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
  12 enum trace_type {
  13         __TRACE_FIRST_TYPE = 0,
  14
  15         TRACE_FN,
  16         TRACE_CTX,
  17         TRACE_WAKE,
  18         TRACE_CONT,
  19         TRACE_STACK,
  20         TRACE_PRINT,
  21         TRACE_SPECIAL,
  22         TRACE_MMIO_RW,
  23         TRACE_MMIO_MAP,
  24         TRACE_BOOT,
  25
  26         __TRACE_LAST_TYPE
  27 };
  28
  29 /*
  30  * The trace entry - the most basic unit of tracing. This is what
  31  * is printed in the end as a single line in the trace output, such as:
  32  *
  33  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
  34  */
  35 struct trace_entry {
  36         unsigned char           type;
  37         unsigned char           cpu;
  38         unsigned char           flags;
  39         unsigned char           preempt_count;
  40         int                     pid;
  41 };
  42
  43 /*
  44  * Function trace entry - function address and parent function addres:
  45  */
  46 struct ftrace_entry {
  47         struct trace_entry      ent;
  48         unsigned long           ip;
  49         unsigned long           parent_ip;
  50 };
  51 extern struct tracer boot_tracer;
  52 extern struct tracer sched_switch_trace; /* Used by the boot tracer */
  53
  54 /*
  55  * Context switch trace entry - which task (and prio) we switched from/to:
  56  */
  57 struct ctx_switch_entry {
  58         struct trace_entry      ent;
  59         unsigned int            prev_pid;
  60         unsigned char           prev_prio;
  61         unsigned char           prev_state;
  62         unsigned int            next_pid;
  63         unsigned char           next_prio;
  64         unsigned char           next_state;
  65         unsigned int            next_cpu;
  66 };
  67
  68 /*
  69  * Special (free-form) trace entry:
  70  */
  71 struct special_entry {
  72         struct trace_entry      ent;
  73         unsigned long           arg1;
  74         unsigned long           arg2;
  75         unsigned long           arg3;
  76 };
  77
  78 /*
  79  * Stack-trace entry:
  80  */
  81
  82 #define FTRACE_STACK_ENTRIES    8
  83
  84 struct stack_entry {
  85         struct trace_entry      ent;
  86         unsigned long           caller[FTRACE_STACK_ENTRIES];
  87 };
  88
  89 /*
  90  * ftrace_printk entry:
  91  */
  92 struct print_entry {
  93         struct trace_entry      ent;
  94         unsigned long           ip;
  95         char                    buf[];
  96 };
  97
  98 #define TRACE_OLD_SIZE          88
  99
 100 struct trace_field_cont {
 101         unsigned char           type;
 102         /* Temporary till we get rid of this completely */
 103         char                    buf[TRACE_OLD_SIZE - 1];
 104 };
 105
 106 struct trace_mmiotrace_rw {
 107         struct trace_entry      ent;
 108         struct mmiotrace_rw     rw;
 109 };
 110
 111 struct trace_mmiotrace_map {
 112         struct trace_entry      ent;
 113         struct mmiotrace_map    map;
 114 };
 115
 116 struct trace_boot {
 117         struct trace_entry      ent;
 118         struct boot_trace       initcall;
 119 };
 120
 121 /*
 122  * trace_flag_type is an enumeration that holds different
 123  * states when a trace occurs. These are:
 124  *  IRQS_OFF            - interrupts were disabled
 125  *  IRQS_NOSUPPORT      - arch does not support irqs_disabled_flags
 126  *  NEED_RESCED         - reschedule is requested
 127  *  HARDIRQ             - inside an interrupt handler
 128  *  SOFTIRQ             - inside a softirq handler
 129  *  CONT                - multiple entries hold the trace item
 130  */
 131 enum trace_flag_type {
 132         TRACE_FLAG_IRQS_OFF             = 0x01,
 133         TRACE_FLAG_IRQS_NOSUPPORT       = 0x02,
 134         TRACE_FLAG_NEED_RESCHED         = 0x04,
 135         TRACE_FLAG_HARDIRQ              = 0x08,
 136         TRACE_FLAG_SOFTIRQ              = 0x10,
 137         TRACE_FLAG_CONT                 = 0x20,
 138 };
 139
 140 #define TRACE_BUF_SIZE          1024
 141
 142 /*
 143  * The CPU trace array - it consists of thousands of trace entries
 144  * plus some other descriptor data: (for example which task started
 145  * the trace, etc.)
 146  */
 147 struct trace_array_cpu {
 148         atomic_t                disabled;
 149
 150         /* these fields get copied into max-trace: */
 151         unsigned long           trace_idx;
 152         unsigned long           overrun;
 153         unsigned long           saved_latency;
 154         unsigned long           critical_start;
 155         unsigned long           critical_end;
 156         unsigned long           critical_sequence;
 157         unsigned long           nice;
 158         unsigned long           policy;
 159         unsigned long           rt_priority;
 160         cycle_t                 preempt_timestamp;
 161         pid_t                   pid;
 162         uid_t                   uid;
 163         char                    comm[TASK_COMM_LEN];
 164 };
 165
 166 struct trace_iterator;
 167
 168 /*
 169  * The trace array - an array of per-CPU trace arrays. This is the
 170  * highest level data structure that individual tracers deal with.
 171  * They have on/off state as well:
 172  */
 173 struct trace_array {
 174         struct ring_buffer      *buffer;
 175         unsigned long           entries;
 176         long                    ctrl;
 177         int                     cpu;
 178         cycle_t                 time_start;
 179         struct task_struct      *waiter;
 180         struct trace_array_cpu  *data[NR_CPUS];
 181 };
 182
 183 #define FTRACE_CMP_TYPE(var, type) \
 184         __builtin_types_compatible_p(typeof(var), type *)
 185
 186 #undef IF_ASSIGN
 187 #define IF_ASSIGN(var, entry, etype, id)                \
 188         if (FTRACE_CMP_TYPE(var, etype)) {              \
 189                 var = (typeof(var))(entry);             \
 190                 WARN_ON(id && (entry)->type != id);     \
 191                 break;                                  \
 192         }
 193
 194 /* Will cause compile errors if type is not found. */
 195 extern void __ftrace_bad_type(void);
 196
 197 /*
 198  * The trace_assign_type is a verifier that the entry type is
 199  * the same as the type being assigned. To add new types simply
 200  * add a line with the following format:
 201  *
 202  * IF_ASSIGN(var, ent, type, id);
 203  *
 204  *  Where "type" is the trace type that includes the trace_entry
 205  *  as the "ent" item. And "id" is the trace identifier that is
 206  *  used in the trace_type enum.
 207  *
 208  *  If the type can have more than one id, then use zero.
 209  */
 210 #define trace_assign_type(var, ent)                                     \
 211         do {                                                            \
 212                 IF_ASSIGN(var, ent, struct ftrace_entry, TRACE_FN);     \
 213                 IF_ASSIGN(var, ent, struct ctx_switch_entry, 0);        \
 214                 IF_ASSIGN(var, ent, struct trace_field_cont, TRACE_CONT); \
 215                 IF_ASSIGN(var, ent, struct stack_entry, TRACE_STACK);   \
 216                 IF_ASSIGN(var, ent, struct print_entry, TRACE_PRINT);   \
 217                 IF_ASSIGN(var, ent, struct special_entry, 0);           \
 218                 IF_ASSIGN(var, ent, struct trace_mmiotrace_rw,          \
 219                           TRACE_MMIO_RW);                               \
 220                 IF_ASSIGN(var, ent, struct trace_mmiotrace_map,         \
 221                           TRACE_MMIO_MAP);                              \
 222                 IF_ASSIGN(var, ent, struct trace_boot, TRACE_BOOT);     \
 223                 __ftrace_bad_type();                                    \
 224         } while (0)
 225
 226 /* Return values for print_line callback */
 227 enum print_line_t {
 228         TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
 229         TRACE_TYPE_HANDLED      = 1,
 230         TRACE_TYPE_UNHANDLED    = 2     /* Relay to other output functions */
 231 };
 232
 233 /*
 234  * A specific tracer, represented by methods that operate on a trace array:
 235  */
 236 struct tracer {
 237         const char              *name;
 238         void                    (*init)(struct trace_array *tr);
 239         void                    (*reset)(struct trace_array *tr);
 240         void                    (*open)(struct trace_iterator *iter);
 241         void                    (*pipe_open)(struct trace_iterator *iter);
 242         void                    (*close)(struct trace_iterator *iter);
 243         void                    (*start)(struct trace_iterator *iter);
 244         void                    (*stop)(struct trace_iterator *iter);
 245         ssize_t                 (*read)(struct trace_iterator *iter,
 246                                         struct file *filp, char __user *ubuf,
 247                                         size_t cnt, loff_t *ppos);
 248         void                    (*ctrl_update)(struct trace_array *tr);
 249 #ifdef CONFIG_FTRACE_STARTUP_TEST
 250         int                     (*selftest)(struct tracer *trace,
 251                                             struct trace_array *tr);
 252 #endif
 253         enum print_line_t       (*print_line)(struct trace_iterator *iter);
 254         struct tracer           *next;
 255         int                     print_max;
 256 };
 257
 258 struct trace_seq {
 259         unsigned char           buffer[PAGE_SIZE];
 260         unsigned int            len;
 261         unsigned int            readpos;
 262 };
 263
 264 /*
 265  * Trace iterator - used by printout routines who present trace
 266  * results to users and which routines might sleep, etc:
 267  */
 268 struct trace_iterator {
 269         struct trace_array      *tr;
 270         struct tracer           *trace;
 271         void                    *private;
 272         struct ring_buffer_iter *buffer_iter[NR_CPUS];
 273
 274         /* The below is zeroed out in pipe_read */
 275         struct trace_seq        seq;
 276         struct trace_entry      *ent;
 277         int                     cpu;
 278         u64                     ts;
 279
 280         unsigned long           iter_flags;
 281         loff_t                  pos;
 282         long                    idx;
 283 };
 284
 285 void trace_wake_up(void);
 286 void tracing_reset(struct trace_array *tr, int cpu);
 287 int tracing_open_generic(struct inode *inode, struct file *filp);
 288 struct dentry *tracing_init_dentry(void);
 289 void init_tracer_sysprof_debugfs(struct dentry *d_tracer);
 290
 291 struct trace_entry *tracing_get_trace_entry(struct trace_array *tr,
 292                                                 struct trace_array_cpu *data);
 293 void tracing_generic_entry_update(struct trace_entry *entry,
 294                                   unsigned long flags,
 295                                   int pc);
 296
 297 void ftrace(struct trace_array *tr,
 298                             struct trace_array_cpu *data,
 299                             unsigned long ip,
 300                             unsigned long parent_ip,
 301                             unsigned long flags, int pc);
 302 void tracing_sched_switch_trace(struct trace_array *tr,
 303                                 struct trace_array_cpu *data,
 304                                 struct task_struct *prev,
 305                                 struct task_struct *next,
 306                                 unsigned long flags, int pc);
 307 void tracing_record_cmdline(struct task_struct *tsk);
 308
 309 void tracing_sched_wakeup_trace(struct trace_array *tr,
 310                                 struct trace_array_cpu *data,
 311                                 struct task_struct *wakee,
 312                                 struct task_struct *cur,
 313                                 unsigned long flags, int pc);
 314 void trace_special(struct trace_array *tr,
 315                    struct trace_array_cpu *data,
 316                    unsigned long arg1,
 317                    unsigned long arg2,
 318                    unsigned long arg3, int pc);
 319 void trace_function(struct trace_array *tr,
 320                     struct trace_array_cpu *data,
 321                     unsigned long ip,
 322                     unsigned long parent_ip,
 323                     unsigned long flags, int pc);
 324
 325 void tracing_start_cmdline_record(void);
 326 void tracing_stop_cmdline_record(void);
 327 int register_tracer(struct tracer *type);
 328 void unregister_tracer(struct tracer *type);
 329
 330 extern unsigned long nsecs_to_usecs(unsigned long nsecs);
 331
 332 extern unsigned long tracing_max_latency;
 333 extern unsigned long tracing_thresh;
 334
 335 void update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu);
 336 void update_max_tr_single(struct trace_array *tr,
 337                           struct task_struct *tsk, int cpu);
 338
 339 extern cycle_t ftrace_now(int cpu);
 340
 341 #ifdef CONFIG_FUNCTION_TRACER
 342 void tracing_start_function_trace(void);
 343 void tracing_stop_function_trace(void);
 344 #else
 345 # define tracing_start_function_trace()         do { } while (0)
 346 # define tracing_stop_function_trace()          do { } while (0)
 347 #endif
 348
 349 #ifdef CONFIG_CONTEXT_SWITCH_TRACER
 350 typedef void
 351 (*tracer_switch_func_t)(void *private,
 352                         void *__rq,
 353                         struct task_struct *prev,
 354                         struct task_struct *next);
 355
 356 struct tracer_switch_ops {
 357         tracer_switch_func_t            func;
 358         void                            *private;
 359         struct tracer_switch_ops        *next;
 360 };
 361
 362 #endif /* CONFIG_CONTEXT_SWITCH_TRACER */
 363
 364 #ifdef CONFIG_DYNAMIC_FTRACE
 365 extern unsigned long ftrace_update_tot_cnt;
 366 #define DYN_FTRACE_TEST_NAME trace_selftest_dynamic_test_func
 367 extern int DYN_FTRACE_TEST_NAME(void);
 368 #endif
 369
 370 #ifdef CONFIG_FTRACE_STARTUP_TEST
 371 extern int trace_selftest_startup_function(struct tracer *trace,
 372                                            struct trace_array *tr);
 373 extern int trace_selftest_startup_irqsoff(struct tracer *trace,
 374                                           struct trace_array *tr);
 375 extern int trace_selftest_startup_preemptoff(struct tracer *trace,
 376                                              struct trace_array *tr);
 377 extern int trace_selftest_startup_preemptirqsoff(struct tracer *trace,
 378                                                  struct trace_array *tr);
 379 extern int trace_selftest_startup_wakeup(struct tracer *trace,
 380                                          struct trace_array *tr);
 381 extern int trace_selftest_startup_nop(struct tracer *trace,
 382                                          struct trace_array *tr);
 383 extern int trace_selftest_startup_sched_switch(struct tracer *trace,
 384                                                struct trace_array *tr);
 385 extern int trace_selftest_startup_sysprof(struct tracer *trace,
 386                                                struct trace_array *tr);
 387 #endif /* CONFIG_FTRACE_STARTUP_TEST */
 388
 389 extern void *head_page(struct trace_array_cpu *data);
 390 extern int trace_seq_printf(struct trace_seq *s, const char *fmt, ...);
 391 extern void trace_seq_print_cont(struct trace_seq *s,
 392                                  struct trace_iterator *iter);
 393 extern ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf,
 394                                  size_t cnt);
 395 extern long ns2usecs(cycle_t nsec);
 396 extern int trace_vprintk(unsigned long ip, const char *fmt, va_list args);
 397
 398 extern unsigned long trace_flags;
 399
 400 /*
 401  * trace_iterator_flags is an enumeration that defines bit
 402  * positions into trace_flags that controls the output.
 403  *
 404  * NOTE: These bits must match the trace_options array in
 405  *       trace.c.
 406  */
 407 enum trace_iterator_flags {
 408         TRACE_ITER_PRINT_PARENT         = 0x01,
 409         TRACE_ITER_SYM_OFFSET           = 0x02,
 410         TRACE_ITER_SYM_ADDR             = 0x04,
 411         TRACE_ITER_VERBOSE              = 0x08,
 412         TRACE_ITER_RAW                  = 0x10,
 413         TRACE_ITER_HEX                  = 0x20,
 414         TRACE_ITER_BIN                  = 0x40,
 415         TRACE_ITER_BLOCK                = 0x80,
 416         TRACE_ITER_STACKTRACE           = 0x100,
 417         TRACE_ITER_SCHED_TREE           = 0x200,
 418         TRACE_ITER_PRINTK               = 0x400,
 419         TRACE_ITER_PREEMPTONLY          = 0x800,
 420 };
 421
 422 extern struct tracer nop_trace;
 423
 424 /**
 425  * ftrace_preempt_disable - disable preemption scheduler safe
 426  *
 427  * When tracing can happen inside the scheduler, there exists
 428  * cases that the tracing might happen before the need_resched
 429  * flag is checked. If this happens and the tracer calls
 430  * preempt_enable (after a disable), a schedule might take place
 431  * causing an infinite recursion.
 432  *
 433  * To prevent this, we read the need_recshed flag before
 434  * disabling preemption. When we want to enable preemption we
 435  * check the flag, if it is set, then we call preempt_enable_no_resched.
 436  * Otherwise, we call preempt_enable.
 437  *
 438  * The rational for doing the above is that if need resched is set
 439  * and we have yet to reschedule, we are either in an atomic location
 440  * (where we do not need to check for scheduling) or we are inside
 441  * the scheduler and do not want to resched.
 442  */
 443 static inline int ftrace_preempt_disable(void)
 444 {
 445         int resched;
 446
 447         resched = need_resched();
 448         preempt_disable_notrace();
 449
 450         return resched;
 451 }
 452
 453 /**
 454  * ftrace_preempt_enable - enable preemption scheduler safe
 455  * @resched: the return value from ftrace_preempt_disable
 456  *
 457  * This is a scheduler safe way to enable preemption and not miss
 458  * any preemption checks. The disabled saved the state of preemption.
 459  * If resched is set, then we were either inside an atomic or
 460  * are inside the scheduler (we would have already scheduled
 461  * otherwise). In this case, we do not want to call normal
 462  * preempt_enable, but preempt_enable_no_resched instead.
 463  */
 464 static inline void ftrace_preempt_enable(int resched)
 465 {
 466         if (resched)
 467                 preempt_enable_no_resched_notrace();
 468         else
 469                 preempt_enable_notrace();
 470 }
 471
 472 #endif /* _LINUX_KERNEL_TRACE_H */