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