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 struct perf_event *group_leader;
665 struct pmu *pmu;
666 void *pmu_private;
667
668 enum perf_event_state state;
669 unsigned int attach_state;
670 local64_t count;
671 atomic64_t child_count;
672
673 /*
674 * These are the total time in nanoseconds that the event
675 * has been enabled (i.e. eligible to run, and the task has
676 * been scheduled in, if this is a per-task event)
677 * and running (scheduled onto the CPU), respectively.
678 */
679 u64 total_time_enabled;
680 u64 total_time_running;
681 u64 tstamp;
682
683 /*
684 * timestamp shadows the actual context timing but it can
685 * be safely used in NMI interrupt context. It reflects the
686 * context time as it was when the event was last scheduled in.
687 *
688 * ctx_time already accounts for ctx->timestamp. Therefore to
689 * compute ctx_time for a sample, simply add perf_clock().
690 */
691 u64 shadow_ctx_time;
692
693 struct perf_event_attr attr;
694 u16 header_size;
695 u16 id_header_size;
696 u16 read_size;
697 struct hw_perf_event hw;
698
699 struct perf_event_context *ctx;
700 atomic_long_t refcount;
701
702 /*
703 * These accumulate total time (in nanoseconds) that children
704 * events have been enabled and running, respectively.
705 */
706 atomic64_t child_total_time_enabled;
707 atomic64_t child_total_time_running;
708
709 /*
710 * Protect attach/detach and child_list:
711 */
712 struct mutex child_mutex;
713 struct list_head child_list;
714 struct perf_event *parent;
715
716 int oncpu;
717 int cpu;
718
719 struct list_head owner_entry;
720 struct task_struct *owner;
721
722 /* mmap bits */
723 struct mutex mmap_mutex;
724 atomic_t mmap_count;
725
726 struct perf_buffer *rb;
727 struct list_head rb_entry;
728 unsigned long rcu_batches;
729 int rcu_pending;
730
731 /* poll related */
732 wait_queue_head_t waitq;
733 struct fasync_struct *fasync;
734
735 /* delayed work for NMIs and such */
736 int pending_wakeup;
737 int pending_kill;
738 int pending_disable;
739 unsigned long pending_addr; /* SIGTRAP */
740 struct irq_work pending;
741
742 atomic_t event_limit;
743
744 /* address range filters */
745 struct perf_addr_filters_head addr_filters;
746 /* vma address array for file-based filders */
747 struct perf_addr_filter_range *addr_filter_ranges;
748 unsigned long addr_filters_gen;
749
750 /* for aux_output events */
751 struct perf_event *aux_event;
752
753 void (*destroy)(struct perf_event *);
754 struct rcu_head rcu_head;
755
756 struct pid_namespace *ns;
757 u64 id;
758
759 u64 (*clock)(void);
760 perf_overflow_handler_t overflow_handler;
761 void *overflow_handler_context;
762 #ifdef CONFIG_BPF_SYSCALL
763 perf_overflow_handler_t orig_overflow_handler;
764 struct bpf_prog *prog;
765 #endif
766
767 #ifdef CONFIG_EVENT_TRACING
768 struct trace_event_call *tp_event;
769 struct event_filter *filter;
770 #ifdef CONFIG_FUNCTION_TRACER
771 struct ftrace_ops ftrace_ops;
772 #endif
773 #endif
774
775 #ifdef CONFIG_CGROUP_PERF
776 struct perf_cgroup *cgrp; /* cgroup event is attach to */
777 #endif
778
779 #ifdef CONFIG_SECURITY
780 void *security;
781 #endif
782 struct list_head sb_list;
783 #endif /* CONFIG_PERF_EVENTS */
784 };
785
786
787 struct perf_event_groups {
788 struct rb_root tree;
789 u64 index;
790 };
791
792 /**
793 * struct perf_event_context - event context structure
794 *
795 * Used as a container for task events and CPU events as well:
796 */
797 struct perf_event_context {
798 struct pmu *pmu;
799 /*
800 * Protect the states of the events in the list,
801 * nr_active, and the list:
802 */
803 raw_spinlock_t lock;
804 /*
805 * Protect the list of events. Locking either mutex or lock
806 * is sufficient to ensure the list doesn't change; to change
807 * the list you need to lock both the mutex and the spinlock.
808 */
809 struct mutex mutex;
810
811 struct list_head active_ctx_list;
812 struct perf_event_groups pinned_groups;
813 struct perf_event_groups flexible_groups;
814 struct list_head event_list;
815
816 struct list_head pinned_active;
817 struct list_head flexible_active;
818
819 int nr_events;
820 int nr_active;
821 int is_active;
822 int nr_stat;
823 int nr_freq;
824 int rotate_disable;
825 /*
826 * Set when nr_events != nr_active, except tolerant to events not
827 * necessary to be active due to scheduling constraints, such as cgroups.
828 */
829 int rotate_necessary;
830 refcount_t refcount;
831 struct task_struct *task;
832
833 /*
834 * Context clock, runs when context enabled.
835 */
836 u64 time;
837 u64 timestamp;
838
839 /*
840 * These fields let us detect when two contexts have both
841 * been cloned (inherited) from a common ancestor.
842 */
843 struct perf_event_context *parent_ctx;
844 u64 parent_gen;
845 u64 generation;
846 int pin_count;
847 #ifdef CONFIG_CGROUP_PERF
848 int nr_cgroups; /* cgroup evts */
849 #endif
850 void *task_ctx_data; /* pmu specific data */
851 struct rcu_head rcu_head;
852 };
853
854 /*
855 * Number of contexts where an event can trigger:
856 * task, softirq, hardirq, nmi.
857 */
858 #define PERF_NR_CONTEXTS 4
859
860 /**
861 * struct perf_event_cpu_context - per cpu event context structure
862 */
863 struct perf_cpu_context {
864 struct perf_event_context ctx;
865 struct perf_event_context *task_ctx;
866 int active_oncpu;
867 int exclusive;
868
869 raw_spinlock_t hrtimer_lock;
870 struct hrtimer hrtimer;
871 ktime_t hrtimer_interval;
872 unsigned int hrtimer_active;
873
874 #ifdef CONFIG_CGROUP_PERF
875 struct perf_cgroup *cgrp;
876 struct list_head cgrp_cpuctx_entry;
877 #endif
878
879 struct list_head sched_cb_entry;
880 int sched_cb_usage;
881
882 int online;
883 /*
884 * Per-CPU storage for iterators used in visit_groups_merge. The default
885 * storage is of size 2 to hold the CPU and any CPU event iterators.
886 */
887 int heap_size;
888 struct perf_event **heap;
889 struct perf_event *heap_default[2];
890 };
891
892 struct perf_output_handle {
893 struct perf_event *event;
894 struct perf_buffer *rb;
895 unsigned long wakeup;
896 unsigned long size;
897 u64 aux_flags;
898 union {
899 void *addr;
900 unsigned long head;
901 };
902 int page;
903 };
904
905 struct bpf_perf_event_data_kern {
906 bpf_user_pt_regs_t *regs;
907 struct perf_sample_data *data;
908 struct perf_event *event;
909 };
910
911 #ifdef CONFIG_CGROUP_PERF
912
913 /*
914 * perf_cgroup_info keeps track of time_enabled for a cgroup.
915 * This is a per-cpu dynamically allocated data structure.
916 */
917 struct perf_cgroup_info {
918 u64 time;
919 u64 timestamp;
920 };
921
922 struct perf_cgroup {
923 struct cgroup_subsys_state css;
924 struct perf_cgroup_info __percpu *info;
925 };
926
927 /*
928 * Must ensure cgroup is pinned (css_get) before calling
929 * this function. In other words, we cannot call this function
930 * if there is no cgroup event for the current CPU context.
931 */
932 static inline struct perf_cgroup *
perf_cgroup_from_task(struct task_struct * task,struct perf_event_context * ctx)933 perf_cgroup_from_task(struct task_struct *task, struct perf_event_context *ctx)
934 {
935 return container_of(task_css_check(task, perf_event_cgrp_id,
936 ctx ? lockdep_is_held(&ctx->lock)
937 : true),
938 struct perf_cgroup, css);
939 }
940 #endif /* CONFIG_CGROUP_PERF */
941
942 #ifdef CONFIG_PERF_EVENTS
943
944 extern void *perf_aux_output_begin(struct perf_output_handle *handle,
945 struct perf_event *event);
946 extern void perf_aux_output_end(struct perf_output_handle *handle,
947 unsigned long size);
948 extern int perf_aux_output_skip(struct perf_output_handle *handle,
949 unsigned long size);
950 extern void *perf_get_aux(struct perf_output_handle *handle);
951 extern void perf_aux_output_flag(struct perf_output_handle *handle, u64 flags);
952 extern void perf_event_itrace_started(struct perf_event *event);
953
954 extern int perf_pmu_register(struct pmu *pmu, const char *name, int type);
955 extern void perf_pmu_unregister(struct pmu *pmu);
956
957 extern void __perf_event_task_sched_in(struct task_struct *prev,
958 struct task_struct *task);
959 extern void __perf_event_task_sched_out(struct task_struct *prev,
960 struct task_struct *next);
961 extern int perf_event_init_task(struct task_struct *child, u64 clone_flags);
962 extern void perf_event_exit_task(struct task_struct *child);
963 extern void perf_event_free_task(struct task_struct *task);
964 extern void perf_event_delayed_put(struct task_struct *task);
965 extern struct file *perf_event_get(unsigned int fd);
966 extern const struct perf_event *perf_get_event(struct file *file);
967 extern const struct perf_event_attr *perf_event_attrs(struct perf_event *event);
968 extern void perf_event_print_debug(void);
969 extern void perf_pmu_disable(struct pmu *pmu);
970 extern void perf_pmu_enable(struct pmu *pmu);
971 extern void perf_sched_cb_dec(struct pmu *pmu);
972 extern void perf_sched_cb_inc(struct pmu *pmu);
973 extern int perf_event_task_disable(void);
974 extern int perf_event_task_enable(void);
975
976 extern void perf_pmu_resched(struct pmu *pmu);
977
978 extern int perf_event_refresh(struct perf_event *event, int refresh);
979 extern void perf_event_update_userpage(struct perf_event *event);
980 extern int perf_event_release_kernel(struct perf_event *event);
981 extern struct perf_event *
982 perf_event_create_kernel_counter(struct perf_event_attr *attr,
983 int cpu,
984 struct task_struct *task,
985 perf_overflow_handler_t callback,
986 void *context);
987 extern void perf_pmu_migrate_context(struct pmu *pmu,
988 int src_cpu, int dst_cpu);
989 int perf_event_read_local(struct perf_event *event, u64 *value,
990 u64 *enabled, u64 *running);
991 extern u64 perf_event_read_value(struct perf_event *event,
992 u64 *enabled, u64 *running);
993
994
995 struct perf_sample_data {
996 /*
997 * Fields set by perf_sample_data_init(), group so as to
998 * minimize the cachelines touched.
999 */
1000 u64 addr;
1001 struct perf_raw_record *raw;
1002 struct perf_branch_stack *br_stack;
1003 u64 period;
1004 union perf_sample_weight weight;
1005 u64 txn;
1006 union perf_mem_data_src data_src;
1007
1008 /*
1009 * The other fields, optionally {set,used} by
1010 * perf_{prepare,output}_sample().
1011 */
1012 u64 type;
1013 u64 ip;
1014 struct {
1015 u32 pid;
1016 u32 tid;
1017 } tid_entry;
1018 u64 time;
1019 u64 id;
1020 u64 stream_id;
1021 struct {
1022 u32 cpu;
1023 u32 reserved;
1024 } cpu_entry;
1025 struct perf_callchain_entry *callchain;
1026 u64 aux_size;
1027
1028 struct perf_regs regs_user;
1029 struct perf_regs regs_intr;
1030 u64 stack_user_size;
1031
1032 u64 phys_addr;
1033 u64 cgroup;
1034 u64 data_page_size;
1035 u64 code_page_size;
1036 } ____cacheline_aligned;
1037
1038 /* default value for data source */
1039 #define PERF_MEM_NA (PERF_MEM_S(OP, NA) |\
1040 PERF_MEM_S(LVL, NA) |\
1041 PERF_MEM_S(SNOOP, NA) |\
1042 PERF_MEM_S(LOCK, NA) |\
1043 PERF_MEM_S(TLB, NA))
1044
perf_sample_data_init(struct perf_sample_data * data,u64 addr,u64 period)1045 static inline void perf_sample_data_init(struct perf_sample_data *data,
1046 u64 addr, u64 period)
1047 {
1048 /* remaining struct members initialized in perf_prepare_sample() */
1049 data->addr = addr;
1050 data->raw = NULL;
1051 data->br_stack = NULL;
1052 data->period = period;
1053 data->weight.full = 0;
1054 data->data_src.val = PERF_MEM_NA;
1055 data->txn = 0;
1056 }
1057
1058 extern void perf_output_sample(struct perf_output_handle *handle,
1059 struct perf_event_header *header,
1060 struct perf_sample_data *data,
1061 struct perf_event *event);
1062 extern void perf_prepare_sample(struct perf_event_header *header,
1063 struct perf_sample_data *data,
1064 struct perf_event *event,
1065 struct pt_regs *regs);
1066
1067 extern int perf_event_overflow(struct perf_event *event,
1068 struct perf_sample_data *data,
1069 struct pt_regs *regs);
1070
1071 extern void perf_event_output_forward(struct perf_event *event,
1072 struct perf_sample_data *data,
1073 struct pt_regs *regs);
1074 extern void perf_event_output_backward(struct perf_event *event,
1075 struct perf_sample_data *data,
1076 struct pt_regs *regs);
1077 extern int perf_event_output(struct perf_event *event,
1078 struct perf_sample_data *data,
1079 struct pt_regs *regs);
1080
1081 static inline bool
is_default_overflow_handler(struct perf_event * event)1082 is_default_overflow_handler(struct perf_event *event)
1083 {
1084 if (likely(event->overflow_handler == perf_event_output_forward))
1085 return true;
1086 if (unlikely(event->overflow_handler == perf_event_output_backward))
1087 return true;
1088 return false;
1089 }
1090
1091 extern void
1092 perf_event_header__init_id(struct perf_event_header *header,
1093 struct perf_sample_data *data,
1094 struct perf_event *event);
1095 extern void
1096 perf_event__output_id_sample(struct perf_event *event,
1097 struct perf_output_handle *handle,
1098 struct perf_sample_data *sample);
1099
1100 extern void
1101 perf_log_lost_samples(struct perf_event *event, u64 lost);
1102
event_has_any_exclude_flag(struct perf_event * event)1103 static inline bool event_has_any_exclude_flag(struct perf_event *event)
1104 {
1105 struct perf_event_attr *attr = &event->attr;
1106
1107 return attr->exclude_idle || attr->exclude_user ||
1108 attr->exclude_kernel || attr->exclude_hv ||
1109 attr->exclude_guest || attr->exclude_host;
1110 }
1111
is_sampling_event(struct perf_event * event)1112 static inline bool is_sampling_event(struct perf_event *event)
1113 {
1114 return event->attr.sample_period != 0;
1115 }
1116
1117 /*
1118 * Return 1 for a software event, 0 for a hardware event
1119 */
is_software_event(struct perf_event * event)1120 static inline int is_software_event(struct perf_event *event)
1121 {
1122 return event->event_caps & PERF_EV_CAP_SOFTWARE;
1123 }
1124
1125 /*
1126 * Return 1 for event in sw context, 0 for event in hw context
1127 */
in_software_context(struct perf_event * event)1128 static inline int in_software_context(struct perf_event *event)
1129 {
1130 return event->ctx->pmu->task_ctx_nr == perf_sw_context;
1131 }
1132
is_exclusive_pmu(struct pmu * pmu)1133 static inline int is_exclusive_pmu(struct pmu *pmu)
1134 {
1135 return pmu->capabilities & PERF_PMU_CAP_EXCLUSIVE;
1136 }
1137
1138 extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX];
1139
1140 extern void ___perf_sw_event(u32, u64, struct pt_regs *, u64);
1141 extern void __perf_sw_event(u32, u64, struct pt_regs *, u64);
1142
1143 #ifndef perf_arch_fetch_caller_regs
perf_arch_fetch_caller_regs(struct pt_regs * regs,unsigned long ip)1144 static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { }
1145 #endif
1146
1147 /*
1148 * When generating a perf sample in-line, instead of from an interrupt /
1149 * exception, we lack a pt_regs. This is typically used from software events
1150 * like: SW_CONTEXT_SWITCHES, SW_MIGRATIONS and the tie-in with tracepoints.
1151 *
1152 * We typically don't need a full set, but (for x86) do require:
1153 * - ip for PERF_SAMPLE_IP
1154 * - cs for user_mode() tests
1155 * - sp for PERF_SAMPLE_CALLCHAIN
1156 * - eflags for MISC bits and CALLCHAIN (see: perf_hw_regs())
1157 *
1158 * NOTE: assumes @regs is otherwise already 0 filled; this is important for
1159 * things like PERF_SAMPLE_REGS_INTR.
1160 */
perf_fetch_caller_regs(struct pt_regs * regs)1161 static inline void perf_fetch_caller_regs(struct pt_regs *regs)
1162 {
1163 perf_arch_fetch_caller_regs(regs, CALLER_ADDR0);
1164 }
1165
1166 static __always_inline void
perf_sw_event(u32 event_id,u64 nr,struct pt_regs * regs,u64 addr)1167 perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)
1168 {
1169 if (static_key_false(&perf_swevent_enabled[event_id]))
1170 __perf_sw_event(event_id, nr, regs, addr);
1171 }
1172
1173 DECLARE_PER_CPU(struct pt_regs, __perf_regs[4]);
1174
1175 /*
1176 * 'Special' version for the scheduler, it hard assumes no recursion,
1177 * which is guaranteed by us not actually scheduling inside other swevents
1178 * because those disable preemption.
1179 */
__perf_sw_event_sched(u32 event_id,u64 nr,u64 addr)1180 static __always_inline void __perf_sw_event_sched(u32 event_id, u64 nr, u64 addr)
1181 {
1182 struct pt_regs *regs = this_cpu_ptr(&__perf_regs[0]);
1183
1184 perf_fetch_caller_regs(regs);
1185 ___perf_sw_event(event_id, nr, regs, addr);
1186 }
1187
1188 extern struct static_key_false perf_sched_events;
1189
__perf_sw_enabled(int swevt)1190 static __always_inline bool __perf_sw_enabled(int swevt)
1191 {
1192 return static_key_false(&perf_swevent_enabled[swevt]);
1193 }
1194
perf_event_task_migrate(struct task_struct * task)1195 static inline void perf_event_task_migrate(struct task_struct *task)
1196 {
1197 if (__perf_sw_enabled(PERF_COUNT_SW_CPU_MIGRATIONS))
1198 task->sched_migrated = 1;
1199 }
1200
perf_event_task_sched_in(struct task_struct * prev,struct task_struct * task)1201 static inline void perf_event_task_sched_in(struct task_struct *prev,
1202 struct task_struct *task)
1203 {
1204 if (static_branch_unlikely(&perf_sched_events))
1205 __perf_event_task_sched_in(prev, task);
1206
1207 if (__perf_sw_enabled(PERF_COUNT_SW_CPU_MIGRATIONS) &&
1208 task->sched_migrated) {
1209 __perf_sw_event_sched(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 0);
1210 task->sched_migrated = 0;
1211 }
1212 }
1213
perf_event_task_sched_out(struct task_struct * prev,struct task_struct * next)1214 static inline void perf_event_task_sched_out(struct task_struct *prev,
1215 struct task_struct *next)
1216 {
1217 if (__perf_sw_enabled(PERF_COUNT_SW_CONTEXT_SWITCHES))
1218 __perf_sw_event_sched(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 0);
1219
1220 #ifdef CONFIG_CGROUP_PERF
1221 if (__perf_sw_enabled(PERF_COUNT_SW_CGROUP_SWITCHES) &&
1222 perf_cgroup_from_task(prev, NULL) !=
1223 perf_cgroup_from_task(next, NULL))
1224 __perf_sw_event_sched(PERF_COUNT_SW_CGROUP_SWITCHES, 1, 0);
1225 #endif
1226
1227 if (static_branch_unlikely(&perf_sched_events))
1228 __perf_event_task_sched_out(prev, next);
1229 }
1230
1231 extern void perf_event_mmap(struct vm_area_struct *vma);
1232
1233 extern void perf_event_ksymbol(u16 ksym_type, u64 addr, u32 len,
1234 bool unregister, const char *sym);
1235 extern void perf_event_bpf_event(struct bpf_prog *prog,
1236 enum perf_bpf_event_type type,
1237 u16 flags);
1238
1239 extern struct perf_guest_info_callbacks *perf_guest_cbs;
1240 extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks);
1241 extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks);
1242
1243 extern void perf_event_exec(void);
1244 extern void perf_event_comm(struct task_struct *tsk, bool exec);
1245 extern void perf_event_namespaces(struct task_struct *tsk);
1246 extern void perf_event_fork(struct task_struct *tsk);
1247 extern void perf_event_text_poke(const void *addr,
1248 const void *old_bytes, size_t old_len,
1249 const void *new_bytes, size_t new_len);
1250
1251 /* Callchains */
1252 DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry);
1253
1254 extern void perf_callchain_user(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs);
1255 extern void perf_callchain_kernel(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs);
1256 extern struct perf_callchain_entry *
1257 get_perf_callchain(struct pt_regs *regs, u32 init_nr, bool kernel, bool user,
1258 u32 max_stack, bool crosstask, bool add_mark);
1259 extern struct perf_callchain_entry *perf_callchain(struct perf_event *event, struct pt_regs *regs);
1260 extern int get_callchain_buffers(int max_stack);
1261 extern void put_callchain_buffers(void);
1262 extern struct perf_callchain_entry *get_callchain_entry(int *rctx);
1263 extern void put_callchain_entry(int rctx);
1264
1265 extern int sysctl_perf_event_max_stack;
1266 extern int sysctl_perf_event_max_contexts_per_stack;
1267
perf_callchain_store_context(struct perf_callchain_entry_ctx * ctx,u64 ip)1268 static inline int perf_callchain_store_context(struct perf_callchain_entry_ctx *ctx, u64 ip)
1269 {
1270 if (ctx->contexts < sysctl_perf_event_max_contexts_per_stack) {
1271 struct perf_callchain_entry *entry = ctx->entry;
1272 entry->ip[entry->nr++] = ip;
1273 ++ctx->contexts;
1274 return 0;
1275 } else {
1276 ctx->contexts_maxed = true;
1277 return -1; /* no more room, stop walking the stack */
1278 }
1279 }
1280
perf_callchain_store(struct perf_callchain_entry_ctx * ctx,u64 ip)1281 static inline int perf_callchain_store(struct perf_callchain_entry_ctx *ctx, u64 ip)
1282 {
1283 if (ctx->nr < ctx->max_stack && !ctx->contexts_maxed) {
1284 struct perf_callchain_entry *entry = ctx->entry;
1285 entry->ip[entry->nr++] = ip;
1286 ++ctx->nr;
1287 return 0;
1288 } else {
1289 return -1; /* no more room, stop walking the stack */
1290 }
1291 }
1292
1293 extern int sysctl_perf_event_paranoid;
1294 extern int sysctl_perf_event_mlock;
1295 extern int sysctl_perf_event_sample_rate;
1296 extern int sysctl_perf_cpu_time_max_percent;
1297
1298 extern void perf_sample_event_took(u64 sample_len_ns);
1299
1300 int perf_proc_update_handler(struct ctl_table *table, int write,
1301 void *buffer, size_t *lenp, loff_t *ppos);
1302 int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write,
1303 void *buffer, size_t *lenp, loff_t *ppos);
1304 int perf_event_max_stack_handler(struct ctl_table *table, int write,
1305 void *buffer, size_t *lenp, loff_t *ppos);
1306
1307 /* Access to perf_event_open(2) syscall. */
1308 #define PERF_SECURITY_OPEN 0
1309
1310 /* Finer grained perf_event_open(2) access control. */
1311 #define PERF_SECURITY_CPU 1
1312 #define PERF_SECURITY_KERNEL 2
1313 #define PERF_SECURITY_TRACEPOINT 3
1314
perf_is_paranoid(void)1315 static inline int perf_is_paranoid(void)
1316 {
1317 return sysctl_perf_event_paranoid > -1;
1318 }
1319
perf_allow_kernel(struct perf_event_attr * attr)1320 static inline int perf_allow_kernel(struct perf_event_attr *attr)
1321 {
1322 if (sysctl_perf_event_paranoid > 1 && !perfmon_capable())
1323 return -EACCES;
1324
1325 return security_perf_event_open(attr, PERF_SECURITY_KERNEL);
1326 }
1327
perf_allow_cpu(struct perf_event_attr * attr)1328 static inline int perf_allow_cpu(struct perf_event_attr *attr)
1329 {
1330 if (sysctl_perf_event_paranoid > 0 && !perfmon_capable())
1331 return -EACCES;
1332
1333 return security_perf_event_open(attr, PERF_SECURITY_CPU);
1334 }
1335
perf_allow_tracepoint(struct perf_event_attr * attr)1336 static inline int perf_allow_tracepoint(struct perf_event_attr *attr)
1337 {
1338 if (sysctl_perf_event_paranoid > -1 && !perfmon_capable())
1339 return -EPERM;
1340
1341 return security_perf_event_open(attr, PERF_SECURITY_TRACEPOINT);
1342 }
1343
1344 extern void perf_event_init(void);
1345 extern void perf_tp_event(u16 event_type, u64 count, void *record,
1346 int entry_size, struct pt_regs *regs,
1347 struct hlist_head *head, int rctx,
1348 struct task_struct *task);
1349 extern void perf_bp_event(struct perf_event *event, void *data);
1350
1351 #ifndef perf_misc_flags
1352 # define perf_misc_flags(regs) \
1353 (user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL)
1354 # define perf_instruction_pointer(regs) instruction_pointer(regs)
1355 #endif
1356 #ifndef perf_arch_bpf_user_pt_regs
1357 # define perf_arch_bpf_user_pt_regs(regs) regs
1358 #endif
1359
has_branch_stack(struct perf_event * event)1360 static inline bool has_branch_stack(struct perf_event *event)
1361 {
1362 return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK;
1363 }
1364
needs_branch_stack(struct perf_event * event)1365 static inline bool needs_branch_stack(struct perf_event *event)
1366 {
1367 return event->attr.branch_sample_type != 0;
1368 }
1369
has_aux(struct perf_event * event)1370 static inline bool has_aux(struct perf_event *event)
1371 {
1372 return event->pmu->setup_aux;
1373 }
1374
is_write_backward(struct perf_event * event)1375 static inline bool is_write_backward(struct perf_event *event)
1376 {
1377 return !!event->attr.write_backward;
1378 }
1379
has_addr_filter(struct perf_event * event)1380 static inline bool has_addr_filter(struct perf_event *event)
1381 {
1382 return event->pmu->nr_addr_filters;
1383 }
1384
1385 /*
1386 * An inherited event uses parent's filters
1387 */
1388 static inline struct perf_addr_filters_head *
perf_event_addr_filters(struct perf_event * event)1389 perf_event_addr_filters(struct perf_event *event)
1390 {
1391 struct perf_addr_filters_head *ifh = &event->addr_filters;
1392
1393 if (event->parent)
1394 ifh = &event->parent->addr_filters;
1395
1396 return ifh;
1397 }
1398
1399 extern void perf_event_addr_filters_sync(struct perf_event *event);
1400
1401 extern int perf_output_begin(struct perf_output_handle *handle,
1402 struct perf_sample_data *data,
1403 struct perf_event *event, unsigned int size);
1404 extern int perf_output_begin_forward(struct perf_output_handle *handle,
1405 struct perf_sample_data *data,
1406 struct perf_event *event,
1407 unsigned int size);
1408 extern int perf_output_begin_backward(struct perf_output_handle *handle,
1409 struct perf_sample_data *data,
1410 struct perf_event *event,
1411 unsigned int size);
1412
1413 extern void perf_output_end(struct perf_output_handle *handle);
1414 extern unsigned int perf_output_copy(struct perf_output_handle *handle,
1415 const void *buf, unsigned int len);
1416 extern unsigned int perf_output_skip(struct perf_output_handle *handle,
1417 unsigned int len);
1418 extern long perf_output_copy_aux(struct perf_output_handle *aux_handle,
1419 struct perf_output_handle *handle,
1420 unsigned long from, unsigned long to);
1421 extern int perf_swevent_get_recursion_context(void);
1422 extern void perf_swevent_put_recursion_context(int rctx);
1423 extern u64 perf_swevent_set_period(struct perf_event *event);
1424 extern void perf_event_enable(struct perf_event *event);
1425 extern void perf_event_disable(struct perf_event *event);
1426 extern void perf_event_disable_local(struct perf_event *event);
1427 extern void perf_event_disable_inatomic(struct perf_event *event);
1428 extern void perf_event_task_tick(void);
1429 extern int perf_event_account_interrupt(struct perf_event *event);
1430 extern int perf_event_period(struct perf_event *event, u64 value);
1431 extern u64 perf_event_pause(struct perf_event *event, bool reset);
1432 #else /* !CONFIG_PERF_EVENTS: */
1433 static inline void *
perf_aux_output_begin(struct perf_output_handle * handle,struct perf_event * event)1434 perf_aux_output_begin(struct perf_output_handle *handle,
1435 struct perf_event *event) { return NULL; }
1436 static inline void
perf_aux_output_end(struct perf_output_handle * handle,unsigned long size)1437 perf_aux_output_end(struct perf_output_handle *handle, unsigned long size)
1438 { }
1439 static inline int
perf_aux_output_skip(struct perf_output_handle * handle,unsigned long size)1440 perf_aux_output_skip(struct perf_output_handle *handle,
1441 unsigned long size) { return -EINVAL; }
1442 static inline void *
perf_get_aux(struct perf_output_handle * handle)1443 perf_get_aux(struct perf_output_handle *handle) { return NULL; }
1444 static inline void
perf_event_task_migrate(struct task_struct * task)1445 perf_event_task_migrate(struct task_struct *task) { }
1446 static inline void
perf_event_task_sched_in(struct task_struct * prev,struct task_struct * task)1447 perf_event_task_sched_in(struct task_struct *prev,
1448 struct task_struct *task) { }
1449 static inline void
perf_event_task_sched_out(struct task_struct * prev,struct task_struct * next)1450 perf_event_task_sched_out(struct task_struct *prev,
1451 struct task_struct *next) { }
perf_event_init_task(struct task_struct * child,u64 clone_flags)1452 static inline int perf_event_init_task(struct task_struct *child,
1453 u64 clone_flags) { return 0; }
perf_event_exit_task(struct task_struct * child)1454 static inline void perf_event_exit_task(struct task_struct *child) { }
perf_event_free_task(struct task_struct * task)1455 static inline void perf_event_free_task(struct task_struct *task) { }
perf_event_delayed_put(struct task_struct * task)1456 static inline void perf_event_delayed_put(struct task_struct *task) { }
perf_event_get(unsigned int fd)1457 static inline struct file *perf_event_get(unsigned int fd) { return ERR_PTR(-EINVAL); }
perf_get_event(struct file * file)1458 static inline const struct perf_event *perf_get_event(struct file *file)
1459 {
1460 return ERR_PTR(-EINVAL);
1461 }
perf_event_attrs(struct perf_event * event)1462 static inline const struct perf_event_attr *perf_event_attrs(struct perf_event *event)
1463 {
1464 return ERR_PTR(-EINVAL);
1465 }
perf_event_read_local(struct perf_event * event,u64 * value,u64 * enabled,u64 * running)1466 static inline int perf_event_read_local(struct perf_event *event, u64 *value,
1467 u64 *enabled, u64 *running)
1468 {
1469 return -EINVAL;
1470 }
perf_event_print_debug(void)1471 static inline void perf_event_print_debug(void) { }
perf_event_task_disable(void)1472 static inline int perf_event_task_disable(void) { return -EINVAL; }
perf_event_task_enable(void)1473 static inline int perf_event_task_enable(void) { return -EINVAL; }
perf_event_refresh(struct perf_event * event,int refresh)1474 static inline int perf_event_refresh(struct perf_event *event, int refresh)
1475 {
1476 return -EINVAL;
1477 }
1478
1479 static inline void
perf_sw_event(u32 event_id,u64 nr,struct pt_regs * regs,u64 addr)1480 perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) { }
1481 static inline void
perf_bp_event(struct perf_event * event,void * data)1482 perf_bp_event(struct perf_event *event, void *data) { }
1483
perf_register_guest_info_callbacks(struct perf_guest_info_callbacks * callbacks)1484 static inline int perf_register_guest_info_callbacks
1485 (struct perf_guest_info_callbacks *callbacks) { return 0; }
perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks * callbacks)1486 static inline int perf_unregister_guest_info_callbacks
1487 (struct perf_guest_info_callbacks *callbacks) { return 0; }
1488
perf_event_mmap(struct vm_area_struct * vma)1489 static inline void perf_event_mmap(struct vm_area_struct *vma) { }
1490
1491 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)1492 static inline void perf_event_ksymbol(u16 ksym_type, u64 addr, u32 len,
1493 bool unregister, const char *sym) { }
perf_event_bpf_event(struct bpf_prog * prog,enum perf_bpf_event_type type,u16 flags)1494 static inline void perf_event_bpf_event(struct bpf_prog *prog,
1495 enum perf_bpf_event_type type,
1496 u16 flags) { }
perf_event_exec(void)1497 static inline void perf_event_exec(void) { }
perf_event_comm(struct task_struct * tsk,bool exec)1498 static inline void perf_event_comm(struct task_struct *tsk, bool exec) { }
perf_event_namespaces(struct task_struct * tsk)1499 static inline void perf_event_namespaces(struct task_struct *tsk) { }
perf_event_fork(struct task_struct * tsk)1500 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)1501 static inline void perf_event_text_poke(const void *addr,
1502 const void *old_bytes,
1503 size_t old_len,
1504 const void *new_bytes,
1505 size_t new_len) { }
perf_event_init(void)1506 static inline void perf_event_init(void) { }
perf_swevent_get_recursion_context(void)1507 static inline int perf_swevent_get_recursion_context(void) { return -1; }
perf_swevent_put_recursion_context(int rctx)1508 static inline void perf_swevent_put_recursion_context(int rctx) { }
perf_swevent_set_period(struct perf_event * event)1509 static inline u64 perf_swevent_set_period(struct perf_event *event) { return 0; }
perf_event_enable(struct perf_event * event)1510 static inline void perf_event_enable(struct perf_event *event) { }
perf_event_disable(struct perf_event * event)1511 static inline void perf_event_disable(struct perf_event *event) { }
__perf_event_disable(void * info)1512 static inline int __perf_event_disable(void *info) { return -1; }
perf_event_task_tick(void)1513 static inline void perf_event_task_tick(void) { }
perf_event_release_kernel(struct perf_event * event)1514 static inline int perf_event_release_kernel(struct perf_event *event) { return 0; }
perf_event_period(struct perf_event * event,u64 value)1515 static inline int perf_event_period(struct perf_event *event, u64 value)
1516 {
1517 return -EINVAL;
1518 }
perf_event_pause(struct perf_event * event,bool reset)1519 static inline u64 perf_event_pause(struct perf_event *event, bool reset)
1520 {
1521 return 0;
1522 }
1523 #endif
1524
1525 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
1526 extern void perf_restore_debug_store(void);
1527 #else
perf_restore_debug_store(void)1528 static inline void perf_restore_debug_store(void) { }
1529 #endif
1530
perf_raw_frag_last(const struct perf_raw_frag * frag)1531 static __always_inline bool perf_raw_frag_last(const struct perf_raw_frag *frag)
1532 {
1533 return frag->pad < sizeof(u64);
1534 }
1535
1536 #define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x))
1537
1538 struct perf_pmu_events_attr {
1539 struct device_attribute attr;
1540 u64 id;
1541 const char *event_str;
1542 };
1543
1544 struct perf_pmu_events_ht_attr {
1545 struct device_attribute attr;
1546 u64 id;
1547 const char *event_str_ht;
1548 const char *event_str_noht;
1549 };
1550
1551 struct perf_pmu_events_hybrid_attr {
1552 struct device_attribute attr;
1553 u64 id;
1554 const char *event_str;
1555 u64 pmu_type;
1556 };
1557
1558 struct perf_pmu_format_hybrid_attr {
1559 struct device_attribute attr;
1560 u64 pmu_type;
1561 };
1562
1563 ssize_t perf_event_sysfs_show(struct device *dev, struct device_attribute *attr,
1564 char *page);
1565
1566 #define PMU_EVENT_ATTR(_name, _var, _id, _show) \
1567 static struct perf_pmu_events_attr _var = { \
1568 .attr = __ATTR(_name, 0444, _show, NULL), \
1569 .id = _id, \
1570 };
1571
1572 #define PMU_EVENT_ATTR_STRING(_name, _var, _str) \
1573 static struct perf_pmu_events_attr _var = { \
1574 .attr = __ATTR(_name, 0444, perf_event_sysfs_show, NULL), \
1575 .id = 0, \
1576 .event_str = _str, \
1577 };
1578
1579 #define PMU_FORMAT_ATTR(_name, _format) \
1580 static ssize_t \
1581 _name##_show(struct device *dev, \
1582 struct device_attribute *attr, \
1583 char *page) \
1584 { \
1585 BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE); \
1586 return sprintf(page, _format "\n"); \
1587 } \
1588 \
1589 static struct device_attribute format_attr_##_name = __ATTR_RO(_name)
1590
1591 /* Performance counter hotplug functions */
1592 #ifdef CONFIG_PERF_EVENTS
1593 int perf_event_init_cpu(unsigned int cpu);
1594 int perf_event_exit_cpu(unsigned int cpu);
1595 #else
1596 #define perf_event_init_cpu NULL
1597 #define perf_event_exit_cpu NULL
1598 #endif
1599
1600 extern void __weak arch_perf_update_userpage(struct perf_event *event,
1601 struct perf_event_mmap_page *userpg,
1602 u64 now);
1603
1604 #ifdef CONFIG_MMU
1605 extern __weak u64 arch_perf_get_page_size(struct mm_struct *mm, unsigned long addr);
1606 #endif
1607
1608 #endif /* _LINUX_PERF_EVENT_H */
1609