void perf_restore_debug_store(void);
1551 #else
perf_restore_debug_store(void)1552 static inline void perf_restore_debug_store(void)			{ }
1553 #endif
1554 
perf_raw_frag_last(const struct perf_raw_frag * frag)1555 static __always_inline bool perf_raw_frag_last(const struct perf_raw_frag *frag)
1556 {
1557 	return frag->pad < sizeof(u64);
1558 }
1559 
1560 #define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x))
1561 
1562 struct perf_pmu_events_attr {
1563 	struct device_attribute attr;
1564 	u64 id;
1565 	const char *event_str;
1566 };
1567 
1568 struct perf_pmu_events_ht_attr {
1569 	struct device_attribute			attr;
1570 	u64					id;
1571 	const char				*event_str_ht;
1572 	const char				*event_str_noht;
1573 };
1574 
1575 struct perf_pmu_events_hybrid_attr {
1576 	struct device_attribute			attr;
1577 	u64					id;
1578 	const char				*event_str;
1579 	u64					pmu_type;
1580 };
1581 
1582 struct perf_pmu_format_hybrid_attr {
1583 	struct device_attribute			attr;
1584 	u64					pmu_type;
1585 };
1586 
1587 ssize_t perf_event_sysfs_show(struct device *dev, struct device_attribute *attr,
1588 			      char *page);
1589 
1590 #define PMU_EVENT_ATTR(_name, _var, _id, _show)				\
1591 static struct perf_pmu_events_attr _var = {				\
1592 	.attr = __ATTR(_name, 0444, _show, NULL),			\
1593 	.id   =  _id,							\
1594 };
1595 
1596 #define PMU_EVENT_ATTR_STRING(_name, _var, _str)			    \
1597 static struct perf_pmu_events_attr _var = {				    \
1598 	.attr		= __ATTR(_name, 0444, perf_event_sysfs_show, NULL), \
1599 	.id		= 0,						    \
1600 	.event_str	= _str,						    \
1601 };
1602 
1603 #define PMU_EVENT_ATTR_ID(_name, _show, _id)				\
1604 	(&((struct perf_pmu_events_attr[]) {				\
1605 		{ .attr = __ATTR(_name, 0444, _show, NULL),		\
1606 		  .id = _id, }						\
1607 	})[0].attr.attr)
1608 
1609 #define PMU_FORMAT_ATTR(_name, _format)					\
1610 static ssize_t								\
1611 _name##_show(struct device *dev,					\
1612 			       struct device_attribute *attr,		\
1613 			       char *page)				\
1614 {									\
1615 	BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE);			\
1616 	return sprintf(page, _format "\n");				\
1617 }									\
1618 									\
1619 static struct device_attribute format_attr_##_name = __ATTR_RO(_name)
1620 
1621 /* Performance counter hotplug functions */
1622 #ifdef CONFIG_PERF_EVENTS
1623 int perf_event_init_cpu(unsigned int cpu);
1624 int perf_event_exit_cpu(unsigned int cpu);
1625 #else
1626 #define perf_event_init_cpu	NULL
1627 #define perf_event_exit_cpu	NULL
1628 #endif
1629 
1630 extern void __weak arch_perf_update_userpage(struct perf_event *event,
1631 					     struct perf_event_mmap_page *userpg,
1632 					     u64 now);
1633 
1634 #ifdef CONFIG_MMU
1635 extern __weak u64 arch_perf_get_page_size(struct mm_struct *mm, unsigned long addr);
1636 #endif
1637 
1638 #endif /* _LINUX_PERF_EVENT_H */
1639