]>
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 | |
53b25335 VW |
169 | /** |
170 | * pmu::capabilities flags | |
171 | */ | |
172 | #define PERF_PMU_CAP_NO_INTERRUPT 0x01 | |
173 | ||
621a01ea | 174 | /** |
4aeb0b42 | 175 | * struct pmu - generic performance monitoring unit |
621a01ea | 176 | */ |
4aeb0b42 | 177 | struct pmu { |
b0a873eb PZ |
178 | struct list_head entry; |
179 | ||
c464c76e | 180 | struct module *module; |
abe43400 | 181 | struct device *dev; |
0c9d42ed | 182 | const struct attribute_group **attr_groups; |
03d8e80b | 183 | const char *name; |
2e80a82a PZ |
184 | int type; |
185 | ||
53b25335 VW |
186 | /* |
187 | * various common per-pmu feature flags | |
188 | */ | |
189 | int capabilities; | |
190 | ||
108b02cf PZ |
191 | int * __percpu pmu_disable_count; |
192 | struct perf_cpu_context * __percpu pmu_cpu_context; | |
8dc85d54 | 193 | int task_ctx_nr; |
62b85639 | 194 | int hrtimer_interval_ms; |
6bde9b6c LM |
195 | |
196 | /* | |
a4eaf7f1 PZ |
197 | * Fully disable/enable this PMU, can be used to protect from the PMI |
198 | * as well as for lazy/batch writing of the MSRs. | |
6bde9b6c | 199 | */ |
ad5133b7 PZ |
200 | void (*pmu_enable) (struct pmu *pmu); /* optional */ |
201 | void (*pmu_disable) (struct pmu *pmu); /* optional */ | |
6bde9b6c | 202 | |
8d2cacbb | 203 | /* |
a4eaf7f1 | 204 | * Try and initialize the event for this PMU. |
24cd7f54 | 205 | * Should return -ENOENT when the @event doesn't match this PMU. |
8d2cacbb | 206 | */ |
b0a873eb PZ |
207 | int (*event_init) (struct perf_event *event); |
208 | ||
a4eaf7f1 PZ |
209 | #define PERF_EF_START 0x01 /* start the counter when adding */ |
210 | #define PERF_EF_RELOAD 0x02 /* reload the counter when starting */ | |
211 | #define PERF_EF_UPDATE 0x04 /* update the counter when stopping */ | |
212 | ||
8d2cacbb | 213 | /* |
a4eaf7f1 PZ |
214 | * Adds/Removes a counter to/from the PMU, can be done inside |
215 | * a transaction, see the ->*_txn() methods. | |
216 | */ | |
217 | int (*add) (struct perf_event *event, int flags); | |
218 | void (*del) (struct perf_event *event, int flags); | |
219 | ||
220 | /* | |
221 | * Starts/Stops a counter present on the PMU. The PMI handler | |
222 | * should stop the counter when perf_event_overflow() returns | |
223 | * !0. ->start() will be used to continue. | |
224 | */ | |
225 | void (*start) (struct perf_event *event, int flags); | |
226 | void (*stop) (struct perf_event *event, int flags); | |
227 | ||
228 | /* | |
229 | * Updates the counter value of the event. | |
230 | */ | |
cdd6c482 | 231 | void (*read) (struct perf_event *event); |
6bde9b6c LM |
232 | |
233 | /* | |
24cd7f54 PZ |
234 | * Group events scheduling is treated as a transaction, add |
235 | * group events as a whole and perform one schedulability test. | |
236 | * If the test fails, roll back the whole group | |
a4eaf7f1 PZ |
237 | * |
238 | * Start the transaction, after this ->add() doesn't need to | |
24cd7f54 | 239 | * do schedulability tests. |
8d2cacbb | 240 | */ |
e7e7ee2e | 241 | void (*start_txn) (struct pmu *pmu); /* optional */ |
8d2cacbb | 242 | /* |
a4eaf7f1 | 243 | * If ->start_txn() disabled the ->add() schedulability test |
8d2cacbb PZ |
244 | * then ->commit_txn() is required to perform one. On success |
245 | * the transaction is closed. On error the transaction is kept | |
246 | * open until ->cancel_txn() is called. | |
247 | */ | |
e7e7ee2e | 248 | int (*commit_txn) (struct pmu *pmu); /* optional */ |
8d2cacbb | 249 | /* |
a4eaf7f1 | 250 | * Will cancel the transaction, assumes ->del() is called |
25985edc | 251 | * for each successful ->add() during the transaction. |
8d2cacbb | 252 | */ |
e7e7ee2e | 253 | void (*cancel_txn) (struct pmu *pmu); /* optional */ |
35edc2a5 PZ |
254 | |
255 | /* | |
256 | * Will return the value for perf_event_mmap_page::index for this event, | |
257 | * if no implementation is provided it will default to: event->hw.idx + 1. | |
258 | */ | |
259 | int (*event_idx) (struct perf_event *event); /*optional */ | |
d010b332 SE |
260 | |
261 | /* | |
262 | * flush branch stack on context-switches (needed in cpu-wide mode) | |
263 | */ | |
264 | void (*flush_branch_stack) (void); | |
621a01ea IM |
265 | }; |
266 | ||
6a930700 | 267 | /** |
cdd6c482 | 268 | * enum perf_event_active_state - the states of a event |
6a930700 | 269 | */ |
cdd6c482 | 270 | enum perf_event_active_state { |
57c0c15b | 271 | PERF_EVENT_STATE_ERROR = -2, |
cdd6c482 IM |
272 | PERF_EVENT_STATE_OFF = -1, |
273 | PERF_EVENT_STATE_INACTIVE = 0, | |
57c0c15b | 274 | PERF_EVENT_STATE_ACTIVE = 1, |
6a930700 IM |
275 | }; |
276 | ||
9b51f66d | 277 | struct file; |
453f19ee PZ |
278 | struct perf_sample_data; |
279 | ||
a8b0ca17 | 280 | typedef void (*perf_overflow_handler_t)(struct perf_event *, |
b326e956 FW |
281 | struct perf_sample_data *, |
282 | struct pt_regs *regs); | |
283 | ||
d6f962b5 | 284 | enum perf_group_flag { |
e7e7ee2e | 285 | PERF_GROUP_SOFTWARE = 0x1, |
d6f962b5 FW |
286 | }; |
287 | ||
e7e7ee2e IM |
288 | #define SWEVENT_HLIST_BITS 8 |
289 | #define SWEVENT_HLIST_SIZE (1 << SWEVENT_HLIST_BITS) | |
76e1d904 FW |
290 | |
291 | struct swevent_hlist { | |
e7e7ee2e IM |
292 | struct hlist_head heads[SWEVENT_HLIST_SIZE]; |
293 | struct rcu_head rcu_head; | |
76e1d904 FW |
294 | }; |
295 | ||
8a49542c PZ |
296 | #define PERF_ATTACH_CONTEXT 0x01 |
297 | #define PERF_ATTACH_GROUP 0x02 | |
d580ff86 | 298 | #define PERF_ATTACH_TASK 0x04 |
8a49542c | 299 | |
877c6856 | 300 | struct perf_cgroup; |
76369139 FW |
301 | struct ring_buffer; |
302 | ||
0793a61d | 303 | /** |
cdd6c482 | 304 | * struct perf_event - performance event kernel representation: |
0793a61d | 305 | */ |
cdd6c482 IM |
306 | struct perf_event { |
307 | #ifdef CONFIG_PERF_EVENTS | |
9886167d PZ |
308 | /* |
309 | * entry onto perf_event_context::event_list; | |
310 | * modifications require ctx->lock | |
311 | * RCU safe iterations. | |
312 | */ | |
592903cd | 313 | struct list_head event_entry; |
9886167d PZ |
314 | |
315 | /* | |
316 | * XXX: group_entry and sibling_list should be mutually exclusive; | |
317 | * either you're a sibling on a group, or you're the group leader. | |
318 | * Rework the code to always use the same list element. | |
319 | * | |
320 | * Locked for modification by both ctx->mutex and ctx->lock; holding | |
321 | * either sufficies for read. | |
322 | */ | |
323 | struct list_head group_entry; | |
04289bb9 | 324 | struct list_head sibling_list; |
9886167d PZ |
325 | |
326 | /* | |
327 | * We need storage to track the entries in perf_pmu_migrate_context; we | |
328 | * cannot use the event_entry because of RCU and we want to keep the | |
329 | * group in tact which avoids us using the other two entries. | |
330 | */ | |
331 | struct list_head migrate_entry; | |
332 | ||
f3ae75de SE |
333 | struct hlist_node hlist_entry; |
334 | struct list_head active_entry; | |
0127c3ea | 335 | int nr_siblings; |
d6f962b5 | 336 | int group_flags; |
cdd6c482 | 337 | struct perf_event *group_leader; |
a4eaf7f1 | 338 | struct pmu *pmu; |
04289bb9 | 339 | |
cdd6c482 | 340 | enum perf_event_active_state state; |
8a49542c | 341 | unsigned int attach_state; |
e7850595 | 342 | local64_t count; |
a6e6dea6 | 343 | atomic64_t child_count; |
ee06094f | 344 | |
53cfbf59 | 345 | /* |
cdd6c482 | 346 | * These are the total time in nanoseconds that the event |
53cfbf59 | 347 | * has been enabled (i.e. eligible to run, and the task has |
cdd6c482 | 348 | * been scheduled in, if this is a per-task event) |
53cfbf59 PM |
349 | * and running (scheduled onto the CPU), respectively. |
350 | * | |
351 | * They are computed from tstamp_enabled, tstamp_running and | |
cdd6c482 | 352 | * tstamp_stopped when the event is in INACTIVE or ACTIVE state. |
53cfbf59 PM |
353 | */ |
354 | u64 total_time_enabled; | |
355 | u64 total_time_running; | |
356 | ||
357 | /* | |
358 | * These are timestamps used for computing total_time_enabled | |
cdd6c482 | 359 | * and total_time_running when the event is in INACTIVE or |
53cfbf59 PM |
360 | * ACTIVE state, measured in nanoseconds from an arbitrary point |
361 | * in time. | |
cdd6c482 IM |
362 | * tstamp_enabled: the notional time when the event was enabled |
363 | * tstamp_running: the notional time when the event was scheduled on | |
53cfbf59 | 364 | * tstamp_stopped: in INACTIVE state, the notional time when the |
cdd6c482 | 365 | * event was scheduled off. |
53cfbf59 PM |
366 | */ |
367 | u64 tstamp_enabled; | |
368 | u64 tstamp_running; | |
369 | u64 tstamp_stopped; | |
370 | ||
eed01528 SE |
371 | /* |
372 | * timestamp shadows the actual context timing but it can | |
373 | * be safely used in NMI interrupt context. It reflects the | |
374 | * context time as it was when the event was last scheduled in. | |
375 | * | |
376 | * ctx_time already accounts for ctx->timestamp. Therefore to | |
377 | * compute ctx_time for a sample, simply add perf_clock(). | |
378 | */ | |
379 | u64 shadow_ctx_time; | |
380 | ||
24f1e32c | 381 | struct perf_event_attr attr; |
c320c7b7 | 382 | u16 header_size; |
6844c09d | 383 | u16 id_header_size; |
c320c7b7 | 384 | u16 read_size; |
cdd6c482 | 385 | struct hw_perf_event hw; |
0793a61d | 386 | |
cdd6c482 | 387 | struct perf_event_context *ctx; |
a6fa941d | 388 | atomic_long_t refcount; |
0793a61d | 389 | |
53cfbf59 PM |
390 | /* |
391 | * These accumulate total time (in nanoseconds) that children | |
cdd6c482 | 392 | * events have been enabled and running, respectively. |
53cfbf59 PM |
393 | */ |
394 | atomic64_t child_total_time_enabled; | |
395 | atomic64_t child_total_time_running; | |
396 | ||
0793a61d | 397 | /* |
d859e29f | 398 | * Protect attach/detach and child_list: |
0793a61d | 399 | */ |
fccc714b PZ |
400 | struct mutex child_mutex; |
401 | struct list_head child_list; | |
cdd6c482 | 402 | struct perf_event *parent; |
0793a61d TG |
403 | |
404 | int oncpu; | |
405 | int cpu; | |
406 | ||
082ff5a2 PZ |
407 | struct list_head owner_entry; |
408 | struct task_struct *owner; | |
409 | ||
7b732a75 PZ |
410 | /* mmap bits */ |
411 | struct mutex mmap_mutex; | |
412 | atomic_t mmap_count; | |
26cb63ad | 413 | |
76369139 | 414 | struct ring_buffer *rb; |
10c6db11 | 415 | struct list_head rb_entry; |
b69cf536 PZ |
416 | unsigned long rcu_batches; |
417 | int rcu_pending; | |
37d81828 | 418 | |
7b732a75 | 419 | /* poll related */ |
0793a61d | 420 | wait_queue_head_t waitq; |
3c446b3d | 421 | struct fasync_struct *fasync; |
79f14641 PZ |
422 | |
423 | /* delayed work for NMIs and such */ | |
424 | int pending_wakeup; | |
4c9e2542 | 425 | int pending_kill; |
79f14641 | 426 | int pending_disable; |
e360adbe | 427 | struct irq_work pending; |
592903cd | 428 | |
79f14641 PZ |
429 | atomic_t event_limit; |
430 | ||
cdd6c482 | 431 | void (*destroy)(struct perf_event *); |
592903cd | 432 | struct rcu_head rcu_head; |
709e50cf PZ |
433 | |
434 | struct pid_namespace *ns; | |
8e5799b1 | 435 | u64 id; |
6fb2915d | 436 | |
b326e956 | 437 | perf_overflow_handler_t overflow_handler; |
4dc0da86 | 438 | void *overflow_handler_context; |
453f19ee | 439 | |
07b139c8 | 440 | #ifdef CONFIG_EVENT_TRACING |
1c024eca | 441 | struct ftrace_event_call *tp_event; |
6fb2915d | 442 | struct event_filter *filter; |
ced39002 JO |
443 | #ifdef CONFIG_FUNCTION_TRACER |
444 | struct ftrace_ops ftrace_ops; | |
445 | #endif | |
ee06094f | 446 | #endif |
6fb2915d | 447 | |
e5d1367f SE |
448 | #ifdef CONFIG_CGROUP_PERF |
449 | struct perf_cgroup *cgrp; /* cgroup event is attach to */ | |
450 | int cgrp_defer_enabled; | |
451 | #endif | |
452 | ||
6fb2915d | 453 | #endif /* CONFIG_PERF_EVENTS */ |
0793a61d TG |
454 | }; |
455 | ||
b04243ef PZ |
456 | enum perf_event_context_type { |
457 | task_context, | |
458 | cpu_context, | |
459 | }; | |
460 | ||
0793a61d | 461 | /** |
cdd6c482 | 462 | * struct perf_event_context - event context structure |
0793a61d | 463 | * |
cdd6c482 | 464 | * Used as a container for task events and CPU events as well: |
0793a61d | 465 | */ |
cdd6c482 | 466 | struct perf_event_context { |
108b02cf | 467 | struct pmu *pmu; |
ee643c41 | 468 | enum perf_event_context_type type; |
0793a61d | 469 | /* |
cdd6c482 | 470 | * Protect the states of the events in the list, |
d859e29f | 471 | * nr_active, and the list: |
0793a61d | 472 | */ |
e625cce1 | 473 | raw_spinlock_t lock; |
d859e29f | 474 | /* |
cdd6c482 | 475 | * Protect the list of events. Locking either mutex or lock |
d859e29f PM |
476 | * is sufficient to ensure the list doesn't change; to change |
477 | * the list you need to lock both the mutex and the spinlock. | |
478 | */ | |
a308444c | 479 | struct mutex mutex; |
04289bb9 | 480 | |
889ff015 FW |
481 | struct list_head pinned_groups; |
482 | struct list_head flexible_groups; | |
a308444c | 483 | struct list_head event_list; |
cdd6c482 | 484 | int nr_events; |
a308444c IM |
485 | int nr_active; |
486 | int is_active; | |
bfbd3381 | 487 | int nr_stat; |
0f5a2601 | 488 | int nr_freq; |
dddd3379 | 489 | int rotate_disable; |
a308444c IM |
490 | atomic_t refcount; |
491 | struct task_struct *task; | |
53cfbf59 PM |
492 | |
493 | /* | |
4af4998b | 494 | * Context clock, runs when context enabled. |
53cfbf59 | 495 | */ |
a308444c IM |
496 | u64 time; |
497 | u64 timestamp; | |
564c2b21 PM |
498 | |
499 | /* | |
500 | * These fields let us detect when two contexts have both | |
501 | * been cloned (inherited) from a common ancestor. | |
502 | */ | |
cdd6c482 | 503 | struct perf_event_context *parent_ctx; |
a308444c IM |
504 | u64 parent_gen; |
505 | u64 generation; | |
506 | int pin_count; | |
d010b332 SE |
507 | int nr_cgroups; /* cgroup evts */ |
508 | int nr_branch_stack; /* branch_stack evt */ | |
28009ce4 | 509 | struct rcu_head rcu_head; |
0793a61d TG |
510 | }; |
511 | ||
7ae07ea3 FW |
512 | /* |
513 | * Number of contexts where an event can trigger: | |
e7e7ee2e | 514 | * task, softirq, hardirq, nmi. |
7ae07ea3 FW |
515 | */ |
516 | #define PERF_NR_CONTEXTS 4 | |
517 | ||
0793a61d | 518 | /** |
cdd6c482 | 519 | * struct perf_event_cpu_context - per cpu event context structure |
0793a61d TG |
520 | */ |
521 | struct perf_cpu_context { | |
cdd6c482 IM |
522 | struct perf_event_context ctx; |
523 | struct perf_event_context *task_ctx; | |
0793a61d | 524 | int active_oncpu; |
3b6f9e5c | 525 | int exclusive; |
9e630205 SE |
526 | struct hrtimer hrtimer; |
527 | ktime_t hrtimer_interval; | |
e9d2b064 | 528 | struct list_head rotation_list; |
3f1f3320 | 529 | struct pmu *unique_pmu; |
e5d1367f | 530 | struct perf_cgroup *cgrp; |
0793a61d TG |
531 | }; |
532 | ||
5622f295 | 533 | struct perf_output_handle { |
57c0c15b | 534 | struct perf_event *event; |
76369139 | 535 | struct ring_buffer *rb; |
6d1acfd5 | 536 | unsigned long wakeup; |
5d967a8b PZ |
537 | unsigned long size; |
538 | void *addr; | |
539 | int page; | |
5622f295 MM |
540 | }; |
541 | ||
cdd6c482 | 542 | #ifdef CONFIG_PERF_EVENTS |
829b42dd | 543 | |
03d8e80b | 544 | extern int perf_pmu_register(struct pmu *pmu, const char *name, int type); |
b0a873eb | 545 | extern void perf_pmu_unregister(struct pmu *pmu); |
621a01ea | 546 | |
3bf101ba | 547 | extern int perf_num_counters(void); |
84c79910 | 548 | extern const char *perf_pmu_name(void); |
ab0cce56 JO |
549 | extern void __perf_event_task_sched_in(struct task_struct *prev, |
550 | struct task_struct *task); | |
551 | extern void __perf_event_task_sched_out(struct task_struct *prev, | |
552 | struct task_struct *next); | |
cdd6c482 IM |
553 | extern int perf_event_init_task(struct task_struct *child); |
554 | extern void perf_event_exit_task(struct task_struct *child); | |
555 | extern void perf_event_free_task(struct task_struct *task); | |
4e231c79 | 556 | extern void perf_event_delayed_put(struct task_struct *task); |
cdd6c482 | 557 | extern void perf_event_print_debug(void); |
33696fc0 PZ |
558 | extern void perf_pmu_disable(struct pmu *pmu); |
559 | extern void perf_pmu_enable(struct pmu *pmu); | |
cdd6c482 IM |
560 | extern int perf_event_task_disable(void); |
561 | extern int perf_event_task_enable(void); | |
26ca5c11 | 562 | extern int perf_event_refresh(struct perf_event *event, int refresh); |
cdd6c482 | 563 | extern void perf_event_update_userpage(struct perf_event *event); |
fb0459d7 AV |
564 | extern int perf_event_release_kernel(struct perf_event *event); |
565 | extern struct perf_event * | |
566 | perf_event_create_kernel_counter(struct perf_event_attr *attr, | |
567 | int cpu, | |
38a81da2 | 568 | struct task_struct *task, |
4dc0da86 AK |
569 | perf_overflow_handler_t callback, |
570 | void *context); | |
0cda4c02 YZ |
571 | extern void perf_pmu_migrate_context(struct pmu *pmu, |
572 | int src_cpu, int dst_cpu); | |
59ed446f PZ |
573 | extern u64 perf_event_read_value(struct perf_event *event, |
574 | u64 *enabled, u64 *running); | |
5c92d124 | 575 | |
d010b332 | 576 | |
df1a132b | 577 | struct perf_sample_data { |
5622f295 MM |
578 | u64 type; |
579 | ||
580 | u64 ip; | |
581 | struct { | |
582 | u32 pid; | |
583 | u32 tid; | |
584 | } tid_entry; | |
585 | u64 time; | |
a308444c | 586 | u64 addr; |
5622f295 MM |
587 | u64 id; |
588 | u64 stream_id; | |
589 | struct { | |
590 | u32 cpu; | |
591 | u32 reserved; | |
592 | } cpu_entry; | |
a308444c | 593 | u64 period; |
d6be9ad6 | 594 | union perf_mem_data_src data_src; |
5622f295 | 595 | struct perf_callchain_entry *callchain; |
3a43ce68 | 596 | struct perf_raw_record *raw; |
bce38cd5 | 597 | struct perf_branch_stack *br_stack; |
4018994f | 598 | struct perf_regs_user regs_user; |
c5ebcedb | 599 | u64 stack_user_size; |
c3feedf2 | 600 | u64 weight; |
fdfbbd07 AK |
601 | /* |
602 | * Transaction flags for abort events: | |
603 | */ | |
604 | u64 txn; | |
df1a132b PZ |
605 | }; |
606 | ||
fd0d000b RR |
607 | static inline void perf_sample_data_init(struct perf_sample_data *data, |
608 | u64 addr, u64 period) | |
dc1d628a | 609 | { |
fd0d000b | 610 | /* remaining struct members initialized in perf_prepare_sample() */ |
dc1d628a PZ |
611 | data->addr = addr; |
612 | data->raw = NULL; | |
bce38cd5 | 613 | data->br_stack = NULL; |
4018994f JO |
614 | data->period = period; |
615 | data->regs_user.abi = PERF_SAMPLE_REGS_ABI_NONE; | |
616 | data->regs_user.regs = NULL; | |
c5ebcedb | 617 | data->stack_user_size = 0; |
c3feedf2 | 618 | data->weight = 0; |
d6be9ad6 | 619 | data->data_src.val = 0; |
fdfbbd07 | 620 | data->txn = 0; |
dc1d628a PZ |
621 | } |
622 | ||
5622f295 MM |
623 | extern void perf_output_sample(struct perf_output_handle *handle, |
624 | struct perf_event_header *header, | |
625 | struct perf_sample_data *data, | |
cdd6c482 | 626 | struct perf_event *event); |
5622f295 MM |
627 | extern void perf_prepare_sample(struct perf_event_header *header, |
628 | struct perf_sample_data *data, | |
cdd6c482 | 629 | struct perf_event *event, |
5622f295 MM |
630 | struct pt_regs *regs); |
631 | ||
a8b0ca17 | 632 | extern int perf_event_overflow(struct perf_event *event, |
5622f295 MM |
633 | struct perf_sample_data *data, |
634 | struct pt_regs *regs); | |
df1a132b | 635 | |
6c7e550f FBH |
636 | static inline bool is_sampling_event(struct perf_event *event) |
637 | { | |
638 | return event->attr.sample_period != 0; | |
639 | } | |
640 | ||
3b6f9e5c | 641 | /* |
cdd6c482 | 642 | * Return 1 for a software event, 0 for a hardware event |
3b6f9e5c | 643 | */ |
cdd6c482 | 644 | static inline int is_software_event(struct perf_event *event) |
3b6f9e5c | 645 | { |
89a1e187 | 646 | return event->pmu->task_ctx_nr == perf_sw_context; |
3b6f9e5c PM |
647 | } |
648 | ||
c5905afb | 649 | extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX]; |
f29ac756 | 650 | |
a8b0ca17 | 651 | extern void __perf_sw_event(u32, u64, struct pt_regs *, u64); |
f29ac756 | 652 | |
b0f82b81 | 653 | #ifndef perf_arch_fetch_caller_regs |
e7e7ee2e | 654 | static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { } |
b0f82b81 | 655 | #endif |
5331d7b8 FW |
656 | |
657 | /* | |
658 | * Take a snapshot of the regs. Skip ip and frame pointer to | |
659 | * the nth caller. We only need a few of the regs: | |
660 | * - ip for PERF_SAMPLE_IP | |
661 | * - cs for user_mode() tests | |
662 | * - bp for callchains | |
663 | * - eflags, for future purposes, just in case | |
664 | */ | |
b0f82b81 | 665 | static inline void perf_fetch_caller_regs(struct pt_regs *regs) |
5331d7b8 | 666 | { |
5331d7b8 FW |
667 | memset(regs, 0, sizeof(*regs)); |
668 | ||
b0f82b81 | 669 | perf_arch_fetch_caller_regs(regs, CALLER_ADDR0); |
5331d7b8 FW |
670 | } |
671 | ||
7e54a5a0 | 672 | static __always_inline void |
a8b0ca17 | 673 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) |
e49a5bd3 | 674 | { |
7e54a5a0 PZ |
675 | struct pt_regs hot_regs; |
676 | ||
c5905afb | 677 | if (static_key_false(&perf_swevent_enabled[event_id])) { |
d430d3d7 JB |
678 | if (!regs) { |
679 | perf_fetch_caller_regs(&hot_regs); | |
680 | regs = &hot_regs; | |
681 | } | |
a8b0ca17 | 682 | __perf_sw_event(event_id, nr, regs, addr); |
e49a5bd3 FW |
683 | } |
684 | } | |
685 | ||
c5905afb | 686 | extern struct static_key_deferred perf_sched_events; |
ee6dcfa4 | 687 | |
ab0cce56 | 688 | static inline void perf_event_task_sched_in(struct task_struct *prev, |
a8d757ef | 689 | struct task_struct *task) |
ab0cce56 JO |
690 | { |
691 | if (static_key_false(&perf_sched_events.key)) | |
692 | __perf_event_task_sched_in(prev, task); | |
693 | } | |
694 | ||
695 | static inline void perf_event_task_sched_out(struct task_struct *prev, | |
696 | struct task_struct *next) | |
ee6dcfa4 | 697 | { |
a8b0ca17 | 698 | perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, NULL, 0); |
ee6dcfa4 | 699 | |
c5905afb | 700 | if (static_key_false(&perf_sched_events.key)) |
ab0cce56 | 701 | __perf_event_task_sched_out(prev, next); |
ee6dcfa4 PZ |
702 | } |
703 | ||
3af9e859 | 704 | extern void perf_event_mmap(struct vm_area_struct *vma); |
39447b38 | 705 | extern struct perf_guest_info_callbacks *perf_guest_cbs; |
dcf46b94 ZY |
706 | extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); |
707 | extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); | |
39447b38 | 708 | |
e041e328 | 709 | extern void perf_event_exec(void); |
cdd6c482 IM |
710 | extern void perf_event_comm(struct task_struct *tsk); |
711 | extern void perf_event_fork(struct task_struct *tsk); | |
8d1b2d93 | 712 | |
56962b44 FW |
713 | /* Callchains */ |
714 | DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry); | |
715 | ||
e7e7ee2e IM |
716 | extern void perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs); |
717 | extern void perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs); | |
394ee076 | 718 | |
e7e7ee2e | 719 | static inline void perf_callchain_store(struct perf_callchain_entry *entry, u64 ip) |
70791ce9 FW |
720 | { |
721 | if (entry->nr < PERF_MAX_STACK_DEPTH) | |
722 | entry->ip[entry->nr++] = ip; | |
723 | } | |
394ee076 | 724 | |
cdd6c482 IM |
725 | extern int sysctl_perf_event_paranoid; |
726 | extern int sysctl_perf_event_mlock; | |
727 | extern int sysctl_perf_event_sample_rate; | |
14c63f17 DH |
728 | extern int sysctl_perf_cpu_time_max_percent; |
729 | ||
730 | extern void perf_sample_event_took(u64 sample_len_ns); | |
1ccd1549 | 731 | |
163ec435 PZ |
732 | extern int perf_proc_update_handler(struct ctl_table *table, int write, |
733 | void __user *buffer, size_t *lenp, | |
734 | loff_t *ppos); | |
14c63f17 DH |
735 | extern int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write, |
736 | void __user *buffer, size_t *lenp, | |
737 | loff_t *ppos); | |
738 | ||
163ec435 | 739 | |
320ebf09 PZ |
740 | static inline bool perf_paranoid_tracepoint_raw(void) |
741 | { | |
742 | return sysctl_perf_event_paranoid > -1; | |
743 | } | |
744 | ||
745 | static inline bool perf_paranoid_cpu(void) | |
746 | { | |
747 | return sysctl_perf_event_paranoid > 0; | |
748 | } | |
749 | ||
750 | static inline bool perf_paranoid_kernel(void) | |
751 | { | |
752 | return sysctl_perf_event_paranoid > 1; | |
753 | } | |
754 | ||
cdd6c482 | 755 | extern void perf_event_init(void); |
1c024eca PZ |
756 | extern void perf_tp_event(u64 addr, u64 count, void *record, |
757 | int entry_size, struct pt_regs *regs, | |
e6dab5ff AV |
758 | struct hlist_head *head, int rctx, |
759 | struct task_struct *task); | |
24f1e32c | 760 | extern void perf_bp_event(struct perf_event *event, void *data); |
0d905bca | 761 | |
9d23a90a | 762 | #ifndef perf_misc_flags |
e7e7ee2e IM |
763 | # define perf_misc_flags(regs) \ |
764 | (user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL) | |
765 | # define perf_instruction_pointer(regs) instruction_pointer(regs) | |
9d23a90a PM |
766 | #endif |
767 | ||
bce38cd5 SE |
768 | static inline bool has_branch_stack(struct perf_event *event) |
769 | { | |
770 | return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK; | |
771 | } | |
772 | ||
5622f295 | 773 | extern int perf_output_begin(struct perf_output_handle *handle, |
a7ac67ea | 774 | struct perf_event *event, unsigned int size); |
5622f295 | 775 | extern void perf_output_end(struct perf_output_handle *handle); |
91d7753a | 776 | extern unsigned int perf_output_copy(struct perf_output_handle *handle, |
5622f295 | 777 | const void *buf, unsigned int len); |
5685e0ff JO |
778 | extern unsigned int perf_output_skip(struct perf_output_handle *handle, |
779 | unsigned int len); | |
4ed7c92d PZ |
780 | extern int perf_swevent_get_recursion_context(void); |
781 | extern void perf_swevent_put_recursion_context(int rctx); | |
ab573844 | 782 | extern u64 perf_swevent_set_period(struct perf_event *event); |
44234adc FW |
783 | extern void perf_event_enable(struct perf_event *event); |
784 | extern void perf_event_disable(struct perf_event *event); | |
500ad2d8 | 785 | extern int __perf_event_disable(void *info); |
e9d2b064 | 786 | extern void perf_event_task_tick(void); |
e041e328 | 787 | #else /* !CONFIG_PERF_EVENTS: */ |
0793a61d | 788 | static inline void |
ab0cce56 JO |
789 | perf_event_task_sched_in(struct task_struct *prev, |
790 | struct task_struct *task) { } | |
791 | static inline void | |
792 | perf_event_task_sched_out(struct task_struct *prev, | |
793 | struct task_struct *next) { } | |
cdd6c482 IM |
794 | static inline int perf_event_init_task(struct task_struct *child) { return 0; } |
795 | static inline void perf_event_exit_task(struct task_struct *child) { } | |
796 | static inline void perf_event_free_task(struct task_struct *task) { } | |
4e231c79 | 797 | static inline void perf_event_delayed_put(struct task_struct *task) { } |
57c0c15b | 798 | static inline void perf_event_print_debug(void) { } |
57c0c15b IM |
799 | static inline int perf_event_task_disable(void) { return -EINVAL; } |
800 | static inline int perf_event_task_enable(void) { return -EINVAL; } | |
26ca5c11 AK |
801 | static inline int perf_event_refresh(struct perf_event *event, int refresh) |
802 | { | |
803 | return -EINVAL; | |
804 | } | |
15dbf27c | 805 | |
925d519a | 806 | static inline void |
a8b0ca17 | 807 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) { } |
24f1e32c | 808 | static inline void |
184f412c | 809 | perf_bp_event(struct perf_event *event, void *data) { } |
0a4a9391 | 810 | |
39447b38 | 811 | static inline int perf_register_guest_info_callbacks |
e7e7ee2e | 812 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 813 | static inline int perf_unregister_guest_info_callbacks |
e7e7ee2e | 814 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 815 | |
57c0c15b | 816 | static inline void perf_event_mmap(struct vm_area_struct *vma) { } |
e041e328 | 817 | static inline void perf_event_exec(void) { } |
cdd6c482 IM |
818 | static inline void perf_event_comm(struct task_struct *tsk) { } |
819 | static inline void perf_event_fork(struct task_struct *tsk) { } | |
820 | static inline void perf_event_init(void) { } | |
184f412c | 821 | static inline int perf_swevent_get_recursion_context(void) { return -1; } |
4ed7c92d | 822 | static inline void perf_swevent_put_recursion_context(int rctx) { } |
ab573844 | 823 | static inline u64 perf_swevent_set_period(struct perf_event *event) { return 0; } |
44234adc FW |
824 | static inline void perf_event_enable(struct perf_event *event) { } |
825 | static inline void perf_event_disable(struct perf_event *event) { } | |
500ad2d8 | 826 | static inline int __perf_event_disable(void *info) { return -1; } |
e9d2b064 | 827 | static inline void perf_event_task_tick(void) { } |
0793a61d TG |
828 | #endif |
829 | ||
026249ef FW |
830 | #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_NO_HZ_FULL) |
831 | extern bool perf_event_can_stop_tick(void); | |
832 | #else | |
833 | static inline bool perf_event_can_stop_tick(void) { return true; } | |
834 | #endif | |
835 | ||
6c4d3bc9 DR |
836 | #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL) |
837 | extern void perf_restore_debug_store(void); | |
838 | #else | |
1d9d8639 | 839 | static inline void perf_restore_debug_store(void) { } |
0793a61d TG |
840 | #endif |
841 | ||
e7e7ee2e | 842 | #define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x)) |
5622f295 | 843 | |
3f6da390 | 844 | /* |
0a0fca9d | 845 | * This has to have a higher priority than migration_notifier in sched/core.c. |
3f6da390 | 846 | */ |
e7e7ee2e IM |
847 | #define perf_cpu_notifier(fn) \ |
848 | do { \ | |
0db0628d | 849 | static struct notifier_block fn##_nb = \ |
e7e7ee2e | 850 | { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ |
c13d38e4 | 851 | unsigned long cpu = smp_processor_id(); \ |
6760bca9 | 852 | unsigned long flags; \ |
f0bdb5e0 SB |
853 | \ |
854 | cpu_notifier_register_begin(); \ | |
e7e7ee2e | 855 | fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE, \ |
c13d38e4 | 856 | (void *)(unsigned long)cpu); \ |
6760bca9 | 857 | local_irq_save(flags); \ |
e7e7ee2e | 858 | fn(&fn##_nb, (unsigned long)CPU_STARTING, \ |
c13d38e4 | 859 | (void *)(unsigned long)cpu); \ |
6760bca9 | 860 | local_irq_restore(flags); \ |
e7e7ee2e | 861 | fn(&fn##_nb, (unsigned long)CPU_ONLINE, \ |
c13d38e4 | 862 | (void *)(unsigned long)cpu); \ |
f0bdb5e0 SB |
863 | __register_cpu_notifier(&fn##_nb); \ |
864 | cpu_notifier_register_done(); \ | |
3f6da390 PZ |
865 | } while (0) |
866 | ||
f0bdb5e0 SB |
867 | /* |
868 | * Bare-bones version of perf_cpu_notifier(), which doesn't invoke the | |
869 | * callback for already online CPUs. | |
870 | */ | |
871 | #define __perf_cpu_notifier(fn) \ | |
872 | do { \ | |
873 | static struct notifier_block fn##_nb = \ | |
874 | { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ | |
875 | \ | |
876 | __register_cpu_notifier(&fn##_nb); \ | |
877 | } while (0) | |
641cc938 | 878 | |
2663960c SB |
879 | struct perf_pmu_events_attr { |
880 | struct device_attribute attr; | |
881 | u64 id; | |
3a54aaa0 | 882 | const char *event_str; |
2663960c SB |
883 | }; |
884 | ||
885 | #define PMU_EVENT_ATTR(_name, _var, _id, _show) \ | |
886 | static struct perf_pmu_events_attr _var = { \ | |
887 | .attr = __ATTR(_name, 0444, _show, NULL), \ | |
888 | .id = _id, \ | |
889 | }; | |
890 | ||
641cc938 JO |
891 | #define PMU_FORMAT_ATTR(_name, _format) \ |
892 | static ssize_t \ | |
893 | _name##_show(struct device *dev, \ | |
894 | struct device_attribute *attr, \ | |
895 | char *page) \ | |
896 | { \ | |
897 | BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE); \ | |
898 | return sprintf(page, _format "\n"); \ | |
899 | } \ | |
900 | \ | |
901 | static struct device_attribute format_attr_##_name = __ATTR_RO(_name) | |
902 | ||
cdd6c482 | 903 | #endif /* _LINUX_PERF_EVENT_H */ |