include/linux/trace_events.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2
   3 #ifndef _LINUX_TRACE_EVENT_H
   4 #define _LINUX_TRACE_EVENT_H
   5
   6 #include <linux/ring_buffer.h>
   7 #include <linux/trace_seq.h>
   8 #include <linux/percpu.h>
   9 #include <linux/hardirq.h>
  10 #include <linux/perf_event.h>
  11 #include <linux/tracepoint.h>
  12
  13 struct trace_array;
  14 struct array_buffer;
  15 struct tracer;
  16 struct dentry;
  17 struct bpf_prog;
  18
  19 const char *trace_print_flags_seq(struct trace_seq *p, const char *delim,
  20                                   unsigned long flags,
  21                                   const struct trace_print_flags *flag_array);
  22
  23 const char *trace_print_symbols_seq(struct trace_seq *p, unsigned long val,
  24                                     const struct trace_print_flags *symbol_array);
  25
  26 #if BITS_PER_LONG == 32
  27 const char *trace_print_flags_seq_u64(struct trace_seq *p, const char *delim,
  28                       unsigned long long flags,
  29                       const struct trace_print_flags_u64 *flag_array);
  30
  31 const char *trace_print_symbols_seq_u64(struct trace_seq *p,
  32                                         unsigned long long val,
  33                                         const struct trace_print_flags_u64
  34                                                                  *symbol_array);
  35 #endif
  36
  37 const char *trace_print_bitmask_seq(struct trace_seq *p, void *bitmask_ptr,
  38                                     unsigned int bitmask_size);
  39
  40 const char *trace_print_hex_seq(struct trace_seq *p,
  41                                 const unsigned char *buf, int len,
  42                                 bool concatenate);
  43
  44 const char *trace_print_array_seq(struct trace_seq *p,
  45                                    const void *buf, int count,
  46                                    size_t el_size);
  47
  48 const char *
  49 trace_print_hex_dump_seq(struct trace_seq *p, const char *prefix_str,
  50                          int prefix_type, int rowsize, int groupsize,
  51                          const void *buf, size_t len, bool ascii);
  52
  53 struct trace_iterator;
  54 struct trace_event;
  55
  56 int trace_raw_output_prep(struct trace_iterator *iter,
  57                           struct trace_event *event);
  58 extern __printf(2, 3)
  59 void trace_event_printf(struct trace_iterator *iter, const char *fmt, ...);
  60
  61 /*
  62  * The trace entry - the most basic unit of tracing. This is what
  63  * is printed in the end as a single line in the trace output, such as:
  64  *
  65  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
  66  */
  67 struct trace_entry {
  68         unsigned short          type;
  69         unsigned char           flags;
  70         unsigned char           preempt_count;
  71         int                     pid;
  72 };
  73
  74 #define TRACE_EVENT_TYPE_MAX                                            \
  75         ((1 << (sizeof(((struct trace_entry *)0)->type) * 8)) - 1)
  76
  77 /*
  78  * Trace iterator - used by printout routines who present trace
  79  * results to users and which routines might sleep, etc:
  80  */
  81 struct trace_iterator {
  82         struct trace_array      *tr;
  83         struct tracer           *trace;
  84         struct array_buffer     *array_buffer;
  85         void                    *private;
  86         int                     cpu_file;
  87         struct mutex            mutex;
  88         struct ring_buffer_iter **buffer_iter;
  89         unsigned long           iter_flags;
  90         void                    *temp;  /* temp holder */
  91         unsigned int            temp_size;
  92         char                    *fmt;   /* modified format holder */
  93         unsigned int            fmt_size;
  94
  95         /* trace_seq for __print_flags() and __print_symbolic() etc. */
  96         struct trace_seq        tmp_seq;
  97
  98         cpumask_var_t           started;
  99
 100         /* it's true when current open file is snapshot */
 101         bool                    snapshot;
 102
 103         /* The below is zeroed out in pipe_read */
 104         struct trace_seq        seq;
 105         struct trace_entry      *ent;
 106         unsigned long           lost_events;
 107         int                     leftover;
 108         int                     ent_size;
 109         int                     cpu;
 110         u64                     ts;
 111
 112         loff_t                  pos;
 113         long                    idx;
 114
 115         /* All new field here will be zeroed out in pipe_read */
 116 };
 117
 118 enum trace_iter_flags {
 119         TRACE_FILE_LAT_FMT      = 1,
 120         TRACE_FILE_ANNOTATE     = 2,
 121         TRACE_FILE_TIME_IN_NS   = 4,
 122 };
 123
 124
 125 typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
 126                                       int flags, struct trace_event *event);
 127
 128 struct trace_event_functions {
 129         trace_print_func        trace;
 130         trace_print_func        raw;
 131         trace_print_func        hex;
 132         trace_print_func        binary;
 133 };
 134
 135 struct trace_event {
 136         struct hlist_node               node;
 137         struct list_head                list;
 138         int                             type;
 139         struct trace_event_functions    *funcs;
 140 };
 141
 142 extern int register_trace_event(struct trace_event *event);
 143 extern int unregister_trace_event(struct trace_event *event);
 144
 145 /* Return values for print_line callback */
 146 enum print_line_t {
 147         TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
 148         TRACE_TYPE_HANDLED      = 1,
 149         TRACE_TYPE_UNHANDLED    = 2,    /* Relay to other output functions */
 150         TRACE_TYPE_NO_CONSUME   = 3     /* Handled but ask to not consume */
 151 };
 152
 153 enum print_line_t trace_handle_return(struct trace_seq *s);
 154
 155 static inline void tracing_generic_entry_update(struct trace_entry *entry,
 156                                                 unsigned short type,
 157                                                 unsigned int trace_ctx)
 158 {
 159         entry->preempt_count            = trace_ctx & 0xff;
 160         entry->pid                      = current->pid;
 161         entry->type                     = type;
 162         entry->flags =                  trace_ctx >> 16;
 163 }
 164
 165 unsigned int tracing_gen_ctx_irq_test(unsigned int irqs_status);
 166
 167 enum trace_flag_type {
 168         TRACE_FLAG_IRQS_OFF             = 0x01,
 169         TRACE_FLAG_IRQS_NOSUPPORT       = 0x02,
 170         TRACE_FLAG_NEED_RESCHED         = 0x04,
 171         TRACE_FLAG_HARDIRQ              = 0x08,
 172         TRACE_FLAG_SOFTIRQ              = 0x10,
 173         TRACE_FLAG_PREEMPT_RESCHED      = 0x20,
 174         TRACE_FLAG_NMI                  = 0x40,
 175 };
 176
 177 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
 178 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
 179 {
 180         unsigned int irq_status = irqs_disabled_flags(irqflags) ?
 181                 TRACE_FLAG_IRQS_OFF : 0;
 182         return tracing_gen_ctx_irq_test(irq_status);
 183 }
 184 static inline unsigned int tracing_gen_ctx(void)
 185 {
 186         unsigned long irqflags;
 187
 188         local_save_flags(irqflags);
 189         return tracing_gen_ctx_flags(irqflags);
 190 }
 191 #else
 192
 193 static inline unsigned int tracing_gen_ctx_flags(unsigned long irqflags)
 194 {
 195         return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
 196 }
 197 static inline unsigned int tracing_gen_ctx(void)
 198 {
 199         return tracing_gen_ctx_irq_test(TRACE_FLAG_IRQS_NOSUPPORT);
 200 }
 201 #endif
 202
 203 static inline unsigned int tracing_gen_ctx_dec(void)
 204 {
 205         unsigned int trace_ctx;
 206
 207         trace_ctx = tracing_gen_ctx();
 208         /*
 209          * Subtract one from the preemption counter if preemption is enabled,
 210          * see trace_event_buffer_reserve()for details.
 211          */
 212         if (IS_ENABLED(CONFIG_PREEMPTION))
 213                 trace_ctx--;
 214         return trace_ctx;
 215 }
 216
 217 struct trace_event_file;
 218
 219 struct ring_buffer_event *
 220 trace_event_buffer_lock_reserve(struct trace_buffer **current_buffer,
 221                                 struct trace_event_file *trace_file,
 222                                 int type, unsigned long len,
 223                                 unsigned int trace_ctx);
 224
 225 #define TRACE_RECORD_CMDLINE    BIT(0)
 226 #define TRACE_RECORD_TGID       BIT(1)
 227
 228 void tracing_record_taskinfo(struct task_struct *task, int flags);
 229 void tracing_record_taskinfo_sched_switch(struct task_struct *prev,
 230                                           struct task_struct *next, int flags);
 231
 232 void tracing_record_cmdline(struct task_struct *task);
 233 void tracing_record_tgid(struct task_struct *task);
 234
 235 int trace_output_call(struct trace_iterator *iter, char *name, char *fmt, ...);
 236
 237 struct event_filter;
 238
 239 enum trace_reg {
 240         TRACE_REG_REGISTER,
 241         TRACE_REG_UNREGISTER,
 242 #ifdef CONFIG_PERF_EVENTS
 243         TRACE_REG_PERF_REGISTER,
 244         TRACE_REG_PERF_UNREGISTER,
 245         TRACE_REG_PERF_OPEN,
 246         TRACE_REG_PERF_CLOSE,
 247         /*
 248          * These (ADD/DEL) use a 'boolean' return value, where 1 (true) means a
 249          * custom action was taken and the default action is not to be
 250          * performed.
 251          */
 252         TRACE_REG_PERF_ADD,
 253         TRACE_REG_PERF_DEL,
 254 #endif
 255 };
 256
 257 struct trace_event_call;
 258
 259 #define TRACE_FUNCTION_TYPE ((const char *)~0UL)
 260
 261 struct trace_event_fields {
 262         const char *type;
 263         union {
 264                 struct {
 265                         const char *name;
 266                         const int  size;
 267                         const int  align;
 268                         const int  is_signed;
 269                         const int  filter_type;
 270                 };
 271                 int (*define_fields)(struct trace_event_call *);
 272         };
 273 };
 274
 275 struct trace_event_class {
 276         const char              *system;
 277         void                    *probe;
 278 #ifdef CONFIG_PERF_EVENTS
 279         void                    *perf_probe;
 280 #endif
 281         int                     (*reg)(struct trace_event_call *event,
 282                                        enum trace_reg type, void *data);
 283         struct trace_event_fields *fields_array;
 284         struct list_head        *(*get_fields)(struct trace_event_call *);
 285         struct list_head        fields;
 286         int                     (*raw_init)(struct trace_event_call *);
 287 };
 288
 289 extern int trace_event_reg(struct trace_event_call *event,
 290                             enum trace_reg type, void *data);
 291
 292 struct trace_event_buffer {
 293         struct trace_buffer             *buffer;
 294         struct ring_buffer_event        *event;
 295         struct trace_event_file         *trace_file;
 296         void                            *entry;
 297         unsigned int                    trace_ctx;
 298         struct pt_regs                  *regs;
 299 };
 300
 301 void *trace_event_buffer_reserve(struct trace_event_buffer *fbuffer,
 302                                   struct trace_event_file *trace_file,
 303                                   unsigned long len);
 304
 305 void trace_event_buffer_commit(struct trace_event_buffer *fbuffer);
 306
 307 enum {
 308         TRACE_EVENT_FL_FILTERED_BIT,
 309         TRACE_EVENT_FL_CAP_ANY_BIT,
 310         TRACE_EVENT_FL_NO_SET_FILTER_BIT,
 311         TRACE_EVENT_FL_IGNORE_ENABLE_BIT,
 312         TRACE_EVENT_FL_TRACEPOINT_BIT,
 313         TRACE_EVENT_FL_KPROBE_BIT,
 314         TRACE_EVENT_FL_UPROBE_BIT,
 315 };
 316
 317 /*
 318  * Event flags:
 319  *  FILTERED      - The event has a filter attached
 320  *  CAP_ANY       - Any user can enable for perf
 321  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 322  *  IGNORE_ENABLE - For trace internal events, do not enable with debugfs file
 323  *  TRACEPOINT    - Event is a tracepoint
 324  *  KPROBE        - Event is a kprobe
 325  *  UPROBE        - Event is a uprobe
 326  */
 327 enum {
 328         TRACE_EVENT_FL_FILTERED         = (1 << TRACE_EVENT_FL_FILTERED_BIT),
 329         TRACE_EVENT_FL_CAP_ANY          = (1 << TRACE_EVENT_FL_CAP_ANY_BIT),
 330         TRACE_EVENT_FL_NO_SET_FILTER    = (1 << TRACE_EVENT_FL_NO_SET_FILTER_BIT),
 331         TRACE_EVENT_FL_IGNORE_ENABLE    = (1 << TRACE_EVENT_FL_IGNORE_ENABLE_BIT),
 332         TRACE_EVENT_FL_TRACEPOINT       = (1 << TRACE_EVENT_FL_TRACEPOINT_BIT),
 333         TRACE_EVENT_FL_KPROBE           = (1 << TRACE_EVENT_FL_KPROBE_BIT),
 334         TRACE_EVENT_FL_UPROBE           = (1 << TRACE_EVENT_FL_UPROBE_BIT),
 335 };
 336
 337 #define TRACE_EVENT_FL_UKPROBE (TRACE_EVENT_FL_KPROBE | TRACE_EVENT_FL_UPROBE)
 338
 339 struct trace_event_call {
 340         struct list_head        list;
 341         struct trace_event_class *class;
 342         union {
 343                 char                    *name;
 344                 /* Set TRACE_EVENT_FL_TRACEPOINT flag when using "tp" */
 345                 struct tracepoint       *tp;
 346         };
 347         struct trace_event      event;
 348         char                    *print_fmt;
 349         struct event_filter     *filter;
 350         void                    *mod;
 351         void                    *data;
 352
 353         /* See the TRACE_EVENT_FL_* flags above */
 354         int                     flags; /* static flags of different events */
 355
 356 #ifdef CONFIG_PERF_EVENTS
 357         int                             perf_refcount;
 358         struct hlist_head __percpu      *perf_events;
 359         struct bpf_prog_array __rcu     *prog_array;
 360
 361         int     (*perf_perm)(struct trace_event_call *,
 362                              struct perf_event *);
 363 #endif
 364 };
 365
 366 #ifdef CONFIG_PERF_EVENTS
 367 static inline bool bpf_prog_array_valid(struct trace_event_call *call)
 368 {
 369         /*
 370          * This inline function checks whether call->prog_array
 371          * is valid or not. The function is called in various places,
 372          * outside rcu_read_lock/unlock, as a heuristic to speed up execution.
 373          *
 374          * If this function returns true, and later call->prog_array
 375          * becomes false inside rcu_read_lock/unlock region,
 376          * we bail out then. If this function return false,
 377          * there is a risk that we might miss a few events if the checking
 378          * were delayed until inside rcu_read_lock/unlock region and
 379          * call->prog_array happened to become non-NULL then.
 380          *
 381          * Here, READ_ONCE() is used instead of rcu_access_pointer().
 382          * rcu_access_pointer() requires the actual definition of
 383          * "struct bpf_prog_array" while READ_ONCE() only needs
 384          * a declaration of the same type.
 385          */
 386         return !!READ_ONCE(call->prog_array);
 387 }
 388 #endif
 389
 390 static inline const char *
 391 trace_event_name(struct trace_event_call *call)
 392 {
 393         if (call->flags & TRACE_EVENT_FL_TRACEPOINT)
 394                 return call->tp ? call->tp->name : NULL;
 395         else
 396                 return call->name;
 397 }
 398
 399 static inline struct list_head *
 400 trace_get_fields(struct trace_event_call *event_call)
 401 {
 402         if (!event_call->class->get_fields)
 403                 return &event_call->class->fields;
 404         return event_call->class->get_fields(event_call);
 405 }
 406
 407 struct trace_subsystem_dir;
 408
 409 enum {
 410         EVENT_FILE_FL_ENABLED_BIT,
 411         EVENT_FILE_FL_RECORDED_CMD_BIT,
 412         EVENT_FILE_FL_RECORDED_TGID_BIT,
 413         EVENT_FILE_FL_FILTERED_BIT,
 414         EVENT_FILE_FL_NO_SET_FILTER_BIT,
 415         EVENT_FILE_FL_SOFT_MODE_BIT,
 416         EVENT_FILE_FL_SOFT_DISABLED_BIT,
 417         EVENT_FILE_FL_TRIGGER_MODE_BIT,
 418         EVENT_FILE_FL_TRIGGER_COND_BIT,
 419         EVENT_FILE_FL_PID_FILTER_BIT,
 420         EVENT_FILE_FL_WAS_ENABLED_BIT,
 421 };
 422
 423 extern struct trace_event_file *trace_get_event_file(const char *instance,
 424                                                      const char *system,
 425                                                      const char *event);
 426 extern void trace_put_event_file(struct trace_event_file *file);
 427
 428 #define MAX_DYNEVENT_CMD_LEN    (2048)
 429
 430 enum dynevent_type {
 431         DYNEVENT_TYPE_SYNTH = 1,
 432         DYNEVENT_TYPE_KPROBE,
 433         DYNEVENT_TYPE_NONE,
 434 };
 435
 436 struct dynevent_cmd;
 437
 438 typedef int (*dynevent_create_fn_t)(struct dynevent_cmd *cmd);
 439
 440 struct dynevent_cmd {
 441         struct seq_buf          seq;
 442         const char              *event_name;
 443         unsigned int            n_fields;
 444         enum dynevent_type      type;
 445         dynevent_create_fn_t    run_command;
 446         void                    *private_data;
 447 };
 448
 449 extern int dynevent_create(struct dynevent_cmd *cmd);
 450
 451 extern int synth_event_delete(const char *name);
 452
 453 extern void synth_event_cmd_init(struct dynevent_cmd *cmd,
 454                                  char *buf, int maxlen);
 455
 456 extern int __synth_event_gen_cmd_start(struct dynevent_cmd *cmd,
 457                                        const char *name,
 458                                        struct module *mod, ...);
 459
 460 #define synth_event_gen_cmd_start(cmd, name, mod, ...)  \
 461         __synth_event_gen_cmd_start(cmd, name, mod, ## __VA_ARGS__, NULL)
 462
 463 struct synth_field_desc {
 464         const char *type;
 465         const char *name;
 466 };
 467
 468 extern int synth_event_gen_cmd_array_start(struct dynevent_cmd *cmd,
 469                                            const char *name,
 470                                            struct module *mod,
 471                                            struct synth_field_desc *fields,
 472                                            unsigned int n_fields);
 473 extern int synth_event_create(const char *name,
 474                               struct synth_field_desc *fields,
 475                               unsigned int n_fields, struct module *mod);
 476
 477 extern int synth_event_add_field(struct dynevent_cmd *cmd,
 478                                  const char *type,
 479                                  const char *name);
 480 extern int synth_event_add_field_str(struct dynevent_cmd *cmd,
 481                                      const char *type_name);
 482 extern int synth_event_add_fields(struct dynevent_cmd *cmd,
 483                                   struct synth_field_desc *fields,
 484                                   unsigned int n_fields);
 485
 486 #define synth_event_gen_cmd_end(cmd)    \
 487         dynevent_create(cmd)
 488
 489 struct synth_event;
 490
 491 struct synth_event_trace_state {
 492         struct trace_event_buffer fbuffer;
 493         struct synth_trace_event *entry;
 494         struct trace_buffer *buffer;
 495         struct synth_event *event;
 496         unsigned int cur_field;
 497         unsigned int n_u64;
 498         bool disabled;
 499         bool add_next;
 500         bool add_name;
 501 };
 502
 503 extern int synth_event_trace(struct trace_event_file *file,
 504                              unsigned int n_vals, ...);
 505 extern int synth_event_trace_array(struct trace_event_file *file, u64 *vals,
 506                                    unsigned int n_vals);
 507 extern int synth_event_trace_start(struct trace_event_file *file,
 508                                    struct synth_event_trace_state *trace_state);
 509 extern int synth_event_add_next_val(u64 val,
 510                                     struct synth_event_trace_state *trace_state);
 511 extern int synth_event_add_val(const char *field_name, u64 val,
 512                                struct synth_event_trace_state *trace_state);
 513 extern int synth_event_trace_end(struct synth_event_trace_state *trace_state);
 514
 515 extern int kprobe_event_delete(const char *name);
 516
 517 extern void kprobe_event_cmd_init(struct dynevent_cmd *cmd,
 518                                   char *buf, int maxlen);
 519
 520 #define kprobe_event_gen_cmd_start(cmd, name, loc, ...)                 \
 521         __kprobe_event_gen_cmd_start(cmd, false, name, loc, ## __VA_ARGS__, NULL)
 522
 523 #define kretprobe_event_gen_cmd_start(cmd, name, loc, ...)              \
 524         __kprobe_event_gen_cmd_start(cmd, true, name, loc, ## __VA_ARGS__, NULL)
 525
 526 extern int __kprobe_event_gen_cmd_start(struct dynevent_cmd *cmd,
 527                                         bool kretprobe,
 528                                         const char *name,
 529                                         const char *loc, ...);
 530
 531 #define kprobe_event_add_fields(cmd, ...)       \
 532         __kprobe_event_add_fields(cmd, ## __VA_ARGS__, NULL)
 533
 534 #define kprobe_event_add_field(cmd, field)      \
 535         __kprobe_event_add_fields(cmd, field, NULL)
 536
 537 extern int __kprobe_event_add_fields(struct dynevent_cmd *cmd, ...);
 538
 539 #define kprobe_event_gen_cmd_end(cmd)           \
 540         dynevent_create(cmd)
 541
 542 #define kretprobe_event_gen_cmd_end(cmd)        \
 543         dynevent_create(cmd)
 544
 545 /*
 546  * Event file flags:
 547  *  ENABLED       - The event is enabled
 548  *  RECORDED_CMD  - The comms should be recorded at sched_switch
 549  *  RECORDED_TGID - The tgids should be recorded at sched_switch
 550  *  FILTERED      - The event has a filter attached
 551  *  NO_SET_FILTER - Set when filter has error and is to be ignored
 552  *  SOFT_MODE     - The event is enabled/disabled by SOFT_DISABLED
 553  *  SOFT_DISABLED - When set, do not trace the event (even though its
 554  *                   tracepoint may be enabled)
 555  *  TRIGGER_MODE  - When set, invoke the triggers associated with the event
 556  *  TRIGGER_COND  - When set, one or more triggers has an associated filter
 557  *  PID_FILTER    - When set, the event is filtered based on pid
 558  *  WAS_ENABLED   - Set when enabled to know to clear trace on module removal
 559  */
 560 enum {
 561         EVENT_FILE_FL_ENABLED           = (1 << EVENT_FILE_FL_ENABLED_BIT),
 562         EVENT_FILE_FL_RECORDED_CMD      = (1 << EVENT_FILE_FL_RECORDED_CMD_BIT),
 563         EVENT_FILE_FL_RECORDED_TGID     = (1 << EVENT_FILE_FL_RECORDED_TGID_BIT),
 564         EVENT_FILE_FL_FILTERED          = (1 << EVENT_FILE_FL_FILTERED_BIT),
 565         EVENT_FILE_FL_NO_SET_FILTER     = (1 << EVENT_FILE_FL_NO_SET_FILTER_BIT),
 566         EVENT_FILE_FL_SOFT_MODE         = (1 << EVENT_FILE_FL_SOFT_MODE_BIT),
 567         EVENT_FILE_FL_SOFT_DISABLED     = (1 << EVENT_FILE_FL_SOFT_DISABLED_BIT),
 568         EVENT_FILE_FL_TRIGGER_MODE      = (1 << EVENT_FILE_FL_TRIGGER_MODE_BIT),
 569         EVENT_FILE_FL_TRIGGER_COND      = (1 << EVENT_FILE_FL_TRIGGER_COND_BIT),
 570         EVENT_FILE_FL_PID_FILTER        = (1 << EVENT_FILE_FL_PID_FILTER_BIT),
 571         EVENT_FILE_FL_WAS_ENABLED       = (1 << EVENT_FILE_FL_WAS_ENABLED_BIT),
 572 };
 573
 574 struct trace_event_file {
 575         struct list_head                list;
 576         struct trace_event_call         *event_call;
 577         struct event_filter __rcu       *filter;
 578         struct dentry                   *dir;
 579         struct trace_array              *tr;
 580         struct trace_subsystem_dir      *system;
 581         struct list_head                triggers;
 582
 583         /*
 584          * 32 bit flags:
 585          *   bit 0:             enabled
 586          *   bit 1:             enabled cmd record
 587          *   bit 2:             enable/disable with the soft disable bit
 588          *   bit 3:             soft disabled
 589          *   bit 4:             trigger enabled
 590          *
 591          * Note: The bits must be set atomically to prevent races
 592          * from other writers. Reads of flags do not need to be in
 593          * sync as they occur in critical sections. But the way flags
 594          * is currently used, these changes do not affect the code
 595          * except that when a change is made, it may have a slight
 596          * delay in propagating the changes to other CPUs due to
 597          * caching and such. Which is mostly OK ;-)
 598          */
 599         unsigned long           flags;
 600         atomic_t                sm_ref; /* soft-mode reference counter */
 601         atomic_t                tm_ref; /* trigger-mode reference counter */
 602 };
 603
 604 #define __TRACE_EVENT_FLAGS(name, value)                                \
 605         static int __init trace_init_flags_##name(void)                 \
 606         {                                                               \
 607                 event_##name.flags |= value;                            \
 608                 return 0;                                               \
 609         }                                                               \
 610         early_initcall(trace_init_flags_##name);
 611
 612 #define __TRACE_EVENT_PERF_PERM(name, expr...)                          \
 613         static int perf_perm_##name(struct trace_event_call *tp_event, \
 614                                     struct perf_event *p_event)         \
 615         {                                                               \
 616                 return ({ expr; });                                     \
 617         }                                                               \
 618         static int __init trace_init_perf_perm_##name(void)             \
 619         {                                                               \
 620                 event_##name.perf_perm = &perf_perm_##name;             \
 621                 return 0;                                               \
 622         }                                                               \
 623         early_initcall(trace_init_perf_perm_##name);
 624
 625 #define PERF_MAX_TRACE_SIZE     2048
 626
 627 #define MAX_FILTER_STR_VAL      256     /* Should handle KSYM_SYMBOL_LEN */
 628
 629 enum event_trigger_type {
 630         ETT_NONE                = (0),
 631         ETT_TRACE_ONOFF         = (1 << 0),
 632         ETT_SNAPSHOT            = (1 << 1),
 633         ETT_STACKTRACE          = (1 << 2),
 634         ETT_EVENT_ENABLE        = (1 << 3),
 635         ETT_EVENT_HIST          = (1 << 4),
 636         ETT_HIST_ENABLE         = (1 << 5),
 637 };
 638
 639 extern int filter_match_preds(struct event_filter *filter, void *rec);
 640
 641 extern enum event_trigger_type
 642 event_triggers_call(struct trace_event_file *file,
 643                     struct trace_buffer *buffer, void *rec,
 644                     struct ring_buffer_event *event);
 645 extern void
 646 event_triggers_post_call(struct trace_event_file *file,
 647                          enum event_trigger_type tt);
 648
 649 bool trace_event_ignore_this_pid(struct trace_event_file *trace_file);
 650
 651 /**
 652  * trace_trigger_soft_disabled - do triggers and test if soft disabled
 653  * @file: The file pointer of the event to test
 654  *
 655  * If any triggers without filters are attached to this event, they
 656  * will be called here. If the event is soft disabled and has no
 657  * triggers that require testing the fields, it will return true,
 658  * otherwise false.
 659  */
 660 static inline bool
 661 trace_trigger_soft_disabled(struct trace_event_file *file)
 662 {
 663         unsigned long eflags = file->flags;
 664
 665         if (!(eflags & EVENT_FILE_FL_TRIGGER_COND)) {
 666                 if (eflags & EVENT_FILE_FL_TRIGGER_MODE)
 667                         event_triggers_call(file, NULL, NULL, NULL);
 668                 if (eflags & EVENT_FILE_FL_SOFT_DISABLED)
 669                         return true;
 670                 if (eflags & EVENT_FILE_FL_PID_FILTER)
 671                         return trace_event_ignore_this_pid(file);
 672         }
 673         return false;
 674 }
 675
 676 #ifdef CONFIG_BPF_EVENTS
 677 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx);
 678 int perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog);
 679 void perf_event_detach_bpf_prog(struct perf_event *event);
 680 int perf_event_query_prog_array(struct perf_event *event, void __user *info);
 681 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
 682 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog);
 683 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name);
 684 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp);
 685 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
 686                             u32 *fd_type, const char **buf,
 687                             u64 *probe_offset, u64 *probe_addr);
 688 #else
 689 static inline unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
 690 {
 691         return 1;
 692 }
 693
 694 static inline int
 695 perf_event_attach_bpf_prog(struct perf_event *event, struct bpf_prog *prog)
 696 {
 697         return -EOPNOTSUPP;
 698 }
 699
 700 static inline void perf_event_detach_bpf_prog(struct perf_event *event) { }
 701
 702 static inline int
 703 perf_event_query_prog_array(struct perf_event *event, void __user *info)
 704 {
 705         return -EOPNOTSUPP;
 706 }
 707 static inline int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *p)
 708 {
 709         return -EOPNOTSUPP;
 710 }
 711 static inline int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *p)
 712 {
 713         return -EOPNOTSUPP;
 714 }
 715 static inline struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
 716 {
 717         return NULL;
 718 }
 719 static inline void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
 720 {
 721 }
 722 static inline int bpf_get_perf_event_info(const struct perf_event *event,
 723                                           u32 *prog_id, u32 *fd_type,
 724                                           const char **buf, u64 *probe_offset,
 725                                           u64 *probe_addr)
 726 {
 727         return -EOPNOTSUPP;
 728 }
 729 #endif
 730
 731 enum {
 732         FILTER_OTHER = 0,
 733         FILTER_STATIC_STRING,
 734         FILTER_DYN_STRING,
 735         FILTER_PTR_STRING,
 736         FILTER_TRACE_FN,
 737         FILTER_COMM,
 738         FILTER_CPU,
 739 };
 740
 741 extern int trace_event_raw_init(struct trace_event_call *call);
 742 extern int trace_define_field(struct trace_event_call *call, const char *type,
 743                               const char *name, int offset, int size,
 744                               int is_signed, int filter_type);
 745 extern int trace_add_event_call(struct trace_event_call *call);
 746 extern int trace_remove_event_call(struct trace_event_call *call);
 747 extern int trace_event_get_offsets(struct trace_event_call *call);
 748
 749 #define is_signed_type(type)    (((type)(-1)) < (type)1)
 750
 751 int ftrace_set_clr_event(struct trace_array *tr, char *buf, int set);
 752 int trace_set_clr_event(const char *system, const char *event, int set);
 753 int trace_array_set_clr_event(struct trace_array *tr, const char *system,
 754                 const char *event, bool enable);
 755 /*
 756  * The double __builtin_constant_p is because gcc will give us an error
 757  * if we try to allocate the static variable to fmt if it is not a
 758  * constant. Even with the outer if statement optimizing out.
 759  */
 760 #define event_trace_printk(ip, fmt, args...)                            \
 761 do {                                                                    \
 762         __trace_printk_check_format(fmt, ##args);                       \
 763         tracing_record_cmdline(current);                                \
 764         if (__builtin_constant_p(fmt)) {                                \
 765                 static const char *trace_printk_fmt                     \
 766                   __section("__trace_printk_fmt") =                     \
 767                         __builtin_constant_p(fmt) ? fmt : NULL;         \
 768                                                                         \
 769                 __trace_bprintk(ip, trace_printk_fmt, ##args);          \
 770         } else                                                          \
 771                 __trace_printk(ip, fmt, ##args);                        \
 772 } while (0)
 773
 774 #ifdef CONFIG_PERF_EVENTS
 775 struct perf_event;
 776
 777 DECLARE_PER_CPU(struct pt_regs, perf_trace_regs);
 778 DECLARE_PER_CPU(int, bpf_kprobe_override);
 779
 780 extern int  perf_trace_init(struct perf_event *event);
 781 extern void perf_trace_destroy(struct perf_event *event);
 782 extern int  perf_trace_add(struct perf_event *event, int flags);
 783 extern void perf_trace_del(struct perf_event *event, int flags);
 784 #ifdef CONFIG_KPROBE_EVENTS
 785 extern int  perf_kprobe_init(struct perf_event *event, bool is_retprobe);
 786 extern void perf_kprobe_destroy(struct perf_event *event);
 787 extern int bpf_get_kprobe_info(const struct perf_event *event,
 788                                u32 *fd_type, const char **symbol,
 789                                u64 *probe_offset, u64 *probe_addr,
 790                                bool perf_type_tracepoint);
 791 #endif
 792 #ifdef CONFIG_UPROBE_EVENTS
 793 extern int  perf_uprobe_init(struct perf_event *event,
 794                              unsigned long ref_ctr_offset, bool is_retprobe);
 795 extern void perf_uprobe_destroy(struct perf_event *event);
 796 extern int bpf_get_uprobe_info(const struct perf_event *event,
 797                                u32 *fd_type, const char **filename,
 798                                u64 *probe_offset, bool perf_type_tracepoint);
 799 #endif
 800 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
 801                                      char *filter_str);
 802 extern void ftrace_profile_free_filter(struct perf_event *event);
 803 void perf_trace_buf_update(void *record, u16 type);
 804 void *perf_trace_buf_alloc(int size, struct pt_regs **regs, int *rctxp);
 805
 806 void bpf_trace_run1(struct bpf_prog *prog, u64 arg1);
 807 void bpf_trace_run2(struct bpf_prog *prog, u64 arg1, u64 arg2);
 808 void bpf_trace_run3(struct bpf_prog *prog, u64 arg1, u64 arg2,
 809                     u64 arg3);
 810 void bpf_trace_run4(struct bpf_prog *prog, u64 arg1, u64 arg2,
 811                     u64 arg3, u64 arg4);
 812 void bpf_trace_run5(struct bpf_prog *prog, u64 arg1, u64 arg2,
 813                     u64 arg3, u64 arg4, u64 arg5);
 814 void bpf_trace_run6(struct bpf_prog *prog, u64 arg1, u64 arg2,
 815                     u64 arg3, u64 arg4, u64 arg5, u64 arg6);
 816 void bpf_trace_run7(struct bpf_prog *prog, u64 arg1, u64 arg2,
 817                     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7);
 818 void bpf_trace_run8(struct bpf_prog *prog, u64 arg1, u64 arg2,
 819                     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
 820                     u64 arg8);
 821 void bpf_trace_run9(struct bpf_prog *prog, u64 arg1, u64 arg2,
 822                     u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
 823                     u64 arg8, u64 arg9);
 824 void bpf_trace_run10(struct bpf_prog *prog, u64 arg1, u64 arg2,
 825                      u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
 826                      u64 arg8, u64 arg9, u64 arg10);
 827 void bpf_trace_run11(struct bpf_prog *prog, u64 arg1, u64 arg2,
 828                      u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
 829                      u64 arg8, u64 arg9, u64 arg10, u64 arg11);
 830 void bpf_trace_run12(struct bpf_prog *prog, u64 arg1, u64 arg2,
 831                      u64 arg3, u64 arg4, u64 arg5, u64 arg6, u64 arg7,
 832                      u64 arg8, u64 arg9, u64 arg10, u64 arg11, u64 arg12);
 833 void perf_trace_run_bpf_submit(void *raw_data, int size, int rctx,
 834                                struct trace_event_call *call, u64 count,
 835                                struct pt_regs *regs, struct hlist_head *head,
 836                                struct task_struct *task);
 837
 838 static inline void
 839 perf_trace_buf_submit(void *raw_data, int size, int rctx, u16 type,
 840                        u64 count, struct pt_regs *regs, void *head,
 841                        struct task_struct *task)
 842 {
 843         perf_tp_event(type, count, raw_data, size, regs, head, rctx, task);
 844 }
 845
 846 #endif
 847
 848 #endif /* _LINUX_TRACE_EVENT_H */