]>
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; | |
c5dfd78e | 61 | __u64 ip[0]; /* /proc/sys/kernel/perf_event_max_stack */ |
f9188e02 PZ |
62 | }; |
63 | ||
cfbcf468 ACM |
64 | struct perf_callchain_entry_ctx { |
65 | struct perf_callchain_entry *entry; | |
66 | u32 max_stack; | |
3b1fff08 | 67 | u32 nr; |
c85b0334 ACM |
68 | short contexts; |
69 | bool contexts_maxed; | |
cfbcf468 ACM |
70 | }; |
71 | ||
7e3f977e | 72 | typedef unsigned long (*perf_copy_f)(void *dst, const void *src, |
aa7145c1 | 73 | unsigned long off, unsigned long len); |
7e3f977e DB |
74 | |
75 | struct perf_raw_frag { | |
76 | union { | |
77 | struct perf_raw_frag *next; | |
78 | unsigned long pad; | |
79 | }; | |
80 | perf_copy_f copy; | |
81 | void *data; | |
82 | u32 size; | |
83 | } __packed; | |
84 | ||
3a43ce68 | 85 | struct perf_raw_record { |
7e3f977e | 86 | struct perf_raw_frag frag; |
3a43ce68 | 87 | u32 size; |
f413cdb8 FW |
88 | }; |
89 | ||
bce38cd5 SE |
90 | /* |
91 | * branch stack layout: | |
92 | * nr: number of taken branches stored in entries[] | |
93 | * | |
94 | * Note that nr can vary from sample to sample | |
95 | * branches (to, from) are stored from most recent | |
96 | * to least recent, i.e., entries[0] contains the most | |
97 | * recent branch. | |
98 | */ | |
caff2bef PZ |
99 | struct perf_branch_stack { |
100 | __u64 nr; | |
101 | struct perf_branch_entry entries[0]; | |
102 | }; | |
103 | ||
f3dfd265 PM |
104 | struct task_struct; |
105 | ||
efc9f05d SE |
106 | /* |
107 | * extra PMU register associated with an event | |
108 | */ | |
109 | struct hw_perf_event_extra { | |
110 | u64 config; /* register value */ | |
111 | unsigned int reg; /* register address or index */ | |
112 | int alloc; /* extra register already allocated */ | |
113 | int idx; /* index in shared_regs->regs[] */ | |
114 | }; | |
115 | ||
0793a61d | 116 | /** |
cdd6c482 | 117 | * struct hw_perf_event - performance event hardware details: |
0793a61d | 118 | */ |
cdd6c482 IM |
119 | struct hw_perf_event { |
120 | #ifdef CONFIG_PERF_EVENTS | |
d6d020e9 PZ |
121 | union { |
122 | struct { /* hardware */ | |
a308444c | 123 | u64 config; |
447a194b | 124 | u64 last_tag; |
a308444c | 125 | unsigned long config_base; |
cdd6c482 | 126 | unsigned long event_base; |
c48b6053 | 127 | int event_base_rdpmc; |
a308444c | 128 | int idx; |
447a194b | 129 | int last_cpu; |
9fac2cf3 | 130 | int flags; |
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 | 138 | struct { /* tracepoint */ |
f22c1bb6 ON |
139 | /* for tp_event->class */ |
140 | struct list_head tp_list; | |
141 | }; | |
4afbb24c MF |
142 | struct { /* intel_cqm */ |
143 | int cqm_state; | |
b3df4ec4 | 144 | u32 cqm_rmid; |
a223c1c7 | 145 | int is_group_event; |
4afbb24c MF |
146 | struct list_head cqm_events_entry; |
147 | struct list_head cqm_groups_entry; | |
148 | struct list_head cqm_group_entry; | |
149 | }; | |
ec0d7729 AS |
150 | struct { /* itrace */ |
151 | int itrace_started; | |
152 | }; | |
c7ab62bf HR |
153 | struct { /* amd_power */ |
154 | u64 pwr_acc; | |
155 | u64 ptsc; | |
156 | }; | |
24f1e32c | 157 | #ifdef CONFIG_HAVE_HW_BREAKPOINT |
45a73372 | 158 | struct { /* breakpoint */ |
d580ff86 PZ |
159 | /* |
160 | * Crufty hack to avoid the chicken and egg | |
161 | * problem hw_breakpoint has with context | |
162 | * creation and event initalization. | |
163 | */ | |
f22c1bb6 ON |
164 | struct arch_hw_breakpoint info; |
165 | struct list_head bp_list; | |
45a73372 | 166 | }; |
24f1e32c | 167 | #endif |
d6d020e9 | 168 | }; |
b0e87875 PZ |
169 | /* |
170 | * If the event is a per task event, this will point to the task in | |
171 | * question. See the comment in perf_event_alloc(). | |
172 | */ | |
50f16a8b | 173 | struct task_struct *target; |
b0e87875 | 174 | |
375637bc AS |
175 | /* |
176 | * PMU would store hardware filter configuration | |
177 | * here. | |
178 | */ | |
179 | void *addr_filters; | |
180 | ||
181 | /* Last sync'ed generation of filters */ | |
182 | unsigned long addr_filters_gen; | |
183 | ||
b0e87875 PZ |
184 | /* |
185 | * hw_perf_event::state flags; used to track the PERF_EF_* state. | |
186 | */ | |
187 | #define PERF_HES_STOPPED 0x01 /* the counter is stopped */ | |
188 | #define PERF_HES_UPTODATE 0x02 /* event->count up-to-date */ | |
189 | #define PERF_HES_ARCH 0x04 | |
190 | ||
a4eaf7f1 | 191 | int state; |
b0e87875 PZ |
192 | |
193 | /* | |
194 | * The last observed hardware counter value, updated with a | |
195 | * local64_cmpxchg() such that pmu::read() can be called nested. | |
196 | */ | |
e7850595 | 197 | local64_t prev_count; |
b0e87875 PZ |
198 | |
199 | /* | |
200 | * The period to start the next sample with. | |
201 | */ | |
b23f3325 | 202 | u64 sample_period; |
b0e87875 PZ |
203 | |
204 | /* | |
205 | * The period we started this sample with. | |
206 | */ | |
9e350de3 | 207 | u64 last_period; |
b0e87875 PZ |
208 | |
209 | /* | |
210 | * However much is left of the current period; note that this is | |
211 | * a full 64bit value and allows for generation of periods longer | |
212 | * than hardware might allow. | |
213 | */ | |
e7850595 | 214 | local64_t period_left; |
b0e87875 PZ |
215 | |
216 | /* | |
217 | * State for throttling the event, see __perf_event_overflow() and | |
218 | * perf_adjust_freq_unthr_context(). | |
219 | */ | |
e050e3f0 | 220 | u64 interrupts_seq; |
60db5e09 | 221 | u64 interrupts; |
6a24ed6c | 222 | |
b0e87875 PZ |
223 | /* |
224 | * State for freq target events, see __perf_event_overflow() and | |
225 | * perf_adjust_freq_unthr_context(). | |
226 | */ | |
abd50713 PZ |
227 | u64 freq_time_stamp; |
228 | u64 freq_count_stamp; | |
ee06094f | 229 | #endif |
0793a61d TG |
230 | }; |
231 | ||
cdd6c482 | 232 | struct perf_event; |
621a01ea | 233 | |
8d2cacbb PZ |
234 | /* |
235 | * Common implementation detail of pmu::{start,commit,cancel}_txn | |
236 | */ | |
fbbe0701 | 237 | #define PERF_PMU_TXN_ADD 0x1 /* txn to add/schedule event on PMU */ |
4a00c16e | 238 | #define PERF_PMU_TXN_READ 0x2 /* txn to read event group from PMU */ |
fbbe0701 | 239 | |
53b25335 VW |
240 | /** |
241 | * pmu::capabilities flags | |
242 | */ | |
243 | #define PERF_PMU_CAP_NO_INTERRUPT 0x01 | |
34f43927 | 244 | #define PERF_PMU_CAP_NO_NMI 0x02 |
0a4e38e6 | 245 | #define PERF_PMU_CAP_AUX_NO_SG 0x04 |
6a279230 | 246 | #define PERF_PMU_CAP_AUX_SW_DOUBLEBUF 0x08 |
bed5b25a | 247 | #define PERF_PMU_CAP_EXCLUSIVE 0x10 |
ec0d7729 | 248 | #define PERF_PMU_CAP_ITRACE 0x20 |
5101ef20 | 249 | #define PERF_PMU_CAP_HETEROGENEOUS_CPUS 0x40 |
53b25335 | 250 | |
621a01ea | 251 | /** |
4aeb0b42 | 252 | * struct pmu - generic performance monitoring unit |
621a01ea | 253 | */ |
4aeb0b42 | 254 | struct pmu { |
b0a873eb PZ |
255 | struct list_head entry; |
256 | ||
c464c76e | 257 | struct module *module; |
abe43400 | 258 | struct device *dev; |
0c9d42ed | 259 | const struct attribute_group **attr_groups; |
03d8e80b | 260 | const char *name; |
2e80a82a PZ |
261 | int type; |
262 | ||
53b25335 VW |
263 | /* |
264 | * various common per-pmu feature flags | |
265 | */ | |
266 | int capabilities; | |
267 | ||
108b02cf PZ |
268 | int * __percpu pmu_disable_count; |
269 | struct perf_cpu_context * __percpu pmu_cpu_context; | |
bed5b25a | 270 | atomic_t exclusive_cnt; /* < 0: cpu; > 0: tsk */ |
8dc85d54 | 271 | int task_ctx_nr; |
62b85639 | 272 | int hrtimer_interval_ms; |
6bde9b6c | 273 | |
375637bc AS |
274 | /* number of address filters this PMU can do */ |
275 | unsigned int nr_addr_filters; | |
276 | ||
6bde9b6c | 277 | /* |
a4eaf7f1 PZ |
278 | * Fully disable/enable this PMU, can be used to protect from the PMI |
279 | * as well as for lazy/batch writing of the MSRs. | |
6bde9b6c | 280 | */ |
ad5133b7 PZ |
281 | void (*pmu_enable) (struct pmu *pmu); /* optional */ |
282 | void (*pmu_disable) (struct pmu *pmu); /* optional */ | |
6bde9b6c | 283 | |
8d2cacbb | 284 | /* |
a4eaf7f1 | 285 | * Try and initialize the event for this PMU. |
b0e87875 PZ |
286 | * |
287 | * Returns: | |
288 | * -ENOENT -- @event is not for this PMU | |
289 | * | |
290 | * -ENODEV -- @event is for this PMU but PMU not present | |
291 | * -EBUSY -- @event is for this PMU but PMU temporarily unavailable | |
292 | * -EINVAL -- @event is for this PMU but @event is not valid | |
293 | * -EOPNOTSUPP -- @event is for this PMU, @event is valid, but not supported | |
294 | * -EACCESS -- @event is for this PMU, @event is valid, but no privilidges | |
295 | * | |
296 | * 0 -- @event is for this PMU and valid | |
297 | * | |
298 | * Other error return values are allowed. | |
8d2cacbb | 299 | */ |
b0a873eb PZ |
300 | int (*event_init) (struct perf_event *event); |
301 | ||
1e0fb9ec AL |
302 | /* |
303 | * Notification that the event was mapped or unmapped. Called | |
304 | * in the context of the mapping task. | |
305 | */ | |
306 | void (*event_mapped) (struct perf_event *event); /*optional*/ | |
307 | void (*event_unmapped) (struct perf_event *event); /*optional*/ | |
308 | ||
b0e87875 PZ |
309 | /* |
310 | * Flags for ->add()/->del()/ ->start()/->stop(). There are | |
311 | * matching hw_perf_event::state flags. | |
312 | */ | |
a4eaf7f1 PZ |
313 | #define PERF_EF_START 0x01 /* start the counter when adding */ |
314 | #define PERF_EF_RELOAD 0x02 /* reload the counter when starting */ | |
315 | #define PERF_EF_UPDATE 0x04 /* update the counter when stopping */ | |
316 | ||
8d2cacbb | 317 | /* |
b0e87875 PZ |
318 | * Adds/Removes a counter to/from the PMU, can be done inside a |
319 | * transaction, see the ->*_txn() methods. | |
320 | * | |
321 | * The add/del callbacks will reserve all hardware resources required | |
322 | * to service the event, this includes any counter constraint | |
323 | * scheduling etc. | |
324 | * | |
325 | * Called with IRQs disabled and the PMU disabled on the CPU the event | |
326 | * is on. | |
327 | * | |
328 | * ->add() called without PERF_EF_START should result in the same state | |
329 | * as ->add() followed by ->stop(). | |
330 | * | |
331 | * ->del() must always PERF_EF_UPDATE stop an event. If it calls | |
332 | * ->stop() that must deal with already being stopped without | |
333 | * PERF_EF_UPDATE. | |
a4eaf7f1 PZ |
334 | */ |
335 | int (*add) (struct perf_event *event, int flags); | |
336 | void (*del) (struct perf_event *event, int flags); | |
337 | ||
338 | /* | |
b0e87875 PZ |
339 | * Starts/Stops a counter present on the PMU. |
340 | * | |
341 | * The PMI handler should stop the counter when perf_event_overflow() | |
342 | * returns !0. ->start() will be used to continue. | |
343 | * | |
344 | * Also used to change the sample period. | |
345 | * | |
346 | * Called with IRQs disabled and the PMU disabled on the CPU the event | |
347 | * is on -- will be called from NMI context with the PMU generates | |
348 | * NMIs. | |
349 | * | |
350 | * ->stop() with PERF_EF_UPDATE will read the counter and update | |
351 | * period/count values like ->read() would. | |
352 | * | |
353 | * ->start() with PERF_EF_RELOAD will reprogram the the counter | |
354 | * value, must be preceded by a ->stop() with PERF_EF_UPDATE. | |
a4eaf7f1 PZ |
355 | */ |
356 | void (*start) (struct perf_event *event, int flags); | |
357 | void (*stop) (struct perf_event *event, int flags); | |
358 | ||
359 | /* | |
360 | * Updates the counter value of the event. | |
b0e87875 PZ |
361 | * |
362 | * For sampling capable PMUs this will also update the software period | |
363 | * hw_perf_event::period_left field. | |
a4eaf7f1 | 364 | */ |
cdd6c482 | 365 | void (*read) (struct perf_event *event); |
6bde9b6c LM |
366 | |
367 | /* | |
24cd7f54 PZ |
368 | * Group events scheduling is treated as a transaction, add |
369 | * group events as a whole and perform one schedulability test. | |
370 | * If the test fails, roll back the whole group | |
a4eaf7f1 PZ |
371 | * |
372 | * Start the transaction, after this ->add() doesn't need to | |
24cd7f54 | 373 | * do schedulability tests. |
fbbe0701 SB |
374 | * |
375 | * Optional. | |
8d2cacbb | 376 | */ |
fbbe0701 | 377 | void (*start_txn) (struct pmu *pmu, unsigned int txn_flags); |
8d2cacbb | 378 | /* |
a4eaf7f1 | 379 | * If ->start_txn() disabled the ->add() schedulability test |
8d2cacbb PZ |
380 | * then ->commit_txn() is required to perform one. On success |
381 | * the transaction is closed. On error the transaction is kept | |
382 | * open until ->cancel_txn() is called. | |
fbbe0701 SB |
383 | * |
384 | * Optional. | |
8d2cacbb | 385 | */ |
fbbe0701 | 386 | int (*commit_txn) (struct pmu *pmu); |
8d2cacbb | 387 | /* |
a4eaf7f1 | 388 | * Will cancel the transaction, assumes ->del() is called |
25985edc | 389 | * for each successful ->add() during the transaction. |
fbbe0701 SB |
390 | * |
391 | * Optional. | |
8d2cacbb | 392 | */ |
fbbe0701 | 393 | void (*cancel_txn) (struct pmu *pmu); |
35edc2a5 PZ |
394 | |
395 | /* | |
396 | * Will return the value for perf_event_mmap_page::index for this event, | |
397 | * if no implementation is provided it will default to: event->hw.idx + 1. | |
398 | */ | |
399 | int (*event_idx) (struct perf_event *event); /*optional */ | |
d010b332 | 400 | |
ba532500 YZ |
401 | /* |
402 | * context-switches callback | |
403 | */ | |
404 | void (*sched_task) (struct perf_event_context *ctx, | |
405 | bool sched_in); | |
4af57ef2 YZ |
406 | /* |
407 | * PMU specific data size | |
408 | */ | |
409 | size_t task_ctx_size; | |
ba532500 | 410 | |
eacd3ecc MF |
411 | |
412 | /* | |
413 | * Return the count value for a counter. | |
414 | */ | |
415 | u64 (*count) (struct perf_event *event); /*optional*/ | |
45bfb2e5 PZ |
416 | |
417 | /* | |
418 | * Set up pmu-private data structures for an AUX area | |
419 | */ | |
420 | void *(*setup_aux) (int cpu, void **pages, | |
421 | int nr_pages, bool overwrite); | |
422 | /* optional */ | |
423 | ||
424 | /* | |
425 | * Free pmu-private AUX data structures | |
426 | */ | |
427 | void (*free_aux) (void *aux); /* optional */ | |
66eb579e | 428 | |
375637bc AS |
429 | /* |
430 | * Validate address range filters: make sure the HW supports the | |
431 | * requested configuration and number of filters; return 0 if the | |
432 | * supplied filters are valid, -errno otherwise. | |
433 | * | |
434 | * Runs in the context of the ioctl()ing process and is not serialized | |
435 | * with the rest of the PMU callbacks. | |
436 | */ | |
437 | int (*addr_filters_validate) (struct list_head *filters); | |
438 | /* optional */ | |
439 | ||
440 | /* | |
441 | * Synchronize address range filter configuration: | |
442 | * translate hw-agnostic filters into hardware configuration in | |
443 | * event::hw::addr_filters. | |
444 | * | |
445 | * Runs as a part of filter sync sequence that is done in ->start() | |
446 | * callback by calling perf_event_addr_filters_sync(). | |
447 | * | |
448 | * May (and should) traverse event::addr_filters::list, for which its | |
449 | * caller provides necessary serialization. | |
450 | */ | |
451 | void (*addr_filters_sync) (struct perf_event *event); | |
452 | /* optional */ | |
453 | ||
66eb579e MR |
454 | /* |
455 | * Filter events for PMU-specific reasons. | |
456 | */ | |
457 | int (*filter_match) (struct perf_event *event); /* optional */ | |
621a01ea IM |
458 | }; |
459 | ||
375637bc AS |
460 | /** |
461 | * struct perf_addr_filter - address range filter definition | |
462 | * @entry: event's filter list linkage | |
463 | * @inode: object file's inode for file-based filters | |
464 | * @offset: filter range offset | |
465 | * @size: filter range size | |
466 | * @range: 1: range, 0: address | |
467 | * @filter: 1: filter/start, 0: stop | |
468 | * | |
469 | * This is a hardware-agnostic filter configuration as specified by the user. | |
470 | */ | |
471 | struct perf_addr_filter { | |
472 | struct list_head entry; | |
473 | struct inode *inode; | |
474 | unsigned long offset; | |
475 | unsigned long size; | |
476 | unsigned int range : 1, | |
477 | filter : 1; | |
478 | }; | |
479 | ||
480 | /** | |
481 | * struct perf_addr_filters_head - container for address range filters | |
482 | * @list: list of filters for this event | |
483 | * @lock: spinlock that serializes accesses to the @list and event's | |
484 | * (and its children's) filter generations. | |
6ce77bfd | 485 | * @nr_file_filters: number of file-based filters |
375637bc AS |
486 | * |
487 | * A child event will use parent's @list (and therefore @lock), so they are | |
488 | * bundled together; see perf_event_addr_filters(). | |
489 | */ | |
490 | struct perf_addr_filters_head { | |
491 | struct list_head list; | |
492 | raw_spinlock_t lock; | |
6ce77bfd | 493 | unsigned int nr_file_filters; |
375637bc AS |
494 | }; |
495 | ||
6a930700 | 496 | /** |
cdd6c482 | 497 | * enum perf_event_active_state - the states of a event |
6a930700 | 498 | */ |
cdd6c482 | 499 | enum perf_event_active_state { |
a69b0ca4 | 500 | PERF_EVENT_STATE_DEAD = -4, |
179033b3 | 501 | PERF_EVENT_STATE_EXIT = -3, |
57c0c15b | 502 | PERF_EVENT_STATE_ERROR = -2, |
cdd6c482 IM |
503 | PERF_EVENT_STATE_OFF = -1, |
504 | PERF_EVENT_STATE_INACTIVE = 0, | |
57c0c15b | 505 | PERF_EVENT_STATE_ACTIVE = 1, |
6a930700 IM |
506 | }; |
507 | ||
9b51f66d | 508 | struct file; |
453f19ee PZ |
509 | struct perf_sample_data; |
510 | ||
a8b0ca17 | 511 | typedef void (*perf_overflow_handler_t)(struct perf_event *, |
b326e956 FW |
512 | struct perf_sample_data *, |
513 | struct pt_regs *regs); | |
514 | ||
4ff6a8de DCC |
515 | /* |
516 | * Event capabilities. For event_caps and groups caps. | |
517 | * | |
518 | * PERF_EV_CAP_SOFTWARE: Is a software event. | |
d6a2f903 DCC |
519 | * PERF_EV_CAP_READ_ACTIVE_PKG: A CPU event (or cgroup event) that can be read |
520 | * from any CPU in the package where it is active. | |
4ff6a8de DCC |
521 | */ |
522 | #define PERF_EV_CAP_SOFTWARE BIT(0) | |
d6a2f903 | 523 | #define PERF_EV_CAP_READ_ACTIVE_PKG BIT(1) |
d6f962b5 | 524 | |
e7e7ee2e IM |
525 | #define SWEVENT_HLIST_BITS 8 |
526 | #define SWEVENT_HLIST_SIZE (1 << SWEVENT_HLIST_BITS) | |
76e1d904 FW |
527 | |
528 | struct swevent_hlist { | |
e7e7ee2e IM |
529 | struct hlist_head heads[SWEVENT_HLIST_SIZE]; |
530 | struct rcu_head rcu_head; | |
76e1d904 FW |
531 | }; |
532 | ||
8a49542c PZ |
533 | #define PERF_ATTACH_CONTEXT 0x01 |
534 | #define PERF_ATTACH_GROUP 0x02 | |
d580ff86 | 535 | #define PERF_ATTACH_TASK 0x04 |
4af57ef2 | 536 | #define PERF_ATTACH_TASK_DATA 0x08 |
8a49542c | 537 | |
877c6856 | 538 | struct perf_cgroup; |
76369139 FW |
539 | struct ring_buffer; |
540 | ||
f2fb6bef KL |
541 | struct pmu_event_list { |
542 | raw_spinlock_t lock; | |
543 | struct list_head list; | |
544 | }; | |
545 | ||
0793a61d | 546 | /** |
cdd6c482 | 547 | * struct perf_event - performance event kernel representation: |
0793a61d | 548 | */ |
cdd6c482 IM |
549 | struct perf_event { |
550 | #ifdef CONFIG_PERF_EVENTS | |
9886167d PZ |
551 | /* |
552 | * entry onto perf_event_context::event_list; | |
553 | * modifications require ctx->lock | |
554 | * RCU safe iterations. | |
555 | */ | |
592903cd | 556 | struct list_head event_entry; |
9886167d PZ |
557 | |
558 | /* | |
559 | * XXX: group_entry and sibling_list should be mutually exclusive; | |
560 | * either you're a sibling on a group, or you're the group leader. | |
561 | * Rework the code to always use the same list element. | |
562 | * | |
563 | * Locked for modification by both ctx->mutex and ctx->lock; holding | |
564 | * either sufficies for read. | |
565 | */ | |
566 | struct list_head group_entry; | |
04289bb9 | 567 | struct list_head sibling_list; |
9886167d PZ |
568 | |
569 | /* | |
570 | * We need storage to track the entries in perf_pmu_migrate_context; we | |
571 | * cannot use the event_entry because of RCU and we want to keep the | |
572 | * group in tact which avoids us using the other two entries. | |
573 | */ | |
574 | struct list_head migrate_entry; | |
575 | ||
f3ae75de SE |
576 | struct hlist_node hlist_entry; |
577 | struct list_head active_entry; | |
0127c3ea | 578 | int nr_siblings; |
4ff6a8de DCC |
579 | |
580 | /* Not serialized. Only written during event initialization. */ | |
581 | int event_caps; | |
582 | /* The cumulative AND of all event_caps for events in this group. */ | |
583 | int group_caps; | |
584 | ||
cdd6c482 | 585 | struct perf_event *group_leader; |
a4eaf7f1 | 586 | struct pmu *pmu; |
54d751d4 | 587 | void *pmu_private; |
04289bb9 | 588 | |
cdd6c482 | 589 | enum perf_event_active_state state; |
8a49542c | 590 | unsigned int attach_state; |
e7850595 | 591 | local64_t count; |
a6e6dea6 | 592 | atomic64_t child_count; |
ee06094f | 593 | |
53cfbf59 | 594 | /* |
cdd6c482 | 595 | * These are the total time in nanoseconds that the event |
53cfbf59 | 596 | * has been enabled (i.e. eligible to run, and the task has |
cdd6c482 | 597 | * been scheduled in, if this is a per-task event) |
53cfbf59 PM |
598 | * and running (scheduled onto the CPU), respectively. |
599 | * | |
600 | * They are computed from tstamp_enabled, tstamp_running and | |
cdd6c482 | 601 | * tstamp_stopped when the event is in INACTIVE or ACTIVE state. |
53cfbf59 PM |
602 | */ |
603 | u64 total_time_enabled; | |
604 | u64 total_time_running; | |
605 | ||
606 | /* | |
607 | * These are timestamps used for computing total_time_enabled | |
cdd6c482 | 608 | * and total_time_running when the event is in INACTIVE or |
53cfbf59 PM |
609 | * ACTIVE state, measured in nanoseconds from an arbitrary point |
610 | * in time. | |
cdd6c482 IM |
611 | * tstamp_enabled: the notional time when the event was enabled |
612 | * tstamp_running: the notional time when the event was scheduled on | |
53cfbf59 | 613 | * tstamp_stopped: in INACTIVE state, the notional time when the |
cdd6c482 | 614 | * event was scheduled off. |
53cfbf59 PM |
615 | */ |
616 | u64 tstamp_enabled; | |
617 | u64 tstamp_running; | |
618 | u64 tstamp_stopped; | |
619 | ||
eed01528 SE |
620 | /* |
621 | * timestamp shadows the actual context timing but it can | |
622 | * be safely used in NMI interrupt context. It reflects the | |
623 | * context time as it was when the event was last scheduled in. | |
624 | * | |
625 | * ctx_time already accounts for ctx->timestamp. Therefore to | |
626 | * compute ctx_time for a sample, simply add perf_clock(). | |
627 | */ | |
628 | u64 shadow_ctx_time; | |
629 | ||
24f1e32c | 630 | struct perf_event_attr attr; |
c320c7b7 | 631 | u16 header_size; |
6844c09d | 632 | u16 id_header_size; |
c320c7b7 | 633 | u16 read_size; |
cdd6c482 | 634 | struct hw_perf_event hw; |
0793a61d | 635 | |
cdd6c482 | 636 | struct perf_event_context *ctx; |
a6fa941d | 637 | atomic_long_t refcount; |
0793a61d | 638 | |
53cfbf59 PM |
639 | /* |
640 | * These accumulate total time (in nanoseconds) that children | |
cdd6c482 | 641 | * events have been enabled and running, respectively. |
53cfbf59 PM |
642 | */ |
643 | atomic64_t child_total_time_enabled; | |
644 | atomic64_t child_total_time_running; | |
645 | ||
0793a61d | 646 | /* |
d859e29f | 647 | * Protect attach/detach and child_list: |
0793a61d | 648 | */ |
fccc714b PZ |
649 | struct mutex child_mutex; |
650 | struct list_head child_list; | |
cdd6c482 | 651 | struct perf_event *parent; |
0793a61d TG |
652 | |
653 | int oncpu; | |
654 | int cpu; | |
655 | ||
082ff5a2 PZ |
656 | struct list_head owner_entry; |
657 | struct task_struct *owner; | |
658 | ||
7b732a75 PZ |
659 | /* mmap bits */ |
660 | struct mutex mmap_mutex; | |
661 | atomic_t mmap_count; | |
26cb63ad | 662 | |
76369139 | 663 | struct ring_buffer *rb; |
10c6db11 | 664 | struct list_head rb_entry; |
b69cf536 PZ |
665 | unsigned long rcu_batches; |
666 | int rcu_pending; | |
37d81828 | 667 | |
7b732a75 | 668 | /* poll related */ |
0793a61d | 669 | wait_queue_head_t waitq; |
3c446b3d | 670 | struct fasync_struct *fasync; |
79f14641 PZ |
671 | |
672 | /* delayed work for NMIs and such */ | |
673 | int pending_wakeup; | |
4c9e2542 | 674 | int pending_kill; |
79f14641 | 675 | int pending_disable; |
e360adbe | 676 | struct irq_work pending; |
592903cd | 677 | |
79f14641 PZ |
678 | atomic_t event_limit; |
679 | ||
375637bc AS |
680 | /* address range filters */ |
681 | struct perf_addr_filters_head addr_filters; | |
682 | /* vma address array for file-based filders */ | |
683 | unsigned long *addr_filters_offs; | |
684 | unsigned long addr_filters_gen; | |
685 | ||
cdd6c482 | 686 | void (*destroy)(struct perf_event *); |
592903cd | 687 | struct rcu_head rcu_head; |
709e50cf PZ |
688 | |
689 | struct pid_namespace *ns; | |
8e5799b1 | 690 | u64 id; |
6fb2915d | 691 | |
34f43927 | 692 | u64 (*clock)(void); |
b326e956 | 693 | perf_overflow_handler_t overflow_handler; |
4dc0da86 | 694 | void *overflow_handler_context; |
aa6a5f3c AS |
695 | #ifdef CONFIG_BPF_SYSCALL |
696 | perf_overflow_handler_t orig_overflow_handler; | |
697 | struct bpf_prog *prog; | |
698 | #endif | |
453f19ee | 699 | |
07b139c8 | 700 | #ifdef CONFIG_EVENT_TRACING |
2425bcb9 | 701 | struct trace_event_call *tp_event; |
6fb2915d | 702 | struct event_filter *filter; |
ced39002 JO |
703 | #ifdef CONFIG_FUNCTION_TRACER |
704 | struct ftrace_ops ftrace_ops; | |
705 | #endif | |
ee06094f | 706 | #endif |
6fb2915d | 707 | |
e5d1367f SE |
708 | #ifdef CONFIG_CGROUP_PERF |
709 | struct perf_cgroup *cgrp; /* cgroup event is attach to */ | |
710 | int cgrp_defer_enabled; | |
711 | #endif | |
712 | ||
f2fb6bef | 713 | struct list_head sb_list; |
6fb2915d | 714 | #endif /* CONFIG_PERF_EVENTS */ |
0793a61d TG |
715 | }; |
716 | ||
717 | /** | |
cdd6c482 | 718 | * struct perf_event_context - event context structure |
0793a61d | 719 | * |
cdd6c482 | 720 | * Used as a container for task events and CPU events as well: |
0793a61d | 721 | */ |
cdd6c482 | 722 | struct perf_event_context { |
108b02cf | 723 | struct pmu *pmu; |
0793a61d | 724 | /* |
cdd6c482 | 725 | * Protect the states of the events in the list, |
d859e29f | 726 | * nr_active, and the list: |
0793a61d | 727 | */ |
e625cce1 | 728 | raw_spinlock_t lock; |
d859e29f | 729 | /* |
cdd6c482 | 730 | * Protect the list of events. Locking either mutex or lock |
d859e29f PM |
731 | * is sufficient to ensure the list doesn't change; to change |
732 | * the list you need to lock both the mutex and the spinlock. | |
733 | */ | |
a308444c | 734 | struct mutex mutex; |
04289bb9 | 735 | |
2fde4f94 | 736 | struct list_head active_ctx_list; |
889ff015 FW |
737 | struct list_head pinned_groups; |
738 | struct list_head flexible_groups; | |
a308444c | 739 | struct list_head event_list; |
cdd6c482 | 740 | int nr_events; |
a308444c IM |
741 | int nr_active; |
742 | int is_active; | |
bfbd3381 | 743 | int nr_stat; |
0f5a2601 | 744 | int nr_freq; |
dddd3379 | 745 | int rotate_disable; |
a308444c IM |
746 | atomic_t refcount; |
747 | struct task_struct *task; | |
53cfbf59 PM |
748 | |
749 | /* | |
4af4998b | 750 | * Context clock, runs when context enabled. |
53cfbf59 | 751 | */ |
a308444c IM |
752 | u64 time; |
753 | u64 timestamp; | |
564c2b21 PM |
754 | |
755 | /* | |
756 | * These fields let us detect when two contexts have both | |
757 | * been cloned (inherited) from a common ancestor. | |
758 | */ | |
cdd6c482 | 759 | struct perf_event_context *parent_ctx; |
a308444c IM |
760 | u64 parent_gen; |
761 | u64 generation; | |
762 | int pin_count; | |
db4a8356 | 763 | #ifdef CONFIG_CGROUP_PERF |
d010b332 | 764 | int nr_cgroups; /* cgroup evts */ |
db4a8356 | 765 | #endif |
4af57ef2 | 766 | void *task_ctx_data; /* pmu specific data */ |
28009ce4 | 767 | struct rcu_head rcu_head; |
0793a61d TG |
768 | }; |
769 | ||
7ae07ea3 FW |
770 | /* |
771 | * Number of contexts where an event can trigger: | |
e7e7ee2e | 772 | * task, softirq, hardirq, nmi. |
7ae07ea3 FW |
773 | */ |
774 | #define PERF_NR_CONTEXTS 4 | |
775 | ||
0793a61d | 776 | /** |
cdd6c482 | 777 | * struct perf_event_cpu_context - per cpu event context structure |
0793a61d TG |
778 | */ |
779 | struct perf_cpu_context { | |
cdd6c482 IM |
780 | struct perf_event_context ctx; |
781 | struct perf_event_context *task_ctx; | |
0793a61d | 782 | int active_oncpu; |
3b6f9e5c | 783 | int exclusive; |
4cfafd30 PZ |
784 | |
785 | raw_spinlock_t hrtimer_lock; | |
9e630205 SE |
786 | struct hrtimer hrtimer; |
787 | ktime_t hrtimer_interval; | |
4cfafd30 PZ |
788 | unsigned int hrtimer_active; |
789 | ||
db4a8356 | 790 | #ifdef CONFIG_CGROUP_PERF |
e5d1367f | 791 | struct perf_cgroup *cgrp; |
058fe1c0 | 792 | struct list_head cgrp_cpuctx_entry; |
db4a8356 | 793 | #endif |
e48c1788 PZ |
794 | |
795 | struct list_head sched_cb_entry; | |
796 | int sched_cb_usage; | |
0793a61d TG |
797 | }; |
798 | ||
5622f295 | 799 | struct perf_output_handle { |
57c0c15b | 800 | struct perf_event *event; |
76369139 | 801 | struct ring_buffer *rb; |
6d1acfd5 | 802 | unsigned long wakeup; |
5d967a8b | 803 | unsigned long size; |
fdc26706 AS |
804 | union { |
805 | void *addr; | |
806 | unsigned long head; | |
807 | }; | |
5d967a8b | 808 | int page; |
5622f295 MM |
809 | }; |
810 | ||
0515e599 AS |
811 | struct bpf_perf_event_data_kern { |
812 | struct pt_regs *regs; | |
813 | struct perf_sample_data *data; | |
814 | }; | |
815 | ||
39bed6cb MF |
816 | #ifdef CONFIG_CGROUP_PERF |
817 | ||
818 | /* | |
819 | * perf_cgroup_info keeps track of time_enabled for a cgroup. | |
820 | * This is a per-cpu dynamically allocated data structure. | |
821 | */ | |
822 | struct perf_cgroup_info { | |
823 | u64 time; | |
824 | u64 timestamp; | |
825 | }; | |
826 | ||
827 | struct perf_cgroup { | |
828 | struct cgroup_subsys_state css; | |
829 | struct perf_cgroup_info __percpu *info; | |
830 | }; | |
831 | ||
832 | /* | |
833 | * Must ensure cgroup is pinned (css_get) before calling | |
834 | * this function. In other words, we cannot call this function | |
835 | * if there is no cgroup event for the current CPU context. | |
836 | */ | |
837 | static inline struct perf_cgroup * | |
614e4c4e | 838 | perf_cgroup_from_task(struct task_struct *task, struct perf_event_context *ctx) |
39bed6cb | 839 | { |
614e4c4e SE |
840 | return container_of(task_css_check(task, perf_event_cgrp_id, |
841 | ctx ? lockdep_is_held(&ctx->lock) | |
842 | : true), | |
39bed6cb MF |
843 | struct perf_cgroup, css); |
844 | } | |
845 | #endif /* CONFIG_CGROUP_PERF */ | |
846 | ||
cdd6c482 | 847 | #ifdef CONFIG_PERF_EVENTS |
829b42dd | 848 | |
fdc26706 AS |
849 | extern void *perf_aux_output_begin(struct perf_output_handle *handle, |
850 | struct perf_event *event); | |
851 | extern void perf_aux_output_end(struct perf_output_handle *handle, | |
852 | unsigned long size, bool truncated); | |
853 | extern int perf_aux_output_skip(struct perf_output_handle *handle, | |
854 | unsigned long size); | |
855 | extern void *perf_get_aux(struct perf_output_handle *handle); | |
856 | ||
03d8e80b | 857 | extern int perf_pmu_register(struct pmu *pmu, const char *name, int type); |
b0a873eb | 858 | extern void perf_pmu_unregister(struct pmu *pmu); |
621a01ea | 859 | |
3bf101ba | 860 | extern int perf_num_counters(void); |
84c79910 | 861 | extern const char *perf_pmu_name(void); |
ab0cce56 JO |
862 | extern void __perf_event_task_sched_in(struct task_struct *prev, |
863 | struct task_struct *task); | |
864 | extern void __perf_event_task_sched_out(struct task_struct *prev, | |
865 | struct task_struct *next); | |
cdd6c482 IM |
866 | extern int perf_event_init_task(struct task_struct *child); |
867 | extern void perf_event_exit_task(struct task_struct *child); | |
868 | extern void perf_event_free_task(struct task_struct *task); | |
4e231c79 | 869 | extern void perf_event_delayed_put(struct task_struct *task); |
e03e7ee3 | 870 | extern struct file *perf_event_get(unsigned int fd); |
ffe8690c | 871 | extern const struct perf_event_attr *perf_event_attrs(struct perf_event *event); |
cdd6c482 | 872 | extern void perf_event_print_debug(void); |
33696fc0 PZ |
873 | extern void perf_pmu_disable(struct pmu *pmu); |
874 | extern void perf_pmu_enable(struct pmu *pmu); | |
ba532500 YZ |
875 | extern void perf_sched_cb_dec(struct pmu *pmu); |
876 | extern void perf_sched_cb_inc(struct pmu *pmu); | |
cdd6c482 IM |
877 | extern int perf_event_task_disable(void); |
878 | extern int perf_event_task_enable(void); | |
26ca5c11 | 879 | extern int perf_event_refresh(struct perf_event *event, int refresh); |
cdd6c482 | 880 | extern void perf_event_update_userpage(struct perf_event *event); |
fb0459d7 AV |
881 | extern int perf_event_release_kernel(struct perf_event *event); |
882 | extern struct perf_event * | |
883 | perf_event_create_kernel_counter(struct perf_event_attr *attr, | |
884 | int cpu, | |
38a81da2 | 885 | struct task_struct *task, |
4dc0da86 AK |
886 | perf_overflow_handler_t callback, |
887 | void *context); | |
0cda4c02 YZ |
888 | extern void perf_pmu_migrate_context(struct pmu *pmu, |
889 | int src_cpu, int dst_cpu); | |
ffe8690c | 890 | extern u64 perf_event_read_local(struct perf_event *event); |
59ed446f PZ |
891 | extern u64 perf_event_read_value(struct perf_event *event, |
892 | u64 *enabled, u64 *running); | |
5c92d124 | 893 | |
d010b332 | 894 | |
df1a132b | 895 | struct perf_sample_data { |
2565711f PZ |
896 | /* |
897 | * Fields set by perf_sample_data_init(), group so as to | |
898 | * minimize the cachelines touched. | |
899 | */ | |
900 | u64 addr; | |
901 | struct perf_raw_record *raw; | |
902 | struct perf_branch_stack *br_stack; | |
903 | u64 period; | |
904 | u64 weight; | |
905 | u64 txn; | |
906 | union perf_mem_data_src data_src; | |
5622f295 | 907 | |
2565711f PZ |
908 | /* |
909 | * The other fields, optionally {set,used} by | |
910 | * perf_{prepare,output}_sample(). | |
911 | */ | |
912 | u64 type; | |
5622f295 MM |
913 | u64 ip; |
914 | struct { | |
915 | u32 pid; | |
916 | u32 tid; | |
917 | } tid_entry; | |
918 | u64 time; | |
5622f295 MM |
919 | u64 id; |
920 | u64 stream_id; | |
921 | struct { | |
922 | u32 cpu; | |
923 | u32 reserved; | |
924 | } cpu_entry; | |
5622f295 | 925 | struct perf_callchain_entry *callchain; |
88a7c26a AL |
926 | |
927 | /* | |
928 | * regs_user may point to task_pt_regs or to regs_user_copy, depending | |
929 | * on arch details. | |
930 | */ | |
60e2364e | 931 | struct perf_regs regs_user; |
88a7c26a AL |
932 | struct pt_regs regs_user_copy; |
933 | ||
60e2364e | 934 | struct perf_regs regs_intr; |
c5ebcedb | 935 | u64 stack_user_size; |
2565711f | 936 | } ____cacheline_aligned; |
df1a132b | 937 | |
770eee1f SE |
938 | /* default value for data source */ |
939 | #define PERF_MEM_NA (PERF_MEM_S(OP, NA) |\ | |
940 | PERF_MEM_S(LVL, NA) |\ | |
941 | PERF_MEM_S(SNOOP, NA) |\ | |
942 | PERF_MEM_S(LOCK, NA) |\ | |
943 | PERF_MEM_S(TLB, NA)) | |
944 | ||
fd0d000b RR |
945 | static inline void perf_sample_data_init(struct perf_sample_data *data, |
946 | u64 addr, u64 period) | |
dc1d628a | 947 | { |
fd0d000b | 948 | /* remaining struct members initialized in perf_prepare_sample() */ |
dc1d628a PZ |
949 | data->addr = addr; |
950 | data->raw = NULL; | |
bce38cd5 | 951 | data->br_stack = NULL; |
4018994f | 952 | data->period = period; |
c3feedf2 | 953 | data->weight = 0; |
770eee1f | 954 | data->data_src.val = PERF_MEM_NA; |
fdfbbd07 | 955 | data->txn = 0; |
dc1d628a PZ |
956 | } |
957 | ||
5622f295 MM |
958 | extern void perf_output_sample(struct perf_output_handle *handle, |
959 | struct perf_event_header *header, | |
960 | struct perf_sample_data *data, | |
cdd6c482 | 961 | struct perf_event *event); |
5622f295 MM |
962 | extern void perf_prepare_sample(struct perf_event_header *header, |
963 | struct perf_sample_data *data, | |
cdd6c482 | 964 | struct perf_event *event, |
5622f295 MM |
965 | struct pt_regs *regs); |
966 | ||
a8b0ca17 | 967 | extern int perf_event_overflow(struct perf_event *event, |
5622f295 MM |
968 | struct perf_sample_data *data, |
969 | struct pt_regs *regs); | |
df1a132b | 970 | |
9ecda41a WN |
971 | extern void perf_event_output_forward(struct perf_event *event, |
972 | struct perf_sample_data *data, | |
973 | struct pt_regs *regs); | |
974 | extern void perf_event_output_backward(struct perf_event *event, | |
975 | struct perf_sample_data *data, | |
976 | struct pt_regs *regs); | |
21509084 | 977 | extern void perf_event_output(struct perf_event *event, |
9ecda41a WN |
978 | struct perf_sample_data *data, |
979 | struct pt_regs *regs); | |
21509084 | 980 | |
1879445d WN |
981 | static inline bool |
982 | is_default_overflow_handler(struct perf_event *event) | |
983 | { | |
9ecda41a WN |
984 | if (likely(event->overflow_handler == perf_event_output_forward)) |
985 | return true; | |
986 | if (unlikely(event->overflow_handler == perf_event_output_backward)) | |
987 | return true; | |
988 | return false; | |
1879445d WN |
989 | } |
990 | ||
21509084 YZ |
991 | extern void |
992 | perf_event_header__init_id(struct perf_event_header *header, | |
993 | struct perf_sample_data *data, | |
994 | struct perf_event *event); | |
995 | extern void | |
996 | perf_event__output_id_sample(struct perf_event *event, | |
997 | struct perf_output_handle *handle, | |
998 | struct perf_sample_data *sample); | |
999 | ||
f38b0dbb KL |
1000 | extern void |
1001 | perf_log_lost_samples(struct perf_event *event, u64 lost); | |
1002 | ||
6c7e550f FBH |
1003 | static inline bool is_sampling_event(struct perf_event *event) |
1004 | { | |
1005 | return event->attr.sample_period != 0; | |
1006 | } | |
1007 | ||
3b6f9e5c | 1008 | /* |
cdd6c482 | 1009 | * Return 1 for a software event, 0 for a hardware event |
3b6f9e5c | 1010 | */ |
cdd6c482 | 1011 | static inline int is_software_event(struct perf_event *event) |
3b6f9e5c | 1012 | { |
4ff6a8de | 1013 | return event->event_caps & PERF_EV_CAP_SOFTWARE; |
3b6f9e5c PM |
1014 | } |
1015 | ||
c5905afb | 1016 | extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX]; |
f29ac756 | 1017 | |
86038c5e | 1018 | extern void ___perf_sw_event(u32, u64, struct pt_regs *, u64); |
a8b0ca17 | 1019 | extern void __perf_sw_event(u32, u64, struct pt_regs *, u64); |
f29ac756 | 1020 | |
b0f82b81 | 1021 | #ifndef perf_arch_fetch_caller_regs |
e7e7ee2e | 1022 | static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { } |
b0f82b81 | 1023 | #endif |
5331d7b8 FW |
1024 | |
1025 | /* | |
1026 | * Take a snapshot of the regs. Skip ip and frame pointer to | |
1027 | * the nth caller. We only need a few of the regs: | |
1028 | * - ip for PERF_SAMPLE_IP | |
1029 | * - cs for user_mode() tests | |
1030 | * - bp for callchains | |
1031 | * - eflags, for future purposes, just in case | |
1032 | */ | |
b0f82b81 | 1033 | static inline void perf_fetch_caller_regs(struct pt_regs *regs) |
5331d7b8 | 1034 | { |
b0f82b81 | 1035 | perf_arch_fetch_caller_regs(regs, CALLER_ADDR0); |
5331d7b8 FW |
1036 | } |
1037 | ||
7e54a5a0 | 1038 | static __always_inline void |
a8b0ca17 | 1039 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) |
e49a5bd3 | 1040 | { |
86038c5e PZI |
1041 | if (static_key_false(&perf_swevent_enabled[event_id])) |
1042 | __perf_sw_event(event_id, nr, regs, addr); | |
1043 | } | |
1044 | ||
1045 | DECLARE_PER_CPU(struct pt_regs, __perf_regs[4]); | |
7e54a5a0 | 1046 | |
86038c5e PZI |
1047 | /* |
1048 | * 'Special' version for the scheduler, it hard assumes no recursion, | |
1049 | * which is guaranteed by us not actually scheduling inside other swevents | |
1050 | * because those disable preemption. | |
1051 | */ | |
1052 | static __always_inline void | |
1053 | perf_sw_event_sched(u32 event_id, u64 nr, u64 addr) | |
1054 | { | |
c5905afb | 1055 | if (static_key_false(&perf_swevent_enabled[event_id])) { |
86038c5e PZI |
1056 | struct pt_regs *regs = this_cpu_ptr(&__perf_regs[0]); |
1057 | ||
1058 | perf_fetch_caller_regs(regs); | |
1059 | ___perf_sw_event(event_id, nr, regs, addr); | |
e49a5bd3 FW |
1060 | } |
1061 | } | |
1062 | ||
9107c89e | 1063 | extern struct static_key_false perf_sched_events; |
ee6dcfa4 | 1064 | |
ff303e66 PZ |
1065 | static __always_inline bool |
1066 | perf_sw_migrate_enabled(void) | |
1067 | { | |
1068 | if (static_key_false(&perf_swevent_enabled[PERF_COUNT_SW_CPU_MIGRATIONS])) | |
1069 | return true; | |
1070 | return false; | |
1071 | } | |
1072 | ||
1073 | static inline void perf_event_task_migrate(struct task_struct *task) | |
1074 | { | |
1075 | if (perf_sw_migrate_enabled()) | |
1076 | task->sched_migrated = 1; | |
1077 | } | |
1078 | ||
ab0cce56 | 1079 | static inline void perf_event_task_sched_in(struct task_struct *prev, |
a8d757ef | 1080 | struct task_struct *task) |
ab0cce56 | 1081 | { |
9107c89e | 1082 | if (static_branch_unlikely(&perf_sched_events)) |
ab0cce56 | 1083 | __perf_event_task_sched_in(prev, task); |
ff303e66 PZ |
1084 | |
1085 | if (perf_sw_migrate_enabled() && task->sched_migrated) { | |
1086 | struct pt_regs *regs = this_cpu_ptr(&__perf_regs[0]); | |
1087 | ||
1088 | perf_fetch_caller_regs(regs); | |
1089 | ___perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, regs, 0); | |
1090 | task->sched_migrated = 0; | |
1091 | } | |
ab0cce56 JO |
1092 | } |
1093 | ||
1094 | static inline void perf_event_task_sched_out(struct task_struct *prev, | |
1095 | struct task_struct *next) | |
ee6dcfa4 | 1096 | { |
86038c5e | 1097 | perf_sw_event_sched(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 0); |
ee6dcfa4 | 1098 | |
9107c89e | 1099 | if (static_branch_unlikely(&perf_sched_events)) |
ab0cce56 | 1100 | __perf_event_task_sched_out(prev, next); |
ee6dcfa4 PZ |
1101 | } |
1102 | ||
eacd3ecc MF |
1103 | static inline u64 __perf_event_count(struct perf_event *event) |
1104 | { | |
1105 | return local64_read(&event->count) + atomic64_read(&event->child_count); | |
1106 | } | |
1107 | ||
3af9e859 | 1108 | extern void perf_event_mmap(struct vm_area_struct *vma); |
39447b38 | 1109 | extern struct perf_guest_info_callbacks *perf_guest_cbs; |
dcf46b94 ZY |
1110 | extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); |
1111 | extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); | |
39447b38 | 1112 | |
e041e328 | 1113 | extern void perf_event_exec(void); |
82b89778 | 1114 | extern void perf_event_comm(struct task_struct *tsk, bool exec); |
cdd6c482 | 1115 | extern void perf_event_fork(struct task_struct *tsk); |
8d1b2d93 | 1116 | |
56962b44 FW |
1117 | /* Callchains */ |
1118 | DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry); | |
1119 | ||
cfbcf468 ACM |
1120 | extern void perf_callchain_user(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs); |
1121 | extern void perf_callchain_kernel(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs); | |
568b329a AS |
1122 | extern struct perf_callchain_entry * |
1123 | get_perf_callchain(struct pt_regs *regs, u32 init_nr, bool kernel, bool user, | |
cfbcf468 | 1124 | u32 max_stack, bool crosstask, bool add_mark); |
97c79a38 | 1125 | extern int get_callchain_buffers(int max_stack); |
568b329a | 1126 | extern void put_callchain_buffers(void); |
394ee076 | 1127 | |
c5dfd78e | 1128 | extern int sysctl_perf_event_max_stack; |
c85b0334 | 1129 | extern int sysctl_perf_event_max_contexts_per_stack; |
c5dfd78e | 1130 | |
c85b0334 ACM |
1131 | static inline int perf_callchain_store_context(struct perf_callchain_entry_ctx *ctx, u64 ip) |
1132 | { | |
1133 | if (ctx->contexts < sysctl_perf_event_max_contexts_per_stack) { | |
1134 | struct perf_callchain_entry *entry = ctx->entry; | |
1135 | entry->ip[entry->nr++] = ip; | |
1136 | ++ctx->contexts; | |
1137 | return 0; | |
1138 | } else { | |
1139 | ctx->contexts_maxed = true; | |
1140 | return -1; /* no more room, stop walking the stack */ | |
1141 | } | |
1142 | } | |
3e4de4ec | 1143 | |
cfbcf468 | 1144 | static inline int perf_callchain_store(struct perf_callchain_entry_ctx *ctx, u64 ip) |
70791ce9 | 1145 | { |
c85b0334 | 1146 | if (ctx->nr < ctx->max_stack && !ctx->contexts_maxed) { |
3b1fff08 | 1147 | struct perf_callchain_entry *entry = ctx->entry; |
70791ce9 | 1148 | entry->ip[entry->nr++] = ip; |
3b1fff08 | 1149 | ++ctx->nr; |
568b329a AS |
1150 | return 0; |
1151 | } else { | |
1152 | return -1; /* no more room, stop walking the stack */ | |
1153 | } | |
70791ce9 | 1154 | } |
394ee076 | 1155 | |
cdd6c482 IM |
1156 | extern int sysctl_perf_event_paranoid; |
1157 | extern int sysctl_perf_event_mlock; | |
1158 | extern int sysctl_perf_event_sample_rate; | |
14c63f17 DH |
1159 | extern int sysctl_perf_cpu_time_max_percent; |
1160 | ||
1161 | extern void perf_sample_event_took(u64 sample_len_ns); | |
1ccd1549 | 1162 | |
163ec435 PZ |
1163 | extern int perf_proc_update_handler(struct ctl_table *table, int write, |
1164 | void __user *buffer, size_t *lenp, | |
1165 | loff_t *ppos); | |
14c63f17 DH |
1166 | extern int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write, |
1167 | void __user *buffer, size_t *lenp, | |
1168 | loff_t *ppos); | |
1169 | ||
c5dfd78e ACM |
1170 | int perf_event_max_stack_handler(struct ctl_table *table, int write, |
1171 | void __user *buffer, size_t *lenp, loff_t *ppos); | |
163ec435 | 1172 | |
320ebf09 PZ |
1173 | static inline bool perf_paranoid_tracepoint_raw(void) |
1174 | { | |
1175 | return sysctl_perf_event_paranoid > -1; | |
1176 | } | |
1177 | ||
1178 | static inline bool perf_paranoid_cpu(void) | |
1179 | { | |
1180 | return sysctl_perf_event_paranoid > 0; | |
1181 | } | |
1182 | ||
1183 | static inline bool perf_paranoid_kernel(void) | |
1184 | { | |
1185 | return sysctl_perf_event_paranoid > 1; | |
1186 | } | |
1187 | ||
cdd6c482 | 1188 | extern void perf_event_init(void); |
1e1dcd93 | 1189 | extern void perf_tp_event(u16 event_type, u64 count, void *record, |
1c024eca | 1190 | int entry_size, struct pt_regs *regs, |
e6dab5ff AV |
1191 | struct hlist_head *head, int rctx, |
1192 | struct task_struct *task); | |
24f1e32c | 1193 | extern void perf_bp_event(struct perf_event *event, void *data); |
0d905bca | 1194 | |
9d23a90a | 1195 | #ifndef perf_misc_flags |
e7e7ee2e IM |
1196 | # define perf_misc_flags(regs) \ |
1197 | (user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL) | |
1198 | # define perf_instruction_pointer(regs) instruction_pointer(regs) | |
9d23a90a PM |
1199 | #endif |
1200 | ||
bce38cd5 SE |
1201 | static inline bool has_branch_stack(struct perf_event *event) |
1202 | { | |
1203 | return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK; | |
a46a2300 YZ |
1204 | } |
1205 | ||
1206 | static inline bool needs_branch_stack(struct perf_event *event) | |
1207 | { | |
1208 | return event->attr.branch_sample_type != 0; | |
bce38cd5 SE |
1209 | } |
1210 | ||
45bfb2e5 PZ |
1211 | static inline bool has_aux(struct perf_event *event) |
1212 | { | |
1213 | return event->pmu->setup_aux; | |
1214 | } | |
1215 | ||
9ecda41a WN |
1216 | static inline bool is_write_backward(struct perf_event *event) |
1217 | { | |
1218 | return !!event->attr.write_backward; | |
1219 | } | |
1220 | ||
375637bc AS |
1221 | static inline bool has_addr_filter(struct perf_event *event) |
1222 | { | |
1223 | return event->pmu->nr_addr_filters; | |
1224 | } | |
1225 | ||
1226 | /* | |
1227 | * An inherited event uses parent's filters | |
1228 | */ | |
1229 | static inline struct perf_addr_filters_head * | |
1230 | perf_event_addr_filters(struct perf_event *event) | |
1231 | { | |
1232 | struct perf_addr_filters_head *ifh = &event->addr_filters; | |
1233 | ||
1234 | if (event->parent) | |
1235 | ifh = &event->parent->addr_filters; | |
1236 | ||
1237 | return ifh; | |
1238 | } | |
1239 | ||
1240 | extern void perf_event_addr_filters_sync(struct perf_event *event); | |
1241 | ||
5622f295 | 1242 | extern int perf_output_begin(struct perf_output_handle *handle, |
a7ac67ea | 1243 | struct perf_event *event, unsigned int size); |
9ecda41a WN |
1244 | extern int perf_output_begin_forward(struct perf_output_handle *handle, |
1245 | struct perf_event *event, | |
1246 | unsigned int size); | |
1247 | extern int perf_output_begin_backward(struct perf_output_handle *handle, | |
1248 | struct perf_event *event, | |
1249 | unsigned int size); | |
1250 | ||
5622f295 | 1251 | extern void perf_output_end(struct perf_output_handle *handle); |
91d7753a | 1252 | extern unsigned int perf_output_copy(struct perf_output_handle *handle, |
5622f295 | 1253 | const void *buf, unsigned int len); |
5685e0ff JO |
1254 | extern unsigned int perf_output_skip(struct perf_output_handle *handle, |
1255 | unsigned int len); | |
4ed7c92d PZ |
1256 | extern int perf_swevent_get_recursion_context(void); |
1257 | extern void perf_swevent_put_recursion_context(int rctx); | |
ab573844 | 1258 | extern u64 perf_swevent_set_period(struct perf_event *event); |
44234adc FW |
1259 | extern void perf_event_enable(struct perf_event *event); |
1260 | extern void perf_event_disable(struct perf_event *event); | |
fae3fde6 | 1261 | extern void perf_event_disable_local(struct perf_event *event); |
5aab90ce | 1262 | extern void perf_event_disable_inatomic(struct perf_event *event); |
e9d2b064 | 1263 | extern void perf_event_task_tick(void); |
475113d9 | 1264 | extern int perf_event_account_interrupt(struct perf_event *event); |
e041e328 | 1265 | #else /* !CONFIG_PERF_EVENTS: */ |
fdc26706 AS |
1266 | static inline void * |
1267 | perf_aux_output_begin(struct perf_output_handle *handle, | |
1268 | struct perf_event *event) { return NULL; } | |
1269 | static inline void | |
1270 | perf_aux_output_end(struct perf_output_handle *handle, unsigned long size, | |
1271 | bool truncated) { } | |
1272 | static inline int | |
1273 | perf_aux_output_skip(struct perf_output_handle *handle, | |
1274 | unsigned long size) { return -EINVAL; } | |
1275 | static inline void * | |
1276 | perf_get_aux(struct perf_output_handle *handle) { return NULL; } | |
0793a61d | 1277 | static inline void |
ff303e66 PZ |
1278 | perf_event_task_migrate(struct task_struct *task) { } |
1279 | static inline void | |
ab0cce56 JO |
1280 | perf_event_task_sched_in(struct task_struct *prev, |
1281 | struct task_struct *task) { } | |
1282 | static inline void | |
1283 | perf_event_task_sched_out(struct task_struct *prev, | |
1284 | struct task_struct *next) { } | |
cdd6c482 IM |
1285 | static inline int perf_event_init_task(struct task_struct *child) { return 0; } |
1286 | static inline void perf_event_exit_task(struct task_struct *child) { } | |
1287 | static inline void perf_event_free_task(struct task_struct *task) { } | |
4e231c79 | 1288 | static inline void perf_event_delayed_put(struct task_struct *task) { } |
e03e7ee3 | 1289 | static inline struct file *perf_event_get(unsigned int fd) { return ERR_PTR(-EINVAL); } |
ffe8690c KX |
1290 | static inline const struct perf_event_attr *perf_event_attrs(struct perf_event *event) |
1291 | { | |
1292 | return ERR_PTR(-EINVAL); | |
1293 | } | |
1294 | static inline u64 perf_event_read_local(struct perf_event *event) { return -EINVAL; } | |
57c0c15b | 1295 | static inline void perf_event_print_debug(void) { } |
57c0c15b IM |
1296 | static inline int perf_event_task_disable(void) { return -EINVAL; } |
1297 | static inline int perf_event_task_enable(void) { return -EINVAL; } | |
26ca5c11 AK |
1298 | static inline int perf_event_refresh(struct perf_event *event, int refresh) |
1299 | { | |
1300 | return -EINVAL; | |
1301 | } | |
15dbf27c | 1302 | |
925d519a | 1303 | static inline void |
a8b0ca17 | 1304 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) { } |
24f1e32c | 1305 | static inline void |
86038c5e PZI |
1306 | perf_sw_event_sched(u32 event_id, u64 nr, u64 addr) { } |
1307 | static inline void | |
184f412c | 1308 | perf_bp_event(struct perf_event *event, void *data) { } |
0a4a9391 | 1309 | |
39447b38 | 1310 | static inline int perf_register_guest_info_callbacks |
e7e7ee2e | 1311 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 1312 | static inline int perf_unregister_guest_info_callbacks |
e7e7ee2e | 1313 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 1314 | |
57c0c15b | 1315 | static inline void perf_event_mmap(struct vm_area_struct *vma) { } |
e041e328 | 1316 | static inline void perf_event_exec(void) { } |
82b89778 | 1317 | static inline void perf_event_comm(struct task_struct *tsk, bool exec) { } |
cdd6c482 IM |
1318 | static inline void perf_event_fork(struct task_struct *tsk) { } |
1319 | static inline void perf_event_init(void) { } | |
184f412c | 1320 | static inline int perf_swevent_get_recursion_context(void) { return -1; } |
4ed7c92d | 1321 | static inline void perf_swevent_put_recursion_context(int rctx) { } |
ab573844 | 1322 | static inline u64 perf_swevent_set_period(struct perf_event *event) { return 0; } |
44234adc FW |
1323 | static inline void perf_event_enable(struct perf_event *event) { } |
1324 | static inline void perf_event_disable(struct perf_event *event) { } | |
500ad2d8 | 1325 | static inline int __perf_event_disable(void *info) { return -1; } |
e9d2b064 | 1326 | static inline void perf_event_task_tick(void) { } |
ffe8690c | 1327 | static inline int perf_event_release_kernel(struct perf_event *event) { return 0; } |
0793a61d TG |
1328 | #endif |
1329 | ||
6c4d3bc9 DR |
1330 | #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL) |
1331 | extern void perf_restore_debug_store(void); | |
1332 | #else | |
1d9d8639 | 1333 | static inline void perf_restore_debug_store(void) { } |
0793a61d TG |
1334 | #endif |
1335 | ||
7e3f977e DB |
1336 | static __always_inline bool perf_raw_frag_last(const struct perf_raw_frag *frag) |
1337 | { | |
1338 | return frag->pad < sizeof(u64); | |
1339 | } | |
1340 | ||
e7e7ee2e | 1341 | #define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x)) |
5622f295 | 1342 | |
2663960c SB |
1343 | struct perf_pmu_events_attr { |
1344 | struct device_attribute attr; | |
1345 | u64 id; | |
3a54aaa0 | 1346 | const char *event_str; |
2663960c SB |
1347 | }; |
1348 | ||
fc07e9f9 AK |
1349 | struct perf_pmu_events_ht_attr { |
1350 | struct device_attribute attr; | |
1351 | u64 id; | |
1352 | const char *event_str_ht; | |
1353 | const char *event_str_noht; | |
1354 | }; | |
1355 | ||
fd979c01 CS |
1356 | ssize_t perf_event_sysfs_show(struct device *dev, struct device_attribute *attr, |
1357 | char *page); | |
1358 | ||
2663960c SB |
1359 | #define PMU_EVENT_ATTR(_name, _var, _id, _show) \ |
1360 | static struct perf_pmu_events_attr _var = { \ | |
1361 | .attr = __ATTR(_name, 0444, _show, NULL), \ | |
1362 | .id = _id, \ | |
1363 | }; | |
1364 | ||
f0405b81 CS |
1365 | #define PMU_EVENT_ATTR_STRING(_name, _var, _str) \ |
1366 | static struct perf_pmu_events_attr _var = { \ | |
1367 | .attr = __ATTR(_name, 0444, perf_event_sysfs_show, NULL), \ | |
1368 | .id = 0, \ | |
1369 | .event_str = _str, \ | |
1370 | }; | |
1371 | ||
641cc938 JO |
1372 | #define PMU_FORMAT_ATTR(_name, _format) \ |
1373 | static ssize_t \ | |
1374 | _name##_show(struct device *dev, \ | |
1375 | struct device_attribute *attr, \ | |
1376 | char *page) \ | |
1377 | { \ | |
1378 | BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE); \ | |
1379 | return sprintf(page, _format "\n"); \ | |
1380 | } \ | |
1381 | \ | |
1382 | static struct device_attribute format_attr_##_name = __ATTR_RO(_name) | |
1383 | ||
00e16c3d TG |
1384 | /* Performance counter hotplug functions */ |
1385 | #ifdef CONFIG_PERF_EVENTS | |
1386 | int perf_event_init_cpu(unsigned int cpu); | |
1387 | int perf_event_exit_cpu(unsigned int cpu); | |
1388 | #else | |
1389 | #define perf_event_init_cpu NULL | |
1390 | #define perf_event_exit_cpu NULL | |
1391 | #endif | |
1392 | ||
cdd6c482 | 1393 | #endif /* _LINUX_PERF_EVENT_H */ |