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 #include <trace/kmemtrace.h>
  13
  14 enum trace_type {
  15         __TRACE_FIRST_TYPE = 0,
  16
  17         TRACE_FN,
  18         TRACE_CTX,
  19         TRACE_WAKE,
  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_GRAPH_RET,
  29         TRACE_GRAPH_ENT,
  30         TRACE_USER_STACK,
  31         TRACE_HW_BRANCHES,
  32         TRACE_KMEM_ALLOC,
  33         TRACE_KMEM_FREE,
  34         TRACE_POWER,
  35
  36         __TRACE_LAST_TYPE,
  37 };
  38
  39 /*
  40  * The trace entry - the most basic unit of tracing. This is what
  41  * is printed in the end as a single line in the trace output, such as:
  42  *
  43  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
  44  */
  45 struct trace_entry {
  46         unsigned char           type;
  47         unsigned char           cpu;
  48         unsigned char           flags;
  49         unsigned char           preempt_count;
  50         int                     pid;
  51         int                     tgid;
  52 };
  53
  54 /*
  55  * Function trace entry - function address and parent function addres:
  56  */
  57 struct ftrace_entry {
  58         struct trace_entry      ent;
  59         unsigned long           ip;
  60         unsigned long           parent_ip;
  61 };
  62
  63 /* Function call entry */
  64 struct ftrace_graph_ent_entry {
  65         struct trace_entry                      ent;
  66         struct ftrace_graph_ent         graph_ent;
  67 };
  68
  69 /* Function return entry */
  70 struct ftrace_graph_ret_entry {
  71         struct trace_entry                      ent;
  72         struct ftrace_graph_ret         ret;
  73 };
  74 extern struct tracer boot_tracer;
  75
  76 /*
  77  * Context switch trace entry - which task (and prio) we switched from/to:
  78  */
  79 struct ctx_switch_entry {
  80         struct trace_entry      ent;
  81         unsigned int            prev_pid;
  82         unsigned char           prev_prio;
  83         unsigned char           prev_state;
  84         unsigned int            next_pid;
  85         unsigned char           next_prio;
  86         unsigned char           next_state;
  87         unsigned int            next_cpu;
  88 };
  89
  90 /*
  91  * Special (free-form) trace entry:
  92  */
  93 struct special_entry {
  94         struct trace_entry      ent;
  95         unsigned long           arg1;
  96         unsigned long           arg2;
  97         unsigned long           arg3;
  98 };
  99
 100 /*
 101  * Stack-trace entry:
 102  */
 103
 104 #define FTRACE_STACK_ENTRIES    8
 105
 106 struct stack_entry {
 107         struct trace_entry      ent;
 108         unsigned long           caller[FTRACE_STACK_ENTRIES];
 109 };
 110
 111 struct userstack_entry {
 112         struct trace_entry      ent;
 113         unsigned long           caller[FTRACE_STACK_ENTRIES];
 114 };
 115
 116 /*
 117  * ftrace_printk entry:
 118  */
 119 struct print_entry {
 120         struct trace_entry      ent;
 121         unsigned long           ip;
 122         int                     depth;
 123         char                    buf[];
 124 };
 125
 126 #define TRACE_OLD_SIZE          88
 127
 128 struct trace_field_cont {
 129         unsigned char           type;
 130         /* Temporary till we get rid of this completely */
 131         char                    buf[TRACE_OLD_SIZE - 1];
 132 };
 133
 134 struct trace_mmiotrace_rw {
 135         struct trace_entry      ent;
 136         struct mmiotrace_rw     rw;
 137 };
 138
 139 struct trace_mmiotrace_map {
 140         struct trace_entry      ent;
 141         struct mmiotrace_map    map;
 142 };
 143
 144 struct trace_boot_call {
 145         struct trace_entry      ent;
 146         struct boot_trace_call boot_call;
 147 };
 148
 149 struct trace_boot_ret {
 150         struct trace_entry      ent;
 151         struct boot_trace_ret boot_ret;
 152 };
 153
 154 #define TRACE_FUNC_SIZE 30
 155 #define TRACE_FILE_SIZE 20
 156 struct trace_branch {
 157         struct trace_entry      ent;
 158         unsigned                line;
 159         char                    func[TRACE_FUNC_SIZE+1];
 160         char                    file[TRACE_FILE_SIZE+1];
 161         char                    correct;
 162 };
 163
 164 struct hw_branch_entry {
 165         struct trace_entry      ent;
 166         u64                     from;
 167         u64                     to;
 168 };
 169
 170 struct trace_power {
 171         struct trace_entry      ent;
 172         struct power_trace      state_data;
 173 };
 174
 175 struct kmemtrace_alloc_entry {
 176         struct trace_entry      ent;
 177         enum kmemtrace_type_id type_id;
 178         unsigned long call_site;
 179         const void *ptr;
 180         size_t bytes_req;
 181         size_t bytes_alloc;
 182         gfp_t gfp_flags;
 183         int node;
 184 };
 185
 186 struct kmemtrace_free_entry {
 187         struct trace_entry      ent;
 188         enum kmemtrace_type_id type_id;
 189         unsigned long call_site;
 190         const void *ptr;
 191 };
 192
 193 /*
 194  * trace_flag_type is an enumeration that holds different
 195  * states when a trace occurs. These are:
 196  *  IRQS_OFF            - interrupts were disabled
 197  *  IRQS_NOSUPPORT      - arch does not support irqs_disabled_flags
 198  *  NEED_RESCED         - reschedule is requested
 199  *  HARDIRQ             - inside an interrupt handler
 200  *  SOFTIRQ             - inside a softirq handler
 201  */
 202 enum trace_flag_type {
 203         TRACE_FLAG_IRQS_OFF             = 0x01,
 204         TRACE_FLAG_IRQS_NOSUPPORT       = 0x02,
 205         TRACE_FLAG_NEED_RESCHED         = 0x04,
 206         TRACE_FLAG_HARDIRQ              = 0x08,
 207         TRACE_FLAG_SOFTIRQ              = 0x10,
 208 };
 209
 210 #define TRACE_BUF_SIZE          1024
 211
 212 /*
 213  * The CPU trace array - it consists of thousands of trace entries
 214  * plus some other descriptor data: (for example which task started
 215  * the trace, etc.)
 216  */
 217 struct trace_array_cpu {
 218         atomic_t                disabled;
 219
 220         /* these fields get copied into max-trace: */
 221         unsigned long           trace_idx;
 222         unsigned long           overrun;
 223         unsigned long           saved_latency;
 224         unsigned long           critical_start;
 225         unsigned long           critical_end;
 226         unsigned long           critical_sequence;
 227         unsigned long           nice;
 228         unsigned long           policy;
 229         unsigned long           rt_priority;
 230         cycle_t                 preempt_timestamp;
 231         pid_t                   pid;
 232         uid_t                   uid;
 233         char                    comm[TASK_COMM_LEN];
 234 };
 235
 236 struct trace_iterator;
 237
 238 /*
 239  * The trace array - an array of per-CPU trace arrays. This is the
 240  * highest level data structure that individual tracers deal with.
 241  * They have on/off state as well:
 242  */
 243 struct trace_array {
 244         struct ring_buffer      *buffer;
 245         unsigned long           entries;
 246         int                     cpu;
 247         cycle_t                 time_start;
 248         struct task_struct      *waiter;
 249         struct trace_array_cpu  *data[NR_CPUS];
 250 };
 251
 252 #define FTRACE_CMP_TYPE(var, type) \
 253         __builtin_types_compatible_p(typeof(var), type *)
 254
 255 #undef IF_ASSIGN
 256 #define IF_ASSIGN(var, entry, etype, id)                \
 257         if (FTRACE_CMP_TYPE(var, etype)) {              \
 258                 var = (typeof(var))(entry);             \
 259                 WARN_ON(id && (entry)->type != id);     \
 260                 break;                                  \
 261         }
 262
 263 /* Will cause compile errors if type is not found. */
 264 extern void __ftrace_bad_type(void);
 265
 266 /*
 267  * The trace_assign_type is a verifier that the entry type is
 268  * the same as the type being assigned. To add new types simply
 269  * add a line with the following format:
 270  *
 271  * IF_ASSIGN(var, ent, type, id);
 272  *
 273  *  Where "type" is the trace type that includes the trace_entry
 274  *  as the "ent" item. And "id" is the trace identifier that is
 275  *  used in the trace_type enum.
 276  *
 277  *  If the type can have more than one id, then use zero.
 278  */
 279 #define trace_assign_type(var, ent)                                     \
 280         do {                                                            \
 281                 IF_ASSIGN(var, ent, struct ftrace_entry, TRACE_FN);     \
 282                 IF_ASSIGN(var, ent, struct ctx_switch_entry, 0);        \
 283                 IF_ASSIGN(var, ent, struct stack_entry, TRACE_STACK);   \
 284                 IF_ASSIGN(var, ent, struct userstack_entry, TRACE_USER_STACK);\
 285                 IF_ASSIGN(var, ent, struct print_entry, TRACE_PRINT);   \
 286                 IF_ASSIGN(var, ent, struct special_entry, 0);           \
 287                 IF_ASSIGN(var, ent, struct trace_mmiotrace_rw,          \
 288                           TRACE_MMIO_RW);                               \
 289                 IF_ASSIGN(var, ent, struct trace_mmiotrace_map,         \
 290                           TRACE_MMIO_MAP);                              \
 291                 IF_ASSIGN(var, ent, struct trace_boot_call, TRACE_BOOT_CALL);\
 292                 IF_ASSIGN(var, ent, struct trace_boot_ret, TRACE_BOOT_RET);\
 293                 IF_ASSIGN(var, ent, struct trace_branch, TRACE_BRANCH); \
 294                 IF_ASSIGN(var, ent, struct ftrace_graph_ent_entry,      \
 295                           TRACE_GRAPH_ENT);             \
 296                 IF_ASSIGN(var, ent, struct ftrace_graph_ret_entry,      \
 297                           TRACE_GRAPH_RET);             \
 298                 IF_ASSIGN(var, ent, struct hw_branch_entry, TRACE_HW_BRANCHES);\
 299                 IF_ASSIGN(var, ent, struct trace_power, TRACE_POWER); \
 300                 IF_ASSIGN(var, ent, struct kmemtrace_alloc_entry,       \
 301                           TRACE_KMEM_ALLOC);    \
 302                 IF_ASSIGN(var, ent, struct kmemtrace_free_entry,        \
 303                           TRACE_KMEM_FREE);     \
 304                 __ftrace_bad_type();                                    \
 305         } while (0)
 306
 307 /* Return values for print_line callback */
 308 enum print_line_t {
 309         TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
 310         TRACE_TYPE_HANDLED      = 1,
 311         TRACE_TYPE_UNHANDLED    = 2     /* Relay to other output functions */
 312 };
 313
 314
 315 /*
 316  * An option specific to a tracer. This is a boolean value.
 317  * The bit is the bit index that sets its value on the
 318  * flags value in struct tracer_flags.
 319  */
 320 struct tracer_opt {
 321         const char      *name; /* Will appear on the trace_options file */
 322         u32             bit; /* Mask assigned in val field in tracer_flags */
 323 };
 324
 325 /*
 326  * The set of specific options for a tracer. Your tracer
 327  * have to set the initial value of the flags val.
 328  */
 329 struct tracer_flags {
 330         u32                     val;
 331         struct tracer_opt       *opts;
 332 };
 333
 334 /* Makes more easy to define a tracer opt */
 335 #define TRACER_OPT(s, b)        .name = #s, .bit = b
 336
 337 /*
 338  * If you want to provide a stat file (one-shot statistics), fill
 339  * an iterator with stat_start/stat_next and a stat_show callbacks.
 340  * The others callbacks are optional.
 341  */
 342 struct tracer_stat {
 343         /* The name of your stat file */
 344         const char              *name;
 345         /* Iteration over statistic entries */
 346         void                    *(*stat_start)(void);
 347         void                    *(*stat_next)(void *prev, int idx);
 348         /* Compare two entries for sorting (optional) for stats */
 349         int                     (*stat_cmp)(void *p1, void *p2);
 350         /* Print a stat entry */
 351         int                     (*stat_show)(struct seq_file *s, void *p);
 352         /* Print the headers of your stat entries */
 353         int                     (*stat_headers)(struct seq_file *s);
 354 };
 355
 356 /*
 357  * A specific tracer, represented by methods that operate on a trace array:
 358  */
 359 struct tracer {
 360         const char              *name;
 361         /* Your tracer should raise a warning if init fails */
 362         int                     (*init)(struct trace_array *tr);
 363         void                    (*reset)(struct trace_array *tr);
 364         void                    (*start)(struct trace_array *tr);
 365         void                    (*stop)(struct trace_array *tr);
 366         void                    (*open)(struct trace_iterator *iter);
 367         void                    (*pipe_open)(struct trace_iterator *iter);
 368         void                    (*close)(struct trace_iterator *iter);
 369         ssize_t                 (*read)(struct trace_iterator *iter,
 370                                         struct file *filp, char __user *ubuf,
 371                                         size_t cnt, loff_t *ppos);
 372 #ifdef CONFIG_FTRACE_STARTUP_TEST
 373         int                     (*selftest)(struct tracer *trace,
 374                                             struct trace_array *tr);
 375 #endif
 376         void                    (*print_header)(struct seq_file *m);
 377         enum print_line_t       (*print_line)(struct trace_iterator *iter);
 378         /* If you handled the flag setting, return 0 */
 379         int                     (*set_flag)(u32 old_flags, u32 bit, int set);
 380         struct tracer           *next;
 381         int                     print_max;
 382         struct tracer_flags     *flags;
 383         struct tracer_stat      *stats;
 384 };
 385
 386 struct trace_seq {
 387         unsigned char           buffer[PAGE_SIZE];
 388         unsigned int            len;
 389         unsigned int            readpos;
 390 };
 391
 392 /*
 393  * Trace iterator - used by printout routines who present trace
 394  * results to users and which routines might sleep, etc:
 395  */
 396 struct trace_iterator {
 397         struct trace_array      *tr;
 398         struct tracer           *trace;
 399         void                    *private;
 400         struct ring_buffer_iter *buffer_iter[NR_CPUS];
 401
 402         /* The below is zeroed out in pipe_read */
 403         struct trace_seq        seq;
 404         struct trace_entry      *ent;
 405         int                     cpu;
 406         u64                     ts;
 407
 408         unsigned long           iter_flags;
 409         loff_t                  pos;
 410         long                    idx;
 411
 412         cpumask_var_t           started;
 413 };
 414
 415 int tracing_is_enabled(void);
 416 void trace_wake_up(void);
 417 void tracing_reset(struct trace_array *tr, int cpu);
 418 void tracing_reset_online_cpus(struct trace_array *tr);
 419 int tracing_open_generic(struct inode *inode, struct file *filp);
 420 struct dentry *tracing_init_dentry(void);
 421 void init_tracer_sysprof_debugfs(struct dentry *d_tracer);
 422
 423 struct trace_entry *tracing_get_trace_entry(struct trace_array *tr,
 424                                                 struct trace_array_cpu *data);
 425 void tracing_generic_entry_update(struct trace_entry *entry,
 426                                   unsigned long flags,
 427                                   int pc);
 428
 429 void ftrace(struct trace_array *tr,
 430                             struct trace_array_cpu *data,
 431                             unsigned long ip,
 432                             unsigned long parent_ip,
 433                             unsigned long flags, int pc);
 434 void tracing_sched_switch_trace(struct trace_array *tr,
 435                                 struct trace_array_cpu *data,
 436                                 struct task_struct *prev,
 437                                 struct task_struct *next,
 438                                 unsigned long flags, int pc);
 439 void tracing_record_cmdline(struct task_struct *tsk);
 440
 441 void tracing_sched_wakeup_trace(struct trace_array *tr,
 442                                 struct trace_array_cpu *data,
 443                                 struct task_struct *wakee,
 444                                 struct task_struct *cur,
 445                                 unsigned long flags, int pc);
 446 void trace_special(struct trace_array *tr,
 447                    struct trace_array_cpu *data,
 448                    unsigned long arg1,
 449                    unsigned long arg2,
 450                    unsigned long arg3, int pc);
 451 void trace_function(struct trace_array *tr,
 452                     struct trace_array_cpu *data,
 453                     unsigned long ip,
 454                     unsigned long parent_ip,
 455                     unsigned long flags, int pc);
 456
 457 void trace_graph_return(struct ftrace_graph_ret *trace);
 458 int trace_graph_entry(struct ftrace_graph_ent *trace);
 459 void trace_hw_branch(struct trace_array *tr, u64 from, u64 to);
 460
 461 void tracing_start_cmdline_record(void);
 462 void tracing_stop_cmdline_record(void);
 463 void tracing_sched_switch_assign_trace(struct trace_array *tr);
 464 void tracing_stop_sched_switch_record(void);
 465 void tracing_start_sched_switch_record(void);
 466 int register_tracer(struct tracer *type);
 467 void unregister_tracer(struct tracer *type);
 468
 469 void init_tracer_stat(struct tracer *trace);
 470
 471 extern unsigned long nsecs_to_usecs(unsigned long nsecs);
 472
 473 extern unsigned long tracing_max_latency;
 474 extern unsigned long tracing_thresh;
 475
 476 void update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu);
 477 void update_max_tr_single(struct trace_array *tr,
 478                           struct task_struct *tsk, int cpu);
 479
 480 extern cycle_t ftrace_now(int cpu);
 481
 482 #ifdef CONFIG_FUNCTION_TRACER
 483 void tracing_start_function_trace(void);
 484 void tracing_stop_function_trace(void);
 485 #else
 486 # define tracing_start_function_trace()         do { } while (0)
 487 # define tracing_stop_function_trace()          do { } while (0)
 488 #endif
 489
 490 #ifdef CONFIG_CONTEXT_SWITCH_TRACER
 491 typedef void
 492 (*tracer_switch_func_t)(void *private,
 493                         void *__rq,
 494                         struct task_struct *prev,
 495                         struct task_struct *next);
 496
 497 struct tracer_switch_ops {
 498         tracer_switch_func_t            func;
 499         void                            *private;
 500         struct tracer_switch_ops        *next;
 501 };
 502 #endif /* CONFIG_CONTEXT_SWITCH_TRACER */
 503
 504 extern char *trace_find_cmdline(int pid);
 505
 506 #ifdef CONFIG_DYNAMIC_FTRACE
 507 extern unsigned long ftrace_update_tot_cnt;
 508 #define DYN_FTRACE_TEST_NAME trace_selftest_dynamic_test_func
 509 extern int DYN_FTRACE_TEST_NAME(void);
 510 #endif
 511
 512 #ifdef CONFIG_FTRACE_STARTUP_TEST
 513 extern int trace_selftest_startup_function(struct tracer *trace,
 514                                            struct trace_array *tr);
 515 extern int trace_selftest_startup_irqsoff(struct tracer *trace,
 516                                           struct trace_array *tr);
 517 extern int trace_selftest_startup_preemptoff(struct tracer *trace,
 518                                              struct trace_array *tr);
 519 extern int trace_selftest_startup_preemptirqsoff(struct tracer *trace,
 520                                                  struct trace_array *tr);
 521 extern int trace_selftest_startup_wakeup(struct tracer *trace,
 522                                          struct trace_array *tr);
 523 extern int trace_selftest_startup_nop(struct tracer *trace,
 524                                          struct trace_array *tr);
 525 extern int trace_selftest_startup_sched_switch(struct tracer *trace,
 526                                                struct trace_array *tr);
 527 extern int trace_selftest_startup_sysprof(struct tracer *trace,
 528                                                struct trace_array *tr);
 529 extern int trace_selftest_startup_branch(struct tracer *trace,
 530                                          struct trace_array *tr);
 531 #endif /* CONFIG_FTRACE_STARTUP_TEST */
 532
 533 extern void *head_page(struct trace_array_cpu *data);
 534 extern long ns2usecs(cycle_t nsec);
 535 extern int
 536 trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args);
 537
 538 extern unsigned long trace_flags;
 539
 540 /* Standard output formatting function used for function return traces */
 541 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 542 extern enum print_line_t print_graph_function(struct trace_iterator *iter);
 543
 544 #ifdef CONFIG_DYNAMIC_FTRACE
 545 /* TODO: make this variable */
 546 #define FTRACE_GRAPH_MAX_FUNCS          32
 547 extern int ftrace_graph_count;
 548 extern unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS];
 549
 550 static inline int ftrace_graph_addr(unsigned long addr)
 551 {
 552         int i;
 553
 554         if (!ftrace_graph_count || test_tsk_trace_graph(current))
 555                 return 1;
 556
 557         for (i = 0; i < ftrace_graph_count; i++) {
 558                 if (addr == ftrace_graph_funcs[i])
 559                         return 1;
 560         }
 561
 562         return 0;
 563 }
 564 #else
 565 static inline int ftrace_trace_addr(unsigned long addr)
 566 {
 567         return 1;
 568 }
 569 static inline int ftrace_graph_addr(unsigned long addr)
 570 {
 571         return 1;
 572 }
 573 #endif /* CONFIG_DYNAMIC_FTRACE */
 574
 575 #else /* CONFIG_FUNCTION_GRAPH_TRACER */
 576 static inline enum print_line_t
 577 print_graph_function(struct trace_iterator *iter)
 578 {
 579         return TRACE_TYPE_UNHANDLED;
 580 }
 581 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
 582
 583 extern struct pid *ftrace_pid_trace;
 584
 585 static inline int ftrace_trace_task(struct task_struct *task)
 586 {
 587         if (!ftrace_pid_trace)
 588                 return 1;
 589
 590         return test_tsk_trace_trace(task);
 591 }
 592
 593 /*
 594  * trace_iterator_flags is an enumeration that defines bit
 595  * positions into trace_flags that controls the output.
 596  *
 597  * NOTE: These bits must match the trace_options array in
 598  *       trace.c.
 599  */
 600 enum trace_iterator_flags {
 601         TRACE_ITER_PRINT_PARENT         = 0x01,
 602         TRACE_ITER_SYM_OFFSET           = 0x02,
 603         TRACE_ITER_SYM_ADDR             = 0x04,
 604         TRACE_ITER_VERBOSE              = 0x08,
 605         TRACE_ITER_RAW                  = 0x10,
 606         TRACE_ITER_HEX                  = 0x20,
 607         TRACE_ITER_BIN                  = 0x40,
 608         TRACE_ITER_BLOCK                = 0x80,
 609         TRACE_ITER_STACKTRACE           = 0x100,
 610         TRACE_ITER_SCHED_TREE           = 0x200,
 611         TRACE_ITER_PRINTK               = 0x400,
 612         TRACE_ITER_PREEMPTONLY          = 0x800,
 613         TRACE_ITER_BRANCH               = 0x1000,
 614         TRACE_ITER_ANNOTATE             = 0x2000,
 615         TRACE_ITER_USERSTACKTRACE       = 0x4000,
 616         TRACE_ITER_SYM_USEROBJ          = 0x8000,
 617         TRACE_ITER_PRINTK_MSGONLY       = 0x10000
 618 };
 619
 620 /*
 621  * TRACE_ITER_SYM_MASK masks the options in trace_flags that
 622  * control the output of kernel symbols.
 623  */
 624 #define TRACE_ITER_SYM_MASK \
 625         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
 626
 627 extern struct tracer nop_trace;
 628
 629 /**
 630  * ftrace_preempt_disable - disable preemption scheduler safe
 631  *
 632  * When tracing can happen inside the scheduler, there exists
 633  * cases that the tracing might happen before the need_resched
 634  * flag is checked. If this happens and the tracer calls
 635  * preempt_enable (after a disable), a schedule might take place
 636  * causing an infinite recursion.
 637  *
 638  * To prevent this, we read the need_recshed flag before
 639  * disabling preemption. When we want to enable preemption we
 640  * check the flag, if it is set, then we call preempt_enable_no_resched.
 641  * Otherwise, we call preempt_enable.
 642  *
 643  * The rational for doing the above is that if need resched is set
 644  * and we have yet to reschedule, we are either in an atomic location
 645  * (where we do not need to check for scheduling) or we are inside
 646  * the scheduler and do not want to resched.
 647  */
 648 static inline int ftrace_preempt_disable(void)
 649 {
 650         int resched;
 651
 652         resched = need_resched();
 653         preempt_disable_notrace();
 654
 655         return resched;
 656 }
 657
 658 /**
 659  * ftrace_preempt_enable - enable preemption scheduler safe
 660  * @resched: the return value from ftrace_preempt_disable
 661  *
 662  * This is a scheduler safe way to enable preemption and not miss
 663  * any preemption checks. The disabled saved the state of preemption.
 664  * If resched is set, then we were either inside an atomic or
 665  * are inside the scheduler (we would have already scheduled
 666  * otherwise). In this case, we do not want to call normal
 667  * preempt_enable, but preempt_enable_no_resched instead.
 668  */
 669 static inline void ftrace_preempt_enable(int resched)
 670 {
 671         if (resched)
 672                 preempt_enable_no_resched_notrace();
 673         else
 674                 preempt_enable_notrace();
 675 }
 676
 677 #ifdef CONFIG_BRANCH_TRACER
 678 extern int enable_branch_tracing(struct trace_array *tr);
 679 extern void disable_branch_tracing(void);
 680 static inline int trace_branch_enable(struct trace_array *tr)
 681 {
 682         if (trace_flags & TRACE_ITER_BRANCH)
 683                 return enable_branch_tracing(tr);
 684         return 0;
 685 }
 686 static inline void trace_branch_disable(void)
 687 {
 688         /* due to races, always disable */
 689         disable_branch_tracing();
 690 }
 691 #else
 692 static inline int trace_branch_enable(struct trace_array *tr)
 693 {
 694         return 0;
 695 }
 696 static inline void trace_branch_disable(void)
 697 {
 698 }
 699 #endif /* CONFIG_BRANCH_TRACER */
 700
 701 #endif /* _LINUX_KERNEL_TRACE_H */