perf_counter: unify and fix delayed counter wakeup
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / include / linux / perf_counter.h
CommitLineData
0793a61d
TG
1/*
2 * Performance counters:
3 *
4 * Copyright(C) 2008, Thomas Gleixner <tglx@linutronix.de>
5 * Copyright(C) 2008, Red Hat, Inc., Ingo Molnar
6 *
7 * Data type definitions, declarations, prototypes.
8 *
9 * Started by: Thomas Gleixner and Ingo Molnar
10 *
11 * For licencing details see kernel-base/COPYING
12 */
13#ifndef _LINUX_PERF_COUNTER_H
14#define _LINUX_PERF_COUNTER_H
15
f3dfd265
PM
16#include <linux/types.h>
17#include <linux/ioctl.h>
9aaa131a 18#include <asm/byteorder.h>
0793a61d
TG
19
20/*
9f66a381
IM
21 * User-space ABI bits:
22 */
23
24/*
b8e83514 25 * hw_event.type
0793a61d 26 */
b8e83514
PZ
27enum perf_event_types {
28 PERF_TYPE_HARDWARE = 0,
29 PERF_TYPE_SOFTWARE = 1,
30 PERF_TYPE_TRACEPOINT = 2,
31
0793a61d 32 /*
b8e83514 33 * available TYPE space, raw is the max value.
0793a61d 34 */
9f66a381 35
b8e83514
PZ
36 PERF_TYPE_RAW = 128,
37};
6c594c21 38
b8e83514
PZ
39/*
40 * Generalized performance counter event types, used by the hw_event.event_id
41 * parameter of the sys_perf_counter_open() syscall:
42 */
43enum hw_event_ids {
9f66a381 44 /*
b8e83514 45 * Common hardware events, generalized by the kernel:
9f66a381 46 */
b8e83514
PZ
47 PERF_COUNT_CPU_CYCLES = 0,
48 PERF_COUNT_INSTRUCTIONS = 1,
49 PERF_COUNT_CACHE_REFERENCES = 2,
50 PERF_COUNT_CACHE_MISSES = 3,
51 PERF_COUNT_BRANCH_INSTRUCTIONS = 4,
52 PERF_COUNT_BRANCH_MISSES = 5,
53 PERF_COUNT_BUS_CYCLES = 6,
54
55 PERF_HW_EVENTS_MAX = 7,
56};
e077df4f 57
b8e83514
PZ
58/*
59 * Special "software" counters provided by the kernel, even if the hardware
60 * does not support performance counters. These counters measure various
61 * physical and sw events of the kernel (and allow the profiling of them as
62 * well):
63 */
64enum sw_event_ids {
65 PERF_COUNT_CPU_CLOCK = 0,
66 PERF_COUNT_TASK_CLOCK = 1,
67 PERF_COUNT_PAGE_FAULTS = 2,
68 PERF_COUNT_CONTEXT_SWITCHES = 3,
69 PERF_COUNT_CPU_MIGRATIONS = 4,
70 PERF_COUNT_PAGE_FAULTS_MIN = 5,
71 PERF_COUNT_PAGE_FAULTS_MAJ = 6,
72
73 PERF_SW_EVENTS_MAX = 7,
0793a61d
TG
74};
75
76/*
77 * IRQ-notification data record type:
78 */
9f66a381 79enum perf_counter_record_type {
b8e83514
PZ
80 PERF_RECORD_SIMPLE = 0,
81 PERF_RECORD_IRQ = 1,
82 PERF_RECORD_GROUP = 2,
0793a61d
TG
83};
84
f4a2deb4
PZ
85#define __PERF_COUNTER_MASK(name) \
86 (((1ULL << PERF_COUNTER_##name##_BITS) - 1) << \
87 PERF_COUNTER_##name##_SHIFT)
88
89#define PERF_COUNTER_RAW_BITS 1
90#define PERF_COUNTER_RAW_SHIFT 63
91#define PERF_COUNTER_RAW_MASK __PERF_COUNTER_MASK(RAW)
92
93#define PERF_COUNTER_CONFIG_BITS 63
94#define PERF_COUNTER_CONFIG_SHIFT 0
95#define PERF_COUNTER_CONFIG_MASK __PERF_COUNTER_MASK(CONFIG)
96
97#define PERF_COUNTER_TYPE_BITS 7
98#define PERF_COUNTER_TYPE_SHIFT 56
99#define PERF_COUNTER_TYPE_MASK __PERF_COUNTER_MASK(TYPE)
100
101#define PERF_COUNTER_EVENT_BITS 56
102#define PERF_COUNTER_EVENT_SHIFT 0
103#define PERF_COUNTER_EVENT_MASK __PERF_COUNTER_MASK(EVENT)
104
53cfbf59
PM
105/*
106 * Bits that can be set in hw_event.read_format to request that
107 * reads on the counter should return the indicated quantities,
108 * in increasing order of bit value, after the counter value.
109 */
110enum perf_counter_read_format {
111 PERF_FORMAT_TOTAL_TIME_ENABLED = 1,
112 PERF_FORMAT_TOTAL_TIME_RUNNING = 2,
113};
114
9f66a381
IM
115/*
116 * Hardware event to monitor via a performance monitoring counter:
117 */
118struct perf_counter_hw_event {
f4a2deb4
PZ
119 /*
120 * The MSB of the config word signifies if the rest contains cpu
121 * specific (raw) counter configuration data, if unset, the next
122 * 7 bits are an event type and the rest of the bits are the event
123 * identifier.
124 */
125 __u64 config;
9f66a381 126
f3dfd265 127 __u64 irq_period;
2743a5b0
PM
128 __u64 record_type;
129 __u64 read_format;
9f66a381 130
2743a5b0 131 __u64 disabled : 1, /* off by default */
0475f9ea 132 nmi : 1, /* NMI sampling */
0475f9ea
PM
133 inherit : 1, /* children inherit it */
134 pinned : 1, /* must always be on PMU */
135 exclusive : 1, /* only group on PMU */
136 exclude_user : 1, /* don't count user */
137 exclude_kernel : 1, /* ditto kernel */
138 exclude_hv : 1, /* ditto hypervisor */
2743a5b0 139 exclude_idle : 1, /* don't count when idle */
ea5d20cf 140 include_tid : 1, /* include the tid */
0475f9ea 141
ea5d20cf 142 __reserved_1 : 54;
2743a5b0
PM
143
144 __u32 extra_config_len;
145 __u32 __reserved_4;
9f66a381 146
f3dfd265 147 __u64 __reserved_2;
2743a5b0 148 __u64 __reserved_3;
eab656ae
TG
149};
150
d859e29f
PM
151/*
152 * Ioctls that can be done on a perf counter fd:
153 */
154#define PERF_COUNTER_IOC_ENABLE _IO('$', 0)
155#define PERF_COUNTER_IOC_DISABLE _IO('$', 1)
156
37d81828
PM
157/*
158 * Structure of the page that can be mapped via mmap
159 */
160struct perf_counter_mmap_page {
161 __u32 version; /* version number of this structure */
162 __u32 compat_version; /* lowest version this is compat with */
163 __u32 lock; /* seqlock for synchronization */
164 __u32 index; /* hardware counter identifier */
165 __s64 offset; /* add to hardware counter value */
7b732a75
PZ
166
167 __u32 data_head; /* head in the data section */
37d81828
PM
168};
169
5c148194
PZ
170struct perf_event_header {
171 __u32 type;
172 __u32 size;
173};
174
175enum perf_event_type {
176 PERF_EVENT_IP = 0,
177 PERF_EVENT_GROUP = 1,
ea5d20cf
PZ
178
179 __PERF_EVENT_TID = 0x100,
5c148194
PZ
180};
181
f3dfd265 182#ifdef __KERNEL__
9f66a381 183/*
f3dfd265 184 * Kernel-internal data types and definitions:
9f66a381
IM
185 */
186
f3dfd265
PM
187#ifdef CONFIG_PERF_COUNTERS
188# include <asm/perf_counter.h>
189#endif
190
191#include <linux/list.h>
192#include <linux/mutex.h>
193#include <linux/rculist.h>
194#include <linux/rcupdate.h>
195#include <linux/spinlock.h>
d6d020e9 196#include <linux/hrtimer.h>
f3dfd265
PM
197#include <asm/atomic.h>
198
199struct task_struct;
200
f4a2deb4
PZ
201static inline u64 perf_event_raw(struct perf_counter_hw_event *hw_event)
202{
203 return hw_event->config & PERF_COUNTER_RAW_MASK;
204}
205
206static inline u64 perf_event_config(struct perf_counter_hw_event *hw_event)
207{
208 return hw_event->config & PERF_COUNTER_CONFIG_MASK;
209}
210
211static inline u64 perf_event_type(struct perf_counter_hw_event *hw_event)
212{
213 return (hw_event->config & PERF_COUNTER_TYPE_MASK) >>
214 PERF_COUNTER_TYPE_SHIFT;
215}
216
217static inline u64 perf_event_id(struct perf_counter_hw_event *hw_event)
218{
219 return hw_event->config & PERF_COUNTER_EVENT_MASK;
220}
221
0793a61d 222/**
9f66a381 223 * struct hw_perf_counter - performance counter hardware details:
0793a61d
TG
224 */
225struct hw_perf_counter {
ee06094f 226#ifdef CONFIG_PERF_COUNTERS
d6d020e9
PZ
227 union {
228 struct { /* hardware */
229 u64 config;
230 unsigned long config_base;
231 unsigned long counter_base;
232 int nmi;
233 unsigned int idx;
234 };
235 union { /* software */
236 atomic64_t count;
237 struct hrtimer hrtimer;
238 };
239 };
ee06094f 240 atomic64_t prev_count;
9f66a381 241 u64 irq_period;
ee06094f
IM
242 atomic64_t period_left;
243#endif
0793a61d
TG
244};
245
621a01ea
IM
246struct perf_counter;
247
248/**
249 * struct hw_perf_counter_ops - performance counter hw ops
250 */
251struct hw_perf_counter_ops {
95cdd2e7 252 int (*enable) (struct perf_counter *counter);
7671581f
IM
253 void (*disable) (struct perf_counter *counter);
254 void (*read) (struct perf_counter *counter);
621a01ea
IM
255};
256
6a930700
IM
257/**
258 * enum perf_counter_active_state - the states of a counter
259 */
260enum perf_counter_active_state {
3b6f9e5c 261 PERF_COUNTER_STATE_ERROR = -2,
6a930700
IM
262 PERF_COUNTER_STATE_OFF = -1,
263 PERF_COUNTER_STATE_INACTIVE = 0,
264 PERF_COUNTER_STATE_ACTIVE = 1,
265};
266
9b51f66d
IM
267struct file;
268
7b732a75
PZ
269struct perf_mmap_data {
270 struct rcu_head rcu_head;
271 int nr_pages;
c7138f37 272 atomic_t wakeup;
7b732a75
PZ
273 atomic_t head;
274 struct perf_counter_mmap_page *user_page;
275 void *data_pages[0];
276};
277
925d519a
PZ
278struct perf_wakeup_entry {
279 struct perf_wakeup_entry *next;
280};
281
0793a61d
TG
282/**
283 * struct perf_counter - performance counter kernel representation:
284 */
285struct perf_counter {
ee06094f 286#ifdef CONFIG_PERF_COUNTERS
04289bb9 287 struct list_head list_entry;
592903cd 288 struct list_head event_entry;
04289bb9 289 struct list_head sibling_list;
5c148194 290 int nr_siblings;
04289bb9 291 struct perf_counter *group_leader;
5c92d124 292 const struct hw_perf_counter_ops *hw_ops;
04289bb9 293
6a930700 294 enum perf_counter_active_state state;
c07c99b6 295 enum perf_counter_active_state prev_state;
0793a61d 296 atomic64_t count;
ee06094f 297
53cfbf59
PM
298 /*
299 * These are the total time in nanoseconds that the counter
300 * has been enabled (i.e. eligible to run, and the task has
301 * been scheduled in, if this is a per-task counter)
302 * and running (scheduled onto the CPU), respectively.
303 *
304 * They are computed from tstamp_enabled, tstamp_running and
305 * tstamp_stopped when the counter is in INACTIVE or ACTIVE state.
306 */
307 u64 total_time_enabled;
308 u64 total_time_running;
309
310 /*
311 * These are timestamps used for computing total_time_enabled
312 * and total_time_running when the counter is in INACTIVE or
313 * ACTIVE state, measured in nanoseconds from an arbitrary point
314 * in time.
315 * tstamp_enabled: the notional time when the counter was enabled
316 * tstamp_running: the notional time when the counter was scheduled on
317 * tstamp_stopped: in INACTIVE state, the notional time when the
318 * counter was scheduled off.
319 */
320 u64 tstamp_enabled;
321 u64 tstamp_running;
322 u64 tstamp_stopped;
323
9f66a381 324 struct perf_counter_hw_event hw_event;
0793a61d
TG
325 struct hw_perf_counter hw;
326
327 struct perf_counter_context *ctx;
328 struct task_struct *task;
9b51f66d 329 struct file *filp;
0793a61d 330
9b51f66d 331 struct perf_counter *parent;
d859e29f
PM
332 struct list_head child_list;
333
53cfbf59
PM
334 /*
335 * These accumulate total time (in nanoseconds) that children
336 * counters have been enabled and running, respectively.
337 */
338 atomic64_t child_total_time_enabled;
339 atomic64_t child_total_time_running;
340
0793a61d 341 /*
d859e29f 342 * Protect attach/detach and child_list:
0793a61d
TG
343 */
344 struct mutex mutex;
345
346 int oncpu;
347 int cpu;
348
7b732a75
PZ
349 /* mmap bits */
350 struct mutex mmap_mutex;
351 atomic_t mmap_count;
352 struct perf_mmap_data *data;
37d81828 353
7b732a75 354 /* poll related */
0793a61d
TG
355 wait_queue_head_t waitq;
356 /* optional: for NMIs */
925d519a 357 struct perf_wakeup_entry wakeup;
592903cd 358
e077df4f 359 void (*destroy)(struct perf_counter *);
592903cd 360 struct rcu_head rcu_head;
ee06094f 361#endif
0793a61d
TG
362};
363
364/**
365 * struct perf_counter_context - counter context structure
366 *
367 * Used as a container for task counters and CPU counters as well:
368 */
369struct perf_counter_context {
370#ifdef CONFIG_PERF_COUNTERS
371 /*
d859e29f
PM
372 * Protect the states of the counters in the list,
373 * nr_active, and the list:
0793a61d
TG
374 */
375 spinlock_t lock;
d859e29f
PM
376 /*
377 * Protect the list of counters. Locking either mutex or lock
378 * is sufficient to ensure the list doesn't change; to change
379 * the list you need to lock both the mutex and the spinlock.
380 */
381 struct mutex mutex;
04289bb9
IM
382
383 struct list_head counter_list;
592903cd 384 struct list_head event_list;
0793a61d
TG
385 int nr_counters;
386 int nr_active;
d859e29f 387 int is_active;
0793a61d 388 struct task_struct *task;
53cfbf59
PM
389
390 /*
391 * time_now is the current time in nanoseconds since an arbitrary
392 * point in the past. For per-task counters, this is based on the
393 * task clock, and for per-cpu counters it is based on the cpu clock.
394 * time_lost is an offset from the task/cpu clock, used to make it
395 * appear that time only passes while the context is scheduled in.
396 */
397 u64 time_now;
398 u64 time_lost;
0793a61d
TG
399#endif
400};
401
402/**
403 * struct perf_counter_cpu_context - per cpu counter context structure
404 */
405struct perf_cpu_context {
406 struct perf_counter_context ctx;
407 struct perf_counter_context *task_ctx;
408 int active_oncpu;
409 int max_pertask;
3b6f9e5c 410 int exclusive;
96f6d444
PZ
411
412 /*
413 * Recursion avoidance:
414 *
415 * task, softirq, irq, nmi context
416 */
417 int recursion[4];
0793a61d
TG
418};
419
420/*
421 * Set by architecture code:
422 */
423extern int perf_max_counters;
424
425#ifdef CONFIG_PERF_COUNTERS
5c92d124 426extern const struct hw_perf_counter_ops *
621a01ea
IM
427hw_perf_counter_init(struct perf_counter *counter);
428
0793a61d
TG
429extern void perf_counter_task_sched_in(struct task_struct *task, int cpu);
430extern void perf_counter_task_sched_out(struct task_struct *task, int cpu);
431extern void perf_counter_task_tick(struct task_struct *task, int cpu);
9b51f66d
IM
432extern void perf_counter_init_task(struct task_struct *child);
433extern void perf_counter_exit_task(struct task_struct *child);
925d519a 434extern void perf_counter_do_pending(void);
0793a61d 435extern void perf_counter_print_debug(void);
1b023a96 436extern void perf_counter_unthrottle(void);
01b2838c
IM
437extern u64 hw_perf_save_disable(void);
438extern void hw_perf_restore(u64 ctrl);
1d1c7ddb
IM
439extern int perf_counter_task_disable(void);
440extern int perf_counter_task_enable(void);
3cbed429
PM
441extern int hw_perf_group_sched_in(struct perf_counter *group_leader,
442 struct perf_cpu_context *cpuctx,
443 struct perf_counter_context *ctx, int cpu);
37d81828 444extern void perf_counter_update_userpage(struct perf_counter *counter);
5c92d124 445
0322cd6e
PZ
446extern void perf_counter_output(struct perf_counter *counter,
447 int nmi, struct pt_regs *regs);
3b6f9e5c
PM
448/*
449 * Return 1 for a software counter, 0 for a hardware counter
450 */
451static inline int is_software_counter(struct perf_counter *counter)
452{
f4a2deb4
PZ
453 return !perf_event_raw(&counter->hw_event) &&
454 perf_event_type(&counter->hw_event) != PERF_TYPE_HARDWARE;
3b6f9e5c
PM
455}
456
b8e83514 457extern void perf_swcounter_event(u32, u64, int, struct pt_regs *);
15dbf27c 458
0793a61d
TG
459#else
460static inline void
461perf_counter_task_sched_in(struct task_struct *task, int cpu) { }
462static inline void
463perf_counter_task_sched_out(struct task_struct *task, int cpu) { }
464static inline void
465perf_counter_task_tick(struct task_struct *task, int cpu) { }
9b51f66d
IM
466static inline void perf_counter_init_task(struct task_struct *child) { }
467static inline void perf_counter_exit_task(struct task_struct *child) { }
925d519a 468static inline void perf_counter_do_pending(void) { }
0793a61d 469static inline void perf_counter_print_debug(void) { }
1b023a96 470static inline void perf_counter_unthrottle(void) { }
15dbf27c 471static inline void hw_perf_restore(u64 ctrl) { }
01b2838c 472static inline u64 hw_perf_save_disable(void) { return 0; }
1d1c7ddb
IM
473static inline int perf_counter_task_disable(void) { return -EINVAL; }
474static inline int perf_counter_task_enable(void) { return -EINVAL; }
15dbf27c 475
925d519a
PZ
476static inline void
477perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs) { }
478
0793a61d
TG
479#endif
480
f3dfd265 481#endif /* __KERNEL__ */
0793a61d 482#endif /* _LINUX_PERF_COUNTER_H */