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