6bd71fa1e1c7deb1ca9521595ba4e7d39c415f68
[safe/jmp/linux-2.6] / 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_STACK,
20         TRACE_PRINT,
21         TRACE_SPECIAL,
22         TRACE_MMIO_RW,
23         TRACE_MMIO_MAP,
24         TRACE_BRANCH,
25         TRACE_BOOT_CALL,
26         TRACE_BOOT_RET,
27         TRACE_GRAPH_RET,
28         TRACE_GRAPH_ENT,
29         TRACE_USER_STACK,
30         TRACE_HW_BRANCHES,
31         TRACE_POWER,
32
33         __TRACE_LAST_TYPE,
34 };
35
36 /*
37  * The trace entry - the most basic unit of tracing. This is what
38  * is printed in the end as a single line in the trace output, such as:
39  *
40  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
41  */
42 struct trace_entry {
43         unsigned char           type;
44         unsigned char           cpu;
45         unsigned char           flags;
46         unsigned char           preempt_count;
47         int                     pid;
48         int                     tgid;
49 };
50
51 /*
52  * Function trace entry - function address and parent function addres:
53  */
54 struct ftrace_entry {
55         struct trace_entry      ent;
56         unsigned long           ip;
57         unsigned long           parent_ip;
58 };
59
60 /* Function call entry */
61 struct ftrace_graph_ent_entry {
62         struct trace_entry                      ent;
63         struct ftrace_graph_ent         graph_ent;
64 };
65
66 /* Function return entry */
67 struct ftrace_graph_ret_entry {
68         struct trace_entry                      ent;
69         struct ftrace_graph_ret         ret;
70 };
71 extern struct tracer boot_tracer;
72
73 /*
74  * Context switch trace entry - which task (and prio) we switched from/to:
75  */
76 struct ctx_switch_entry {
77         struct trace_entry      ent;
78         unsigned int            prev_pid;
79         unsigned char           prev_prio;
80         unsigned char           prev_state;
81         unsigned int            next_pid;
82         unsigned char           next_prio;
83         unsigned char           next_state;
84         unsigned int            next_cpu;
85 };
86
87 /*
88  * Special (free-form) trace entry:
89  */
90 struct special_entry {
91         struct trace_entry      ent;
92         unsigned long           arg1;
93         unsigned long           arg2;
94         unsigned long           arg3;
95 };
96
97 /*
98  * Stack-trace entry:
99  */
100
101 #define FTRACE_STACK_ENTRIES    8
102
103 struct stack_entry {
104         struct trace_entry      ent;
105         unsigned long           caller[FTRACE_STACK_ENTRIES];
106 };
107
108 struct userstack_entry {
109         struct trace_entry      ent;
110         unsigned long           caller[FTRACE_STACK_ENTRIES];
111 };
112
113 /*
114  * ftrace_printk entry:
115  */
116 struct print_entry {
117         struct trace_entry      ent;
118         unsigned long           ip;
119         int                     depth;
120         char                    buf[];
121 };
122
123 #define TRACE_OLD_SIZE          88
124
125 struct trace_field_cont {
126         unsigned char           type;
127         /* Temporary till we get rid of this completely */
128         char                    buf[TRACE_OLD_SIZE - 1];
129 };
130
131 struct trace_mmiotrace_rw {
132         struct trace_entry      ent;
133         struct mmiotrace_rw     rw;
134 };
135
136 struct trace_mmiotrace_map {
137         struct trace_entry      ent;
138         struct mmiotrace_map    map;
139 };
140
141 struct trace_boot_call {
142         struct trace_entry      ent;
143         struct boot_trace_call boot_call;
144 };
145
146 struct trace_boot_ret {
147         struct trace_entry      ent;
148         struct boot_trace_ret boot_ret;
149 };
150
151 #define TRACE_FUNC_SIZE 30
152 #define TRACE_FILE_SIZE 20
153 struct trace_branch {
154         struct trace_entry      ent;
155         unsigned                line;
156         char                    func[TRACE_FUNC_SIZE+1];
157         char                    file[TRACE_FILE_SIZE+1];
158         char                    correct;
159 };
160
161 struct hw_branch_entry {
162         struct trace_entry      ent;
163         u64                     from;
164         u64                     to;
165 };
166
167 struct trace_power {
168         struct trace_entry      ent;
169         struct power_trace      state_data;
170 };
171
172 /*
173  * trace_flag_type is an enumeration that holds different
174  * states when a trace occurs. These are:
175  *  IRQS_OFF            - interrupts were disabled
176  *  IRQS_NOSUPPORT      - arch does not support irqs_disabled_flags
177  *  NEED_RESCED         - reschedule is requested
178  *  HARDIRQ             - inside an interrupt handler
179  *  SOFTIRQ             - inside a softirq handler
180  */
181 enum trace_flag_type {
182         TRACE_FLAG_IRQS_OFF             = 0x01,
183         TRACE_FLAG_IRQS_NOSUPPORT       = 0x02,
184         TRACE_FLAG_NEED_RESCHED         = 0x04,
185         TRACE_FLAG_HARDIRQ              = 0x08,
186         TRACE_FLAG_SOFTIRQ              = 0x10,
187 };
188
189 #define TRACE_BUF_SIZE          1024
190
191 /*
192  * The CPU trace array - it consists of thousands of trace entries
193  * plus some other descriptor data: (for example which task started
194  * the trace, etc.)
195  */
196 struct trace_array_cpu {
197         atomic_t                disabled;
198
199         /* these fields get copied into max-trace: */
200         unsigned long           trace_idx;
201         unsigned long           overrun;
202         unsigned long           saved_latency;
203         unsigned long           critical_start;
204         unsigned long           critical_end;
205         unsigned long           critical_sequence;
206         unsigned long           nice;
207         unsigned long           policy;
208         unsigned long           rt_priority;
209         cycle_t                 preempt_timestamp;
210         pid_t                   pid;
211         uid_t                   uid;
212         char                    comm[TASK_COMM_LEN];
213 };
214
215 struct trace_iterator;
216
217 /*
218  * The trace array - an array of per-CPU trace arrays. This is the
219  * highest level data structure that individual tracers deal with.
220  * They have on/off state as well:
221  */
222 struct trace_array {
223         struct ring_buffer      *buffer;
224         unsigned long           entries;
225         int                     cpu;
226         cycle_t                 time_start;
227         struct task_struct      *waiter;
228         struct trace_array_cpu  *data[NR_CPUS];
229 };
230
231 #define FTRACE_CMP_TYPE(var, type) \
232         __builtin_types_compatible_p(typeof(var), type *)
233
234 #undef IF_ASSIGN
235 #define IF_ASSIGN(var, entry, etype, id)                \
236         if (FTRACE_CMP_TYPE(var, etype)) {              \
237                 var = (typeof(var))(entry);             \
238                 WARN_ON(id && (entry)->type != id);     \
239                 break;                                  \
240         }
241
242 /* Will cause compile errors if type is not found. */
243 extern void __ftrace_bad_type(void);
244
245 /*
246  * The trace_assign_type is a verifier that the entry type is
247  * the same as the type being assigned. To add new types simply
248  * add a line with the following format:
249  *
250  * IF_ASSIGN(var, ent, type, id);
251  *
252  *  Where "type" is the trace type that includes the trace_entry
253  *  as the "ent" item. And "id" is the trace identifier that is
254  *  used in the trace_type enum.
255  *
256  *  If the type can have more than one id, then use zero.
257  */
258 #define trace_assign_type(var, ent)                                     \
259         do {                                                            \
260                 IF_ASSIGN(var, ent, struct ftrace_entry, TRACE_FN);     \
261                 IF_ASSIGN(var, ent, struct ctx_switch_entry, 0);        \
262                 IF_ASSIGN(var, ent, struct stack_entry, TRACE_STACK);   \
263                 IF_ASSIGN(var, ent, struct userstack_entry, TRACE_USER_STACK);\
264                 IF_ASSIGN(var, ent, struct print_entry, TRACE_PRINT);   \
265                 IF_ASSIGN(var, ent, struct special_entry, 0);           \
266                 IF_ASSIGN(var, ent, struct trace_mmiotrace_rw,          \
267                           TRACE_MMIO_RW);                               \
268                 IF_ASSIGN(var, ent, struct trace_mmiotrace_map,         \
269                           TRACE_MMIO_MAP);                              \
270                 IF_ASSIGN(var, ent, struct trace_boot_call, TRACE_BOOT_CALL);\
271                 IF_ASSIGN(var, ent, struct trace_boot_ret, TRACE_BOOT_RET);\
272                 IF_ASSIGN(var, ent, struct trace_branch, TRACE_BRANCH); \
273                 IF_ASSIGN(var, ent, struct ftrace_graph_ent_entry,      \
274                           TRACE_GRAPH_ENT);             \
275                 IF_ASSIGN(var, ent, struct ftrace_graph_ret_entry,      \
276                           TRACE_GRAPH_RET);             \
277                 IF_ASSIGN(var, ent, struct hw_branch_entry, TRACE_HW_BRANCHES);\
278                 IF_ASSIGN(var, ent, struct trace_power, TRACE_POWER); \
279                 __ftrace_bad_type();                                    \
280         } while (0)
281
282 /* Return values for print_line callback */
283 enum print_line_t {
284         TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
285         TRACE_TYPE_HANDLED      = 1,
286         TRACE_TYPE_UNHANDLED    = 2     /* Relay to other output functions */
287 };
288
289
290 /*
291  * An option specific to a tracer. This is a boolean value.
292  * The bit is the bit index that sets its value on the
293  * flags value in struct tracer_flags.
294  */
295 struct tracer_opt {
296         const char      *name; /* Will appear on the trace_options file */
297         u32             bit; /* Mask assigned in val field in tracer_flags */
298 };
299
300 /*
301  * The set of specific options for a tracer. Your tracer
302  * have to set the initial value of the flags val.
303  */
304 struct tracer_flags {
305         u32                     val;
306         struct tracer_opt       *opts;
307 };
308
309 /* Makes more easy to define a tracer opt */
310 #define TRACER_OPT(s, b)        .name = #s, .bit = b
311
312 /*
313  * A specific tracer, represented by methods that operate on a trace array:
314  */
315 struct tracer {
316         const char              *name;
317         /* Your tracer should raise a warning if init fails */
318         int                     (*init)(struct trace_array *tr);
319         void                    (*reset)(struct trace_array *tr);
320         void                    (*start)(struct trace_array *tr);
321         void                    (*stop)(struct trace_array *tr);
322         void                    (*open)(struct trace_iterator *iter);
323         void                    (*pipe_open)(struct trace_iterator *iter);
324         void                    (*close)(struct trace_iterator *iter);
325         ssize_t                 (*read)(struct trace_iterator *iter,
326                                         struct file *filp, char __user *ubuf,
327                                         size_t cnt, loff_t *ppos);
328 #ifdef CONFIG_FTRACE_STARTUP_TEST
329         int                     (*selftest)(struct tracer *trace,
330                                             struct trace_array *tr);
331 #endif
332         void                    (*print_header)(struct seq_file *m);
333         enum print_line_t       (*print_line)(struct trace_iterator *iter);
334         /* If you handled the flag setting, return 0 */
335         int                     (*set_flag)(u32 old_flags, u32 bit, int set);
336         struct tracer           *next;
337         int                     print_max;
338         struct tracer_flags     *flags;
339 };
340
341 struct trace_seq {
342         unsigned char           buffer[PAGE_SIZE];
343         unsigned int            len;
344         unsigned int            readpos;
345 };
346
347 /*
348  * Trace iterator - used by printout routines who present trace
349  * results to users and which routines might sleep, etc:
350  */
351 struct trace_iterator {
352         struct trace_array      *tr;
353         struct tracer           *trace;
354         void                    *private;
355         struct ring_buffer_iter *buffer_iter[NR_CPUS];
356
357         /* The below is zeroed out in pipe_read */
358         struct trace_seq        seq;
359         struct trace_entry      *ent;
360         int                     cpu;
361         u64                     ts;
362
363         unsigned long           iter_flags;
364         loff_t                  pos;
365         long                    idx;
366
367         cpumask_t               started;
368 };
369
370 int tracing_is_enabled(void);
371 void trace_wake_up(void);
372 void tracing_reset(struct trace_array *tr, int cpu);
373 void tracing_reset_online_cpus(struct trace_array *tr);
374 int tracing_open_generic(struct inode *inode, struct file *filp);
375 struct dentry *tracing_init_dentry(void);
376 void init_tracer_sysprof_debugfs(struct dentry *d_tracer);
377
378 struct trace_entry *tracing_get_trace_entry(struct trace_array *tr,
379                                                 struct trace_array_cpu *data);
380 void tracing_generic_entry_update(struct trace_entry *entry,
381                                   unsigned long flags,
382                                   int pc);
383
384 void ftrace(struct trace_array *tr,
385                             struct trace_array_cpu *data,
386                             unsigned long ip,
387                             unsigned long parent_ip,
388                             unsigned long flags, int pc);
389 void tracing_sched_switch_trace(struct trace_array *tr,
390                                 struct trace_array_cpu *data,
391                                 struct task_struct *prev,
392                                 struct task_struct *next,
393                                 unsigned long flags, int pc);
394 void tracing_record_cmdline(struct task_struct *tsk);
395
396 void tracing_sched_wakeup_trace(struct trace_array *tr,
397                                 struct trace_array_cpu *data,
398                                 struct task_struct *wakee,
399                                 struct task_struct *cur,
400                                 unsigned long flags, int pc);
401 void trace_special(struct trace_array *tr,
402                    struct trace_array_cpu *data,
403                    unsigned long arg1,
404                    unsigned long arg2,
405                    unsigned long arg3, int pc);
406 void trace_function(struct trace_array *tr,
407                     struct trace_array_cpu *data,
408                     unsigned long ip,
409                     unsigned long parent_ip,
410                     unsigned long flags, int pc);
411
412 void trace_graph_return(struct ftrace_graph_ret *trace);
413 int trace_graph_entry(struct ftrace_graph_ent *trace);
414 void trace_hw_branch(struct trace_array *tr, u64 from, u64 to);
415
416 void tracing_start_cmdline_record(void);
417 void tracing_stop_cmdline_record(void);
418 void tracing_sched_switch_assign_trace(struct trace_array *tr);
419 void tracing_stop_sched_switch_record(void);
420 void tracing_start_sched_switch_record(void);
421 int register_tracer(struct tracer *type);
422 void unregister_tracer(struct tracer *type);
423
424 extern unsigned long nsecs_to_usecs(unsigned long nsecs);
425
426 extern unsigned long tracing_max_latency;
427 extern unsigned long tracing_thresh;
428
429 void update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu);
430 void update_max_tr_single(struct trace_array *tr,
431                           struct task_struct *tsk, int cpu);
432
433 extern cycle_t ftrace_now(int cpu);
434
435 #ifdef CONFIG_FUNCTION_TRACER
436 void tracing_start_function_trace(void);
437 void tracing_stop_function_trace(void);
438 #else
439 # define tracing_start_function_trace()         do { } while (0)
440 # define tracing_stop_function_trace()          do { } while (0)
441 #endif
442
443 #ifdef CONFIG_CONTEXT_SWITCH_TRACER
444 typedef void
445 (*tracer_switch_func_t)(void *private,
446                         void *__rq,
447                         struct task_struct *prev,
448                         struct task_struct *next);
449
450 struct tracer_switch_ops {
451         tracer_switch_func_t            func;
452         void                            *private;
453         struct tracer_switch_ops        *next;
454 };
455
456 char *trace_find_cmdline(int pid);
457 #endif /* CONFIG_CONTEXT_SWITCH_TRACER */
458
459 #ifdef CONFIG_DYNAMIC_FTRACE
460 extern unsigned long ftrace_update_tot_cnt;
461 #define DYN_FTRACE_TEST_NAME trace_selftest_dynamic_test_func
462 extern int DYN_FTRACE_TEST_NAME(void);
463 #endif
464
465 #ifdef CONFIG_FTRACE_STARTUP_TEST
466 extern int trace_selftest_startup_function(struct tracer *trace,
467                                            struct trace_array *tr);
468 extern int trace_selftest_startup_irqsoff(struct tracer *trace,
469                                           struct trace_array *tr);
470 extern int trace_selftest_startup_preemptoff(struct tracer *trace,
471                                              struct trace_array *tr);
472 extern int trace_selftest_startup_preemptirqsoff(struct tracer *trace,
473                                                  struct trace_array *tr);
474 extern int trace_selftest_startup_wakeup(struct tracer *trace,
475                                          struct trace_array *tr);
476 extern int trace_selftest_startup_nop(struct tracer *trace,
477                                          struct trace_array *tr);
478 extern int trace_selftest_startup_sched_switch(struct tracer *trace,
479                                                struct trace_array *tr);
480 extern int trace_selftest_startup_sysprof(struct tracer *trace,
481                                                struct trace_array *tr);
482 extern int trace_selftest_startup_branch(struct tracer *trace,
483                                          struct trace_array *tr);
484 #endif /* CONFIG_FTRACE_STARTUP_TEST */
485
486 extern void *head_page(struct trace_array_cpu *data);
487 extern long ns2usecs(cycle_t nsec);
488 extern int
489 trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args);
490
491 extern unsigned long trace_flags;
492
493 /* Standard output formatting function used for function return traces */
494 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
495 extern enum print_line_t print_graph_function(struct trace_iterator *iter);
496
497 #ifdef CONFIG_DYNAMIC_FTRACE
498 /* TODO: make this variable */
499 #define FTRACE_GRAPH_MAX_FUNCS          32
500 extern int ftrace_graph_count;
501 extern unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS];
502
503 static inline int ftrace_graph_addr(unsigned long addr)
504 {
505         int i;
506
507         if (!ftrace_graph_count || test_tsk_trace_graph(current))
508                 return 1;
509
510         for (i = 0; i < ftrace_graph_count; i++) {
511                 if (addr == ftrace_graph_funcs[i])
512                         return 1;
513         }
514
515         return 0;
516 }
517 #else
518 static inline int ftrace_trace_addr(unsigned long addr)
519 {
520         return 1;
521 }
522 static inline int ftrace_graph_addr(unsigned long addr)
523 {
524         return 1;
525 }
526 #endif /* CONFIG_DYNAMIC_FTRACE */
527
528 #else /* CONFIG_FUNCTION_GRAPH_TRACER */
529 static inline enum print_line_t
530 print_graph_function(struct trace_iterator *iter)
531 {
532         return TRACE_TYPE_UNHANDLED;
533 }
534 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
535
536 extern struct pid *ftrace_pid_trace;
537
538 static inline int ftrace_trace_task(struct task_struct *task)
539 {
540         if (!ftrace_pid_trace)
541                 return 1;
542
543         return test_tsk_trace_trace(task);
544 }
545
546 /*
547  * trace_iterator_flags is an enumeration that defines bit
548  * positions into trace_flags that controls the output.
549  *
550  * NOTE: These bits must match the trace_options array in
551  *       trace.c.
552  */
553 enum trace_iterator_flags {
554         TRACE_ITER_PRINT_PARENT         = 0x01,
555         TRACE_ITER_SYM_OFFSET           = 0x02,
556         TRACE_ITER_SYM_ADDR             = 0x04,
557         TRACE_ITER_VERBOSE              = 0x08,
558         TRACE_ITER_RAW                  = 0x10,
559         TRACE_ITER_HEX                  = 0x20,
560         TRACE_ITER_BIN                  = 0x40,
561         TRACE_ITER_BLOCK                = 0x80,
562         TRACE_ITER_STACKTRACE           = 0x100,
563         TRACE_ITER_SCHED_TREE           = 0x200,
564         TRACE_ITER_PRINTK               = 0x400,
565         TRACE_ITER_PREEMPTONLY          = 0x800,
566         TRACE_ITER_BRANCH               = 0x1000,
567         TRACE_ITER_ANNOTATE             = 0x2000,
568         TRACE_ITER_USERSTACKTRACE       = 0x4000,
569         TRACE_ITER_SYM_USEROBJ          = 0x8000,
570         TRACE_ITER_PRINTK_MSGONLY       = 0x10000
571 };
572
573 /*
574  * TRACE_ITER_SYM_MASK masks the options in trace_flags that
575  * control the output of kernel symbols.
576  */
577 #define TRACE_ITER_SYM_MASK \
578         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
579
580 extern struct tracer nop_trace;
581
582 /**
583  * ftrace_preempt_disable - disable preemption scheduler safe
584  *
585  * When tracing can happen inside the scheduler, there exists
586  * cases that the tracing might happen before the need_resched
587  * flag is checked. If this happens and the tracer calls
588  * preempt_enable (after a disable), a schedule might take place
589  * causing an infinite recursion.
590  *
591  * To prevent this, we read the need_recshed flag before
592  * disabling preemption. When we want to enable preemption we
593  * check the flag, if it is set, then we call preempt_enable_no_resched.
594  * Otherwise, we call preempt_enable.
595  *
596  * The rational for doing the above is that if need resched is set
597  * and we have yet to reschedule, we are either in an atomic location
598  * (where we do not need to check for scheduling) or we are inside
599  * the scheduler and do not want to resched.
600  */
601 static inline int ftrace_preempt_disable(void)
602 {
603         int resched;
604
605         resched = need_resched();
606         preempt_disable_notrace();
607
608         return resched;
609 }
610
611 /**
612  * ftrace_preempt_enable - enable preemption scheduler safe
613  * @resched: the return value from ftrace_preempt_disable
614  *
615  * This is a scheduler safe way to enable preemption and not miss
616  * any preemption checks. The disabled saved the state of preemption.
617  * If resched is set, then we were either inside an atomic or
618  * are inside the scheduler (we would have already scheduled
619  * otherwise). In this case, we do not want to call normal
620  * preempt_enable, but preempt_enable_no_resched instead.
621  */
622 static inline void ftrace_preempt_enable(int resched)
623 {
624         if (resched)
625                 preempt_enable_no_resched_notrace();
626         else
627                 preempt_enable_notrace();
628 }
629
630 #ifdef CONFIG_BRANCH_TRACER
631 extern int enable_branch_tracing(struct trace_array *tr);
632 extern void disable_branch_tracing(void);
633 static inline int trace_branch_enable(struct trace_array *tr)
634 {
635         if (trace_flags & TRACE_ITER_BRANCH)
636                 return enable_branch_tracing(tr);
637         return 0;
638 }
639 static inline void trace_branch_disable(void)
640 {
641         /* due to races, always disable */
642         disable_branch_tracing();
643 }
644 #else
645 static inline int trace_branch_enable(struct trace_array *tr)
646 {
647         return 0;
648 }
649 static inline void trace_branch_disable(void)
650 {
651 }
652 #endif /* CONFIG_BRANCH_TRACER */
653
654 #endif /* _LINUX_KERNEL_TRACE_H */