~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/include/linux/perf_event.h

Version: ~ [ linux-5.17-rc1 ] ~ [ linux-5.16.2 ] ~ [ linux-5.15.16 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.93 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.173 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.225 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.262 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.297 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.299 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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               0x01
264 #define PERF_PMU_CAP_NO_NMI                     0x02
265 #define PERF_PMU_CAP_AUX_NO_SG                  0x04
266 #define PERF_PMU_CAP_EXTENDED_REGS              0x08
267 #define PERF_PMU_CAP_EXCLUSIVE                  0x10
268 #define PERF_PMU_CAP_ITRACE                     0x20
269 #define PERF_PMU_CAP_HETEROGENEOUS_CPUS         0x40
270 #define PERF_PMU_CAP_NO_EXCLUDE                 0x80
271 #define PERF_PMU_CAP_AUX_OUTPUT                 0x100
272 
273 struct perf_output_handle;
274 
275 /**
276  * struct pmu - generic performance monitoring unit
277  */
278 struct pmu {
279         struct list_head                entry;
280 
281         struct module                   *module;
282         struct device                   *dev;
283         const struct attribute_group    **attr_groups;
284         const struct attribute_group    **attr_update;
285         const char                      *name;
286         int                             type;
287 
288         /*
289          * various common per-pmu feature flags
290          */
291         int                             capabilities;
292 
293         int __percpu                    *pmu_disable_count;
294         struct perf_cpu_context __percpu *pmu_cpu_context;
295         atomic_t                        exclusive_cnt; /* < 0: cpu; > 0: tsk */
296         int                             task_ctx_nr;
297         int                             hrtimer_interval_ms;
298 
299         /* number of address filters this PMU can do */
300         unsigned int                    nr_addr_filters;
301 
302         /*
303          * Fully disable/enable this PMU, can be used to protect from the PMI
304          * as well as for lazy/batch writing of the MSRs.
305          */
306         void (*pmu_enable)              (struct pmu *pmu); /* optional */
307         void (*pmu_disable)             (struct pmu *pmu); /* optional */
308 
309         /*
310          * Try and initialize the event for this PMU.
311          *
312          * Returns:
313          *  -ENOENT     -- @event is not for this PMU
314          *
315          *  -ENODEV     -- @event is for this PMU but PMU not present
316          *  -EBUSY      -- @event is for this PMU but PMU temporarily unavailable
317          *  -EINVAL     -- @event is for this PMU but @event is not valid
318          *  -EOPNOTSUPP -- @event is for this PMU, @event is valid, but not supported
319          *  -EACCES     -- @event is for this PMU, @event is valid, but no privileges
320          *
321          *  0           -- @event is for this PMU and valid
322          *
323          * Other error return values are allowed.
324          */
325         int (*event_init)               (struct perf_event *event);
326 
327         /*
328          * Notification that the event was mapped or unmapped.  Called
329          * in the context of the mapping task.
330          */
331         void (*event_mapped)            (struct perf_event *event, struct mm_struct *mm); /* optional */
332         void (*event_unmapped)          (struct perf_event *event, struct mm_struct *mm); /* optional */
333 
334         /*
335          * Flags for ->add()/->del()/ ->start()/->stop(). There are
336          * matching hw_perf_event::state flags.
337          */
338 #define PERF_EF_START   0x01            /* start the counter when adding    */
339 #define PERF_EF_RELOAD  0x02            /* reload the counter when starting */
340 #define PERF_EF_UPDATE  0x04            /* update the counter when stopping */
341 
342         /*
343          * Adds/Removes a counter to/from the PMU, can be done inside a
344          * transaction, see the ->*_txn() methods.
345          *
346          * The add/del callbacks will reserve all hardware resources required
347          * to service the event, this includes any counter constraint
348          * scheduling etc.
349          *
350          * Called with IRQs disabled and the PMU disabled on the CPU the event
351          * is on.
352          *
353          * ->add() called without PERF_EF_START should result in the same state
354          *  as ->add() followed by ->stop().
355          *
356          * ->del() must always PERF_EF_UPDATE stop an event. If it calls
357          *  ->stop() that must deal with already being stopped without
358          *  PERF_EF_UPDATE.
359          */
360         int  (*add)                     (struct perf_event *event, int flags);
361         void (*del)                     (struct perf_event *event, int flags);
362 
363         /*
364          * Starts/Stops a counter present on the PMU.
365          *
366          * The PMI handler should stop the counter when perf_event_overflow()
367          * returns !0. ->start() will be used to continue.
368          *
369          * Also used to change the sample period.
370          *
371          * Called with IRQs disabled and the PMU disabled on the CPU the event
372          * is on -- will be called from NMI context with the PMU generates
373          * NMIs.
374          *
375          * ->stop() with PERF_EF_UPDATE will read the counter and update
376          *  period/count values like ->read() would.
377          *
378          * ->start() with PERF_EF_RELOAD will reprogram the counter
379          *  value, must be preceded by a ->stop() with PERF_EF_UPDATE.
380          */
381         void (*start)                   (struct perf_event *event, int flags);
382         void (*stop)                    (struct perf_event *event, int flags);
383 
384         /*
385          * Updates the counter value of the event.
386          *
387          * For sampling capable PMUs this will also update the software period
388          * hw_perf_event::period_left field.
389          */
390         void (*read)                    (struct perf_event *event);
391 
392         /*
393          * Group events scheduling is treated as a transaction, add
394          * group events as a whole and perform one schedulability test.
395          * If the test fails, roll back the whole group
396          *
397          * Start the transaction, after this ->add() doesn't need to
398          * do schedulability tests.
399          *
400          * Optional.
401          */
402         void (*start_txn)               (struct pmu *pmu, unsigned int txn_flags);
403         /*
404          * If ->start_txn() disabled the ->add() schedulability test
405          * then ->commit_txn() is required to perform one. On success
406          * the transaction is closed. On error the transaction is kept
407          * open until ->cancel_txn() is called.
408          *
409          * Optional.
410          */
411         int  (*commit_txn)              (struct pmu *pmu);
412         /*
413          * Will cancel the transaction, assumes ->del() is called
414          * for each successful ->add() during the transaction.
415          *
416          * Optional.
417          */
418         void (*cancel_txn)              (struct pmu *pmu);
419 
420         /*
421          * Will return the value for perf_event_mmap_page::index for this event,
422          * if no implementation is provided it will default to: event->hw.idx + 1.
423          */
424         int (*event_idx)                (struct perf_event *event); /*optional */
425 
426         /*
427          * context-switches callback
428          */
429         void (*sched_task)              (struct perf_event_context *ctx,
430                                         bool sched_in);
431 
432         /*
433          * Kmem cache of PMU specific data
434          */
435         struct kmem_cache               *task_ctx_cache;
436 
437         /*
438          * PMU specific parts of task perf event context (i.e. ctx->task_ctx_data)
439          * can be synchronized using this function. See Intel LBR callstack support
440          * implementation and Perf core context switch handling callbacks for usage
441          * examples.
442          */
443         void (*swap_task_ctx)           (struct perf_event_context *prev,
444                                          struct perf_event_context *next);
445                                         /* optional */
446 
447         /*
448          * Set up pmu-private data structures for an AUX area
449          */
450         void *(*setup_aux)              (struct perf_event *event, void **pages,
451                                          int nr_pages, bool overwrite);
452                                         /* optional */
453 
454         /*
455          * Free pmu-private AUX data structures
456          */
457         void (*free_aux)                (void *aux); /* optional */
458 
459         /*
460          * Take a snapshot of the AUX buffer without touching the event
461          * state, so that preempting ->start()/->stop() callbacks does
462          * not interfere with their logic. Called in PMI context.
463          *
464          * Returns the size of AUX data copied to the output handle.
465          *
466          * Optional.
467          */
468         long (*snapshot_aux)            (struct perf_event *event,
469                                          struct perf_output_handle *handle,
470                                          unsigned long size);
471 
472         /*
473          * Validate address range filters: make sure the HW supports the
474          * requested configuration and number of filters; return 0 if the
475          * supplied filters are valid, -errno otherwise.
476          *
477          * Runs in the context of the ioctl()ing process and is not serialized
478          * with the rest of the PMU callbacks.
479          */
480         int (*addr_filters_validate)    (struct list_head *filters);
481                                         /* optional */
482 
483         /*
484          * Synchronize address range filter configuration:
485          * translate hw-agnostic filters into hardware configuration in
486          * event::hw::addr_filters.
487          *
488          * Runs as a part of filter sync sequence that is done in ->start()
489          * callback by calling perf_event_addr_filters_sync().
490          *
491          * May (and should) traverse event::addr_filters::list, for which its
492          * caller provides necessary serialization.
493          */
494         void (*addr_filters_sync)       (struct perf_event *event);
495                                         /* optional */
496 
497         /*
498          * Check if event can be used for aux_output purposes for
499          * events of this PMU.
500          *
501          * Runs from perf_event_open(). Should return 0 for "no match"
502          * or non-zero for "match".
503          */
504         int (*aux_output_match)         (struct perf_event *event);
505                                         /* optional */
506 
507         /*
508          * Filter events for PMU-specific reasons.
509          */
510         int (*filter_match)             (struct perf_event *event); /* optional */
511 
512         /*
513          * Check period value for PERF_EVENT_IOC_PERIOD ioctl.
514          */
515         int (*check_period)             (struct perf_event *event, u64 value); /* optional */
516 };
517 
518 enum perf_addr_filter_action_t {
519         PERF_ADDR_FILTER_ACTION_STOP = 0,
520         PERF_ADDR_FILTER_ACTION_START,
521         PERF_ADDR_FILTER_ACTION_FILTER,
522 };
523 
524 /**
525  * struct perf_addr_filter - address range filter definition
526  * @entry:      event's filter list linkage
527  * @path:       object file's path for file-based filters
528  * @offset:     filter range offset
529  * @size:       filter range size (size==0 means single address trigger)
530  * @action:     filter/start/stop
531  *
532  * This is a hardware-agnostic filter configuration as specified by the user.
533  */
534 struct perf_addr_filter {
535         struct list_head        entry;
536         struct path             path;
537         unsigned long           offset;
538         unsigned long           size;
539         enum perf_addr_filter_action_t  action;
540 };
541 
542 /**
543  * struct perf_addr_filters_head - container for address range filters
544  * @list:       list of filters for this event
545  * @lock:       spinlock that serializes accesses to the @list and event's
546  *              (and its children's) filter generations.
547  * @nr_file_filters:    number of file-based filters
548  *
549  * A child event will use parent's @list (and therefore @lock), so they are
550  * bundled together; see perf_event_addr_filters().
551  */
552 struct perf_addr_filters_head {
553         struct list_head        list;
554         raw_spinlock_t          lock;
555         unsigned int            nr_file_filters;
556 };
557 
558 struct perf_addr_filter_range {
559         unsigned long           start;
560         unsigned long           size;
561 };
562 
563 /**
564  * enum perf_event_state - the states of an event:
565  */
566 enum perf_event_state {
567         PERF_EVENT_STATE_DEAD           = -4,
568         PERF_EVENT_STATE_EXIT           = -3,
569         PERF_EVENT_STATE_ERROR          = -2,
570         PERF_EVENT_STATE_OFF            = -1,
571         PERF_EVENT_STATE_INACTIVE       =  0,
572         PERF_EVENT_STATE_ACTIVE         =  1,
573 };
574 
575 struct file;
576 struct perf_sample_data;
577 
578 typedef void (*perf_overflow_handler_t)(struct perf_event *,
579                                         struct perf_sample_data *,
580                                         struct pt_regs *regs);
581 
582 /*
583  * Event capabilities. For event_caps and groups caps.
584  *
585  * PERF_EV_CAP_SOFTWARE: Is a software event.
586  * PERF_EV_CAP_READ_ACTIVE_PKG: A CPU event (or cgroup event) that can be read
587  * from any CPU in the package where it is active.
588  * PERF_EV_CAP_SIBLING: An event with this flag must be a group sibling and
589  * cannot be a group leader. If an event with this flag is detached from the
590  * group it is scheduled out and moved into an unrecoverable ERROR state.
591  */
592 #define PERF_EV_CAP_SOFTWARE            BIT(0)
593 #define PERF_EV_CAP_READ_ACTIVE_PKG     BIT(1)
594 #define PERF_EV_CAP_SIBLING             BIT(2)
595 
596 #define SWEVENT_HLIST_BITS              8
597 #define SWEVENT_HLIST_SIZE              (1 << SWEVENT_HLIST_BITS)
598 
599 struct swevent_hlist {
600         struct hlist_head               heads[SWEVENT_HLIST_SIZE];
601         struct rcu_head                 rcu_head;
602 };
603 
604 #define PERF_ATTACH_CONTEXT     0x01
605 #define PERF_ATTACH_GROUP       0x02
606 #define PERF_ATTACH_TASK        0x04
607 #define PERF_ATTACH_TASK_DATA   0x08
608 #define PERF_ATTACH_ITRACE      0x10
609 #define PERF_ATTACH_SCHED_CB    0x20
610 #define PERF_ATTACH_CHILD       0x40
611 
612 struct perf_cgroup;
613 struct perf_buffer;
614 
615 struct pmu_event_list {
616         raw_spinlock_t          lock;
617         struct list_head        list;
618 };
619 
620 #define for_each_sibling_event(sibling, event)                  \
621         if ((event)->group_leader == (event))                   \
622                 list_for_each_entry((sibling), &(event)->sibling_list, sibling_list)
623 
624 /**
625  * struct perf_event - performance event kernel representation:
626  */
627 struct perf_event {
628 #ifdef CONFIG_PERF_EVENTS
629         /*
630          * entry onto perf_event_context::event_list;
631          *   modifications require ctx->lock
632          *   RCU safe iterations.
633          */
634         struct list_head                event_entry;
635 
636         /*
637          * Locked for modification by both ctx->mutex and ctx->lock; holding
638          * either sufficies for read.
639          */
640         struct list_head                sibling_list;
641         struct list_head                active_list;
642         /*
643          * Node on the pinned or flexible tree located at the event context;
644          */
645         struct rb_node                  group_node;
646         u64                             group_index;
647         /*
648          * We need storage to track the entries in perf_pmu_migrate_context; we
649          * cannot use the event_entry because of RCU and we want to keep the
650          * group in tact which avoids us using the other two entries.
651          */
652         struct list_head                migrate_entry;
653 
654         struct hlist_node               hlist_entry;
655         struct list_head                active_entry;
656         int                             nr_siblings;
657 
658         /* Not serialized. Only written during event initialization. */
659         int                             event_caps;
660         /* The cumulative AND of all event_caps for events in this group. */
661         int                             group_caps;
662 
663         struct perf_event               *group_leader;
664         struct pmu                      *pmu;
665         void                            *pmu_private;
666 
667         enum perf_event_state           state;
668         unsigned int                    attach_state;
669         local64_t                       count;
670         atomic64_t                      child_count;
671 
672         /*
673          * These are the total time in nanoseconds that the event
674          * has been enabled (i.e. eligible to run, and the task has
675          * been scheduled in, if this is a per-task event)
676          * and running (scheduled onto the CPU), respectively.
677          */
678         u64                             total_time_enabled;
679         u64                             total_time_running;
680         u64                             tstamp;
681 
682         /*
683          * timestamp shadows the actual context timing but it can
684          * be safely used in NMI interrupt context. It reflects the
685          * context time as it was when the event was last scheduled in.
686          *
687          * ctx_time already accounts for ctx->timestamp. Therefore to
688          * compute ctx_time for a sample, simply add perf_clock().
689          */
690         u64                             shadow_ctx_time;
691 
692         struct perf_event_attr          attr;
693         u16                             header_size;
694         u16                             id_header_size;
695         u16                             read_size;
696         struct hw_perf_event            hw;
697 
698         struct perf_event_context       *ctx;
699         atomic_long_t                   refcount;
700 
701         /*
702          * These accumulate total time (in nanoseconds) that children
703          * events have been enabled and running, respectively.
704          */
705         atomic64_t                      child_total_time_enabled;
706         atomic64_t                      child_total_time_running;
707 
708         /*
709          * Protect attach/detach and child_list:
710          */
711         struct mutex                    child_mutex;
712         struct list_head                child_list;
713         struct perf_event               *parent;
714 
715         int                             oncpu;
716         int                             cpu;
717 
718         struct list_head                owner_entry;
719         struct task_struct              *owner;
720 
721         /* mmap bits */
722         struct mutex                    mmap_mutex;
723         atomic_t                        mmap_count;
724 
725         struct perf_buffer              *rb;
726         struct list_head                rb_entry;
727         unsigned long                   rcu_batches;
728         int                             rcu_pending;
729 
730         /* poll related */
731         wait_queue_head_t               waitq;
732         struct fasync_struct            *fasync;
733 
734         /* delayed work for NMIs and such */
735         int                             pending_wakeup;
736         int                             pending_kill;
737         int                             pending_disable;
738         struct irq_work                 pending;
739 
740         atomic_t                        event_limit;
741 
742         /* address range filters */
743         struct perf_addr_filters_head   addr_filters;
744         /* vma address array for file-based filders */
745         struct perf_addr_filter_range   *addr_filter_ranges;
746         unsigned long                   addr_filters_gen;
747 
748         /* for aux_output events */
749         struct perf_event               *aux_event;
750 
751         void (*destroy)(struct perf_event *);
752         struct rcu_head                 rcu_head;
753 
754         struct pid_namespace            *ns;
755         u64                             id;
756 
757         u64                             (*clock)(void);
758         perf_overflow_handler_t         overflow_handler;
759         void                            *overflow_handler_context;
760 #ifdef CONFIG_BPF_SYSCALL
761         perf_overflow_handler_t         orig_overflow_handler;
762         struct bpf_prog                 *prog;
763 #endif
764 
765 #ifdef CONFIG_EVENT_TRACING
766         struct trace_event_call         *tp_event;
767         struct event_filter             *filter;
768 #ifdef CONFIG_FUNCTION_TRACER
769         struct ftrace_ops               ftrace_ops;
770 #endif
771 #endif
772 
773 #ifdef CONFIG_CGROUP_PERF
774         struct perf_cgroup              *cgrp; /* cgroup event is attach to */
775 #endif
776 
777 #ifdef CONFIG_SECURITY
778         void *security;
779 #endif
780         struct list_head                sb_list;
781 #endif /* CONFIG_PERF_EVENTS */
782 };
783 
784 
785 struct perf_event_groups {
786         struct rb_root  tree;
787         u64             index;
788 };
789 
790 /**
791  * struct perf_event_context - event context structure
792  *
793  * Used as a container for task events and CPU events as well:
794  */
795 struct perf_event_context {
796         struct pmu                      *pmu;
797         /*
798          * Protect the states of the events in the list,
799          * nr_active, and the list:
800          */
801         raw_spinlock_t                  lock;
802         /*
803          * Protect the list of events.  Locking either mutex or lock
804          * is sufficient to ensure the list doesn't change; to change
805          * the list you need to lock both the mutex and the spinlock.
806          */
807         struct mutex                    mutex;
808 
809         struct list_head                active_ctx_list;
810         struct perf_event_groups        pinned_groups;
811         struct perf_event_groups        flexible_groups;
812         struct list_head                event_list;
813 
814         struct list_head                pinned_active;
815         struct list_head                flexible_active;
816 
817         int                             nr_events;
818         int                             nr_active;
819         int                             is_active;
820         int                             nr_stat;
821         int                             nr_freq;
822         int                             rotate_disable;
823         /*
824          * Set when nr_events != nr_active, except tolerant to events not
825          * necessary to be active due to scheduling constraints, such as cgroups.
826          */
827         int                             rotate_necessary;
828         refcount_t                      refcount;
829         struct task_struct              *task;
830 
831         /*
832          * Context clock, runs when context enabled.
833          */
834         u64                             time;
835         u64                             timestamp;
836 
837         /*
838          * These fields let us detect when two contexts have both
839          * been cloned (inherited) from a common ancestor.
840          */
841         struct perf_event_context       *parent_ctx;
842         u64                             parent_gen;
843         u64                             generation;
844         int                             pin_count;
845 #ifdef CONFIG_CGROUP_PERF
846         int                             nr_cgroups;      /* cgroup evts */
847 #endif
848         void                            *task_ctx_data; /* pmu specific data */
849         struct rcu_head                 rcu_head;
850 };
851 
852 /*
853  * Number of contexts where an event can trigger:
854  *      task, softirq, hardirq, nmi.
855  */
856 #define PERF_NR_CONTEXTS        4
857 
858 /**
859  * struct perf_event_cpu_context - per cpu event context structure
860  */
861 struct perf_cpu_context {
862         struct perf_event_context       ctx;
863         struct perf_event_context       *task_ctx;
864         int                             active_oncpu;
865         int                             exclusive;
866 
867         raw_spinlock_t                  hrtimer_lock;
868         struct hrtimer                  hrtimer;
869         ktime_t                         hrtimer_interval;
870         unsigned int                    hrtimer_active;
871 
872 #ifdef CONFIG_CGROUP_PERF
873         struct perf_cgroup              *cgrp;
874         struct list_head                cgrp_cpuctx_entry;
875 #endif
876 
877         struct list_head                sched_cb_entry;
878         int                             sched_cb_usage;
879 
880         int                             online;
881         /*
882          * Per-CPU storage for iterators used in visit_groups_merge. The default
883          * storage is of size 2 to hold the CPU and any CPU event iterators.
884          */
885         int                             heap_size;
886         struct perf_event               **heap;
887         struct perf_event               *heap_default[2];
888 };
889 
890 struct perf_output_handle {
891         struct perf_event               *event;
892         struct perf_buffer              *rb;
893         unsigned long                   wakeup;
894         unsigned long                   size;
895         u64                             aux_flags;
896         union {
897                 void                    *addr;
898                 unsigned long           head;
899         };
900         int                             page;
901 };
902 
903 struct bpf_perf_event_data_kern {
904         bpf_user_pt_regs_t *regs;
905         struct perf_sample_data *data;
906         struct perf_event *event;
907 };
908 
909 #ifdef CONFIG_CGROUP_PERF
910 
911 /*
912  * perf_cgroup_info keeps track of time_enabled for a cgroup.
913  * This is a per-cpu dynamically allocated data structure.
914  */
915 struct perf_cgroup_info {
916         u64                             time;
917         u64                             timestamp;
918 };
919 
920 struct perf_cgroup {
921         struct cgroup_subsys_state      css;
922         struct perf_cgroup_info __percpu *info;
923 };
924 
925 /*
926  * Must ensure cgroup is pinned (css_get) before calling
927  * this function. In other words, we cannot call this function
928  * if there is no cgroup event for the current CPU context.
929  */
930 static inline struct perf_cgroup *
931 perf_cgroup_from_task(struct task_struct *task, struct perf_event_context *ctx)
932 {
933         return container_of(task_css_check(task, perf_event_cgrp_id,
934                                            ctx ? lockdep_is_held(&ctx->lock)
935                                                : true),
936                             struct perf_cgroup, css);
937 }
938 #endif /* CONFIG_CGROUP_PERF */
939 
940 #ifdef CONFIG_PERF_EVENTS
941 
942 extern void *perf_aux_output_begin(struct perf_output_handle *handle,
943                                    struct perf_event *event);
944 extern void perf_aux_output_end(struct perf_output_handle *handle,
945                                 unsigned long size);
946 extern int perf_aux_output_skip(struct perf_output_handle *handle,
947                                 unsigned long size);
948 extern void *perf_get_aux(struct perf_output_handle *handle);
949 extern void perf_aux_output_flag(struct perf_output_handle *handle, u64 flags);
950 extern void perf_event_itrace_started(struct perf_event *event);
951 
952 extern int perf_pmu_register(struct pmu *pmu, const char *name, int type);
953 extern void perf_pmu_unregister(struct pmu *pmu);
954 
955 extern int perf_num_counters(void);
956 extern const char *perf_pmu_name(void);
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);
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 
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
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 
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 
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  */
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  */
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 
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
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  */
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
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  */
1180 static __always_inline void
1181 perf_sw_event_sched(u32 event_id, u64 nr, u64 addr)
1182 {
1183         if (static_key_false(&perf_swevent_enabled[event_id])) {
1184                 struct pt_regs *regs = this_cpu_ptr(&__perf_regs[0]);
1185 
1186                 perf_fetch_caller_regs(regs);
1187                 ___perf_sw_event(event_id, nr, regs, addr);
1188         }
1189 }
1190 
1191 extern struct static_key_false perf_sched_events;
1192 
1193 static __always_inline bool
1194 perf_sw_migrate_enabled(void)
1195 {
1196         if (static_key_false(&perf_swevent_enabled[PERF_COUNT_SW_CPU_MIGRATIONS]))
1197                 return true;
1198         return false;
1199 }
1200 
1201 static inline void perf_event_task_migrate(struct task_struct *task)
1202 {
1203         if (perf_sw_migrate_enabled())
1204                 task->sched_migrated = 1;
1205 }
1206 
1207 static inline void perf_event_task_sched_in(struct task_struct *prev,
1208                                             struct task_struct *task)
1209 {
1210         if (static_branch_unlikely(&perf_sched_events))
1211                 __perf_event_task_sched_in(prev, task);
1212 
1213         if (perf_sw_migrate_enabled() && task->sched_migrated) {
1214                 struct pt_regs *regs = this_cpu_ptr(&__perf_regs[0]);
1215 
1216                 perf_fetch_caller_regs(regs);
1217                 ___perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, regs, 0);
1218                 task->sched_migrated = 0;
1219         }
1220 }
1221 
1222 static inline void perf_event_task_sched_out(struct task_struct *prev,
1223                                              struct task_struct *next)
1224 {
1225         perf_sw_event_sched(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 0);
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 
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 
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 
1315 static inline int perf_is_paranoid(void)
1316 {
1317         return sysctl_perf_event_paranoid > -1;
1318 }
1319 
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 
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 
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 
1360 static inline bool has_branch_stack(struct perf_event *event)
1361 {
1362         return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK;
1363 }
1364 
1365 static inline bool needs_branch_stack(struct perf_event *event)
1366 {
1367         return event->attr.branch_sample_type != 0;
1368 }
1369 
1370 static inline bool has_aux(struct perf_event *event)
1371 {
1372         return event->pmu->setup_aux;
1373 }
1374 
1375 static inline bool is_write_backward(struct perf_event *event)
1376 {
1377         return !!event->attr.write_backward;
1378 }
1379 
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 *
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 *
1434 perf_aux_output_begin(struct perf_output_handle *handle,
1435                       struct perf_event *event)                         { return NULL; }
1436 static inline void
1437 perf_aux_output_end(struct perf_output_handle *handle, unsigned long size)
1438                                                                         { }
1439 static inline int
1440 perf_aux_output_skip(struct perf_output_handle *handle,
1441                      unsigned long size)                                { return -EINVAL; }
1442 static inline void *
1443 perf_get_aux(struct perf_output_handle *handle)                         { return NULL; }
1444 static inline void
1445 perf_event_task_migrate(struct task_struct *task)                       { }
1446 static inline void
1447 perf_event_task_sched_in(struct task_struct *prev,
1448                          struct task_struct *task)                      { }
1449 static inline void
1450 perf_event_task_sched_out(struct task_struct *prev,
1451                           struct task_struct *next)                     { }
1452 static inline int perf_event_init_task(struct task_struct *child)       { return 0; }
1453 static inline void perf_event_exit_task(struct task_struct *child)      { }
1454 static inline void perf_event_free_task(struct task_struct *task)       { }
1455 static inline void perf_event_delayed_put(struct task_struct *task)     { }
1456 static inline struct file *perf_event_get(unsigned int fd)      { return ERR_PTR(-EINVAL); }
1457 static inline const struct perf_event *perf_get_event(struct file *file)
1458 {
1459         return ERR_PTR(-EINVAL);
1460 }
1461 static inline const struct perf_event_attr *perf_event_attrs(struct perf_event *event)
1462 {
1463         return ERR_PTR(-EINVAL);
1464 }
1465 static inline int perf_event_read_local(struct perf_event *event, u64 *value,
1466                                         u64 *enabled, u64 *running)
1467 {
1468         return -EINVAL;
1469 }
1470 static inline void perf_event_print_debug(void)                         { }
1471 static inline int perf_event_task_disable(void)                         { return -EINVAL; }
1472 static inline int perf_event_task_enable(void)                          { return -EINVAL; }
1473 static inline int perf_event_refresh(struct perf_event *event, int refresh)
1474 {
1475         return -EINVAL;
1476 }
1477 
1478 static inline void
1479 perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr)     { }
1480 static inline void
1481 perf_sw_event_sched(u32 event_id, u64 nr, u64 addr)                     { }
1482 static inline void
1483 perf_bp_event(struct perf_event *event, void *data)                     { }
1484 
1485 static inline int perf_register_guest_info_callbacks
1486 (struct perf_guest_info_callbacks *callbacks)                           { return 0; }
1487 static inline int perf_unregister_guest_info_callbacks
1488 (struct perf_guest_info_callbacks *callbacks)                           { return 0; }
1489 
1490 static inline void perf_event_mmap(struct vm_area_struct *vma)          { }
1491 
1492 typedef int (perf_ksymbol_get_name_f)(char *name, int name_len, void *data);
1493 static inline void perf_event_ksymbol(u16 ksym_type, u64 addr, u32 len,
1494                                       bool unregister, const char *sym) { }
1495 static inline void perf_event_bpf_event(struct bpf_prog *prog,
1496                                         enum perf_bpf_event_type type,
1497                                         u16 flags)                      { }
1498 static inline void perf_event_exec(void)                                { }
1499 static inline void perf_event_comm(struct task_struct *tsk, bool exec)  { }
1500 static inline void perf_event_namespaces(struct task_struct *tsk)       { }
1501 static inline void perf_event_fork(struct task_struct *tsk)             { }
1502 static inline void perf_event_text_poke(const void *addr,
1503                                         const void *old_bytes,
1504                                         size_t old_len,
1505                                         const void *new_bytes,
1506                                         size_t new_len)                 { }
1507 static inline void perf_event_init(void)                                { }
1508 static inline int  perf_swevent_get_recursion_context(void)             { return -1; }
1509 static inline void perf_swevent_put_recursion_context(int rctx)         { }
1510 static inline u64 perf_swevent_set_period(struct perf_event *event)     { return 0; }
1511 static inline void perf_event_enable(struct perf_event *event)          { }
1512 static inline void perf_event_disable(struct perf_event *event)         { }
1513 static inline int __perf_event_disable(void *info)                      { return -1; }
1514 static inline void perf_event_task_tick(void)                           { }
1515 static inline int perf_event_release_kernel(struct perf_event *event)   { return 0; }
1516 static inline int perf_event_period(struct perf_event *event, u64 value)
1517 {
1518         return -EINVAL;
1519 }
1520 static inline u64 perf_event_pause(struct perf_event *event, bool reset)
1521 {
1522         return 0;
1523 }
1524 #endif
1525 
1526 #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL)
1527 extern void perf_restore_debug_store(void);
1528 #else
1529 static inline void perf_restore_debug_store(void)                       { }
1530 #endif
1531 
1532 static __always_inline bool perf_raw_frag_last(const struct perf_raw_frag *frag)
1533 {
1534         return frag->pad < sizeof(u64);
1535 }
1536 
1537 #define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x))
1538 
1539 struct perf_pmu_events_attr {
1540         struct device_attribute attr;
1541         u64 id;
1542         const char *event_str;
1543 };
1544 
1545 struct perf_pmu_events_ht_attr {
1546         struct device_attribute                 attr;
1547         u64                                     id;
1548         const char                              *event_str_ht;
1549         const char                              *event_str_noht;
1550 };
1551 
1552 ssize_t perf_event_sysfs_show(struct device *dev, struct device_attribute *attr,
1553                               char *page);
1554 
1555 #define PMU_EVENT_ATTR(_name, _var, _id, _show)                         \
1556 static struct perf_pmu_events_attr _var = {                             \
1557         .attr = __ATTR(_name, 0444, _show, NULL),                       \
1558         .id   =  _id,                                                   \
1559 };
1560 
1561 #define PMU_EVENT_ATTR_STRING(_name, _var, _str)                            \
1562 static struct perf_pmu_events_attr _var = {                                 \
1563         .attr           = __ATTR(_name, 0444, perf_event_sysfs_show, NULL), \
1564         .id             = 0,                                                \
1565         .event_str      = _str,                                             \
1566 };
1567 
1568 #define PMU_FORMAT_ATTR(_name, _format)                                 \
1569 static ssize_t                                                          \
1570 _name##_show(struct device *dev,                                        \
1571                                struct device_attribute *attr,           \
1572                                char *page)                              \
1573 {                                                                       \
1574         BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE);                     \
1575         return sprintf(page, _format "\n");                             \
1576 }                                                                       \
1577                                                                         \
1578 static struct device_attribute format_attr_##_name = __ATTR_RO(_name)
1579 
1580 /* Performance counter hotplug functions */
1581 #ifdef CONFIG_PERF_EVENTS
1582 int perf_event_init_cpu(unsigned int cpu);
1583 int perf_event_exit_cpu(unsigned int cpu);
1584 #else
1585 #define perf_event_init_cpu     NULL
1586 #define perf_event_exit_cpu     NULL
1587 #endif
1588 
1589 extern void __weak arch_perf_update_userpage(struct perf_event *event,
1590                                              struct perf_event_mmap_page *userpg,
1591                                              u64 now);
1592 
1593 #ifdef CONFIG_MMU
1594 extern __weak u64 arch_perf_get_page_size(struct mm_struct *mm, unsigned long addr);
1595 #endif
1596 
1597 #endif /* _LINUX_PERF_EVENT_H */
1598 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp