kernel/trace/trace.h

   1 #ifndef _LINUX_KERNEL_TRACE_H
   2 #define _LINUX_KERNEL_TRACE_H
   3
   4 #include <linux/fs.h>
   5 #include <asm/atomic.h>
   6 #include <linux/sched.h>
   7 #include <linux/clocksource.h>
   8 #include <linux/ring_buffer.h>
   9 #include <linux/mmiotrace.h>
  10 #include <linux/ftrace.h>
  11 #include <trace/boot.h>
  12
  13 enum trace_type {
  14         __TRACE_FIRST_TYPE = 0,
  15
  16         TRACE_FN,
  17         TRACE_CTX,
  18         TRACE_WAKE,
  19         TRACE_CONT,
  20         TRACE_STACK,
  21         TRACE_PRINT,
  22         TRACE_SPECIAL,
  23         TRACE_MMIO_RW,
  24         TRACE_MMIO_MAP,
  25         TRACE_BRANCH,
  26         TRACE_BOOT_CALL,
  27         TRACE_BOOT_RET,
  28         TRACE_FN_RET,
  29         TRACE_USER_STACK,
  30         TRACE_BTS,
  31
  32         __TRACE_LAST_TYPE
  33 };
  34
  35 /*
  36  * The trace entry - the most basic unit of tracing. This is what
  37  * is printed in the end as a single line in the trace output, such as:
  38  *
  39  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
  40  */
  41 struct trace_entry {
  42         unsigned char           type;
  43         unsigned char           cpu;
  44         unsigned char           flags;
  45         unsigned char           preempt_count;
  46         int                     pid;
  47         int                     tgid;
  48 };
  49
  50 /*
  51  * Function trace entry - function address and parent function addres:
  52  */
  53 struct ftrace_entry {
  54         struct trace_entry      ent;
  55         unsigned long           ip;
  56         unsigned long           parent_ip;
  57 };
  58
  59 /* Function return entry */
  60 struct ftrace_ret_entry {
  61         struct trace_entry      ent;
  62         unsigned long           ip;
  63         unsigned long           parent_ip;
  64         unsigned long long      calltime;
  65         unsigned long long      rettime;
  66         unsigned long           overrun;
  67 };
  68 extern struct tracer boot_tracer;
  69
  70 /*
  71  * Context switch trace entry - which task (and prio) we switched from/to:
  72  */
  73 struct ctx_switch_entry {
  74         struct trace_entry      ent;
  75         unsigned int            prev_pid;
  76         unsigned char           prev_prio;
  77         unsigned char           prev_state;
  78         unsigned int            next_pid;
  79         unsigned char           next_prio;
  80         unsigned char           next_state;
  81         unsigned int            next_cpu;
  82 };
  83
  84 /*
  85  * Special (free-form) trace entry:
  86  */
  87 struct special_entry {
  88         struct trace_entry      ent;
  89         unsigned long           arg1;
  90         unsigned long           arg2;
  91         unsigned long           arg3;
  92 };
  93
  94 /*
  95  * Stack-trace entry:
  96  */
  97
  98 #define FTRACE_STACK_ENTRIES    8
  99
 100 struct stack_entry {
 101         struct trace_entry      ent;
 102         unsigned long           caller[FTRACE_STACK_ENTRIES];
 103 };
 104
 105 struct userstack_entry {
 106         struct trace_entry      ent;
 107         unsigned long           caller[FTRACE_STACK_ENTRIES];
 108 };
 109
 110 /*
 111  * ftrace_printk entry:
 112  */
 113 struct print_entry {
 114         struct trace_entry      ent;
 115         unsigned long           ip;
 116         char                    buf[];
 117 };
 118
 119 #define TRACE_OLD_SIZE          88
 120
 121 struct trace_field_cont {
 122         unsigned char           type;
 123         /* Temporary till we get rid of this completely */
 124         char                    buf[TRACE_OLD_SIZE - 1];
 125 };
 126
 127 struct trace_mmiotrace_rw {
 128         struct trace_entry      ent;
 129         struct mmiotrace_rw     rw;
 130 };
 131
 132 struct trace_mmiotrace_map {
 133         struct trace_entry      ent;
 134         struct mmiotrace_map    map;
 135 };
 136
 137 struct trace_boot_call {
 138         struct trace_entry      ent;
 139         struct boot_trace_call boot_call;
 140 };
 141
 142 struct trace_boot_ret {
 143         struct trace_entry      ent;
 144         struct boot_trace_ret boot_ret;
 145 };
 146
 147 #define TRACE_FUNC_SIZE 30
 148 #define TRACE_FILE_SIZE 20
 149 struct trace_branch {
 150         struct trace_entry      ent;
 151         unsigned                line;
 152         char                    func[TRACE_FUNC_SIZE+1];
 153         char                    file[TRACE_FILE_SIZE+1];
 154         char                    correct;
 155 };
 156
 157 struct bts_entry {
 158         struct trace_entry      ent;
 159         unsigned long           from;
 160         unsigned long           to;
 161 };
 162
 163 /*
 164  * trace_flag_type is an enumeration that holds different
 165  * states when a trace occurs. These are:
 166  *  IRQS_OFF            - interrupts were disabled
 167  *  IRQS_NOSUPPORT      - arch does not support irqs_disabled_flags
 168  *  NEED_RESCED         - reschedule is requested
 169  *  HARDIRQ             - inside an interrupt handler
 170  *  SOFTIRQ             - inside a softirq handler
 171  *  CONT                - multiple entries hold the trace item
 172  */
 173 enum trace_flag_type {
 174         TRACE_FLAG_IRQS_OFF             = 0x01,
 175         TRACE_FLAG_IRQS_NOSUPPORT       = 0x02,
 176         TRACE_FLAG_NEED_RESCHED         = 0x04,
 177         TRACE_FLAG_HARDIRQ              = 0x08,
 178         TRACE_FLAG_SOFTIRQ              = 0x10,
 179         TRACE_FLAG_CONT                 = 0x20,
 180 };
 181
 182 #define TRACE_BUF_SIZE          1024
 183
 184 /*
 185  * The CPU trace array - it consists of thousands of trace entries
 186  * plus some other descriptor data: (for example which task started
 187  * the trace, etc.)
 188  */
 189 struct trace_array_cpu {
 190         atomic_t                disabled;
 191
 192         /* these fields get copied into max-trace: */
 193         unsigned long           trace_idx;
 194         unsigned long           overrun;
 195         unsigned long           saved_latency;
 196         unsigned long           critical_start;
 197         unsigned long           critical_end;
 198         unsigned long           critical_sequence;
 199         unsigned long           nice;
 200         unsigned long           policy;
 201         unsigned long           rt_priority;
 202         cycle_t                 preempt_timestamp;
 203         pid_t                   pid;
 204         uid_t                   uid;
 205         char                    comm[TASK_COMM_LEN];
 206 };
 207
 208 struct trace_iterator;
 209
 210 /*
 211  * The trace array - an array of per-CPU trace arrays. This is the
 212  * highest level data structure that individual tracers deal with.
 213  * They have on/off state as well:
 214  */
 215 struct trace_array {
 216         struct ring_buffer      *buffer;
 217         unsigned long           entries;
 218         int                     cpu;
 219         cycle_t                 time_start;
 220         struct task_struct      *waiter;
 221         struct trace_array_cpu  *data[NR_CPUS];
 222 };
 223
 224 #define FTRACE_CMP_TYPE(var, type) \
 225         __builtin_types_compatible_p(typeof(var), type *)
 226
 227 #undef IF_ASSIGN
 228 #define IF_ASSIGN(var, entry, etype, id)                \
 229         if (FTRACE_CMP_TYPE(var, etype)) {              \
 230                 var = (typeof(var))(entry);             \
 231                 WARN_ON(id && (entry)->type != id);     \
 232                 break;                                  \
 233         }
 234
 235 /* Will cause compile errors if type is not found. */
 236 extern void __ftrace_bad_type(void);
 237
 238 /*
 239  * The trace_assign_type is a verifier that the entry type is
 240  * the same as the type being assigned. To add new types simply
 241  * add a line with the following format:
 242  *
 243  * IF_ASSIGN(var, ent, type, id);
 244  *
 245  *  Where "type" is the trace type that includes the trace_entry
 246  *  as the "ent" item. And "id" is the trace identifier that is
 247  *  used in the trace_type enum.
 248  *
 249  *  If the type can have more than one id, then use zero.
 250  */
 251 #define trace_assign_type(var, ent)                                     \
 252         do {                                                            \
 253                 IF_ASSIGN(var, ent, struct ftrace_entry, TRACE_FN);     \
 254                 IF_ASSIGN(var, ent, struct ctx_switch_entry, 0);        \
 255                 IF_ASSIGN(var, ent, struct trace_field_cont, TRACE_CONT); \
 256                 IF_ASSIGN(var, ent, struct stack_entry, TRACE_STACK);   \
 257                 IF_ASSIGN(var, ent, struct userstack_entry, TRACE_USER_STACK);\
 258                 IF_ASSIGN(var, ent, struct print_entry, TRACE_PRINT);   \
 259                 IF_ASSIGN(var, ent, struct special_entry, 0);           \
 260                 IF_ASSIGN(var, ent, struct trace_mmiotrace_rw,          \
 261                           TRACE_MMIO_RW);                               \
 262                 IF_ASSIGN(var, ent, struct trace_mmiotrace_map,         \
 263                           TRACE_MMIO_MAP);                              \
 264                 IF_ASSIGN(var, ent, struct trace_boot_call, TRACE_BOOT_CALL);\
 265                 IF_ASSIGN(var, ent, struct trace_boot_ret, TRACE_BOOT_RET);\
 266                 IF_ASSIGN(var, ent, struct trace_branch, TRACE_BRANCH); \
 267                 IF_ASSIGN(var, ent, struct ftrace_ret_entry, TRACE_FN_RET);\
 268                 IF_ASSIGN(var, ent, struct bts_entry, TRACE_BTS);\
 269                 __ftrace_bad_type();                                    \
 270         } while (0)
 271
 272 /* Return values for print_line callback */
 273 enum print_line_t {
 274         TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
 275         TRACE_TYPE_HANDLED      = 1,
 276         TRACE_TYPE_UNHANDLED    = 2     /* Relay to other output functions */
 277 };
 278
 279
 280 /*
 281  * An option specific to a tracer. This is a boolean value.
 282  * The bit is the bit index that sets its value on the
 283  * flags value in struct tracer_flags.
 284  */
 285 struct tracer_opt {
 286         const char      *name; /* Will appear on the trace_options file */
 287         u32             bit; /* Mask assigned in val field in tracer_flags */
 288 };
 289
 290 /*
 291  * The set of specific options for a tracer. Your tracer
 292  * have to set the initial value of the flags val.
 293  */
 294 struct tracer_flags {
 295         u32                     val;
 296         struct tracer_opt       *opts;
 297 };
 298
 299 /* Makes more easy to define a tracer opt */
 300 #define TRACER_OPT(s, b)        .name = #s, .bit = b
 301
 302 /*
 303  * A specific tracer, represented by methods that operate on a trace array:
 304  */
 305 struct tracer {
 306         const char              *name;
 307         /* Your tracer should raise a warning if init fails */
 308         int                     (*init)(struct trace_array *tr);
 309         void                    (*reset)(struct trace_array *tr);
 310         void                    (*start)(struct trace_array *tr);
 311         void                    (*stop)(struct trace_array *tr);
 312         void                    (*open)(struct trace_iterator *iter);
 313         void                    (*pipe_open)(struct trace_iterator *iter);
 314         void                    (*close)(struct trace_iterator *iter);
 315         ssize_t                 (*read)(struct trace_iterator *iter,
 316                                         struct file *filp, char __user *ubuf,
 317                                         size_t cnt, loff_t *ppos);
 318 #ifdef CONFIG_FTRACE_STARTUP_TEST
 319         int                     (*selftest)(struct tracer *trace,
 320                                             struct trace_array *tr);
 321 #endif
 322         void                    (*print_header)(struct seq_file *m);
 323         enum print_line_t       (*print_line)(struct trace_iterator *iter);
 324         /* If you handled the flag setting, return 0 */
 325         int                     (*set_flag)(u32 old_flags, u32 bit, int set);
 326         struct tracer           *next;
 327         int                     print_max;
 328         struct tracer_flags     *flags;
 329 };
 330
 331 struct trace_seq {
 332         unsigned char           buffer[PAGE_SIZE];
 333         unsigned int            len;
 334         unsigned int            readpos;
 335 };
 336
 337 /*
 338  * Trace iterator - used by printout routines who present trace
 339  * results to users and which routines might sleep, etc:
 340  */
 341 struct trace_iterator {
 342         struct trace_array      *tr;
 343         struct tracer           *trace;
 344         void                    *private;
 345         struct ring_buffer_iter *buffer_iter[NR_CPUS];
 346
 347         /* The below is zeroed out in pipe_read */
 348         struct trace_seq        seq;
 349         struct trace_entry      *ent;
 350         int                     cpu;
 351         u64                     ts;
 352
 353         unsigned long           iter_flags;
 354         loff_t                  pos;
 355         long                    idx;
 356
 357         cpumask_t               started;
 358 };
 359
 360 int tracing_is_enabled(void);
 361 void trace_wake_up(void);
 362 void tracing_reset(struct trace_array *tr, int cpu);
 363 int tracing_open_generic(struct inode *inode, struct file *filp);
 364 struct dentry *tracing_init_dentry(void);
 365 void init_tracer_sysprof_debugfs(struct dentry *d_tracer);
 366
 367 struct trace_entry *tracing_get_trace_entry(struct trace_array *tr,
 368                                                 struct trace_array_cpu *data);
 369 void tracing_generic_entry_update(struct trace_entry *entry,
 370                                   unsigned long flags,
 371                                   int pc);
 372
 373 void ftrace(struct trace_array *tr,
 374                             struct trace_array_cpu *data,
 375                             unsigned long ip,
 376                             unsigned long parent_ip,
 377                             unsigned long flags, int pc);
 378 void tracing_sched_switch_trace(struct trace_array *tr,
 379                                 struct trace_array_cpu *data,
 380                                 struct task_struct *prev,
 381                                 struct task_struct *next,
 382                                 unsigned long flags, int pc);
 383 void tracing_record_cmdline(struct task_struct *tsk);
 384
 385 void tracing_sched_wakeup_trace(struct trace_array *tr,
 386                                 struct trace_array_cpu *data,
 387                                 struct task_struct *wakee,
 388                                 struct task_struct *cur,
 389                                 unsigned long flags, int pc);
 390 void trace_special(struct trace_array *tr,
 391                    struct trace_array_cpu *data,
 392                    unsigned long arg1,
 393                    unsigned long arg2,
 394                    unsigned long arg3, int pc);
 395 void trace_function(struct trace_array *tr,
 396                     struct trace_array_cpu *data,
 397                     unsigned long ip,
 398                     unsigned long parent_ip,
 399                     unsigned long flags, int pc);
 400 void
 401 trace_function_return(struct ftrace_retfunc *trace);
 402
 403 void trace_bts(struct trace_array *tr,
 404                unsigned long from,
 405                unsigned long to);
 406
 407 void tracing_start_cmdline_record(void);
 408 void tracing_stop_cmdline_record(void);
 409 void tracing_sched_switch_assign_trace(struct trace_array *tr);
 410 void tracing_stop_sched_switch_record(void);
 411 void tracing_start_sched_switch_record(void);
 412 int register_tracer(struct tracer *type);
 413 void unregister_tracer(struct tracer *type);
 414
 415 extern unsigned long nsecs_to_usecs(unsigned long nsecs);
 416
 417 extern unsigned long tracing_max_latency;
 418 extern unsigned long tracing_thresh;
 419
 420 void update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu);
 421 void update_max_tr_single(struct trace_array *tr,
 422                           struct task_struct *tsk, int cpu);
 423
 424 extern cycle_t ftrace_now(int cpu);
 425
 426 #ifdef CONFIG_FUNCTION_TRACER
 427 void tracing_start_function_trace(void);
 428 void tracing_stop_function_trace(void);
 429 #else
 430 # define tracing_start_function_trace()         do { } while (0)
 431 # define tracing_stop_function_trace()          do { } while (0)
 432 #endif
 433
 434 #ifdef CONFIG_CONTEXT_SWITCH_TRACER
 435 typedef void
 436 (*tracer_switch_func_t)(void *private,
 437                         void *__rq,
 438                         struct task_struct *prev,
 439                         struct task_struct *next);
 440
 441 struct tracer_switch_ops {
 442         tracer_switch_func_t            func;
 443         void                            *private;
 444         struct tracer_switch_ops        *next;
 445 };
 446
 447 #endif /* CONFIG_CONTEXT_SWITCH_TRACER */
 448
 449 #ifdef CONFIG_DYNAMIC_FTRACE
 450 extern unsigned long ftrace_update_tot_cnt;
 451 #define DYN_FTRACE_TEST_NAME trace_selftest_dynamic_test_func
 452 extern int DYN_FTRACE_TEST_NAME(void);
 453 #endif
 454
 455 #ifdef CONFIG_FTRACE_STARTUP_TEST
 456 extern int trace_selftest_startup_function(struct tracer *trace,
 457                                            struct trace_array *tr);
 458 extern int trace_selftest_startup_irqsoff(struct tracer *trace,
 459                                           struct trace_array *tr);
 460 extern int trace_selftest_startup_preemptoff(struct tracer *trace,
 461                                              struct trace_array *tr);
 462 extern int trace_selftest_startup_preemptirqsoff(struct tracer *trace,
 463                                                  struct trace_array *tr);
 464 extern int trace_selftest_startup_wakeup(struct tracer *trace,
 465                                          struct trace_array *tr);
 466 extern int trace_selftest_startup_nop(struct tracer *trace,
 467                                          struct trace_array *tr);
 468 extern int trace_selftest_startup_sched_switch(struct tracer *trace,
 469                                                struct trace_array *tr);
 470 extern int trace_selftest_startup_sysprof(struct tracer *trace,
 471                                                struct trace_array *tr);
 472 extern int trace_selftest_startup_branch(struct tracer *trace,
 473                                          struct trace_array *tr);
 474 #endif /* CONFIG_FTRACE_STARTUP_TEST */
 475
 476 extern void *head_page(struct trace_array_cpu *data);
 477 extern int trace_seq_printf(struct trace_seq *s, const char *fmt, ...);
 478 extern void trace_seq_print_cont(struct trace_seq *s,
 479                                  struct trace_iterator *iter);
 480
 481 extern int
 482 seq_print_ip_sym(struct trace_seq *s, unsigned long ip,
 483                 unsigned long sym_flags);
 484 extern ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf,
 485                                  size_t cnt);
 486 extern long ns2usecs(cycle_t nsec);
 487 extern int trace_vprintk(unsigned long ip, const char *fmt, va_list args);
 488
 489 extern unsigned long trace_flags;
 490
 491 /* Standard output formatting function used for function return traces */
 492 #ifdef CONFIG_FUNCTION_RET_TRACER
 493 extern enum print_line_t print_return_function(struct trace_iterator *iter);
 494 #else
 495 static inline enum print_line_t
 496 print_return_function(struct trace_iterator *iter)
 497 {
 498         return TRACE_TYPE_UNHANDLED;
 499 }
 500 #endif
 501
 502 /*
 503  * trace_iterator_flags is an enumeration that defines bit
 504  * positions into trace_flags that controls the output.
 505  *
 506  * NOTE: These bits must match the trace_options array in
 507  *       trace.c.
 508  */
 509 enum trace_iterator_flags {
 510         TRACE_ITER_PRINT_PARENT         = 0x01,
 511         TRACE_ITER_SYM_OFFSET           = 0x02,
 512         TRACE_ITER_SYM_ADDR             = 0x04,
 513         TRACE_ITER_VERBOSE              = 0x08,
 514         TRACE_ITER_RAW                  = 0x10,
 515         TRACE_ITER_HEX                  = 0x20,
 516         TRACE_ITER_BIN                  = 0x40,
 517         TRACE_ITER_BLOCK                = 0x80,
 518         TRACE_ITER_STACKTRACE           = 0x100,
 519         TRACE_ITER_SCHED_TREE           = 0x200,
 520         TRACE_ITER_PRINTK               = 0x400,
 521         TRACE_ITER_PREEMPTONLY          = 0x800,
 522         TRACE_ITER_BRANCH               = 0x1000,
 523         TRACE_ITER_ANNOTATE             = 0x2000,
 524         TRACE_ITER_USERSTACKTRACE       = 0x4000,
 525         TRACE_ITER_SYM_USEROBJ          = 0x8000
 526 };
 527
 528 /*
 529  * TRACE_ITER_SYM_MASK masks the options in trace_flags that
 530  * control the output of kernel symbols.
 531  */
 532 #define TRACE_ITER_SYM_MASK \
 533         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 534
 535 extern struct tracer nop_trace;
 536
 537 /**
 538  * ftrace_preempt_disable - disable preemption scheduler safe
 539  *
 540  * When tracing can happen inside the scheduler, there exists
 541  * cases that the tracing might happen before the need_resched
 542  * flag is checked. If this happens and the tracer calls
 543  * preempt_enable (after a disable), a schedule might take place
 544  * causing an infinite recursion.
 545  *
 546  * To prevent this, we read the need_recshed flag before
 547  * disabling preemption. When we want to enable preemption we
 548  * check the flag, if it is set, then we call preempt_enable_no_resched.
 549  * Otherwise, we call preempt_enable.
 550  *
 551  * The rational for doing the above is that if need resched is set
 552  * and we have yet to reschedule, we are either in an atomic location
 553  * (where we do not need to check for scheduling) or we are inside
 554  * the scheduler and do not want to resched.
 555  */
 556 static inline int ftrace_preempt_disable(void)
 557 {
 558         int resched;
 559
 560         resched = need_resched();
 561         preempt_disable_notrace();
 562
 563         return resched;
 564 }
 565
 566 /**
 567  * ftrace_preempt_enable - enable preemption scheduler safe
 568  * @resched: the return value from ftrace_preempt_disable
 569  *
 570  * This is a scheduler safe way to enable preemption and not miss
 571  * any preemption checks. The disabled saved the state of preemption.
 572  * If resched is set, then we were either inside an atomic or
 573  * are inside the scheduler (we would have already scheduled
 574  * otherwise). In this case, we do not want to call normal
 575  * preempt_enable, but preempt_enable_no_resched instead.
 576  */
 577 static inline void ftrace_preempt_enable(int resched)
 578 {
 579         if (resched)
 580                 preempt_enable_no_resched_notrace();
 581         else
 582                 preempt_enable_notrace();
 583 }
 584
 585 #ifdef CONFIG_BRANCH_TRACER
 586 extern int enable_branch_tracing(struct trace_array *tr);
 587 extern void disable_branch_tracing(void);
 588 static inline int trace_branch_enable(struct trace_array *tr)
 589 {
 590         if (trace_flags & TRACE_ITER_BRANCH)
 591                 return enable_branch_tracing(tr);
 592         return 0;
 593 }
 594 static inline void trace_branch_disable(void)
 595 {
 596         /* due to races, always disable */
 597         disable_branch_tracing();
 598 }
 599 #else
 600 static inline int trace_branch_enable(struct trace_array *tr)
 601 {
 602         return 0;
 603 }
 604 static inline void trace_branch_disable(void)
 605 {
 606 }
 607 #endif /* CONFIG_BRANCH_TRACER */
 608
 609 #endif /* _LINUX_KERNEL_TRACE_H */