]> bbs.cooldavid.org Git - net-next-2.6.git/blame - include/linux/perf_counter.h
Merge branch 'linus' into perfcounters/rename
[net-next-2.6.git] / include / linux / perf_counter.h
CommitLineData
0793a61d
TG
1/*
2 * Performance counters:
3 *
a308444c
IM
4 * Copyright (C) 2008-2009, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright (C) 2008-2009, Red Hat, Inc., Ingo Molnar
6 * Copyright (C) 2008-2009, Red Hat, Inc., Peter Zijlstra
0793a61d
TG
7 *
8 * Data type definitions, declarations, prototypes.
9 *
a308444c 10 * Started by: Thomas Gleixner and Ingo Molnar
0793a61d
TG
11 *
12 * For licencing details see kernel-base/COPYING
13 */
14#ifndef _LINUX_PERF_COUNTER_H
15#define _LINUX_PERF_COUNTER_H
16
f3dfd265
PM
17#include <linux/types.h>
18#include <linux/ioctl.h>
9aaa131a 19#include <asm/byteorder.h>
0793a61d
TG
20
21/*
9f66a381
IM
22 * User-space ABI bits:
23 */
24
25/*
0d48696f 26 * attr.type
0793a61d 27 */
1c432d89 28enum perf_type_id {
a308444c
IM
29 PERF_TYPE_HARDWARE = 0,
30 PERF_TYPE_SOFTWARE = 1,
31 PERF_TYPE_TRACEPOINT = 2,
32 PERF_TYPE_HW_CACHE = 3,
33 PERF_TYPE_RAW = 4,
b8e83514 34
a308444c 35 PERF_TYPE_MAX, /* non-ABI */
b8e83514 36};
6c594c21 37
b8e83514 38/*
a308444c
IM
39 * Generalized performance counter event types, used by the
40 * attr.event_id parameter of the sys_perf_counter_open()
41 * syscall:
b8e83514 42 */
1c432d89 43enum perf_hw_id {
9f66a381 44 /*
b8e83514 45 * Common hardware events, generalized by the kernel:
9f66a381 46 */
f4dbfa8f
PZ
47 PERF_COUNT_HW_CPU_CYCLES = 0,
48 PERF_COUNT_HW_INSTRUCTIONS = 1,
49 PERF_COUNT_HW_CACHE_REFERENCES = 2,
50 PERF_COUNT_HW_CACHE_MISSES = 3,
51 PERF_COUNT_HW_BRANCH_INSTRUCTIONS = 4,
52 PERF_COUNT_HW_BRANCH_MISSES = 5,
53 PERF_COUNT_HW_BUS_CYCLES = 6,
54
a308444c 55 PERF_COUNT_HW_MAX, /* non-ABI */
b8e83514 56};
e077df4f 57
8326f44d
IM
58/*
59 * Generalized hardware cache counters:
60 *
8be6e8f3 61 * { L1-D, L1-I, LLC, ITLB, DTLB, BPU } x
8326f44d
IM
62 * { read, write, prefetch } x
63 * { accesses, misses }
64 */
1c432d89 65enum perf_hw_cache_id {
a308444c
IM
66 PERF_COUNT_HW_CACHE_L1D = 0,
67 PERF_COUNT_HW_CACHE_L1I = 1,
68 PERF_COUNT_HW_CACHE_LL = 2,
69 PERF_COUNT_HW_CACHE_DTLB = 3,
70 PERF_COUNT_HW_CACHE_ITLB = 4,
71 PERF_COUNT_HW_CACHE_BPU = 5,
72
73 PERF_COUNT_HW_CACHE_MAX, /* non-ABI */
8326f44d
IM
74};
75
1c432d89 76enum perf_hw_cache_op_id {
a308444c
IM
77 PERF_COUNT_HW_CACHE_OP_READ = 0,
78 PERF_COUNT_HW_CACHE_OP_WRITE = 1,
79 PERF_COUNT_HW_CACHE_OP_PREFETCH = 2,
8326f44d 80
a308444c 81 PERF_COUNT_HW_CACHE_OP_MAX, /* non-ABI */
8326f44d
IM
82};
83
1c432d89
PZ
84enum perf_hw_cache_op_result_id {
85 PERF_COUNT_HW_CACHE_RESULT_ACCESS = 0,
86 PERF_COUNT_HW_CACHE_RESULT_MISS = 1,
8326f44d 87
a308444c 88 PERF_COUNT_HW_CACHE_RESULT_MAX, /* non-ABI */
8326f44d
IM
89};
90
b8e83514
PZ
91/*
92 * Special "software" counters provided by the kernel, even if the hardware
93 * does not support performance counters. These counters measure various
94 * physical and sw events of the kernel (and allow the profiling of them as
95 * well):
96 */
1c432d89 97enum perf_sw_ids {
a308444c
IM
98 PERF_COUNT_SW_CPU_CLOCK = 0,
99 PERF_COUNT_SW_TASK_CLOCK = 1,
100 PERF_COUNT_SW_PAGE_FAULTS = 2,
101 PERF_COUNT_SW_CONTEXT_SWITCHES = 3,
102 PERF_COUNT_SW_CPU_MIGRATIONS = 4,
103 PERF_COUNT_SW_PAGE_FAULTS_MIN = 5,
104 PERF_COUNT_SW_PAGE_FAULTS_MAJ = 6,
105
106 PERF_COUNT_SW_MAX, /* non-ABI */
0793a61d
TG
107};
108
8a057d84 109/*
0d48696f 110 * Bits that can be set in attr.sample_type to request information
8a057d84
PZ
111 * in the overflow packets.
112 */
b23f3325 113enum perf_counter_sample_format {
a308444c
IM
114 PERF_SAMPLE_IP = 1U << 0,
115 PERF_SAMPLE_TID = 1U << 1,
116 PERF_SAMPLE_TIME = 1U << 2,
117 PERF_SAMPLE_ADDR = 1U << 3,
3dab77fb 118 PERF_SAMPLE_READ = 1U << 4,
a308444c
IM
119 PERF_SAMPLE_CALLCHAIN = 1U << 5,
120 PERF_SAMPLE_ID = 1U << 6,
121 PERF_SAMPLE_CPU = 1U << 7,
122 PERF_SAMPLE_PERIOD = 1U << 8,
7f453c24 123 PERF_SAMPLE_STREAM_ID = 1U << 9,
3a43ce68 124 PERF_SAMPLE_RAW = 1U << 10,
974802ea 125
f413cdb8 126 PERF_SAMPLE_MAX = 1U << 11, /* non-ABI */
8a057d84
PZ
127};
128
53cfbf59 129/*
3dab77fb
PZ
130 * The format of the data returned by read() on a perf counter fd,
131 * as specified by attr.read_format:
132 *
133 * struct read_format {
134 * { u64 value;
135 * { u64 time_enabled; } && PERF_FORMAT_ENABLED
136 * { u64 time_running; } && PERF_FORMAT_RUNNING
137 * { u64 id; } && PERF_FORMAT_ID
138 * } && !PERF_FORMAT_GROUP
139 *
140 * { u64 nr;
141 * { u64 time_enabled; } && PERF_FORMAT_ENABLED
142 * { u64 time_running; } && PERF_FORMAT_RUNNING
143 * { u64 value;
144 * { u64 id; } && PERF_FORMAT_ID
145 * } cntr[nr];
146 * } && PERF_FORMAT_GROUP
147 * };
53cfbf59
PM
148 */
149enum perf_counter_read_format {
a308444c
IM
150 PERF_FORMAT_TOTAL_TIME_ENABLED = 1U << 0,
151 PERF_FORMAT_TOTAL_TIME_RUNNING = 1U << 1,
152 PERF_FORMAT_ID = 1U << 2,
3dab77fb 153 PERF_FORMAT_GROUP = 1U << 3,
974802ea 154
3dab77fb 155 PERF_FORMAT_MAX = 1U << 4, /* non-ABI */
53cfbf59
PM
156};
157
974802ea
PZ
158#define PERF_ATTR_SIZE_VER0 64 /* sizeof first published struct */
159
9f66a381
IM
160/*
161 * Hardware event to monitor via a performance monitoring counter:
162 */
0d48696f 163struct perf_counter_attr {
974802ea 164
f4a2deb4 165 /*
a21ca2ca
IM
166 * Major type: hardware/software/tracepoint/etc.
167 */
168 __u32 type;
974802ea
PZ
169
170 /*
171 * Size of the attr structure, for fwd/bwd compat.
172 */
173 __u32 size;
a21ca2ca
IM
174
175 /*
176 * Type specific configuration information.
f4a2deb4
PZ
177 */
178 __u64 config;
9f66a381 179
60db5e09 180 union {
b23f3325
PZ
181 __u64 sample_period;
182 __u64 sample_freq;
60db5e09
PZ
183 };
184
b23f3325
PZ
185 __u64 sample_type;
186 __u64 read_format;
9f66a381 187
2743a5b0 188 __u64 disabled : 1, /* off by default */
0475f9ea
PM
189 inherit : 1, /* children inherit it */
190 pinned : 1, /* must always be on PMU */
191 exclusive : 1, /* only group on PMU */
192 exclude_user : 1, /* don't count user */
193 exclude_kernel : 1, /* ditto kernel */
194 exclude_hv : 1, /* ditto hypervisor */
2743a5b0 195 exclude_idle : 1, /* don't count when idle */
0a4a9391 196 mmap : 1, /* include mmap data */
8d1b2d93 197 comm : 1, /* include comm data */
60db5e09 198 freq : 1, /* use freq, not period */
bfbd3381 199 inherit_stat : 1, /* per task counts */
57e7986e 200 enable_on_exec : 1, /* next exec enables */
9f498cc5 201 task : 1, /* trace fork/exit */
2667de81 202 watermark : 1, /* wakeup_watermark */
0475f9ea 203
2667de81 204 __reserved_1 : 49;
2743a5b0 205
2667de81
PZ
206 union {
207 __u32 wakeup_events; /* wakeup every n events */
208 __u32 wakeup_watermark; /* bytes before wakeup */
209 };
974802ea 210 __u32 __reserved_2;
9f66a381 211
974802ea 212 __u64 __reserved_3;
eab656ae
TG
213};
214
d859e29f
PM
215/*
216 * Ioctls that can be done on a perf counter fd:
217 */
08247e31
PZ
218#define PERF_COUNTER_IOC_ENABLE _IO ('$', 0)
219#define PERF_COUNTER_IOC_DISABLE _IO ('$', 1)
220#define PERF_COUNTER_IOC_REFRESH _IO ('$', 2)
221#define PERF_COUNTER_IOC_RESET _IO ('$', 3)
222#define PERF_COUNTER_IOC_PERIOD _IOW('$', 4, u64)
a4be7c27 223#define PERF_COUNTER_IOC_SET_OUTPUT _IO ('$', 5)
3df5edad
PZ
224
225enum perf_counter_ioc_flags {
226 PERF_IOC_FLAG_GROUP = 1U << 0,
227};
d859e29f 228
37d81828
PM
229/*
230 * Structure of the page that can be mapped via mmap
231 */
232struct perf_counter_mmap_page {
233 __u32 version; /* version number of this structure */
234 __u32 compat_version; /* lowest version this is compat with */
38ff667b
PZ
235
236 /*
237 * Bits needed to read the hw counters in user-space.
238 *
92f22a38
PZ
239 * u32 seq;
240 * s64 count;
38ff667b 241 *
a2e87d06
PZ
242 * do {
243 * seq = pc->lock;
38ff667b 244 *
a2e87d06
PZ
245 * barrier()
246 * if (pc->index) {
247 * count = pmc_read(pc->index - 1);
248 * count += pc->offset;
249 * } else
250 * goto regular_read;
38ff667b 251 *
a2e87d06
PZ
252 * barrier();
253 * } while (pc->lock != seq);
38ff667b 254 *
92f22a38
PZ
255 * NOTE: for obvious reason this only works on self-monitoring
256 * processes.
38ff667b 257 */
37d81828
PM
258 __u32 lock; /* seqlock for synchronization */
259 __u32 index; /* hardware counter identifier */
260 __s64 offset; /* add to hardware counter value */
7f8b4e4e
PZ
261 __u64 time_enabled; /* time counter active */
262 __u64 time_running; /* time counter on cpu */
7b732a75 263
41f95331
PZ
264 /*
265 * Hole for extension of the self monitor capabilities
266 */
267
7f8b4e4e 268 __u64 __reserved[123]; /* align to 1k */
41f95331 269
38ff667b
PZ
270 /*
271 * Control data for the mmap() data buffer.
272 *
43a21ea8
PZ
273 * User-space reading the @data_head value should issue an rmb(), on
274 * SMP capable platforms, after reading this value -- see
275 * perf_counter_wakeup().
276 *
277 * When the mapping is PROT_WRITE the @data_tail value should be
278 * written by userspace to reflect the last read data. In this case
279 * the kernel will not over-write unread data.
38ff667b 280 */
8e3747c1 281 __u64 data_head; /* head in the data section */
43a21ea8 282 __u64 data_tail; /* user-space written tail */
37d81828
PM
283};
284
a308444c
IM
285#define PERF_EVENT_MISC_CPUMODE_MASK (3 << 0)
286#define PERF_EVENT_MISC_CPUMODE_UNKNOWN (0 << 0)
287#define PERF_EVENT_MISC_KERNEL (1 << 0)
288#define PERF_EVENT_MISC_USER (2 << 0)
289#define PERF_EVENT_MISC_HYPERVISOR (3 << 0)
6fab0192 290
5c148194
PZ
291struct perf_event_header {
292 __u32 type;
6fab0192
PZ
293 __u16 misc;
294 __u16 size;
5c148194
PZ
295};
296
297enum perf_event_type {
5ed00415 298
0c593b34
PZ
299 /*
300 * The MMAP events record the PROT_EXEC mappings so that we can
301 * correlate userspace IPs to code. They have the following structure:
302 *
303 * struct {
0127c3ea 304 * struct perf_event_header header;
0c593b34 305 *
0127c3ea
IM
306 * u32 pid, tid;
307 * u64 addr;
308 * u64 len;
309 * u64 pgoff;
310 * char filename[];
0c593b34
PZ
311 * };
312 */
8a057d84 313 PERF_EVENT_MMAP = 1,
0a4a9391 314
43a21ea8
PZ
315 /*
316 * struct {
317 * struct perf_event_header header;
318 * u64 id;
319 * u64 lost;
320 * };
321 */
322 PERF_EVENT_LOST = 2,
323
8d1b2d93
PZ
324 /*
325 * struct {
0127c3ea 326 * struct perf_event_header header;
8d1b2d93 327 *
0127c3ea
IM
328 * u32 pid, tid;
329 * char comm[];
8d1b2d93
PZ
330 * };
331 */
332 PERF_EVENT_COMM = 3,
333
9f498cc5
PZ
334 /*
335 * struct {
336 * struct perf_event_header header;
337 * u32 pid, ppid;
338 * u32 tid, ptid;
393b2ad8 339 * u64 time;
9f498cc5
PZ
340 * };
341 */
342 PERF_EVENT_EXIT = 4,
343
26b119bc
PZ
344 /*
345 * struct {
0127c3ea
IM
346 * struct perf_event_header header;
347 * u64 time;
689802b2 348 * u64 id;
7f453c24 349 * u64 stream_id;
a78ac325
PZ
350 * };
351 */
352 PERF_EVENT_THROTTLE = 5,
353 PERF_EVENT_UNTHROTTLE = 6,
354
60313ebe
PZ
355 /*
356 * struct {
a21ca2ca
IM
357 * struct perf_event_header header;
358 * u32 pid, ppid;
9f498cc5 359 * u32 tid, ptid;
393b2ad8 360 * { u64 time; } && PERF_SAMPLE_TIME
60313ebe
PZ
361 * };
362 */
363 PERF_EVENT_FORK = 7,
364
38b200d6
PZ
365 /*
366 * struct {
367 * struct perf_event_header header;
368 * u32 pid, tid;
3dab77fb
PZ
369 *
370 * struct read_format values;
38b200d6
PZ
371 * };
372 */
373 PERF_EVENT_READ = 8,
374
8a057d84 375 /*
0c593b34 376 * struct {
0127c3ea 377 * struct perf_event_header header;
0c593b34 378 *
43a21ea8
PZ
379 * { u64 ip; } && PERF_SAMPLE_IP
380 * { u32 pid, tid; } && PERF_SAMPLE_TID
381 * { u64 time; } && PERF_SAMPLE_TIME
382 * { u64 addr; } && PERF_SAMPLE_ADDR
e6e18ec7 383 * { u64 id; } && PERF_SAMPLE_ID
7f453c24 384 * { u64 stream_id;} && PERF_SAMPLE_STREAM_ID
43a21ea8 385 * { u32 cpu, res; } && PERF_SAMPLE_CPU
e6e18ec7 386 * { u64 period; } && PERF_SAMPLE_PERIOD
0c593b34 387 *
3dab77fb 388 * { struct read_format values; } && PERF_SAMPLE_READ
0c593b34 389 *
f9188e02 390 * { u64 nr,
43a21ea8 391 * u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN
3dab77fb
PZ
392 *
393 * #
394 * # The RAW record below is opaque data wrt the ABI
395 * #
396 * # That is, the ABI doesn't make any promises wrt to
397 * # the stability of its content, it may vary depending
398 * # on event, hardware, kernel version and phase of
399 * # the moon.
400 * #
401 * # In other words, PERF_SAMPLE_RAW contents are not an ABI.
402 * #
403 *
a044560c
PZ
404 * { u32 size;
405 * char data[size];}&& PERF_SAMPLE_RAW
0c593b34 406 * };
8a057d84 407 */
e6e18ec7
PZ
408 PERF_EVENT_SAMPLE = 9,
409
410 PERF_EVENT_MAX, /* non-ABI */
5c148194
PZ
411};
412
f9188e02
PZ
413enum perf_callchain_context {
414 PERF_CONTEXT_HV = (__u64)-32,
415 PERF_CONTEXT_KERNEL = (__u64)-128,
416 PERF_CONTEXT_USER = (__u64)-512,
7522060c 417
f9188e02
PZ
418 PERF_CONTEXT_GUEST = (__u64)-2048,
419 PERF_CONTEXT_GUEST_KERNEL = (__u64)-2176,
420 PERF_CONTEXT_GUEST_USER = (__u64)-2560,
421
422 PERF_CONTEXT_MAX = (__u64)-4095,
7522060c
IM
423};
424
a4be7c27
PZ
425#define PERF_FLAG_FD_NO_GROUP (1U << 0)
426#define PERF_FLAG_FD_OUTPUT (1U << 1)
427
f3dfd265 428#ifdef __KERNEL__
9f66a381 429/*
f3dfd265 430 * Kernel-internal data types and definitions:
9f66a381
IM
431 */
432
f3dfd265
PM
433#ifdef CONFIG_PERF_COUNTERS
434# include <asm/perf_counter.h>
435#endif
436
437#include <linux/list.h>
438#include <linux/mutex.h>
439#include <linux/rculist.h>
440#include <linux/rcupdate.h>
441#include <linux/spinlock.h>
d6d020e9 442#include <linux/hrtimer.h>
3c446b3d 443#include <linux/fs.h>
709e50cf 444#include <linux/pid_namespace.h>
f3dfd265
PM
445#include <asm/atomic.h>
446
f9188e02
PZ
447#define PERF_MAX_STACK_DEPTH 255
448
449struct perf_callchain_entry {
450 __u64 nr;
451 __u64 ip[PERF_MAX_STACK_DEPTH];
452};
453
3a43ce68
FW
454struct perf_raw_record {
455 u32 size;
456 void *data;
f413cdb8
FW
457};
458
f3dfd265
PM
459struct task_struct;
460
0793a61d 461/**
9f66a381 462 * struct hw_perf_counter - performance counter hardware details:
0793a61d
TG
463 */
464struct hw_perf_counter {
ee06094f 465#ifdef CONFIG_PERF_COUNTERS
d6d020e9
PZ
466 union {
467 struct { /* hardware */
a308444c
IM
468 u64 config;
469 unsigned long config_base;
470 unsigned long counter_base;
471 int idx;
d6d020e9
PZ
472 };
473 union { /* software */
a308444c
IM
474 atomic64_t count;
475 struct hrtimer hrtimer;
d6d020e9
PZ
476 };
477 };
ee06094f 478 atomic64_t prev_count;
b23f3325 479 u64 sample_period;
9e350de3 480 u64 last_period;
ee06094f 481 atomic64_t period_left;
60db5e09 482 u64 interrupts;
6a24ed6c
PZ
483
484 u64 freq_count;
485 u64 freq_interrupts;
bd2b5b12 486 u64 freq_stamp;
ee06094f 487#endif
0793a61d
TG
488};
489
621a01ea
IM
490struct perf_counter;
491
492/**
4aeb0b42 493 * struct pmu - generic performance monitoring unit
621a01ea 494 */
4aeb0b42 495struct pmu {
95cdd2e7 496 int (*enable) (struct perf_counter *counter);
7671581f
IM
497 void (*disable) (struct perf_counter *counter);
498 void (*read) (struct perf_counter *counter);
a78ac325 499 void (*unthrottle) (struct perf_counter *counter);
621a01ea
IM
500};
501
6a930700
IM
502/**
503 * enum perf_counter_active_state - the states of a counter
504 */
505enum perf_counter_active_state {
3b6f9e5c 506 PERF_COUNTER_STATE_ERROR = -2,
6a930700
IM
507 PERF_COUNTER_STATE_OFF = -1,
508 PERF_COUNTER_STATE_INACTIVE = 0,
509 PERF_COUNTER_STATE_ACTIVE = 1,
510};
511
9b51f66d
IM
512struct file;
513
7b732a75
PZ
514struct perf_mmap_data {
515 struct rcu_head rcu_head;
8740f941 516 int nr_pages; /* nr of data pages */
43a21ea8 517 int writable; /* are we writable */
c5078f78 518 int nr_locked; /* nr pages mlocked */
8740f941 519
c33a0bc4 520 atomic_t poll; /* POLL_ for wakeups */
8740f941
PZ
521 atomic_t events; /* event limit */
522
8e3747c1
PZ
523 atomic_long_t head; /* write position */
524 atomic_long_t done_head; /* completed head */
525
c33a0bc4 526 atomic_t lock; /* concurrent writes */
c66de4a5 527 atomic_t wakeup; /* needs a wakeup */
43a21ea8 528 atomic_t lost; /* nr records lost */
c66de4a5 529
2667de81
PZ
530 long watermark; /* wakeup watermark */
531
7b732a75 532 struct perf_counter_mmap_page *user_page;
0127c3ea 533 void *data_pages[0];
7b732a75
PZ
534};
535
671dec5d
PZ
536struct perf_pending_entry {
537 struct perf_pending_entry *next;
538 void (*func)(struct perf_pending_entry *);
925d519a
PZ
539};
540
0793a61d
TG
541/**
542 * struct perf_counter - performance counter kernel representation:
543 */
544struct perf_counter {
ee06094f 545#ifdef CONFIG_PERF_COUNTERS
04289bb9 546 struct list_head list_entry;
592903cd 547 struct list_head event_entry;
04289bb9 548 struct list_head sibling_list;
0127c3ea 549 int nr_siblings;
04289bb9 550 struct perf_counter *group_leader;
a4be7c27 551 struct perf_counter *output;
4aeb0b42 552 const struct pmu *pmu;
04289bb9 553
6a930700 554 enum perf_counter_active_state state;
0793a61d 555 atomic64_t count;
ee06094f 556
53cfbf59
PM
557 /*
558 * These are the total time in nanoseconds that the counter
559 * has been enabled (i.e. eligible to run, and the task has
560 * been scheduled in, if this is a per-task counter)
561 * and running (scheduled onto the CPU), respectively.
562 *
563 * They are computed from tstamp_enabled, tstamp_running and
564 * tstamp_stopped when the counter is in INACTIVE or ACTIVE state.
565 */
566 u64 total_time_enabled;
567 u64 total_time_running;
568
569 /*
570 * These are timestamps used for computing total_time_enabled
571 * and total_time_running when the counter is in INACTIVE or
572 * ACTIVE state, measured in nanoseconds from an arbitrary point
573 * in time.
574 * tstamp_enabled: the notional time when the counter was enabled
575 * tstamp_running: the notional time when the counter was scheduled on
576 * tstamp_stopped: in INACTIVE state, the notional time when the
577 * counter was scheduled off.
578 */
579 u64 tstamp_enabled;
580 u64 tstamp_running;
581 u64 tstamp_stopped;
582
0d48696f 583 struct perf_counter_attr attr;
0793a61d
TG
584 struct hw_perf_counter hw;
585
586 struct perf_counter_context *ctx;
9b51f66d 587 struct file *filp;
0793a61d 588
53cfbf59
PM
589 /*
590 * These accumulate total time (in nanoseconds) that children
591 * counters have been enabled and running, respectively.
592 */
593 atomic64_t child_total_time_enabled;
594 atomic64_t child_total_time_running;
595
0793a61d 596 /*
d859e29f 597 * Protect attach/detach and child_list:
0793a61d 598 */
fccc714b
PZ
599 struct mutex child_mutex;
600 struct list_head child_list;
601 struct perf_counter *parent;
0793a61d
TG
602
603 int oncpu;
604 int cpu;
605
082ff5a2
PZ
606 struct list_head owner_entry;
607 struct task_struct *owner;
608
7b732a75
PZ
609 /* mmap bits */
610 struct mutex mmap_mutex;
611 atomic_t mmap_count;
612 struct perf_mmap_data *data;
37d81828 613
7b732a75 614 /* poll related */
0793a61d 615 wait_queue_head_t waitq;
3c446b3d 616 struct fasync_struct *fasync;
79f14641
PZ
617
618 /* delayed work for NMIs and such */
619 int pending_wakeup;
4c9e2542 620 int pending_kill;
79f14641 621 int pending_disable;
671dec5d 622 struct perf_pending_entry pending;
592903cd 623
79f14641
PZ
624 atomic_t event_limit;
625
e077df4f 626 void (*destroy)(struct perf_counter *);
592903cd 627 struct rcu_head rcu_head;
709e50cf
PZ
628
629 struct pid_namespace *ns;
8e5799b1 630 u64 id;
ee06094f 631#endif
0793a61d
TG
632};
633
634/**
635 * struct perf_counter_context - counter context structure
636 *
637 * Used as a container for task counters and CPU counters as well:
638 */
639struct perf_counter_context {
0793a61d 640 /*
d859e29f
PM
641 * Protect the states of the counters in the list,
642 * nr_active, and the list:
0793a61d 643 */
a308444c 644 spinlock_t lock;
d859e29f
PM
645 /*
646 * Protect the list of counters. Locking either mutex or lock
647 * is sufficient to ensure the list doesn't change; to change
648 * the list you need to lock both the mutex and the spinlock.
649 */
a308444c 650 struct mutex mutex;
04289bb9 651
a308444c
IM
652 struct list_head counter_list;
653 struct list_head event_list;
654 int nr_counters;
655 int nr_active;
656 int is_active;
bfbd3381 657 int nr_stat;
a308444c
IM
658 atomic_t refcount;
659 struct task_struct *task;
53cfbf59
PM
660
661 /*
4af4998b 662 * Context clock, runs when context enabled.
53cfbf59 663 */
a308444c
IM
664 u64 time;
665 u64 timestamp;
564c2b21
PM
666
667 /*
668 * These fields let us detect when two contexts have both
669 * been cloned (inherited) from a common ancestor.
670 */
a308444c
IM
671 struct perf_counter_context *parent_ctx;
672 u64 parent_gen;
673 u64 generation;
674 int pin_count;
675 struct rcu_head rcu_head;
0793a61d
TG
676};
677
678/**
679 * struct perf_counter_cpu_context - per cpu counter context structure
680 */
681struct perf_cpu_context {
682 struct perf_counter_context ctx;
683 struct perf_counter_context *task_ctx;
684 int active_oncpu;
685 int max_pertask;
3b6f9e5c 686 int exclusive;
96f6d444
PZ
687
688 /*
689 * Recursion avoidance:
690 *
691 * task, softirq, irq, nmi context
692 */
22a4f650 693 int recursion[4];
0793a61d
TG
694};
695
5622f295
MM
696struct perf_output_handle {
697 struct perf_counter *counter;
698 struct perf_mmap_data *data;
699 unsigned long head;
700 unsigned long offset;
701 int nmi;
702 int sample;
703 int locked;
704 unsigned long flags;
705};
706
829b42dd
RR
707#ifdef CONFIG_PERF_COUNTERS
708
0793a61d
TG
709/*
710 * Set by architecture code:
711 */
712extern int perf_max_counters;
713
4aeb0b42 714extern const struct pmu *hw_perf_counter_init(struct perf_counter *counter);
621a01ea 715
0793a61d 716extern void perf_counter_task_sched_in(struct task_struct *task, int cpu);
564c2b21
PM
717extern void perf_counter_task_sched_out(struct task_struct *task,
718 struct task_struct *next, int cpu);
0793a61d 719extern void perf_counter_task_tick(struct task_struct *task, int cpu);
6ab423e0 720extern int perf_counter_init_task(struct task_struct *child);
9b51f66d 721extern void perf_counter_exit_task(struct task_struct *child);
bbbee908 722extern void perf_counter_free_task(struct task_struct *task);
9974458e 723extern void set_perf_counter_pending(void);
925d519a 724extern void perf_counter_do_pending(void);
0793a61d 725extern void perf_counter_print_debug(void);
9e35ad38
PZ
726extern void __perf_disable(void);
727extern bool __perf_enable(void);
728extern void perf_disable(void);
729extern void perf_enable(void);
1d1c7ddb
IM
730extern int perf_counter_task_disable(void);
731extern int perf_counter_task_enable(void);
3cbed429
PM
732extern int hw_perf_group_sched_in(struct perf_counter *group_leader,
733 struct perf_cpu_context *cpuctx,
734 struct perf_counter_context *ctx, int cpu);
37d81828 735extern void perf_counter_update_userpage(struct perf_counter *counter);
5c92d124 736
df1a132b 737struct perf_sample_data {
5622f295
MM
738 u64 type;
739
740 u64 ip;
741 struct {
742 u32 pid;
743 u32 tid;
744 } tid_entry;
745 u64 time;
a308444c 746 u64 addr;
5622f295
MM
747 u64 id;
748 u64 stream_id;
749 struct {
750 u32 cpu;
751 u32 reserved;
752 } cpu_entry;
a308444c 753 u64 period;
5622f295 754 struct perf_callchain_entry *callchain;
3a43ce68 755 struct perf_raw_record *raw;
df1a132b
PZ
756};
757
5622f295
MM
758extern void perf_output_sample(struct perf_output_handle *handle,
759 struct perf_event_header *header,
760 struct perf_sample_data *data,
761 struct perf_counter *counter);
762extern void perf_prepare_sample(struct perf_event_header *header,
763 struct perf_sample_data *data,
764 struct perf_counter *counter,
765 struct pt_regs *regs);
766
df1a132b 767extern int perf_counter_overflow(struct perf_counter *counter, int nmi,
5622f295
MM
768 struct perf_sample_data *data,
769 struct pt_regs *regs);
df1a132b 770
3b6f9e5c
PM
771/*
772 * Return 1 for a software counter, 0 for a hardware counter
773 */
774static inline int is_software_counter(struct perf_counter *counter)
775{
a21ca2ca 776 return (counter->attr.type != PERF_TYPE_RAW) &&
f1a3c979
PZ
777 (counter->attr.type != PERF_TYPE_HARDWARE) &&
778 (counter->attr.type != PERF_TYPE_HW_CACHE);
3b6f9e5c
PM
779}
780
f29ac756
PZ
781extern atomic_t perf_swcounter_enabled[PERF_COUNT_SW_MAX];
782
783extern void __perf_swcounter_event(u32, u64, int, struct pt_regs *, u64);
784
785static inline void
786perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs, u64 addr)
787{
788 if (atomic_read(&perf_swcounter_enabled[event]))
789 __perf_swcounter_event(event, nr, nmi, regs, addr);
790}
15dbf27c 791
089dd79d
PZ
792extern void __perf_counter_mmap(struct vm_area_struct *vma);
793
794static inline void perf_counter_mmap(struct vm_area_struct *vma)
795{
796 if (vma->vm_flags & VM_EXEC)
797 __perf_counter_mmap(vma);
798}
0a4a9391 799
8d1b2d93 800extern void perf_counter_comm(struct task_struct *tsk);
60313ebe 801extern void perf_counter_fork(struct task_struct *tsk);
8d1b2d93 802
394ee076
PZ
803extern struct perf_callchain_entry *perf_callchain(struct pt_regs *regs);
804
0764771d 805extern int sysctl_perf_counter_paranoid;
c5078f78 806extern int sysctl_perf_counter_mlock;
df58ab24 807extern int sysctl_perf_counter_sample_rate;
1ccd1549 808
0d905bca 809extern void perf_counter_init(void);
f4b5ffcc
JB
810extern void perf_tpcounter_event(int event_id, u64 addr, u64 count,
811 void *record, int entry_size);
0d905bca 812
9d23a90a
PM
813#ifndef perf_misc_flags
814#define perf_misc_flags(regs) (user_mode(regs) ? PERF_EVENT_MISC_USER : \
815 PERF_EVENT_MISC_KERNEL)
816#define perf_instruction_pointer(regs) instruction_pointer(regs)
817#endif
818
5622f295
MM
819extern int perf_output_begin(struct perf_output_handle *handle,
820 struct perf_counter *counter, unsigned int size,
821 int nmi, int sample);
822extern void perf_output_end(struct perf_output_handle *handle);
823extern void perf_output_copy(struct perf_output_handle *handle,
824 const void *buf, unsigned int len);
0793a61d
TG
825#else
826static inline void
827perf_counter_task_sched_in(struct task_struct *task, int cpu) { }
828static inline void
910431c7
IM
829perf_counter_task_sched_out(struct task_struct *task,
830 struct task_struct *next, int cpu) { }
0793a61d
TG
831static inline void
832perf_counter_task_tick(struct task_struct *task, int cpu) { }
d3e78ee3 833static inline int perf_counter_init_task(struct task_struct *child) { return 0; }
9b51f66d 834static inline void perf_counter_exit_task(struct task_struct *child) { }
bbbee908 835static inline void perf_counter_free_task(struct task_struct *task) { }
925d519a 836static inline void perf_counter_do_pending(void) { }
0793a61d 837static inline void perf_counter_print_debug(void) { }
9e35ad38
PZ
838static inline void perf_disable(void) { }
839static inline void perf_enable(void) { }
1d1c7ddb
IM
840static inline int perf_counter_task_disable(void) { return -EINVAL; }
841static inline int perf_counter_task_enable(void) { return -EINVAL; }
15dbf27c 842
925d519a 843static inline void
78f13e95
PZ
844perf_swcounter_event(u32 event, u64 nr, int nmi,
845 struct pt_regs *regs, u64 addr) { }
0a4a9391 846
089dd79d 847static inline void perf_counter_mmap(struct vm_area_struct *vma) { }
8d1b2d93 848static inline void perf_counter_comm(struct task_struct *tsk) { }
60313ebe 849static inline void perf_counter_fork(struct task_struct *tsk) { }
0d905bca 850static inline void perf_counter_init(void) { }
5622f295 851
0793a61d
TG
852#endif
853
5622f295
MM
854#define perf_output_put(handle, x) \
855 perf_output_copy((handle), &(x), sizeof(x))
856
f3dfd265 857#endif /* __KERNEL__ */
0793a61d 858#endif /* _LINUX_PERF_COUNTER_H */