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 | }; | |
c7ab62bf HR |
142 | struct { /* amd_power */ |
143 | u64 pwr_acc; | |
144 | u64 ptsc; | |
145 | }; | |
24f1e32c | 146 | #ifdef CONFIG_HAVE_HW_BREAKPOINT |
45a73372 | 147 | struct { /* breakpoint */ |
d580ff86 PZ |
148 | /* |
149 | * Crufty hack to avoid the chicken and egg | |
150 | * problem hw_breakpoint has with context | |
151 | * creation and event initalization. | |
152 | */ | |
f22c1bb6 ON |
153 | struct arch_hw_breakpoint info; |
154 | struct list_head bp_list; | |
45a73372 | 155 | }; |
24f1e32c | 156 | #endif |
cf25f904 SS |
157 | struct { /* amd_iommu */ |
158 | u8 iommu_bank; | |
159 | u8 iommu_cntr; | |
160 | u16 padding; | |
161 | u64 conf; | |
162 | u64 conf1; | |
163 | }; | |
d6d020e9 | 164 | }; |
b0e87875 PZ |
165 | /* |
166 | * If the event is a per task event, this will point to the task in | |
167 | * question. See the comment in perf_event_alloc(). | |
168 | */ | |
50f16a8b | 169 | struct task_struct *target; |
b0e87875 | 170 | |
375637bc AS |
171 | /* |
172 | * PMU would store hardware filter configuration | |
173 | * here. | |
174 | */ | |
175 | void *addr_filters; | |
176 | ||
177 | /* Last sync'ed generation of filters */ | |
178 | unsigned long addr_filters_gen; | |
179 | ||
b0e87875 PZ |
180 | /* |
181 | * hw_perf_event::state flags; used to track the PERF_EF_* state. | |
182 | */ | |
183 | #define PERF_HES_STOPPED 0x01 /* the counter is stopped */ | |
184 | #define PERF_HES_UPTODATE 0x02 /* event->count up-to-date */ | |
185 | #define PERF_HES_ARCH 0x04 | |
186 | ||
a4eaf7f1 | 187 | int state; |
b0e87875 PZ |
188 | |
189 | /* | |
190 | * The last observed hardware counter value, updated with a | |
191 | * local64_cmpxchg() such that pmu::read() can be called nested. | |
192 | */ | |
e7850595 | 193 | local64_t prev_count; |
b0e87875 PZ |
194 | |
195 | /* | |
196 | * The period to start the next sample with. | |
197 | */ | |
b23f3325 | 198 | u64 sample_period; |
b0e87875 PZ |
199 | |
200 | /* | |
201 | * The period we started this sample with. | |
202 | */ | |
9e350de3 | 203 | u64 last_period; |
b0e87875 PZ |
204 | |
205 | /* | |
206 | * However much is left of the current period; note that this is | |
207 | * a full 64bit value and allows for generation of periods longer | |
208 | * than hardware might allow. | |
209 | */ | |
e7850595 | 210 | local64_t period_left; |
b0e87875 PZ |
211 | |
212 | /* | |
213 | * State for throttling the event, see __perf_event_overflow() and | |
214 | * perf_adjust_freq_unthr_context(). | |
215 | */ | |
e050e3f0 | 216 | u64 interrupts_seq; |
60db5e09 | 217 | u64 interrupts; |
6a24ed6c | 218 | |
b0e87875 PZ |
219 | /* |
220 | * State for freq target events, see __perf_event_overflow() and | |
221 | * perf_adjust_freq_unthr_context(). | |
222 | */ | |
abd50713 PZ |
223 | u64 freq_time_stamp; |
224 | u64 freq_count_stamp; | |
ee06094f | 225 | #endif |
0793a61d TG |
226 | }; |
227 | ||
cdd6c482 | 228 | struct perf_event; |
621a01ea | 229 | |
8d2cacbb PZ |
230 | /* |
231 | * Common implementation detail of pmu::{start,commit,cancel}_txn | |
232 | */ | |
fbbe0701 | 233 | #define PERF_PMU_TXN_ADD 0x1 /* txn to add/schedule event on PMU */ |
4a00c16e | 234 | #define PERF_PMU_TXN_READ 0x2 /* txn to read event group from PMU */ |
fbbe0701 | 235 | |
53b25335 VW |
236 | /** |
237 | * pmu::capabilities flags | |
238 | */ | |
239 | #define PERF_PMU_CAP_NO_INTERRUPT 0x01 | |
34f43927 | 240 | #define PERF_PMU_CAP_NO_NMI 0x02 |
0a4e38e6 | 241 | #define PERF_PMU_CAP_AUX_NO_SG 0x04 |
6a279230 | 242 | #define PERF_PMU_CAP_AUX_SW_DOUBLEBUF 0x08 |
bed5b25a | 243 | #define PERF_PMU_CAP_EXCLUSIVE 0x10 |
ec0d7729 | 244 | #define PERF_PMU_CAP_ITRACE 0x20 |
5101ef20 | 245 | #define PERF_PMU_CAP_HETEROGENEOUS_CPUS 0x40 |
53b25335 | 246 | |
621a01ea | 247 | /** |
4aeb0b42 | 248 | * struct pmu - generic performance monitoring unit |
621a01ea | 249 | */ |
4aeb0b42 | 250 | struct pmu { |
b0a873eb PZ |
251 | struct list_head entry; |
252 | ||
c464c76e | 253 | struct module *module; |
abe43400 | 254 | struct device *dev; |
0c9d42ed | 255 | const struct attribute_group **attr_groups; |
03d8e80b | 256 | const char *name; |
2e80a82a PZ |
257 | int type; |
258 | ||
53b25335 VW |
259 | /* |
260 | * various common per-pmu feature flags | |
261 | */ | |
262 | int capabilities; | |
263 | ||
108b02cf PZ |
264 | int * __percpu pmu_disable_count; |
265 | struct perf_cpu_context * __percpu pmu_cpu_context; | |
bed5b25a | 266 | atomic_t exclusive_cnt; /* < 0: cpu; > 0: tsk */ |
8dc85d54 | 267 | int task_ctx_nr; |
62b85639 | 268 | int hrtimer_interval_ms; |
6bde9b6c | 269 | |
375637bc AS |
270 | /* number of address filters this PMU can do */ |
271 | unsigned int nr_addr_filters; | |
272 | ||
6bde9b6c | 273 | /* |
a4eaf7f1 PZ |
274 | * Fully disable/enable this PMU, can be used to protect from the PMI |
275 | * as well as for lazy/batch writing of the MSRs. | |
6bde9b6c | 276 | */ |
ad5133b7 PZ |
277 | void (*pmu_enable) (struct pmu *pmu); /* optional */ |
278 | void (*pmu_disable) (struct pmu *pmu); /* optional */ | |
6bde9b6c | 279 | |
8d2cacbb | 280 | /* |
a4eaf7f1 | 281 | * Try and initialize the event for this PMU. |
b0e87875 PZ |
282 | * |
283 | * Returns: | |
284 | * -ENOENT -- @event is not for this PMU | |
285 | * | |
286 | * -ENODEV -- @event is for this PMU but PMU not present | |
287 | * -EBUSY -- @event is for this PMU but PMU temporarily unavailable | |
288 | * -EINVAL -- @event is for this PMU but @event is not valid | |
289 | * -EOPNOTSUPP -- @event is for this PMU, @event is valid, but not supported | |
290 | * -EACCESS -- @event is for this PMU, @event is valid, but no privilidges | |
291 | * | |
292 | * 0 -- @event is for this PMU and valid | |
293 | * | |
294 | * Other error return values are allowed. | |
8d2cacbb | 295 | */ |
b0a873eb PZ |
296 | int (*event_init) (struct perf_event *event); |
297 | ||
1e0fb9ec AL |
298 | /* |
299 | * Notification that the event was mapped or unmapped. Called | |
300 | * in the context of the mapping task. | |
301 | */ | |
bfe33492 PZ |
302 | void (*event_mapped) (struct perf_event *event, struct mm_struct *mm); /* optional */ |
303 | void (*event_unmapped) (struct perf_event *event, struct mm_struct *mm); /* optional */ | |
1e0fb9ec | 304 | |
b0e87875 PZ |
305 | /* |
306 | * Flags for ->add()/->del()/ ->start()/->stop(). There are | |
307 | * matching hw_perf_event::state flags. | |
308 | */ | |
a4eaf7f1 PZ |
309 | #define PERF_EF_START 0x01 /* start the counter when adding */ |
310 | #define PERF_EF_RELOAD 0x02 /* reload the counter when starting */ | |
311 | #define PERF_EF_UPDATE 0x04 /* update the counter when stopping */ | |
312 | ||
8d2cacbb | 313 | /* |
b0e87875 PZ |
314 | * Adds/Removes a counter to/from the PMU, can be done inside a |
315 | * transaction, see the ->*_txn() methods. | |
316 | * | |
317 | * The add/del callbacks will reserve all hardware resources required | |
318 | * to service the event, this includes any counter constraint | |
319 | * scheduling etc. | |
320 | * | |
321 | * Called with IRQs disabled and the PMU disabled on the CPU the event | |
322 | * is on. | |
323 | * | |
324 | * ->add() called without PERF_EF_START should result in the same state | |
325 | * as ->add() followed by ->stop(). | |
326 | * | |
327 | * ->del() must always PERF_EF_UPDATE stop an event. If it calls | |
328 | * ->stop() that must deal with already being stopped without | |
329 | * PERF_EF_UPDATE. | |
a4eaf7f1 PZ |
330 | */ |
331 | int (*add) (struct perf_event *event, int flags); | |
332 | void (*del) (struct perf_event *event, int flags); | |
333 | ||
334 | /* | |
b0e87875 PZ |
335 | * Starts/Stops a counter present on the PMU. |
336 | * | |
337 | * The PMI handler should stop the counter when perf_event_overflow() | |
338 | * returns !0. ->start() will be used to continue. | |
339 | * | |
340 | * Also used to change the sample period. | |
341 | * | |
342 | * Called with IRQs disabled and the PMU disabled on the CPU the event | |
343 | * is on -- will be called from NMI context with the PMU generates | |
344 | * NMIs. | |
345 | * | |
346 | * ->stop() with PERF_EF_UPDATE will read the counter and update | |
347 | * period/count values like ->read() would. | |
348 | * | |
349 | * ->start() with PERF_EF_RELOAD will reprogram the the counter | |
350 | * value, must be preceded by a ->stop() with PERF_EF_UPDATE. | |
a4eaf7f1 PZ |
351 | */ |
352 | void (*start) (struct perf_event *event, int flags); | |
353 | void (*stop) (struct perf_event *event, int flags); | |
354 | ||
355 | /* | |
356 | * Updates the counter value of the event. | |
b0e87875 PZ |
357 | * |
358 | * For sampling capable PMUs this will also update the software period | |
359 | * hw_perf_event::period_left field. | |
a4eaf7f1 | 360 | */ |
cdd6c482 | 361 | void (*read) (struct perf_event *event); |
6bde9b6c LM |
362 | |
363 | /* | |
24cd7f54 PZ |
364 | * Group events scheduling is treated as a transaction, add |
365 | * group events as a whole and perform one schedulability test. | |
366 | * If the test fails, roll back the whole group | |
a4eaf7f1 PZ |
367 | * |
368 | * Start the transaction, after this ->add() doesn't need to | |
24cd7f54 | 369 | * do schedulability tests. |
fbbe0701 SB |
370 | * |
371 | * Optional. | |
8d2cacbb | 372 | */ |
fbbe0701 | 373 | void (*start_txn) (struct pmu *pmu, unsigned int txn_flags); |
8d2cacbb | 374 | /* |
a4eaf7f1 | 375 | * If ->start_txn() disabled the ->add() schedulability test |
8d2cacbb PZ |
376 | * then ->commit_txn() is required to perform one. On success |
377 | * the transaction is closed. On error the transaction is kept | |
378 | * open until ->cancel_txn() is called. | |
fbbe0701 SB |
379 | * |
380 | * Optional. | |
8d2cacbb | 381 | */ |
fbbe0701 | 382 | int (*commit_txn) (struct pmu *pmu); |
8d2cacbb | 383 | /* |
a4eaf7f1 | 384 | * Will cancel the transaction, assumes ->del() is called |
25985edc | 385 | * for each successful ->add() during the transaction. |
fbbe0701 SB |
386 | * |
387 | * Optional. | |
8d2cacbb | 388 | */ |
fbbe0701 | 389 | void (*cancel_txn) (struct pmu *pmu); |
35edc2a5 PZ |
390 | |
391 | /* | |
392 | * Will return the value for perf_event_mmap_page::index for this event, | |
393 | * if no implementation is provided it will default to: event->hw.idx + 1. | |
394 | */ | |
395 | int (*event_idx) (struct perf_event *event); /*optional */ | |
d010b332 | 396 | |
ba532500 YZ |
397 | /* |
398 | * context-switches callback | |
399 | */ | |
400 | void (*sched_task) (struct perf_event_context *ctx, | |
401 | bool sched_in); | |
4af57ef2 YZ |
402 | /* |
403 | * PMU specific data size | |
404 | */ | |
405 | size_t task_ctx_size; | |
ba532500 | 406 | |
eacd3ecc | 407 | |
45bfb2e5 PZ |
408 | /* |
409 | * Set up pmu-private data structures for an AUX area | |
410 | */ | |
411 | void *(*setup_aux) (int cpu, void **pages, | |
412 | int nr_pages, bool overwrite); | |
413 | /* optional */ | |
414 | ||
415 | /* | |
416 | * Free pmu-private AUX data structures | |
417 | */ | |
418 | void (*free_aux) (void *aux); /* optional */ | |
66eb579e | 419 | |
375637bc AS |
420 | /* |
421 | * Validate address range filters: make sure the HW supports the | |
422 | * requested configuration and number of filters; return 0 if the | |
423 | * supplied filters are valid, -errno otherwise. | |
424 | * | |
425 | * Runs in the context of the ioctl()ing process and is not serialized | |
426 | * with the rest of the PMU callbacks. | |
427 | */ | |
428 | int (*addr_filters_validate) (struct list_head *filters); | |
429 | /* optional */ | |
430 | ||
431 | /* | |
432 | * Synchronize address range filter configuration: | |
433 | * translate hw-agnostic filters into hardware configuration in | |
434 | * event::hw::addr_filters. | |
435 | * | |
436 | * Runs as a part of filter sync sequence that is done in ->start() | |
437 | * callback by calling perf_event_addr_filters_sync(). | |
438 | * | |
439 | * May (and should) traverse event::addr_filters::list, for which its | |
440 | * caller provides necessary serialization. | |
441 | */ | |
442 | void (*addr_filters_sync) (struct perf_event *event); | |
443 | /* optional */ | |
444 | ||
66eb579e MR |
445 | /* |
446 | * Filter events for PMU-specific reasons. | |
447 | */ | |
448 | int (*filter_match) (struct perf_event *event); /* optional */ | |
621a01ea IM |
449 | }; |
450 | ||
375637bc AS |
451 | /** |
452 | * struct perf_addr_filter - address range filter definition | |
453 | * @entry: event's filter list linkage | |
454 | * @inode: object file's inode for file-based filters | |
455 | * @offset: filter range offset | |
456 | * @size: filter range size | |
457 | * @range: 1: range, 0: address | |
458 | * @filter: 1: filter/start, 0: stop | |
459 | * | |
460 | * This is a hardware-agnostic filter configuration as specified by the user. | |
461 | */ | |
462 | struct perf_addr_filter { | |
463 | struct list_head entry; | |
464 | struct inode *inode; | |
465 | unsigned long offset; | |
466 | unsigned long size; | |
467 | unsigned int range : 1, | |
468 | filter : 1; | |
469 | }; | |
470 | ||
471 | /** | |
472 | * struct perf_addr_filters_head - container for address range filters | |
473 | * @list: list of filters for this event | |
474 | * @lock: spinlock that serializes accesses to the @list and event's | |
475 | * (and its children's) filter generations. | |
6ce77bfd | 476 | * @nr_file_filters: number of file-based filters |
375637bc AS |
477 | * |
478 | * A child event will use parent's @list (and therefore @lock), so they are | |
479 | * bundled together; see perf_event_addr_filters(). | |
480 | */ | |
481 | struct perf_addr_filters_head { | |
482 | struct list_head list; | |
483 | raw_spinlock_t lock; | |
6ce77bfd | 484 | unsigned int nr_file_filters; |
375637bc AS |
485 | }; |
486 | ||
6a930700 | 487 | /** |
cdd6c482 | 488 | * enum perf_event_active_state - the states of a event |
6a930700 | 489 | */ |
cdd6c482 | 490 | enum perf_event_active_state { |
a69b0ca4 | 491 | PERF_EVENT_STATE_DEAD = -4, |
179033b3 | 492 | PERF_EVENT_STATE_EXIT = -3, |
57c0c15b | 493 | PERF_EVENT_STATE_ERROR = -2, |
cdd6c482 IM |
494 | PERF_EVENT_STATE_OFF = -1, |
495 | PERF_EVENT_STATE_INACTIVE = 0, | |
57c0c15b | 496 | PERF_EVENT_STATE_ACTIVE = 1, |
6a930700 IM |
497 | }; |
498 | ||
9b51f66d | 499 | struct file; |
453f19ee PZ |
500 | struct perf_sample_data; |
501 | ||
a8b0ca17 | 502 | typedef void (*perf_overflow_handler_t)(struct perf_event *, |
b326e956 FW |
503 | struct perf_sample_data *, |
504 | struct pt_regs *regs); | |
505 | ||
4ff6a8de DCC |
506 | /* |
507 | * Event capabilities. For event_caps and groups caps. | |
508 | * | |
509 | * PERF_EV_CAP_SOFTWARE: Is a software event. | |
d6a2f903 DCC |
510 | * PERF_EV_CAP_READ_ACTIVE_PKG: A CPU event (or cgroup event) that can be read |
511 | * from any CPU in the package where it is active. | |
4ff6a8de DCC |
512 | */ |
513 | #define PERF_EV_CAP_SOFTWARE BIT(0) | |
d6a2f903 | 514 | #define PERF_EV_CAP_READ_ACTIVE_PKG BIT(1) |
d6f962b5 | 515 | |
e7e7ee2e IM |
516 | #define SWEVENT_HLIST_BITS 8 |
517 | #define SWEVENT_HLIST_SIZE (1 << SWEVENT_HLIST_BITS) | |
76e1d904 FW |
518 | |
519 | struct swevent_hlist { | |
e7e7ee2e IM |
520 | struct hlist_head heads[SWEVENT_HLIST_SIZE]; |
521 | struct rcu_head rcu_head; | |
76e1d904 FW |
522 | }; |
523 | ||
8a49542c PZ |
524 | #define PERF_ATTACH_CONTEXT 0x01 |
525 | #define PERF_ATTACH_GROUP 0x02 | |
d580ff86 | 526 | #define PERF_ATTACH_TASK 0x04 |
4af57ef2 | 527 | #define PERF_ATTACH_TASK_DATA 0x08 |
8d4e6c4c | 528 | #define PERF_ATTACH_ITRACE 0x10 |
8a49542c | 529 | |
877c6856 | 530 | struct perf_cgroup; |
76369139 FW |
531 | struct ring_buffer; |
532 | ||
f2fb6bef KL |
533 | struct pmu_event_list { |
534 | raw_spinlock_t lock; | |
535 | struct list_head list; | |
536 | }; | |
537 | ||
0793a61d | 538 | /** |
cdd6c482 | 539 | * struct perf_event - performance event kernel representation: |
0793a61d | 540 | */ |
cdd6c482 IM |
541 | struct perf_event { |
542 | #ifdef CONFIG_PERF_EVENTS | |
9886167d PZ |
543 | /* |
544 | * entry onto perf_event_context::event_list; | |
545 | * modifications require ctx->lock | |
546 | * RCU safe iterations. | |
547 | */ | |
592903cd | 548 | struct list_head event_entry; |
9886167d PZ |
549 | |
550 | /* | |
551 | * XXX: group_entry and sibling_list should be mutually exclusive; | |
552 | * either you're a sibling on a group, or you're the group leader. | |
553 | * Rework the code to always use the same list element. | |
554 | * | |
555 | * Locked for modification by both ctx->mutex and ctx->lock; holding | |
556 | * either sufficies for read. | |
557 | */ | |
558 | struct list_head group_entry; | |
04289bb9 | 559 | struct list_head sibling_list; |
9886167d PZ |
560 | |
561 | /* | |
562 | * We need storage to track the entries in perf_pmu_migrate_context; we | |
563 | * cannot use the event_entry because of RCU and we want to keep the | |
564 | * group in tact which avoids us using the other two entries. | |
565 | */ | |
566 | struct list_head migrate_entry; | |
567 | ||
f3ae75de SE |
568 | struct hlist_node hlist_entry; |
569 | struct list_head active_entry; | |
0127c3ea | 570 | int nr_siblings; |
4ff6a8de DCC |
571 | |
572 | /* Not serialized. Only written during event initialization. */ | |
573 | int event_caps; | |
574 | /* The cumulative AND of all event_caps for events in this group. */ | |
575 | int group_caps; | |
576 | ||
cdd6c482 | 577 | struct perf_event *group_leader; |
a4eaf7f1 | 578 | struct pmu *pmu; |
54d751d4 | 579 | void *pmu_private; |
04289bb9 | 580 | |
cdd6c482 | 581 | enum perf_event_active_state state; |
8a49542c | 582 | unsigned int attach_state; |
e7850595 | 583 | local64_t count; |
a6e6dea6 | 584 | atomic64_t child_count; |
ee06094f | 585 | |
53cfbf59 | 586 | /* |
cdd6c482 | 587 | * These are the total time in nanoseconds that the event |
53cfbf59 | 588 | * has been enabled (i.e. eligible to run, and the task has |
cdd6c482 | 589 | * been scheduled in, if this is a per-task event) |
53cfbf59 PM |
590 | * and running (scheduled onto the CPU), respectively. |
591 | * | |
592 | * They are computed from tstamp_enabled, tstamp_running and | |
cdd6c482 | 593 | * tstamp_stopped when the event is in INACTIVE or ACTIVE state. |
53cfbf59 PM |
594 | */ |
595 | u64 total_time_enabled; | |
596 | u64 total_time_running; | |
597 | ||
598 | /* | |
599 | * These are timestamps used for computing total_time_enabled | |
cdd6c482 | 600 | * and total_time_running when the event is in INACTIVE or |
53cfbf59 PM |
601 | * ACTIVE state, measured in nanoseconds from an arbitrary point |
602 | * in time. | |
cdd6c482 IM |
603 | * tstamp_enabled: the notional time when the event was enabled |
604 | * tstamp_running: the notional time when the event was scheduled on | |
53cfbf59 | 605 | * tstamp_stopped: in INACTIVE state, the notional time when the |
cdd6c482 | 606 | * event was scheduled off. |
53cfbf59 PM |
607 | */ |
608 | u64 tstamp_enabled; | |
609 | u64 tstamp_running; | |
610 | u64 tstamp_stopped; | |
611 | ||
eed01528 SE |
612 | /* |
613 | * timestamp shadows the actual context timing but it can | |
614 | * be safely used in NMI interrupt context. It reflects the | |
615 | * context time as it was when the event was last scheduled in. | |
616 | * | |
617 | * ctx_time already accounts for ctx->timestamp. Therefore to | |
618 | * compute ctx_time for a sample, simply add perf_clock(). | |
619 | */ | |
620 | u64 shadow_ctx_time; | |
621 | ||
24f1e32c | 622 | struct perf_event_attr attr; |
c320c7b7 | 623 | u16 header_size; |
6844c09d | 624 | u16 id_header_size; |
c320c7b7 | 625 | u16 read_size; |
cdd6c482 | 626 | struct hw_perf_event hw; |
0793a61d | 627 | |
cdd6c482 | 628 | struct perf_event_context *ctx; |
a6fa941d | 629 | atomic_long_t refcount; |
0793a61d | 630 | |
53cfbf59 PM |
631 | /* |
632 | * These accumulate total time (in nanoseconds) that children | |
cdd6c482 | 633 | * events have been enabled and running, respectively. |
53cfbf59 PM |
634 | */ |
635 | atomic64_t child_total_time_enabled; | |
636 | atomic64_t child_total_time_running; | |
637 | ||
0793a61d | 638 | /* |
d859e29f | 639 | * Protect attach/detach and child_list: |
0793a61d | 640 | */ |
fccc714b PZ |
641 | struct mutex child_mutex; |
642 | struct list_head child_list; | |
cdd6c482 | 643 | struct perf_event *parent; |
0793a61d TG |
644 | |
645 | int oncpu; | |
646 | int cpu; | |
647 | ||
082ff5a2 PZ |
648 | struct list_head owner_entry; |
649 | struct task_struct *owner; | |
650 | ||
7b732a75 PZ |
651 | /* mmap bits */ |
652 | struct mutex mmap_mutex; | |
653 | atomic_t mmap_count; | |
26cb63ad | 654 | |
76369139 | 655 | struct ring_buffer *rb; |
10c6db11 | 656 | struct list_head rb_entry; |
b69cf536 PZ |
657 | unsigned long rcu_batches; |
658 | int rcu_pending; | |
37d81828 | 659 | |
7b732a75 | 660 | /* poll related */ |
0793a61d | 661 | wait_queue_head_t waitq; |
3c446b3d | 662 | struct fasync_struct *fasync; |
79f14641 PZ |
663 | |
664 | /* delayed work for NMIs and such */ | |
665 | int pending_wakeup; | |
4c9e2542 | 666 | int pending_kill; |
79f14641 | 667 | int pending_disable; |
e360adbe | 668 | struct irq_work pending; |
592903cd | 669 | |
79f14641 PZ |
670 | atomic_t event_limit; |
671 | ||
375637bc AS |
672 | /* address range filters */ |
673 | struct perf_addr_filters_head addr_filters; | |
674 | /* vma address array for file-based filders */ | |
675 | unsigned long *addr_filters_offs; | |
676 | unsigned long addr_filters_gen; | |
677 | ||
cdd6c482 | 678 | void (*destroy)(struct perf_event *); |
592903cd | 679 | struct rcu_head rcu_head; |
709e50cf PZ |
680 | |
681 | struct pid_namespace *ns; | |
8e5799b1 | 682 | u64 id; |
6fb2915d | 683 | |
34f43927 | 684 | u64 (*clock)(void); |
b326e956 | 685 | perf_overflow_handler_t overflow_handler; |
4dc0da86 | 686 | void *overflow_handler_context; |
aa6a5f3c AS |
687 | #ifdef CONFIG_BPF_SYSCALL |
688 | perf_overflow_handler_t orig_overflow_handler; | |
689 | struct bpf_prog *prog; | |
690 | #endif | |
453f19ee | 691 | |
07b139c8 | 692 | #ifdef CONFIG_EVENT_TRACING |
2425bcb9 | 693 | struct trace_event_call *tp_event; |
6fb2915d | 694 | struct event_filter *filter; |
ced39002 JO |
695 | #ifdef CONFIG_FUNCTION_TRACER |
696 | struct ftrace_ops ftrace_ops; | |
697 | #endif | |
ee06094f | 698 | #endif |
6fb2915d | 699 | |
e5d1367f SE |
700 | #ifdef CONFIG_CGROUP_PERF |
701 | struct perf_cgroup *cgrp; /* cgroup event is attach to */ | |
702 | int cgrp_defer_enabled; | |
703 | #endif | |
704 | ||
f2fb6bef | 705 | struct list_head sb_list; |
6fb2915d | 706 | #endif /* CONFIG_PERF_EVENTS */ |
0793a61d TG |
707 | }; |
708 | ||
709 | /** | |
cdd6c482 | 710 | * struct perf_event_context - event context structure |
0793a61d | 711 | * |
cdd6c482 | 712 | * Used as a container for task events and CPU events as well: |
0793a61d | 713 | */ |
cdd6c482 | 714 | struct perf_event_context { |
108b02cf | 715 | struct pmu *pmu; |
0793a61d | 716 | /* |
cdd6c482 | 717 | * Protect the states of the events in the list, |
d859e29f | 718 | * nr_active, and the list: |
0793a61d | 719 | */ |
e625cce1 | 720 | raw_spinlock_t lock; |
d859e29f | 721 | /* |
cdd6c482 | 722 | * Protect the list of events. Locking either mutex or lock |
d859e29f PM |
723 | * is sufficient to ensure the list doesn't change; to change |
724 | * the list you need to lock both the mutex and the spinlock. | |
725 | */ | |
a308444c | 726 | struct mutex mutex; |
04289bb9 | 727 | |
2fde4f94 | 728 | struct list_head active_ctx_list; |
889ff015 FW |
729 | struct list_head pinned_groups; |
730 | struct list_head flexible_groups; | |
a308444c | 731 | struct list_head event_list; |
cdd6c482 | 732 | int nr_events; |
a308444c IM |
733 | int nr_active; |
734 | int is_active; | |
bfbd3381 | 735 | int nr_stat; |
0f5a2601 | 736 | int nr_freq; |
dddd3379 | 737 | int rotate_disable; |
a308444c IM |
738 | atomic_t refcount; |
739 | struct task_struct *task; | |
53cfbf59 PM |
740 | |
741 | /* | |
4af4998b | 742 | * Context clock, runs when context enabled. |
53cfbf59 | 743 | */ |
a308444c IM |
744 | u64 time; |
745 | u64 timestamp; | |
564c2b21 PM |
746 | |
747 | /* | |
748 | * These fields let us detect when two contexts have both | |
749 | * been cloned (inherited) from a common ancestor. | |
750 | */ | |
cdd6c482 | 751 | struct perf_event_context *parent_ctx; |
a308444c IM |
752 | u64 parent_gen; |
753 | u64 generation; | |
754 | int pin_count; | |
db4a8356 | 755 | #ifdef CONFIG_CGROUP_PERF |
d010b332 | 756 | int nr_cgroups; /* cgroup evts */ |
db4a8356 | 757 | #endif |
4af57ef2 | 758 | void *task_ctx_data; /* pmu specific data */ |
28009ce4 | 759 | struct rcu_head rcu_head; |
0793a61d TG |
760 | }; |
761 | ||
7ae07ea3 FW |
762 | /* |
763 | * Number of contexts where an event can trigger: | |
e7e7ee2e | 764 | * task, softirq, hardirq, nmi. |
7ae07ea3 FW |
765 | */ |
766 | #define PERF_NR_CONTEXTS 4 | |
767 | ||
0793a61d | 768 | /** |
cdd6c482 | 769 | * struct perf_event_cpu_context - per cpu event context structure |
0793a61d TG |
770 | */ |
771 | struct perf_cpu_context { | |
cdd6c482 IM |
772 | struct perf_event_context ctx; |
773 | struct perf_event_context *task_ctx; | |
0793a61d | 774 | int active_oncpu; |
3b6f9e5c | 775 | int exclusive; |
4cfafd30 PZ |
776 | |
777 | raw_spinlock_t hrtimer_lock; | |
9e630205 SE |
778 | struct hrtimer hrtimer; |
779 | ktime_t hrtimer_interval; | |
4cfafd30 PZ |
780 | unsigned int hrtimer_active; |
781 | ||
db4a8356 | 782 | #ifdef CONFIG_CGROUP_PERF |
e5d1367f | 783 | struct perf_cgroup *cgrp; |
058fe1c0 | 784 | struct list_head cgrp_cpuctx_entry; |
db4a8356 | 785 | #endif |
e48c1788 PZ |
786 | |
787 | struct list_head sched_cb_entry; | |
788 | int sched_cb_usage; | |
a63fbed7 TG |
789 | |
790 | int online; | |
0793a61d TG |
791 | }; |
792 | ||
5622f295 | 793 | struct perf_output_handle { |
57c0c15b | 794 | struct perf_event *event; |
76369139 | 795 | struct ring_buffer *rb; |
6d1acfd5 | 796 | unsigned long wakeup; |
5d967a8b | 797 | unsigned long size; |
f4c0b0aa | 798 | u64 aux_flags; |
fdc26706 AS |
799 | union { |
800 | void *addr; | |
801 | unsigned long head; | |
802 | }; | |
5d967a8b | 803 | int page; |
5622f295 MM |
804 | }; |
805 | ||
0515e599 AS |
806 | struct bpf_perf_event_data_kern { |
807 | struct pt_regs *regs; | |
808 | struct perf_sample_data *data; | |
809 | }; | |
810 | ||
39bed6cb MF |
811 | #ifdef CONFIG_CGROUP_PERF |
812 | ||
813 | /* | |
814 | * perf_cgroup_info keeps track of time_enabled for a cgroup. | |
815 | * This is a per-cpu dynamically allocated data structure. | |
816 | */ | |
817 | struct perf_cgroup_info { | |
818 | u64 time; | |
819 | u64 timestamp; | |
820 | }; | |
821 | ||
822 | struct perf_cgroup { | |
823 | struct cgroup_subsys_state css; | |
824 | struct perf_cgroup_info __percpu *info; | |
825 | }; | |
826 | ||
827 | /* | |
828 | * Must ensure cgroup is pinned (css_get) before calling | |
829 | * this function. In other words, we cannot call this function | |
830 | * if there is no cgroup event for the current CPU context. | |
831 | */ | |
832 | static inline struct perf_cgroup * | |
614e4c4e | 833 | perf_cgroup_from_task(struct task_struct *task, struct perf_event_context *ctx) |
39bed6cb | 834 | { |
614e4c4e SE |
835 | return container_of(task_css_check(task, perf_event_cgrp_id, |
836 | ctx ? lockdep_is_held(&ctx->lock) | |
837 | : true), | |
39bed6cb MF |
838 | struct perf_cgroup, css); |
839 | } | |
840 | #endif /* CONFIG_CGROUP_PERF */ | |
841 | ||
cdd6c482 | 842 | #ifdef CONFIG_PERF_EVENTS |
829b42dd | 843 | |
fdc26706 AS |
844 | extern void *perf_aux_output_begin(struct perf_output_handle *handle, |
845 | struct perf_event *event); | |
846 | extern void perf_aux_output_end(struct perf_output_handle *handle, | |
f4c0b0aa | 847 | unsigned long size); |
fdc26706 AS |
848 | extern int perf_aux_output_skip(struct perf_output_handle *handle, |
849 | unsigned long size); | |
850 | extern void *perf_get_aux(struct perf_output_handle *handle); | |
f4c0b0aa | 851 | extern void perf_aux_output_flag(struct perf_output_handle *handle, u64 flags); |
8d4e6c4c | 852 | extern void perf_event_itrace_started(struct perf_event *event); |
fdc26706 | 853 | |
03d8e80b | 854 | extern int perf_pmu_register(struct pmu *pmu, const char *name, int type); |
b0a873eb | 855 | extern void perf_pmu_unregister(struct pmu *pmu); |
621a01ea | 856 | |
3bf101ba | 857 | extern int perf_num_counters(void); |
84c79910 | 858 | extern const char *perf_pmu_name(void); |
ab0cce56 JO |
859 | extern void __perf_event_task_sched_in(struct task_struct *prev, |
860 | struct task_struct *task); | |
861 | extern void __perf_event_task_sched_out(struct task_struct *prev, | |
862 | struct task_struct *next); | |
cdd6c482 IM |
863 | extern int perf_event_init_task(struct task_struct *child); |
864 | extern void perf_event_exit_task(struct task_struct *child); | |
865 | extern void perf_event_free_task(struct task_struct *task); | |
4e231c79 | 866 | extern void perf_event_delayed_put(struct task_struct *task); |
e03e7ee3 | 867 | extern struct file *perf_event_get(unsigned int fd); |
ffe8690c | 868 | extern const struct perf_event_attr *perf_event_attrs(struct perf_event *event); |
cdd6c482 | 869 | extern void perf_event_print_debug(void); |
33696fc0 PZ |
870 | extern void perf_pmu_disable(struct pmu *pmu); |
871 | extern void perf_pmu_enable(struct pmu *pmu); | |
ba532500 YZ |
872 | extern void perf_sched_cb_dec(struct pmu *pmu); |
873 | extern void perf_sched_cb_inc(struct pmu *pmu); | |
cdd6c482 IM |
874 | extern int perf_event_task_disable(void); |
875 | extern int perf_event_task_enable(void); | |
26ca5c11 | 876 | extern int perf_event_refresh(struct perf_event *event, int refresh); |
cdd6c482 | 877 | extern void perf_event_update_userpage(struct perf_event *event); |
fb0459d7 AV |
878 | extern int perf_event_release_kernel(struct perf_event *event); |
879 | extern struct perf_event * | |
880 | perf_event_create_kernel_counter(struct perf_event_attr *attr, | |
881 | int cpu, | |
38a81da2 | 882 | struct task_struct *task, |
4dc0da86 AK |
883 | perf_overflow_handler_t callback, |
884 | void *context); | |
0cda4c02 YZ |
885 | extern void perf_pmu_migrate_context(struct pmu *pmu, |
886 | int src_cpu, int dst_cpu); | |
f91840a3 | 887 | int perf_event_read_local(struct perf_event *event, u64 *value); |
59ed446f PZ |
888 | extern u64 perf_event_read_value(struct perf_event *event, |
889 | u64 *enabled, u64 *running); | |
5c92d124 | 890 | |
d010b332 | 891 | |
df1a132b | 892 | struct perf_sample_data { |
2565711f PZ |
893 | /* |
894 | * Fields set by perf_sample_data_init(), group so as to | |
895 | * minimize the cachelines touched. | |
896 | */ | |
897 | u64 addr; | |
898 | struct perf_raw_record *raw; | |
899 | struct perf_branch_stack *br_stack; | |
900 | u64 period; | |
901 | u64 weight; | |
902 | u64 txn; | |
903 | union perf_mem_data_src data_src; | |
5622f295 | 904 | |
2565711f PZ |
905 | /* |
906 | * The other fields, optionally {set,used} by | |
907 | * perf_{prepare,output}_sample(). | |
908 | */ | |
909 | u64 type; | |
5622f295 MM |
910 | u64 ip; |
911 | struct { | |
912 | u32 pid; | |
913 | u32 tid; | |
914 | } tid_entry; | |
915 | u64 time; | |
5622f295 MM |
916 | u64 id; |
917 | u64 stream_id; | |
918 | struct { | |
919 | u32 cpu; | |
920 | u32 reserved; | |
921 | } cpu_entry; | |
5622f295 | 922 | struct perf_callchain_entry *callchain; |
88a7c26a AL |
923 | |
924 | /* | |
925 | * regs_user may point to task_pt_regs or to regs_user_copy, depending | |
926 | * on arch details. | |
927 | */ | |
60e2364e | 928 | struct perf_regs regs_user; |
88a7c26a AL |
929 | struct pt_regs regs_user_copy; |
930 | ||
60e2364e | 931 | struct perf_regs regs_intr; |
c5ebcedb | 932 | u64 stack_user_size; |
fc7ce9c7 KL |
933 | |
934 | u64 phys_addr; | |
2565711f | 935 | } ____cacheline_aligned; |
df1a132b | 936 | |
770eee1f SE |
937 | /* default value for data source */ |
938 | #define PERF_MEM_NA (PERF_MEM_S(OP, NA) |\ | |
939 | PERF_MEM_S(LVL, NA) |\ | |
940 | PERF_MEM_S(SNOOP, NA) |\ | |
941 | PERF_MEM_S(LOCK, NA) |\ | |
942 | PERF_MEM_S(TLB, NA)) | |
943 | ||
fd0d000b RR |
944 | static inline void perf_sample_data_init(struct perf_sample_data *data, |
945 | u64 addr, u64 period) | |
dc1d628a | 946 | { |
fd0d000b | 947 | /* remaining struct members initialized in perf_prepare_sample() */ |
dc1d628a PZ |
948 | data->addr = addr; |
949 | data->raw = NULL; | |
bce38cd5 | 950 | data->br_stack = NULL; |
4018994f | 951 | data->period = period; |
c3feedf2 | 952 | data->weight = 0; |
770eee1f | 953 | data->data_src.val = PERF_MEM_NA; |
fdfbbd07 | 954 | data->txn = 0; |
dc1d628a PZ |
955 | } |
956 | ||
5622f295 MM |
957 | extern void perf_output_sample(struct perf_output_handle *handle, |
958 | struct perf_event_header *header, | |
959 | struct perf_sample_data *data, | |
cdd6c482 | 960 | struct perf_event *event); |
5622f295 MM |
961 | extern void perf_prepare_sample(struct perf_event_header *header, |
962 | struct perf_sample_data *data, | |
cdd6c482 | 963 | struct perf_event *event, |
5622f295 MM |
964 | struct pt_regs *regs); |
965 | ||
a8b0ca17 | 966 | extern int perf_event_overflow(struct perf_event *event, |
5622f295 MM |
967 | struct perf_sample_data *data, |
968 | struct pt_regs *regs); | |
df1a132b | 969 | |
9ecda41a WN |
970 | extern void perf_event_output_forward(struct perf_event *event, |
971 | struct perf_sample_data *data, | |
972 | struct pt_regs *regs); | |
973 | extern void perf_event_output_backward(struct perf_event *event, | |
974 | struct perf_sample_data *data, | |
975 | struct pt_regs *regs); | |
21509084 | 976 | extern void perf_event_output(struct perf_event *event, |
9ecda41a WN |
977 | struct perf_sample_data *data, |
978 | struct pt_regs *regs); | |
21509084 | 979 | |
1879445d WN |
980 | static inline bool |
981 | is_default_overflow_handler(struct perf_event *event) | |
982 | { | |
9ecda41a WN |
983 | if (likely(event->overflow_handler == perf_event_output_forward)) |
984 | return true; | |
985 | if (unlikely(event->overflow_handler == perf_event_output_backward)) | |
986 | return true; | |
987 | return false; | |
1879445d WN |
988 | } |
989 | ||
21509084 YZ |
990 | extern void |
991 | perf_event_header__init_id(struct perf_event_header *header, | |
992 | struct perf_sample_data *data, | |
993 | struct perf_event *event); | |
994 | extern void | |
995 | perf_event__output_id_sample(struct perf_event *event, | |
996 | struct perf_output_handle *handle, | |
997 | struct perf_sample_data *sample); | |
998 | ||
f38b0dbb KL |
999 | extern void |
1000 | perf_log_lost_samples(struct perf_event *event, u64 lost); | |
1001 | ||
6c7e550f FBH |
1002 | static inline bool is_sampling_event(struct perf_event *event) |
1003 | { | |
1004 | return event->attr.sample_period != 0; | |
1005 | } | |
1006 | ||
3b6f9e5c | 1007 | /* |
cdd6c482 | 1008 | * Return 1 for a software event, 0 for a hardware event |
3b6f9e5c | 1009 | */ |
cdd6c482 | 1010 | static inline int is_software_event(struct perf_event *event) |
3b6f9e5c | 1011 | { |
4ff6a8de | 1012 | return event->event_caps & PERF_EV_CAP_SOFTWARE; |
3b6f9e5c PM |
1013 | } |
1014 | ||
c5905afb | 1015 | extern struct static_key perf_swevent_enabled[PERF_COUNT_SW_MAX]; |
f29ac756 | 1016 | |
86038c5e | 1017 | extern void ___perf_sw_event(u32, u64, struct pt_regs *, u64); |
a8b0ca17 | 1018 | extern void __perf_sw_event(u32, u64, struct pt_regs *, u64); |
f29ac756 | 1019 | |
b0f82b81 | 1020 | #ifndef perf_arch_fetch_caller_regs |
e7e7ee2e | 1021 | static inline void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip) { } |
b0f82b81 | 1022 | #endif |
5331d7b8 FW |
1023 | |
1024 | /* | |
1025 | * Take a snapshot of the regs. Skip ip and frame pointer to | |
1026 | * the nth caller. We only need a few of the regs: | |
1027 | * - ip for PERF_SAMPLE_IP | |
1028 | * - cs for user_mode() tests | |
1029 | * - bp for callchains | |
1030 | * - eflags, for future purposes, just in case | |
1031 | */ | |
b0f82b81 | 1032 | static inline void perf_fetch_caller_regs(struct pt_regs *regs) |
5331d7b8 | 1033 | { |
b0f82b81 | 1034 | perf_arch_fetch_caller_regs(regs, CALLER_ADDR0); |
5331d7b8 FW |
1035 | } |
1036 | ||
7e54a5a0 | 1037 | static __always_inline void |
a8b0ca17 | 1038 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) |
e49a5bd3 | 1039 | { |
86038c5e PZI |
1040 | if (static_key_false(&perf_swevent_enabled[event_id])) |
1041 | __perf_sw_event(event_id, nr, regs, addr); | |
1042 | } | |
1043 | ||
1044 | DECLARE_PER_CPU(struct pt_regs, __perf_regs[4]); | |
7e54a5a0 | 1045 | |
86038c5e PZI |
1046 | /* |
1047 | * 'Special' version for the scheduler, it hard assumes no recursion, | |
1048 | * which is guaranteed by us not actually scheduling inside other swevents | |
1049 | * because those disable preemption. | |
1050 | */ | |
1051 | static __always_inline void | |
1052 | perf_sw_event_sched(u32 event_id, u64 nr, u64 addr) | |
1053 | { | |
c5905afb | 1054 | if (static_key_false(&perf_swevent_enabled[event_id])) { |
86038c5e PZI |
1055 | struct pt_regs *regs = this_cpu_ptr(&__perf_regs[0]); |
1056 | ||
1057 | perf_fetch_caller_regs(regs); | |
1058 | ___perf_sw_event(event_id, nr, regs, addr); | |
e49a5bd3 FW |
1059 | } |
1060 | } | |
1061 | ||
9107c89e | 1062 | extern struct static_key_false perf_sched_events; |
ee6dcfa4 | 1063 | |
ff303e66 PZ |
1064 | static __always_inline bool |
1065 | perf_sw_migrate_enabled(void) | |
1066 | { | |
1067 | if (static_key_false(&perf_swevent_enabled[PERF_COUNT_SW_CPU_MIGRATIONS])) | |
1068 | return true; | |
1069 | return false; | |
1070 | } | |
1071 | ||
1072 | static inline void perf_event_task_migrate(struct task_struct *task) | |
1073 | { | |
1074 | if (perf_sw_migrate_enabled()) | |
1075 | task->sched_migrated = 1; | |
1076 | } | |
1077 | ||
ab0cce56 | 1078 | static inline void perf_event_task_sched_in(struct task_struct *prev, |
a8d757ef | 1079 | struct task_struct *task) |
ab0cce56 | 1080 | { |
9107c89e | 1081 | if (static_branch_unlikely(&perf_sched_events)) |
ab0cce56 | 1082 | __perf_event_task_sched_in(prev, task); |
ff303e66 PZ |
1083 | |
1084 | if (perf_sw_migrate_enabled() && task->sched_migrated) { | |
1085 | struct pt_regs *regs = this_cpu_ptr(&__perf_regs[0]); | |
1086 | ||
1087 | perf_fetch_caller_regs(regs); | |
1088 | ___perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, regs, 0); | |
1089 | task->sched_migrated = 0; | |
1090 | } | |
ab0cce56 JO |
1091 | } |
1092 | ||
1093 | static inline void perf_event_task_sched_out(struct task_struct *prev, | |
1094 | struct task_struct *next) | |
ee6dcfa4 | 1095 | { |
86038c5e | 1096 | perf_sw_event_sched(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 0); |
ee6dcfa4 | 1097 | |
9107c89e | 1098 | if (static_branch_unlikely(&perf_sched_events)) |
ab0cce56 | 1099 | __perf_event_task_sched_out(prev, next); |
ee6dcfa4 PZ |
1100 | } |
1101 | ||
3af9e859 | 1102 | extern void perf_event_mmap(struct vm_area_struct *vma); |
39447b38 | 1103 | extern struct perf_guest_info_callbacks *perf_guest_cbs; |
dcf46b94 ZY |
1104 | extern int perf_register_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); |
1105 | extern int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *callbacks); | |
39447b38 | 1106 | |
e041e328 | 1107 | extern void perf_event_exec(void); |
82b89778 | 1108 | extern void perf_event_comm(struct task_struct *tsk, bool exec); |
e4222673 | 1109 | extern void perf_event_namespaces(struct task_struct *tsk); |
cdd6c482 | 1110 | extern void perf_event_fork(struct task_struct *tsk); |
8d1b2d93 | 1111 | |
56962b44 FW |
1112 | /* Callchains */ |
1113 | DECLARE_PER_CPU(struct perf_callchain_entry, perf_callchain_entry); | |
1114 | ||
cfbcf468 ACM |
1115 | extern void perf_callchain_user(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs); |
1116 | extern void perf_callchain_kernel(struct perf_callchain_entry_ctx *entry, struct pt_regs *regs); | |
568b329a AS |
1117 | extern struct perf_callchain_entry * |
1118 | get_perf_callchain(struct pt_regs *regs, u32 init_nr, bool kernel, bool user, | |
cfbcf468 | 1119 | u32 max_stack, bool crosstask, bool add_mark); |
97c79a38 | 1120 | extern int get_callchain_buffers(int max_stack); |
568b329a | 1121 | extern void put_callchain_buffers(void); |
394ee076 | 1122 | |
c5dfd78e | 1123 | extern int sysctl_perf_event_max_stack; |
c85b0334 | 1124 | extern int sysctl_perf_event_max_contexts_per_stack; |
c5dfd78e | 1125 | |
c85b0334 ACM |
1126 | static inline int perf_callchain_store_context(struct perf_callchain_entry_ctx *ctx, u64 ip) |
1127 | { | |
1128 | if (ctx->contexts < sysctl_perf_event_max_contexts_per_stack) { | |
1129 | struct perf_callchain_entry *entry = ctx->entry; | |
1130 | entry->ip[entry->nr++] = ip; | |
1131 | ++ctx->contexts; | |
1132 | return 0; | |
1133 | } else { | |
1134 | ctx->contexts_maxed = true; | |
1135 | return -1; /* no more room, stop walking the stack */ | |
1136 | } | |
1137 | } | |
3e4de4ec | 1138 | |
cfbcf468 | 1139 | static inline int perf_callchain_store(struct perf_callchain_entry_ctx *ctx, u64 ip) |
70791ce9 | 1140 | { |
c85b0334 | 1141 | if (ctx->nr < ctx->max_stack && !ctx->contexts_maxed) { |
3b1fff08 | 1142 | struct perf_callchain_entry *entry = ctx->entry; |
70791ce9 | 1143 | entry->ip[entry->nr++] = ip; |
3b1fff08 | 1144 | ++ctx->nr; |
568b329a AS |
1145 | return 0; |
1146 | } else { | |
1147 | return -1; /* no more room, stop walking the stack */ | |
1148 | } | |
70791ce9 | 1149 | } |
394ee076 | 1150 | |
cdd6c482 IM |
1151 | extern int sysctl_perf_event_paranoid; |
1152 | extern int sysctl_perf_event_mlock; | |
1153 | extern int sysctl_perf_event_sample_rate; | |
14c63f17 DH |
1154 | extern int sysctl_perf_cpu_time_max_percent; |
1155 | ||
1156 | extern void perf_sample_event_took(u64 sample_len_ns); | |
1ccd1549 | 1157 | |
163ec435 PZ |
1158 | extern int perf_proc_update_handler(struct ctl_table *table, int write, |
1159 | void __user *buffer, size_t *lenp, | |
1160 | loff_t *ppos); | |
14c63f17 DH |
1161 | extern int perf_cpu_time_max_percent_handler(struct ctl_table *table, int write, |
1162 | void __user *buffer, size_t *lenp, | |
1163 | loff_t *ppos); | |
1164 | ||
c5dfd78e ACM |
1165 | int perf_event_max_stack_handler(struct ctl_table *table, int write, |
1166 | void __user *buffer, size_t *lenp, loff_t *ppos); | |
163ec435 | 1167 | |
320ebf09 PZ |
1168 | static inline bool perf_paranoid_tracepoint_raw(void) |
1169 | { | |
1170 | return sysctl_perf_event_paranoid > -1; | |
1171 | } | |
1172 | ||
1173 | static inline bool perf_paranoid_cpu(void) | |
1174 | { | |
1175 | return sysctl_perf_event_paranoid > 0; | |
1176 | } | |
1177 | ||
1178 | static inline bool perf_paranoid_kernel(void) | |
1179 | { | |
1180 | return sysctl_perf_event_paranoid > 1; | |
1181 | } | |
1182 | ||
cdd6c482 | 1183 | extern void perf_event_init(void); |
1e1dcd93 | 1184 | extern void perf_tp_event(u16 event_type, u64 count, void *record, |
1c024eca | 1185 | int entry_size, struct pt_regs *regs, |
e6dab5ff | 1186 | struct hlist_head *head, int rctx, |
75e83876 | 1187 | struct task_struct *task, struct perf_event *event); |
24f1e32c | 1188 | extern void perf_bp_event(struct perf_event *event, void *data); |
0d905bca | 1189 | |
9d23a90a | 1190 | #ifndef perf_misc_flags |
e7e7ee2e IM |
1191 | # define perf_misc_flags(regs) \ |
1192 | (user_mode(regs) ? PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL) | |
1193 | # define perf_instruction_pointer(regs) instruction_pointer(regs) | |
9d23a90a PM |
1194 | #endif |
1195 | ||
bce38cd5 SE |
1196 | static inline bool has_branch_stack(struct perf_event *event) |
1197 | { | |
1198 | return event->attr.sample_type & PERF_SAMPLE_BRANCH_STACK; | |
a46a2300 YZ |
1199 | } |
1200 | ||
1201 | static inline bool needs_branch_stack(struct perf_event *event) | |
1202 | { | |
1203 | return event->attr.branch_sample_type != 0; | |
bce38cd5 SE |
1204 | } |
1205 | ||
45bfb2e5 PZ |
1206 | static inline bool has_aux(struct perf_event *event) |
1207 | { | |
1208 | return event->pmu->setup_aux; | |
1209 | } | |
1210 | ||
9ecda41a WN |
1211 | static inline bool is_write_backward(struct perf_event *event) |
1212 | { | |
1213 | return !!event->attr.write_backward; | |
1214 | } | |
1215 | ||
375637bc AS |
1216 | static inline bool has_addr_filter(struct perf_event *event) |
1217 | { | |
1218 | return event->pmu->nr_addr_filters; | |
1219 | } | |
1220 | ||
1221 | /* | |
1222 | * An inherited event uses parent's filters | |
1223 | */ | |
1224 | static inline struct perf_addr_filters_head * | |
1225 | perf_event_addr_filters(struct perf_event *event) | |
1226 | { | |
1227 | struct perf_addr_filters_head *ifh = &event->addr_filters; | |
1228 | ||
1229 | if (event->parent) | |
1230 | ifh = &event->parent->addr_filters; | |
1231 | ||
1232 | return ifh; | |
1233 | } | |
1234 | ||
1235 | extern void perf_event_addr_filters_sync(struct perf_event *event); | |
1236 | ||
5622f295 | 1237 | extern int perf_output_begin(struct perf_output_handle *handle, |
a7ac67ea | 1238 | struct perf_event *event, unsigned int size); |
9ecda41a WN |
1239 | extern int perf_output_begin_forward(struct perf_output_handle *handle, |
1240 | struct perf_event *event, | |
1241 | unsigned int size); | |
1242 | extern int perf_output_begin_backward(struct perf_output_handle *handle, | |
1243 | struct perf_event *event, | |
1244 | unsigned int size); | |
1245 | ||
5622f295 | 1246 | extern void perf_output_end(struct perf_output_handle *handle); |
91d7753a | 1247 | extern unsigned int perf_output_copy(struct perf_output_handle *handle, |
5622f295 | 1248 | const void *buf, unsigned int len); |
5685e0ff JO |
1249 | extern unsigned int perf_output_skip(struct perf_output_handle *handle, |
1250 | unsigned int len); | |
4ed7c92d PZ |
1251 | extern int perf_swevent_get_recursion_context(void); |
1252 | extern void perf_swevent_put_recursion_context(int rctx); | |
ab573844 | 1253 | extern u64 perf_swevent_set_period(struct perf_event *event); |
44234adc FW |
1254 | extern void perf_event_enable(struct perf_event *event); |
1255 | extern void perf_event_disable(struct perf_event *event); | |
fae3fde6 | 1256 | extern void perf_event_disable_local(struct perf_event *event); |
5aab90ce | 1257 | extern void perf_event_disable_inatomic(struct perf_event *event); |
e9d2b064 | 1258 | extern void perf_event_task_tick(void); |
475113d9 | 1259 | extern int perf_event_account_interrupt(struct perf_event *event); |
e041e328 | 1260 | #else /* !CONFIG_PERF_EVENTS: */ |
fdc26706 AS |
1261 | static inline void * |
1262 | perf_aux_output_begin(struct perf_output_handle *handle, | |
1263 | struct perf_event *event) { return NULL; } | |
1264 | static inline void | |
f4c0b0aa WD |
1265 | perf_aux_output_end(struct perf_output_handle *handle, unsigned long size) |
1266 | { } | |
fdc26706 AS |
1267 | static inline int |
1268 | perf_aux_output_skip(struct perf_output_handle *handle, | |
1269 | unsigned long size) { return -EINVAL; } | |
1270 | static inline void * | |
1271 | perf_get_aux(struct perf_output_handle *handle) { return NULL; } | |
0793a61d | 1272 | static inline void |
ff303e66 PZ |
1273 | perf_event_task_migrate(struct task_struct *task) { } |
1274 | static inline void | |
ab0cce56 JO |
1275 | perf_event_task_sched_in(struct task_struct *prev, |
1276 | struct task_struct *task) { } | |
1277 | static inline void | |
1278 | perf_event_task_sched_out(struct task_struct *prev, | |
1279 | struct task_struct *next) { } | |
cdd6c482 IM |
1280 | static inline int perf_event_init_task(struct task_struct *child) { return 0; } |
1281 | static inline void perf_event_exit_task(struct task_struct *child) { } | |
1282 | static inline void perf_event_free_task(struct task_struct *task) { } | |
4e231c79 | 1283 | static inline void perf_event_delayed_put(struct task_struct *task) { } |
e03e7ee3 | 1284 | static inline struct file *perf_event_get(unsigned int fd) { return ERR_PTR(-EINVAL); } |
ffe8690c KX |
1285 | static inline const struct perf_event_attr *perf_event_attrs(struct perf_event *event) |
1286 | { | |
1287 | return ERR_PTR(-EINVAL); | |
1288 | } | |
f91840a3 AS |
1289 | static inline int perf_event_read_local(struct perf_event *event, u64 *value) |
1290 | { | |
1291 | return -EINVAL; | |
1292 | } | |
57c0c15b | 1293 | static inline void perf_event_print_debug(void) { } |
57c0c15b IM |
1294 | static inline int perf_event_task_disable(void) { return -EINVAL; } |
1295 | static inline int perf_event_task_enable(void) { return -EINVAL; } | |
26ca5c11 AK |
1296 | static inline int perf_event_refresh(struct perf_event *event, int refresh) |
1297 | { | |
1298 | return -EINVAL; | |
1299 | } | |
15dbf27c | 1300 | |
925d519a | 1301 | static inline void |
a8b0ca17 | 1302 | perf_sw_event(u32 event_id, u64 nr, struct pt_regs *regs, u64 addr) { } |
24f1e32c | 1303 | static inline void |
86038c5e PZI |
1304 | perf_sw_event_sched(u32 event_id, u64 nr, u64 addr) { } |
1305 | static inline void | |
184f412c | 1306 | perf_bp_event(struct perf_event *event, void *data) { } |
0a4a9391 | 1307 | |
39447b38 | 1308 | static inline int perf_register_guest_info_callbacks |
e7e7ee2e | 1309 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 1310 | static inline int perf_unregister_guest_info_callbacks |
e7e7ee2e | 1311 | (struct perf_guest_info_callbacks *callbacks) { return 0; } |
39447b38 | 1312 | |
57c0c15b | 1313 | static inline void perf_event_mmap(struct vm_area_struct *vma) { } |
e041e328 | 1314 | static inline void perf_event_exec(void) { } |
82b89778 | 1315 | static inline void perf_event_comm(struct task_struct *tsk, bool exec) { } |
e4222673 | 1316 | static inline void perf_event_namespaces(struct task_struct *tsk) { } |
cdd6c482 IM |
1317 | static inline void perf_event_fork(struct task_struct *tsk) { } |
1318 | static inline void perf_event_init(void) { } | |
184f412c | 1319 | static inline int perf_swevent_get_recursion_context(void) { return -1; } |
4ed7c92d | 1320 | static inline void perf_swevent_put_recursion_context(int rctx) { } |
ab573844 | 1321 | static inline u64 perf_swevent_set_period(struct perf_event *event) { return 0; } |
44234adc FW |
1322 | static inline void perf_event_enable(struct perf_event *event) { } |
1323 | static inline void perf_event_disable(struct perf_event *event) { } | |
500ad2d8 | 1324 | static inline int __perf_event_disable(void *info) { return -1; } |
e9d2b064 | 1325 | static inline void perf_event_task_tick(void) { } |
ffe8690c | 1326 | static inline int perf_event_release_kernel(struct perf_event *event) { return 0; } |
0793a61d TG |
1327 | #endif |
1328 | ||
6c4d3bc9 DR |
1329 | #if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_CPU_SUP_INTEL) |
1330 | extern void perf_restore_debug_store(void); | |
1331 | #else | |
1d9d8639 | 1332 | static inline void perf_restore_debug_store(void) { } |
0793a61d TG |
1333 | #endif |
1334 | ||
7e3f977e DB |
1335 | static __always_inline bool perf_raw_frag_last(const struct perf_raw_frag *frag) |
1336 | { | |
1337 | return frag->pad < sizeof(u64); | |
1338 | } | |
1339 | ||
e7e7ee2e | 1340 | #define perf_output_put(handle, x) perf_output_copy((handle), &(x), sizeof(x)) |
5622f295 | 1341 | |
2663960c SB |
1342 | struct perf_pmu_events_attr { |
1343 | struct device_attribute attr; | |
1344 | u64 id; | |
3a54aaa0 | 1345 | const char *event_str; |
2663960c SB |
1346 | }; |
1347 | ||
fc07e9f9 AK |
1348 | struct perf_pmu_events_ht_attr { |
1349 | struct device_attribute attr; | |
1350 | u64 id; | |
1351 | const char *event_str_ht; | |
1352 | const char *event_str_noht; | |
1353 | }; | |
1354 | ||
fd979c01 CS |
1355 | ssize_t perf_event_sysfs_show(struct device *dev, struct device_attribute *attr, |
1356 | char *page); | |
1357 | ||
2663960c SB |
1358 | #define PMU_EVENT_ATTR(_name, _var, _id, _show) \ |
1359 | static struct perf_pmu_events_attr _var = { \ | |
1360 | .attr = __ATTR(_name, 0444, _show, NULL), \ | |
1361 | .id = _id, \ | |
1362 | }; | |
1363 | ||
f0405b81 CS |
1364 | #define PMU_EVENT_ATTR_STRING(_name, _var, _str) \ |
1365 | static struct perf_pmu_events_attr _var = { \ | |
1366 | .attr = __ATTR(_name, 0444, perf_event_sysfs_show, NULL), \ | |
1367 | .id = 0, \ | |
1368 | .event_str = _str, \ | |
1369 | }; | |
1370 | ||
641cc938 JO |
1371 | #define PMU_FORMAT_ATTR(_name, _format) \ |
1372 | static ssize_t \ | |
1373 | _name##_show(struct device *dev, \ | |
1374 | struct device_attribute *attr, \ | |
1375 | char *page) \ | |
1376 | { \ | |
1377 | BUILD_BUG_ON(sizeof(_format) >= PAGE_SIZE); \ | |
1378 | return sprintf(page, _format "\n"); \ | |
1379 | } \ | |
1380 | \ | |
1381 | static struct device_attribute format_attr_##_name = __ATTR_RO(_name) | |
1382 | ||
00e16c3d TG |
1383 | /* Performance counter hotplug functions */ |
1384 | #ifdef CONFIG_PERF_EVENTS | |
1385 | int perf_event_init_cpu(unsigned int cpu); | |
1386 | int perf_event_exit_cpu(unsigned int cpu); | |
1387 | #else | |
1388 | #define perf_event_init_cpu NULL | |
1389 | #define perf_event_exit_cpu NULL | |
1390 | #endif | |
1391 | ||
cdd6c482 | 1392 | #endif /* _LINUX_PERF_EVENT_H */ |