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 | 23 | #ifdef CONFIG_PERF_EVENTS |
e5d1367f | 24 | # include <linux/cgroup.h> |
cdd6c482 | 25 | # include <asm/perf_event.h> |
7be79236 | 26 | # include <asm/local64.h> |
f3dfd265 PM |
27 | #endif |
28 | ||
39447b38 | 29 | struct perf_guest_info_callbacks { |
e7e7ee2e IM |
30 | int (*is_in_guest)(void); |
31 | int (*is_user_mode)(void); | |
32 | unsigned long (*get_guest_ip)(void); | |
39447b38 ZY |
33 | }; |
34 | ||
2ff6cfd7 AB |
35 | #ifdef CONFIG_HAVE_HW_BREAKPOINT |
36 | #include <asm/hw_breakpoint.h> | |
37 | #endif | |
38 | ||
f3dfd265 PM |
39 | #include <linux/list.h> |
40 | #include <linux/mutex.h> | |
41 | #include <linux/rculist.h> | |
42 | #include <linux/rcupdate.h> | |
43 | #include <linux/spinlock.h> | |
d6d020e9 | 44 | #include <linux/hrtimer.h> |
3c446b3d | 45 | #include <linux/fs.h> |
709e50cf | 46 | #include <linux/pid_namespace.h> |
906010b2 | 47 | #include <linux/workqueue.h> |
5331d7b8 | 48 | #include <linux/ftrace.h> |
85cfabbc | 49 | #include <linux/cpu.h> |
e360adbe | 50 | #include <linux/irq_work.h> |
c5905afb | 51 | #include <linux/static_key.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 | * single taken branch record layout: | |
69 | * | |
70 | * from: source instruction (may not always be a branch insn) | |
71 | * to: branch target | |
72 | * mispred: branch target was mispredicted | |
73 | * predicted: branch target was predicted | |
74 | * | |
75 | * support for mispred, predicted is optional. In case it | |
76 | * is not supported mispred = predicted = 0. | |
77 | */ | |
caff2bef | 78 | struct perf_branch_entry { |
bce38cd5 SE |
79 | __u64 from; |
80 | __u64 to; | |
81 | __u64 mispred:1, /* target mispredicted */ | |
82 | predicted:1,/* target predicted */ | |
83 | reserved:62; | |
caff2bef PZ |
84 | }; |
85 | ||
bce38cd5 SE |
86 | /* |
87 | * branch stack layout: | |
88 | * nr: number of taken branches stored in entries[] | |
89 | * | |
90 | * Note that nr can vary from sample to sample | |
91 | * branches (to, from) are stored from most recent | |
92 | * to least recent, i.e., entries[0] contains the most | |
93 | * recent branch. | |
94 | */ | |
caff2bef PZ |
95 | struct perf_branch_stack { |
96 | __u64 nr; | |
97 | struct perf_branch_entry entries[0]; | |
98 | }; | |
99 | ||
4018994f JO |
100 | struct perf_regs_user { |
101 | __u64 abi; | |
102 | struct pt_regs *regs; | |
103 | }; | |
104 | ||
f3dfd265 PM |
105 | struct task_struct; |
106 | ||
efc9f05d SE |
107 | /* |
108 | * extra PMU register associated with an event | |
109 | */ | |
110 | struct hw_perf_event_extra { | |
111 | u64 config; /* register value */ | |
112 | unsigned int reg; /* register address or index */ | |
113 | int alloc; /* extra register already allocated */ | |
114 | int idx; /* index in shared_regs->regs[] */ | |
115 | }; | |
116 | ||
0793a61d | 117 | /** |
cdd6c482 | 118 | * struct hw_perf_event - performance event hardware details: |
0793a61d | 119 | */ |
cdd6c482 IM |
120 | struct hw_perf_event { |
121 | #ifdef CONFIG_PERF_EVENTS | |
d6d020e9 PZ |
122 | union { |
123 | struct { /* hardware */ | |
a308444c | 124 | u64 config; |
447a194b | 125 | u64 last_tag; |
a308444c | 126 | unsigned long config_base; |
cdd6c482 | 127 | unsigned long event_base; |
c48b6053 | 128 | int event_base_rdpmc; |
a308444c | 129 | int idx; |
447a194b | 130 | int last_cpu; |
bce38cd5 | 131 | |
efc9f05d | 132 | struct hw_perf_event_extra extra_reg; |
bce38cd5 | 133 | struct hw_perf_event_extra branch_reg; |
d6d020e9 | 134 | }; |
721a669b | 135 | struct { /* software */ |
a308444c | 136 | struct hrtimer hrtimer; |
d6d020e9 | 137 | }; |
f22c1bb6 ON |
138 | struct { /* tracepoint */ |
139 | struct task_struct *tp_target; | |
140 | /* for tp_event->class */ | |
141 | struct list_head tp_list; | |
142 | }; | |
24f1e32c | 143 | #ifdef CONFIG_HAVE_HW_BREAKPOINT |
45a73372 | 144 | struct { /* breakpoint */ |
d580ff86 PZ |
145 | /* |
146 | * Crufty hack to avoid the chicken and egg | |
147 | * problem hw_breakpoint has with context | |
148 | * creation and event initalization. | |
149 | */ | |
150 | struct task_struct *bp_target; | |
f22c1bb6 ON |
151 | struct arch_hw_breakpoint info; |
152 | struct list_head bp_list; | |
45a73372 | 153 | }; |
24f1e32c | 154 | #endif |
d6d020e9 | 155 | }; |
a4eaf7f1 | 156 | int state; |
e7850595 | 157 | local64_t prev_count; |
b23f3325 | 158 | u64 sample_period; |
9e350de3 | 159 | u64 last_period; |
e7850595 | 160 | local64_t period_left; |
e050e3f0 | 161 | u64 interrupts_seq; |
60db5e09 | 162 | u64 interrupts; |
6a24ed6c | 163 | |
abd50713 PZ |
164 | u64 freq_time_stamp; |
165 | u64 freq_count_stamp; | |
ee06094f | 166 | #endif |
0793a61d TG |
167 | }; |
168 | ||
a4eaf7f1 PZ |
169 | /* |
170 | * hw_perf_event::state flags | |
171 | */ | |
172 | #define PERF_HES_STOPPED 0x01 /* the counter is stopped */ | |
173 | #define PERF_HES_UPTODATE 0x02 /* event->count up-to-date */ | |
174 | #define PERF_HES_ARCH 0x04 | |
175 | ||
cdd6c482 | 176 | struct perf_event; |
621a01ea | 177 | |
8d2cacbb PZ |
178 | /* |
179 | * Common implementation detail of pmu::{start,commit,cancel}_txn | |
180 | */ | |
181 | #define PERF_EVENT_TXN 0x1 | |
6bde9b6c | 182 | |
621a01ea | 183 | /** |
4aeb0b42 | 184 | * struct pmu - generic performance monitoring unit |
621a01ea | 185 | */ |
4aeb0b42 | 186 | struct pmu { |
b0a873eb PZ |
187 | struct list_head entry; |
188 | ||
abe43400 | 189 | struct device *dev; |
0c9d42ed | 190 | const struct attribute_group **attr_groups; |
2e80a82a PZ |
191 | char *name; |
192 | int type; | |
193 | ||
108b02cf PZ |
194 | int * __percpu pmu_disable_count; |
195 | struct perf_cpu_context * __percpu pmu_cpu_context; | |
8dc85d54 | 196 | int task_ctx_nr; |
6bde9b6c LM |
197 | |
198 | /* | |
a4eaf7f1 PZ |
199 | * Fully disable/enable this PMU, can be used to protect from the PMI |
200 | * as well as for lazy/batch writing of the MSRs. | |
6bde9b6c | 201 | */ |
ad5133b7 PZ |
202 | void (*pmu_enable) (struct pmu *pmu); /* optional */ |
203 | void (*pmu_disable) (struct pmu *pmu); /* optional */ | |
6bde9b6c | 204 | |
8d2cacbb | 205 | /* |
a4eaf7f1 | 206 | * Try and initialize the event for this PMU. |
24cd7f54 | 207 | * Should return -ENOENT when the @event doesn't match this PMU. |
8d2cacbb | 208 | */ |
b0a873eb PZ |
209 | int (*event_init) (struct perf_event *event); |
210 | ||
a4eaf7f1 PZ |
211 | #define PERF_EF_START 0x01 /* start the counter when adding */ |
212 | #define PERF_EF_RELOAD 0x02 /* reload the counter when starting */ | |
213 | #define PERF_EF_UPDATE 0x04 /* update the counter when stopping */ | |
214 | ||
8d2cacbb | 215 | /* |
a4eaf7f1 PZ |
216 | * Adds/Removes a counter to/from the PMU, can be done inside |
217 | * a transaction, see the ->*_txn() methods. | |
218 | */ | |
219 | int (*add) (struct perf_event *event, int flags); | |
220 | void (*del) (struct perf_event *event, int flags); | |
221 | ||
222 | /* | |
223 | * Starts/Stops a counter present on the PMU. The PMI handler | |
224 | * should stop the counter when perf_event_overflow() returns | |
225 | * !0. ->start() will be used to continue. | |
226 | */ | |
227 | void (*start) (struct perf_event *event, int flags); | |
228 | void (*stop) (struct perf_event *event, int flags); | |
229 | ||
230 | /* | |
231 | * Updates the counter value of the event. | |
232 | */ | |
cdd6c482 | 233 | void (*read) (struct perf_event *event); |
6bde9b6c LM |
234 | |
235 | /* | |
24cd7f54 PZ |
236 | * Group events scheduling is treated as a transaction, add |
237 | * group events as a whole and perform one schedulability test. | |
238 | * If the test fails, roll back the whole group | |
a4eaf7f1 PZ |
239 | * |
240 | * Start the transaction, after this ->add() doesn't need to | |
24cd7f54 | 241 | * do schedulability tests. |
8d2cacbb | 242 | */ |
e7e7ee2e | 243 | void (*start_txn) (struct pmu *pmu); /* optional */ |
8d2cacbb | 244 | /* |
a4eaf7f1 | 245 | * If ->start_txn() disabled the ->add() schedulability test |
8d2cacbb PZ |
246 | * then ->commit_txn() is required to perform one. On success |
247 | * the transaction is closed. On error the transaction is kept | |
248 | * open until ->cancel_txn() is called. | |
249 | */ | |
e7e7ee2e | 250 | int (*commit_txn) (struct pmu *pmu); /* optional */ |
8d2cacbb | 251 | /* |
a4eaf7f1 | 252 | * Will cancel the transaction, assumes ->del() is called |
25985edc | 253 | * for each successful ->add() during the transaction. |
8d2cacbb | 254 | */ |
e7e7ee2e | 255 | void (*cancel_txn) (struct pmu *pmu); /* optional */ |
35edc2a5 PZ |
256 | |
257 | /* | |
258 | * Will return the value for perf_event_mmap_page::index for this event, | |
259 | * if no implementation is provided it will default to: event->hw.idx + 1. | |
260 | */ | |
261 | int (*event_idx) (struct perf_event *event); /*optional */ | |
d010b332 SE |
262 | |
263 | /* | |
264 | * flush branch stack on context-switches (needed in cpu-wide mode) | |
265 | */ | |
266 | void (*flush_branch_stack) (void); | |
621a01ea IM |
267 | }; |
268 | ||
6a930700 | 269 | /** |
cdd6c482 | 270 | * enum perf_event_active_state - the states of a event |
6a930700 | 271 | */ |
cdd6c482 | 272 | enum perf_event_active_state { |
57c0c15b | 273 | PERF_EVENT_STATE_ERROR = -2, |
cdd6c482 IM |
274 | PERF_EVENT_STATE_OFF = -1, |
275 | PERF_EVENT_STATE_INACTIVE = 0, | |
57c0c15b | 276 | PERF_EVENT_STATE_ACTIVE = 1, |
6a930700 IM |
277 | }; |
278 | ||
9b51f66d | 279 | struct file; |
453f19ee PZ |
280 | struct perf_sample_data; |
281 | ||
a8b0ca17 | 282 | typedef void (*perf_overflow_handler_t)(struct perf_event *, |
b326e956 FW |
283 | struct perf_sample_data *, |
284 | struct pt_regs *regs); | |
285 | ||
d6f962b5 | 286 | enum perf_group_flag { |
e7e7ee2e | 287 | PERF_GROUP_SOFTWARE = 0x1, |
d6f962b5 FW |
288 | }; |
289 | ||
e7e7ee2e IM |
290 | #define SWEVENT_HLIST_BITS 8 |
291 | #define SWEVENT_HLIST_SIZE (1 << SWEVENT_HLIST_BITS) | |
76e1d904 FW |
292 | |
293 | struct swevent_hlist { | |
e7e7ee2e IM |
294 | struct hlist_head heads[SWEVENT_HLIST_SIZE]; |
295 | struct rcu_head rcu_head; | |
76e1d904 FW |
296 | }; |
297 | ||
8a49542c PZ |
298 | #define PERF_ATTACH_CONTEXT 0x01 |
299 | #define PERF_ATTACH_GROUP 0x02 | |
d580ff86 | 300 | #define PERF_ATTACH_TASK 0x04 |
8a49542c | 301 | |
e5d1367f SE |
302 | #ifdef CONFIG_CGROUP_PERF |
303 | /* | |
304 | * perf_cgroup_info keeps track of time_enabled for a cgroup. | |
305 | * This is a per-cpu dynamically allocated data structure. | |
306 | */ | |
307 | struct perf_cgroup_info { | |
e7e7ee2e IM |
308 | u64 time; |
309 | u64 timestamp; | |
e5d1367f SE |
310 | }; |
311 | ||
312 | struct perf_cgroup { | |
e7e7ee2e IM |
313 | struct cgroup_subsys_state css; |
314 | struct perf_cgroup_info *info; /* timing info, one per cpu */ | |
e5d1367f SE |
315 | }; |
316 | #endif | |
317 | ||
76369139 FW |
318 | struct ring_buffer; |
319 | ||
0793a61d | 320 | /** |
cdd6c482 | 321 | * struct perf_event - performance event kernel representation: |
0793a61d | 322 | */ |
cdd6c482 IM |
323 | struct perf_event { |
324 | #ifdef CONFIG_PERF_EVENTS | |
65abc865 | 325 | struct list_head group_entry; |
592903cd | 326 | struct list_head event_entry; |
04289bb9 | 327 | struct list_head sibling_list; |
76e1d904 | 328 | struct hlist_node hlist_entry; |
0127c3ea | 329 | int nr_siblings; |
d6f962b5 | 330 | int group_flags; |
cdd6c482 | 331 | struct perf_event *group_leader; |
a4eaf7f1 | 332 | struct pmu *pmu; |
04289bb9 | 333 | |
cdd6c482 | 334 | enum perf_event_active_state state; |
8a49542c | 335 | unsigned int attach_state; |
e7850595 | 336 | local64_t count; |
a6e6dea6 | 337 | atomic64_t child_count; |
ee06094f | 338 | |
53cfbf59 | 339 | /* |
cdd6c482 | 340 | * These are the total time in nanoseconds that the event |
53cfbf59 | 341 | * has been enabled (i.e. eligible to run, and the task has |
cdd6c482 | 342 | * been scheduled in, if this is a per-task event) |
53cfbf59 PM |
343 | * and running (scheduled onto the CPU), respectively. |
344 | * | |
345 | * They are computed from tstamp_enabled, tstamp_running and | |
cdd6c482 | 346 | * tstamp_stopped when the event is in INACTIVE or ACTIVE state. |
53cfbf59 PM |
347 | */ |
348 | u64 total_time_enabled; | |
349 | u64 total_time_running; | |
350 | ||
351 | /* | |
352 | * These are timestamps used for computing total_time_enabled | |
cdd6c482 | 353 | * and total_time_running when the event is in INACTIVE or |
53cfbf59 PM |
354 | * ACTIVE state, measured in nanoseconds from an arbitrary point |
355 | * in time. | |
cdd6c482 IM |
356 | * tstamp_enabled: the notional time when the event was enabled |
357 | * tstamp_running: the notional time when the event was scheduled on | |
53cfbf59 | 358 | * tstamp_stopped: in INACTIVE state, the notional time when the |
cdd6c482 | 359 | * event was scheduled off. |
53cfbf59 PM |
360 | */ |
361 | u64 tstamp_enabled; | |
362 | u64 tstamp_running; | |
363 | u64 tstamp_stopped; | |
364 | ||
eed01528 SE |
365 | /* |
366 | * timestamp shadows the actual context timing but it can | |
367 | * be safely used in NMI interrupt context. It reflects the | |
368 | * context time as it was when the event was last scheduled in. | |
369 | * | |
370 | * ctx_time already accounts for ctx->timestamp. Therefore to | |
371 | * compute ctx_time for a sample, simply add perf_clock(). | |
372 | */ | |
373 | u64 shadow_ctx_time; | |
374 | ||
24f1e32c | 375 | struct perf_event_attr attr; |
c320c7b7 | 376 | u16 header_size; |
6844c09d | 377 | u16 id_header_size; |
c320c7b7 | 378 | u16 read_size; |
cdd6c482 | 379 | struct hw_perf_event hw; |
0793a61d | 380 | |
cdd6c482 | 381 | struct perf_event_context *ctx; |
a6fa941d | 382 | atomic_long_t refcount; |
0793a61d | 383 | |
53cfbf59 PM |
384 | /* |
385 | * These accumulate total time (in nanoseconds) that children | |
cdd6c482 | 386 | * events have been enabled and running, respectively. |
53cfbf59 PM |
387 | */ |
388 | atomic64_t child_total_time_enabled; | |
389 | atomic64_t child_total_time_running; | |
390 | ||
0793a61d | 391 | /* |
d859e29f | 392 | * Protect attach/detach and child_list: |
0793a61d | 393 | */ |
fccc714b PZ |
394 | struct mutex child_mutex; |
395 | struct list_head child_list; | |
cdd6c482 | 396 | struct perf_event *parent; |
0793a61d TG |
397 | |
398 | int oncpu; | |
399 | int cpu; | |
400 | ||
082ff5a2 PZ |
401 | struct list_head owner_entry; |
402 | struct task_struct *owner; | |
403 | ||
7b732a75 PZ |
404 | /* mmap bits */ |
405 | struct mutex mmap_mutex; | |
406 | atomic_t mmap_count; | |
ac9721f3 PZ |
407 | int mmap_locked; |
408 | struct user_struct *mmap_user; | |
76369139 | 409 | struct ring_buffer *rb; |
10c6db11 | 410 | struct list_head rb_entry; |
37d81828 | 411 | |
7b732a75 | 412 | /* poll related */ |
0793a61d | 413 | wait_queue_head_t waitq; |
3c446b3d | 414 | struct fasync_struct *fasync; |
79f14641 PZ |
415 | |
416 | /* delayed work for NMIs and such */ | |
417 | int pending_wakeup; | |
4c9e2542 | 418 | int pending_kill; |
79f14641 | 419 | int pending_disable; |
e360adbe | 420 | struct irq_work pending; |
592903cd | 421 | |
79f14641 PZ |
422 | atomic_t event_limit; |
423 | ||
cdd6c482 | 424 | void (*destroy)(struct perf_event *); |
592903cd | 425 | struct rcu_head rcu_head; |
709e50cf PZ |
426 | |
427 | struct pid_namespace *ns; | |
8e5799b1 | 428 | u64 id; |
6fb2915d | 429 | |
b326e956 | 430 | perf_overflow_handler_t overflow_handler; |
4dc0da86 | 431 | void *overflow_handler_context; |
453f19ee | 432 | |
07b139c8 | 433 | #ifdef CONFIG_EVENT_TRACING |
1c024eca | 434 | struct ftrace_event_call *tp_event; |
6fb2915d | 435 | struct event_filter *filter; |
ced39002 JO |
436 | #ifdef CONFIG_FUNCTION_TRACER |
437 | struct ftrace_ops ftrace_ops; | |
438 | #endif | |
ee06094f | 439 | #endif |
6fb2915d | 440 | |
e5d1367f SE |
441 | #ifdef CONFIG_CGROUP_PERF |
442 | struct perf_cgroup *cgrp; /* cgroup event is attach to */ | |
443 | int cgrp_defer_enabled; | |
444 | #endif | |
445 | ||
6fb2915d | 446 | #endif /* CONFIG_PERF_EVENTS */ |
0793a61d TG |
447 | }; |
448 | ||
b04243ef PZ |
449 | enum perf_event_context_type { |
450 | task_context, | |
451 | cpu_context, | |
452 | }; | |
453 | ||
0793a61d | 454 | /** |
cdd6c482 | 455 | * struct perf_event_context - event context structure |
0793a61d | 456 | * |
cdd6c482 | 457 | * Used as a container for task events and CPU events as well: |
0793a61d | 458 | */ |
cdd6c482 | 459 | struct perf_event_context { |
108b02cf | 460 | struct pmu *pmu; |
ee643c41 | 461 | enum perf_event_context_type type; |
0793a61d | 462 | /* |
cdd6c482 | 463 | * Protect the states of the events in the list, |
d859e29f | 464 | * nr_active, and the list: |
0793a61d | 465 | */ |
e625cce1 | 466 | raw_spinlock_t lock; |
d859e29f | 467 | /* |
cdd6c482 | 468 | * Protect the list of events. Locking either mutex or lock |
d859e29f PM |
469 | * is sufficient to ensure the list doesn't change; to change |
470 | * the list you need to lock both the mutex and the spinlock. | |
471 | */ | |
a308444c | 472 | struct mutex mutex; |
04289bb9 | 473 | |
889ff015 FW |
474 | struct list_head pinned_groups; |
475 | struct list_head flexible_groups; | |
a308444c | 476 | struct list_head event_list; |
cdd6c482 | 477 | int nr_events; |
a308444c IM |
478 | int nr_active; |
479 | int is_active; | |
bfbd3381 | 480 | int nr_stat; |
0f5a2601 | 481 | int nr_freq; |
dddd3379 | 482 | int rotate_disable; |
a308444c IM |
483 | atomic_t refcount; |
484 | struct task_struct *task; | |
53cfbf59 PM |
485 | |
486 | /* | |
4af4998b | 487 | * Context clock, runs when context enabled. |
53cfbf59 | 488 | */ |
a308444c IM |
489 | u64 time; |
490 | u64 timestamp; | |
564c2b21 PM |
491 | |
492 | /* | |
493 | * These fields let us detect when two contexts have both | |
494 | * been cloned (inherited) from a common ancestor. | |
495 | */ | |
cdd6c482 | 496 | struct perf_event_context *parent_ctx; |
a308444c IM |
497 | u64 parent_gen; |
498 | u64 generation; | |
499 | int pin_count; | |
d010b332 SE |
500 | int nr_cgroups; /* cgroup evts */ |
501 | int nr_branch_stack; /* branch_stack evt */ | |
28009ce4 | 502 | struct rcu_head rcu_head; |
0793a61d TG |
503 | }; |
504 | ||
7ae07ea3 FW |
505 | /* |
506 | * Number of contexts where an event can trigger: | |
e7e7ee2e | 507 | * task, softirq, hardirq, nmi. |
7ae07ea3 FW |
508 | */ |
509 | #define PERF_NR_CONTEXTS 4 | |
510 | ||
0793a61d | 511 | /** |
cdd6c482 | 512 | * struct perf_event_cpu_context - per cpu event context structure |
0793a61d TG |
513 | */ |
514 | struct perf_cpu_context { | |
cdd6c482 IM |
515 | struct perf_event_context ctx; |
516 | struct perf_event_context *task_ctx; | |
0793a61d | 517 | int active_oncpu; |
3b6f9e5c | 518 | int exclusive; |
e9d2b064 PZ |
519 | struct list_head rotation_list; |
520 | int jiffies_interval; | |
3f1f3320 | 521 | struct pmu *unique_pmu; |
e5d1367f | 522 | struct perf_cgroup *cgrp; |
0793a61d TG |
523 | }; |
524 | ||
5622f295 | 525 | struct perf_output_handle { |
57c0c15b | 526 | struct perf_event *event; |
76369139 | 527 | struct ring_buffer *rb; |
6d1acfd5 | 528 | unsigned long wakeup; |
5d967a8b PZ |
529 | unsigned long size; |
530 | void *addr; | |
531 | int page; | |
5622f295 MM |
532 | }; |
533 | ||
cdd6c482 | 534 | #ifdef CONFIG_PERF_EVENTS |
829b42dd | 535 | |
2e80a82a | 536 | extern int perf_pmu_register(struct pmu *pmu, char *name, int type); |
b0a873eb | 537 | extern void perf_pmu_unregister(struct pmu *pmu); |
621a01ea | 538 | |
3bf101ba | 539 | extern int perf_num_counters(void); |
84c79910 | 540 | extern const char *perf_pmu_name(void); |
ab0cce56 JO |
541 | extern void __perf_event_task_sched_in(struct task_struct *prev, |
542 | struct task_struct *task); | |
543 | extern void __perf_event_task_sched_out(struct task_struct *prev, | |
544 | struct task_struct *next); | |
cdd6c482 IM |
545 | extern int perf_event_init_task(struct task_struct *child); |
546 | extern void perf_event_exit_task(struct task_struct *child); | |
547 | extern void perf_event_free_task(struct task_struct *task); | |
4e231c79 | 548 | extern void perf_event_delayed_put(struct task_struct *task); |
cdd6c482 | 549 | extern void perf_event_print_debug(void); |
33696fc0 PZ |
550 | extern void perf_pmu_disable(struct pmu *pmu); |
551 | extern void perf_pmu_enable(struct pmu *pmu); | |
cdd6c482 IM |
552 | extern int perf_event_task_disable(void); |
553 | extern int perf_event_task_enable(void); | |
26ca5c11 | 554 | extern int perf_event_refresh(struct perf_event *event, int refresh); |
cdd6c482 | 555 | extern void perf_event_update_userpage(struct perf_event *event); |
fb0459d7 AV |
556 | extern int perf_event_release_kernel(struct perf_event *event); |
557 | extern struct perf_event * | |
558 | perf_event_create_kernel_counter(struct perf_event_attr *attr, | |
559 | int cpu, | |
38a81da2 | 560 | struct task_struct *task, |
4dc0da86 AK |
561 | perf_overflow_handler_t callback, |
562 | void *context); | |
0cda4c02 YZ |
563 | extern void perf_pmu_migrate_context(struct pmu *pmu, |
564 | int src_cpu, int dst_cpu); | |
59ed446f PZ |
565 | extern u64 perf_event_read_value(struct perf_event *event, |
566 | u64 *enabled, u64 *running); | |
5c92d124 | 567 | |
d010b332 | 568 | |
df1a132b | 569 | struct perf_sample_data { |
5622f295 MM |
570 | u64 type; |
571 | ||
572 | u64 ip; | |
573 | struct { | |
574 | u32 pid; | |
575 | u32 tid; | |
576 | } tid_entry; | |
577 | u64 time; | |
a308444c | 578 | u64 addr; |
5622f295 MM |
579 | u64 id; |
580 | u64 stream_id; | |
581 | struct { | |
582 | u32 cpu; | |
583 | u32 reserved; | |
584 | } cpu_entry; | |
a308444c | 585 | u64 period; |
5622f295 | 586 | struct perf_callchain_entry *callchain; |
3a43ce68 | 587 | struct perf_raw_record *raw; |
bce38cd5 | 588 | struct perf_branch_stack *br_stack; |
4018994f | 589 | struct perf_regs_user regs_user; |
c5ebcedb | 590 | u64 stack_user_size; |
df1a132b PZ |
591 | }; |
592 | ||
fd0d000b RR |
593 | static inline void perf_sample_data_init(struct perf_sample_data *data, |
594 | u64 addr, u64 period) | |
dc1d628a | 595 | { |
fd0d000b | 596 | /* remaining struct members initialized in perf_prepare_sample() */ |
dc1d628a PZ |
597 | data->addr = addr; |
598 | data->raw = NULL; | |
bce38cd5 | 599 | data->br_stack = NULL; |
4018994f JO |
600 | data->period = period; |
601 | data->regs_user.abi = PERF_SAMPLE_REGS_ABI_NONE; | |
602 | data->regs_user.regs = NULL; | |
c5ebcedb | 603 | data->stack_user_size = 0; |
dc1d628a PZ |
604 | } |
605 | ||
5622f295 MM |
606 | extern void perf_output_sample(struct perf_output_handle *handle, |
607 | struct perf_event_header *header, | |
608 | struct perf_sample_data *data, | |
cdd6c482 | 609 | struct perf_event *event); |
5622f295 MM |
610 | extern void perf_prepare_sample(struct perf_event_header *header, |
611 | struct perf_sample_data *data, | |
cdd6c482 | 612 | struct perf_event *event, |
5622f295 MM |
613 | struct pt_regs *regs); |
614 | ||
a8b0ca17 | 615 | extern int perf_event_overflow(struct perf_event *event, |
5622f295 MM |
616 | struct perf_sample_data *data, |
617 | struct pt_regs *regs); | |
df1a132b | 618 | |
6c7e550f FBH |
619 | static inline bool is_sampling_event(struct perf_event *event) |
620 | { | |
621 | return event->attr.sample_period != 0; | |
622 | } | |
623 | ||
3b6f9e5c | 624 | /* |
cdd6c482 | 625 | * Return 1 for a software event, 0 for a hardware event |
3b6f9e5c | 626 | */ |
cdd6c482 | 627 | static inline int is_software_event(struct perf_event *event) |
3b6f9e5c | 628 | { |
89a1e187 | 629 | return event->pmu->task_ctx_nr == perf_sw_context; |
3b6f9e5c PM |
630 | } |
631 | ||
c5905afb | 632 | extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX]; |
f29ac756 | 633 | |
a8b0ca17 | 634 | extern void __perf_sw_event(u32, u64, struct pt_regs *, u64); |
f29ac756 | 635 | |
b0f82b81 | 636 | #ifndef perf_arch_fetch_caller_regs |
e7e7ee2e | 637 | static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { } |
b0f82b81 | 638 | #endif |
5331d7b8 FW |
639 | |
640 | /* | |
641 | * Take a snapshot of the regs. Skip ip and frame pointer to | |
642 | * the nth caller. We only need a few of the regs: | |
643 | * - ip for PERF_SAMPLE_IP | |
644 | * - cs for user_mode() tests | |
645 | * - bp for callchains | |
646 | * - eflags, for future purposes, just in case | |
647 | */ | |
b0f82b81 | 648 | static inline void perf_fetch_caller_regs(struct pt_regs *regs) |
5331d7b8 | 649 | { |
5331d7b8 FW |
650 | memset(regs, 0, sizeof(*regs)); |
651 | ||
b0f82b81 | 652 | perf_arch_fetch_caller_regs(regs, CALLER_ADDR0); |
5331d7b8 FW |
653 | } |
654 | ||
7e54a5a0 | 655 | static __always_inline void |
a8b0ca17 | 656 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) |
e49a5bd3 | 657 | { |
7e54a5a0 PZ |
658 | struct pt_regs hot_regs; |
659 | ||
c5905afb | 660 | if (static_key_false(&perf_swevent_enabled[event_id])) { |
d430d3d7 JB |
661 | if (!regs) { |
662 | perf_fetch_caller_regs(&hot_regs); | |
663 | regs = &hot_regs; | |
664 | } | |
a8b0ca17 | 665 | __perf_sw_event(event_id, nr, regs, addr); |
e49a5bd3 FW |
666 | } |
667 | } | |
668 | ||
c5905afb | 669 | extern struct static_key_deferred perf_sched_events; |
ee6dcfa4 | 670 | |
ab0cce56 | 671 | static inline void perf_event_task_sched_in(struct task_struct *prev, |
a8d757ef | 672 | struct task_struct *task) |
ab0cce56 JO |
673 | { |
674 | if (static_key_false(&perf_sched_events.key)) | |
675 | __perf_event_task_sched_in(prev, task); | |
676 | } | |
677 | ||
678 | static inline void perf_event_task_sched_out(struct task_struct *prev, | |
679 | struct task_struct *next) | |
ee6dcfa4 | 680 | { |
a8b0ca17 | 681 | perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, NULL, 0); |
ee6dcfa4 | 682 | |
c5905afb | 683 | if (static_key_false(&perf_sched_events.key)) |
ab0cce56 | 684 | __perf_event_task_sched_out(prev, next); |
ee6dcfa4 PZ |
685 | } |
686 | ||
3af9e859 | 687 | extern void perf_event_mmap(struct vm_area_struct *vma); |
39447b38 | 688 | extern struct perf_guest_info_callbacks *perf_guest_cbs; |
dcf46b94 ZY |
689 | extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); |
690 | extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); | |
39447b38 | 691 | |
cdd6c482 IM |
692 | extern void perf_event_comm(struct task_struct *tsk); |
693 | extern void perf_event_fork(struct task_struct *tsk); | |
8d1b2d93 | 694 | |
56962b44 FW |
695 | /* Callchains */ |
696 | DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry); | |
697 | ||
e7e7ee2e IM |
698 | extern void perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs); |
699 | extern void perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs); | |
394ee076 | 700 | |
e7e7ee2e | 701 | static inline void perf_callchain_store(struct perf_callchain_entry *entry, u64 ip) |
70791ce9 FW |
702 | { |
703 | if (entry->nr < PERF_MAX_STACK_DEPTH) | |
704 | entry->ip[entry->nr++] = ip; | |
705 | } | |
394ee076 | 706 | |
cdd6c482 IM |
707 | extern int sysctl_perf_event_paranoid; |
708 | extern int sysctl_perf_event_mlock; | |
709 | extern int sysctl_perf_event_sample_rate; | |
1ccd1549 | 710 | |
163ec435 PZ |
711 | extern int perf_proc_update_handler(struct ctl_table *table, int write, |
712 | void __user *buffer, size_t *lenp, | |
713 | loff_t *ppos); | |
714 | ||
320ebf09 PZ |
715 | static inline bool perf_paranoid_tracepoint_raw(void) |
716 | { | |
717 | return sysctl_perf_event_paranoid > -1; | |
718 | } | |
719 | ||
720 | static inline bool perf_paranoid_cpu(void) | |
721 | { | |
722 | return sysctl_perf_event_paranoid > 0; | |
723 | } | |
724 | ||
725 | static inline bool perf_paranoid_kernel(void) | |
726 | { | |
727 | return sysctl_perf_event_paranoid > 1; | |
728 | } | |
729 | ||
cdd6c482 | 730 | extern void perf_event_init(void); |
1c024eca PZ |
731 | extern void perf_tp_event(u64 addr, u64 count, void *record, |
732 | int entry_size, struct pt_regs *regs, | |
e6dab5ff AV |
733 | struct hlist_head *head, int rctx, |
734 | struct task_struct *task); | |
24f1e32c | 735 | extern void perf_bp_event(struct perf_event *event, void *data); |
0d905bca | 736 | |
9d23a90a | 737 | #ifndef perf_misc_flags |
e7e7ee2e IM |
738 | # define perf_misc_flags(regs) \ |
739 | (user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL) | |
740 | # define perf_instruction_pointer(regs) instruction_pointer(regs) | |
9d23a90a PM |
741 | #endif |
742 | ||
bce38cd5 SE |
743 | static inline bool has_branch_stack(struct perf_event *event) |
744 | { | |
745 | return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK; | |
746 | } | |
747 | ||
5622f295 | 748 | extern int perf_output_begin(struct perf_output_handle *handle, |
a7ac67ea | 749 | struct perf_event *event, unsigned int size); |
5622f295 | 750 | extern void perf_output_end(struct perf_output_handle *handle); |
91d7753a | 751 | extern unsigned int perf_output_copy(struct perf_output_handle *handle, |
5622f295 | 752 | const void *buf, unsigned int len); |
5685e0ff JO |
753 | extern unsigned int perf_output_skip(struct perf_output_handle *handle, |
754 | unsigned int len); | |
4ed7c92d PZ |
755 | extern int perf_swevent_get_recursion_context(void); |
756 | extern void perf_swevent_put_recursion_context(int rctx); | |
44234adc FW |
757 | extern void perf_event_enable(struct perf_event *event); |
758 | extern void perf_event_disable(struct perf_event *event); | |
500ad2d8 | 759 | extern int __perf_event_disable(void *info); |
e9d2b064 | 760 | extern void perf_event_task_tick(void); |
1d9d8639 | 761 | extern void perf_restore_debug_store(void); |
0793a61d TG |
762 | #else |
763 | static inline void | |
ab0cce56 JO |
764 | perf_event_task_sched_in(struct task_struct *prev, |
765 | struct task_struct *task) { } | |
766 | static inline void | |
767 | perf_event_task_sched_out(struct task_struct *prev, | |
768 | struct task_struct *next) { } | |
cdd6c482 IM |
769 | static inline int perf_event_init_task(struct task_struct *child) { return 0; } |
770 | static inline void perf_event_exit_task(struct task_struct *child) { } | |
771 | static inline void perf_event_free_task(struct task_struct *task) { } | |
4e231c79 | 772 | static inline void perf_event_delayed_put(struct task_struct *task) { } |
57c0c15b | 773 | static inline void perf_event_print_debug(void) { } |
57c0c15b IM |
774 | static inline int perf_event_task_disable(void) { return -EINVAL; } |
775 | static inline int perf_event_task_enable(void) { return -EINVAL; } | |
26ca5c11 AK |
776 | static inline int perf_event_refresh(struct perf_event *event, int refresh) |
777 | { | |
778 | return -EINVAL; | |
779 | } | |
15dbf27c | 780 | |
925d519a | 781 | static inline void |
a8b0ca17 | 782 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) { } |
24f1e32c | 783 | static inline void |
184f412c | 784 | perf_bp_event(struct perf_event *event, void *data) { } |
0a4a9391 | 785 | |
39447b38 | 786 | static inline int perf_register_guest_info_callbacks |
e7e7ee2e | 787 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 788 | static inline int perf_unregister_guest_info_callbacks |
e7e7ee2e | 789 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 790 | |
57c0c15b | 791 | static inline void perf_event_mmap(struct vm_area_struct *vma) { } |
cdd6c482 IM |
792 | static inline void perf_event_comm(struct task_struct *tsk) { } |
793 | static inline void perf_event_fork(struct task_struct *tsk) { } | |
794 | static inline void perf_event_init(void) { } | |
184f412c | 795 | static inline int perf_swevent_get_recursion_context(void) { return -1; } |
4ed7c92d | 796 | static inline void perf_swevent_put_recursion_context(int rctx) { } |
44234adc FW |
797 | static inline void perf_event_enable(struct perf_event *event) { } |
798 | static inline void perf_event_disable(struct perf_event *event) { } | |
500ad2d8 | 799 | static inline int __perf_event_disable(void *info) { return -1; } |
e9d2b064 | 800 | static inline void perf_event_task_tick(void) { } |
1d9d8639 | 801 | static inline void perf_restore_debug_store(void) { } |
0793a61d TG |
802 | #endif |
803 | ||
e7e7ee2e | 804 | #define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x)) |
5622f295 | 805 | |
3f6da390 PZ |
806 | /* |
807 | * This has to have a higher priority than migration_notifier in sched.c. | |
808 | */ | |
e7e7ee2e IM |
809 | #define perf_cpu_notifier(fn) \ |
810 | do { \ | |
811 | static struct notifier_block fn##_nb __cpuinitdata = \ | |
812 | { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ | |
c13d38e4 | 813 | unsigned long cpu = smp_processor_id(); \ |
6760bca9 | 814 | unsigned long flags; \ |
e7e7ee2e | 815 | fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE, \ |
c13d38e4 | 816 | (void *)(unsigned long)cpu); \ |
6760bca9 | 817 | local_irq_save(flags); \ |
e7e7ee2e | 818 | fn(&fn##_nb, (unsigned long)CPU_STARTING, \ |
c13d38e4 | 819 | (void *)(unsigned long)cpu); \ |
6760bca9 | 820 | local_irq_restore(flags); \ |
e7e7ee2e | 821 | fn(&fn##_nb, (unsigned long)CPU_ONLINE, \ |
c13d38e4 | 822 | (void *)(unsigned long)cpu); \ |
e7e7ee2e | 823 | register_cpu_notifier(&fn##_nb); \ |
3f6da390 PZ |
824 | } while (0) |
825 | ||
641cc938 | 826 | |
2663960c SB |
827 | struct perf_pmu_events_attr { |
828 | struct device_attribute attr; | |
829 | u64 id; | |
830 | }; | |
831 | ||
832 | #define PMU_EVENT_ATTR(_name, _var, _id, _show) \ | |
833 | static struct perf_pmu_events_attr _var = { \ | |
834 | .attr = __ATTR(_name, 0444, _show, NULL), \ | |
835 | .id = _id, \ | |
836 | }; | |
837 | ||
641cc938 JO |
838 | #define PMU_FORMAT_ATTR(_name, _format) \ |
839 | static ssize_t \ | |
840 | _name##_show(struct device *dev, \ | |
841 | struct device_attribute *attr, \ | |
842 | char *page) \ | |
843 | { \ | |
844 | BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE); \ | |
845 | return sprintf(page, _format "\n"); \ | |
846 | } \ | |
847 | \ | |
848 | static struct device_attribute format_attr_##_name = __ATTR_RO(_name) | |
849 | ||
cdd6c482 | 850 | #endif /* _LINUX_PERF_EVENT_H */ |