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> |
39bed6cb | 56 | #include <linux/cgroup.h> |
fa588151 | 57 | #include <asm/local.h> |
f3dfd265 | 58 | |
f9188e02 PZ |
59 | struct perf_callchain_entry { |
60 | __u64 nr; | |
61 | __u64 ip[PERF_MAX_STACK_DEPTH]; | |
62 | }; | |
63 | ||
3a43ce68 FW |
64 | struct perf_raw_record { |
65 | u32 size; | |
66 | void *data; | |
f413cdb8 FW |
67 | }; |
68 | ||
bce38cd5 SE |
69 | /* |
70 | * branch stack layout: | |
71 | * nr: number of taken branches stored in entries[] | |
72 | * | |
73 | * Note that nr can vary from sample to sample | |
74 | * branches (to, from) are stored from most recent | |
75 | * to least recent, i.e., entries[0] contains the most | |
76 | * recent branch. | |
77 | */ | |
caff2bef PZ |
78 | struct perf_branch_stack { |
79 | __u64 nr; | |
80 | struct perf_branch_entry entries[0]; | |
81 | }; | |
82 | ||
f3dfd265 PM |
83 | struct task_struct; |
84 | ||
efc9f05d SE |
85 | /* |
86 | * extra PMU register associated with an event | |
87 | */ | |
88 | struct hw_perf_event_extra { | |
89 | u64 config; /* register value */ | |
90 | unsigned int reg; /* register address or index */ | |
91 | int alloc; /* extra register already allocated */ | |
92 | int idx; /* index in shared_regs->regs[] */ | |
93 | }; | |
94 | ||
0793a61d | 95 | /** |
cdd6c482 | 96 | * struct hw_perf_event - performance event hardware details: |
0793a61d | 97 | */ |
cdd6c482 IM |
98 | struct hw_perf_event { |
99 | #ifdef CONFIG_PERF_EVENTS | |
d6d020e9 PZ |
100 | union { |
101 | struct { /* hardware */ | |
a308444c | 102 | u64 config; |
447a194b | 103 | u64 last_tag; |
a308444c | 104 | unsigned long config_base; |
cdd6c482 | 105 | unsigned long event_base; |
c48b6053 | 106 | int event_base_rdpmc; |
a308444c | 107 | int idx; |
447a194b | 108 | int last_cpu; |
9fac2cf3 | 109 | int flags; |
bce38cd5 | 110 | |
efc9f05d | 111 | struct hw_perf_event_extra extra_reg; |
bce38cd5 | 112 | struct hw_perf_event_extra branch_reg; |
d6d020e9 | 113 | }; |
721a669b | 114 | struct { /* software */ |
a308444c | 115 | struct hrtimer hrtimer; |
d6d020e9 | 116 | }; |
f22c1bb6 | 117 | struct { /* tracepoint */ |
f22c1bb6 ON |
118 | /* for tp_event->class */ |
119 | struct list_head tp_list; | |
120 | }; | |
4afbb24c MF |
121 | struct { /* intel_cqm */ |
122 | int cqm_state; | |
b3df4ec4 | 123 | u32 cqm_rmid; |
4afbb24c MF |
124 | struct list_head cqm_events_entry; |
125 | struct list_head cqm_groups_entry; | |
126 | struct list_head cqm_group_entry; | |
127 | }; | |
ec0d7729 AS |
128 | struct { /* itrace */ |
129 | int itrace_started; | |
130 | }; | |
24f1e32c | 131 | #ifdef CONFIG_HAVE_HW_BREAKPOINT |
45a73372 | 132 | struct { /* breakpoint */ |
d580ff86 PZ |
133 | /* |
134 | * Crufty hack to avoid the chicken and egg | |
135 | * problem hw_breakpoint has with context | |
136 | * creation and event initalization. | |
137 | */ | |
f22c1bb6 ON |
138 | struct arch_hw_breakpoint info; |
139 | struct list_head bp_list; | |
45a73372 | 140 | }; |
24f1e32c | 141 | #endif |
d6d020e9 | 142 | }; |
b0e87875 PZ |
143 | /* |
144 | * If the event is a per task event, this will point to the task in | |
145 | * question. See the comment in perf_event_alloc(). | |
146 | */ | |
50f16a8b | 147 | struct task_struct *target; |
b0e87875 PZ |
148 | |
149 | /* | |
150 | * hw_perf_event::state flags; used to track the PERF_EF_* state. | |
151 | */ | |
152 | #define PERF_HES_STOPPED 0x01 /* the counter is stopped */ | |
153 | #define PERF_HES_UPTODATE 0x02 /* event->count up-to-date */ | |
154 | #define PERF_HES_ARCH 0x04 | |
155 | ||
a4eaf7f1 | 156 | int state; |
b0e87875 PZ |
157 | |
158 | /* | |
159 | * The last observed hardware counter value, updated with a | |
160 | * local64_cmpxchg() such that pmu::read() can be called nested. | |
161 | */ | |
e7850595 | 162 | local64_t prev_count; |
b0e87875 PZ |
163 | |
164 | /* | |
165 | * The period to start the next sample with. | |
166 | */ | |
b23f3325 | 167 | u64 sample_period; |
b0e87875 PZ |
168 | |
169 | /* | |
170 | * The period we started this sample with. | |
171 | */ | |
9e350de3 | 172 | u64 last_period; |
b0e87875 PZ |
173 | |
174 | /* | |
175 | * However much is left of the current period; note that this is | |
176 | * a full 64bit value and allows for generation of periods longer | |
177 | * than hardware might allow. | |
178 | */ | |
e7850595 | 179 | local64_t period_left; |
b0e87875 PZ |
180 | |
181 | /* | |
182 | * State for throttling the event, see __perf_event_overflow() and | |
183 | * perf_adjust_freq_unthr_context(). | |
184 | */ | |
e050e3f0 | 185 | u64 interrupts_seq; |
60db5e09 | 186 | u64 interrupts; |
6a24ed6c | 187 | |
b0e87875 PZ |
188 | /* |
189 | * State for freq target events, see __perf_event_overflow() and | |
190 | * perf_adjust_freq_unthr_context(). | |
191 | */ | |
abd50713 PZ |
192 | u64 freq_time_stamp; |
193 | u64 freq_count_stamp; | |
ee06094f | 194 | #endif |
0793a61d TG |
195 | }; |
196 | ||
cdd6c482 | 197 | struct perf_event; |
621a01ea | 198 | |
8d2cacbb PZ |
199 | /* |
200 | * Common implementation detail of pmu::{start,commit,cancel}_txn | |
201 | */ | |
fbbe0701 | 202 | #define PERF_PMU_TXN_ADD 0x1 /* txn to add/schedule event on PMU */ |
4a00c16e | 203 | #define PERF_PMU_TXN_READ 0x2 /* txn to read event group from PMU */ |
fbbe0701 | 204 | |
53b25335 VW |
205 | /** |
206 | * pmu::capabilities flags | |
207 | */ | |
208 | #define PERF_PMU_CAP_NO_INTERRUPT 0x01 | |
34f43927 | 209 | #define PERF_PMU_CAP_NO_NMI 0x02 |
0a4e38e6 | 210 | #define PERF_PMU_CAP_AUX_NO_SG 0x04 |
6a279230 | 211 | #define PERF_PMU_CAP_AUX_SW_DOUBLEBUF 0x08 |
bed5b25a | 212 | #define PERF_PMU_CAP_EXCLUSIVE 0x10 |
ec0d7729 | 213 | #define PERF_PMU_CAP_ITRACE 0x20 |
53b25335 | 214 | |
621a01ea | 215 | /** |
4aeb0b42 | 216 | * struct pmu - generic performance monitoring unit |
621a01ea | 217 | */ |
4aeb0b42 | 218 | struct pmu { |
b0a873eb PZ |
219 | struct list_head entry; |
220 | ||
c464c76e | 221 | struct module *module; |
abe43400 | 222 | struct device *dev; |
0c9d42ed | 223 | const struct attribute_group **attr_groups; |
03d8e80b | 224 | const char *name; |
2e80a82a PZ |
225 | int type; |
226 | ||
53b25335 VW |
227 | /* |
228 | * various common per-pmu feature flags | |
229 | */ | |
230 | int capabilities; | |
231 | ||
108b02cf PZ |
232 | int * __percpu pmu_disable_count; |
233 | struct perf_cpu_context * __percpu pmu_cpu_context; | |
bed5b25a | 234 | atomic_t exclusive_cnt; /* < 0: cpu; > 0: tsk */ |
8dc85d54 | 235 | int task_ctx_nr; |
62b85639 | 236 | int hrtimer_interval_ms; |
6bde9b6c LM |
237 | |
238 | /* | |
a4eaf7f1 PZ |
239 | * Fully disable/enable this PMU, can be used to protect from the PMI |
240 | * as well as for lazy/batch writing of the MSRs. | |
6bde9b6c | 241 | */ |
ad5133b7 PZ |
242 | void (*pmu_enable) (struct pmu *pmu); /* optional */ |
243 | void (*pmu_disable) (struct pmu *pmu); /* optional */ | |
6bde9b6c | 244 | |
8d2cacbb | 245 | /* |
a4eaf7f1 | 246 | * Try and initialize the event for this PMU. |
b0e87875 PZ |
247 | * |
248 | * Returns: | |
249 | * -ENOENT -- @event is not for this PMU | |
250 | * | |
251 | * -ENODEV -- @event is for this PMU but PMU not present | |
252 | * -EBUSY -- @event is for this PMU but PMU temporarily unavailable | |
253 | * -EINVAL -- @event is for this PMU but @event is not valid | |
254 | * -EOPNOTSUPP -- @event is for this PMU, @event is valid, but not supported | |
255 | * -EACCESS -- @event is for this PMU, @event is valid, but no privilidges | |
256 | * | |
257 | * 0 -- @event is for this PMU and valid | |
258 | * | |
259 | * Other error return values are allowed. | |
8d2cacbb | 260 | */ |
b0a873eb PZ |
261 | int (*event_init) (struct perf_event *event); |
262 | ||
1e0fb9ec AL |
263 | /* |
264 | * Notification that the event was mapped or unmapped. Called | |
265 | * in the context of the mapping task. | |
266 | */ | |
267 | void (*event_mapped) (struct perf_event *event); /*optional*/ | |
268 | void (*event_unmapped) (struct perf_event *event); /*optional*/ | |
269 | ||
b0e87875 PZ |
270 | /* |
271 | * Flags for ->add()/->del()/ ->start()/->stop(). There are | |
272 | * matching hw_perf_event::state flags. | |
273 | */ | |
a4eaf7f1 PZ |
274 | #define PERF_EF_START 0x01 /* start the counter when adding */ |
275 | #define PERF_EF_RELOAD 0x02 /* reload the counter when starting */ | |
276 | #define PERF_EF_UPDATE 0x04 /* update the counter when stopping */ | |
277 | ||
8d2cacbb | 278 | /* |
b0e87875 PZ |
279 | * Adds/Removes a counter to/from the PMU, can be done inside a |
280 | * transaction, see the ->*_txn() methods. | |
281 | * | |
282 | * The add/del callbacks will reserve all hardware resources required | |
283 | * to service the event, this includes any counter constraint | |
284 | * scheduling etc. | |
285 | * | |
286 | * Called with IRQs disabled and the PMU disabled on the CPU the event | |
287 | * is on. | |
288 | * | |
289 | * ->add() called without PERF_EF_START should result in the same state | |
290 | * as ->add() followed by ->stop(). | |
291 | * | |
292 | * ->del() must always PERF_EF_UPDATE stop an event. If it calls | |
293 | * ->stop() that must deal with already being stopped without | |
294 | * PERF_EF_UPDATE. | |
a4eaf7f1 PZ |
295 | */ |
296 | int (*add) (struct perf_event *event, int flags); | |
297 | void (*del) (struct perf_event *event, int flags); | |
298 | ||
299 | /* | |
b0e87875 PZ |
300 | * Starts/Stops a counter present on the PMU. |
301 | * | |
302 | * The PMI handler should stop the counter when perf_event_overflow() | |
303 | * returns !0. ->start() will be used to continue. | |
304 | * | |
305 | * Also used to change the sample period. | |
306 | * | |
307 | * Called with IRQs disabled and the PMU disabled on the CPU the event | |
308 | * is on -- will be called from NMI context with the PMU generates | |
309 | * NMIs. | |
310 | * | |
311 | * ->stop() with PERF_EF_UPDATE will read the counter and update | |
312 | * period/count values like ->read() would. | |
313 | * | |
314 | * ->start() with PERF_EF_RELOAD will reprogram the the counter | |
315 | * value, must be preceded by a ->stop() with PERF_EF_UPDATE. | |
a4eaf7f1 PZ |
316 | */ |
317 | void (*start) (struct perf_event *event, int flags); | |
318 | void (*stop) (struct perf_event *event, int flags); | |
319 | ||
320 | /* | |
321 | * Updates the counter value of the event. | |
b0e87875 PZ |
322 | * |
323 | * For sampling capable PMUs this will also update the software period | |
324 | * hw_perf_event::period_left field. | |
a4eaf7f1 | 325 | */ |
cdd6c482 | 326 | void (*read) (struct perf_event *event); |
6bde9b6c LM |
327 | |
328 | /* | |
24cd7f54 PZ |
329 | * Group events scheduling is treated as a transaction, add |
330 | * group events as a whole and perform one schedulability test. | |
331 | * If the test fails, roll back the whole group | |
a4eaf7f1 PZ |
332 | * |
333 | * Start the transaction, after this ->add() doesn't need to | |
24cd7f54 | 334 | * do schedulability tests. |
fbbe0701 SB |
335 | * |
336 | * Optional. | |
8d2cacbb | 337 | */ |
fbbe0701 | 338 | void (*start_txn) (struct pmu *pmu, unsigned int txn_flags); |
8d2cacbb | 339 | /* |
a4eaf7f1 | 340 | * If ->start_txn() disabled the ->add() schedulability test |
8d2cacbb PZ |
341 | * then ->commit_txn() is required to perform one. On success |
342 | * the transaction is closed. On error the transaction is kept | |
343 | * open until ->cancel_txn() is called. | |
fbbe0701 SB |
344 | * |
345 | * Optional. | |
8d2cacbb | 346 | */ |
fbbe0701 | 347 | int (*commit_txn) (struct pmu *pmu); |
8d2cacbb | 348 | /* |
a4eaf7f1 | 349 | * Will cancel the transaction, assumes ->del() is called |
25985edc | 350 | * for each successful ->add() during the transaction. |
fbbe0701 SB |
351 | * |
352 | * Optional. | |
8d2cacbb | 353 | */ |
fbbe0701 | 354 | void (*cancel_txn) (struct pmu *pmu); |
35edc2a5 PZ |
355 | |
356 | /* | |
357 | * Will return the value for perf_event_mmap_page::index for this event, | |
358 | * if no implementation is provided it will default to: event->hw.idx + 1. | |
359 | */ | |
360 | int (*event_idx) (struct perf_event *event); /*optional */ | |
d010b332 | 361 | |
ba532500 YZ |
362 | /* |
363 | * context-switches callback | |
364 | */ | |
365 | void (*sched_task) (struct perf_event_context *ctx, | |
366 | bool sched_in); | |
4af57ef2 YZ |
367 | /* |
368 | * PMU specific data size | |
369 | */ | |
370 | size_t task_ctx_size; | |
ba532500 | 371 | |
eacd3ecc MF |
372 | |
373 | /* | |
374 | * Return the count value for a counter. | |
375 | */ | |
376 | u64 (*count) (struct perf_event *event); /*optional*/ | |
45bfb2e5 PZ |
377 | |
378 | /* | |
379 | * Set up pmu-private data structures for an AUX area | |
380 | */ | |
381 | void *(*setup_aux) (int cpu, void **pages, | |
382 | int nr_pages, bool overwrite); | |
383 | /* optional */ | |
384 | ||
385 | /* | |
386 | * Free pmu-private AUX data structures | |
387 | */ | |
388 | void (*free_aux) (void *aux); /* optional */ | |
66eb579e MR |
389 | |
390 | /* | |
391 | * Filter events for PMU-specific reasons. | |
392 | */ | |
393 | int (*filter_match) (struct perf_event *event); /* optional */ | |
621a01ea IM |
394 | }; |
395 | ||
6a930700 | 396 | /** |
cdd6c482 | 397 | * enum perf_event_active_state - the states of a event |
6a930700 | 398 | */ |
cdd6c482 | 399 | enum perf_event_active_state { |
179033b3 | 400 | PERF_EVENT_STATE_EXIT = -3, |
57c0c15b | 401 | PERF_EVENT_STATE_ERROR = -2, |
cdd6c482 IM |
402 | PERF_EVENT_STATE_OFF = -1, |
403 | PERF_EVENT_STATE_INACTIVE = 0, | |
57c0c15b | 404 | PERF_EVENT_STATE_ACTIVE = 1, |
6a930700 IM |
405 | }; |
406 | ||
9b51f66d | 407 | struct file; |
453f19ee PZ |
408 | struct perf_sample_data; |
409 | ||
a8b0ca17 | 410 | typedef void (*perf_overflow_handler_t)(struct perf_event *, |
b326e956 FW |
411 | struct perf_sample_data *, |
412 | struct pt_regs *regs); | |
413 | ||
d6f962b5 | 414 | enum perf_group_flag { |
e7e7ee2e | 415 | PERF_GROUP_SOFTWARE = 0x1, |
d6f962b5 FW |
416 | }; |
417 | ||
e7e7ee2e IM |
418 | #define SWEVENT_HLIST_BITS 8 |
419 | #define SWEVENT_HLIST_SIZE (1 << SWEVENT_HLIST_BITS) | |
76e1d904 FW |
420 | |
421 | struct swevent_hlist { | |
e7e7ee2e IM |
422 | struct hlist_head heads[SWEVENT_HLIST_SIZE]; |
423 | struct rcu_head rcu_head; | |
76e1d904 FW |
424 | }; |
425 | ||
8a49542c PZ |
426 | #define PERF_ATTACH_CONTEXT 0x01 |
427 | #define PERF_ATTACH_GROUP 0x02 | |
d580ff86 | 428 | #define PERF_ATTACH_TASK 0x04 |
4af57ef2 | 429 | #define PERF_ATTACH_TASK_DATA 0x08 |
8a49542c | 430 | |
877c6856 | 431 | struct perf_cgroup; |
76369139 FW |
432 | struct ring_buffer; |
433 | ||
0793a61d | 434 | /** |
cdd6c482 | 435 | * struct perf_event - performance event kernel representation: |
0793a61d | 436 | */ |
cdd6c482 IM |
437 | struct perf_event { |
438 | #ifdef CONFIG_PERF_EVENTS | |
9886167d PZ |
439 | /* |
440 | * entry onto perf_event_context::event_list; | |
441 | * modifications require ctx->lock | |
442 | * RCU safe iterations. | |
443 | */ | |
592903cd | 444 | struct list_head event_entry; |
9886167d PZ |
445 | |
446 | /* | |
447 | * XXX: group_entry and sibling_list should be mutually exclusive; | |
448 | * either you're a sibling on a group, or you're the group leader. | |
449 | * Rework the code to always use the same list element. | |
450 | * | |
451 | * Locked for modification by both ctx->mutex and ctx->lock; holding | |
452 | * either sufficies for read. | |
453 | */ | |
454 | struct list_head group_entry; | |
04289bb9 | 455 | struct list_head sibling_list; |
9886167d PZ |
456 | |
457 | /* | |
458 | * We need storage to track the entries in perf_pmu_migrate_context; we | |
459 | * cannot use the event_entry because of RCU and we want to keep the | |
460 | * group in tact which avoids us using the other two entries. | |
461 | */ | |
462 | struct list_head migrate_entry; | |
463 | ||
f3ae75de SE |
464 | struct hlist_node hlist_entry; |
465 | struct list_head active_entry; | |
0127c3ea | 466 | int nr_siblings; |
d6f962b5 | 467 | int group_flags; |
cdd6c482 | 468 | struct perf_event *group_leader; |
a4eaf7f1 | 469 | struct pmu *pmu; |
04289bb9 | 470 | |
cdd6c482 | 471 | enum perf_event_active_state state; |
8a49542c | 472 | unsigned int attach_state; |
e7850595 | 473 | local64_t count; |
a6e6dea6 | 474 | atomic64_t child_count; |
ee06094f | 475 | |
53cfbf59 | 476 | /* |
cdd6c482 | 477 | * These are the total time in nanoseconds that the event |
53cfbf59 | 478 | * has been enabled (i.e. eligible to run, and the task has |
cdd6c482 | 479 | * been scheduled in, if this is a per-task event) |
53cfbf59 PM |
480 | * and running (scheduled onto the CPU), respectively. |
481 | * | |
482 | * They are computed from tstamp_enabled, tstamp_running and | |
cdd6c482 | 483 | * tstamp_stopped when the event is in INACTIVE or ACTIVE state. |
53cfbf59 PM |
484 | */ |
485 | u64 total_time_enabled; | |
486 | u64 total_time_running; | |
487 | ||
488 | /* | |
489 | * These are timestamps used for computing total_time_enabled | |
cdd6c482 | 490 | * and total_time_running when the event is in INACTIVE or |
53cfbf59 PM |
491 | * ACTIVE state, measured in nanoseconds from an arbitrary point |
492 | * in time. | |
cdd6c482 IM |
493 | * tstamp_enabled: the notional time when the event was enabled |
494 | * tstamp_running: the notional time when the event was scheduled on | |
53cfbf59 | 495 | * tstamp_stopped: in INACTIVE state, the notional time when the |
cdd6c482 | 496 | * event was scheduled off. |
53cfbf59 PM |
497 | */ |
498 | u64 tstamp_enabled; | |
499 | u64 tstamp_running; | |
500 | u64 tstamp_stopped; | |
501 | ||
eed01528 SE |
502 | /* |
503 | * timestamp shadows the actual context timing but it can | |
504 | * be safely used in NMI interrupt context. It reflects the | |
505 | * context time as it was when the event was last scheduled in. | |
506 | * | |
507 | * ctx_time already accounts for ctx->timestamp. Therefore to | |
508 | * compute ctx_time for a sample, simply add perf_clock(). | |
509 | */ | |
510 | u64 shadow_ctx_time; | |
511 | ||
24f1e32c | 512 | struct perf_event_attr attr; |
c320c7b7 | 513 | u16 header_size; |
6844c09d | 514 | u16 id_header_size; |
c320c7b7 | 515 | u16 read_size; |
cdd6c482 | 516 | struct hw_perf_event hw; |
0793a61d | 517 | |
cdd6c482 | 518 | struct perf_event_context *ctx; |
a6fa941d | 519 | atomic_long_t refcount; |
0793a61d | 520 | |
53cfbf59 PM |
521 | /* |
522 | * These accumulate total time (in nanoseconds) that children | |
cdd6c482 | 523 | * events have been enabled and running, respectively. |
53cfbf59 PM |
524 | */ |
525 | atomic64_t child_total_time_enabled; | |
526 | atomic64_t child_total_time_running; | |
527 | ||
0793a61d | 528 | /* |
d859e29f | 529 | * Protect attach/detach and child_list: |
0793a61d | 530 | */ |
fccc714b PZ |
531 | struct mutex child_mutex; |
532 | struct list_head child_list; | |
cdd6c482 | 533 | struct perf_event *parent; |
0793a61d TG |
534 | |
535 | int oncpu; | |
536 | int cpu; | |
537 | ||
082ff5a2 PZ |
538 | struct list_head owner_entry; |
539 | struct task_struct *owner; | |
540 | ||
7b732a75 PZ |
541 | /* mmap bits */ |
542 | struct mutex mmap_mutex; | |
543 | atomic_t mmap_count; | |
26cb63ad | 544 | |
76369139 | 545 | struct ring_buffer *rb; |
10c6db11 | 546 | struct list_head rb_entry; |
b69cf536 PZ |
547 | unsigned long rcu_batches; |
548 | int rcu_pending; | |
37d81828 | 549 | |
7b732a75 | 550 | /* poll related */ |
0793a61d | 551 | wait_queue_head_t waitq; |
3c446b3d | 552 | struct fasync_struct *fasync; |
79f14641 PZ |
553 | |
554 | /* delayed work for NMIs and such */ | |
555 | int pending_wakeup; | |
4c9e2542 | 556 | int pending_kill; |
79f14641 | 557 | int pending_disable; |
e360adbe | 558 | struct irq_work pending; |
592903cd | 559 | |
79f14641 PZ |
560 | atomic_t event_limit; |
561 | ||
cdd6c482 | 562 | void (*destroy)(struct perf_event *); |
592903cd | 563 | struct rcu_head rcu_head; |
709e50cf PZ |
564 | |
565 | struct pid_namespace *ns; | |
8e5799b1 | 566 | u64 id; |
6fb2915d | 567 | |
34f43927 | 568 | u64 (*clock)(void); |
b326e956 | 569 | perf_overflow_handler_t overflow_handler; |
4dc0da86 | 570 | void *overflow_handler_context; |
453f19ee | 571 | |
07b139c8 | 572 | #ifdef CONFIG_EVENT_TRACING |
2425bcb9 | 573 | struct trace_event_call *tp_event; |
6fb2915d | 574 | struct event_filter *filter; |
ced39002 JO |
575 | #ifdef CONFIG_FUNCTION_TRACER |
576 | struct ftrace_ops ftrace_ops; | |
577 | #endif | |
ee06094f | 578 | #endif |
6fb2915d | 579 | |
e5d1367f SE |
580 | #ifdef CONFIG_CGROUP_PERF |
581 | struct perf_cgroup *cgrp; /* cgroup event is attach to */ | |
582 | int cgrp_defer_enabled; | |
583 | #endif | |
584 | ||
6fb2915d | 585 | #endif /* CONFIG_PERF_EVENTS */ |
0793a61d TG |
586 | }; |
587 | ||
588 | /** | |
cdd6c482 | 589 | * struct perf_event_context - event context structure |
0793a61d | 590 | * |
cdd6c482 | 591 | * Used as a container for task events and CPU events as well: |
0793a61d | 592 | */ |
cdd6c482 | 593 | struct perf_event_context { |
108b02cf | 594 | struct pmu *pmu; |
0793a61d | 595 | /* |
cdd6c482 | 596 | * Protect the states of the events in the list, |
d859e29f | 597 | * nr_active, and the list: |
0793a61d | 598 | */ |
e625cce1 | 599 | raw_spinlock_t lock; |
d859e29f | 600 | /* |
cdd6c482 | 601 | * Protect the list of events. Locking either mutex or lock |
d859e29f PM |
602 | * is sufficient to ensure the list doesn't change; to change |
603 | * the list you need to lock both the mutex and the spinlock. | |
604 | */ | |
a308444c | 605 | struct mutex mutex; |
04289bb9 | 606 | |
2fde4f94 | 607 | struct list_head active_ctx_list; |
889ff015 FW |
608 | struct list_head pinned_groups; |
609 | struct list_head flexible_groups; | |
a308444c | 610 | struct list_head event_list; |
cdd6c482 | 611 | int nr_events; |
a308444c IM |
612 | int nr_active; |
613 | int is_active; | |
bfbd3381 | 614 | int nr_stat; |
0f5a2601 | 615 | int nr_freq; |
dddd3379 | 616 | int rotate_disable; |
a308444c IM |
617 | atomic_t refcount; |
618 | struct task_struct *task; | |
53cfbf59 PM |
619 | |
620 | /* | |
4af4998b | 621 | * Context clock, runs when context enabled. |
53cfbf59 | 622 | */ |
a308444c IM |
623 | u64 time; |
624 | u64 timestamp; | |
564c2b21 PM |
625 | |
626 | /* | |
627 | * These fields let us detect when two contexts have both | |
628 | * been cloned (inherited) from a common ancestor. | |
629 | */ | |
cdd6c482 | 630 | struct perf_event_context *parent_ctx; |
a308444c IM |
631 | u64 parent_gen; |
632 | u64 generation; | |
633 | int pin_count; | |
d010b332 | 634 | int nr_cgroups; /* cgroup evts */ |
4af57ef2 | 635 | void *task_ctx_data; /* pmu specific data */ |
28009ce4 | 636 | struct rcu_head rcu_head; |
0793a61d TG |
637 | }; |
638 | ||
7ae07ea3 FW |
639 | /* |
640 | * Number of contexts where an event can trigger: | |
e7e7ee2e | 641 | * task, softirq, hardirq, nmi. |
7ae07ea3 FW |
642 | */ |
643 | #define PERF_NR_CONTEXTS 4 | |
644 | ||
0793a61d | 645 | /** |
cdd6c482 | 646 | * struct perf_event_cpu_context - per cpu event context structure |
0793a61d TG |
647 | */ |
648 | struct perf_cpu_context { | |
cdd6c482 IM |
649 | struct perf_event_context ctx; |
650 | struct perf_event_context *task_ctx; | |
0793a61d | 651 | int active_oncpu; |
3b6f9e5c | 652 | int exclusive; |
4cfafd30 PZ |
653 | |
654 | raw_spinlock_t hrtimer_lock; | |
9e630205 SE |
655 | struct hrtimer hrtimer; |
656 | ktime_t hrtimer_interval; | |
4cfafd30 PZ |
657 | unsigned int hrtimer_active; |
658 | ||
3f1f3320 | 659 | struct pmu *unique_pmu; |
e5d1367f | 660 | struct perf_cgroup *cgrp; |
0793a61d TG |
661 | }; |
662 | ||
5622f295 | 663 | struct perf_output_handle { |
57c0c15b | 664 | struct perf_event *event; |
76369139 | 665 | struct ring_buffer *rb; |
6d1acfd5 | 666 | unsigned long wakeup; |
5d967a8b | 667 | unsigned long size; |
fdc26706 AS |
668 | union { |
669 | void *addr; | |
670 | unsigned long head; | |
671 | }; | |
5d967a8b | 672 | int page; |
5622f295 MM |
673 | }; |
674 | ||
39bed6cb MF |
675 | #ifdef CONFIG_CGROUP_PERF |
676 | ||
677 | /* | |
678 | * perf_cgroup_info keeps track of time_enabled for a cgroup. | |
679 | * This is a per-cpu dynamically allocated data structure. | |
680 | */ | |
681 | struct perf_cgroup_info { | |
682 | u64 time; | |
683 | u64 timestamp; | |
684 | }; | |
685 | ||
686 | struct perf_cgroup { | |
687 | struct cgroup_subsys_state css; | |
688 | struct perf_cgroup_info __percpu *info; | |
689 | }; | |
690 | ||
691 | /* | |
692 | * Must ensure cgroup is pinned (css_get) before calling | |
693 | * this function. In other words, we cannot call this function | |
694 | * if there is no cgroup event for the current CPU context. | |
695 | */ | |
696 | static inline struct perf_cgroup * | |
614e4c4e | 697 | perf_cgroup_from_task(struct task_struct *task, struct perf_event_context *ctx) |
39bed6cb | 698 | { |
614e4c4e SE |
699 | return container_of(task_css_check(task, perf_event_cgrp_id, |
700 | ctx ? lockdep_is_held(&ctx->lock) | |
701 | : true), | |
39bed6cb MF |
702 | struct perf_cgroup, css); |
703 | } | |
704 | #endif /* CONFIG_CGROUP_PERF */ | |
705 | ||
cdd6c482 | 706 | #ifdef CONFIG_PERF_EVENTS |
829b42dd | 707 | |
fdc26706 AS |
708 | extern void *perf_aux_output_begin(struct perf_output_handle *handle, |
709 | struct perf_event *event); | |
710 | extern void perf_aux_output_end(struct perf_output_handle *handle, | |
711 | unsigned long size, bool truncated); | |
712 | extern int perf_aux_output_skip(struct perf_output_handle *handle, | |
713 | unsigned long size); | |
714 | extern void *perf_get_aux(struct perf_output_handle *handle); | |
715 | ||
03d8e80b | 716 | extern int perf_pmu_register(struct pmu *pmu, const char *name, int type); |
b0a873eb | 717 | extern void perf_pmu_unregister(struct pmu *pmu); |
621a01ea | 718 | |
3bf101ba | 719 | extern int perf_num_counters(void); |
84c79910 | 720 | extern const char *perf_pmu_name(void); |
ab0cce56 JO |
721 | extern void __perf_event_task_sched_in(struct task_struct *prev, |
722 | struct task_struct *task); | |
723 | extern void __perf_event_task_sched_out(struct task_struct *prev, | |
724 | struct task_struct *next); | |
cdd6c482 IM |
725 | extern int perf_event_init_task(struct task_struct *child); |
726 | extern void perf_event_exit_task(struct task_struct *child); | |
727 | extern void perf_event_free_task(struct task_struct *task); | |
4e231c79 | 728 | extern void perf_event_delayed_put(struct task_struct *task); |
e03e7ee3 | 729 | extern struct file *perf_event_get(unsigned int fd); |
ffe8690c | 730 | extern const struct perf_event_attr *perf_event_attrs(struct perf_event *event); |
cdd6c482 | 731 | extern void perf_event_print_debug(void); |
33696fc0 PZ |
732 | extern void perf_pmu_disable(struct pmu *pmu); |
733 | extern void perf_pmu_enable(struct pmu *pmu); | |
ba532500 YZ |
734 | extern void perf_sched_cb_dec(struct pmu *pmu); |
735 | extern void perf_sched_cb_inc(struct pmu *pmu); | |
cdd6c482 IM |
736 | extern int perf_event_task_disable(void); |
737 | extern int perf_event_task_enable(void); | |
26ca5c11 | 738 | extern int perf_event_refresh(struct perf_event *event, int refresh); |
cdd6c482 | 739 | extern void perf_event_update_userpage(struct perf_event *event); |
fb0459d7 AV |
740 | extern int perf_event_release_kernel(struct perf_event *event); |
741 | extern struct perf_event * | |
742 | perf_event_create_kernel_counter(struct perf_event_attr *attr, | |
743 | int cpu, | |
38a81da2 | 744 | struct task_struct *task, |
4dc0da86 AK |
745 | perf_overflow_handler_t callback, |
746 | void *context); | |
0cda4c02 YZ |
747 | extern void perf_pmu_migrate_context(struct pmu *pmu, |
748 | int src_cpu, int dst_cpu); | |
ffe8690c | 749 | extern u64 perf_event_read_local(struct perf_event *event); |
59ed446f PZ |
750 | extern u64 perf_event_read_value(struct perf_event *event, |
751 | u64 *enabled, u64 *running); | |
5c92d124 | 752 | |
d010b332 | 753 | |
df1a132b | 754 | struct perf_sample_data { |
2565711f PZ |
755 | /* |
756 | * Fields set by perf_sample_data_init(), group so as to | |
757 | * minimize the cachelines touched. | |
758 | */ | |
759 | u64 addr; | |
760 | struct perf_raw_record *raw; | |
761 | struct perf_branch_stack *br_stack; | |
762 | u64 period; | |
763 | u64 weight; | |
764 | u64 txn; | |
765 | union perf_mem_data_src data_src; | |
5622f295 | 766 | |
2565711f PZ |
767 | /* |
768 | * The other fields, optionally {set,used} by | |
769 | * perf_{prepare,output}_sample(). | |
770 | */ | |
771 | u64 type; | |
5622f295 MM |
772 | u64 ip; |
773 | struct { | |
774 | u32 pid; | |
775 | u32 tid; | |
776 | } tid_entry; | |
777 | u64 time; | |
5622f295 MM |
778 | u64 id; |
779 | u64 stream_id; | |
780 | struct { | |
781 | u32 cpu; | |
782 | u32 reserved; | |
783 | } cpu_entry; | |
5622f295 | 784 | struct perf_callchain_entry *callchain; |
88a7c26a AL |
785 | |
786 | /* | |
787 | * regs_user may point to task_pt_regs or to regs_user_copy, depending | |
788 | * on arch details. | |
789 | */ | |
60e2364e | 790 | struct perf_regs regs_user; |
88a7c26a AL |
791 | struct pt_regs regs_user_copy; |
792 | ||
60e2364e | 793 | struct perf_regs regs_intr; |
c5ebcedb | 794 | u64 stack_user_size; |
2565711f | 795 | } ____cacheline_aligned; |
df1a132b | 796 | |
770eee1f SE |
797 | /* default value for data source */ |
798 | #define PERF_MEM_NA (PERF_MEM_S(OP, NA) |\ | |
799 | PERF_MEM_S(LVL, NA) |\ | |
800 | PERF_MEM_S(SNOOP, NA) |\ | |
801 | PERF_MEM_S(LOCK, NA) |\ | |
802 | PERF_MEM_S(TLB, NA)) | |
803 | ||
fd0d000b RR |
804 | static inline void perf_sample_data_init(struct perf_sample_data *data, |
805 | u64 addr, u64 period) | |
dc1d628a | 806 | { |
fd0d000b | 807 | /* remaining struct members initialized in perf_prepare_sample() */ |
dc1d628a PZ |
808 | data->addr = addr; |
809 | data->raw = NULL; | |
bce38cd5 | 810 | data->br_stack = NULL; |
4018994f | 811 | data->period = period; |
c3feedf2 | 812 | data->weight = 0; |
770eee1f | 813 | data->data_src.val = PERF_MEM_NA; |
fdfbbd07 | 814 | data->txn = 0; |
dc1d628a PZ |
815 | } |
816 | ||
5622f295 MM |
817 | extern void perf_output_sample(struct perf_output_handle *handle, |
818 | struct perf_event_header *header, | |
819 | struct perf_sample_data *data, | |
cdd6c482 | 820 | struct perf_event *event); |
5622f295 MM |
821 | extern void perf_prepare_sample(struct perf_event_header *header, |
822 | struct perf_sample_data *data, | |
cdd6c482 | 823 | struct perf_event *event, |
5622f295 MM |
824 | struct pt_regs *regs); |
825 | ||
a8b0ca17 | 826 | extern int perf_event_overflow(struct perf_event *event, |
5622f295 MM |
827 | struct perf_sample_data *data, |
828 | struct pt_regs *regs); | |
df1a132b | 829 | |
21509084 YZ |
830 | extern void perf_event_output(struct perf_event *event, |
831 | struct perf_sample_data *data, | |
832 | struct pt_regs *regs); | |
833 | ||
834 | extern void | |
835 | perf_event_header__init_id(struct perf_event_header *header, | |
836 | struct perf_sample_data *data, | |
837 | struct perf_event *event); | |
838 | extern void | |
839 | perf_event__output_id_sample(struct perf_event *event, | |
840 | struct perf_output_handle *handle, | |
841 | struct perf_sample_data *sample); | |
842 | ||
f38b0dbb KL |
843 | extern void |
844 | perf_log_lost_samples(struct perf_event *event, u64 lost); | |
845 | ||
6c7e550f FBH |
846 | static inline bool is_sampling_event(struct perf_event *event) |
847 | { | |
848 | return event->attr.sample_period != 0; | |
849 | } | |
850 | ||
3b6f9e5c | 851 | /* |
cdd6c482 | 852 | * Return 1 for a software event, 0 for a hardware event |
3b6f9e5c | 853 | */ |
cdd6c482 | 854 | static inline int is_software_event(struct perf_event *event) |
3b6f9e5c | 855 | { |
89a1e187 | 856 | return event->pmu->task_ctx_nr == perf_sw_context; |
3b6f9e5c PM |
857 | } |
858 | ||
c5905afb | 859 | extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX]; |
f29ac756 | 860 | |
86038c5e | 861 | extern void ___perf_sw_event(u32, u64, struct pt_regs *, u64); |
a8b0ca17 | 862 | extern void __perf_sw_event(u32, u64, struct pt_regs *, u64); |
f29ac756 | 863 | |
b0f82b81 | 864 | #ifndef perf_arch_fetch_caller_regs |
e7e7ee2e | 865 | static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { } |
b0f82b81 | 866 | #endif |
5331d7b8 FW |
867 | |
868 | /* | |
869 | * Take a snapshot of the regs. Skip ip and frame pointer to | |
870 | * the nth caller. We only need a few of the regs: | |
871 | * - ip for PERF_SAMPLE_IP | |
872 | * - cs for user_mode() tests | |
873 | * - bp for callchains | |
874 | * - eflags, for future purposes, just in case | |
875 | */ | |
b0f82b81 | 876 | static inline void perf_fetch_caller_regs(struct pt_regs *regs) |
5331d7b8 | 877 | { |
5331d7b8 FW |
878 | memset(regs, 0, sizeof(*regs)); |
879 | ||
b0f82b81 | 880 | perf_arch_fetch_caller_regs(regs, CALLER_ADDR0); |
5331d7b8 FW |
881 | } |
882 | ||
7e54a5a0 | 883 | static __always_inline void |
a8b0ca17 | 884 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) |
e49a5bd3 | 885 | { |
86038c5e PZI |
886 | if (static_key_false(&perf_swevent_enabled[event_id])) |
887 | __perf_sw_event(event_id, nr, regs, addr); | |
888 | } | |
889 | ||
890 | DECLARE_PER_CPU(struct pt_regs, __perf_regs[4]); | |
7e54a5a0 | 891 | |
86038c5e PZI |
892 | /* |
893 | * 'Special' version for the scheduler, it hard assumes no recursion, | |
894 | * which is guaranteed by us not actually scheduling inside other swevents | |
895 | * because those disable preemption. | |
896 | */ | |
897 | static __always_inline void | |
898 | perf_sw_event_sched(u32 event_id, u64 nr, u64 addr) | |
899 | { | |
c5905afb | 900 | if (static_key_false(&perf_swevent_enabled[event_id])) { |
86038c5e PZI |
901 | struct pt_regs *regs = this_cpu_ptr(&__perf_regs[0]); |
902 | ||
903 | perf_fetch_caller_regs(regs); | |
904 | ___perf_sw_event(event_id, nr, regs, addr); | |
e49a5bd3 FW |
905 | } |
906 | } | |
907 | ||
c5905afb | 908 | extern struct static_key_deferred perf_sched_events; |
ee6dcfa4 | 909 | |
ff303e66 PZ |
910 | static __always_inline bool |
911 | perf_sw_migrate_enabled(void) | |
912 | { | |
913 | if (static_key_false(&perf_swevent_enabled[PERF_COUNT_SW_CPU_MIGRATIONS])) | |
914 | return true; | |
915 | return false; | |
916 | } | |
917 | ||
918 | static inline void perf_event_task_migrate(struct task_struct *task) | |
919 | { | |
920 | if (perf_sw_migrate_enabled()) | |
921 | task->sched_migrated = 1; | |
922 | } | |
923 | ||
ab0cce56 | 924 | static inline void perf_event_task_sched_in(struct task_struct *prev, |
a8d757ef | 925 | struct task_struct *task) |
ab0cce56 JO |
926 | { |
927 | if (static_key_false(&perf_sched_events.key)) | |
928 | __perf_event_task_sched_in(prev, task); | |
ff303e66 PZ |
929 | |
930 | if (perf_sw_migrate_enabled() && task->sched_migrated) { | |
931 | struct pt_regs *regs = this_cpu_ptr(&__perf_regs[0]); | |
932 | ||
933 | perf_fetch_caller_regs(regs); | |
934 | ___perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, regs, 0); | |
935 | task->sched_migrated = 0; | |
936 | } | |
ab0cce56 JO |
937 | } |
938 | ||
939 | static inline void perf_event_task_sched_out(struct task_struct *prev, | |
940 | struct task_struct *next) | |
ee6dcfa4 | 941 | { |
86038c5e | 942 | perf_sw_event_sched(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 0); |
ee6dcfa4 | 943 | |
c5905afb | 944 | if (static_key_false(&perf_sched_events.key)) |
ab0cce56 | 945 | __perf_event_task_sched_out(prev, next); |
ee6dcfa4 PZ |
946 | } |
947 | ||
eacd3ecc MF |
948 | static inline u64 __perf_event_count(struct perf_event *event) |
949 | { | |
950 | return local64_read(&event->count) + atomic64_read(&event->child_count); | |
951 | } | |
952 | ||
3af9e859 | 953 | extern void perf_event_mmap(struct vm_area_struct *vma); |
39447b38 | 954 | extern struct perf_guest_info_callbacks *perf_guest_cbs; |
dcf46b94 ZY |
955 | extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); |
956 | extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); | |
39447b38 | 957 | |
e041e328 | 958 | extern void perf_event_exec(void); |
82b89778 | 959 | extern void perf_event_comm(struct task_struct *tsk, bool exec); |
cdd6c482 | 960 | extern void perf_event_fork(struct task_struct *tsk); |
8d1b2d93 | 961 | |
56962b44 FW |
962 | /* Callchains */ |
963 | DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry); | |
964 | ||
e7e7ee2e IM |
965 | extern void perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs); |
966 | extern void perf_callchain_kernel(struct perf_callchain_entry *entry, struct pt_regs *regs); | |
394ee076 | 967 | |
e7e7ee2e | 968 | static inline void perf_callchain_store(struct perf_callchain_entry *entry, u64 ip) |
70791ce9 FW |
969 | { |
970 | if (entry->nr < PERF_MAX_STACK_DEPTH) | |
971 | entry->ip[entry->nr++] = ip; | |
972 | } | |
394ee076 | 973 | |
cdd6c482 IM |
974 | extern int sysctl_perf_event_paranoid; |
975 | extern int sysctl_perf_event_mlock; | |
976 | extern int sysctl_perf_event_sample_rate; | |
14c63f17 DH |
977 | extern int sysctl_perf_cpu_time_max_percent; |
978 | ||
979 | extern void perf_sample_event_took(u64 sample_len_ns); | |
1ccd1549 | 980 | |
163ec435 PZ |
981 | extern int perf_proc_update_handler(struct ctl_table *table, int write, |
982 | void __user *buffer, size_t *lenp, | |
983 | loff_t *ppos); | |
14c63f17 DH |
984 | extern int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write, |
985 | void __user *buffer, size_t *lenp, | |
986 | loff_t *ppos); | |
987 | ||
163ec435 | 988 | |
320ebf09 PZ |
989 | static inline bool perf_paranoid_tracepoint_raw(void) |
990 | { | |
991 | return sysctl_perf_event_paranoid > -1; | |
992 | } | |
993 | ||
994 | static inline bool perf_paranoid_cpu(void) | |
995 | { | |
996 | return sysctl_perf_event_paranoid > 0; | |
997 | } | |
998 | ||
999 | static inline bool perf_paranoid_kernel(void) | |
1000 | { | |
1001 | return sysctl_perf_event_paranoid > 1; | |
1002 | } | |
1003 | ||
cdd6c482 | 1004 | extern void perf_event_init(void); |
1c024eca PZ |
1005 | extern void perf_tp_event(u64 addr, u64 count, void *record, |
1006 | int entry_size, struct pt_regs *regs, | |
e6dab5ff AV |
1007 | struct hlist_head *head, int rctx, |
1008 | struct task_struct *task); | |
24f1e32c | 1009 | extern void perf_bp_event(struct perf_event *event, void *data); |
0d905bca | 1010 | |
9d23a90a | 1011 | #ifndef perf_misc_flags |
e7e7ee2e IM |
1012 | # define perf_misc_flags(regs) \ |
1013 | (user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL) | |
1014 | # define perf_instruction_pointer(regs) instruction_pointer(regs) | |
9d23a90a PM |
1015 | #endif |
1016 | ||
bce38cd5 SE |
1017 | static inline bool has_branch_stack(struct perf_event *event) |
1018 | { | |
1019 | return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK; | |
a46a2300 YZ |
1020 | } |
1021 | ||
1022 | static inline bool needs_branch_stack(struct perf_event *event) | |
1023 | { | |
1024 | return event->attr.branch_sample_type != 0; | |
bce38cd5 SE |
1025 | } |
1026 | ||
45bfb2e5 PZ |
1027 | static inline bool has_aux(struct perf_event *event) |
1028 | { | |
1029 | return event->pmu->setup_aux; | |
1030 | } | |
1031 | ||
5622f295 | 1032 | extern int perf_output_begin(struct perf_output_handle *handle, |
a7ac67ea | 1033 | struct perf_event *event, unsigned int size); |
5622f295 | 1034 | extern void perf_output_end(struct perf_output_handle *handle); |
91d7753a | 1035 | extern unsigned int perf_output_copy(struct perf_output_handle *handle, |
5622f295 | 1036 | const void *buf, unsigned int len); |
5685e0ff JO |
1037 | extern unsigned int perf_output_skip(struct perf_output_handle *handle, |
1038 | unsigned int len); | |
4ed7c92d PZ |
1039 | extern int perf_swevent_get_recursion_context(void); |
1040 | extern void perf_swevent_put_recursion_context(int rctx); | |
ab573844 | 1041 | extern u64 perf_swevent_set_period(struct perf_event *event); |
44234adc FW |
1042 | extern void perf_event_enable(struct perf_event *event); |
1043 | extern void perf_event_disable(struct perf_event *event); | |
fae3fde6 | 1044 | extern void perf_event_disable_local(struct perf_event *event); |
e9d2b064 | 1045 | extern void perf_event_task_tick(void); |
e041e328 | 1046 | #else /* !CONFIG_PERF_EVENTS: */ |
fdc26706 AS |
1047 | static inline void * |
1048 | perf_aux_output_begin(struct perf_output_handle *handle, | |
1049 | struct perf_event *event) { return NULL; } | |
1050 | static inline void | |
1051 | perf_aux_output_end(struct perf_output_handle *handle, unsigned long size, | |
1052 | bool truncated) { } | |
1053 | static inline int | |
1054 | perf_aux_output_skip(struct perf_output_handle *handle, | |
1055 | unsigned long size) { return -EINVAL; } | |
1056 | static inline void * | |
1057 | perf_get_aux(struct perf_output_handle *handle) { return NULL; } | |
0793a61d | 1058 | static inline void |
ff303e66 PZ |
1059 | perf_event_task_migrate(struct task_struct *task) { } |
1060 | static inline void | |
ab0cce56 JO |
1061 | perf_event_task_sched_in(struct task_struct *prev, |
1062 | struct task_struct *task) { } | |
1063 | static inline void | |
1064 | perf_event_task_sched_out(struct task_struct *prev, | |
1065 | struct task_struct *next) { } | |
cdd6c482 IM |
1066 | static inline int perf_event_init_task(struct task_struct *child) { return 0; } |
1067 | static inline void perf_event_exit_task(struct task_struct *child) { } | |
1068 | static inline void perf_event_free_task(struct task_struct *task) { } | |
4e231c79 | 1069 | static inline void perf_event_delayed_put(struct task_struct *task) { } |
e03e7ee3 | 1070 | static inline struct file *perf_event_get(unsigned int fd) { return ERR_PTR(-EINVAL); } |
ffe8690c KX |
1071 | static inline const struct perf_event_attr *perf_event_attrs(struct perf_event *event) |
1072 | { | |
1073 | return ERR_PTR(-EINVAL); | |
1074 | } | |
1075 | static inline u64 perf_event_read_local(struct perf_event *event) { return -EINVAL; } | |
57c0c15b | 1076 | static inline void perf_event_print_debug(void) { } |
57c0c15b IM |
1077 | static inline int perf_event_task_disable(void) { return -EINVAL; } |
1078 | static inline int perf_event_task_enable(void) { return -EINVAL; } | |
26ca5c11 AK |
1079 | static inline int perf_event_refresh(struct perf_event *event, int refresh) |
1080 | { | |
1081 | return -EINVAL; | |
1082 | } | |
15dbf27c | 1083 | |
925d519a | 1084 | static inline void |
a8b0ca17 | 1085 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) { } |
24f1e32c | 1086 | static inline void |
86038c5e PZI |
1087 | perf_sw_event_sched(u32 event_id, u64 nr, u64 addr) { } |
1088 | static inline void | |
184f412c | 1089 | perf_bp_event(struct perf_event *event, void *data) { } |
0a4a9391 | 1090 | |
39447b38 | 1091 | static inline int perf_register_guest_info_callbacks |
e7e7ee2e | 1092 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 1093 | static inline int perf_unregister_guest_info_callbacks |
e7e7ee2e | 1094 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 1095 | |
57c0c15b | 1096 | static inline void perf_event_mmap(struct vm_area_struct *vma) { } |
e041e328 | 1097 | static inline void perf_event_exec(void) { } |
82b89778 | 1098 | static inline void perf_event_comm(struct task_struct *tsk, bool exec) { } |
cdd6c482 IM |
1099 | static inline void perf_event_fork(struct task_struct *tsk) { } |
1100 | static inline void perf_event_init(void) { } | |
184f412c | 1101 | static inline int perf_swevent_get_recursion_context(void) { return -1; } |
4ed7c92d | 1102 | static inline void perf_swevent_put_recursion_context(int rctx) { } |
ab573844 | 1103 | static inline u64 perf_swevent_set_period(struct perf_event *event) { return 0; } |
44234adc FW |
1104 | static inline void perf_event_enable(struct perf_event *event) { } |
1105 | static inline void perf_event_disable(struct perf_event *event) { } | |
500ad2d8 | 1106 | static inline int __perf_event_disable(void *info) { return -1; } |
e9d2b064 | 1107 | static inline void perf_event_task_tick(void) { } |
ffe8690c | 1108 | static inline int perf_event_release_kernel(struct perf_event *event) { return 0; } |
0793a61d TG |
1109 | #endif |
1110 | ||
026249ef FW |
1111 | #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_NO_HZ_FULL) |
1112 | extern bool perf_event_can_stop_tick(void); | |
1113 | #else | |
1114 | static inline bool perf_event_can_stop_tick(void) { return true; } | |
1115 | #endif | |
1116 | ||
6c4d3bc9 DR |
1117 | #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL) |
1118 | extern void perf_restore_debug_store(void); | |
1119 | #else | |
1d9d8639 | 1120 | static inline void perf_restore_debug_store(void) { } |
0793a61d TG |
1121 | #endif |
1122 | ||
e7e7ee2e | 1123 | #define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x)) |
5622f295 | 1124 | |
3f6da390 | 1125 | /* |
0a0fca9d | 1126 | * This has to have a higher priority than migration_notifier in sched/core.c. |
3f6da390 | 1127 | */ |
e7e7ee2e IM |
1128 | #define perf_cpu_notifier(fn) \ |
1129 | do { \ | |
0db0628d | 1130 | static struct notifier_block fn##_nb = \ |
e7e7ee2e | 1131 | { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ |
c13d38e4 | 1132 | unsigned long cpu = smp_processor_id(); \ |
6760bca9 | 1133 | unsigned long flags; \ |
f0bdb5e0 SB |
1134 | \ |
1135 | cpu_notifier_register_begin(); \ | |
e7e7ee2e | 1136 | fn(&fn##_nb, (unsigned long)CPU_UP_PREPARE, \ |
c13d38e4 | 1137 | (void *)(unsigned long)cpu); \ |
6760bca9 | 1138 | local_irq_save(flags); \ |
e7e7ee2e | 1139 | fn(&fn##_nb, (unsigned long)CPU_STARTING, \ |
c13d38e4 | 1140 | (void *)(unsigned long)cpu); \ |
6760bca9 | 1141 | local_irq_restore(flags); \ |
e7e7ee2e | 1142 | fn(&fn##_nb, (unsigned long)CPU_ONLINE, \ |
c13d38e4 | 1143 | (void *)(unsigned long)cpu); \ |
f0bdb5e0 SB |
1144 | __register_cpu_notifier(&fn##_nb); \ |
1145 | cpu_notifier_register_done(); \ | |
3f6da390 PZ |
1146 | } while (0) |
1147 | ||
f0bdb5e0 SB |
1148 | /* |
1149 | * Bare-bones version of perf_cpu_notifier(), which doesn't invoke the | |
1150 | * callback for already online CPUs. | |
1151 | */ | |
1152 | #define __perf_cpu_notifier(fn) \ | |
1153 | do { \ | |
1154 | static struct notifier_block fn##_nb = \ | |
1155 | { .notifier_call = fn, .priority = CPU_PRI_PERF }; \ | |
1156 | \ | |
1157 | __register_cpu_notifier(&fn##_nb); \ | |
1158 | } while (0) | |
641cc938 | 1159 | |
2663960c SB |
1160 | struct perf_pmu_events_attr { |
1161 | struct device_attribute attr; | |
1162 | u64 id; | |
3a54aaa0 | 1163 | const char *event_str; |
2663960c SB |
1164 | }; |
1165 | ||
fd979c01 CS |
1166 | ssize_t perf_event_sysfs_show(struct device *dev, struct device_attribute *attr, |
1167 | char *page); | |
1168 | ||
2663960c SB |
1169 | #define PMU_EVENT_ATTR(_name, _var, _id, _show) \ |
1170 | static struct perf_pmu_events_attr _var = { \ | |
1171 | .attr = __ATTR(_name, 0444, _show, NULL), \ | |
1172 | .id = _id, \ | |
1173 | }; | |
1174 | ||
f0405b81 CS |
1175 | #define PMU_EVENT_ATTR_STRING(_name, _var, _str) \ |
1176 | static struct perf_pmu_events_attr _var = { \ | |
1177 | .attr = __ATTR(_name, 0444, perf_event_sysfs_show, NULL), \ | |
1178 | .id = 0, \ | |
1179 | .event_str = _str, \ | |
1180 | }; | |
1181 | ||
641cc938 JO |
1182 | #define PMU_FORMAT_ATTR(_name, _format) \ |
1183 | static ssize_t \ | |
1184 | _name##_show(struct device *dev, \ | |
1185 | struct device_attribute *attr, \ | |
1186 | char *page) \ | |
1187 | { \ | |
1188 | BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE); \ | |
1189 | return sprintf(page, _format "\n"); \ | |
1190 | } \ | |
1191 | \ | |
1192 | static struct device_attribute format_attr_##_name = __ATTR_RO(_name) | |
1193 | ||
cdd6c482 | 1194 | #endif /* _LINUX_PERF_EVENT_H */ |