include/linux/perf_event.h

   1 /*
   2  * Performance events:
   3  *
   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
   7  *
   8  * Data type definitions, declarations, prototypes.
   9  *
  10  *    Started by: Thomas Gleixner and Ingo Molnar
  11  *
  12  * For licencing details see kernel-base/COPYING
  13  */
  14 #ifndef _LINUX_PERF_EVENT_H
  15 #define _LINUX_PERF_EVENT_H
  16
  17 #include <linux/types.h>
  18 #include <linux/ioctl.h>
  19 #include <asm/byteorder.h>
  20
  21 /*
  22  * User-space ABI bits:
  23  */
  24
  25 /*
  26  * attr.type
  27  */
  28 enum perf_type_id {
  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,
  34
  35         PERF_TYPE_MAX,                          /* non-ABI */
  36 };
  37
  38 /*
  39  * Generalized performance event event_id types, used by the
  40  * attr.event_id parameter of the sys_perf_event_open()
  41  * syscall:
  42  */
  43 enum perf_hw_id {
  44         /*
  45          * Common hardware events, generalized by the kernel:
  46          */
  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
  55         PERF_COUNT_HW_MAX,                      /* non-ABI */
  56 };
  57
  58 /*
  59  * Generalized hardware cache events:
  60  *
  61  *       { L1-D, L1-I, LLC, ITLB, DTLB, BPU } x
  62  *       { read, write, prefetch } x
  63  *       { accesses, misses }
  64  */
  65 enum perf_hw_cache_id {
  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 */
  74 };
  75
  76 enum perf_hw_cache_op_id {
  77         PERF_COUNT_HW_CACHE_OP_READ             = 0,
  78         PERF_COUNT_HW_CACHE_OP_WRITE            = 1,
  79         PERF_COUNT_HW_CACHE_OP_PREFETCH         = 2,
  80
  81         PERF_COUNT_HW_CACHE_OP_MAX,             /* non-ABI */
  82 };
  83
  84 enum perf_hw_cache_op_result_id {
  85         PERF_COUNT_HW_CACHE_RESULT_ACCESS       = 0,
  86         PERF_COUNT_HW_CACHE_RESULT_MISS         = 1,
  87
  88         PERF_COUNT_HW_CACHE_RESULT_MAX,         /* non-ABI */
  89 };
  90
  91 /*
  92  * Special "software" events provided by the kernel, even if the hardware
  93  * does not support performance events. These events measure various
  94  * physical and sw events of the kernel (and allow the profiling of them as
  95  * well):
  96  */
  97 enum perf_sw_ids {
  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 */
 107 };
 108
 109 /*
 110  * Bits that can be set in attr.sample_type to request information
 111  * in the overflow packets.
 112  */
 113 enum perf_event_sample_format {
 114         PERF_SAMPLE_IP                          = 1U << 0,
 115         PERF_SAMPLE_TID                         = 1U << 1,
 116         PERF_SAMPLE_TIME                        = 1U << 2,
 117         PERF_SAMPLE_ADDR                        = 1U << 3,
 118         PERF_SAMPLE_READ                        = 1U << 4,
 119         PERF_SAMPLE_CALLCHAIN                   = 1U << 5,
 120         PERF_SAMPLE_ID                          = 1U << 6,
 121         PERF_SAMPLE_CPU                         = 1U << 7,
 122         PERF_SAMPLE_PERIOD                      = 1U << 8,
 123         PERF_SAMPLE_STREAM_ID                   = 1U << 9,
 124         PERF_SAMPLE_RAW                         = 1U << 10,
 125
 126         PERF_SAMPLE_MAX = 1U << 11,             /* non-ABI */
 127 };
 128
 129 /*
 130  * The format of the data returned by read() on a perf event 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  * };
 148  */
 149 enum perf_event_read_format {
 150         PERF_FORMAT_TOTAL_TIME_ENABLED          = 1U << 0,
 151         PERF_FORMAT_TOTAL_TIME_RUNNING          = 1U << 1,
 152         PERF_FORMAT_ID                          = 1U << 2,
 153         PERF_FORMAT_GROUP                       = 1U << 3,
 154
 155         PERF_FORMAT_MAX = 1U << 4,              /* non-ABI */
 156 };
 157
 158 #define PERF_ATTR_SIZE_VER0     64      /* sizeof first published struct */
 159
 160 /*
 161  * Hardware event_id to monitor via a performance monitoring event:
 162  */
 163 struct perf_event_attr {
 164
 165         /*
 166          * Major type: hardware/software/tracepoint/etc.
 167          */
 168         __u32                   type;
 169
 170         /*
 171          * Size of the attr structure, for fwd/bwd compat.
 172          */
 173         __u32                   size;
 174
 175         /*
 176          * Type specific configuration information.
 177          */
 178         __u64                   config;
 179
 180         union {
 181                 __u64           sample_period;
 182                 __u64           sample_freq;
 183         };
 184
 185         __u64                   sample_type;
 186         __u64                   read_format;
 187
 188         __u64                   disabled       :  1, /* off by default        */
 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      */
 195                                 exclude_idle   :  1, /* don't count when idle */
 196                                 mmap           :  1, /* include mmap data     */
 197                                 comm           :  1, /* include comm data     */
 198                                 freq           :  1, /* use freq, not period  */
 199                                 inherit_stat   :  1, /* per task counts       */
 200                                 enable_on_exec :  1, /* next exec enables     */
 201                                 task           :  1, /* trace fork/exit       */
 202                                 watermark      :  1, /* wakeup_watermark      */
 203
 204                                 __reserved_1   : 49;
 205
 206         union {
 207                 __u32           wakeup_events;    /* wakeup every n events */
 208                 __u32           wakeup_watermark; /* bytes before wakeup   */
 209         };
 210         __u32                   __reserved_2;
 211
 212         __u64                   __reserved_3;
 213 };
 214
 215 /*
 216  * Ioctls that can be done on a perf event fd:
 217  */
 218 #define PERF_EVENT_IOC_ENABLE           _IO ('$', 0)
 219 #define PERF_EVENT_IOC_DISABLE          _IO ('$', 1)
 220 #define PERF_EVENT_IOC_REFRESH          _IO ('$', 2)
 221 #define PERF_EVENT_IOC_RESET            _IO ('$', 3)
 222 #define PERF_EVENT_IOC_PERIOD           _IOW('$', 4, u64)
 223 #define PERF_EVENT_IOC_SET_OUTPUT       _IO ('$', 5)
 224
 225 enum perf_event_ioc_flags {
 226         PERF_IOC_FLAG_GROUP             = 1U << 0,
 227 };
 228
 229 /*
 230  * Structure of the page that can be mapped via mmap
 231  */
 232 struct perf_event_mmap_page {
 233         __u32   version;                /* version number of this structure */
 234         __u32   compat_version;         /* lowest version this is compat with */
 235
 236         /*
 237          * Bits needed to read the hw events in user-space.
 238          *
 239          *   u32 seq;
 240          *   s64 count;
 241          *
 242          *   do {
 243          *     seq = pc->lock;
 244          *
 245          *     barrier()
 246          *     if (pc->index) {
 247          *       count = pmc_read(pc->index - 1);
 248          *       count += pc->offset;
 249          *     } else
 250          *       goto regular_read;
 251          *
 252          *     barrier();
 253          *   } while (pc->lock != seq);
 254          *
 255          * NOTE: for obvious reason this only works on self-monitoring
 256          *       processes.
 257          */
 258         __u32   lock;                   /* seqlock for synchronization */
 259         __u32   index;                  /* hardware event identifier */
 260         __s64   offset;                 /* add to hardware event value */
 261         __u64   time_enabled;           /* time event active */
 262         __u64   time_running;           /* time event on cpu */
 263
 264                 /*
 265                  * Hole for extension of the self monitor capabilities
 266                  */
 267
 268         __u64   __reserved[123];        /* align to 1k */
 269
 270         /*
 271          * Control data for the mmap() data buffer.
 272          *
 273          * User-space reading the @data_head value should issue an rmb(), on
 274          * SMP capable platforms, after reading this value -- see
 275          * perf_event_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.
 280          */
 281         __u64   data_head;              /* head in the data section */
 282         __u64   data_tail;              /* user-space written tail */
 283 };
 284
 285 #define PERF_RECORD_MISC_CPUMODE_MASK           (3 << 0)
 286 #define PERF_RECORD_MISC_CPUMODE_UNKNOWN                (0 << 0)
 287 #define PERF_RECORD_MISC_KERNEL                 (1 << 0)
 288 #define PERF_RECORD_MISC_USER                   (2 << 0)
 289 #define PERF_RECORD_MISC_HYPERVISOR             (3 << 0)
 290
 291 struct perf_event_header {
 292         __u32   type;
 293         __u16   misc;
 294         __u16   size;
 295 };
 296
 297 enum perf_event_type {
 298
 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 {
 304          *      struct perf_event_header        header;
 305          *
 306          *      u32                             pid, tid;
 307          *      u64                             addr;
 308          *      u64                             len;
 309          *      u64                             pgoff;
 310          *      char                            filename[];
 311          * };
 312          */
 313         PERF_RECORD_MMAP                        = 1,
 314
 315         /*
 316          * struct {
 317          *      struct perf_event_header        header;
 318          *      u64                             id;
 319          *      u64                             lost;
 320          * };
 321          */
 322         PERF_RECORD_LOST                        = 2,
 323
 324         /*
 325          * struct {
 326          *      struct perf_event_header        header;
 327          *
 328          *      u32                             pid, tid;
 329          *      char                            comm[];
 330          * };
 331          */
 332         PERF_RECORD_COMM                        = 3,
 333
 334         /*
 335          * struct {
 336          *      struct perf_event_header        header;
 337          *      u32                             pid, ppid;
 338          *      u32                             tid, ptid;
 339          *      u64                             time;
 340          * };
 341          */
 342         PERF_RECORD_EXIT                        = 4,
 343
 344         /*
 345          * struct {
 346          *      struct perf_event_header        header;
 347          *      u64                             time;
 348          *      u64                             id;
 349          *      u64                             stream_id;
 350          * };
 351          */
 352         PERF_RECORD_THROTTLE            = 5,
 353         PERF_RECORD_UNTHROTTLE          = 6,
 354
 355         /*
 356          * struct {
 357          *      struct perf_event_header        header;
 358          *      u32                             pid, ppid;
 359          *      u32                             tid, ptid;
 360          *      u64                             time;
 361          * };
 362          */
 363         PERF_RECORD_FORK                        = 7,
 364
 365         /*
 366          * struct {
 367          *      struct perf_event_header        header;
 368          *      u32                             pid, tid;
 369          *
 370          *      struct read_format              values;
 371          * };
 372          */
 373         PERF_RECORD_READ                        = 8,
 374
 375         /*
 376          * struct {
 377          *      struct perf_event_header        header;
 378          *
 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
 383          *      { u64                   id;       } && PERF_SAMPLE_ID
 384          *      { u64                   stream_id;} && PERF_SAMPLE_STREAM_ID
 385          *      { u32                   cpu, res; } && PERF_SAMPLE_CPU
 386          *      { u64                   period;   } && PERF_SAMPLE_PERIOD
 387          *
 388          *      { struct read_format    values;   } && PERF_SAMPLE_READ
 389          *
 390          *      { u64                   nr,
 391          *        u64                   ips[nr];  } && PERF_SAMPLE_CALLCHAIN
 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          *
 404          *      { u32                   size;
 405          *        char                  data[size];}&& PERF_SAMPLE_RAW
 406          * };
 407          */
 408         PERF_RECORD_SAMPLE              = 9,
 409
 410         PERF_RECORD_MAX,                        /* non-ABI */
 411 };
 412
 413 enum perf_callchain_context {
 414         PERF_CONTEXT_HV                 = (__u64)-32,
 415         PERF_CONTEXT_KERNEL             = (__u64)-128,
 416         PERF_CONTEXT_USER               = (__u64)-512,
 417
 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,
 423 };
 424
 425 #define PERF_FLAG_FD_NO_GROUP   (1U << 0)
 426 #define PERF_FLAG_FD_OUTPUT     (1U << 1)
 427
 428 #ifdef __KERNEL__
 429 /*
 430  * Kernel-internal data types and definitions:
 431  */
 432
 433 #ifdef CONFIG_PERF_EVENTS
 434 # include <asm/perf_event.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>
 442 #include <linux/hrtimer.h>
 443 #include <linux/fs.h>
 444 #include <linux/pid_namespace.h>
 445 #include <asm/atomic.h>
 446
 447 #define PERF_MAX_STACK_DEPTH            255
 448
 449 struct perf_callchain_entry {
 450         __u64                           nr;
 451         __u64                           ip[PERF_MAX_STACK_DEPTH];
 452 };
 453
 454 struct perf_raw_record {
 455         u32                             size;
 456         void                            *data;
 457 };
 458
 459 struct task_struct;
 460
 461 /**
 462  * struct hw_perf_event - performance event hardware details:
 463  */
 464 struct hw_perf_event {
 465 #ifdef CONFIG_PERF_EVENTS
 466         union {
 467                 struct { /* hardware */
 468                         u64             config;
 469                         unsigned long   config_base;
 470                         unsigned long   event_base;
 471                         int             idx;
 472                 };
 473                 union { /* software */
 474                         atomic64_t      count;
 475                         struct hrtimer  hrtimer;
 476                 };
 477         };
 478         atomic64_t                      prev_count;
 479         u64                             sample_period;
 480         u64                             last_period;
 481         atomic64_t                      period_left;
 482         u64                             interrupts;
 483
 484         u64                             freq_count;
 485         u64                             freq_interrupts;
 486         u64                             freq_stamp;
 487 #endif
 488 };
 489
 490 struct perf_event;
 491
 492 /**
 493  * struct pmu - generic performance monitoring unit
 494  */
 495 struct pmu {
 496         int (*enable)                   (struct perf_event *event);
 497         void (*disable)                 (struct perf_event *event);
 498         void (*read)                    (struct perf_event *event);
 499         void (*unthrottle)              (struct perf_event *event);
 500 };
 501
 502 /**
 503  * enum perf_event_active_state - the states of a event
 504  */
 505 enum perf_event_active_state {
 506         PERF_EVENT_STATE_ERROR          = -2,
 507         PERF_EVENT_STATE_OFF            = -1,
 508         PERF_EVENT_STATE_INACTIVE       =  0,
 509         PERF_EVENT_STATE_ACTIVE         =  1,
 510 };
 511
 512 struct file;
 513
 514 struct perf_mmap_data {
 515         struct rcu_head                 rcu_head;
 516         int                             nr_pages;       /* nr of data pages  */
 517         int                             writable;       /* are we writable   */
 518         int                             nr_locked;      /* nr pages mlocked  */
 519
 520         atomic_t                        poll;           /* POLL_ for wakeups */
 521         atomic_t                        events;         /* event_id limit       */
 522
 523         atomic_long_t                   head;           /* write position    */
 524         atomic_long_t                   done_head;      /* completed head    */
 525
 526         atomic_t                        lock;           /* concurrent writes */
 527         atomic_t                        wakeup;         /* needs a wakeup    */
 528         atomic_t                        lost;           /* nr records lost   */
 529
 530         long                            watermark;      /* wakeup watermark  */
 531
 532         struct perf_event_mmap_page     *user_page;
 533         void                            *data_pages[0];
 534 };
 535
 536 struct perf_pending_entry {
 537         struct perf_pending_entry *next;
 538         void (*func)(struct perf_pending_entry *);
 539 };
 540
 541 /**
 542  * struct perf_event - performance event kernel representation:
 543  */
 544 struct perf_event {
 545 #ifdef CONFIG_PERF_EVENTS
 546         struct list_head                group_entry;
 547         struct list_head                event_entry;
 548         struct list_head                sibling_list;
 549         int                             nr_siblings;
 550         struct perf_event               *group_leader;
 551         struct perf_event               *output;
 552         const struct pmu                *pmu;
 553
 554         enum perf_event_active_state    state;
 555         atomic64_t                      count;
 556
 557         /*
 558          * These are the total time in nanoseconds that the event
 559          * has been enabled (i.e. eligible to run, and the task has
 560          * been scheduled in, if this is a per-task event)
 561          * and running (scheduled onto the CPU), respectively.
 562          *
 563          * They are computed from tstamp_enabled, tstamp_running and
 564          * tstamp_stopped when the event 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 event 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 event was enabled
 575          * tstamp_running: the notional time when the event was scheduled on
 576          * tstamp_stopped: in INACTIVE state, the notional time when the
 577          *      event was scheduled off.
 578          */
 579         u64                             tstamp_enabled;
 580         u64                             tstamp_running;
 581         u64                             tstamp_stopped;
 582
 583         struct perf_event_attr  attr;
 584         struct hw_perf_event            hw;
 585
 586         struct perf_event_context       *ctx;
 587         struct file                     *filp;
 588
 589         /*
 590          * These accumulate total time (in nanoseconds) that children
 591          * events have been enabled and running, respectively.
 592          */
 593         atomic64_t                      child_total_time_enabled;
 594         atomic64_t                      child_total_time_running;
 595
 596         /*
 597          * Protect attach/detach and child_list:
 598          */
 599         struct mutex                    child_mutex;
 600         struct list_head                child_list;
 601         struct perf_event               *parent;
 602
 603         int                             oncpu;
 604         int                             cpu;
 605
 606         struct list_head                owner_entry;
 607         struct task_struct              *owner;
 608
 609         /* mmap bits */
 610         struct mutex                    mmap_mutex;
 611         atomic_t                        mmap_count;
 612         struct perf_mmap_data           *data;
 613
 614         /* poll related */
 615         wait_queue_head_t               waitq;
 616         struct fasync_struct            *fasync;
 617
 618         /* delayed work for NMIs and such */
 619         int                             pending_wakeup;
 620         int                             pending_kill;
 621         int                             pending_disable;
 622         struct perf_pending_entry       pending;
 623
 624         atomic_t                        event_limit;
 625
 626         void (*destroy)(struct perf_event *);
 627         struct rcu_head                 rcu_head;
 628
 629         struct pid_namespace            *ns;
 630         u64                             id;
 631 #endif
 632 };
 633
 634 /**
 635  * struct perf_event_context - event context structure
 636  *
 637  * Used as a container for task events and CPU events as well:
 638  */
 639 struct perf_event_context {
 640         /*
 641          * Protect the states of the events in the list,
 642          * nr_active, and the list:
 643          */
 644         spinlock_t                      lock;
 645         /*
 646          * Protect the list of events.  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          */
 650         struct mutex                    mutex;
 651
 652         struct list_head                group_list;
 653         struct list_head                event_list;
 654         int                             nr_events;
 655         int                             nr_active;
 656         int                             is_active;
 657         int                             nr_stat;
 658         atomic_t                        refcount;
 659         struct task_struct              *task;
 660
 661         /*
 662          * Context clock, runs when context enabled.
 663          */
 664         u64                             time;
 665         u64                             timestamp;
 666
 667         /*
 668          * These fields let us detect when two contexts have both
 669          * been cloned (inherited) from a common ancestor.
 670          */
 671         struct perf_event_context       *parent_ctx;
 672         u64                             parent_gen;
 673         u64                             generation;
 674         int                             pin_count;
 675         struct rcu_head                 rcu_head;
 676 };
 677
 678 /**
 679  * struct perf_event_cpu_context - per cpu event context structure
 680  */
 681 struct perf_cpu_context {
 682         struct perf_event_context       ctx;
 683         struct perf_event_context       *task_ctx;
 684         int                             active_oncpu;
 685         int                             max_pertask;
 686         int                             exclusive;
 687
 688         /*
 689          * Recursion avoidance:
 690          *
 691          * task, softirq, irq, nmi context
 692          */
 693         int                             recursion[4];
 694 };
 695
 696 struct perf_output_handle {
 697         struct perf_event               *event;
 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
 707 #ifdef CONFIG_PERF_EVENTS
 708
 709 /*
 710  * Set by architecture code:
 711  */
 712 extern int perf_max_events;
 713
 714 extern const struct pmu *hw_perf_event_init(struct perf_event *event);
 715
 716 extern void perf_event_task_sched_in(struct task_struct *task, int cpu);
 717 extern void perf_event_task_sched_out(struct task_struct *task,
 718                                         struct task_struct *next, int cpu);
 719 extern void perf_event_task_tick(struct task_struct *task, int cpu);
 720 extern int perf_event_init_task(struct task_struct *child);
 721 extern void perf_event_exit_task(struct task_struct *child);
 722 extern void perf_event_free_task(struct task_struct *task);
 723 extern void set_perf_event_pending(void);
 724 extern void perf_event_do_pending(void);
 725 extern void perf_event_print_debug(void);
 726 extern void __perf_disable(void);
 727 extern bool __perf_enable(void);
 728 extern void perf_disable(void);
 729 extern void perf_enable(void);
 730 extern int perf_event_task_disable(void);
 731 extern int perf_event_task_enable(void);
 732 extern int hw_perf_group_sched_in(struct perf_event *group_leader,
 733                struct perf_cpu_context *cpuctx,
 734                struct perf_event_context *ctx, int cpu);
 735 extern void perf_event_update_userpage(struct perf_event *event);
 736
 737 struct perf_sample_data {
 738         u64                             type;
 739
 740         u64                             ip;
 741         struct {
 742                 u32     pid;
 743                 u32     tid;
 744         }                               tid_entry;
 745         u64                             time;
 746         u64                             addr;
 747         u64                             id;
 748         u64                             stream_id;
 749         struct {
 750                 u32     cpu;
 751                 u32     reserved;
 752         }                               cpu_entry;
 753         u64                             period;
 754         struct perf_callchain_entry     *callchain;
 755         struct perf_raw_record          *raw;
 756 };
 757
 758 extern void perf_output_sample(struct perf_output_handle *handle,
 759                                struct perf_event_header *header,
 760                                struct perf_sample_data *data,
 761                                struct perf_event *event);
 762 extern void perf_prepare_sample(struct perf_event_header *header,
 763                                 struct perf_sample_data *data,
 764                                 struct perf_event *event,
 765                                 struct pt_regs *regs);
 766
 767 extern int perf_event_overflow(struct perf_event *event, int nmi,
 768                                  struct perf_sample_data *data,
 769                                  struct pt_regs *regs);
 770
 771 /*
 772  * Return 1 for a software event, 0 for a hardware event
 773  */
 774 static inline int is_software_event(struct perf_event *event)
 775 {
 776         return (event->attr.type != PERF_TYPE_RAW) &&
 777                 (event->attr.type != PERF_TYPE_HARDWARE) &&
 778                 (event->attr.type != PERF_TYPE_HW_CACHE);
 779 }
 780
 781 extern atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX];
 782
 783 extern void __perf_sw_event(u32, u64, int, struct pt_regs *, u64);
 784
 785 static inline void
 786 perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr)
 787 {
 788         if (atomic_read(&perf_swevent_enabled[event_id]))
 789                 __perf_sw_event(event_id, nr, nmi, regs, addr);
 790 }
 791
 792 extern void __perf_event_mmap(struct vm_area_struct *vma);
 793
 794 static inline void perf_event_mmap(struct vm_area_struct *vma)
 795 {
 796         if (vma->vm_flags & VM_EXEC)
 797                 __perf_event_mmap(vma);
 798 }
 799
 800 extern void perf_event_comm(struct task_struct *tsk);
 801 extern void perf_event_fork(struct task_struct *tsk);
 802
 803 extern struct perf_callchain_entry *perf_callchain(struct pt_regs *regs);
 804
 805 extern int sysctl_perf_event_paranoid;
 806 extern int sysctl_perf_event_mlock;
 807 extern int sysctl_perf_event_sample_rate;
 808
 809 extern void perf_event_init(void);
 810 extern void perf_tp_event(int event_id, u64 addr, u64 count,
 811                                  void *record, int entry_size);
 812
 813 #ifndef perf_misc_flags
 814 #define perf_misc_flags(regs)   (user_mode(regs) ? PERF_RECORD_MISC_USER : \
 815                                  PERF_RECORD_MISC_KERNEL)
 816 #define perf_instruction_pointer(regs)  instruction_pointer(regs)
 817 #endif
 818
 819 extern int perf_output_begin(struct perf_output_handle *handle,
 820                              struct perf_event *event, unsigned int size,
 821                              int nmi, int sample);
 822 extern void perf_output_end(struct perf_output_handle *handle);
 823 extern void perf_output_copy(struct perf_output_handle *handle,
 824                              const void *buf, unsigned int len);
 825 #else
 826 static inline void
 827 perf_event_task_sched_in(struct task_struct *task, int cpu)             { }
 828 static inline void
 829 perf_event_task_sched_out(struct task_struct *task,
 830                             struct task_struct *next, int cpu)          { }
 831 static inline void
 832 perf_event_task_tick(struct task_struct *task, int cpu)                 { }
 833 static inline int perf_event_init_task(struct task_struct *child)       { return 0; }
 834 static inline void perf_event_exit_task(struct task_struct *child)      { }
 835 static inline void perf_event_free_task(struct task_struct *task)       { }
 836 static inline void perf_event_do_pending(void)                          { }
 837 static inline void perf_event_print_debug(void)                         { }
 838 static inline void perf_disable(void)                                   { }
 839 static inline void perf_enable(void)                                    { }
 840 static inline int perf_event_task_disable(void)                         { return -EINVAL; }
 841 static inline int perf_event_task_enable(void)                          { return -EINVAL; }
 842
 843 static inline void
 844 perf_sw_event(u32 event_id, u64 nr, int nmi,
 845                      struct pt_regs *regs, u64 addr)                    { }
 846
 847 static inline void perf_event_mmap(struct vm_area_struct *vma)          { }
 848 static inline void perf_event_comm(struct task_struct *tsk)             { }
 849 static inline void perf_event_fork(struct task_struct *tsk)             { }
 850 static inline void perf_event_init(void)                                { }
 851
 852 #endif
 853
 854 #define perf_output_put(handle, x) \
 855         perf_output_copy((handle), &(x), sizeof(x))
 856
 857 #endif /* __KERNEL__ */
 858 #endif /* _LINUX_PERF_EVENT_H */