e603a291134bdbff476bac1bb2c3713662a1b99b
[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 #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         TRACE_BLK,
36
37         __TRACE_LAST_TYPE,
38 };
39
40 /*
41  * The trace entry - the most basic unit of tracing. This is what
42  * is printed in the end as a single line in the trace output, such as:
43  *
44  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
45  */
46 struct trace_entry {
47         unsigned char           type;
48         unsigned char           cpu;
49         unsigned char           flags;
50         unsigned char           preempt_count;
51         int                     pid;
52         int                     tgid;
53 };
54
55 /*
56  * Function trace entry - function address and parent function addres:
57  */
58 struct ftrace_entry {
59         struct trace_entry      ent;
60         unsigned long           ip;
61         unsigned long           parent_ip;
62 };
63
64 /* Function call entry */
65 struct ftrace_graph_ent_entry {
66         struct trace_entry                      ent;
67         struct ftrace_graph_ent         graph_ent;
68 };
69
70 /* Function return entry */
71 struct ftrace_graph_ret_entry {
72         struct trace_entry                      ent;
73         struct ftrace_graph_ret         ret;
74 };
75 extern struct tracer boot_tracer;
76
77 /*
78  * Context switch trace entry - which task (and prio) we switched from/to:
79  */
80 struct ctx_switch_entry {
81         struct trace_entry      ent;
82         unsigned int            prev_pid;
83         unsigned char           prev_prio;
84         unsigned char           prev_state;
85         unsigned int            next_pid;
86         unsigned char           next_prio;
87         unsigned char           next_state;
88         unsigned int            next_cpu;
89 };
90
91 /*
92  * Special (free-form) trace entry:
93  */
94 struct special_entry {
95         struct trace_entry      ent;
96         unsigned long           arg1;
97         unsigned long           arg2;
98         unsigned long           arg3;
99 };
100
101 /*
102  * Stack-trace entry:
103  */
104
105 #define FTRACE_STACK_ENTRIES    8
106
107 struct stack_entry {
108         struct trace_entry      ent;
109         unsigned long           caller[FTRACE_STACK_ENTRIES];
110 };
111
112 struct userstack_entry {
113         struct trace_entry      ent;
114         unsigned long           caller[FTRACE_STACK_ENTRIES];
115 };
116
117 /*
118  * ftrace_printk entry:
119  */
120 struct print_entry {
121         struct trace_entry      ent;
122         unsigned long           ip;
123         int                     depth;
124         char                    buf[];
125 };
126
127 #define TRACE_OLD_SIZE          88
128
129 struct trace_field_cont {
130         unsigned char           type;
131         /* Temporary till we get rid of this completely */
132         char                    buf[TRACE_OLD_SIZE - 1];
133 };
134
135 struct trace_mmiotrace_rw {
136         struct trace_entry      ent;
137         struct mmiotrace_rw     rw;
138 };
139
140 struct trace_mmiotrace_map {
141         struct trace_entry      ent;
142         struct mmiotrace_map    map;
143 };
144
145 struct trace_boot_call {
146         struct trace_entry      ent;
147         struct boot_trace_call boot_call;
148 };
149
150 struct trace_boot_ret {
151         struct trace_entry      ent;
152         struct boot_trace_ret boot_ret;
153 };
154
155 #define TRACE_FUNC_SIZE 30
156 #define TRACE_FILE_SIZE 20
157 struct trace_branch {
158         struct trace_entry      ent;
159         unsigned                line;
160         char                    func[TRACE_FUNC_SIZE+1];
161         char                    file[TRACE_FILE_SIZE+1];
162         char                    correct;
163 };
164
165 struct hw_branch_entry {
166         struct trace_entry      ent;
167         u64                     from;
168         u64                     to;
169 };
170
171 struct trace_power {
172         struct trace_entry      ent;
173         struct power_trace      state_data;
174 };
175
176 struct kmemtrace_alloc_entry {
177         struct trace_entry      ent;
178         enum kmemtrace_type_id type_id;
179         unsigned long call_site;
180         const void *ptr;
181         size_t bytes_req;
182         size_t bytes_alloc;
183         gfp_t gfp_flags;
184         int node;
185 };
186
187 struct kmemtrace_free_entry {
188         struct trace_entry      ent;
189         enum kmemtrace_type_id type_id;
190         unsigned long call_site;
191         const void *ptr;
192 };
193
194 /*
195  * trace_flag_type is an enumeration that holds different
196  * states when a trace occurs. These are:
197  *  IRQS_OFF            - interrupts were disabled
198  *  IRQS_NOSUPPORT      - arch does not support irqs_disabled_flags
199  *  NEED_RESCED         - reschedule is requested
200  *  HARDIRQ             - inside an interrupt handler
201  *  SOFTIRQ             - inside a softirq handler
202  */
203 enum trace_flag_type {
204         TRACE_FLAG_IRQS_OFF             = 0x01,
205         TRACE_FLAG_IRQS_NOSUPPORT       = 0x02,
206         TRACE_FLAG_NEED_RESCHED         = 0x04,
207         TRACE_FLAG_HARDIRQ              = 0x08,
208         TRACE_FLAG_SOFTIRQ              = 0x10,
209 };
210
211 #define TRACE_BUF_SIZE          1024
212
213 /*
214  * The CPU trace array - it consists of thousands of trace entries
215  * plus some other descriptor data: (for example which task started
216  * the trace, etc.)
217  */
218 struct trace_array_cpu {
219         atomic_t                disabled;
220
221         /* these fields get copied into max-trace: */
222         unsigned long           trace_idx;
223         unsigned long           overrun;
224         unsigned long           saved_latency;
225         unsigned long           critical_start;
226         unsigned long           critical_end;
227         unsigned long           critical_sequence;
228         unsigned long           nice;
229         unsigned long           policy;
230         unsigned long           rt_priority;
231         cycle_t                 preempt_timestamp;
232         pid_t                   pid;
233         uid_t                   uid;
234         char                    comm[TASK_COMM_LEN];
235 };
236
237 struct trace_iterator;
238
239 /*
240  * The trace array - an array of per-CPU trace arrays. This is the
241  * highest level data structure that individual tracers deal with.
242  * They have on/off state as well:
243  */
244 struct trace_array {
245         struct ring_buffer      *buffer;
246         unsigned long           entries;
247         int                     cpu;
248         cycle_t                 time_start;
249         struct task_struct      *waiter;
250         struct trace_array_cpu  *data[NR_CPUS];
251 };
252
253 #define FTRACE_CMP_TYPE(var, type) \
254         __builtin_types_compatible_p(typeof(var), type *)
255
256 #undef IF_ASSIGN
257 #define IF_ASSIGN(var, entry, etype, id)                \
258         if (FTRACE_CMP_TYPE(var, etype)) {              \
259                 var = (typeof(var))(entry);             \
260                 WARN_ON(id && (entry)->type != id);     \
261                 break;                                  \
262         }
263
264 /* Will cause compile errors if type is not found. */
265 extern void __ftrace_bad_type(void);
266
267 /*
268  * The trace_assign_type is a verifier that the entry type is
269  * the same as the type being assigned. To add new types simply
270  * add a line with the following format:
271  *
272  * IF_ASSIGN(var, ent, type, id);
273  *
274  *  Where "type" is the trace type that includes the trace_entry
275  *  as the "ent" item. And "id" is the trace identifier that is
276  *  used in the trace_type enum.
277  *
278  *  If the type can have more than one id, then use zero.
279  */
280 #define trace_assign_type(var, ent)                                     \
281         do {                                                            \
282                 IF_ASSIGN(var, ent, struct ftrace_entry, TRACE_FN);     \
283                 IF_ASSIGN(var, ent, struct ctx_switch_entry, 0);        \
284                 IF_ASSIGN(var, ent, struct stack_entry, TRACE_STACK);   \
285                 IF_ASSIGN(var, ent, struct userstack_entry, TRACE_USER_STACK);\
286                 IF_ASSIGN(var, ent, struct print_entry, TRACE_PRINT);   \
287                 IF_ASSIGN(var, ent, struct special_entry, 0);           \
288                 IF_ASSIGN(var, ent, struct trace_mmiotrace_rw,          \
289                           TRACE_MMIO_RW);                               \
290                 IF_ASSIGN(var, ent, struct trace_mmiotrace_map,         \
291                           TRACE_MMIO_MAP);                              \
292                 IF_ASSIGN(var, ent, struct trace_boot_call, TRACE_BOOT_CALL);\
293                 IF_ASSIGN(var, ent, struct trace_boot_ret, TRACE_BOOT_RET);\
294                 IF_ASSIGN(var, ent, struct trace_branch, TRACE_BRANCH); \
295                 IF_ASSIGN(var, ent, struct ftrace_graph_ent_entry,      \
296                           TRACE_GRAPH_ENT);             \
297                 IF_ASSIGN(var, ent, struct ftrace_graph_ret_entry,      \
298                           TRACE_GRAPH_RET);             \
299                 IF_ASSIGN(var, ent, struct hw_branch_entry, TRACE_HW_BRANCHES);\
300                 IF_ASSIGN(var, ent, struct trace_power, TRACE_POWER); \
301                 IF_ASSIGN(var, ent, struct kmemtrace_alloc_entry,       \
302                           TRACE_KMEM_ALLOC);    \
303                 IF_ASSIGN(var, ent, struct kmemtrace_free_entry,        \
304                           TRACE_KMEM_FREE);     \
305                 __ftrace_bad_type();                                    \
306         } while (0)
307
308 /* Return values for print_line callback */
309 enum print_line_t {
310         TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
311         TRACE_TYPE_HANDLED      = 1,
312         TRACE_TYPE_UNHANDLED    = 2     /* Relay to other output functions */
313 };
314
315
316 /*
317  * An option specific to a tracer. This is a boolean value.
318  * The bit is the bit index that sets its value on the
319  * flags value in struct tracer_flags.
320  */
321 struct tracer_opt {
322         const char      *name; /* Will appear on the trace_options file */
323         u32             bit; /* Mask assigned in val field in tracer_flags */
324 };
325
326 /*
327  * The set of specific options for a tracer. Your tracer
328  * have to set the initial value of the flags val.
329  */
330 struct tracer_flags {
331         u32                     val;
332         struct tracer_opt       *opts;
333 };
334
335 /* Makes more easy to define a tracer opt */
336 #define TRACER_OPT(s, b)        .name = #s, .bit = b
337
338
339 /*
340  * A specific tracer, represented by methods that operate on a trace array:
341  */
342 struct tracer {
343         const char              *name;
344         /* Your tracer should raise a warning if init fails */
345         int                     (*init)(struct trace_array *tr);
346         void                    (*reset)(struct trace_array *tr);
347         void                    (*start)(struct trace_array *tr);
348         void                    (*stop)(struct trace_array *tr);
349         void                    (*open)(struct trace_iterator *iter);
350         void                    (*pipe_open)(struct trace_iterator *iter);
351         void                    (*close)(struct trace_iterator *iter);
352         ssize_t                 (*read)(struct trace_iterator *iter,
353                                         struct file *filp, char __user *ubuf,
354                                         size_t cnt, loff_t *ppos);
355 #ifdef CONFIG_FTRACE_STARTUP_TEST
356         int                     (*selftest)(struct tracer *trace,
357                                             struct trace_array *tr);
358 #endif
359         void                    (*print_header)(struct seq_file *m);
360         enum print_line_t       (*print_line)(struct trace_iterator *iter);
361         /* If you handled the flag setting, return 0 */
362         int                     (*set_flag)(u32 old_flags, u32 bit, int set);
363         struct tracer           *next;
364         int                     print_max;
365         struct tracer_flags     *flags;
366         struct tracer_stat      *stats;
367 };
368
369 struct trace_seq {
370         unsigned char           buffer[PAGE_SIZE];
371         unsigned int            len;
372         unsigned int            readpos;
373 };
374
375 /*
376  * Trace iterator - used by printout routines who present trace
377  * results to users and which routines might sleep, etc:
378  */
379 struct trace_iterator {
380         struct trace_array      *tr;
381         struct tracer           *trace;
382         void                    *private;
383         struct ring_buffer_iter *buffer_iter[NR_CPUS];
384
385         /* The below is zeroed out in pipe_read */
386         struct trace_seq        seq;
387         struct trace_entry      *ent;
388         int                     cpu;
389         u64                     ts;
390
391         unsigned long           iter_flags;
392         loff_t                  pos;
393         long                    idx;
394
395         cpumask_var_t           started;
396 };
397
398 int tracing_is_enabled(void);
399 void trace_wake_up(void);
400 void tracing_reset(struct trace_array *tr, int cpu);
401 void tracing_reset_online_cpus(struct trace_array *tr);
402 int tracing_open_generic(struct inode *inode, struct file *filp);
403 struct dentry *tracing_init_dentry(void);
404 void init_tracer_sysprof_debugfs(struct dentry *d_tracer);
405
406 struct trace_entry *tracing_get_trace_entry(struct trace_array *tr,
407                                                 struct trace_array_cpu *data);
408 void tracing_generic_entry_update(struct trace_entry *entry,
409                                   unsigned long flags,
410                                   int pc);
411
412 void ftrace(struct trace_array *tr,
413                             struct trace_array_cpu *data,
414                             unsigned long ip,
415                             unsigned long parent_ip,
416                             unsigned long flags, int pc);
417 void tracing_sched_switch_trace(struct trace_array *tr,
418                                 struct trace_array_cpu *data,
419                                 struct task_struct *prev,
420                                 struct task_struct *next,
421                                 unsigned long flags, int pc);
422 void tracing_record_cmdline(struct task_struct *tsk);
423
424 void tracing_sched_wakeup_trace(struct trace_array *tr,
425                                 struct trace_array_cpu *data,
426                                 struct task_struct *wakee,
427                                 struct task_struct *cur,
428                                 unsigned long flags, int pc);
429 void trace_special(struct trace_array *tr,
430                    struct trace_array_cpu *data,
431                    unsigned long arg1,
432                    unsigned long arg2,
433                    unsigned long arg3, int pc);
434 void trace_function(struct trace_array *tr,
435                     struct trace_array_cpu *data,
436                     unsigned long ip,
437                     unsigned long parent_ip,
438                     unsigned long flags, int pc);
439
440 void trace_graph_return(struct ftrace_graph_ret *trace);
441 int trace_graph_entry(struct ftrace_graph_ent *trace);
442
443 void tracing_start_cmdline_record(void);
444 void tracing_stop_cmdline_record(void);
445 void tracing_sched_switch_assign_trace(struct trace_array *tr);
446 void tracing_stop_sched_switch_record(void);
447 void tracing_start_sched_switch_record(void);
448 int register_tracer(struct tracer *type);
449 void unregister_tracer(struct tracer *type);
450
451 extern unsigned long nsecs_to_usecs(unsigned long nsecs);
452
453 extern unsigned long tracing_max_latency;
454 extern unsigned long tracing_thresh;
455
456 void update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu);
457 void update_max_tr_single(struct trace_array *tr,
458                           struct task_struct *tsk, int cpu);
459
460 void __trace_stack(struct trace_array *tr,
461                    struct trace_array_cpu *data,
462                    unsigned long flags,
463                    int skip, int pc);
464
465 extern cycle_t ftrace_now(int cpu);
466
467 #ifdef CONFIG_CONTEXT_SWITCH_TRACER
468 typedef void
469 (*tracer_switch_func_t)(void *private,
470                         void *__rq,
471                         struct task_struct *prev,
472                         struct task_struct *next);
473
474 struct tracer_switch_ops {
475         tracer_switch_func_t            func;
476         void                            *private;
477         struct tracer_switch_ops        *next;
478 };
479 #endif /* CONFIG_CONTEXT_SWITCH_TRACER */
480
481 extern char *trace_find_cmdline(int pid);
482
483 #ifdef CONFIG_DYNAMIC_FTRACE
484 extern unsigned long ftrace_update_tot_cnt;
485 #define DYN_FTRACE_TEST_NAME trace_selftest_dynamic_test_func
486 extern int DYN_FTRACE_TEST_NAME(void);
487 #endif
488
489 #ifdef CONFIG_FTRACE_STARTUP_TEST
490 extern int trace_selftest_startup_function(struct tracer *trace,
491                                            struct trace_array *tr);
492 extern int trace_selftest_startup_irqsoff(struct tracer *trace,
493                                           struct trace_array *tr);
494 extern int trace_selftest_startup_preemptoff(struct tracer *trace,
495                                              struct trace_array *tr);
496 extern int trace_selftest_startup_preemptirqsoff(struct tracer *trace,
497                                                  struct trace_array *tr);
498 extern int trace_selftest_startup_wakeup(struct tracer *trace,
499                                          struct trace_array *tr);
500 extern int trace_selftest_startup_nop(struct tracer *trace,
501                                          struct trace_array *tr);
502 extern int trace_selftest_startup_sched_switch(struct tracer *trace,
503                                                struct trace_array *tr);
504 extern int trace_selftest_startup_sysprof(struct tracer *trace,
505                                                struct trace_array *tr);
506 extern int trace_selftest_startup_branch(struct tracer *trace,
507                                          struct trace_array *tr);
508 #endif /* CONFIG_FTRACE_STARTUP_TEST */
509
510 extern void *head_page(struct trace_array_cpu *data);
511 extern long ns2usecs(cycle_t nsec);
512 extern int
513 trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args);
514
515 extern unsigned long trace_flags;
516
517 /* Standard output formatting function used for function return traces */
518 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
519 extern enum print_line_t print_graph_function(struct trace_iterator *iter);
520
521 #ifdef CONFIG_DYNAMIC_FTRACE
522 /* TODO: make this variable */
523 #define FTRACE_GRAPH_MAX_FUNCS          32
524 extern int ftrace_graph_count;
525 extern unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS];
526
527 static inline int ftrace_graph_addr(unsigned long addr)
528 {
529         int i;
530
531         if (!ftrace_graph_count || test_tsk_trace_graph(current))
532                 return 1;
533
534         for (i = 0; i < ftrace_graph_count; i++) {
535                 if (addr == ftrace_graph_funcs[i])
536                         return 1;
537         }
538
539         return 0;
540 }
541 #else
542 static inline int ftrace_trace_addr(unsigned long addr)
543 {
544         return 1;
545 }
546 static inline int ftrace_graph_addr(unsigned long addr)
547 {
548         return 1;
549 }
550 #endif /* CONFIG_DYNAMIC_FTRACE */
551
552 #else /* CONFIG_FUNCTION_GRAPH_TRACER */
553 static inline enum print_line_t
554 print_graph_function(struct trace_iterator *iter)
555 {
556         return TRACE_TYPE_UNHANDLED;
557 }
558 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
559
560 extern struct pid *ftrace_pid_trace;
561
562 static inline int ftrace_trace_task(struct task_struct *task)
563 {
564         if (!ftrace_pid_trace)
565                 return 1;
566
567         return test_tsk_trace_trace(task);
568 }
569
570 /*
571  * trace_iterator_flags is an enumeration that defines bit
572  * positions into trace_flags that controls the output.
573  *
574  * NOTE: These bits must match the trace_options array in
575  *       trace.c.
576  */
577 enum trace_iterator_flags {
578         TRACE_ITER_PRINT_PARENT         = 0x01,
579         TRACE_ITER_SYM_OFFSET           = 0x02,
580         TRACE_ITER_SYM_ADDR             = 0x04,
581         TRACE_ITER_VERBOSE              = 0x08,
582         TRACE_ITER_RAW                  = 0x10,
583         TRACE_ITER_HEX                  = 0x20,
584         TRACE_ITER_BIN                  = 0x40,
585         TRACE_ITER_BLOCK                = 0x80,
586         TRACE_ITER_STACKTRACE           = 0x100,
587         TRACE_ITER_SCHED_TREE           = 0x200,
588         TRACE_ITER_PRINTK               = 0x400,
589         TRACE_ITER_PREEMPTONLY          = 0x800,
590         TRACE_ITER_BRANCH               = 0x1000,
591         TRACE_ITER_ANNOTATE             = 0x2000,
592         TRACE_ITER_USERSTACKTRACE       = 0x4000,
593         TRACE_ITER_SYM_USEROBJ          = 0x8000,
594         TRACE_ITER_PRINTK_MSGONLY       = 0x10000
595 };
596
597 /*
598  * TRACE_ITER_SYM_MASK masks the options in trace_flags that
599  * control the output of kernel symbols.
600  */
601 #define TRACE_ITER_SYM_MASK \
602         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
603
604 extern struct tracer nop_trace;
605
606 /**
607  * ftrace_preempt_disable - disable preemption scheduler safe
608  *
609  * When tracing can happen inside the scheduler, there exists
610  * cases that the tracing might happen before the need_resched
611  * flag is checked. If this happens and the tracer calls
612  * preempt_enable (after a disable), a schedule might take place
613  * causing an infinite recursion.
614  *
615  * To prevent this, we read the need_recshed flag before
616  * disabling preemption. When we want to enable preemption we
617  * check the flag, if it is set, then we call preempt_enable_no_resched.
618  * Otherwise, we call preempt_enable.
619  *
620  * The rational for doing the above is that if need resched is set
621  * and we have yet to reschedule, we are either in an atomic location
622  * (where we do not need to check for scheduling) or we are inside
623  * the scheduler and do not want to resched.
624  */
625 static inline int ftrace_preempt_disable(void)
626 {
627         int resched;
628
629         resched = need_resched();
630         preempt_disable_notrace();
631
632         return resched;
633 }
634
635 /**
636  * ftrace_preempt_enable - enable preemption scheduler safe
637  * @resched: the return value from ftrace_preempt_disable
638  *
639  * This is a scheduler safe way to enable preemption and not miss
640  * any preemption checks. The disabled saved the state of preemption.
641  * If resched is set, then we were either inside an atomic or
642  * are inside the scheduler (we would have already scheduled
643  * otherwise). In this case, we do not want to call normal
644  * preempt_enable, but preempt_enable_no_resched instead.
645  */
646 static inline void ftrace_preempt_enable(int resched)
647 {
648         if (resched)
649                 preempt_enable_no_resched_notrace();
650         else
651                 preempt_enable_notrace();
652 }
653
654 #ifdef CONFIG_BRANCH_TRACER
655 extern int enable_branch_tracing(struct trace_array *tr);
656 extern void disable_branch_tracing(void);
657 static inline int trace_branch_enable(struct trace_array *tr)
658 {
659         if (trace_flags & TRACE_ITER_BRANCH)
660                 return enable_branch_tracing(tr);
661         return 0;
662 }
663 static inline void trace_branch_disable(void)
664 {
665         /* due to races, always disable */
666         disable_branch_tracing();
667 }
668 #else
669 static inline int trace_branch_enable(struct trace_array *tr)
670 {
671         return 0;
672 }
673 static inline void trace_branch_disable(void)
674 {
675 }
676 #endif /* CONFIG_BRANCH_TRACER */
677
678 #endif /* _LINUX_KERNEL_TRACE_H */