]>
Commit | Line | Data |
---|---|---|
0793a61d | 1 | /* |
57c0c15b | 2 | * Performance events: |
0793a61d | 3 | * |
a308444c | 4 | * Copyright (C) 2008-2009, Thomas Gleixner <tglx@linutronix.de> |
e7e7ee2e IM |
5 | * Copyright (C) 2008-2011, Red Hat, Inc., Ingo Molnar |
6 | * Copyright (C) 2008-2011, Red Hat, Inc., Peter Zijlstra | |
0793a61d | 7 | * |
57c0c15b | 8 | * Data type definitions, declarations, prototypes. |
0793a61d | 9 | * |
a308444c | 10 | * Started by: Thomas Gleixner and Ingo Molnar |
0793a61d | 11 | * |
57c0c15b | 12 | * For licencing details see kernel-base/COPYING |
0793a61d | 13 | */ |
cdd6c482 IM |
14 | #ifndef _LINUX_PERF_EVENT_H |
15 | #define _LINUX_PERF_EVENT_H | |
0793a61d | 16 | |
607ca46e | 17 | #include <uapi/linux/perf_event.h> |
0793a61d | 18 | |
9f66a381 | 19 | /* |
f3dfd265 | 20 | * Kernel-internal data types and definitions: |
9f66a381 IM |
21 | */ |
22 | ||
cdd6c482 IM |
23 | #ifdef CONFIG_PERF_EVENTS |
24 | # include <asm/perf_event.h> | |
7be79236 | 25 | # include <asm/local64.h> |
f3dfd265 PM |
26 | #endif |
27 | ||
39447b38 | 28 | struct perf_guest_info_callbacks { |
e7e7ee2e IM |
29 | int (*is_in_guest)(void); |
30 | int (*is_user_mode)(void); | |
31 | unsigned long (*get_guest_ip)(void); | |
39447b38 ZY |
32 | }; |
33 | ||
2ff6cfd7 AB |
34 | #ifdef CONFIG_HAVE_HW_BREAKPOINT |
35 | #include <asm/hw_breakpoint.h> | |
36 | #endif | |
37 | ||
f3dfd265 PM |
38 | #include <linux/list.h> |
39 | #include <linux/mutex.h> | |
40 | #include <linux/rculist.h> | |
41 | #include <linux/rcupdate.h> | |
42 | #include <linux/spinlock.h> | |
d6d020e9 | 43 | #include <linux/hrtimer.h> |
3c446b3d | 44 | #include <linux/fs.h> |
709e50cf | 45 | #include <linux/pid_namespace.h> |
906010b2 | 46 | #include <linux/workqueue.h> |
5331d7b8 | 47 | #include <linux/ftrace.h> |
85cfabbc | 48 | #include <linux/cpu.h> |
e360adbe | 49 | #include <linux/irq_work.h> |
c5905afb | 50 | #include <linux/static_key.h> |
851cf6e7 | 51 | #include <linux/jump_label_ratelimit.h> |
60063497 | 52 | #include <linux/atomic.h> |
641cc938 | 53 | #include <linux/sysfs.h> |
4018994f | 54 | #include <linux/perf_regs.h> |
fa588151 | 55 | #include <asm/local.h> |
f3dfd265 | 56 | |
f9188e02 PZ |
57 | struct perf_callchain_entry { |
58 | __u64 nr; | |
59 | __u64 ip[PERF_MAX_STACK_DEPTH]; | |
60 | }; | |
61 | ||
3a43ce68 FW |
62 | struct perf_raw_record { |
63 | u32 size; | |
64 | void *data; | |
f413cdb8 FW |
65 | }; |
66 | ||
bce38cd5 SE |
67 | /* |
68 | * branch stack layout: | |
69 | * nr: number of taken branches stored in entries[] | |
70 | * | |
71 | * Note that nr can vary from sample to sample | |
72 | * branches (to, from) are stored from most recent | |
73 | * to least recent, i.e., entries[0] contains the most | |
74 | * recent branch. | |
75 | */ | |
caff2bef PZ |
76 | struct perf_branch_stack { |
77 | __u64 nr; | |
78 | struct perf_branch_entry entries[0]; | |
79 | }; | |
80 | ||
4018994f JO |
81 | struct perf_regs_user { |
82 | __u64 abi; | |
83 | struct pt_regs *regs; | |
84 | }; | |
85 | ||
f3dfd265 PM |
86 | struct task_struct; |
87 | ||
efc9f05d SE |
88 | /* |
89 | * extra PMU register associated with an event | |
90 | */ | |
91 | struct hw_perf_event_extra { | |
92 | u64 config; /* register value */ | |
93 | unsigned int reg; /* register address or index */ | |
94 | int alloc; /* extra register already allocated */ | |
95 | int idx; /* index in shared_regs->regs[] */ | |
96 | }; | |
97 | ||
43b45780 AH |
98 | struct event_constraint; |
99 | ||
0793a61d | 100 | /** |
cdd6c482 | 101 | * struct hw_perf_event - performance event hardware details: |
0793a61d | 102 | */ |
cdd6c482 IM |
103 | struct hw_perf_event { |
104 | #ifdef CONFIG_PERF_EVENTS | |
d6d020e9 PZ |
105 | union { |
106 | struct { /* hardware */ | |
a308444c | 107 | u64 config; |
447a194b | 108 | u64 last_tag; |
a308444c | 109 | unsigned long config_base; |
cdd6c482 | 110 | unsigned long event_base; |
c48b6053 | 111 | int event_base_rdpmc; |
a308444c | 112 | int idx; |
447a194b | 113 | int last_cpu; |
9fac2cf3 | 114 | int flags; |
bce38cd5 | 115 | |
efc9f05d | 116 | struct hw_perf_event_extra extra_reg; |
bce38cd5 | 117 | struct hw_perf_event_extra branch_reg; |
43b45780 AH |
118 | |
119 | struct event_constraint *constraint; | |
d6d020e9 | 120 | }; |
721a669b | 121 | struct { /* software */ |
a308444c | 122 | struct hrtimer hrtimer; |
d6d020e9 | 123 | }; |
f22c1bb6 ON |
124 | struct { /* tracepoint */ |
125 | struct task_struct *tp_target; | |
126 | /* for tp_event->class */ | |
127 | struct list_head tp_list; | |
128 | }; | |
24f1e32c | 129 | #ifdef CONFIG_HAVE_HW_BREAKPOINT |
45a73372 | 130 | struct { /* breakpoint */ |
d580ff86 PZ |
131 | /* |
132 | * Crufty hack to avoid the chicken and egg | |
133 | * problem hw_breakpoint has with context | |
134 | * creation and event initalization. | |
135 | */ | |
136 | struct task_struct *bp_target; | |
f22c1bb6 ON |
137 | struct arch_hw_breakpoint info; |
138 | struct list_head bp_list; | |
45a73372 | 139 | }; |
24f1e32c | 140 | #endif |
d6d020e9 | 141 | }; |
a4eaf7f1 | 142 | int state; |
e7850595 | 143 | local64_t prev_count; |
b23f3325 | 144 | u64 sample_period; |
9e350de3 | 145 | u64 last_period; |
e7850595 | 146 | local64_t period_left; |
e050e3f0 | 147 | u64 interrupts_seq; |
60db5e09 | 148 | u64 interrupts; |
6a24ed6c | 149 | |
abd50713 PZ |
150 | u64 freq_time_stamp; |
151 | u64 freq_count_stamp; | |
ee06094f | 152 | #endif |
0793a61d TG |
153 | }; |
154 | ||
a4eaf7f1 PZ |
155 | /* |
156 | * hw_perf_event::state flags | |
157 | */ | |
158 | #define PERF_HES_STOPPED 0x01 /* the counter is stopped */ | |
159 | #define PERF_HES_UPTODATE 0x02 /* event->count up-to-date */ | |
160 | #define PERF_HES_ARCH 0x04 | |
161 | ||
cdd6c482 | 162 | struct perf_event; |
621a01ea | 163 | |
8d2cacbb PZ |
164 | /* |
165 | * Common implementation detail of pmu::{start,commit,cancel}_txn | |
166 | */ | |
167 | #define PERF_EVENT_TXN 0x1 | |
6bde9b6c | 168 | |
621a01ea | 169 | /** |
4aeb0b42 | 170 | * struct pmu - generic performance monitoring unit |
621a01ea | 171 | */ |
4aeb0b42 | 172 | struct pmu { |
b0a873eb PZ |
173 | struct list_head entry; |
174 | ||
c464c76e | 175 | struct module *module; |
abe43400 | 176 | struct device *dev; |
0c9d42ed | 177 | const struct attribute_group **attr_groups; |
03d8e80b | 178 | const char *name; |
2e80a82a PZ |
179 | int type; |
180 | ||
108b02cf PZ |
181 | int * __percpu pmu_disable_count; |
182 | struct perf_cpu_context * __percpu pmu_cpu_context; | |
8dc85d54 | 183 | int task_ctx_nr; |
62b85639 | 184 | int hrtimer_interval_ms; |
6bde9b6c LM |
185 | |
186 | /* | |
a4eaf7f1 PZ |
187 | * Fully disable/enable this PMU, can be used to protect from the PMI |
188 | * as well as for lazy/batch writing of the MSRs. | |
6bde9b6c | 189 | */ |
ad5133b7 PZ |
190 | void (*pmu_enable) (struct pmu *pmu); /* optional */ |
191 | void (*pmu_disable) (struct pmu *pmu); /* optional */ | |
6bde9b6c | 192 | |
8d2cacbb | 193 | /* |
a4eaf7f1 | 194 | * Try and initialize the event for this PMU. |
24cd7f54 | 195 | * Should return -ENOENT when the @event doesn't match this PMU. |
8d2cacbb | 196 | */ |
b0a873eb PZ |
197 | int (*event_init) (struct perf_event *event); |
198 | ||
a4eaf7f1 PZ |
199 | #define PERF_EF_START 0x01 /* start the counter when adding */ |
200 | #define PERF_EF_RELOAD 0x02 /* reload the counter when starting */ | |
201 | #define PERF_EF_UPDATE 0x04 /* update the counter when stopping */ | |
202 | ||
8d2cacbb | 203 | /* |
a4eaf7f1 PZ |
204 | * Adds/Removes a counter to/from the PMU, can be done inside |
205 | * a transaction, see the ->*_txn() methods. | |
206 | */ | |
207 | int (*add) (struct perf_event *event, int flags); | |
208 | void (*del) (struct perf_event *event, int flags); | |
209 | ||
210 | /* | |
211 | * Starts/Stops a counter present on the PMU. The PMI handler | |
212 | * should stop the counter when perf_event_overflow() returns | |
213 | * !0. ->start() will be used to continue. | |
214 | */ | |
215 | void (*start) (struct perf_event *event, int flags); | |
216 | void (*stop) (struct perf_event *event, int flags); | |
217 | ||
218 | /* | |
219 | * Updates the counter value of the event. | |
220 | */ | |
cdd6c482 | 221 | void (*read) (struct perf_event *event); |
6bde9b6c LM |
222 | |
223 | /* | |
24cd7f54 PZ |
224 | * Group events scheduling is treated as a transaction, add |
225 | * group events as a whole and perform one schedulability test. | |
226 | * If the test fails, roll back the whole group | |
a4eaf7f1 PZ |
227 | * |
228 | * Start the transaction, after this ->add() doesn't need to | |
24cd7f54 | 229 | * do schedulability tests. |
8d2cacbb | 230 | */ |
e7e7ee2e | 231 | void (*start_txn) (struct pmu *pmu); /* optional */ |
8d2cacbb | 232 | /* |
a4eaf7f1 | 233 | * If ->start_txn() disabled the ->add() schedulability test |
8d2cacbb PZ |
234 | * then ->commit_txn() is required to perform one. On success |
235 | * the transaction is closed. On error the transaction is kept | |
236 | * open until ->cancel_txn() is called. | |
237 | */ | |
e7e7ee2e | 238 | int (*commit_txn) (struct pmu *pmu); /* optional */ |
8d2cacbb | 239 | /* |
a4eaf7f1 | 240 | * Will cancel the transaction, assumes ->del() is called |
25985edc | 241 | * for each successful ->add() during the transaction. |
8d2cacbb | 242 | */ |
e7e7ee2e | 243 | void (*cancel_txn) (struct pmu *pmu); /* optional */ |
35edc2a5 PZ |
244 | |
245 | /* | |
246 | * Will return the value for perf_event_mmap_page::index for this event, | |
247 | * if no implementation is provided it will default to: event->hw.idx + 1. | |
248 | */ | |
249 | int (*event_idx) (struct perf_event *event); /*optional */ | |
d010b332 SE |
250 | |
251 | /* | |
252 | * flush branch stack on context-switches (needed in cpu-wide mode) | |
253 | */ | |
254 | void (*flush_branch_stack) (void); | |
621a01ea IM |
255 | }; |
256 | ||
6a930700 | 257 | /** |
cdd6c482 | 258 | * enum perf_event_active_state - the states of a event |
6a930700 | 259 | */ |
cdd6c482 | 260 | enum perf_event_active_state { |
57c0c15b | 261 | PERF_EVENT_STATE_ERROR = -2, |
cdd6c482 IM |
262 | PERF_EVENT_STATE_OFF = -1, |
263 | PERF_EVENT_STATE_INACTIVE = 0, | |
57c0c15b | 264 | PERF_EVENT_STATE_ACTIVE = 1, |
6a930700 IM |
265 | }; |
266 | ||
9b51f66d | 267 | struct file; |
453f19ee PZ |
268 | struct perf_sample_data; |
269 | ||
a8b0ca17 | 270 | typedef void (*perf_overflow_handler_t)(struct perf_event *, |
b326e956 FW |
271 | struct perf_sample_data *, |
272 | struct pt_regs *regs); | |
273 | ||
d6f962b5 | 274 | enum perf_group_flag { |
e7e7ee2e | 275 | PERF_GROUP_SOFTWARE = 0x1, |
d6f962b5 FW |
276 | }; |
277 | ||
e7e7ee2e IM |
278 | #define SWEVENT_HLIST_BITS 8 |
279 | #define SWEVENT_HLIST_SIZE (1 << SWEVENT_HLIST_BITS) | |
76e1d904 FW |
280 | |
281 | struct swevent_hlist { | |
e7e7ee2e IM |
282 | struct hlist_head heads[SWEVENT_HLIST_SIZE]; |
283 | struct rcu_head rcu_head; | |
76e1d904 FW |
284 | }; |
285 | ||
8a49542c PZ |
286 | #define PERF_ATTACH_CONTEXT 0x01 |
287 | #define PERF_ATTACH_GROUP 0x02 | |
d580ff86 | 288 | #define PERF_ATTACH_TASK 0x04 |
8a49542c | 289 | |
877c6856 | 290 | struct perf_cgroup; |
76369139 FW |
291 | struct ring_buffer; |
292 | ||
0793a61d | 293 | /** |
cdd6c482 | 294 | * struct perf_event - performance event kernel representation: |
0793a61d | 295 | */ |
cdd6c482 IM |
296 | struct perf_event { |
297 | #ifdef CONFIG_PERF_EVENTS | |
9886167d PZ |
298 | /* |
299 | * entry onto perf_event_context::event_list; | |
300 | * modifications require ctx->lock | |
301 | * RCU safe iterations. | |
302 | */ | |
592903cd | 303 | struct list_head event_entry; |
9886167d PZ |
304 | |
305 | /* | |
306 | * XXX: group_entry and sibling_list should be mutually exclusive; | |
307 | * either you're a sibling on a group, or you're the group leader. | |
308 | * Rework the code to always use the same list element. | |
309 | * | |
310 | * Locked for modification by both ctx->mutex and ctx->lock; holding | |
311 | * either sufficies for read. | |
312 | */ | |
313 | struct list_head group_entry; | |
04289bb9 | 314 | struct list_head sibling_list; |
9886167d PZ |
315 | |
316 | /* | |
317 | * We need storage to track the entries in perf_pmu_migrate_context; we | |
318 | * cannot use the event_entry because of RCU and we want to keep the | |
319 | * group in tact which avoids us using the other two entries. | |
320 | */ | |
321 | struct list_head migrate_entry; | |
322 | ||
f3ae75de SE |
323 | struct hlist_node hlist_entry; |
324 | struct list_head active_entry; | |
0127c3ea | 325 | int nr_siblings; |
d6f962b5 | 326 | int group_flags; |
cdd6c482 | 327 | struct perf_event *group_leader; |
a4eaf7f1 | 328 | struct pmu *pmu; |
04289bb9 | 329 | |
cdd6c482 | 330 | enum perf_event_active_state state; |
8a49542c | 331 | unsigned int attach_state; |
e7850595 | 332 | local64_t count; |
a6e6dea6 | 333 | atomic64_t child_count; |
ee06094f | 334 | |
53cfbf59 | 335 | /* |
cdd6c482 | 336 | * These are the total time in nanoseconds that the event |
53cfbf59 | 337 | * has been enabled (i.e. eligible to run, and the task has |
cdd6c482 | 338 | * been scheduled in, if this is a per-task event) |
53cfbf59 PM |
339 | * and running (scheduled onto the CPU), respectively. |
340 | * | |
341 | * They are computed from tstamp_enabled, tstamp_running and | |
cdd6c482 | 342 | * tstamp_stopped when the event is in INACTIVE or ACTIVE state. |
53cfbf59 PM |
343 | */ |
344 | u64 total_time_enabled; | |
345 | u64 total_time_running; | |
346 | ||
347 | /* | |
348 | * These are timestamps used for computing total_time_enabled | |
cdd6c482 | 349 | * and total_time_running when the event is in INACTIVE or |
53cfbf59 PM |
350 | * ACTIVE state, measured in nanoseconds from an arbitrary point |
351 | * in time. | |
cdd6c482 IM |
352 | * tstamp_enabled: the notional time when the event was enabled |
353 | * tstamp_running: the notional time when the event was scheduled on | |
53cfbf59 | 354 | * tstamp_stopped: in INACTIVE state, the notional time when the |
cdd6c482 | 355 | * event was scheduled off. |
53cfbf59 PM |
356 | */ |
357 | u64 tstamp_enabled; | |
358 | u64 tstamp_running; | |
359 | u64 tstamp_stopped; | |
360 | ||
eed01528 SE |
361 | /* |
362 | * timestamp shadows the actual context timing but it can | |
363 | * be safely used in NMI interrupt context. It reflects the | |
364 | * context time as it was when the event was last scheduled in. | |
365 | * | |
366 | * ctx_time already accounts for ctx->timestamp. Therefore to | |
367 | * compute ctx_time for a sample, simply add perf_clock(). | |
368 | */ | |
369 | u64 shadow_ctx_time; | |
370 | ||
24f1e32c | 371 | struct perf_event_attr attr; |
c320c7b7 | 372 | u16 header_size; |
6844c09d | 373 | u16 id_header_size; |
c320c7b7 | 374 | u16 read_size; |
cdd6c482 | 375 | struct hw_perf_event hw; |
0793a61d | 376 | |
cdd6c482 | 377 | struct perf_event_context *ctx; |
a6fa941d | 378 | atomic_long_t refcount; |
0793a61d | 379 | |
53cfbf59 PM |
380 | /* |
381 | * These accumulate total time (in nanoseconds) that children | |
cdd6c482 | 382 | * events have been enabled and running, respectively. |
53cfbf59 PM |
383 | */ |
384 | atomic64_t child_total_time_enabled; | |
385 | atomic64_t child_total_time_running; | |
386 | ||
0793a61d | 387 | /* |
d859e29f | 388 | * Protect attach/detach and child_list: |
0793a61d | 389 | */ |
fccc714b PZ |
390 | struct mutex child_mutex; |
391 | struct list_head child_list; | |
cdd6c482 | 392 | struct perf_event *parent; |
0793a61d TG |
393 | |
394 | int oncpu; | |
395 | int cpu; | |
396 | ||
082ff5a2 PZ |
397 | struct list_head owner_entry; |
398 | struct task_struct *owner; | |
399 | ||
7b732a75 PZ |
400 | /* mmap bits */ |
401 | struct mutex mmap_mutex; | |
402 | atomic_t mmap_count; | |
26cb63ad | 403 | |
76369139 | 404 | struct ring_buffer *rb; |
10c6db11 | 405 | struct list_head rb_entry; |
b69cf536 PZ |
406 | unsigned long rcu_batches; |
407 | int rcu_pending; | |
37d81828 | 408 | |
7b732a75 | 409 | /* poll related */ |
0793a61d | 410 | wait_queue_head_t waitq; |
3c446b3d | 411 | struct fasync_struct *fasync; |
79f14641 PZ |
412 | |
413 | /* delayed work for NMIs and such */ | |
414 | int pending_wakeup; | |
4c9e2542 | 415 | int pending_kill; |
79f14641 | 416 | int pending_disable; |
e360adbe | 417 | struct irq_work pending; |
592903cd | 418 | |
79f14641 PZ |
419 | atomic_t event_limit; |
420 | ||
cdd6c482 | 421 | void (*destroy)(struct perf_event *); |
592903cd | 422 | struct rcu_head rcu_head; |
709e50cf PZ |
423 | |
424 | struct pid_namespace *ns; | |
8e5799b1 | 425 | u64 id; |
6fb2915d | 426 | |
b326e956 | 427 | perf_overflow_handler_t overflow_handler; |
4dc0da86 | 428 | void *overflow_handler_context; |
453f19ee | 429 | |
07b139c8 | 430 | #ifdef CONFIG_EVENT_TRACING |
1c024eca | 431 | struct ftrace_event_call *tp_event; |
6fb2915d | 432 | struct event_filter *filter; |
ced39002 JO |
433 | #ifdef CONFIG_FUNCTION_TRACER |
434 | struct ftrace_ops ftrace_ops; | |
435 | #endif | |
ee06094f | 436 | #endif |
6fb2915d | 437 | |
e5d1367f SE |
438 | #ifdef CONFIG_CGROUP_PERF |
439 | struct perf_cgroup *cgrp; /* cgroup event is attach to */ | |
440 | int cgrp_defer_enabled; | |
441 | #endif | |
442 | ||
6fb2915d | 443 | #endif /* CONFIG_PERF_EVENTS */ |
0793a61d TG |
444 | }; |
445 | ||
b04243ef PZ |
446 | enum perf_event_context_type { |
447 | task_context, | |
448 | cpu_context, | |
449 | }; | |
450 | ||
0793a61d | 451 | /** |
cdd6c482 | 452 | * struct perf_event_context - event context structure |
0793a61d | 453 | * |
cdd6c482 | 454 | * Used as a container for task events and CPU events as well: |
0793a61d | 455 | */ |
cdd6c482 | 456 | struct perf_event_context { |
108b02cf | 457 | struct pmu *pmu; |
ee643c41 | 458 | enum perf_event_context_type type; |
0793a61d | 459 | /* |
cdd6c482 | 460 | * Protect the states of the events in the list, |
d859e29f | 461 | * nr_active, and the list: |
0793a61d | 462 | */ |
e625cce1 | 463 | raw_spinlock_t lock; |
d859e29f | 464 | /* |
cdd6c482 | 465 | * Protect the list of events. Locking either mutex or lock |
d859e29f PM |
466 | * is sufficient to ensure the list doesn't change; to change |
467 | * the list you need to lock both the mutex and the spinlock. | |
468 | */ | |
a308444c | 469 | struct mutex mutex; |
04289bb9 | 470 | |
889ff015 FW |
471 | struct list_head pinned_groups; |
472 | struct list_head flexible_groups; | |
a308444c | 473 | struct list_head event_list; |
cdd6c482 | 474 | int nr_events; |
a308444c IM |
475 | int nr_active; |
476 | int is_active; | |
bfbd3381 | 477 | int nr_stat; |
0f5a2601 | 478 | int nr_freq; |
dddd3379 | 479 | int rotate_disable; |
a308444c IM |
480 | atomic_t refcount; |
481 | struct task_struct *task; | |
53cfbf59 PM |
482 | |
483 | /* | |
4af4998b | 484 | * Context clock, runs when context enabled. |
53cfbf59 | 485 | */ |
a308444c IM |
486 | u64 time; |
487 | u64 timestamp; | |
564c2b21 PM |
488 | |
489 | /* | |
490 | * These fields let us detect when two contexts have both | |
491 | * been cloned (inherited) from a common ancestor. | |
492 | */ | |
cdd6c482 | 493 | struct perf_event_context *parent_ctx; |
a308444c IM |
494 | u64 parent_gen; |
495 | u64 generation; | |
496 | int pin_count; | |
d010b332 SE |
497 | int nr_cgroups; /* cgroup evts */ |
498 | int nr_branch_stack; /* branch_stack evt */ | |
28009ce4 | 499 | struct rcu_head rcu_head; |
0793a61d TG |
500 | }; |
501 | ||
7ae07ea3 FW |
502 | /* |
503 | * Number of contexts where an event can trigger: | |
e7e7ee2e | 504 | * task, softirq, hardirq, nmi. |
7ae07ea3 FW |
505 | */ |
506 | #define PERF_NR_CONTEXTS 4 | |
507 | ||
0793a61d | 508 | /** |
cdd6c482 | 509 | * struct perf_event_cpu_context - per cpu event context structure |
0793a61d TG |
510 | */ |
511 | struct perf_cpu_context { | |
cdd6c482 IM |
512 | struct perf_event_context ctx; |
513 | struct perf_event_context *task_ctx; | |
0793a61d | 514 | int active_oncpu; |
3b6f9e5c | 515 | int exclusive; |
9e630205 SE |
516 | struct hrtimer hrtimer; |
517 | ktime_t hrtimer_interval; | |
e9d2b064 | 518 | struct list_head rotation_list; |
3f1f3320 | 519 | struct pmu *unique_pmu; |
e5d1367f | 520 | struct perf_cgroup *cgrp; |
0793a61d TG |
521 | }; |
522 | ||
5622f295 | 523 | struct perf_output_handle { |
57c0c15b | 524 | struct perf_event *event; |
76369139 | 525 | struct ring_buffer *rb; |
6d1acfd5 | 526 | unsigned long wakeup; |
5d967a8b PZ |
527 | unsigned long size; |
528 | void *addr; | |
529 | int page; | |
5622f295 MM |
530 | }; |
531 | ||
cdd6c482 | 532 | #ifdef CONFIG_PERF_EVENTS |
829b42dd | 533 | |
03d8e80b | 534 | extern int perf_pmu_register(struct pmu *pmu, const char *name, int type); |
b0a873eb | 535 | extern void perf_pmu_unregister(struct pmu *pmu); |
621a01ea | 536 | |
3bf101ba | 537 | extern int perf_num_counters(void); |
84c79910 | 538 | extern const char *perf_pmu_name(void); |
ab0cce56 JO |
539 | extern void __perf_event_task_sched_in(struct task_struct *prev, |
540 | struct task_struct *task); | |
541 | extern void __perf_event_task_sched_out(struct task_struct *prev, | |
542 | struct task_struct *next); | |
cdd6c482 IM |
543 | extern int perf_event_init_task(struct task_struct *child); |
544 | extern void perf_event_exit_task(struct task_struct *child); | |
545 | extern void perf_event_free_task(struct task_struct *task); | |
4e231c79 | 546 | extern void perf_event_delayed_put(struct task_struct *task); |
cdd6c482 | 547 | extern void perf_event_print_debug(void); |
33696fc0 PZ |
548 | extern void perf_pmu_disable(struct pmu *pmu); |
549 | extern void perf_pmu_enable(struct pmu *pmu); | |
cdd6c482 IM |
550 | extern int perf_event_task_disable(void); |
551 | extern int perf_event_task_enable(void); | |
26ca5c11 | 552 | extern int perf_event_refresh(struct perf_event *event, int refresh); |
cdd6c482 | 553 | extern void perf_event_update_userpage(struct perf_event *event); |
fb0459d7 AV |
554 | extern int perf_event_release_kernel(struct perf_event *event); |
555 | extern struct perf_event * | |
556 | perf_event_create_kernel_counter(struct perf_event_attr *attr, | |
557 | int cpu, | |
38a81da2 | 558 | struct task_struct *task, |
4dc0da86 AK |
559 | perf_overflow_handler_t callback, |
560 | void *context); | |
0cda4c02 YZ |
561 | extern void perf_pmu_migrate_context(struct pmu *pmu, |
562 | int src_cpu, int dst_cpu); | |
59ed446f PZ |
563 | extern u64 perf_event_read_value(struct perf_event *event, |
564 | u64 *enabled, u64 *running); | |
5c92d124 | 565 | |
d010b332 | 566 | |
df1a132b | 567 | struct perf_sample_data { |
5622f295 MM |
568 | u64 type; |
569 | ||
570 | u64 ip; | |
571 | struct { | |
572 | u32 pid; | |
573 | u32 tid; | |
574 | } tid_entry; | |
575 | u64 time; | |
a308444c | 576 | u64 addr; |
5622f295 MM |
577 | u64 id; |
578 | u64 stream_id; | |
579 | struct { | |
580 | u32 cpu; | |
581 | u32 reserved; | |
582 | } cpu_entry; | |
a308444c | 583 | u64 period; |
d6be9ad6 | 584 | union perf_mem_data_src data_src; |
5622f295 | 585 | struct perf_callchain_entry *callchain; |
3a43ce68 | 586 | struct perf_raw_record *raw; |
bce38cd5 | 587 | struct perf_branch_stack *br_stack; |
4018994f | 588 | struct perf_regs_user regs_user; |
c5ebcedb | 589 | u64 stack_user_size; |
c3feedf2 | 590 | u64 weight; |
fdfbbd07 AK |
591 | /* |
592 | * Transaction flags for abort events: | |
593 | */ | |
594 | u64 txn; | |
df1a132b PZ |
595 | }; |
596 | ||
fd0d000b RR |
597 | static inline void perf_sample_data_init(struct perf_sample_data *data, |
598 | u64 addr, u64 period) | |
dc1d628a | 599 | { |
fd0d000b | 600 | /* remaining struct members initialized in perf_prepare_sample() */ |
dc1d628a PZ |
601 | data->addr = addr; |
602 | data->raw = NULL; | |
bce38cd5 | 603 | data->br_stack = NULL; |
4018994f JO |
604 | data->period = period; |
605 | data->regs_user.abi = PERF_SAMPLE_REGS_ABI_NONE; | |
606 | data->regs_user.regs = NULL; | |
c5ebcedb | 607 | data->stack_user_size = 0; |
c3feedf2 | 608 | data->weight = 0; |
d6be9ad6 | 609 | data->data_src.val = 0; |
fdfbbd07 | 610 | data->txn = 0; |
dc1d628a PZ |
611 | } |
612 | ||
5622f295 MM |
613 | extern void perf_output_sample(struct perf_output_handle *handle, |
614 | struct perf_event_header *header, | |
615 | struct perf_sample_data *data, | |
cdd6c482 | 616 | struct perf_event *event); |
5622f295 MM |
617 | extern void perf_prepare_sample(struct perf_event_header *header, |
618 | struct perf_sample_data *data, | |
cdd6c482 | 619 | struct perf_event *event, |
5622f295 MM |
620 | struct pt_regs *regs); |
621 | ||
a8b0ca17 | 622 | extern int perf_event_overflow(struct perf_event *event, |
5622f295 MM |
623 | struct perf_sample_data *data, |
624 | struct pt_regs *regs); | |
df1a132b | 625 | |
6c7e550f FBH |
626 | static inline bool is_sampling_event(struct perf_event *event) |
627 | { | |
628 | return event->attr.sample_period != 0; | |
629 | } | |
630 | ||
3b6f9e5c | 631 | /* |
cdd6c482 | 632 | * Return 1 for a software event, 0 for a hardware event |
3b6f9e5c | 633 | */ |
cdd6c482 | 634 | static inline int is_software_event(struct perf_event *event) |
3b6f9e5c | 635 | { |
89a1e187 | 636 | return event->pmu->task_ctx_nr == perf_sw_context; |
3b6f9e5c PM |
637 | } |
638 | ||
c5905afb | 639 | extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX]; |
f29ac756 | 640 | |
a8b0ca17 | 641 | extern void __perf_sw_event(u32, u64, struct pt_regs *, u64); |
f29ac756 | 642 | |
b0f82b81 | 643 | #ifndef perf_arch_fetch_caller_regs |
e7e7ee2e | 644 | static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { } |
b0f82b81 | 645 | #endif |
5331d7b8 FW |
646 | |
647 | /* | |
648 | * Take a snapshot of the regs. Skip ip and frame pointer to | |
649 | * the nth caller. We only need a few of the regs: | |
650 | * - ip for PERF_SAMPLE_IP | |
651 | * - cs for user_mode() tests | |
652 | * - bp for callchains | |
653 | * - eflags, for future purposes, just in case | |
654 | */ | |
b0f82b81 | 655 | static inline void perf_fetch_caller_regs(struct pt_regs *regs) |
5331d7b8 | 656 | { |
5331d7b8 FW |
657 | memset(regs, 0, sizeof(*regs)); |
658 | ||
b0f82b81 | 659 | perf_arch_fetch_caller_regs(regs, CALLER_ADDR0); |
5331d7b8 FW |
660 | } |
661 | ||
7e54a5a0 | 662 | static __always_inline void |
a8b0ca17 | 663 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) |
e49a5bd3 | 664 | { |
7e54a5a0 PZ |
665 | struct pt_regs hot_regs; |
666 | ||
c5905afb | 667 | if (static_key_false(&perf_swevent_enabled[event_id])) { |
d430d3d7 JB |
668 | if (!regs) { |
669 | perf_fetch_caller_regs(&hot_regs); | |
670 | regs = &hot_regs; | |
671 | } | |
a8b0ca17 | 672 | __perf_sw_event(event_id, nr, regs, addr); |
e49a5bd3 FW |
673 | } |
674 | } | |
675 | ||
c5905afb | 676 | extern struct static_key_deferred perf_sched_events; |
ee6dcfa4 | 677 | |
ab0cce56 | 678 | static inline void perf_event_task_sched_in(struct task_struct *prev, |
a8d757ef | 679 | struct task_struct *task) |
ab0cce56 JO |
680 | { |
681 | if (static_key_false(&perf_sched_events.key)) | |
682 | __perf_event_task_sched_in(prev, task); | |
683 | } | |
684 | ||
685 | static inline void perf_event_task_sched_out(struct task_struct *prev, | |
686 | struct task_struct *next) | |
ee6dcfa4 | 687 | { |
a8b0ca17 | 688 | perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, NULL, 0); |
ee6dcfa4 | 689 | |
c5905afb | 690 | if (static_key_false(&perf_sched_events.key)) |
ab0cce56 | 691 | __perf_event_task_sched_out(prev, next); |
ee6dcfa4 PZ |
692 | } |
693 | ||
3af9e859 | 694 | extern void perf_event_mmap(struct vm_area_struct *vma); |
39447b38 | 695 | extern struct perf_guest_info_callbacks *perf_guest_cbs; |
dcf46b94 ZY |
696 | extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); |
697 | extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); | |
39447b38 | 698 | |
cdd6c482 IM |
699 | extern void perf_event_comm(struct task_struct *tsk); |
700 | extern void perf_event_fork(struct task_struct *tsk); | |
8d1b2d93 | 701 | |
56962b44 FW |
702 | /* Callchains */ |
703 | DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry); | |
704 | ||
e7e7ee2e IM |
705 | extern void perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs); |
706 | extern void perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs); | |
394ee076 | 707 | |
e7e7ee2e | 708 | static inline void perf_callchain_store(struct perf_callchain_entry *entry, u64 ip) |
70791ce9 FW |
709 | { |
710 | if (entry->nr < PERF_MAX_STACK_DEPTH) | |
711 | entry->ip[entry->nr++] = ip; | |
712 | } | |
394ee076 | 713 | |
cdd6c482 IM |
714 | extern int sysctl_perf_event_paranoid; |
715 | extern int sysctl_perf_event_mlock; | |
716 | extern int sysctl_perf_event_sample_rate; | |
14c63f17 DH |
717 | extern int sysctl_perf_cpu_time_max_percent; |
718 | ||
719 | extern void perf_sample_event_took(u64 sample_len_ns); | |
1ccd1549 | 720 | |
163ec435 PZ |
721 | extern int perf_proc_update_handler(struct ctl_table *table, int write, |
722 | void __user *buffer, size_t *lenp, | |
723 | loff_t *ppos); | |
14c63f17 DH |
724 | extern int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write, |
725 | void __user *buffer, size_t *lenp, | |
726 | loff_t *ppos); | |
727 | ||
163ec435 | 728 | |
320ebf09 PZ |
729 | static inline bool perf_paranoid_tracepoint_raw(void) |
730 | { | |
731 | return sysctl_perf_event_paranoid > -1; | |
732 | } | |
733 | ||
734 | static inline bool perf_paranoid_cpu(void) | |
735 | { | |
736 | return sysctl_perf_event_paranoid > 0; | |
737 | } | |
738 | ||
739 | static inline bool perf_paranoid_kernel(void) | |
740 | { | |
741 | return sysctl_perf_event_paranoid > 1; | |
742 | } | |
743 | ||
cdd6c482 | 744 | extern void perf_event_init(void); |
1c024eca PZ |
745 | extern void perf_tp_event(u64 addr, u64 count, void *record, |
746 | int entry_size, struct pt_regs *regs, | |
e6dab5ff AV |
747 | struct hlist_head *head, int rctx, |
748 | struct task_struct *task); | |
24f1e32c | 749 | extern void perf_bp_event(struct perf_event *event, void *data); |
0d905bca | 750 | |
9d23a90a | 751 | #ifndef perf_misc_flags |
e7e7ee2e IM |
752 | # define perf_misc_flags(regs) \ |
753 | (user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL) | |
754 | # define perf_instruction_pointer(regs) instruction_pointer(regs) | |
9d23a90a PM |
755 | #endif |
756 | ||
bce38cd5 SE |
757 | static inline bool has_branch_stack(struct perf_event *event) |
758 | { | |
759 | return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK; | |
760 | } | |
761 | ||
5622f295 | 762 | extern int perf_output_begin(struct perf_output_handle *handle, |
a7ac67ea | 763 | struct perf_event *event, unsigned int size); |
5622f295 | 764 | extern void perf_output_end(struct perf_output_handle *handle); |
91d7753a | 765 | extern unsigned int perf_output_copy(struct perf_output_handle *handle, |
5622f295 | 766 | const void *buf, unsigned int len); |
5685e0ff JO |
767 | extern unsigned int perf_output_skip(struct perf_output_handle *handle, |
768 | unsigned int len); | |
4ed7c92d PZ |
769 | extern int perf_swevent_get_recursion_context(void); |
770 | extern void perf_swevent_put_recursion_context(int rctx); | |
ab573844 | 771 | extern u64 perf_swevent_set_period(struct perf_event *event); |
44234adc FW |
772 | extern void perf_event_enable(struct perf_event *event); |
773 | extern void perf_event_disable(struct perf_event *event); | |
500ad2d8 | 774 | extern int __perf_event_disable(void *info); |
e9d2b064 | 775 | extern void perf_event_task_tick(void); |
0793a61d TG |
776 | #else |
777 | static inline void | |
ab0cce56 JO |
778 | perf_event_task_sched_in(struct task_struct *prev, |
779 | struct task_struct *task) { } | |
780 | static inline void | |
781 | perf_event_task_sched_out(struct task_struct *prev, | |
782 | struct task_struct *next) { } | |
cdd6c482 IM |
783 | static inline int perf_event_init_task(struct task_struct *child) { return 0; } |
784 | static inline void perf_event_exit_task(struct task_struct *child) { } | |
785 | static inline void perf_event_free_task(struct task_struct *task) { } | |
4e231c79 | 786 | static inline void perf_event_delayed_put(struct task_struct *task) { } |
57c0c15b | 787 | static inline void perf_event_print_debug(void) { } |
57c0c15b IM |
788 | static inline int perf_event_task_disable(void) { return -EINVAL; } |
789 | static inline int perf_event_task_enable(void) { return -EINVAL; } | |
26ca5c11 AK |
790 | static inline int perf_event_refresh(struct perf_event *event, int refresh) |
791 | { | |
792 | return -EINVAL; | |
793 | } | |
15dbf27c | 794 | |
925d519a | 795 | static inline void |
a8b0ca17 | 796 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) { } |
24f1e32c | 797 | static inline void |
184f412c | 798 | perf_bp_event(struct perf_event *event, void *data) { } |
0a4a9391 | 799 | |
39447b38 | 800 | static inline int perf_register_guest_info_callbacks |
e7e7ee2e | 801 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 802 | static inline int perf_unregister_guest_info_callbacks |
e7e7ee2e | 803 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 804 | |
57c0c15b | 805 | static inline void perf_event_mmap(struct vm_area_struct *vma) { } |
cdd6c482 IM |
806 | static inline void perf_event_comm(struct task_struct *tsk) { } |
807 | static inline void perf_event_fork(struct task_struct *tsk) { } | |
808 | static inline void perf_event_init(void) { } | |
184f412c | 809 | static inline int perf_swevent_get_recursion_context(void) { return -1; } |
4ed7c92d | 810 | static inline void perf_swevent_put_recursion_context(int rctx) { } |
ab573844 | 811 | static inline u64 perf_swevent_set_period(struct perf_event *event) { return 0; } |
44234adc FW |
812 | static inline void perf_event_enable(struct perf_event *event) { } |
813 | static inline void perf_event_disable(struct perf_event *event) { } | |
500ad2d8 | 814 | static inline int __perf_event_disable(void *info) { return -1; } |
e9d2b064 | 815 | static inline void perf_event_task_tick(void) { } |
0793a61d TG |
816 | #endif |
817 | ||
026249ef FW |
818 | #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_NO_HZ_FULL) |
819 | extern bool perf_event_can_stop_tick(void); | |
820 | #else | |
821 | static inline bool perf_event_can_stop_tick(void) { return true; } | |
822 | #endif | |
823 | ||
6c4d3bc9 DR |
824 | #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL) |
825 | extern void perf_restore_debug_store(void); | |
826 | #else | |
1d9d8639 | 827 | static inline void perf_restore_debug_store(void) { } |
0793a61d TG |
828 | #endif |
829 | ||
e7e7ee2e | 830 | #define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x)) |
5622f295 | 831 | |
3f6da390 | 832 | /* |
0a0fca9d | 833 | * This has to have a higher priority than migration_notifier in sched/core.c. |
3f6da390 | 834 | */ |
e7e7ee2e IM |
835 | #define perf_cpu_notifier(fn) \ |
836 | do { \ | |
0db0628d | 837 | static struct notifier_block fn##_nb = \ |
e7e7ee2e | 838 | { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ |
c13d38e4 | 839 | unsigned long cpu = smp_processor_id(); \ |
6760bca9 | 840 | unsigned long flags; \ |
f0bdb5e0 SB |
841 | \ |
842 | cpu_notifier_register_begin(); \ | |
e7e7ee2e | 843 | fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE, \ |
c13d38e4 | 844 | (void *)(unsigned long)cpu); \ |
6760bca9 | 845 | local_irq_save(flags); \ |
e7e7ee2e | 846 | fn(&fn##_nb, (unsigned long)CPU_STARTING, \ |
c13d38e4 | 847 | (void *)(unsigned long)cpu); \ |
6760bca9 | 848 | local_irq_restore(flags); \ |
e7e7ee2e | 849 | fn(&fn##_nb, (unsigned long)CPU_ONLINE, \ |
c13d38e4 | 850 | (void *)(unsigned long)cpu); \ |
f0bdb5e0 SB |
851 | __register_cpu_notifier(&fn##_nb); \ |
852 | cpu_notifier_register_done(); \ | |
3f6da390 PZ |
853 | } while (0) |
854 | ||
f0bdb5e0 SB |
855 | /* |
856 | * Bare-bones version of perf_cpu_notifier(), which doesn't invoke the | |
857 | * callback for already online CPUs. | |
858 | */ | |
859 | #define __perf_cpu_notifier(fn) \ | |
860 | do { \ | |
861 | static struct notifier_block fn##_nb = \ | |
862 | { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ | |
863 | \ | |
864 | __register_cpu_notifier(&fn##_nb); \ | |
865 | } while (0) | |
641cc938 | 866 | |
2663960c SB |
867 | struct perf_pmu_events_attr { |
868 | struct device_attribute attr; | |
869 | u64 id; | |
3a54aaa0 | 870 | const char *event_str; |
2663960c SB |
871 | }; |
872 | ||
873 | #define PMU_EVENT_ATTR(_name, _var, _id, _show) \ | |
874 | static struct perf_pmu_events_attr _var = { \ | |
875 | .attr = __ATTR(_name, 0444, _show, NULL), \ | |
876 | .id = _id, \ | |
877 | }; | |
878 | ||
641cc938 JO |
879 | #define PMU_FORMAT_ATTR(_name, _format) \ |
880 | static ssize_t \ | |
881 | _name##_show(struct device *dev, \ | |
882 | struct device_attribute *attr, \ | |
883 | char *page) \ | |
884 | { \ | |
885 | BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE); \ | |
886 | return sprintf(page, _format "\n"); \ | |
887 | } \ | |
888 | \ | |
889 | static struct device_attribute format_attr_##_name = __ATTR_RO(_name) | |
890 | ||
cdd6c482 | 891 | #endif /* _LINUX_PERF_EVENT_H */ |