Merge git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw
[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_CONT,
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_POWER,
33
34         __TRACE_LAST_TYPE
35 };
36
37 /*
38  * The trace entry - the most basic unit of tracing. This is what
39  * is printed in the end as a single line in the trace output, such as:
40  *
41  *     bash-15816 [01]   235.197585: idle_cpu <- irq_enter
42  */
43 struct trace_entry {
44         unsigned char           type;
45         unsigned char           cpu;
46         unsigned char           flags;
47         unsigned char           preempt_count;
48         int                     pid;
49         int                     tgid;
50 };
51
52 /*
53  * Function trace entry - function address and parent function addres:
54  */
55 struct ftrace_entry {
56         struct trace_entry      ent;
57         unsigned long           ip;
58         unsigned long           parent_ip;
59 };
60
61 /* Function call entry */
62 struct ftrace_graph_ent_entry {
63         struct trace_entry                      ent;
64         struct ftrace_graph_ent         graph_ent;
65 };
66
67 /* Function return entry */
68 struct ftrace_graph_ret_entry {
69         struct trace_entry                      ent;
70         struct ftrace_graph_ret         ret;
71 };
72 extern struct tracer boot_tracer;
73
74 /*
75  * Context switch trace entry - which task (and prio) we switched from/to:
76  */
77 struct ctx_switch_entry {
78         struct trace_entry      ent;
79         unsigned int            prev_pid;
80         unsigned char           prev_prio;
81         unsigned char           prev_state;
82         unsigned int            next_pid;
83         unsigned char           next_prio;
84         unsigned char           next_state;
85         unsigned int            next_cpu;
86 };
87
88 /*
89  * Special (free-form) trace entry:
90  */
91 struct special_entry {
92         struct trace_entry      ent;
93         unsigned long           arg1;
94         unsigned long           arg2;
95         unsigned long           arg3;
96 };
97
98 /*
99  * Stack-trace entry:
100  */
101
102 #define FTRACE_STACK_ENTRIES    8
103
104 struct stack_entry {
105         struct trace_entry      ent;
106         unsigned long           caller[FTRACE_STACK_ENTRIES];
107 };
108
109 struct userstack_entry {
110         struct trace_entry      ent;
111         unsigned long           caller[FTRACE_STACK_ENTRIES];
112 };
113
114 /*
115  * ftrace_printk entry:
116  */
117 struct print_entry {
118         struct trace_entry      ent;
119         unsigned long           ip;
120         int                     depth;
121         char                    buf[];
122 };
123
124 #define TRACE_OLD_SIZE          88
125
126 struct trace_field_cont {
127         unsigned char           type;
128         /* Temporary till we get rid of this completely */
129         char                    buf[TRACE_OLD_SIZE - 1];
130 };
131
132 struct trace_mmiotrace_rw {
133         struct trace_entry      ent;
134         struct mmiotrace_rw     rw;
135 };
136
137 struct trace_mmiotrace_map {
138         struct trace_entry      ent;
139         struct mmiotrace_map    map;
140 };
141
142 struct trace_boot_call {
143         struct trace_entry      ent;
144         struct boot_trace_call boot_call;
145 };
146
147 struct trace_boot_ret {
148         struct trace_entry      ent;
149         struct boot_trace_ret boot_ret;
150 };
151
152 #define TRACE_FUNC_SIZE 30
153 #define TRACE_FILE_SIZE 20
154 struct trace_branch {
155         struct trace_entry      ent;
156         unsigned                line;
157         char                    func[TRACE_FUNC_SIZE+1];
158         char                    file[TRACE_FILE_SIZE+1];
159         char                    correct;
160 };
161
162 struct hw_branch_entry {
163         struct trace_entry      ent;
164         u64                     from;
165         u64                     to;
166 };
167
168 struct trace_power {
169         struct trace_entry      ent;
170         struct power_trace      state_data;
171 };
172
173 /*
174  * trace_flag_type is an enumeration that holds different
175  * states when a trace occurs. These are:
176  *  IRQS_OFF            - interrupts were disabled
177  *  IRQS_NOSUPPORT      - arch does not support irqs_disabled_flags
178  *  NEED_RESCED         - reschedule is requested
179  *  HARDIRQ             - inside an interrupt handler
180  *  SOFTIRQ             - inside a softirq handler
181  *  CONT                - multiple entries hold the trace item
182  */
183 enum trace_flag_type {
184         TRACE_FLAG_IRQS_OFF             = 0x01,
185         TRACE_FLAG_IRQS_NOSUPPORT       = 0x02,
186         TRACE_FLAG_NEED_RESCHED         = 0x04,
187         TRACE_FLAG_HARDIRQ              = 0x08,
188         TRACE_FLAG_SOFTIRQ              = 0x10,
189         TRACE_FLAG_CONT                 = 0x20,
190 };
191
192 #define TRACE_BUF_SIZE          1024
193
194 /*
195  * The CPU trace array - it consists of thousands of trace entries
196  * plus some other descriptor data: (for example which task started
197  * the trace, etc.)
198  */
199 struct trace_array_cpu {
200         atomic_t                disabled;
201
202         /* these fields get copied into max-trace: */
203         unsigned long           trace_idx;
204         unsigned long           overrun;
205         unsigned long           saved_latency;
206         unsigned long           critical_start;
207         unsigned long           critical_end;
208         unsigned long           critical_sequence;
209         unsigned long           nice;
210         unsigned long           policy;
211         unsigned long           rt_priority;
212         cycle_t                 preempt_timestamp;
213         pid_t                   pid;
214         uid_t                   uid;
215         char                    comm[TASK_COMM_LEN];
216 };
217
218 struct trace_iterator;
219
220 /*
221  * The trace array - an array of per-CPU trace arrays. This is the
222  * highest level data structure that individual tracers deal with.
223  * They have on/off state as well:
224  */
225 struct trace_array {
226         struct ring_buffer      *buffer;
227         unsigned long           entries;
228         int                     cpu;
229         cycle_t                 time_start;
230         struct task_struct      *waiter;
231         struct trace_array_cpu  *data[NR_CPUS];
232 };
233
234 #define FTRACE_CMP_TYPE(var, type) \
235         __builtin_types_compatible_p(typeof(var), type *)
236
237 #undef IF_ASSIGN
238 #define IF_ASSIGN(var, entry, etype, id)                \
239         if (FTRACE_CMP_TYPE(var, etype)) {              \
240                 var = (typeof(var))(entry);             \
241                 WARN_ON(id && (entry)->type != id);     \
242                 break;                                  \
243         }
244
245 /* Will cause compile errors if type is not found. */
246 extern void __ftrace_bad_type(void);
247
248 /*
249  * The trace_assign_type is a verifier that the entry type is
250  * the same as the type being assigned. To add new types simply
251  * add a line with the following format:
252  *
253  * IF_ASSIGN(var, ent, type, id);
254  *
255  *  Where "type" is the trace type that includes the trace_entry
256  *  as the "ent" item. And "id" is the trace identifier that is
257  *  used in the trace_type enum.
258  *
259  *  If the type can have more than one id, then use zero.
260  */
261 #define trace_assign_type(var, ent)                                     \
262         do {                                                            \
263                 IF_ASSIGN(var, ent, struct ftrace_entry, TRACE_FN);     \
264                 IF_ASSIGN(var, ent, struct ctx_switch_entry, 0);        \
265                 IF_ASSIGN(var, ent, struct trace_field_cont, TRACE_CONT); \
266                 IF_ASSIGN(var, ent, struct stack_entry, TRACE_STACK);   \
267                 IF_ASSIGN(var, ent, struct userstack_entry, TRACE_USER_STACK);\
268                 IF_ASSIGN(var, ent, struct print_entry, TRACE_PRINT);   \
269                 IF_ASSIGN(var, ent, struct special_entry, 0);           \
270                 IF_ASSIGN(var, ent, struct trace_mmiotrace_rw,          \
271                           TRACE_MMIO_RW);                               \
272                 IF_ASSIGN(var, ent, struct trace_mmiotrace_map,         \
273                           TRACE_MMIO_MAP);                              \
274                 IF_ASSIGN(var, ent, struct trace_boot_call, TRACE_BOOT_CALL);\
275                 IF_ASSIGN(var, ent, struct trace_boot_ret, TRACE_BOOT_RET);\
276                 IF_ASSIGN(var, ent, struct trace_branch, TRACE_BRANCH); \
277                 IF_ASSIGN(var, ent, struct ftrace_graph_ent_entry,      \
278                           TRACE_GRAPH_ENT);             \
279                 IF_ASSIGN(var, ent, struct ftrace_graph_ret_entry,      \
280                           TRACE_GRAPH_RET);             \
281                 IF_ASSIGN(var, ent, struct hw_branch_entry, TRACE_HW_BRANCHES);\
282                 IF_ASSIGN(var, ent, struct trace_power, TRACE_POWER); \
283                 __ftrace_bad_type();                                    \
284         } while (0)
285
286 /* Return values for print_line callback */
287 enum print_line_t {
288         TRACE_TYPE_PARTIAL_LINE = 0,    /* Retry after flushing the seq */
289         TRACE_TYPE_HANDLED      = 1,
290         TRACE_TYPE_UNHANDLED    = 2     /* Relay to other output functions */
291 };
292
293
294 /*
295  * An option specific to a tracer. This is a boolean value.
296  * The bit is the bit index that sets its value on the
297  * flags value in struct tracer_flags.
298  */
299 struct tracer_opt {
300         const char      *name; /* Will appear on the trace_options file */
301         u32             bit; /* Mask assigned in val field in tracer_flags */
302 };
303
304 /*
305  * The set of specific options for a tracer. Your tracer
306  * have to set the initial value of the flags val.
307  */
308 struct tracer_flags {
309         u32                     val;
310         struct tracer_opt       *opts;
311 };
312
313 /* Makes more easy to define a tracer opt */
314 #define TRACER_OPT(s, b)        .name = #s, .bit = b
315
316 /*
317  * A specific tracer, represented by methods that operate on a trace array:
318  */
319 struct tracer {
320         const char              *name;
321         /* Your tracer should raise a warning if init fails */
322         int                     (*init)(struct trace_array *tr);
323         void                    (*reset)(struct trace_array *tr);
324         void                    (*start)(struct trace_array *tr);
325         void                    (*stop)(struct trace_array *tr);
326         void                    (*open)(struct trace_iterator *iter);
327         void                    (*pipe_open)(struct trace_iterator *iter);
328         void                    (*close)(struct trace_iterator *iter);
329         ssize_t                 (*read)(struct trace_iterator *iter,
330                                         struct file *filp, char __user *ubuf,
331                                         size_t cnt, loff_t *ppos);
332 #ifdef CONFIG_FTRACE_STARTUP_TEST
333         int                     (*selftest)(struct tracer *trace,
334                                             struct trace_array *tr);
335 #endif
336         void                    (*print_header)(struct seq_file *m);
337         enum print_line_t       (*print_line)(struct trace_iterator *iter);
338         /* If you handled the flag setting, return 0 */
339         int                     (*set_flag)(u32 old_flags, u32 bit, int set);
340         struct tracer           *next;
341         int                     print_max;
342         struct tracer_flags     *flags;
343 };
344
345 struct trace_seq {
346         unsigned char           buffer[PAGE_SIZE];
347         unsigned int            len;
348         unsigned int            readpos;
349 };
350
351 /*
352  * Trace iterator - used by printout routines who present trace
353  * results to users and which routines might sleep, etc:
354  */
355 struct trace_iterator {
356         struct trace_array      *tr;
357         struct tracer           *trace;
358         void                    *private;
359         struct ring_buffer_iter *buffer_iter[NR_CPUS];
360
361         /* The below is zeroed out in pipe_read */
362         struct trace_seq        seq;
363         struct trace_entry      *ent;
364         int                     cpu;
365         u64                     ts;
366
367         unsigned long           iter_flags;
368         loff_t                  pos;
369         long                    idx;
370
371         cpumask_var_t           started;
372 };
373
374 int tracing_is_enabled(void);
375 void trace_wake_up(void);
376 void tracing_reset(struct trace_array *tr, int cpu);
377 void tracing_reset_online_cpus(struct trace_array *tr);
378 int tracing_open_generic(struct inode *inode, struct file *filp);
379 struct dentry *tracing_init_dentry(void);
380 void init_tracer_sysprof_debugfs(struct dentry *d_tracer);
381
382 struct trace_entry *tracing_get_trace_entry(struct trace_array *tr,
383                                                 struct trace_array_cpu *data);
384 void tracing_generic_entry_update(struct trace_entry *entry,
385                                   unsigned long flags,
386                                   int pc);
387
388 void ftrace(struct trace_array *tr,
389                             struct trace_array_cpu *data,
390                             unsigned long ip,
391                             unsigned long parent_ip,
392                             unsigned long flags, int pc);
393 void tracing_sched_switch_trace(struct trace_array *tr,
394                                 struct trace_array_cpu *data,
395                                 struct task_struct *prev,
396                                 struct task_struct *next,
397                                 unsigned long flags, int pc);
398 void tracing_record_cmdline(struct task_struct *tsk);
399
400 void tracing_sched_wakeup_trace(struct trace_array *tr,
401                                 struct trace_array_cpu *data,
402                                 struct task_struct *wakee,
403                                 struct task_struct *cur,
404                                 unsigned long flags, int pc);
405 void trace_special(struct trace_array *tr,
406                    struct trace_array_cpu *data,
407                    unsigned long arg1,
408                    unsigned long arg2,
409                    unsigned long arg3, int pc);
410 void trace_function(struct trace_array *tr,
411                     struct trace_array_cpu *data,
412                     unsigned long ip,
413                     unsigned long parent_ip,
414                     unsigned long flags, int pc);
415
416 void trace_graph_return(struct ftrace_graph_ret *trace);
417 int trace_graph_entry(struct ftrace_graph_ent *trace);
418 void trace_hw_branch(struct trace_array *tr, u64 from, u64 to);
419
420 void tracing_start_cmdline_record(void);
421 void tracing_stop_cmdline_record(void);
422 void tracing_sched_switch_assign_trace(struct trace_array *tr);
423 void tracing_stop_sched_switch_record(void);
424 void tracing_start_sched_switch_record(void);
425 int register_tracer(struct tracer *type);
426 void unregister_tracer(struct tracer *type);
427
428 extern unsigned long nsecs_to_usecs(unsigned long nsecs);
429
430 extern unsigned long tracing_max_latency;
431 extern unsigned long tracing_thresh;
432
433 void update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu);
434 void update_max_tr_single(struct trace_array *tr,
435                           struct task_struct *tsk, int cpu);
436
437 extern cycle_t ftrace_now(int cpu);
438
439 #ifdef CONFIG_FUNCTION_TRACER
440 void tracing_start_function_trace(void);
441 void tracing_stop_function_trace(void);
442 #else
443 # define tracing_start_function_trace()         do { } while (0)
444 # define tracing_stop_function_trace()          do { } while (0)
445 #endif
446
447 #ifdef CONFIG_CONTEXT_SWITCH_TRACER
448 typedef void
449 (*tracer_switch_func_t)(void *private,
450                         void *__rq,
451                         struct task_struct *prev,
452                         struct task_struct *next);
453
454 struct tracer_switch_ops {
455         tracer_switch_func_t            func;
456         void                            *private;
457         struct tracer_switch_ops        *next;
458 };
459
460 char *trace_find_cmdline(int pid);
461 #endif /* CONFIG_CONTEXT_SWITCH_TRACER */
462
463 #ifdef CONFIG_DYNAMIC_FTRACE
464 extern unsigned long ftrace_update_tot_cnt;
465 #define DYN_FTRACE_TEST_NAME trace_selftest_dynamic_test_func
466 extern int DYN_FTRACE_TEST_NAME(void);
467 #endif
468
469 #ifdef CONFIG_FTRACE_STARTUP_TEST
470 extern int trace_selftest_startup_function(struct tracer *trace,
471                                            struct trace_array *tr);
472 extern int trace_selftest_startup_irqsoff(struct tracer *trace,
473                                           struct trace_array *tr);
474 extern int trace_selftest_startup_preemptoff(struct tracer *trace,
475                                              struct trace_array *tr);
476 extern int trace_selftest_startup_preemptirqsoff(struct tracer *trace,
477                                                  struct trace_array *tr);
478 extern int trace_selftest_startup_wakeup(struct tracer *trace,
479                                          struct trace_array *tr);
480 extern int trace_selftest_startup_nop(struct tracer *trace,
481                                          struct trace_array *tr);
482 extern int trace_selftest_startup_sched_switch(struct tracer *trace,
483                                                struct trace_array *tr);
484 extern int trace_selftest_startup_sysprof(struct tracer *trace,
485                                                struct trace_array *tr);
486 extern int trace_selftest_startup_branch(struct tracer *trace,
487                                          struct trace_array *tr);
488 #endif /* CONFIG_FTRACE_STARTUP_TEST */
489
490 extern void *head_page(struct trace_array_cpu *data);
491 extern int trace_seq_printf(struct trace_seq *s, const char *fmt, ...);
492 extern void trace_seq_print_cont(struct trace_seq *s,
493                                  struct trace_iterator *iter);
494
495 extern int
496 seq_print_ip_sym(struct trace_seq *s, unsigned long ip,
497                 unsigned long sym_flags);
498 extern ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf,
499                                  size_t cnt);
500 extern long ns2usecs(cycle_t nsec);
501 extern int
502 trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args);
503
504 extern unsigned long trace_flags;
505
506 /* Standard output formatting function used for function return traces */
507 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
508 extern enum print_line_t print_graph_function(struct trace_iterator *iter);
509
510 #ifdef CONFIG_DYNAMIC_FTRACE
511 /* TODO: make this variable */
512 #define FTRACE_GRAPH_MAX_FUNCS          32
513 extern int ftrace_graph_count;
514 extern unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS];
515
516 static inline int ftrace_graph_addr(unsigned long addr)
517 {
518         int i;
519
520         if (!ftrace_graph_count || test_tsk_trace_graph(current))
521                 return 1;
522
523         for (i = 0; i < ftrace_graph_count; i++) {
524                 if (addr == ftrace_graph_funcs[i])
525                         return 1;
526         }
527
528         return 0;
529 }
530 #else
531 static inline int ftrace_trace_addr(unsigned long addr)
532 {
533         return 1;
534 }
535 static inline int ftrace_graph_addr(unsigned long addr)
536 {
537         return 1;
538 }
539 #endif /* CONFIG_DYNAMIC_FTRACE */
540
541 #else /* CONFIG_FUNCTION_GRAPH_TRACER */
542 static inline enum print_line_t
543 print_graph_function(struct trace_iterator *iter)
544 {
545         return TRACE_TYPE_UNHANDLED;
546 }
547 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
548
549 extern struct pid *ftrace_pid_trace;
550
551 static inline int ftrace_trace_task(struct task_struct *task)
552 {
553         if (!ftrace_pid_trace)
554                 return 1;
555
556         return test_tsk_trace_trace(task);
557 }
558
559 /*
560  * trace_iterator_flags is an enumeration that defines bit
561  * positions into trace_flags that controls the output.
562  *
563  * NOTE: These bits must match the trace_options array in
564  *       trace.c.
565  */
566 enum trace_iterator_flags {
567         TRACE_ITER_PRINT_PARENT         = 0x01,
568         TRACE_ITER_SYM_OFFSET           = 0x02,
569         TRACE_ITER_SYM_ADDR             = 0x04,
570         TRACE_ITER_VERBOSE              = 0x08,
571         TRACE_ITER_RAW                  = 0x10,
572         TRACE_ITER_HEX                  = 0x20,
573         TRACE_ITER_BIN                  = 0x40,
574         TRACE_ITER_BLOCK                = 0x80,
575         TRACE_ITER_STACKTRACE           = 0x100,
576         TRACE_ITER_SCHED_TREE           = 0x200,
577         TRACE_ITER_PRINTK               = 0x400,
578         TRACE_ITER_PREEMPTONLY          = 0x800,
579         TRACE_ITER_BRANCH               = 0x1000,
580         TRACE_ITER_ANNOTATE             = 0x2000,
581         TRACE_ITER_USERSTACKTRACE       = 0x4000,
582         TRACE_ITER_SYM_USEROBJ          = 0x8000,
583         TRACE_ITER_PRINTK_MSGONLY       = 0x10000
584 };
585
586 /*
587  * TRACE_ITER_SYM_MASK masks the options in trace_flags that
588  * control the output of kernel symbols.
589  */
590 #define TRACE_ITER_SYM_MASK \
591         (TRACE_ITER_PRINT_PARENT|TRACE_ITER_SYM_OFFSET|TRACE_ITER_SYM_ADDR)
592
593 extern struct tracer nop_trace;
594
595 /**
596  * ftrace_preempt_disable - disable preemption scheduler safe
597  *
598  * When tracing can happen inside the scheduler, there exists
599  * cases that the tracing might happen before the need_resched
600  * flag is checked. If this happens and the tracer calls
601  * preempt_enable (after a disable), a schedule might take place
602  * causing an infinite recursion.
603  *
604  * To prevent this, we read the need_recshed flag before
605  * disabling preemption. When we want to enable preemption we
606  * check the flag, if it is set, then we call preempt_enable_no_resched.
607  * Otherwise, we call preempt_enable.
608  *
609  * The rational for doing the above is that if need resched is set
610  * and we have yet to reschedule, we are either in an atomic location
611  * (where we do not need to check for scheduling) or we are inside
612  * the scheduler and do not want to resched.
613  */
614 static inline int ftrace_preempt_disable(void)
615 {
616         int resched;
617
618         resched = need_resched();
619         preempt_disable_notrace();
620
621         return resched;
622 }
623
624 /**
625  * ftrace_preempt_enable - enable preemption scheduler safe
626  * @resched: the return value from ftrace_preempt_disable
627  *
628  * This is a scheduler safe way to enable preemption and not miss
629  * any preemption checks. The disabled saved the state of preemption.
630  * If resched is set, then we were either inside an atomic or
631  * are inside the scheduler (we would have already scheduled
632  * otherwise). In this case, we do not want to call normal
633  * preempt_enable, but preempt_enable_no_resched instead.
634  */
635 static inline void ftrace_preempt_enable(int resched)
636 {
637         if (resched)
638                 preempt_enable_no_resched_notrace();
639         else
640                 preempt_enable_notrace();
641 }
642
643 #ifdef CONFIG_BRANCH_TRACER
644 extern int enable_branch_tracing(struct trace_array *tr);
645 extern void disable_branch_tracing(void);
646 static inline int trace_branch_enable(struct trace_array *tr)
647 {
648         if (trace_flags & TRACE_ITER_BRANCH)
649                 return enable_branch_tracing(tr);
650         return 0;
651 }
652 static inline void trace_branch_disable(void)
653 {
654         /* due to races, always disable */
655         disable_branch_tracing();
656 }
657 #else
658 static inline int trace_branch_enable(struct trace_array *tr)
659 {
660         return 0;
661 }
662 static inline void trace_branch_disable(void)
663 {
664 }
665 #endif /* CONFIG_BRANCH_TRACER */
666
667 #endif /* _LINUX_KERNEL_TRACE_H */