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> |
fadfe7be | 55 | #include <linux/workqueue.h> |
fa588151 | 56 | #include <asm/local.h> |
f3dfd265 | 57 | |
f9188e02 PZ |
58 | struct perf_callchain_entry { |
59 | __u64 nr; | |
60 | __u64 ip[PERF_MAX_STACK_DEPTH]; | |
61 | }; | |
62 | ||
3a43ce68 FW |
63 | struct perf_raw_record { |
64 | u32 size; | |
65 | void *data; | |
f413cdb8 FW |
66 | }; |
67 | ||
bce38cd5 SE |
68 | /* |
69 | * branch stack layout: | |
70 | * nr: number of taken branches stored in entries[] | |
71 | * | |
72 | * Note that nr can vary from sample to sample | |
73 | * branches (to, from) are stored from most recent | |
74 | * to least recent, i.e., entries[0] contains the most | |
75 | * recent branch. | |
76 | */ | |
caff2bef PZ |
77 | struct perf_branch_stack { |
78 | __u64 nr; | |
79 | struct perf_branch_entry entries[0]; | |
80 | }; | |
81 | ||
f3dfd265 PM |
82 | struct task_struct; |
83 | ||
efc9f05d SE |
84 | /* |
85 | * extra PMU register associated with an event | |
86 | */ | |
87 | struct hw_perf_event_extra { | |
88 | u64 config; /* register value */ | |
89 | unsigned int reg; /* register address or index */ | |
90 | int alloc; /* extra register already allocated */ | |
91 | int idx; /* index in shared_regs->regs[] */ | |
92 | }; | |
93 | ||
43b45780 AH |
94 | struct event_constraint; |
95 | ||
0793a61d | 96 | /** |
cdd6c482 | 97 | * struct hw_perf_event - performance event hardware details: |
0793a61d | 98 | */ |
cdd6c482 IM |
99 | struct hw_perf_event { |
100 | #ifdef CONFIG_PERF_EVENTS | |
d6d020e9 PZ |
101 | union { |
102 | struct { /* hardware */ | |
a308444c | 103 | u64 config; |
447a194b | 104 | u64 last_tag; |
a308444c | 105 | unsigned long config_base; |
cdd6c482 | 106 | unsigned long event_base; |
c48b6053 | 107 | int event_base_rdpmc; |
a308444c | 108 | int idx; |
447a194b | 109 | int last_cpu; |
9fac2cf3 | 110 | int flags; |
bce38cd5 | 111 | |
efc9f05d | 112 | struct hw_perf_event_extra extra_reg; |
bce38cd5 | 113 | struct hw_perf_event_extra branch_reg; |
43b45780 AH |
114 | |
115 | struct event_constraint *constraint; | |
d6d020e9 | 116 | }; |
721a669b | 117 | struct { /* software */ |
a308444c | 118 | struct hrtimer hrtimer; |
d6d020e9 | 119 | }; |
f22c1bb6 ON |
120 | struct { /* tracepoint */ |
121 | struct task_struct *tp_target; | |
122 | /* for tp_event->class */ | |
123 | struct list_head tp_list; | |
124 | }; | |
24f1e32c | 125 | #ifdef CONFIG_HAVE_HW_BREAKPOINT |
45a73372 | 126 | struct { /* breakpoint */ |
d580ff86 PZ |
127 | /* |
128 | * Crufty hack to avoid the chicken and egg | |
129 | * problem hw_breakpoint has with context | |
130 | * creation and event initalization. | |
131 | */ | |
132 | struct task_struct *bp_target; | |
f22c1bb6 ON |
133 | struct arch_hw_breakpoint info; |
134 | struct list_head bp_list; | |
45a73372 | 135 | }; |
24f1e32c | 136 | #endif |
d6d020e9 | 137 | }; |
a4eaf7f1 | 138 | int state; |
e7850595 | 139 | local64_t prev_count; |
b23f3325 | 140 | u64 sample_period; |
9e350de3 | 141 | u64 last_period; |
e7850595 | 142 | local64_t period_left; |
e050e3f0 | 143 | u64 interrupts_seq; |
60db5e09 | 144 | u64 interrupts; |
6a24ed6c | 145 | |
abd50713 PZ |
146 | u64 freq_time_stamp; |
147 | u64 freq_count_stamp; | |
ee06094f | 148 | #endif |
0793a61d TG |
149 | }; |
150 | ||
a4eaf7f1 PZ |
151 | /* |
152 | * hw_perf_event::state flags | |
153 | */ | |
154 | #define PERF_HES_STOPPED 0x01 /* the counter is stopped */ | |
155 | #define PERF_HES_UPTODATE 0x02 /* event->count up-to-date */ | |
156 | #define PERF_HES_ARCH 0x04 | |
157 | ||
cdd6c482 | 158 | struct perf_event; |
621a01ea | 159 | |
8d2cacbb PZ |
160 | /* |
161 | * Common implementation detail of pmu::{start,commit,cancel}_txn | |
162 | */ | |
163 | #define PERF_EVENT_TXN 0x1 | |
6bde9b6c | 164 | |
53b25335 VW |
165 | /** |
166 | * pmu::capabilities flags | |
167 | */ | |
168 | #define PERF_PMU_CAP_NO_INTERRUPT 0x01 | |
169 | ||
621a01ea | 170 | /** |
4aeb0b42 | 171 | * struct pmu - generic performance monitoring unit |
621a01ea | 172 | */ |
4aeb0b42 | 173 | struct pmu { |
b0a873eb PZ |
174 | struct list_head entry; |
175 | ||
c464c76e | 176 | struct module *module; |
abe43400 | 177 | struct device *dev; |
0c9d42ed | 178 | const struct attribute_group **attr_groups; |
03d8e80b | 179 | const char *name; |
2e80a82a PZ |
180 | int type; |
181 | ||
53b25335 VW |
182 | /* |
183 | * various common per-pmu feature flags | |
184 | */ | |
185 | int capabilities; | |
186 | ||
108b02cf PZ |
187 | int * __percpu pmu_disable_count; |
188 | struct perf_cpu_context * __percpu pmu_cpu_context; | |
8dc85d54 | 189 | int task_ctx_nr; |
62b85639 | 190 | int hrtimer_interval_ms; |
6bde9b6c LM |
191 | |
192 | /* | |
a4eaf7f1 PZ |
193 | * Fully disable/enable this PMU, can be used to protect from the PMI |
194 | * as well as for lazy/batch writing of the MSRs. | |
6bde9b6c | 195 | */ |
ad5133b7 PZ |
196 | void (*pmu_enable) (struct pmu *pmu); /* optional */ |
197 | void (*pmu_disable) (struct pmu *pmu); /* optional */ | |
6bde9b6c | 198 | |
8d2cacbb | 199 | /* |
a4eaf7f1 | 200 | * Try and initialize the event for this PMU. |
24cd7f54 | 201 | * Should return -ENOENT when the @event doesn't match this PMU. |
8d2cacbb | 202 | */ |
b0a873eb PZ |
203 | int (*event_init) (struct perf_event *event); |
204 | ||
a4eaf7f1 PZ |
205 | #define PERF_EF_START 0x01 /* start the counter when adding */ |
206 | #define PERF_EF_RELOAD 0x02 /* reload the counter when starting */ | |
207 | #define PERF_EF_UPDATE 0x04 /* update the counter when stopping */ | |
208 | ||
8d2cacbb | 209 | /* |
a4eaf7f1 PZ |
210 | * Adds/Removes a counter to/from the PMU, can be done inside |
211 | * a transaction, see the ->*_txn() methods. | |
212 | */ | |
213 | int (*add) (struct perf_event *event, int flags); | |
214 | void (*del) (struct perf_event *event, int flags); | |
215 | ||
216 | /* | |
217 | * Starts/Stops a counter present on the PMU. The PMI handler | |
218 | * should stop the counter when perf_event_overflow() returns | |
219 | * !0. ->start() will be used to continue. | |
220 | */ | |
221 | void (*start) (struct perf_event *event, int flags); | |
222 | void (*stop) (struct perf_event *event, int flags); | |
223 | ||
224 | /* | |
225 | * Updates the counter value of the event. | |
226 | */ | |
cdd6c482 | 227 | void (*read) (struct perf_event *event); |
6bde9b6c LM |
228 | |
229 | /* | |
24cd7f54 PZ |
230 | * Group events scheduling is treated as a transaction, add |
231 | * group events as a whole and perform one schedulability test. | |
232 | * If the test fails, roll back the whole group | |
a4eaf7f1 PZ |
233 | * |
234 | * Start the transaction, after this ->add() doesn't need to | |
24cd7f54 | 235 | * do schedulability tests. |
8d2cacbb | 236 | */ |
e7e7ee2e | 237 | void (*start_txn) (struct pmu *pmu); /* optional */ |
8d2cacbb | 238 | /* |
a4eaf7f1 | 239 | * If ->start_txn() disabled the ->add() schedulability test |
8d2cacbb PZ |
240 | * then ->commit_txn() is required to perform one. On success |
241 | * the transaction is closed. On error the transaction is kept | |
242 | * open until ->cancel_txn() is called. | |
243 | */ | |
e7e7ee2e | 244 | int (*commit_txn) (struct pmu *pmu); /* optional */ |
8d2cacbb | 245 | /* |
a4eaf7f1 | 246 | * Will cancel the transaction, assumes ->del() is called |
25985edc | 247 | * for each successful ->add() during the transaction. |
8d2cacbb | 248 | */ |
e7e7ee2e | 249 | void (*cancel_txn) (struct pmu *pmu); /* optional */ |
35edc2a5 PZ |
250 | |
251 | /* | |
252 | * Will return the value for perf_event_mmap_page::index for this event, | |
253 | * if no implementation is provided it will default to: event->hw.idx + 1. | |
254 | */ | |
255 | int (*event_idx) (struct perf_event *event); /*optional */ | |
d010b332 SE |
256 | |
257 | /* | |
258 | * flush branch stack on context-switches (needed in cpu-wide mode) | |
259 | */ | |
260 | void (*flush_branch_stack) (void); | |
621a01ea IM |
261 | }; |
262 | ||
6a930700 | 263 | /** |
cdd6c482 | 264 | * enum perf_event_active_state - the states of a event |
6a930700 | 265 | */ |
cdd6c482 | 266 | enum perf_event_active_state { |
179033b3 | 267 | PERF_EVENT_STATE_EXIT = -3, |
57c0c15b | 268 | PERF_EVENT_STATE_ERROR = -2, |
cdd6c482 IM |
269 | PERF_EVENT_STATE_OFF = -1, |
270 | PERF_EVENT_STATE_INACTIVE = 0, | |
57c0c15b | 271 | PERF_EVENT_STATE_ACTIVE = 1, |
6a930700 IM |
272 | }; |
273 | ||
9b51f66d | 274 | struct file; |
453f19ee PZ |
275 | struct perf_sample_data; |
276 | ||
a8b0ca17 | 277 | typedef void (*perf_overflow_handler_t)(struct perf_event *, |
b326e956 FW |
278 | struct perf_sample_data *, |
279 | struct pt_regs *regs); | |
280 | ||
d6f962b5 | 281 | enum perf_group_flag { |
e7e7ee2e | 282 | PERF_GROUP_SOFTWARE = 0x1, |
d6f962b5 FW |
283 | }; |
284 | ||
e7e7ee2e IM |
285 | #define SWEVENT_HLIST_BITS 8 |
286 | #define SWEVENT_HLIST_SIZE (1 << SWEVENT_HLIST_BITS) | |
76e1d904 FW |
287 | |
288 | struct swevent_hlist { | |
e7e7ee2e IM |
289 | struct hlist_head heads[SWEVENT_HLIST_SIZE]; |
290 | struct rcu_head rcu_head; | |
76e1d904 FW |
291 | }; |
292 | ||
8a49542c PZ |
293 | #define PERF_ATTACH_CONTEXT 0x01 |
294 | #define PERF_ATTACH_GROUP 0x02 | |
d580ff86 | 295 | #define PERF_ATTACH_TASK 0x04 |
8a49542c | 296 | |
877c6856 | 297 | struct perf_cgroup; |
76369139 FW |
298 | struct ring_buffer; |
299 | ||
0793a61d | 300 | /** |
cdd6c482 | 301 | * struct perf_event - performance event kernel representation: |
0793a61d | 302 | */ |
cdd6c482 IM |
303 | struct perf_event { |
304 | #ifdef CONFIG_PERF_EVENTS | |
9886167d PZ |
305 | /* |
306 | * entry onto perf_event_context::event_list; | |
307 | * modifications require ctx->lock | |
308 | * RCU safe iterations. | |
309 | */ | |
592903cd | 310 | struct list_head event_entry; |
9886167d PZ |
311 | |
312 | /* | |
313 | * XXX: group_entry and sibling_list should be mutually exclusive; | |
314 | * either you're a sibling on a group, or you're the group leader. | |
315 | * Rework the code to always use the same list element. | |
316 | * | |
317 | * Locked for modification by both ctx->mutex and ctx->lock; holding | |
318 | * either sufficies for read. | |
319 | */ | |
320 | struct list_head group_entry; | |
04289bb9 | 321 | struct list_head sibling_list; |
9886167d PZ |
322 | |
323 | /* | |
324 | * We need storage to track the entries in perf_pmu_migrate_context; we | |
325 | * cannot use the event_entry because of RCU and we want to keep the | |
326 | * group in tact which avoids us using the other two entries. | |
327 | */ | |
328 | struct list_head migrate_entry; | |
329 | ||
f3ae75de SE |
330 | struct hlist_node hlist_entry; |
331 | struct list_head active_entry; | |
0127c3ea | 332 | int nr_siblings; |
d6f962b5 | 333 | int group_flags; |
cdd6c482 | 334 | struct perf_event *group_leader; |
a4eaf7f1 | 335 | struct pmu *pmu; |
04289bb9 | 336 | |
cdd6c482 | 337 | enum perf_event_active_state state; |
8a49542c | 338 | unsigned int attach_state; |
e7850595 | 339 | local64_t count; |
a6e6dea6 | 340 | atomic64_t child_count; |
ee06094f | 341 | |
53cfbf59 | 342 | /* |
cdd6c482 | 343 | * These are the total time in nanoseconds that the event |
53cfbf59 | 344 | * has been enabled (i.e. eligible to run, and the task has |
cdd6c482 | 345 | * been scheduled in, if this is a per-task event) |
53cfbf59 PM |
346 | * and running (scheduled onto the CPU), respectively. |
347 | * | |
348 | * They are computed from tstamp_enabled, tstamp_running and | |
cdd6c482 | 349 | * tstamp_stopped when the event is in INACTIVE or ACTIVE state. |
53cfbf59 PM |
350 | */ |
351 | u64 total_time_enabled; | |
352 | u64 total_time_running; | |
353 | ||
354 | /* | |
355 | * These are timestamps used for computing total_time_enabled | |
cdd6c482 | 356 | * and total_time_running when the event is in INACTIVE or |
53cfbf59 PM |
357 | * ACTIVE state, measured in nanoseconds from an arbitrary point |
358 | * in time. | |
cdd6c482 IM |
359 | * tstamp_enabled: the notional time when the event was enabled |
360 | * tstamp_running: the notional time when the event was scheduled on | |
53cfbf59 | 361 | * tstamp_stopped: in INACTIVE state, the notional time when the |
cdd6c482 | 362 | * event was scheduled off. |
53cfbf59 PM |
363 | */ |
364 | u64 tstamp_enabled; | |
365 | u64 tstamp_running; | |
366 | u64 tstamp_stopped; | |
367 | ||
eed01528 SE |
368 | /* |
369 | * timestamp shadows the actual context timing but it can | |
370 | * be safely used in NMI interrupt context. It reflects the | |
371 | * context time as it was when the event was last scheduled in. | |
372 | * | |
373 | * ctx_time already accounts for ctx->timestamp. Therefore to | |
374 | * compute ctx_time for a sample, simply add perf_clock(). | |
375 | */ | |
376 | u64 shadow_ctx_time; | |
377 | ||
24f1e32c | 378 | struct perf_event_attr attr; |
c320c7b7 | 379 | u16 header_size; |
6844c09d | 380 | u16 id_header_size; |
c320c7b7 | 381 | u16 read_size; |
cdd6c482 | 382 | struct hw_perf_event hw; |
0793a61d | 383 | |
cdd6c482 | 384 | struct perf_event_context *ctx; |
a6fa941d | 385 | atomic_long_t refcount; |
0793a61d | 386 | |
53cfbf59 PM |
387 | /* |
388 | * These accumulate total time (in nanoseconds) that children | |
cdd6c482 | 389 | * events have been enabled and running, respectively. |
53cfbf59 PM |
390 | */ |
391 | atomic64_t child_total_time_enabled; | |
392 | atomic64_t child_total_time_running; | |
393 | ||
0793a61d | 394 | /* |
d859e29f | 395 | * Protect attach/detach and child_list: |
0793a61d | 396 | */ |
fccc714b PZ |
397 | struct mutex child_mutex; |
398 | struct list_head child_list; | |
cdd6c482 | 399 | struct perf_event *parent; |
0793a61d TG |
400 | |
401 | int oncpu; | |
402 | int cpu; | |
403 | ||
082ff5a2 PZ |
404 | struct list_head owner_entry; |
405 | struct task_struct *owner; | |
406 | ||
7b732a75 PZ |
407 | /* mmap bits */ |
408 | struct mutex mmap_mutex; | |
409 | atomic_t mmap_count; | |
26cb63ad | 410 | |
76369139 | 411 | struct ring_buffer *rb; |
10c6db11 | 412 | struct list_head rb_entry; |
b69cf536 PZ |
413 | unsigned long rcu_batches; |
414 | int rcu_pending; | |
37d81828 | 415 | |
7b732a75 | 416 | /* poll related */ |
0793a61d | 417 | wait_queue_head_t waitq; |
3c446b3d | 418 | struct fasync_struct *fasync; |
79f14641 PZ |
419 | |
420 | /* delayed work for NMIs and such */ | |
421 | int pending_wakeup; | |
4c9e2542 | 422 | int pending_kill; |
79f14641 | 423 | int pending_disable; |
e360adbe | 424 | struct irq_work pending; |
592903cd | 425 | |
79f14641 PZ |
426 | atomic_t event_limit; |
427 | ||
cdd6c482 | 428 | void (*destroy)(struct perf_event *); |
592903cd | 429 | struct rcu_head rcu_head; |
709e50cf PZ |
430 | |
431 | struct pid_namespace *ns; | |
8e5799b1 | 432 | u64 id; |
6fb2915d | 433 | |
b326e956 | 434 | perf_overflow_handler_t overflow_handler; |
4dc0da86 | 435 | void *overflow_handler_context; |
453f19ee | 436 | |
07b139c8 | 437 | #ifdef CONFIG_EVENT_TRACING |
1c024eca | 438 | struct ftrace_event_call *tp_event; |
6fb2915d | 439 | struct event_filter *filter; |
ced39002 JO |
440 | #ifdef CONFIG_FUNCTION_TRACER |
441 | struct ftrace_ops ftrace_ops; | |
442 | #endif | |
ee06094f | 443 | #endif |
6fb2915d | 444 | |
e5d1367f SE |
445 | #ifdef CONFIG_CGROUP_PERF |
446 | struct perf_cgroup *cgrp; /* cgroup event is attach to */ | |
447 | int cgrp_defer_enabled; | |
448 | #endif | |
449 | ||
6fb2915d | 450 | #endif /* CONFIG_PERF_EVENTS */ |
0793a61d TG |
451 | }; |
452 | ||
453 | /** | |
cdd6c482 | 454 | * struct perf_event_context - event context structure |
0793a61d | 455 | * |
cdd6c482 | 456 | * Used as a container for task events and CPU events as well: |
0793a61d | 457 | */ |
cdd6c482 | 458 | struct perf_event_context { |
108b02cf | 459 | struct pmu *pmu; |
0793a61d | 460 | /* |
cdd6c482 | 461 | * Protect the states of the events in the list, |
d859e29f | 462 | * nr_active, and the list: |
0793a61d | 463 | */ |
e625cce1 | 464 | raw_spinlock_t lock; |
d859e29f | 465 | /* |
cdd6c482 | 466 | * Protect the list of events. Locking either mutex or lock |
d859e29f PM |
467 | * is sufficient to ensure the list doesn't change; to change |
468 | * the list you need to lock both the mutex and the spinlock. | |
469 | */ | |
a308444c | 470 | struct mutex mutex; |
04289bb9 | 471 | |
889ff015 FW |
472 | struct list_head pinned_groups; |
473 | struct list_head flexible_groups; | |
a308444c | 474 | struct list_head event_list; |
cdd6c482 | 475 | int nr_events; |
a308444c IM |
476 | int nr_active; |
477 | int is_active; | |
bfbd3381 | 478 | int nr_stat; |
0f5a2601 | 479 | int nr_freq; |
dddd3379 | 480 | int rotate_disable; |
a308444c IM |
481 | atomic_t refcount; |
482 | struct task_struct *task; | |
53cfbf59 PM |
483 | |
484 | /* | |
4af4998b | 485 | * Context clock, runs when context enabled. |
53cfbf59 | 486 | */ |
a308444c IM |
487 | u64 time; |
488 | u64 timestamp; | |
564c2b21 PM |
489 | |
490 | /* | |
491 | * These fields let us detect when two contexts have both | |
492 | * been cloned (inherited) from a common ancestor. | |
493 | */ | |
cdd6c482 | 494 | struct perf_event_context *parent_ctx; |
a308444c IM |
495 | u64 parent_gen; |
496 | u64 generation; | |
497 | int pin_count; | |
d010b332 SE |
498 | int nr_cgroups; /* cgroup evts */ |
499 | int nr_branch_stack; /* branch_stack evt */ | |
28009ce4 | 500 | struct rcu_head rcu_head; |
fadfe7be JO |
501 | |
502 | struct delayed_work orphans_remove; | |
503 | bool orphans_remove_sched; | |
0793a61d TG |
504 | }; |
505 | ||
7ae07ea3 FW |
506 | /* |
507 | * Number of contexts where an event can trigger: | |
e7e7ee2e | 508 | * task, softirq, hardirq, nmi. |
7ae07ea3 FW |
509 | */ |
510 | #define PERF_NR_CONTEXTS 4 | |
511 | ||
0793a61d | 512 | /** |
cdd6c482 | 513 | * struct perf_event_cpu_context - per cpu event context structure |
0793a61d TG |
514 | */ |
515 | struct perf_cpu_context { | |
cdd6c482 IM |
516 | struct perf_event_context ctx; |
517 | struct perf_event_context *task_ctx; | |
0793a61d | 518 | int active_oncpu; |
3b6f9e5c | 519 | int exclusive; |
9e630205 SE |
520 | struct hrtimer hrtimer; |
521 | ktime_t hrtimer_interval; | |
e9d2b064 | 522 | struct list_head rotation_list; |
3f1f3320 | 523 | struct pmu *unique_pmu; |
e5d1367f | 524 | struct perf_cgroup *cgrp; |
0793a61d TG |
525 | }; |
526 | ||
5622f295 | 527 | struct perf_output_handle { |
57c0c15b | 528 | struct perf_event *event; |
76369139 | 529 | struct ring_buffer *rb; |
6d1acfd5 | 530 | unsigned long wakeup; |
5d967a8b PZ |
531 | unsigned long size; |
532 | void *addr; | |
533 | int page; | |
5622f295 MM |
534 | }; |
535 | ||
cdd6c482 | 536 | #ifdef CONFIG_PERF_EVENTS |
829b42dd | 537 | |
03d8e80b | 538 | extern int perf_pmu_register(struct pmu *pmu, const char *name, int type); |
b0a873eb | 539 | extern void perf_pmu_unregister(struct pmu *pmu); |
621a01ea | 540 | |
3bf101ba | 541 | extern int perf_num_counters(void); |
84c79910 | 542 | extern const char *perf_pmu_name(void); |
ab0cce56 JO |
543 | extern void __perf_event_task_sched_in(struct task_struct *prev, |
544 | struct task_struct *task); | |
545 | extern void __perf_event_task_sched_out(struct task_struct *prev, | |
546 | struct task_struct *next); | |
cdd6c482 IM |
547 | extern int perf_event_init_task(struct task_struct *child); |
548 | extern void perf_event_exit_task(struct task_struct *child); | |
549 | extern void perf_event_free_task(struct task_struct *task); | |
4e231c79 | 550 | extern void perf_event_delayed_put(struct task_struct *task); |
cdd6c482 | 551 | extern void perf_event_print_debug(void); |
33696fc0 PZ |
552 | extern void perf_pmu_disable(struct pmu *pmu); |
553 | extern void perf_pmu_enable(struct pmu *pmu); | |
cdd6c482 IM |
554 | extern int perf_event_task_disable(void); |
555 | extern int perf_event_task_enable(void); | |
26ca5c11 | 556 | extern int perf_event_refresh(struct perf_event *event, int refresh); |
cdd6c482 | 557 | extern void perf_event_update_userpage(struct perf_event *event); |
fb0459d7 AV |
558 | extern int perf_event_release_kernel(struct perf_event *event); |
559 | extern struct perf_event * | |
560 | perf_event_create_kernel_counter(struct perf_event_attr *attr, | |
561 | int cpu, | |
38a81da2 | 562 | struct task_struct *task, |
4dc0da86 AK |
563 | perf_overflow_handler_t callback, |
564 | void *context); | |
0cda4c02 YZ |
565 | extern void perf_pmu_migrate_context(struct pmu *pmu, |
566 | int src_cpu, int dst_cpu); | |
59ed446f PZ |
567 | extern u64 perf_event_read_value(struct perf_event *event, |
568 | u64 *enabled, u64 *running); | |
5c92d124 | 569 | |
d010b332 | 570 | |
df1a132b | 571 | struct perf_sample_data { |
2565711f PZ |
572 | /* |
573 | * Fields set by perf_sample_data_init(), group so as to | |
574 | * minimize the cachelines touched. | |
575 | */ | |
576 | u64 addr; | |
577 | struct perf_raw_record *raw; | |
578 | struct perf_branch_stack *br_stack; | |
579 | u64 period; | |
580 | u64 weight; | |
581 | u64 txn; | |
582 | union perf_mem_data_src data_src; | |
5622f295 | 583 | |
2565711f PZ |
584 | /* |
585 | * The other fields, optionally {set,used} by | |
586 | * perf_{prepare,output}_sample(). | |
587 | */ | |
588 | u64 type; | |
5622f295 MM |
589 | u64 ip; |
590 | struct { | |
591 | u32 pid; | |
592 | u32 tid; | |
593 | } tid_entry; | |
594 | u64 time; | |
5622f295 MM |
595 | u64 id; |
596 | u64 stream_id; | |
597 | struct { | |
598 | u32 cpu; | |
599 | u32 reserved; | |
600 | } cpu_entry; | |
5622f295 | 601 | struct perf_callchain_entry *callchain; |
88a7c26a AL |
602 | |
603 | /* | |
604 | * regs_user may point to task_pt_regs or to regs_user_copy, depending | |
605 | * on arch details. | |
606 | */ | |
60e2364e | 607 | struct perf_regs regs_user; |
88a7c26a AL |
608 | struct pt_regs regs_user_copy; |
609 | ||
60e2364e | 610 | struct perf_regs regs_intr; |
c5ebcedb | 611 | u64 stack_user_size; |
2565711f | 612 | } ____cacheline_aligned; |
df1a132b | 613 | |
770eee1f SE |
614 | /* default value for data source */ |
615 | #define PERF_MEM_NA (PERF_MEM_S(OP, NA) |\ | |
616 | PERF_MEM_S(LVL, NA) |\ | |
617 | PERF_MEM_S(SNOOP, NA) |\ | |
618 | PERF_MEM_S(LOCK, NA) |\ | |
619 | PERF_MEM_S(TLB, NA)) | |
620 | ||
fd0d000b RR |
621 | static inline void perf_sample_data_init(struct perf_sample_data *data, |
622 | u64 addr, u64 period) | |
dc1d628a | 623 | { |
fd0d000b | 624 | /* remaining struct members initialized in perf_prepare_sample() */ |
dc1d628a PZ |
625 | data->addr = addr; |
626 | data->raw = NULL; | |
bce38cd5 | 627 | data->br_stack = NULL; |
4018994f | 628 | data->period = period; |
c3feedf2 | 629 | data->weight = 0; |
770eee1f | 630 | data->data_src.val = PERF_MEM_NA; |
fdfbbd07 | 631 | data->txn = 0; |
dc1d628a PZ |
632 | } |
633 | ||
5622f295 MM |
634 | extern void perf_output_sample(struct perf_output_handle *handle, |
635 | struct perf_event_header *header, | |
636 | struct perf_sample_data *data, | |
cdd6c482 | 637 | struct perf_event *event); |
5622f295 MM |
638 | extern void perf_prepare_sample(struct perf_event_header *header, |
639 | struct perf_sample_data *data, | |
cdd6c482 | 640 | struct perf_event *event, |
5622f295 MM |
641 | struct pt_regs *regs); |
642 | ||
a8b0ca17 | 643 | extern int perf_event_overflow(struct perf_event *event, |
5622f295 MM |
644 | struct perf_sample_data *data, |
645 | struct pt_regs *regs); | |
df1a132b | 646 | |
6c7e550f FBH |
647 | static inline bool is_sampling_event(struct perf_event *event) |
648 | { | |
649 | return event->attr.sample_period != 0; | |
650 | } | |
651 | ||
3b6f9e5c | 652 | /* |
cdd6c482 | 653 | * Return 1 for a software event, 0 for a hardware event |
3b6f9e5c | 654 | */ |
cdd6c482 | 655 | static inline int is_software_event(struct perf_event *event) |
3b6f9e5c | 656 | { |
89a1e187 | 657 | return event->pmu->task_ctx_nr == perf_sw_context; |
3b6f9e5c PM |
658 | } |
659 | ||
c5905afb | 660 | extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX]; |
f29ac756 | 661 | |
a8b0ca17 | 662 | extern void __perf_sw_event(u32, u64, struct pt_regs *, u64); |
f29ac756 | 663 | |
b0f82b81 | 664 | #ifndef perf_arch_fetch_caller_regs |
e7e7ee2e | 665 | static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { } |
b0f82b81 | 666 | #endif |
5331d7b8 FW |
667 | |
668 | /* | |
669 | * Take a snapshot of the regs. Skip ip and frame pointer to | |
670 | * the nth caller. We only need a few of the regs: | |
671 | * - ip for PERF_SAMPLE_IP | |
672 | * - cs for user_mode() tests | |
673 | * - bp for callchains | |
674 | * - eflags, for future purposes, just in case | |
675 | */ | |
b0f82b81 | 676 | static inline void perf_fetch_caller_regs(struct pt_regs *regs) |
5331d7b8 | 677 | { |
5331d7b8 FW |
678 | memset(regs, 0, sizeof(*regs)); |
679 | ||
b0f82b81 | 680 | perf_arch_fetch_caller_regs(regs, CALLER_ADDR0); |
5331d7b8 FW |
681 | } |
682 | ||
7e54a5a0 | 683 | static __always_inline void |
a8b0ca17 | 684 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) |
e49a5bd3 | 685 | { |
7e54a5a0 PZ |
686 | struct pt_regs hot_regs; |
687 | ||
c5905afb | 688 | if (static_key_false(&perf_swevent_enabled[event_id])) { |
d430d3d7 JB |
689 | if (!regs) { |
690 | perf_fetch_caller_regs(&hot_regs); | |
691 | regs = &hot_regs; | |
692 | } | |
a8b0ca17 | 693 | __perf_sw_event(event_id, nr, regs, addr); |
e49a5bd3 FW |
694 | } |
695 | } | |
696 | ||
c5905afb | 697 | extern struct static_key_deferred perf_sched_events; |
ee6dcfa4 | 698 | |
ab0cce56 | 699 | static inline void perf_event_task_sched_in(struct task_struct *prev, |
a8d757ef | 700 | struct task_struct *task) |
ab0cce56 JO |
701 | { |
702 | if (static_key_false(&perf_sched_events.key)) | |
703 | __perf_event_task_sched_in(prev, task); | |
704 | } | |
705 | ||
706 | static inline void perf_event_task_sched_out(struct task_struct *prev, | |
707 | struct task_struct *next) | |
ee6dcfa4 | 708 | { |
a8b0ca17 | 709 | perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, NULL, 0); |
ee6dcfa4 | 710 | |
c5905afb | 711 | if (static_key_false(&perf_sched_events.key)) |
ab0cce56 | 712 | __perf_event_task_sched_out(prev, next); |
ee6dcfa4 PZ |
713 | } |
714 | ||
3af9e859 | 715 | extern void perf_event_mmap(struct vm_area_struct *vma); |
39447b38 | 716 | extern struct perf_guest_info_callbacks *perf_guest_cbs; |
dcf46b94 ZY |
717 | extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); |
718 | extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); | |
39447b38 | 719 | |
e041e328 | 720 | extern void perf_event_exec(void); |
82b89778 | 721 | extern void perf_event_comm(struct task_struct *tsk, bool exec); |
cdd6c482 | 722 | extern void perf_event_fork(struct task_struct *tsk); |
8d1b2d93 | 723 | |
56962b44 FW |
724 | /* Callchains */ |
725 | DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry); | |
726 | ||
e7e7ee2e IM |
727 | extern void perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs); |
728 | extern void perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs); | |
394ee076 | 729 | |
e7e7ee2e | 730 | static inline void perf_callchain_store(struct perf_callchain_entry *entry, u64 ip) |
70791ce9 FW |
731 | { |
732 | if (entry->nr < PERF_MAX_STACK_DEPTH) | |
733 | entry->ip[entry->nr++] = ip; | |
734 | } | |
394ee076 | 735 | |
cdd6c482 IM |
736 | extern int sysctl_perf_event_paranoid; |
737 | extern int sysctl_perf_event_mlock; | |
738 | extern int sysctl_perf_event_sample_rate; | |
14c63f17 DH |
739 | extern int sysctl_perf_cpu_time_max_percent; |
740 | ||
741 | extern void perf_sample_event_took(u64 sample_len_ns); | |
1ccd1549 | 742 | |
163ec435 PZ |
743 | extern int perf_proc_update_handler(struct ctl_table *table, int write, |
744 | void __user *buffer, size_t *lenp, | |
745 | loff_t *ppos); | |
14c63f17 DH |
746 | extern int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write, |
747 | void __user *buffer, size_t *lenp, | |
748 | loff_t *ppos); | |
749 | ||
163ec435 | 750 | |
320ebf09 PZ |
751 | static inline bool perf_paranoid_tracepoint_raw(void) |
752 | { | |
753 | return sysctl_perf_event_paranoid > -1; | |
754 | } | |
755 | ||
756 | static inline bool perf_paranoid_cpu(void) | |
757 | { | |
758 | return sysctl_perf_event_paranoid > 0; | |
759 | } | |
760 | ||
761 | static inline bool perf_paranoid_kernel(void) | |
762 | { | |
763 | return sysctl_perf_event_paranoid > 1; | |
764 | } | |
765 | ||
cdd6c482 | 766 | extern void perf_event_init(void); |
1c024eca PZ |
767 | extern void perf_tp_event(u64 addr, u64 count, void *record, |
768 | int entry_size, struct pt_regs *regs, | |
e6dab5ff AV |
769 | struct hlist_head *head, int rctx, |
770 | struct task_struct *task); | |
24f1e32c | 771 | extern void perf_bp_event(struct perf_event *event, void *data); |
0d905bca | 772 | |
9d23a90a | 773 | #ifndef perf_misc_flags |
e7e7ee2e IM |
774 | # define perf_misc_flags(regs) \ |
775 | (user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL) | |
776 | # define perf_instruction_pointer(regs) instruction_pointer(regs) | |
9d23a90a PM |
777 | #endif |
778 | ||
bce38cd5 SE |
779 | static inline bool has_branch_stack(struct perf_event *event) |
780 | { | |
781 | return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK; | |
782 | } | |
783 | ||
5622f295 | 784 | extern int perf_output_begin(struct perf_output_handle *handle, |
a7ac67ea | 785 | struct perf_event *event, unsigned int size); |
5622f295 | 786 | extern void perf_output_end(struct perf_output_handle *handle); |
91d7753a | 787 | extern unsigned int perf_output_copy(struct perf_output_handle *handle, |
5622f295 | 788 | const void *buf, unsigned int len); |
5685e0ff JO |
789 | extern unsigned int perf_output_skip(struct perf_output_handle *handle, |
790 | unsigned int len); | |
4ed7c92d PZ |
791 | extern int perf_swevent_get_recursion_context(void); |
792 | extern void perf_swevent_put_recursion_context(int rctx); | |
ab573844 | 793 | extern u64 perf_swevent_set_period(struct perf_event *event); |
44234adc FW |
794 | extern void perf_event_enable(struct perf_event *event); |
795 | extern void perf_event_disable(struct perf_event *event); | |
500ad2d8 | 796 | extern int __perf_event_disable(void *info); |
e9d2b064 | 797 | extern void perf_event_task_tick(void); |
e041e328 | 798 | #else /* !CONFIG_PERF_EVENTS: */ |
0793a61d | 799 | static inline void |
ab0cce56 JO |
800 | perf_event_task_sched_in(struct task_struct *prev, |
801 | struct task_struct *task) { } | |
802 | static inline void | |
803 | perf_event_task_sched_out(struct task_struct *prev, | |
804 | struct task_struct *next) { } | |
cdd6c482 IM |
805 | static inline int perf_event_init_task(struct task_struct *child) { return 0; } |
806 | static inline void perf_event_exit_task(struct task_struct *child) { } | |
807 | static inline void perf_event_free_task(struct task_struct *task) { } | |
4e231c79 | 808 | static inline void perf_event_delayed_put(struct task_struct *task) { } |
57c0c15b | 809 | static inline void perf_event_print_debug(void) { } |
57c0c15b IM |
810 | static inline int perf_event_task_disable(void) { return -EINVAL; } |
811 | static inline int perf_event_task_enable(void) { return -EINVAL; } | |
26ca5c11 AK |
812 | static inline int perf_event_refresh(struct perf_event *event, int refresh) |
813 | { | |
814 | return -EINVAL; | |
815 | } | |
15dbf27c | 816 | |
925d519a | 817 | static inline void |
a8b0ca17 | 818 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) { } |
24f1e32c | 819 | static inline void |
184f412c | 820 | perf_bp_event(struct perf_event *event, void *data) { } |
0a4a9391 | 821 | |
39447b38 | 822 | static inline int perf_register_guest_info_callbacks |
e7e7ee2e | 823 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 824 | static inline int perf_unregister_guest_info_callbacks |
e7e7ee2e | 825 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 826 | |
57c0c15b | 827 | static inline void perf_event_mmap(struct vm_area_struct *vma) { } |
e041e328 | 828 | static inline void perf_event_exec(void) { } |
82b89778 | 829 | static inline void perf_event_comm(struct task_struct *tsk, bool exec) { } |
cdd6c482 IM |
830 | static inline void perf_event_fork(struct task_struct *tsk) { } |
831 | static inline void perf_event_init(void) { } | |
184f412c | 832 | static inline int perf_swevent_get_recursion_context(void) { return -1; } |
4ed7c92d | 833 | static inline void perf_swevent_put_recursion_context(int rctx) { } |
ab573844 | 834 | static inline u64 perf_swevent_set_period(struct perf_event *event) { return 0; } |
44234adc FW |
835 | static inline void perf_event_enable(struct perf_event *event) { } |
836 | static inline void perf_event_disable(struct perf_event *event) { } | |
500ad2d8 | 837 | static inline int __perf_event_disable(void *info) { return -1; } |
e9d2b064 | 838 | static inline void perf_event_task_tick(void) { } |
0793a61d TG |
839 | #endif |
840 | ||
026249ef FW |
841 | #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_NO_HZ_FULL) |
842 | extern bool perf_event_can_stop_tick(void); | |
843 | #else | |
844 | static inline bool perf_event_can_stop_tick(void) { return true; } | |
845 | #endif | |
846 | ||
6c4d3bc9 DR |
847 | #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL) |
848 | extern void perf_restore_debug_store(void); | |
849 | #else | |
1d9d8639 | 850 | static inline void perf_restore_debug_store(void) { } |
0793a61d TG |
851 | #endif |
852 | ||
e7e7ee2e | 853 | #define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x)) |
5622f295 | 854 | |
3f6da390 | 855 | /* |
0a0fca9d | 856 | * This has to have a higher priority than migration_notifier in sched/core.c. |
3f6da390 | 857 | */ |
e7e7ee2e IM |
858 | #define perf_cpu_notifier(fn) \ |
859 | do { \ | |
0db0628d | 860 | static struct notifier_block fn##_nb = \ |
e7e7ee2e | 861 | { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ |
c13d38e4 | 862 | unsigned long cpu = smp_processor_id(); \ |
6760bca9 | 863 | unsigned long flags; \ |
f0bdb5e0 SB |
864 | \ |
865 | cpu_notifier_register_begin(); \ | |
e7e7ee2e | 866 | fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE, \ |
c13d38e4 | 867 | (void *)(unsigned long)cpu); \ |
6760bca9 | 868 | local_irq_save(flags); \ |
e7e7ee2e | 869 | fn(&fn##_nb, (unsigned long)CPU_STARTING, \ |
c13d38e4 | 870 | (void *)(unsigned long)cpu); \ |
6760bca9 | 871 | local_irq_restore(flags); \ |
e7e7ee2e | 872 | fn(&fn##_nb, (unsigned long)CPU_ONLINE, \ |
c13d38e4 | 873 | (void *)(unsigned long)cpu); \ |
f0bdb5e0 SB |
874 | __register_cpu_notifier(&fn##_nb); \ |
875 | cpu_notifier_register_done(); \ | |
3f6da390 PZ |
876 | } while (0) |
877 | ||
f0bdb5e0 SB |
878 | /* |
879 | * Bare-bones version of perf_cpu_notifier(), which doesn't invoke the | |
880 | * callback for already online CPUs. | |
881 | */ | |
882 | #define __perf_cpu_notifier(fn) \ | |
883 | do { \ | |
884 | static struct notifier_block fn##_nb = \ | |
885 | { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ | |
886 | \ | |
887 | __register_cpu_notifier(&fn##_nb); \ | |
888 | } while (0) | |
641cc938 | 889 | |
2663960c SB |
890 | struct perf_pmu_events_attr { |
891 | struct device_attribute attr; | |
892 | u64 id; | |
3a54aaa0 | 893 | const char *event_str; |
2663960c SB |
894 | }; |
895 | ||
896 | #define PMU_EVENT_ATTR(_name, _var, _id, _show) \ | |
897 | static struct perf_pmu_events_attr _var = { \ | |
898 | .attr = __ATTR(_name, 0444, _show, NULL), \ | |
899 | .id = _id, \ | |
900 | }; | |
901 | ||
641cc938 JO |
902 | #define PMU_FORMAT_ATTR(_name, _format) \ |
903 | static ssize_t \ | |
904 | _name##_show(struct device *dev, \ | |
905 | struct device_attribute *attr, \ | |
906 | char *page) \ | |
907 | { \ | |
908 | BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE); \ | |
909 | return sprintf(page, _format "\n"); \ | |
910 | } \ | |
911 | \ | |
912 | static struct device_attribute format_attr_##_name = __ATTR_RO(_name) | |
913 | ||
cdd6c482 | 914 | #endif /* _LINUX_PERF_EVENT_H */ |