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 #define PERF_GUEST_ACTIVE	0x01
30 #define PERF_GUEST_USER	0x02
31 
32 struct perf_guest_info_callbacks {
33 	unsigned int			(*state)(void);
34 	unsigned long			(*get_ip)(void);
35 	unsigned int			(*handle_intel_pt_intr)(void);
36 };
37 
38 #ifdef CONFIG_HAVE_HW_BREAKPOINT
39 #include <linux/rhashtable-types.h>
40 #include <asm/hw_breakpoint.h>
41 #endif
42 
43 #include <linux/list.h>
44 #include <linux/mutex.h>
45 #include <linux/rculist.h>
46 #include <linux/rcupdate.h>
47 #include <linux/spinlock.h>
48 #include <linux/hrtimer.h>
49 #include <linux/fs.h>
50 #include <linux/pid_namespace.h>
51 #include <linux/workqueue.h>
52 #include <linux/ftrace.h>
53 #include <linux/cpu.h>
54 #include <linux/irq_work.h>
55 #include <linux/static_key.h>
56 #include <linux/jump_label_ratelimit.h>
57 #include <linux/atomic.h>
58 #include <linux/sysfs.h>
59 #include <linux/perf_regs.h>
60 #include <linux/cgroup.h>
61 #include <linux/refcount.h>
62 #include <linux/security.h>
63 #include <linux/static_call.h>
64 #include <linux/lockdep.h>
65 #include <asm/local.h>
66 
67 struct perf_callchain_entry {
68 	__u64				nr;
69 	__u64				ip[]; /* /proc/sys/kernel/perf_event_max_stack */
70 };
71 
72 struct perf_callchain_entry_ctx {
73 	struct perf_callchain_entry *entry;
74 	u32			    max_stack;
75 	u32			    nr;
76 	short			    contexts;
77 	bool			    contexts_maxed;
78 };
79 
80 typedef unsigned long (*perf_copy_f)(void *dst, const void *src,
81 				     unsigned long off, unsigned long len);
82 
83 struct perf_raw_frag {
84 	union {
85 		struct perf_raw_frag	*next;
86 		unsigned long		pad;
87 	};
88 	perf_copy_f			copy;
89 	void				*data;
90 	u32				size;
91 } __packed;
92 
93 struct perf_raw_record {
94 	struct perf_raw_frag		frag;
95 	u32				size;
96 };
97 
98 /*
99  * branch stack layout:
100  *  nr: number of taken branches stored in entries[]
101  *  hw_idx: The low level index of raw branch records
102  *          for the most recent branch.
103  *          -1ULL means invalid/unknown.
104  *
105  * Note that nr can vary from sample to sample
106  * branches (to, from) are stored from most recent
107  * to least recent, i.e., entries[0] contains the most
108  * recent branch.
109  * The entries[] is an abstraction of raw branch records,
110  * which may not be stored in age order in HW, e.g. Intel LBR.
111  * The hw_idx is to expose the low level index of raw
112  * branch record for the most recent branch aka entries[0].
113  * The hw_idx index is between -1 (unknown) and max depth,
114  * which can be retrieved in /sys/devices/cpu/caps/branches.
115  * For the architectures whose raw branch records are
116  * already stored in age order, the hw_idx should be 0.
117  */
118 struct perf_branch_stack {
119 	__u64				nr;
120 	__u64				hw_idx;
121 	struct perf_branch_entry	entries[];
122 };
123 
124 struct task_struct;
125 
126 /*
127  * extra PMU register associated with an event
128  */
129 struct hw_perf_event_extra {
130 	u64		config;	/* register value */
131 	unsigned int	reg;	/* register address or index */
132 	int		alloc;	/* extra register already allocated */
133 	int		idx;	/* index in shared_regs->regs[] */
134 };
135 
136 /**
137  * hw_perf_event::flag values
138  *
139  * PERF_EVENT_FLAG_ARCH bits are reserved for architecture-specific
140  * usage.
141  */
142 #define PERF_EVENT_FLAG_ARCH			0x000fffff
143 #define PERF_EVENT_FLAG_USER_READ_CNT		0x80000000
144 
145 static_assert((PERF_EVENT_FLAG_USER_READ_CNT & PERF_EVENT_FLAG_ARCH) == 0);
146 
147 /**
148  * struct hw_perf_event - performance event hardware details:
149  */
150 struct hw_perf_event {
151 #ifdef CONFIG_PERF_EVENTS
152 	union {
153 		struct { /* hardware */
154 			u64		config;
155 			u64		last_tag;
156 			unsigned long	config_base;
157 			unsigned long	event_base;
158 			int		event_base_rdpmc;
159 			int		idx;
160 			int		last_cpu;
161 			int		flags;
162 
163 			struct hw_perf_event_extra extra_reg;
164 			struct hw_perf_event_extra branch_reg;
165 		};
166 		struct { /* software */
167 			struct hrtimer	hrtimer;
168 		};
169 		struct { /* tracepoint */
170 			/* for tp_event->class */
171 			struct list_head	tp_list;
172 		};
173 		struct { /* amd_power */
174 			u64	pwr_acc;
175 			u64	ptsc;
176 		};
177 #ifdef CONFIG_HAVE_HW_BREAKPOINT
178 		struct { /* breakpoint */
179 			/*
180 			 * Crufty hack to avoid the chicken and egg
181 			 * problem hw_breakpoint has with context
182 			 * creation and event initalization.
183 			 */
184 			struct arch_hw_breakpoint	info;
185 			struct rhlist_head		bp_list;
186 		};
187 #endif
188 		struct { /* amd_iommu */
189 			u8	iommu_bank;
190 			u8	iommu_cntr;
191 			u16	padding;
192 			u64	conf;
193 			u64	conf1;
194 		};
195 	};
196 	/*
197 	 * If the event is a per task event, this will point to the task in
198 	 * question. See the comment in perf_event_alloc().
199 	 */
200 	struct task_struct		*target;
201 
202 	/*
203 	 * PMU would store hardware filter configuration
204 	 * here.
205 	 */
206 	void				*addr_filters;
207 
208 	/* Last sync'ed generation of filters */
209 	unsigned long			addr_filters_gen;
210 
211 /*
212  * hw_perf_event::state flags; used to track the PERF_EF_* state.
213  */
214 #define PERF_HES_STOPPED	0x01 /* the counter is stopped */
215 #define PERF_HES_UPTODATE	0x02 /* event->count up-to-date */
216 #define PERF_HES_ARCH		0x04
217 
218 	int				state;
219 
220 	/*
221 	 * The last observed hardware counter value, updated with a
222 	 * local64_cmpxchg() such that pmu::read() can be called nested.
223 	 */
224 	local64_t			prev_count;
225 
226 	/*
227 	 * The period to start the next sample with.
228 	 */
229 	u64				sample_period;
230 
231 	union {
232 		struct { /* Sampling */
233 			/*
234 			 * The period we started this sample with.
235 			 */
236 			u64				last_period;
237 
238 			/*
239 			 * However much is left of the current period;
240 			 * note that this is a full 64bit value and
241 			 * allows for generation of periods longer
242 			 * than hardware might allow.
243 			 */
244 			local64_t			period_left;
245 		};
246 		struct { /* Topdown events counting for context switch */
247 			u64				saved_metric;
248 			u64				saved_slots;
249 		};
250 	};
251 
252 	/*
253 	 * State for throttling the event, see __perf_event_overflow() and
254 	 * perf_adjust_freq_unthr_context().
255 	 */
256 	u64                             interrupts_seq;
257 	u64				interrupts;
258 
259 	/*
260 	 * State for freq target events, see __perf_event_overflow() and
261 	 * perf_adjust_freq_unthr_context().
262 	 */
263 	u64				freq_time_stamp;
264 	u64				freq_count_stamp;
265 #endif
266 };
267 
268 struct perf_event;
269 
270 /*
271  * Common implementation detail of pmu::{start,commit,cancel}_txn
272  */
273 #define PERF_PMU_TXN_ADD  0x1		/* txn to add/schedule event on PMU */
274 #define PERF_PMU_TXN_READ 0x2		/* txn to read event group from PMU */
275 
276 /**
277  * pmu::capabilities flags
278  */
279 #define PERF_PMU_CAP_NO_INTERRUPT		0x0001
280 #define PERF_PMU_CAP_NO_NMI			0x0002
281 #define PERF_PMU_CAP_AUX_NO_SG			0x0004
282 #define PERF_PMU_CAP_EXTENDED_REGS		0x0008
283 #define PERF_PMU_CAP_EXCLUSIVE			0x0010
284 #define PERF_PMU_CAP_ITRACE			0x0020
285 #define PERF_PMU_CAP_HETEROGENEOUS_CPUS		0x0040
286 #define PERF_PMU_CAP_NO_EXCLUDE			0x0080
287 #define PERF_PMU_CAP_AUX_OUTPUT			0x0100
288 #define PERF_PMU_CAP_EXTENDED_HW_TYPE		0x0200
289 
290 struct perf_output_handle;
291 
292 /**
293  * struct pmu - generic performance monitoring unit
294  */
295 struct pmu {
296 	struct list_head		entry;
297 
298 	struct module			*module;
299 	struct device			*dev;
300 	const struct attribute_group	**attr_groups;
301 	const struct attribute_group	**attr_update;
302 	const char			*name;
303 	int				type;
304 
305 	/*
306 	 * various common per-pmu feature flags
307 	 */
308 	int				capabilities;
309 
310 	int __percpu			*pmu_disable_count;
311 	struct perf_cpu_context __percpu *pmu_cpu_context;
312 	atomic_t			exclusive_cnt; /* < 0: cpu; > 0: tsk */
313 	int				task_ctx_nr;
314 	int				hrtimer_interval_ms;
315 
316 	/* number of address filters this PMU can do */
317 	unsigned int			nr_addr_filters;
318 
319 	/*
320 	 * Fully disable/enable this PMU, can be used to protect from the PMI
321 	 * as well as for lazy/batch writing of the MSRs.
322 	 */
323 	void (*pmu_enable)		(struct pmu *pmu); /* optional */
324 	void (*pmu_disable)		(struct pmu *pmu); /* optional */
325 
326 	/*
327 	 * Try and initialize the event for this PMU.
328 	 *
329 	 * Returns:
330 	 *  -ENOENT	-- @event is not for this PMU
331 	 *
332 	 *  -ENODEV	-- @event is for this PMU but PMU not present
333 	 *  -EBUSY	-- @event is for this PMU but PMU temporarily unavailable
334 	 *  -EINVAL	-- @event is for this PMU but @event is not valid
335 	 *  -EOPNOTSUPP -- @event is for this PMU, @event is valid, but not supported
336 	 *  -EACCES	-- @event is for this PMU, @event is valid, but no privileges
337 	 *
338 	 *  0		-- @event is for this PMU and valid
339 	 *
340 	 * Other error return values are allowed.
341 	 */
342 	int (*event_init)		(struct perf_event *event);
343 
344 	/*
345 	 * Notification that the event was mapped or unmapped.  Called
346 	 * in the context of the mapping task.
347 	 */
348 	void (*event_mapped)		(struct perf_event *event, struct mm_struct *mm); /* optional */
349 	void (*event_unmapped)		(struct perf_event *event, struct mm_struct *mm); /* optional */
350 
351 	/*
352 	 * Flags for ->add()/->del()/ ->start()/->stop(). There are
353 	 * matching hw_perf_event::state flags.
354 	 */
355 #define PERF_EF_START	0x01		/* start the counter when adding    */
356 #define PERF_EF_RELOAD	0x02		/* reload the counter when starting */
357 #define PERF_EF_UPDATE	0x04		/* update the counter when stopping */
358 
359 	/*
360 	 * Adds/Removes a counter to/from the PMU, can be done inside a
361 	 * transaction, see the ->*_txn() methods.
362 	 *
363 	 * The add/del callbacks will reserve all hardware resources required
364 	 * to service the event, this includes any counter constraint
365 	 * scheduling etc.
366 	 *
367 	 * Called with IRQs disabled and the PMU disabled on the CPU the event
368 	 * is on.
369 	 *
370 	 * ->add() called without PERF_EF_START should result in the same state
371 	 *  as ->add() followed by ->stop().
372 	 *
373 	 * ->del() must always PERF_EF_UPDATE stop an event. If it calls
374 	 *  ->stop() that must deal with already being stopped without
375 	 *  PERF_EF_UPDATE.
376 	 */
377 	int  (*add)			(struct perf_event *event, int flags);
378 	void (*del)			(struct perf_event *event, int flags);
379 
380 	/*
381 	 * Starts/Stops a counter present on the PMU.
382 	 *
383 	 * The PMI handler should stop the counter when perf_event_overflow()
384 	 * returns !0. ->start() will be used to continue.
385 	 *
386 	 * Also used to change the sample period.
387 	 *
388 	 * Called with IRQs disabled and the PMU disabled on the CPU the event
389 	 * is on -- will be called from NMI context with the PMU generates
390 	 * NMIs.
391 	 *
392 	 * ->stop() with PERF_EF_UPDATE will read the counter and update
393 	 *  period/count values like ->read() would.
394 	 *
395 	 * ->start() with PERF_EF_RELOAD will reprogram the counter
396 	 *  value, must be preceded by a ->stop() with PERF_EF_UPDATE.
397 	 */
398 	void (*start)			(struct perf_event *event, int flags);
399 	void (*stop)			(struct perf_event *event, int flags);
400 
401 	/*
402 	 * Updates the counter value of the event.
403 	 *
404 	 * For sampling capable PMUs this will also update the software period
405 	 * hw_perf_event::period_left field.
406 	 */
407 	void (*read)			(struct perf_event *event);
408 
409 	/*
410 	 * Group events scheduling is treated as a transaction, add
411 	 * group events as a whole and perform one schedulability test.
412 	 * If the test fails, roll back the whole group
413 	 *
414 	 * Start the transaction, after this ->add() doesn't need to
415 	 * do schedulability tests.
416 	 *
417 	 * Optional.
418 	 */
419 	void (*start_txn)		(struct pmu *pmu, unsigned int txn_flags);
420 	/*
421 	 * If ->start_txn() disabled the ->add() schedulability test
422 	 * then ->commit_txn() is required to perform one. On success
423 	 * the transaction is closed. On error the transaction is kept
424 	 * open until ->cancel_txn() is called.
425 	 *
426 	 * Optional.
427 	 */
428 	int  (*commit_txn)		(struct pmu *pmu);
429 	/*
430 	 * Will cancel the transaction, assumes ->del() is called
431 	 * for each successful ->add() during the transaction.
432 	 *
433 	 * Optional.
434 	 */
435 	void (*cancel_txn)		(struct pmu *pmu);
436 
437 	/*
438 	 * Will return the value for perf_event_mmap_page::index for this event,
439 	 * if no implementation is provided it will default to: event->hw.idx + 1.
440 	 */
441 	int (*event_idx)		(struct perf_event *event); /*optional */
442 
443 	/*
444 	 * context-switches callback
445 	 */
446 	void (*sched_task)		(struct perf_event_context *ctx,
447 					bool sched_in);
448 
449 	/*
450 	 * Kmem cache of PMU specific data
451 	 */
452 	struct kmem_cache		*task_ctx_cache;
453 
454 	/*
455 	 * PMU specific parts of task perf event context (i.e. ctx->task_ctx_data)
456 	 * can be synchronized using this function. See Intel LBR callstack support
457 	 * implementation and Perf core context switch handling callbacks for usage
458 	 * examples.
459 	 */
460 	void (*swap_task_ctx)		(struct perf_event_context *prev,
461 					 struct perf_event_context *next);
462 					/* optional */
463 
464 	/*
465 	 * Set up pmu-private data structures for an AUX area
466 	 */
467 	void *(*setup_aux)		(struct perf_event *event, void **pages,
468 					 int nr_pages, bool overwrite);
469 					/* optional */
470 
471 	/*
472 	 * Free pmu-private AUX data structures
473 	 */
474 	void (*free_aux)		(void *aux); /* optional */
475 
476 	/*
477 	 * Take a snapshot of the AUX buffer without touching the event
478 	 * state, so that preempting ->start()/->stop() callbacks does
479 	 * not interfere with their logic. Called in PMI context.
480 	 *
481 	 * Returns the size of AUX data copied to the output handle.
482 	 *
483 	 * Optional.
484 	 */
485 	long (*snapshot_aux)		(struct perf_event *event,
486 					 struct perf_output_handle *handle,
487 					 unsigned long size);
488 
489 	/*
490 	 * Validate address range filters: make sure the HW supports the
491 	 * requested configuration and number of filters; return 0 if the
492 	 * supplied filters are valid, -errno otherwise.
493 	 *
494 	 * Runs in the context of the ioctl()ing process and is not serialized
495 	 * with the rest of the PMU callbacks.
496 	 */
497 	int (*addr_filters_validate)	(struct list_head *filters);
498 					/* optional */
499 
500 	/*
501 	 * Synchronize address range filter configuration:
502 	 * translate hw-agnostic filters into hardware configuration in
503 	 * event::hw::addr_filters.
504 	 *
505 	 * Runs as a part of filter sync sequence that is done in ->start()
506 	 * callback by calling perf_event_addr_filters_sync().
507 	 *
508 	 * May (and should) traverse event::addr_filters::list, for which its
509 	 * caller provides necessary serialization.
510 	 */
511 	void (*addr_filters_sync)	(struct perf_event *event);
512 					/* optional */
513 
514 	/*
515 	 * Check if event can be used for aux_output purposes for
516 	 * events of this PMU.
517 	 *
518 	 * Runs from perf_event_open(). Should return 0 for "no match"
519 	 * or non-zero for "match".
520 	 */
521 	int (*aux_output_match)		(struct perf_event *event);
522 					/* optional */
523 
524 	/*
525 	 * Filter events for PMU-specific reasons.
526 	 */
527 	int (*filter_match)		(struct perf_event *event); /* optional */
528 
529 	/*
530 	 * Check period value for PERF_EVENT_IOC_PERIOD ioctl.
531 	 */
532 	int (*check_period)		(struct perf_event *event, u64 value); /* optional */
533 };
534 
535 enum perf_addr_filter_action_t {
536 	PERF_ADDR_FILTER_ACTION_STOP = 0,
537 	PERF_ADDR_FILTER_ACTION_START,
538 	PERF_ADDR_FILTER_ACTION_FILTER,
539 };
540 
541 /**
542  * struct perf_addr_filter - address range filter definition
543  * @entry:	event's filter list linkage
544  * @path:	object file's path for file-based filters
545  * @offset:	filter range offset
546  * @size:	filter range size (size==0 means single address trigger)
547  * @action:	filter/start/stop
548  *
549  * This is a hardware-agnostic filter configuration as specified by the user.
550  */
551 struct perf_addr_filter {
552 	struct list_head	entry;
553 	struct path		path;
554 	unsigned long		offset;
555 	unsigned long		size;
556 	enum perf_addr_filter_action_t	action;
557 };
558 
559 /**
560  * struct perf_addr_filters_head - container for address range filters
561  * @list:	list of filters for this event
562  * @lock:	spinlock that serializes accesses to the @list and event's
563  *		(and its children's) filter generations.
564  * @nr_file_filters:	number of file-based filters
565  *
566  * A child event will use parent's @list (and therefore @lock), so they are
567  * bundled together; see perf_event_addr_filters().
568  */
569 struct perf_addr_filters_head {
570 	struct list_head	list;
571 	raw_spinlock_t		lock;
572 	unsigned int		nr_file_filters;
573 };
574 
575 struct perf_addr_filter_range {
576 	unsigned long		start;
577 	unsigned long		size;
578 };
579 
580 /**
581  * enum perf_event_state - the states of an event:
582  */
583 enum perf_event_state {
584 	PERF_EVENT_STATE_DEAD		= -4,
585 	PERF_EVENT_STATE_EXIT		= -3,
586 	PERF_EVENT_STATE_ERROR		= -2,
587 	PERF_EVENT_STATE_OFF		= -1,
588 	PERF_EVENT_STATE_INACTIVE	=  0,
589 	PERF_EVENT_STATE_ACTIVE		=  1,
590 };
591 
592 struct file;
593 struct perf_sample_data;
594 
595 typedef void (*perf_overflow_handler_t)(struct perf_event *,
596 					struct perf_sample_data *,
597 					struct pt_regs *regs);
598 
599 /*
600  * Event capabilities. For event_caps and groups caps.
601  *
602  * PERF_EV_CAP_SOFTWARE: Is a software event.
603  * PERF_EV_CAP_READ_ACTIVE_PKG: A CPU event (or cgroup event) that can be read
604  * from any CPU in the package where it is active.
605  * PERF_EV_CAP_SIBLING: An event with this flag must be a group sibling and
606  * cannot be a group leader. If an event with this flag is detached from the
607  * group it is scheduled out and moved into an unrecoverable ERROR state.
608  */
609 #define PERF_EV_CAP_SOFTWARE		BIT(0)
610 #define PERF_EV_CAP_READ_ACTIVE_PKG	BIT(1)
611 #define PERF_EV_CAP_SIBLING		BIT(2)
612 
613 #define SWEVENT_HLIST_BITS		8
614 #define SWEVENT_HLIST_SIZE		(1 << SWEVENT_HLIST_BITS)
615 
616 struct swevent_hlist {
617 	struct hlist_head		heads[SWEVENT_HLIST_SIZE];
618 	struct rcu_head			rcu_head;
619 };
620 
621 #define PERF_ATTACH_CONTEXT	0x01
622 #define PERF_ATTACH_GROUP	0x02
623 #define PERF_ATTACH_TASK	0x04
624 #define PERF_ATTACH_TASK_DATA	0x08
625 #define PERF_ATTACH_ITRACE	0x10
626 #define PERF_ATTACH_SCHED_CB	0x20
627 #define PERF_ATTACH_CHILD	0x40
628 
629 struct bpf_prog;
630 struct perf_cgroup;
631 struct perf_buffer;
632 
633 struct pmu_event_list {
634 	raw_spinlock_t		lock;
635 	struct list_head	list;
636 };
637 
638 /*
639  * event->sibling_list is modified whole holding both ctx->lock and ctx->mutex
640  * as such iteration must hold either lock. However, since ctx->lock is an IRQ
641  * safe lock, and is only held by the CPU doing the modification, having IRQs
642  * disabled is sufficient since it will hold-off the IPIs.
643  */
644 #ifdef CONFIG_PROVE_LOCKING
645 #define lockdep_assert_event_ctx(event)				\
646 	WARN_ON_ONCE(__lockdep_enabled &&			\
647 		     (this_cpu_read(hardirqs_enabled) &&	\
648 		      lockdep_is_held(&(event)->ctx->mutex) != LOCK_STATE_HELD))
649 #else
650 #define lockdep_assert_event_ctx(event)
651 #endif
652 
653 #define for_each_sibling_event(sibling, event)			\
654 	lockdep_assert_event_ctx(event);			\
655 	if ((event)->group_leader == (event))			\
656 		list_for_each_entry((sibling), &(event)->sibling_list, sibling_list)
657 
658 /**
659  * struct perf_event - performance event kernel representation:
660  */
661 struct perf_event {
662 #ifdef CONFIG_PERF_EVENTS
663 	/*
664 	 * entry onto perf_event_context::event_list;
665 	 *   modifications require ctx->lock
666 	 *   RCU safe iterations.
667 	 */
668 	struct list_head		event_entry;
669 
670 	/*
671 	 * Locked for modification by both ctx->mutex and ctx->lock; holding
672 	 * either sufficies for read.
673 	 */
674 	struct list_head		sibling_list;
675 	struct list_head		active_list;
676 	/*
677 	 * Node on the pinned or flexible tree located at the event context;
678 	 */
679 	struct rb_node			group_node;
680 	u64				group_index;
681 	/*
682 	 * We need storage to track the entries in perf_pmu_migrate_context; we
683 	 * cannot use the event_entry because of RCU and we want to keep the
684 	 * group in tact which avoids us using the other two entries.
685 	 */
686 	struct list_head		migrate_entry;
687 
688 	struct hlist_node		hlist_entry;
689 	struct list_head		active_entry;
690 	int				nr_siblings;
691 
692 	/* Not serialized. Only written during event initialization. */
693 	int				event_caps;
694 	/* The cumulative AND of all event_caps for events in this group. */
695 	int				group_caps;
696 
697 	struct perf_event		*group_leader;
698 	struct pmu			*pmu;
699 	void				*pmu_private;
700 
701 	enum perf_event_state		state;
702 	unsigned int			attach_state;
703 	local64_t			count;
704 	atomic64_t			child_count;
705 
706 	/*
707 	 * These are the total time in nanoseconds that the event
708 	 * has been enabled (i.e. eligible to run, and the task has
709 	 * been scheduled in, if this is a per-task event)
710 	 * and running (scheduled onto the CPU), respectively.
711 	 */
712 	u64				total_time_enabled;
713 	u64				total_time_running;
714 	u64				tstamp;
715 
716 	struct perf_event_attr		attr;
717 	u16				header_size;
718 	u16				id_header_size;
719 	u16				read_size;
720 	struct hw_perf_event		hw;
721 
722 	struct perf_event_context	*ctx;
723 	atomic_long_t			refcount;
724 
725 	/*
726 	 * These accumulate total time (in nanoseconds) that children
727 	 * events have been enabled and running, respectively.
728 	 */
729 	atomic64_t			child_total_time_enabled;
730 	atomic64_t			child_total_time_running;
731 
732 	/*
733 	 * Protect attach/detach and child_list:
734 	 */
735 	struct mutex			child_mutex;
736 	struct list_head		child_list;
737 	struct perf_event		*parent;
738 
739 	int				oncpu;
740 	int				cpu;
741 
742 	struct list_head		owner_entry;
743 	struct task_struct		*owner;
744 
745 	/* mmap bits */
746 	struct mutex			mmap_mutex;
747 	atomic_t			mmap_count;
748 
749 	struct perf_buffer		*rb;
750 	struct list_head		rb_entry;
751 	unsigned long			rcu_batches;
752 	int				rcu_pending;
753 
754 	/* poll related */
755 	wait_queue_head_t		waitq;
756 	struct fasync_struct		*fasync;
757 
758 	/* delayed work for NMIs and such */
759 	unsigned int			pending_wakeup;
760 	unsigned int			pending_kill;
761 	unsigned int			pending_disable;
762 	unsigned int			pending_sigtrap;
763 	unsigned long			pending_addr;	/* SIGTRAP */
764 	struct irq_work			pending_irq;
765 	struct callback_head		pending_task;
766 	unsigned int			pending_work;
767 
768 	atomic_t			event_limit;
769 
770 	/* address range filters */
771 	struct perf_addr_filters_head	addr_filters;
772 	/* vma address array for file-based filders */
773 	struct perf_addr_filter_range	*addr_filter_ranges;
774 	unsigned long			addr_filters_gen;
775 
776 	/* for aux_output events */
777 	struct perf_event		*aux_event;
778 
779 	void (*destroy)(struct perf_event *);
780 	struct rcu_head			rcu_head;
781 
782 	struct pid_namespace		*ns;
783 	u64				id;
784 
785 	atomic64_t			lost_samples;
786 
787 	u64				(*clock)(void);
788 	perf_overflow_handler_t		overflow_handler;
789 	void				*overflow_handler_context;
790 #ifdef CONFIG_BPF_SYSCALL
791 	perf_overflow_handler_t		orig_overflow_handler;
792 	struct bpf_prog			*prog;
793 	u64				bpf_cookie;
794 #endif
795 
796 #ifdef CONFIG_EVENT_TRACING
797 	struct trace_event_call		*tp_event;
798 	struct event_filter		*filter;
799 #ifdef CONFIG_FUNCTION_TRACER
800 	struct ftrace_ops               ftrace_ops;
801 #endif
802 #endif
803 
804 #ifdef CONFIG_CGROUP_PERF
805 	struct perf_cgroup		*cgrp; /* cgroup event is attach to */
806 #endif
807 
808 #ifdef CONFIG_SECURITY
809 	void *security;
810 #endif
811 	struct list_head		sb_list;
812 #endif /* CONFIG_PERF_EVENTS */
813 };
814 
815 
816 struct perf_event_groups {
817 	struct rb_root	tree;
818 	u64		index;
819 };
820 
821 /**
822  * struct perf_event_context - event context structure
823  *
824  * Used as a container for task events and CPU events as well:
825  */
826 struct perf_event_context {
827 	struct pmu			*pmu;
828 	/*
829 	 * Protect the states of the events in the list,
830 	 * nr_active, and the list:
831 	 */
832 	raw_spinlock_t			lock;
833 	/*
834 	 * Protect the list of events.  Locking either mutex or lock
835 	 * is sufficient to ensure the list doesn't change; to change
836 	 * the list you need to lock both the mutex and the spinlock.
837 	 */
838 	struct mutex			mutex;
839 
840 	struct list_head		active_ctx_list;
841 	struct perf_event_groups	pinned_groups;
842 	struct perf_event_groups	flexible_groups;
843 	struct list_head		event_list;
844 
845 	struct list_head		pinned_active;
846 	struct list_head		flexible_active;
847 
848 	int				nr_events;
849 	int				nr_active;
850 	int				nr_user;
851 	int				is_active;
852 	int				nr_stat;
853 	int				nr_freq;
854 	int				rotate_disable;
855 	/*
856 	 * Set when nr_events != nr_active, except tolerant to events not
857 	 * necessary to be active due to scheduling constraints, such as cgroups.
858 	 */
859 	int				rotate_necessary;
860 	refcount_t			refcount;
861 	struct task_struct		*task;
862 
863 	/*
864 	 * Context clock, runs when context enabled.
865 	 */
866 	u64				time;
867 	u64				timestamp;
868 	u64				timeoffset;
869 
870 	/*
871 	 * These fields let us detect when two contexts have both
872 	 * been cloned (inherited) from a common ancestor.
873 	 */
874 	struct perf_event_context	*parent_ctx;
875 	u64				parent_gen;
876 	u64				generation;
877 	int				pin_count;
878 #ifdef CONFIG_CGROUP_PERF
879 	int				nr_cgroups;	 /* cgroup evts */
880 #endif
881 	void				*task_ctx_data; /* pmu specific data */
882 	struct rcu_head			rcu_head;
883 
884 	/*
885 	 * Sum (event->pending_sigtrap + event->pending_work)
886 	 *
887 	 * The SIGTRAP is targeted at ctx->task, as such it won't do changing
888 	 * that until the signal is delivered.
889 	 */
890 	local_t				nr_pending;
891 };
892 
893 /*
894  * Number of contexts where an event can trigger:
895  *	task, softirq, hardirq, nmi.
896  */
897 #define PERF_NR_CONTEXTS	4
898 
899 /**
900  * struct perf_cpu_context - per cpu event context structure
901  */
902 struct perf_cpu_context {
903 	struct perf_event_context	ctx;
904 	struct perf_event_context	*task_ctx;
905 	int				active_oncpu;
906 	int				exclusive;
907 
908 	raw_spinlock_t			hrtimer_lock;
909 	struct hrtimer			hrtimer;
910 	ktime_t				hrtimer_interval;
911 	unsigned int			hrtimer_active;
912 
913 #ifdef CONFIG_CGROUP_PERF
914 	struct perf_cgroup		*cgrp;
915 	struct list_head		cgrp_cpuctx_entry;
916 #endif
917 
918 	struct list_head		sched_cb_entry;
919 	int				sched_cb_usage;
920 
921 	int				online;
922 	/*
923 	 * Per-CPU storage for iterators used in visit_groups_merge. The default
924 	 * storage is of size 2 to hold the CPU and any CPU event iterators.
925 	 */
926 	int				heap_size;
927 	struct perf_event		**heap;
928 	struct perf_event		*heap_default[2];
929 };
930 
931 struct perf_output_handle {
932 	struct perf_event		*event;
933 	struct perf_buffer		*rb;
934 	unsigned long			wakeup;
935 	unsigned long			size;
936 	u64				aux_flags;
937 	union {
938 		void			*addr;
939 		unsigned long		head;
940 	};
941 	int				page;
942 };
943 
944 struct bpf_perf_event_data_kern {
945 	bpf_user_pt_regs_t *regs;
946 	struct perf_sample_data *data;
947 	struct perf_event *event;
948 };
949 
950 #ifdef CONFIG_CGROUP_PERF
951 
952 /*
953  * perf_cgroup_info keeps track of time_enabled for a cgroup.
954  * This is a per-cpu dynamically allocated data structure.
955  */
956 struct perf_cgroup_info {
957 	u64				time;
958 	u64				timestamp;
959 	u64				timeoffset;
960 	int				active;
961 };
962 
963 struct perf_cgroup {
964 	struct cgroup_subsys_state	css;
965 	struct perf_cgroup_info	__percpu *info;
966 };
967 
968 /*
969  * Must ensure cgroup is pinned (css_get) before calling
970  * this function. In other words, we cannot call this function
971  * if there is no cgroup event for the current CPU context.
972  */
973 static inline struct perf_cgroup *
perf_cgroup_from_task(struct task_struct * task,struct perf_event_context * ctx)974 perf_cgroup_from_task(struct task_struct *task, struct perf_event_context *ctx)
975 {
976 	return container_of(task_css_check(task, perf_event_cgrp_id,
977 					   ctx ? lockdep_is_held(&ctx->lock)
978 					       : true),
979 			    struct perf_cgroup, css);
980 }
981 #endif /* CONFIG_CGROUP_PERF */
982 
983 #ifdef CONFIG_PERF_EVENTS
984 
985 extern void *perf_aux_output_begin(struct perf_output_handle *handle,
986 				   struct perf_event *event);
987 extern void perf_aux_output_end(struct perf_output_handle *handle,
988 				unsigned long size);
989 extern int perf_aux_output_skip(struct perf_output_handle *handle,
990 				unsigned long size);
991 extern void *perf_get_aux(struct perf_output_handle *handle);
992 extern void perf_aux_output_flag(struct perf_output_handle *handle, u64 flags);
993 extern void perf_event_itrace_started(struct perf_event *event);
994 
995 extern int perf_pmu_register(struct pmu *pmu, const char *name, int type);
996 extern void perf_pmu_unregister(struct pmu *pmu);
997 
998 extern void __perf_event_task_sched_in(struct task_struct *prev,
999 				       struct task_struct *task);
1000 extern void __perf_event_task_sched_out(struct task_struct *prev,
1001 					struct task_struct *next);
1002 extern int perf_event_init_task(struct task_struct *child, u64 clone_flags);
1003 extern void perf_event_exit_task(struct task_struct *child);
1004 extern void perf_event_free_task(struct task_struct *task);
1005 extern void perf_event_delayed_put(struct task_struct *task);
1006 extern struct file *perf_event_get(unsigned int fd);
1007 extern const struct perf_event *perf_get_event(struct file *file);
1008 extern const struct perf_event_attr *perf_event_attrs(struct perf_event *event);
1009 extern void perf_event_print_debug(void);
1010 extern void perf_pmu_disable(struct pmu *pmu);
1011 extern void perf_pmu_enable(struct pmu *pmu);
1012 extern void perf_sched_cb_dec(struct pmu *pmu);
1013 extern void perf_sched_cb_inc(struct pmu *pmu);
1014 extern int perf_event_task_disable(void);
1015 extern int perf_event_task_enable(void);
1016 
1017 extern void perf_pmu_resched(struct pmu *pmu);
1018 
1019 extern int perf_event_refresh(struct perf_event *event, int refresh);
1020 extern void perf_event_update_userpage(struct perf_event *event);
1021 extern int perf_event_release_kernel(struct perf_event *event);
1022 extern struct perf_event *
1023 perf_event_create_kernel_counter(struct perf_event_attr *attr,
1024 				int cpu,
1025 				struct task_struct *task,
1026 				perf_overflow_handler_t callback,
1027 				void *context);
1028 extern void perf_pmu_migrate_context(struct pmu *pmu,
1029 				int src_cpu, int dst_cpu);
1030 int perf_event_read_local(struct perf_event *event, u64 *value,
1031 			  u64 *enabled, u64 *running);
1032 extern u64 perf_event_read_value(struct perf_event *event,
1033 				 u64 *enabled, u64 *running);
1034 
1035 
1036 struct perf_sample_data {
1037 	/*
1038 	 * Fields set by perf_sample_data_init(), group so as to
1039 	 * minimize the cachelines touched.
1040 	 */
1041 	u64				sample_flags;
1042 	u64				period;
1043 
1044 	/*
1045 	 * The other fields, optionally {set,used} by
1046 	 * perf_{prepare,output}_sample().
1047 	 */
1048 	struct perf_branch_stack	*br_stack;
1049 	union perf_sample_weight	weight;
1050 	union  perf_mem_data_src	data_src;
1051 	u64				txn;
1052 	u64				addr;
1053 	struct perf_raw_record		*raw;
1054 
1055 	u64				type;
1056 	u64				ip;
1057 	struct {
1058 		u32	pid;
1059 		u32	tid;
1060 	}				tid_entry;
1061 	u64				time;
1062 	u64				id;
1063 	u64				stream_id;
1064 	struct {
1065 		u32	cpu;
1066 		u32	reserved;
1067 	}				cpu_entry;
1068 	struct perf_callchain_entry	*callchain;
1069 	u64				aux_size;
1070 
1071 	struct perf_regs		regs_user;
1072 	struct perf_regs		regs_intr;
1073 	u64				stack_user_size;
1074 
1075 	u64				phys_addr;
1076 	u64				cgroup;
1077 	u64				data_page_size;
1078 	u64				code_page_size;
1079 } ____cacheline_aligned;
1080 
1081 /* default value for data source */
1082 #define PERF_MEM_NA (PERF_MEM_S(OP, NA)   |\
1083 		    PERF_MEM_S(LVL, NA)   |\
1084 		    PERF_MEM_S(SNOOP, NA) |\
1085 		    PERF_MEM_S(LOCK, NA)  |\
1086 		    PERF_MEM_S(TLB, NA))
1087 
perf_sample_data_init(struct perf_sample_data * data,u64 addr,u64 period)1088 static inline void perf_sample_data_init(struct perf_sample_data *data,
1089 					 u64 addr, u64 period)
1090 {
1091 	/* remaining struct members initialized in perf_prepare_sample() */
1092 	data->sample_flags = PERF_SAMPLE_PERIOD;
1093 	data->period = period;
1094 
1095 	if (addr) {
1096 		data->addr = addr;
1097 		data->sample_flags |= PERF_SAMPLE_ADDR;
1098 	}
1099 }
1100 
1101 /*
1102  * Clear all bitfields in the perf_branch_entry.
1103  * The to and from fields are not cleared because they are
1104  * systematically modified by caller.
1105  */
perf_clear_branch_entry_bitfields(struct perf_branch_entry * br)1106 static inline void perf_clear_branch_entry_bitfields(struct perf_branch_entry *br)
1107 {
1108 	br->mispred = 0;
1109 	br->predicted = 0;
1110 	br->in_tx = 0;
1111 	br->abort = 0;
1112 	br->cycles = 0;
1113 	br->type = 0;
1114 	br->spec = PERF_BR_SPEC_NA;
1115 	br->reserved = 0;
1116 }
1117 
1118 extern void perf_output_sample(struct perf_output_handle *handle,
1119 			       struct perf_event_header *header,
1120 			       struct perf_sample_data *data,
1121 			       struct perf_event *event);
1122 extern void perf_prepare_sample(struct perf_event_header *header,
1123 				struct perf_sample_data *data,
1124 				struct perf_event *event,
1125 				struct pt_regs *regs);
1126 
1127 extern int perf_event_overflow(struct perf_event *event,
1128 				 struct perf_sample_data *data,
1129 				 struct pt_regs *regs);
1130 
1131 extern void perf_event_output_forward(struct perf_event *event,
1132 				     struct perf_sample_data *data,
1133 				     struct pt_regs *regs);
1134 extern void perf_event_output_backward(struct perf_event *event,
1135 				       struct perf_sample_data *data,
1136 				       struct pt_regs *regs);
1137 extern int perf_event_output(struct perf_event *event,
1138 			     struct perf_sample_data *data,
1139 			     struct pt_regs *regs);
1140 
1141 static inline bool
is_default_overflow_handler(struct perf_event * event)1142 is_default_overflow_handler(struct perf_event *event)
1143 {
1144 	if (likely(event->overflow_handler == perf_event_output_forward))
1145 		return true;
1146 	if (unlikely(event->overflow_handler == perf_event_output_backward))
1147 		return true;
1148 	return false;
1149 }
1150 
1151 extern void
1152 perf_event_header__init_id(struct perf_event_header *header,
1153 			   struct perf_sample_data *data,
1154 			   struct perf_event *event);
1155 extern void
1156 perf_event__output_id_sample(struct perf_event *event,
1157 			     struct perf_output_handle *handle,
1158 			     struct perf_sample_data *sample);
1159 
1160 extern void
1161 perf_log_lost_samples(struct perf_event *event, u64 lost);
1162 
event_has_any_exclude_flag(struct perf_event * event)1163 static inline bool event_has_any_exclude_flag(struct perf_event *event)
1164 {
1165 	struct perf_event_attr *attr = &event->attr;
1166 
1167 	return attr->exclude_idle || attr->exclude_user ||
1168 	       attr->exclude_kernel || attr->exclude_hv ||
1169 	       attr->exclude_guest || attr->exclude_host;
1170 }
1171 
is_sampling_event(struct perf_event * event)1172 static inline bool is_sampling_event(struct perf_event *event)
1173 {
1174 	return event->attr.sample_period != 0;
1175 }
1176 
1177 /*
1178  * Return 1 for a software event, 0 for a hardware event
1179  */
is_software_event(struct perf_event * event)1180 static inline int is_software_event(struct perf_event *event)
1181 {
1182 	return event->event_caps & PERF_EV_CAP_SOFTWARE;
1183 }
1184 
1185 /*
1186  * Return 1 for event in sw context, 0 for event in hw context
1187  */
in_software_context(struct perf_event * event)1188 static inline int in_software_context(struct perf_event *event)
1189 {
1190 	return event->ctx->pmu->task_ctx_nr == perf_sw_context;
1191 }
1192 
is_exclusive_pmu(struct pmu * pmu)1193 static inline int is_exclusive_pmu(struct pmu *pmu)
1194 {
1195 	return pmu->capabilities & PERF_PMU_CAP_EXCLUSIVE;
1196 }
1197 
1198 extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX];
1199 
1200 extern void ___perf_sw_event(u32, u64, struct pt_regs *, u64);
1201 extern void __perf_sw_event(u32, u64, struct pt_regs *, u64);
1202 
1203 #ifndef perf_arch_fetch_caller_regs
perf_arch_fetch_caller_regs(struct pt_regs * regs,unsigned long ip)1204 static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { }
1205 #endif
1206 
1207 /*
1208  * When generating a perf sample in-line, instead of from an interrupt /
1209  * exception, we lack a pt_regs. This is typically used from software events
1210  * like: SW_CONTEXT_SWITCHES, SW_MIGRATIONS and the tie-in with tracepoints.
1211  *
1212  * We typically don't need a full set, but (for x86) do require:
1213  * - ip for PERF_SAMPLE_IP
1214  * - cs for user_mode() tests
1215  * - sp for PERF_SAMPLE_CALLCHAIN
1216  * - eflags for MISC bits and CALLCHAIN (see: perf_hw_regs())
1217  *
1218  * NOTE: assumes @regs is otherwise already 0 filled; this is important for
1219  * things like PERF_SAMPLE_REGS_INTR.
1220  */
perf_fetch_caller_regs(struct pt_regs * regs)1221 static inline void perf_fetch_caller_regs(struct pt_regs *regs)
1222 {
1223 	perf_arch_fetch_caller_regs(regs, CALLER_ADDR0);
1224 }
1225 
1226 static __always_inline void
perf_sw_event(u32 event_id,u64 nr,struct pt_regs * regs,u64 addr)1227 perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
1228 {
1229 	if (static_key_false(&perf_swevent_enabled[event_id]))
1230 		__perf_sw_event(event_id, nr, regs, addr);
1231 }
1232 
1233 DECLARE_PER_CPU(struct pt_regs, __perf_regs[4]);
1234 
1235 /*
1236  * 'Special' version for the scheduler, it hard assumes no recursion,
1237  * which is guaranteed by us not actually scheduling inside other swevents
1238  * because those disable preemption.
1239  */
__perf_sw_event_sched(u32 event_id,u64 nr,u64 addr)1240 static __always_inline void __perf_sw_event_sched(u32 event_id, u64 nr, u64 addr)
1241 {
1242 	struct pt_regs *regs = this_cpu_ptr(&__perf_regs[0]);
1243 
1244 	perf_fetch_caller_regs(regs);
1245 	___perf_sw_event(event_id, nr, regs, addr);
1246 }
1247 
1248 extern struct static_key_false perf_sched_events;
1249 
__perf_sw_enabled(int swevt)1250 static __always_inline bool __perf_sw_enabled(int swevt)
1251 {
1252 	return static_key_false(&perf_swevent_enabled[swevt]);
1253 }
1254 
perf_event_task_migrate(struct task_struct * task)1255 static inline void perf_event_task_migrate(struct task_struct *task)
1256 {
1257 	if (__perf_sw_enabled(PERF_COUNT_SW_CPU_MIGRATIONS))
1258 		task->sched_migrated = 1;
1259 }
1260 
perf_event_task_sched_in(struct task_struct * prev,struct task_struct * task)1261 static inline void perf_event_task_sched_in(struct task_struct *prev,
1262 					    struct task_struct *task)
1263 {
1264 	if (static_branch_unlikely(&perf_sched_events))
1265 		__perf_event_task_sched_in(prev, task);
1266 
1267 	if (__perf_sw_enabled(PERF_COUNT_SW_CPU_MIGRATIONS) &&
1268 	    task->sched_migrated) {
1269 		__perf_sw_event_sched(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 0);
1270 		task->sched_migrated = 0;
1271 	}
1272 }
1273 
perf_event_task_sched_out(struct task_struct * prev,struct task_struct * next)1274 static inline void perf_event_task_sched_out(struct task_struct *prev,
1275 					     struct task_struct *next)
1276 {
1277 	if (__perf_sw_enabled(PERF_COUNT_SW_CONTEXT_SWITCHES))
1278 		__perf_sw_event_sched(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 0);
1279 
1280 #ifdef CONFIG_CGROUP_PERF
1281 	if (__perf_sw_enabled(PERF_COUNT_SW_CGROUP_SWITCHES) &&
1282 	    perf_cgroup_from_task(prev, NULL) !=
1283 	    perf_cgroup_from_task(next, NULL))
1284 		__perf_sw_event_sched(PERF_COUNT_SW_CGROUP_SWITCHES, 1, 0);
1285 #endif
1286 
1287 	if (static_branch_unlikely(&perf_sched_events))
1288 		__perf_event_task_sched_out(prev, next);
1289 }
1290 
1291 extern void perf_event_mmap(struct vm_area_struct *vma);
1292 
1293 extern void perf_event_ksymbol(u16 ksym_type, u64 addr, u32 len,
1294 			       bool unregister, const char *sym);
1295 extern void perf_event_bpf_event(struct bpf_prog *prog,
1296 				 enum perf_bpf_event_type type,
1297 				 u16 flags);
1298 
1299 #ifdef CONFIG_GUEST_PERF_EVENTS
1300 extern struct perf_guest_info_callbacks __rcu *perf_guest_cbs;
1301 
1302 DECLARE_STATIC_CALL(__perf_guest_state, *perf_guest_cbs->state);
1303 DECLARE_STATIC_CALL(__perf_guest_get_ip, *perf_guest_cbs->get_ip);
1304 DECLARE_STATIC_CALL(__perf_guest_handle_intel_pt_intr, *perf_guest_cbs->handle_intel_pt_intr);
1305 
perf_guest_state(void)1306 static inline unsigned int perf_guest_state(void)
1307 {
1308 	return static_call(__perf_guest_state)();
1309 }
perf_guest_get_ip(void)1310 static inline unsigned long perf_guest_get_ip(void)
1311 {
1312 	return static_call(__perf_guest_get_ip)();
1313 }
perf_guest_handle_intel_pt_intr(void)1314 static inline unsigned int perf_guest_handle_intel_pt_intr(void)
1315 {
1316 	return static_call(__perf_guest_handle_intel_pt_intr)();
1317 }
1318 extern void perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *cbs);
1319 extern void perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *cbs);
1320 #else
perf_guest_state(void)1321 static inline unsigned int perf_guest_state(void)		 { return 0; }
perf_guest_get_ip(void)1322 static inline unsigned long perf_guest_get_ip(void)		 { return 0; }
perf_guest_handle_intel_pt_intr(void)1323 static inline unsigned int perf_guest_handle_intel_pt_intr(void) { return 0; }
1324 #endif /* CONFIG_GUEST_PERF_EVENTS */
1325 
1326 extern void perf_event_exec(void);
1327 extern void perf_event_comm(struct task_struct *tsk, bool exec);
1328 extern void perf_event_namespaces(struct task_struct *tsk);
1329 extern void perf_event_fork(struct task_struct *tsk);
1330 extern void perf_event_text_poke(const void *addr,
1331 				 const void *old_bytes, size_t old_len,
1332 				 const void *new_bytes, size_t new_len);
1333 
1334 /* Callchains */
1335 DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry);
1336 
1337 extern void perf_callchain_user(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs);
1338 extern void perf_callchain_kernel(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs);
1339 extern struct perf_callchain_entry *
1340 get_perf_callchain(struct pt_regs *regs, u32 init_nr, bool kernel, bool user,
1341 		   u32 max_stack, bool crosstask, bool add_mark);
1342 extern struct perf_callchain_entry *perf_callchain(struct perf_event *event, struct pt_regs *regs);
1343 extern int get_callchain_buffers(int max_stack);
1344 extern void put_callchain_buffers(void);
1345 extern struct perf_callchain_entry *get_callchain_entry(int *rctx);
1346 extern void put_callchain_entry(int rctx);
1347 
1348 extern int sysctl_perf_event_max_stack;
1349 extern int sysctl_perf_event_max_contexts_per_stack;
1350 
perf_callchain_store_context(struct perf_callchain_entry_ctx * ctx,u64 ip)1351 static inline int perf_callchain_store_context(struct perf_callchain_entry_ctx *ctx, u64 ip)
1352 {
1353 	if (ctx->contexts < sysctl_perf_event_max_contexts_per_stack) {
1354 		struct perf_callchain_entry *entry = ctx->entry;
1355 		entry->ip[entry->nr++] = ip;
1356 		++ctx->contexts;
1357 		return 0;
1358 	} else {
1359 		ctx->contexts_maxed = true;
1360 		return -1; /* no more room, stop walking the stack */
1361 	}
1362 }
1363 
perf_callchain_store(struct perf_callchain_entry_ctx * ctx,u64 ip)1364 static inline int perf_callchain_store(struct perf_callchain_entry_ctx *ctx, u64 ip)
1365 {
1366 	if (ctx->nr < ctx->max_stack && !ctx->contexts_maxed) {
1367 		struct perf_callchain_entry *entry = ctx->entry;
1368 		entry->ip[entry->nr++] = ip;
1369 		++ctx->nr;
1370 		return 0;
1371 	} else {
1372 		return -1; /* no more room, stop walking the stack */
1373 	}
1374 }
1375 
1376 extern int sysctl_perf_event_paranoid;
1377 extern int sysctl_perf_event_mlock;
1378 extern int sysctl_perf_event_sample_rate;
1379 extern int sysctl_perf_cpu_time_max_percent;
1380 
1381 extern void perf_sample_event_took(u64 sample_len_ns);
1382 
1383 int perf_proc_update_handler(struct ctl_table *table, int write,
1384 		void *buffer, size_t *lenp, loff_t *ppos);
1385 int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write,
1386 		void *buffer, size_t *lenp, loff_t *ppos);
1387 int perf_event_max_stack_handler(struct ctl_table *table, int write,
1388 		void *buffer, size_t *lenp, loff_t *ppos);
1389 
1390 /* Access to perf_event_open(2) syscall. */
1391 #define PERF_SECURITY_OPEN		0
1392 
1393 /* Finer grained perf_event_open(2) access control. */
1394 #define PERF_SECURITY_CPU		1
1395 #define PERF_SECURITY_KERNEL		2
1396 #define PERF_SECURITY_TRACEPOINT	3
1397 
perf_is_paranoid(void)1398 static inline int perf_is_paranoid(void)
1399 {
1400 	return sysctl_perf_event_paranoid > -1;
1401 }
1402 
perf_allow_kernel(struct perf_event_attr * attr)1403 static inline int perf_allow_kernel(struct perf_event_attr *attr)
1404 {
1405 	if (sysctl_perf_event_paranoid > 1 && !perfmon_capable())
1406 		return -EACCES;
1407 
1408 	return security_perf_event_open(attr, PERF_SECURITY_KERNEL);
1409 }
1410 
perf_allow_cpu(struct perf_event_attr * attr)1411 static inline int perf_allow_cpu(struct perf_event_attr *attr)
1412 {
1413 	if (sysctl_perf_event_paranoid > 0 && !perfmon_capable())
1414 		return -EACCES;
1415 
1416 	return security_perf_event_open(attr, PERF_SECURITY_CPU);
1417 }
1418 
perf_allow_tracepoint(struct perf_event_attr * attr)1419 static inline int perf_allow_tracepoint(struct perf_event_attr *attr)
1420 {
1421 	if (sysctl_perf_event_paranoid > -1 && !perfmon_capable())
1422 		return -EPERM;
1423 
1424 	return security_perf_event_open(attr, PERF_SECURITY_TRACEPOINT);
1425 }
1426 
1427 extern void perf_event_init(void);
1428 extern void perf_tp_event(u16 event_type, u64 count, void *record,
1429 			  int entry_size, struct pt_regs *regs,
1430 			  struct hlist_head *head, int rctx,
1431 			  struct task_struct *task);
1432 extern void perf_bp_event(struct perf_event *event, void *data);
1433 
1434 #ifndef perf_misc_flags
1435 # define perf_misc_flags(regs) \
1436 		(user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL)
1437 # define perf_instruction_pointer(regs)	instruction_pointer(regs)
1438 #endif
1439 #ifndef perf_arch_bpf_user_pt_regs
1440 # define perf_arch_bpf_user_pt_regs(regs) regs
1441 #endif
1442 
has_branch_stack(struct perf_event * event)1443 static inline bool has_branch_stack(struct perf_event *event)
1444 {
1445 	return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK;
1446 }
1447 
needs_branch_stack(struct perf_event * event)1448 static inline bool needs_branch_stack(struct perf_event *event)
1449 {
1450 	return event->attr.branch_sample_type != 0;
1451 }
1452 
has_aux(struct perf_event * event)1453 static inline bool has_aux(struct perf_event *event)
1454 {
1455 	return event->pmu->setup_aux;
1456 }
1457 
is_write_backward(struct perf_event * event)1458 static inline bool is_write_backward(struct perf_event *event)
1459 {
1460 	return !!event->attr.write_backward;
1461 }
1462 
has_addr_filter(struct perf_event * event)1463 static inline bool has_addr_filter(struct perf_event *event)
1464 {
1465 	return event->pmu->nr_addr_filters;
1466 }
1467 
1468 /*
1469  * An inherited event uses parent's filters
1470  */
1471 static inline struct perf_addr_filters_head *
perf_event_addr_filters(struct perf_event * event)1472 perf_event_addr_filters(struct perf_event *event)
1473 {
1474 	struct perf_addr_filters_head *ifh = &event->addr_filters;
1475 
1476 	if (event->parent)
1477 		ifh = &event->parent->addr_filters;
1478 
1479 	return ifh;
1480 }
1481 
1482 extern void perf_event_addr_filters_sync(struct perf_event *event);
1483 extern void perf_report_aux_output_id(struct perf_event *event, u64 hw_id);
1484 
1485 extern int perf_output_begin(struct perf_output_handle *handle,
1486 			     struct perf_sample_data *data,
1487 			     struct perf_event *event, unsigned int size);
1488 extern int perf_output_begin_forward(struct perf_output_handle *handle,
1489 				     struct perf_sample_data *data,
1490 				     struct perf_event *event,
1491 				     unsigned int size);
1492 extern int perf_output_begin_backward(struct perf_output_handle *handle,
1493 				      struct perf_sample_data *data,
1494 				      struct perf_event *event,
1495 				      unsigned int size);
1496 
1497 extern void perf_output_end(struct perf_output_handle *handle);
1498 extern unsigned int perf_output_copy(struct perf_output_handle *handle,
1499 			     const void *buf, unsigned int len);
1500 extern unsigned int perf_output_skip(struct perf_output_handle *handle,
1501 				     unsigned int len);
1502 extern long perf_output_copy_aux(struct perf_output_handle *aux_handle,
1503 				 struct perf_output_handle *handle,
1504 				 unsigned long from, unsigned long to);
1505 extern int perf_swevent_get_recursion_context(void);
1506 extern void perf_swevent_put_recursion_context(int rctx);
1507 extern u64 perf_swevent_set_period(struct perf_event *event);
1508 extern void perf_event_enable(struct perf_event *event);
1509 extern void perf_event_disable(struct perf_event *event);
1510 extern void perf_event_disable_local(struct perf_event *event);
1511 extern void perf_event_disable_inatomic(struct perf_event *event);
1512 extern void perf_event_task_tick(void);
1513 extern int perf_event_account_interrupt(struct perf_event *event);
1514 extern int perf_event_period(struct perf_event *event, u64 value);
1515 extern u64 perf_event_pause(struct perf_event *event, bool reset);
1516 #else /* !CONFIG_PERF_EVENTS: */
1517 static inline void *
perf_aux_output_begin(struct perf_output_handle * handle,struct perf_event * event)1518 perf_aux_output_begin(struct perf_output_handle *handle,
1519 		      struct perf_event *event)				{ return NULL; }
1520 static inline void
perf_aux_output_end(struct perf_output_handle * handle,unsigned long size)1521 perf_aux_output_end(struct perf_output_handle *handle, unsigned long size)
1522 									{ }
1523 static inline int
perf_aux_output_skip(struct perf_output_handle * handle,unsigned long size)1524 perf_aux_output_skip(struct perf_output_handle *handle,
1525 		     unsigned long size)				{ return -EINVAL; }
1526 static inline void *
perf_get_aux(struct perf_output_handle * handle)1527 perf_get_aux(struct perf_output_handle *handle)				{ return NULL; }
1528 static inline void
perf_event_task_migrate(struct task_struct * task)1529 perf_event_task_migrate(struct task_struct *task)			{ }
1530 static inline void
perf_event_task_sched_in(struct task_struct * prev,struct task_struct * task)1531 perf_event_task_sched_in(struct task_struct *prev,
1532 			 struct task_struct *task)			{ }
1533 static inline void
perf_event_task_sched_out(struct task_struct * prev,struct task_struct * next)1534 perf_event_task_sched_out(struct task_struct *prev,
1535 			  struct task_struct *next)			{ }
perf_event_init_task(struct task_struct * child,u64 clone_flags)1536 static inline int perf_event_init_task(struct task_struct *child,
1537 				       u64 clone_flags)			{ return 0; }
perf_event_exit_task(struct task_struct * child)1538 static inline void perf_event_exit_task(struct task_struct *child)	{ }
perf_event_free_task(struct task_struct * task)1539 static inline void perf_event_free_task(struct task_struct *task)	{ }
perf_event_delayed_put(struct task_struct * task)1540 static inline void perf_event_delayed_put(struct task_struct *task)	{ }
perf_event_get(unsigned int fd)1541 static inline struct file *perf_event_get(unsigned int fd)	{ return ERR_PTR(-EINVAL); }
perf_get_event(struct file * file)1542 static inline const struct perf_event *perf_get_event(struct file *file)
1543 {
1544 	return ERR_PTR(-EINVAL);
1545 }
perf_event_attrs(struct perf_event * event)1546 static inline const struct perf_event_attr *perf_event_attrs(struct perf_event *event)
1547 {
1548 	return ERR_PTR(-EINVAL);
1549 }
perf_event_read_local(struct perf_event * event,u64 * value,u64 * enabled,u64 * running)1550 static inline int perf_event_read_local(struct perf_event *event, u64 *value,
1551 					u64 *enabled, u64 *running)
1552 {
1553 	return -EINVAL;
1554 }
perf_event_print_debug(void)1555 static inline void perf_event_print_debug(void)				{ }
perf_event_task_disable(void)1556 static inline int perf_event_task_disable(void)				{ return -EINVAL; }
perf_event_task_enable(void)1557 static inline int perf_event_task_enable(void)				{ return -EINVAL; }
perf_event_refresh(struct perf_event * event,int refresh)1558 static inline int perf_event_refresh(struct perf_event *event, int refresh)
1559 {
1560 	return -EINVAL;
1561 }
1562 
1563 static inline void
perf_sw_event(u32 event_id,u64 nr,struct pt_regs * regs,u64 addr)1564 perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)	{ }
1565 static inline void
perf_bp_event(struct perf_event * event,void * data)1566 perf_bp_event(struct perf_event *event, void *data)			{ }
1567 
perf_event_mmap(struct vm_area_struct * vma)1568 static inline void perf_event_mmap(struct vm_area_struct *vma)		{ }
1569 
1570 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)1571 static inline void perf_event_ksymbol(u16 ksym_type, u64 addr, u32 len,
1572 				      bool unregister, const char *sym)	{ }
perf_event_bpf_event(struct bpf_prog * prog,enum perf_bpf_event_type type,u16 flags)1573 static inline void perf_event_bpf_event(struct bpf_prog *prog,
1574 					enum perf_bpf_event_type type,
1575 					u16 flags)			{ }
perf_event_exec(void)1576 static inline void perf_event_exec(void)				{ }
perf_event_comm(struct task_struct * tsk,bool exec)1577 static inline void perf_event_comm(struct task_struct *tsk, bool exec)	{ }
perf_event_namespaces(struct task_struct * tsk)1578 static inline void perf_event_namespaces(struct task_struct *tsk)	{ }
perf_event_fork(struct task_struct * tsk)1579 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)1580 static inline void perf_event_text_poke(const void *addr,
1581 					const void *old_bytes,
1582 					size_t old_len,
1583 					const void *new_bytes,
1584 					size_t new_len)			{ }
perf_event_init(void)1585 static inline void perf_event_init(void)				{ }
perf_swevent_get_recursion_context(void)1586 static inline int  perf_swevent_get_recursion_context(void)		{ return -1; }
perf_swevent_put_recursion_context(int rctx)1587 static inline void perf_swevent_put_recursion_context(int rctx)		{ }
perf_swevent_set_period(struct perf_event * event)1588 static inline u64 perf_swevent_set_period(struct perf_event *event)	{ return 0; }
perf_event_enable(struct perf_event * event)1589 static inline void perf_event_enable(struct perf_event *event)		{ }
perf_event_disable(struct perf_event * event)1590 static inline void perf_event_disable(struct perf_event *event)		{ }
__perf_event_disable(void * info)1591 static inline int __perf_event_disable(void *info)			{ return -1; }
perf_event_task_tick(void)1592 static inline void perf_event_task_tick(void)				{ }
perf_event_release_kernel(struct perf_event * event)1593 static inline int perf_event_release_kernel(struct perf_event *event)	{ return 0; }
perf_event_period(struct perf_event * event,u64 value)1594 static inline int perf_event_period(struct perf_event *event, u64 value)
1595 {
1596 	return -EINVAL;
1597 }
perf_event_pause(struct perf_event * event,bool reset)1598 static inline u64 perf_event_pause(struct perf_event *event, bool reset)
1599 {
1600 	return 0;
1601 }
1602 #endif
1603 
1604 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
1605 extern void perf_restore_debug_store(void);
1606 #else
perf_restore_debug_store(void)1607 static inline void perf_restore_debug_store(void)			{ }
1608 #endif
1609 
perf_raw_frag_last(const struct perf_raw_frag * frag)1610 static __always_inline bool perf_raw_frag_last(const struct perf_raw_frag *frag)
1611 {
1612 	return frag->pad < sizeof(u64);
1613 }
1614 
1615 #define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x))
1616 
1617 struct perf_pmu_events_attr {
1618 	struct device_attribute attr;
1619 	u64 id;
1620 	const char *event_str;
1621 };
1622 
1623 struct perf_pmu_events_ht_attr {
1624 	struct device_attribute			attr;
1625 	u64					id;
1626 	const char				*event_str_ht;
1627 	const char				*event_str_noht;
1628 };
1629 
1630 struct perf_pmu_events_hybrid_attr {
1631 	struct device_attribute			attr;
1632 	u64					id;
1633 	const char				*event_str;
1634 	u64					pmu_type;
1635 };
1636 
1637 struct perf_pmu_format_hybrid_attr {
1638 	struct device_attribute			attr;
1639 	u64					pmu_type;
1640 };
1641 
1642 ssize_t perf_event_sysfs_show(struct device *dev, struct device_attribute *attr,
1643 			      char *page);
1644 
1645 #define PMU_EVENT_ATTR(_name, _var, _id, _show)				\
1646 static struct perf_pmu_events_attr _var = {				\
1647 	.attr = __ATTR(_name, 0444, _show, NULL),			\
1648 	.id   =  _id,							\
1649 };
1650 
1651 #define PMU_EVENT_ATTR_STRING(_name, _var, _str)			    \
1652 static struct perf_pmu_events_attr _var = {				    \
1653 	.attr		= __ATTR(_name, 0444, perf_event_sysfs_show, NULL), \
1654 	.id		= 0,						    \
1655 	.event_str	= _str,						    \
1656 };
1657 
1658 #define PMU_EVENT_ATTR_ID(_name, _show, _id)				\
1659 	(&((struct perf_pmu_events_attr[]) {				\
1660 		{ .attr = __ATTR(_name, 0444, _show, NULL),		\
1661 		  .id = _id, }						\
1662 	})[0].attr.attr)
1663 
1664 #define PMU_FORMAT_ATTR(_name, _format)					\
1665 static ssize_t								\
1666 _name##_show(struct device *dev,					\
1667 			       struct device_attribute *attr,		\
1668 			       char *page)				\
1669 {									\
1670 	BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE);			\
1671 	return sprintf(page, _format "\n");				\
1672 }									\
1673 									\
1674 static struct device_attribute format_attr_##_name = __ATTR_RO(_name)
1675 
1676 /* Performance counter hotplug functions */
1677 #ifdef CONFIG_PERF_EVENTS
1678 int perf_event_init_cpu(unsigned int cpu);
1679 int perf_event_exit_cpu(unsigned int cpu);
1680 #else
1681 #define perf_event_init_cpu	NULL
1682 #define perf_event_exit_cpu	NULL
1683 #endif
1684 
1685 extern void __weak arch_perf_update_userpage(struct perf_event *event,
1686 					     struct perf_event_mmap_page *userpg,
1687 					     u64 now);
1688 
1689 #ifdef CONFIG_MMU
1690 extern __weak u64 arch_perf_get_page_size(struct mm_struct *mm, unsigned long addr);
1691 #endif
1692 
1693 /*
1694  * Snapshot branch stack on software events.
1695  *
1696  * Branch stack can be very useful in understanding software events. For
1697  * example, when a long function, e.g. sys_perf_event_open, returns an
1698  * errno, it is not obvious why the function failed. Branch stack could
1699  * provide very helpful information in this type of scenarios.
1700  *
1701  * On software event, it is necessary to stop the hardware branch recorder
1702  * fast. Otherwise, the hardware register/buffer will be flushed with
1703  * entries of the triggering event. Therefore, static call is used to
1704  * stop the hardware recorder.
1705  */
1706 
1707 /*
1708  * cnt is the number of entries allocated for entries.
1709  * Return number of entries copied to .
1710  */
1711 typedef int (perf_snapshot_branch_stack_t)(struct perf_branch_entry *entries,
1712 					   unsigned int cnt);
1713 DECLARE_STATIC_CALL(perf_snapshot_branch_stack, perf_snapshot_branch_stack_t);
1714 
1715 #ifndef PERF_NEEDS_LOPWR_CB
perf_lopwr_cb(bool mode)1716 static inline void perf_lopwr_cb(bool mode)
1717 {
1718 }
1719 #endif
1720 
1721 #ifdef CONFIG_PERF_EVENTS
branch_sample_no_flags(const struct perf_event * event)1722 static inline bool branch_sample_no_flags(const struct perf_event *event)
1723 {
1724 	return event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_NO_FLAGS;
1725 }
1726 
branch_sample_no_cycles(const struct perf_event * event)1727 static inline bool branch_sample_no_cycles(const struct perf_event *event)
1728 {
1729 	return event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_NO_CYCLES;
1730 }
1731 
branch_sample_type(const struct perf_event * event)1732 static inline bool branch_sample_type(const struct perf_event *event)
1733 {
1734 	return event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_TYPE_SAVE;
1735 }
1736 
branch_sample_hw_index(const struct perf_event * event)1737 static inline bool branch_sample_hw_index(const struct perf_event *event)
1738 {
1739 	return event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_HW_INDEX;
1740 }
1741 
branch_sample_priv(const struct perf_event * event)1742 static inline bool branch_sample_priv(const struct perf_event *event)
1743 {
1744 	return event->attr.branch_sample_type & PERF_SAMPLE_BRANCH_PRIV_SAVE;
1745 }
1746 #endif /* CONFIG_PERF_EVENTS */
1747 #endif /* _LINUX_PERF_EVENT_H */
1748