trace: do not disable wake up tracer on output of trace
[linux-2.6] / kernel / trace / trace.c
1 /*
2  * ring buffer based function tracer
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally taken from the RT patch by:
8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code from the latency_tracer, that is:
11  *  Copyright (C) 2004-2006 Ingo Molnar
12  *  Copyright (C) 2004 William Lee Irwin III
13  */
14 #include <linux/utsrelease.h>
15 #include <linux/kallsyms.h>
16 #include <linux/seq_file.h>
17 #include <linux/notifier.h>
18 #include <linux/debugfs.h>
19 #include <linux/pagemap.h>
20 #include <linux/hardirq.h>
21 #include <linux/linkage.h>
22 #include <linux/uaccess.h>
23 #include <linux/ftrace.h>
24 #include <linux/module.h>
25 #include <linux/percpu.h>
26 #include <linux/kdebug.h>
27 #include <linux/ctype.h>
28 #include <linux/init.h>
29 #include <linux/poll.h>
30 #include <linux/gfp.h>
31 #include <linux/fs.h>
32 #include <linux/kprobes.h>
33 #include <linux/writeback.h>
34
35 #include <linux/stacktrace.h>
36 #include <linux/ring_buffer.h>
37 #include <linux/irqflags.h>
38
39 #include "trace.h"
40 #include "trace_output.h"
41
42 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
43
44 unsigned long __read_mostly     tracing_max_latency;
45 unsigned long __read_mostly     tracing_thresh;
46
47 /*
48  * We need to change this state when a selftest is running.
49  * A selftest will lurk into the ring-buffer to count the
50  * entries inserted during the selftest although some concurrent
51  * insertions into the ring-buffer such as ftrace_printk could occurred
52  * at the same time, giving false positive or negative results.
53  */
54 static bool __read_mostly tracing_selftest_running;
55
56 /* For tracers that don't implement custom flags */
57 static struct tracer_opt dummy_tracer_opt[] = {
58         { }
59 };
60
61 static struct tracer_flags dummy_tracer_flags = {
62         .val = 0,
63         .opts = dummy_tracer_opt
64 };
65
66 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
67 {
68         return 0;
69 }
70
71 /*
72  * Kill all tracing for good (never come back).
73  * It is initialized to 1 but will turn to zero if the initialization
74  * of the tracer is successful. But that is the only place that sets
75  * this back to zero.
76  */
77 int tracing_disabled = 1;
78
79 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
80
81 static inline void ftrace_disable_cpu(void)
82 {
83         preempt_disable();
84         local_inc(&__get_cpu_var(ftrace_cpu_disabled));
85 }
86
87 static inline void ftrace_enable_cpu(void)
88 {
89         local_dec(&__get_cpu_var(ftrace_cpu_disabled));
90         preempt_enable();
91 }
92
93 static cpumask_var_t __read_mostly      tracing_buffer_mask;
94
95 #define for_each_tracing_cpu(cpu)       \
96         for_each_cpu(cpu, tracing_buffer_mask)
97
98 /*
99  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
100  *
101  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
102  * is set, then ftrace_dump is called. This will output the contents
103  * of the ftrace buffers to the console.  This is very useful for
104  * capturing traces that lead to crashes and outputing it to a
105  * serial console.
106  *
107  * It is default off, but you can enable it with either specifying
108  * "ftrace_dump_on_oops" in the kernel command line, or setting
109  * /proc/sys/kernel/ftrace_dump_on_oops to true.
110  */
111 int ftrace_dump_on_oops;
112
113 static int tracing_set_tracer(char *buf);
114
115 static int __init set_ftrace(char *str)
116 {
117         tracing_set_tracer(str);
118         return 1;
119 }
120 __setup("ftrace", set_ftrace);
121
122 static int __init set_ftrace_dump_on_oops(char *str)
123 {
124         ftrace_dump_on_oops = 1;
125         return 1;
126 }
127 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
128
129 long
130 ns2usecs(cycle_t nsec)
131 {
132         nsec += 500;
133         do_div(nsec, 1000);
134         return nsec;
135 }
136
137 cycle_t ftrace_now(int cpu)
138 {
139         u64 ts = ring_buffer_time_stamp(cpu);
140         ring_buffer_normalize_time_stamp(cpu, &ts);
141         return ts;
142 }
143
144 /*
145  * The global_trace is the descriptor that holds the tracing
146  * buffers for the live tracing. For each CPU, it contains
147  * a link list of pages that will store trace entries. The
148  * page descriptor of the pages in the memory is used to hold
149  * the link list by linking the lru item in the page descriptor
150  * to each of the pages in the buffer per CPU.
151  *
152  * For each active CPU there is a data field that holds the
153  * pages for the buffer for that CPU. Each CPU has the same number
154  * of pages allocated for its buffer.
155  */
156 static struct trace_array       global_trace;
157
158 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
159
160 /*
161  * The max_tr is used to snapshot the global_trace when a maximum
162  * latency is reached. Some tracers will use this to store a maximum
163  * trace while it continues examining live traces.
164  *
165  * The buffers for the max_tr are set up the same as the global_trace.
166  * When a snapshot is taken, the link list of the max_tr is swapped
167  * with the link list of the global_trace and the buffers are reset for
168  * the global_trace so the tracing can continue.
169  */
170 static struct trace_array       max_tr;
171
172 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
173
174 /* tracer_enabled is used to toggle activation of a tracer */
175 static int                      tracer_enabled = 1;
176
177 /**
178  * tracing_is_enabled - return tracer_enabled status
179  *
180  * This function is used by other tracers to know the status
181  * of the tracer_enabled flag.  Tracers may use this function
182  * to know if it should enable their features when starting
183  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
184  */
185 int tracing_is_enabled(void)
186 {
187         return tracer_enabled;
188 }
189
190 /*
191  * trace_buf_size is the size in bytes that is allocated
192  * for a buffer. Note, the number of bytes is always rounded
193  * to page size.
194  *
195  * This number is purposely set to a low number of 16384.
196  * If the dump on oops happens, it will be much appreciated
197  * to not have to wait for all that output. Anyway this can be
198  * boot time and run time configurable.
199  */
200 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
201
202 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
203
204 /* trace_types holds a link list of available tracers. */
205 static struct tracer            *trace_types __read_mostly;
206
207 /* current_trace points to the tracer that is currently active */
208 static struct tracer            *current_trace __read_mostly;
209
210 /*
211  * max_tracer_type_len is used to simplify the allocating of
212  * buffers to read userspace tracer names. We keep track of
213  * the longest tracer name registered.
214  */
215 static int                      max_tracer_type_len;
216
217 /*
218  * trace_types_lock is used to protect the trace_types list.
219  * This lock is also used to keep user access serialized.
220  * Accesses from userspace will grab this lock while userspace
221  * activities happen inside the kernel.
222  */
223 static DEFINE_MUTEX(trace_types_lock);
224
225 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
226 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
227
228 /* trace_flags holds trace_options default values */
229 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
230         TRACE_ITER_ANNOTATE;
231
232 /**
233  * trace_wake_up - wake up tasks waiting for trace input
234  *
235  * Simply wakes up any task that is blocked on the trace_wait
236  * queue. These is used with trace_poll for tasks polling the trace.
237  */
238 void trace_wake_up(void)
239 {
240         /*
241          * The runqueue_is_locked() can fail, but this is the best we
242          * have for now:
243          */
244         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
245                 wake_up(&trace_wait);
246 }
247
248 static int __init set_buf_size(char *str)
249 {
250         unsigned long buf_size;
251         int ret;
252
253         if (!str)
254                 return 0;
255         ret = strict_strtoul(str, 0, &buf_size);
256         /* nr_entries can not be zero */
257         if (ret < 0 || buf_size == 0)
258                 return 0;
259         trace_buf_size = buf_size;
260         return 1;
261 }
262 __setup("trace_buf_size=", set_buf_size);
263
264 unsigned long nsecs_to_usecs(unsigned long nsecs)
265 {
266         return nsecs / 1000;
267 }
268
269 /* These must match the bit postions in trace_iterator_flags */
270 static const char *trace_options[] = {
271         "print-parent",
272         "sym-offset",
273         "sym-addr",
274         "verbose",
275         "raw",
276         "hex",
277         "bin",
278         "block",
279         "stacktrace",
280         "sched-tree",
281         "ftrace_printk",
282         "ftrace_preempt",
283         "branch",
284         "annotate",
285         "userstacktrace",
286         "sym-userobj",
287         "printk-msg-only",
288         NULL
289 };
290
291 /*
292  * ftrace_max_lock is used to protect the swapping of buffers
293  * when taking a max snapshot. The buffers themselves are
294  * protected by per_cpu spinlocks. But the action of the swap
295  * needs its own lock.
296  *
297  * This is defined as a raw_spinlock_t in order to help
298  * with performance when lockdep debugging is enabled.
299  */
300 static raw_spinlock_t ftrace_max_lock =
301         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
302
303 /*
304  * Copy the new maximum trace into the separate maximum-trace
305  * structure. (this way the maximum trace is permanently saved,
306  * for later retrieval via /debugfs/tracing/latency_trace)
307  */
308 static void
309 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
310 {
311         struct trace_array_cpu *data = tr->data[cpu];
312
313         max_tr.cpu = cpu;
314         max_tr.time_start = data->preempt_timestamp;
315
316         data = max_tr.data[cpu];
317         data->saved_latency = tracing_max_latency;
318
319         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
320         data->pid = tsk->pid;
321         data->uid = task_uid(tsk);
322         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
323         data->policy = tsk->policy;
324         data->rt_priority = tsk->rt_priority;
325
326         /* record this tasks comm */
327         tracing_record_cmdline(current);
328 }
329
330 static void
331 trace_seq_reset(struct trace_seq *s)
332 {
333         s->len = 0;
334         s->readpos = 0;
335 }
336
337 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
338 {
339         int len;
340         int ret;
341
342         if (s->len <= s->readpos)
343                 return -EBUSY;
344
345         len = s->len - s->readpos;
346         if (cnt > len)
347                 cnt = len;
348         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
349         if (ret)
350                 return -EFAULT;
351
352         s->readpos += len;
353         return cnt;
354 }
355
356 static void
357 trace_print_seq(struct seq_file *m, struct trace_seq *s)
358 {
359         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
360
361         s->buffer[len] = 0;
362         seq_puts(m, s->buffer);
363
364         trace_seq_reset(s);
365 }
366
367 /**
368  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
369  * @tr: tracer
370  * @tsk: the task with the latency
371  * @cpu: The cpu that initiated the trace.
372  *
373  * Flip the buffers between the @tr and the max_tr and record information
374  * about which task was the cause of this latency.
375  */
376 void
377 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
378 {
379         struct ring_buffer *buf = tr->buffer;
380
381         WARN_ON_ONCE(!irqs_disabled());
382         __raw_spin_lock(&ftrace_max_lock);
383
384         tr->buffer = max_tr.buffer;
385         max_tr.buffer = buf;
386
387         ftrace_disable_cpu();
388         ring_buffer_reset(tr->buffer);
389         ftrace_enable_cpu();
390
391         __update_max_tr(tr, tsk, cpu);
392         __raw_spin_unlock(&ftrace_max_lock);
393 }
394
395 /**
396  * update_max_tr_single - only copy one trace over, and reset the rest
397  * @tr - tracer
398  * @tsk - task with the latency
399  * @cpu - the cpu of the buffer to copy.
400  *
401  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
402  */
403 void
404 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
405 {
406         int ret;
407
408         WARN_ON_ONCE(!irqs_disabled());
409         __raw_spin_lock(&ftrace_max_lock);
410
411         ftrace_disable_cpu();
412
413         ring_buffer_reset(max_tr.buffer);
414         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
415
416         ftrace_enable_cpu();
417
418         WARN_ON_ONCE(ret);
419
420         __update_max_tr(tr, tsk, cpu);
421         __raw_spin_unlock(&ftrace_max_lock);
422 }
423
424 /**
425  * register_tracer - register a tracer with the ftrace system.
426  * @type - the plugin for the tracer
427  *
428  * Register a new plugin tracer.
429  */
430 int register_tracer(struct tracer *type)
431 {
432         struct tracer *t;
433         int len;
434         int ret = 0;
435
436         if (!type->name) {
437                 pr_info("Tracer must have a name\n");
438                 return -1;
439         }
440
441         /*
442          * When this gets called we hold the BKL which means that
443          * preemption is disabled. Various trace selftests however
444          * need to disable and enable preemption for successful tests.
445          * So we drop the BKL here and grab it after the tests again.
446          */
447         unlock_kernel();
448         mutex_lock(&trace_types_lock);
449
450         tracing_selftest_running = true;
451
452         for (t = trace_types; t; t = t->next) {
453                 if (strcmp(type->name, t->name) == 0) {
454                         /* already found */
455                         pr_info("Trace %s already registered\n",
456                                 type->name);
457                         ret = -1;
458                         goto out;
459                 }
460         }
461
462         if (!type->set_flag)
463                 type->set_flag = &dummy_set_flag;
464         if (!type->flags)
465                 type->flags = &dummy_tracer_flags;
466         else
467                 if (!type->flags->opts)
468                         type->flags->opts = dummy_tracer_opt;
469
470 #ifdef CONFIG_FTRACE_STARTUP_TEST
471         if (type->selftest) {
472                 struct tracer *saved_tracer = current_trace;
473                 struct trace_array *tr = &global_trace;
474                 int i;
475
476                 /*
477                  * Run a selftest on this tracer.
478                  * Here we reset the trace buffer, and set the current
479                  * tracer to be this tracer. The tracer can then run some
480                  * internal tracing to verify that everything is in order.
481                  * If we fail, we do not register this tracer.
482                  */
483                 for_each_tracing_cpu(i)
484                         tracing_reset(tr, i);
485
486                 current_trace = type;
487                 /* the test is responsible for initializing and enabling */
488                 pr_info("Testing tracer %s: ", type->name);
489                 ret = type->selftest(type, tr);
490                 /* the test is responsible for resetting too */
491                 current_trace = saved_tracer;
492                 if (ret) {
493                         printk(KERN_CONT "FAILED!\n");
494                         goto out;
495                 }
496                 /* Only reset on passing, to avoid touching corrupted buffers */
497                 for_each_tracing_cpu(i)
498                         tracing_reset(tr, i);
499
500                 printk(KERN_CONT "PASSED\n");
501         }
502 #endif
503
504         type->next = trace_types;
505         trace_types = type;
506         len = strlen(type->name);
507         if (len > max_tracer_type_len)
508                 max_tracer_type_len = len;
509
510  out:
511         tracing_selftest_running = false;
512         mutex_unlock(&trace_types_lock);
513         lock_kernel();
514
515         return ret;
516 }
517
518 void unregister_tracer(struct tracer *type)
519 {
520         struct tracer **t;
521         int len;
522
523         mutex_lock(&trace_types_lock);
524         for (t = &trace_types; *t; t = &(*t)->next) {
525                 if (*t == type)
526                         goto found;
527         }
528         pr_info("Trace %s not registered\n", type->name);
529         goto out;
530
531  found:
532         *t = (*t)->next;
533         if (strlen(type->name) != max_tracer_type_len)
534                 goto out;
535
536         max_tracer_type_len = 0;
537         for (t = &trace_types; *t; t = &(*t)->next) {
538                 len = strlen((*t)->name);
539                 if (len > max_tracer_type_len)
540                         max_tracer_type_len = len;
541         }
542  out:
543         mutex_unlock(&trace_types_lock);
544 }
545
546 void tracing_reset(struct trace_array *tr, int cpu)
547 {
548         ftrace_disable_cpu();
549         ring_buffer_reset_cpu(tr->buffer, cpu);
550         ftrace_enable_cpu();
551 }
552
553 void tracing_reset_online_cpus(struct trace_array *tr)
554 {
555         int cpu;
556
557         tr->time_start = ftrace_now(tr->cpu);
558
559         for_each_online_cpu(cpu)
560                 tracing_reset(tr, cpu);
561 }
562
563 #define SAVED_CMDLINES 128
564 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
565 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
566 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
567 static int cmdline_idx;
568 static DEFINE_SPINLOCK(trace_cmdline_lock);
569
570 /* temporary disable recording */
571 atomic_t trace_record_cmdline_disabled __read_mostly;
572
573 static void trace_init_cmdlines(void)
574 {
575         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
576         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
577         cmdline_idx = 0;
578 }
579
580 static int trace_stop_count;
581 static DEFINE_SPINLOCK(tracing_start_lock);
582
583 /**
584  * ftrace_off_permanent - disable all ftrace code permanently
585  *
586  * This should only be called when a serious anomally has
587  * been detected.  This will turn off the function tracing,
588  * ring buffers, and other tracing utilites. It takes no
589  * locks and can be called from any context.
590  */
591 void ftrace_off_permanent(void)
592 {
593         tracing_disabled = 1;
594         ftrace_stop();
595         tracing_off_permanent();
596 }
597
598 /**
599  * tracing_start - quick start of the tracer
600  *
601  * If tracing is enabled but was stopped by tracing_stop,
602  * this will start the tracer back up.
603  */
604 void tracing_start(void)
605 {
606         struct ring_buffer *buffer;
607         unsigned long flags;
608
609         if (tracing_disabled)
610                 return;
611
612         spin_lock_irqsave(&tracing_start_lock, flags);
613         if (--trace_stop_count)
614                 goto out;
615
616         if (trace_stop_count < 0) {
617                 /* Someone screwed up their debugging */
618                 WARN_ON_ONCE(1);
619                 trace_stop_count = 0;
620                 goto out;
621         }
622
623
624         buffer = global_trace.buffer;
625         if (buffer)
626                 ring_buffer_record_enable(buffer);
627
628         buffer = max_tr.buffer;
629         if (buffer)
630                 ring_buffer_record_enable(buffer);
631
632         ftrace_start();
633  out:
634         spin_unlock_irqrestore(&tracing_start_lock, flags);
635 }
636
637 /**
638  * tracing_stop - quick stop of the tracer
639  *
640  * Light weight way to stop tracing. Use in conjunction with
641  * tracing_start.
642  */
643 void tracing_stop(void)
644 {
645         struct ring_buffer *buffer;
646         unsigned long flags;
647
648         ftrace_stop();
649         spin_lock_irqsave(&tracing_start_lock, flags);
650         if (trace_stop_count++)
651                 goto out;
652
653         buffer = global_trace.buffer;
654         if (buffer)
655                 ring_buffer_record_disable(buffer);
656
657         buffer = max_tr.buffer;
658         if (buffer)
659                 ring_buffer_record_disable(buffer);
660
661  out:
662         spin_unlock_irqrestore(&tracing_start_lock, flags);
663 }
664
665 void trace_stop_cmdline_recording(void);
666
667 static void trace_save_cmdline(struct task_struct *tsk)
668 {
669         unsigned map;
670         unsigned idx;
671
672         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
673                 return;
674
675         /*
676          * It's not the end of the world if we don't get
677          * the lock, but we also don't want to spin
678          * nor do we want to disable interrupts,
679          * so if we miss here, then better luck next time.
680          */
681         if (!spin_trylock(&trace_cmdline_lock))
682                 return;
683
684         idx = map_pid_to_cmdline[tsk->pid];
685         if (idx >= SAVED_CMDLINES) {
686                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
687
688                 map = map_cmdline_to_pid[idx];
689                 if (map <= PID_MAX_DEFAULT)
690                         map_pid_to_cmdline[map] = (unsigned)-1;
691
692                 map_pid_to_cmdline[tsk->pid] = idx;
693
694                 cmdline_idx = idx;
695         }
696
697         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
698
699         spin_unlock(&trace_cmdline_lock);
700 }
701
702 char *trace_find_cmdline(int pid)
703 {
704         char *cmdline = "<...>";
705         unsigned map;
706
707         if (!pid)
708                 return "<idle>";
709
710         if (pid > PID_MAX_DEFAULT)
711                 goto out;
712
713         map = map_pid_to_cmdline[pid];
714         if (map >= SAVED_CMDLINES)
715                 goto out;
716
717         cmdline = saved_cmdlines[map];
718
719  out:
720         return cmdline;
721 }
722
723 void tracing_record_cmdline(struct task_struct *tsk)
724 {
725         if (atomic_read(&trace_record_cmdline_disabled))
726                 return;
727
728         trace_save_cmdline(tsk);
729 }
730
731 void
732 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
733                              int pc)
734 {
735         struct task_struct *tsk = current;
736
737         entry->preempt_count            = pc & 0xff;
738         entry->pid                      = (tsk) ? tsk->pid : 0;
739         entry->tgid                     = (tsk) ? tsk->tgid : 0;
740         entry->flags =
741 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
742                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
743 #else
744                 TRACE_FLAG_IRQS_NOSUPPORT |
745 #endif
746                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
747                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
748                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
749 }
750
751 void
752 trace_function(struct trace_array *tr, struct trace_array_cpu *data,
753                unsigned long ip, unsigned long parent_ip, unsigned long flags,
754                int pc)
755 {
756         struct ring_buffer_event *event;
757         struct ftrace_entry *entry;
758         unsigned long irq_flags;
759
760         /* If we are reading the ring buffer, don't trace */
761         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
762                 return;
763
764         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
765                                          &irq_flags);
766         if (!event)
767                 return;
768         entry   = ring_buffer_event_data(event);
769         tracing_generic_entry_update(&entry->ent, flags, pc);
770         entry->ent.type                 = TRACE_FN;
771         entry->ip                       = ip;
772         entry->parent_ip                = parent_ip;
773         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
774 }
775
776 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
777 static void __trace_graph_entry(struct trace_array *tr,
778                                 struct trace_array_cpu *data,
779                                 struct ftrace_graph_ent *trace,
780                                 unsigned long flags,
781                                 int pc)
782 {
783         struct ring_buffer_event *event;
784         struct ftrace_graph_ent_entry *entry;
785         unsigned long irq_flags;
786
787         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
788                 return;
789
790         event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
791                                          &irq_flags);
792         if (!event)
793                 return;
794         entry   = ring_buffer_event_data(event);
795         tracing_generic_entry_update(&entry->ent, flags, pc);
796         entry->ent.type                 = TRACE_GRAPH_ENT;
797         entry->graph_ent                        = *trace;
798         ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
799 }
800
801 static void __trace_graph_return(struct trace_array *tr,
802                                 struct trace_array_cpu *data,
803                                 struct ftrace_graph_ret *trace,
804                                 unsigned long flags,
805                                 int pc)
806 {
807         struct ring_buffer_event *event;
808         struct ftrace_graph_ret_entry *entry;
809         unsigned long irq_flags;
810
811         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
812                 return;
813
814         event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry),
815                                          &irq_flags);
816         if (!event)
817                 return;
818         entry   = ring_buffer_event_data(event);
819         tracing_generic_entry_update(&entry->ent, flags, pc);
820         entry->ent.type                 = TRACE_GRAPH_RET;
821         entry->ret                              = *trace;
822         ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags);
823 }
824 #endif
825
826 void
827 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
828        unsigned long ip, unsigned long parent_ip, unsigned long flags,
829        int pc)
830 {
831         if (likely(!atomic_read(&data->disabled)))
832                 trace_function(tr, data, ip, parent_ip, flags, pc);
833 }
834
835 static void __ftrace_trace_stack(struct trace_array *tr,
836                                  struct trace_array_cpu *data,
837                                  unsigned long flags,
838                                  int skip, int pc)
839 {
840 #ifdef CONFIG_STACKTRACE
841         struct ring_buffer_event *event;
842         struct stack_entry *entry;
843         struct stack_trace trace;
844         unsigned long irq_flags;
845
846         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
847                                          &irq_flags);
848         if (!event)
849                 return;
850         entry   = ring_buffer_event_data(event);
851         tracing_generic_entry_update(&entry->ent, flags, pc);
852         entry->ent.type         = TRACE_STACK;
853
854         memset(&entry->caller, 0, sizeof(entry->caller));
855
856         trace.nr_entries        = 0;
857         trace.max_entries       = FTRACE_STACK_ENTRIES;
858         trace.skip              = skip;
859         trace.entries           = entry->caller;
860
861         save_stack_trace(&trace);
862         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
863 #endif
864 }
865
866 static void ftrace_trace_stack(struct trace_array *tr,
867                                struct trace_array_cpu *data,
868                                unsigned long flags,
869                                int skip, int pc)
870 {
871         if (!(trace_flags & TRACE_ITER_STACKTRACE))
872                 return;
873
874         __ftrace_trace_stack(tr, data, flags, skip, pc);
875 }
876
877 void __trace_stack(struct trace_array *tr,
878                    struct trace_array_cpu *data,
879                    unsigned long flags,
880                    int skip, int pc)
881 {
882         __ftrace_trace_stack(tr, data, flags, skip, pc);
883 }
884
885 static void ftrace_trace_userstack(struct trace_array *tr,
886                    struct trace_array_cpu *data,
887                    unsigned long flags, int pc)
888 {
889 #ifdef CONFIG_STACKTRACE
890         struct ring_buffer_event *event;
891         struct userstack_entry *entry;
892         struct stack_trace trace;
893         unsigned long irq_flags;
894
895         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
896                 return;
897
898         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
899                                          &irq_flags);
900         if (!event)
901                 return;
902         entry   = ring_buffer_event_data(event);
903         tracing_generic_entry_update(&entry->ent, flags, pc);
904         entry->ent.type         = TRACE_USER_STACK;
905
906         memset(&entry->caller, 0, sizeof(entry->caller));
907
908         trace.nr_entries        = 0;
909         trace.max_entries       = FTRACE_STACK_ENTRIES;
910         trace.skip              = 0;
911         trace.entries           = entry->caller;
912
913         save_stack_trace_user(&trace);
914         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
915 #endif
916 }
917
918 void __trace_userstack(struct trace_array *tr,
919                    struct trace_array_cpu *data,
920                    unsigned long flags)
921 {
922         ftrace_trace_userstack(tr, data, flags, preempt_count());
923 }
924
925 static void
926 ftrace_trace_special(void *__tr, void *__data,
927                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
928                      int pc)
929 {
930         struct ring_buffer_event *event;
931         struct trace_array_cpu *data = __data;
932         struct trace_array *tr = __tr;
933         struct special_entry *entry;
934         unsigned long irq_flags;
935
936         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
937                                          &irq_flags);
938         if (!event)
939                 return;
940         entry   = ring_buffer_event_data(event);
941         tracing_generic_entry_update(&entry->ent, 0, pc);
942         entry->ent.type                 = TRACE_SPECIAL;
943         entry->arg1                     = arg1;
944         entry->arg2                     = arg2;
945         entry->arg3                     = arg3;
946         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
947         ftrace_trace_stack(tr, data, irq_flags, 4, pc);
948         ftrace_trace_userstack(tr, data, irq_flags, pc);
949
950         trace_wake_up();
951 }
952
953 void
954 __trace_special(void *__tr, void *__data,
955                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
956 {
957         ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count());
958 }
959
960 void
961 tracing_sched_switch_trace(struct trace_array *tr,
962                            struct trace_array_cpu *data,
963                            struct task_struct *prev,
964                            struct task_struct *next,
965                            unsigned long flags, int pc)
966 {
967         struct ring_buffer_event *event;
968         struct ctx_switch_entry *entry;
969         unsigned long irq_flags;
970
971         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
972                                            &irq_flags);
973         if (!event)
974                 return;
975         entry   = ring_buffer_event_data(event);
976         tracing_generic_entry_update(&entry->ent, flags, pc);
977         entry->ent.type                 = TRACE_CTX;
978         entry->prev_pid                 = prev->pid;
979         entry->prev_prio                = prev->prio;
980         entry->prev_state               = prev->state;
981         entry->next_pid                 = next->pid;
982         entry->next_prio                = next->prio;
983         entry->next_state               = next->state;
984         entry->next_cpu = task_cpu(next);
985         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
986         ftrace_trace_stack(tr, data, flags, 5, pc);
987         ftrace_trace_userstack(tr, data, flags, pc);
988 }
989
990 void
991 tracing_sched_wakeup_trace(struct trace_array *tr,
992                            struct trace_array_cpu *data,
993                            struct task_struct *wakee,
994                            struct task_struct *curr,
995                            unsigned long flags, int pc)
996 {
997         struct ring_buffer_event *event;
998         struct ctx_switch_entry *entry;
999         unsigned long irq_flags;
1000
1001         event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
1002                                            &irq_flags);
1003         if (!event)
1004                 return;
1005         entry   = ring_buffer_event_data(event);
1006         tracing_generic_entry_update(&entry->ent, flags, pc);
1007         entry->ent.type                 = TRACE_WAKE;
1008         entry->prev_pid                 = curr->pid;
1009         entry->prev_prio                = curr->prio;
1010         entry->prev_state               = curr->state;
1011         entry->next_pid                 = wakee->pid;
1012         entry->next_prio                = wakee->prio;
1013         entry->next_state               = wakee->state;
1014         entry->next_cpu                 = task_cpu(wakee);
1015         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1016         ftrace_trace_stack(tr, data, flags, 6, pc);
1017         ftrace_trace_userstack(tr, data, flags, pc);
1018
1019         trace_wake_up();
1020 }
1021
1022 void
1023 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1024 {
1025         struct trace_array *tr = &global_trace;
1026         struct trace_array_cpu *data;
1027         unsigned long flags;
1028         int cpu;
1029         int pc;
1030
1031         if (tracing_disabled)
1032                 return;
1033
1034         pc = preempt_count();
1035         local_irq_save(flags);
1036         cpu = raw_smp_processor_id();
1037         data = tr->data[cpu];
1038
1039         if (likely(atomic_inc_return(&data->disabled) == 1))
1040                 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc);
1041
1042         atomic_dec(&data->disabled);
1043         local_irq_restore(flags);
1044 }
1045
1046 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1047 int trace_graph_entry(struct ftrace_graph_ent *trace)
1048 {
1049         struct trace_array *tr = &global_trace;
1050         struct trace_array_cpu *data;
1051         unsigned long flags;
1052         long disabled;
1053         int cpu;
1054         int pc;
1055
1056         if (!ftrace_trace_task(current))
1057                 return 0;
1058
1059         if (!ftrace_graph_addr(trace->func))
1060                 return 0;
1061
1062         local_irq_save(flags);
1063         cpu = raw_smp_processor_id();
1064         data = tr->data[cpu];
1065         disabled = atomic_inc_return(&data->disabled);
1066         if (likely(disabled == 1)) {
1067                 pc = preempt_count();
1068                 __trace_graph_entry(tr, data, trace, flags, pc);
1069         }
1070         /* Only do the atomic if it is not already set */
1071         if (!test_tsk_trace_graph(current))
1072                 set_tsk_trace_graph(current);
1073         atomic_dec(&data->disabled);
1074         local_irq_restore(flags);
1075
1076         return 1;
1077 }
1078
1079 void trace_graph_return(struct ftrace_graph_ret *trace)
1080 {
1081         struct trace_array *tr = &global_trace;
1082         struct trace_array_cpu *data;
1083         unsigned long flags;
1084         long disabled;
1085         int cpu;
1086         int pc;
1087
1088         local_irq_save(flags);
1089         cpu = raw_smp_processor_id();
1090         data = tr->data[cpu];
1091         disabled = atomic_inc_return(&data->disabled);
1092         if (likely(disabled == 1)) {
1093                 pc = preempt_count();
1094                 __trace_graph_return(tr, data, trace, flags, pc);
1095         }
1096         if (!trace->depth)
1097                 clear_tsk_trace_graph(current);
1098         atomic_dec(&data->disabled);
1099         local_irq_restore(flags);
1100 }
1101 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1102
1103 enum trace_file_type {
1104         TRACE_FILE_LAT_FMT      = 1,
1105         TRACE_FILE_ANNOTATE     = 2,
1106 };
1107
1108 static void trace_iterator_increment(struct trace_iterator *iter)
1109 {
1110         /* Don't allow ftrace to trace into the ring buffers */
1111         ftrace_disable_cpu();
1112
1113         iter->idx++;
1114         if (iter->buffer_iter[iter->cpu])
1115                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1116
1117         ftrace_enable_cpu();
1118 }
1119
1120 static struct trace_entry *
1121 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1122 {
1123         struct ring_buffer_event *event;
1124         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1125
1126         /* Don't allow ftrace to trace into the ring buffers */
1127         ftrace_disable_cpu();
1128
1129         if (buf_iter)
1130                 event = ring_buffer_iter_peek(buf_iter, ts);
1131         else
1132                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1133
1134         ftrace_enable_cpu();
1135
1136         return event ? ring_buffer_event_data(event) : NULL;
1137 }
1138
1139 static struct trace_entry *
1140 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1141 {
1142         struct ring_buffer *buffer = iter->tr->buffer;
1143         struct trace_entry *ent, *next = NULL;
1144         u64 next_ts = 0, ts;
1145         int next_cpu = -1;
1146         int cpu;
1147
1148         for_each_tracing_cpu(cpu) {
1149
1150                 if (ring_buffer_empty_cpu(buffer, cpu))
1151                         continue;
1152
1153                 ent = peek_next_entry(iter, cpu, &ts);
1154
1155                 /*
1156                  * Pick the entry with the smallest timestamp:
1157                  */
1158                 if (ent && (!next || ts < next_ts)) {
1159                         next = ent;
1160                         next_cpu = cpu;
1161                         next_ts = ts;
1162                 }
1163         }
1164
1165         if (ent_cpu)
1166                 *ent_cpu = next_cpu;
1167
1168         if (ent_ts)
1169                 *ent_ts = next_ts;
1170
1171         return next;
1172 }
1173
1174 /* Find the next real entry, without updating the iterator itself */
1175 static struct trace_entry *
1176 find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1177 {
1178         return __find_next_entry(iter, ent_cpu, ent_ts);
1179 }
1180
1181 /* Find the next real entry, and increment the iterator to the next entry */
1182 static void *find_next_entry_inc(struct trace_iterator *iter)
1183 {
1184         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1185
1186         if (iter->ent)
1187                 trace_iterator_increment(iter);
1188
1189         return iter->ent ? iter : NULL;
1190 }
1191
1192 static void trace_consume(struct trace_iterator *iter)
1193 {
1194         /* Don't allow ftrace to trace into the ring buffers */
1195         ftrace_disable_cpu();
1196         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1197         ftrace_enable_cpu();
1198 }
1199
1200 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1201 {
1202         struct trace_iterator *iter = m->private;
1203         int i = (int)*pos;
1204         void *ent;
1205
1206         (*pos)++;
1207
1208         /* can't go backwards */
1209         if (iter->idx > i)
1210                 return NULL;
1211
1212         if (iter->idx < 0)
1213                 ent = find_next_entry_inc(iter);
1214         else
1215                 ent = iter;
1216
1217         while (ent && iter->idx < i)
1218                 ent = find_next_entry_inc(iter);
1219
1220         iter->pos = *pos;
1221
1222         return ent;
1223 }
1224
1225 static void *s_start(struct seq_file *m, loff_t *pos)
1226 {
1227         struct trace_iterator *iter = m->private;
1228         void *p = NULL;
1229         loff_t l = 0;
1230         int cpu;
1231
1232         mutex_lock(&trace_types_lock);
1233
1234         if (!current_trace || current_trace != iter->trace) {
1235                 mutex_unlock(&trace_types_lock);
1236                 return NULL;
1237         }
1238
1239         atomic_inc(&trace_record_cmdline_disabled);
1240
1241         if (*pos != iter->pos) {
1242                 iter->ent = NULL;
1243                 iter->cpu = 0;
1244                 iter->idx = -1;
1245
1246                 ftrace_disable_cpu();
1247
1248                 for_each_tracing_cpu(cpu) {
1249                         ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1250                 }
1251
1252                 ftrace_enable_cpu();
1253
1254                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1255                         ;
1256
1257         } else {
1258                 l = *pos - 1;
1259                 p = s_next(m, p, &l);
1260         }
1261
1262         return p;
1263 }
1264
1265 static void s_stop(struct seq_file *m, void *p)
1266 {
1267         atomic_dec(&trace_record_cmdline_disabled);
1268         mutex_unlock(&trace_types_lock);
1269 }
1270
1271 static void print_lat_help_header(struct seq_file *m)
1272 {
1273         seq_puts(m, "#                  _------=> CPU#            \n");
1274         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1275         seq_puts(m, "#                | / _----=> need-resched    \n");
1276         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1277         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1278         seq_puts(m, "#                |||| /                      \n");
1279         seq_puts(m, "#                |||||     delay             \n");
1280         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1281         seq_puts(m, "#     \\   /      |||||   \\   |   /           \n");
1282 }
1283
1284 static void print_func_help_header(struct seq_file *m)
1285 {
1286         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1287         seq_puts(m, "#              | |       |          |         |\n");
1288 }
1289
1290
1291 static void
1292 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1293 {
1294         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1295         struct trace_array *tr = iter->tr;
1296         struct trace_array_cpu *data = tr->data[tr->cpu];
1297         struct tracer *type = current_trace;
1298         unsigned long total;
1299         unsigned long entries;
1300         const char *name = "preemption";
1301
1302         if (type)
1303                 name = type->name;
1304
1305         entries = ring_buffer_entries(iter->tr->buffer);
1306         total = entries +
1307                 ring_buffer_overruns(iter->tr->buffer);
1308
1309         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1310                    name, UTS_RELEASE);
1311         seq_puts(m, "-----------------------------------"
1312                  "---------------------------------\n");
1313         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1314                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1315                    nsecs_to_usecs(data->saved_latency),
1316                    entries,
1317                    total,
1318                    tr->cpu,
1319 #if defined(CONFIG_PREEMPT_NONE)
1320                    "server",
1321 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1322                    "desktop",
1323 #elif defined(CONFIG_PREEMPT)
1324                    "preempt",
1325 #else
1326                    "unknown",
1327 #endif
1328                    /* These are reserved for later use */
1329                    0, 0, 0, 0);
1330 #ifdef CONFIG_SMP
1331         seq_printf(m, " #P:%d)\n", num_online_cpus());
1332 #else
1333         seq_puts(m, ")\n");
1334 #endif
1335         seq_puts(m, "    -----------------\n");
1336         seq_printf(m, "    | task: %.16s-%d "
1337                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1338                    data->comm, data->pid, data->uid, data->nice,
1339                    data->policy, data->rt_priority);
1340         seq_puts(m, "    -----------------\n");
1341
1342         if (data->critical_start) {
1343                 seq_puts(m, " => started at: ");
1344                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1345                 trace_print_seq(m, &iter->seq);
1346                 seq_puts(m, "\n => ended at:   ");
1347                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1348                 trace_print_seq(m, &iter->seq);
1349                 seq_puts(m, "\n");
1350         }
1351
1352         seq_puts(m, "\n");
1353 }
1354
1355 static void
1356 lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1357 {
1358         int hardirq, softirq;
1359         char *comm;
1360
1361         comm = trace_find_cmdline(entry->pid);
1362
1363         trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1364         trace_seq_printf(s, "%3d", cpu);
1365         trace_seq_printf(s, "%c%c",
1366                         (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1367                          (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1368                         ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1369
1370         hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1371         softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1372         if (hardirq && softirq) {
1373                 trace_seq_putc(s, 'H');
1374         } else {
1375                 if (hardirq) {
1376                         trace_seq_putc(s, 'h');
1377                 } else {
1378                         if (softirq)
1379                                 trace_seq_putc(s, 's');
1380                         else
1381                                 trace_seq_putc(s, '.');
1382                 }
1383         }
1384
1385         if (entry->preempt_count)
1386                 trace_seq_printf(s, "%x", entry->preempt_count);
1387         else
1388                 trace_seq_puts(s, ".");
1389 }
1390
1391 unsigned long preempt_mark_thresh = 100;
1392
1393 static void
1394 lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1395                     unsigned long rel_usecs)
1396 {
1397         trace_seq_printf(s, " %4lldus", abs_usecs);
1398         if (rel_usecs > preempt_mark_thresh)
1399                 trace_seq_puts(s, "!: ");
1400         else if (rel_usecs > 1)
1401                 trace_seq_puts(s, "+: ");
1402         else
1403                 trace_seq_puts(s, " : ");
1404 }
1405
1406 static void test_cpu_buff_start(struct trace_iterator *iter)
1407 {
1408         struct trace_seq *s = &iter->seq;
1409
1410         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1411                 return;
1412
1413         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1414                 return;
1415
1416         if (cpumask_test_cpu(iter->cpu, iter->started))
1417                 return;
1418
1419         cpumask_set_cpu(iter->cpu, iter->started);
1420         trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1421 }
1422
1423 static enum print_line_t
1424 print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1425 {
1426         struct trace_seq *s = &iter->seq;
1427         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1428         struct trace_entry *next_entry;
1429         struct trace_event *event;
1430         unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1431         struct trace_entry *entry = iter->ent;
1432         unsigned long abs_usecs;
1433         unsigned long rel_usecs;
1434         u64 next_ts;
1435         char *comm;
1436         int ret;
1437
1438         test_cpu_buff_start(iter);
1439
1440         next_entry = find_next_entry(iter, NULL, &next_ts);
1441         if (!next_entry)
1442                 next_ts = iter->ts;
1443         rel_usecs = ns2usecs(next_ts - iter->ts);
1444         abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1445
1446         if (verbose) {
1447                 comm = trace_find_cmdline(entry->pid);
1448                 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1449                                  " %ld.%03ldms (+%ld.%03ldms): ",
1450                                  comm,
1451                                  entry->pid, cpu, entry->flags,
1452                                  entry->preempt_count, trace_idx,
1453                                  ns2usecs(iter->ts),
1454                                  abs_usecs/1000,
1455                                  abs_usecs % 1000, rel_usecs/1000,
1456                                  rel_usecs % 1000);
1457         } else {
1458                 lat_print_generic(s, entry, cpu);
1459                 lat_print_timestamp(s, abs_usecs, rel_usecs);
1460         }
1461
1462         event = ftrace_find_event(entry->type);
1463         if (event && event->latency_trace) {
1464                 ret = event->latency_trace(s, entry, sym_flags);
1465                 if (ret)
1466                         return ret;
1467                 return TRACE_TYPE_HANDLED;
1468         }
1469
1470         trace_seq_printf(s, "Unknown type %d\n", entry->type);
1471         return TRACE_TYPE_HANDLED;
1472 }
1473
1474 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1475 {
1476         struct trace_seq *s = &iter->seq;
1477         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1478         struct trace_entry *entry;
1479         struct trace_event *event;
1480         unsigned long usec_rem;
1481         unsigned long long t;
1482         unsigned long secs;
1483         char *comm;
1484         int ret;
1485
1486         entry = iter->ent;
1487
1488         test_cpu_buff_start(iter);
1489
1490         comm = trace_find_cmdline(iter->ent->pid);
1491
1492         t = ns2usecs(iter->ts);
1493         usec_rem = do_div(t, 1000000ULL);
1494         secs = (unsigned long)t;
1495
1496         ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1497         if (!ret)
1498                 return TRACE_TYPE_PARTIAL_LINE;
1499         ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1500         if (!ret)
1501                 return TRACE_TYPE_PARTIAL_LINE;
1502         ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1503         if (!ret)
1504                 return TRACE_TYPE_PARTIAL_LINE;
1505
1506         event = ftrace_find_event(entry->type);
1507         if (event && event->trace) {
1508                 ret = event->trace(s, entry, sym_flags);
1509                 if (ret)
1510                         return ret;
1511                 return TRACE_TYPE_HANDLED;
1512         }
1513         ret = trace_seq_printf(s, "Unknown type %d\n", entry->type);
1514         if (!ret)
1515                 return TRACE_TYPE_PARTIAL_LINE;
1516
1517         return TRACE_TYPE_HANDLED;
1518 }
1519
1520 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1521 {
1522         struct trace_seq *s = &iter->seq;
1523         struct trace_entry *entry;
1524         struct trace_event *event;
1525         int ret;
1526
1527         entry = iter->ent;
1528
1529         ret = trace_seq_printf(s, "%d %d %llu ",
1530                 entry->pid, iter->cpu, iter->ts);
1531         if (!ret)
1532                 return TRACE_TYPE_PARTIAL_LINE;
1533
1534         event = ftrace_find_event(entry->type);
1535         if (event && event->raw) {
1536                 ret = event->raw(s, entry, 0);
1537                 if (ret)
1538                         return ret;
1539                 return TRACE_TYPE_HANDLED;
1540         }
1541         ret = trace_seq_printf(s, "%d ?\n", entry->type);
1542         if (!ret)
1543                 return TRACE_TYPE_PARTIAL_LINE;
1544
1545         return TRACE_TYPE_HANDLED;
1546 }
1547
1548 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1549 {
1550         struct trace_seq *s = &iter->seq;
1551         unsigned char newline = '\n';
1552         struct trace_entry *entry;
1553         struct trace_event *event;
1554
1555         entry = iter->ent;
1556
1557         SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1558         SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1559         SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1560
1561         event = ftrace_find_event(entry->type);
1562         if (event && event->hex)
1563                 event->hex(s, entry, 0);
1564
1565         SEQ_PUT_FIELD_RET(s, newline);
1566
1567         return TRACE_TYPE_HANDLED;
1568 }
1569
1570 static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
1571 {
1572         struct trace_seq *s = &iter->seq;
1573         struct trace_entry *entry = iter->ent;
1574         struct print_entry *field;
1575         int ret;
1576
1577         trace_assign_type(field, entry);
1578
1579         ret = trace_seq_printf(s, field->buf);
1580         if (!ret)
1581                 return TRACE_TYPE_PARTIAL_LINE;
1582
1583         return TRACE_TYPE_HANDLED;
1584 }
1585
1586 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1587 {
1588         struct trace_seq *s = &iter->seq;
1589         struct trace_entry *entry;
1590         struct trace_event *event;
1591
1592         entry = iter->ent;
1593
1594         SEQ_PUT_FIELD_RET(s, entry->pid);
1595         SEQ_PUT_FIELD_RET(s, entry->cpu);
1596         SEQ_PUT_FIELD_RET(s, iter->ts);
1597
1598         event = ftrace_find_event(entry->type);
1599         if (event && event->binary)
1600                 event->binary(s, entry, 0);
1601
1602         return TRACE_TYPE_HANDLED;
1603 }
1604
1605 static int trace_empty(struct trace_iterator *iter)
1606 {
1607         int cpu;
1608
1609         for_each_tracing_cpu(cpu) {
1610                 if (iter->buffer_iter[cpu]) {
1611                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1612                                 return 0;
1613                 } else {
1614                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1615                                 return 0;
1616                 }
1617         }
1618
1619         return 1;
1620 }
1621
1622 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1623 {
1624         enum print_line_t ret;
1625
1626         if (iter->trace && iter->trace->print_line) {
1627                 ret = iter->trace->print_line(iter);
1628                 if (ret != TRACE_TYPE_UNHANDLED)
1629                         return ret;
1630         }
1631
1632         if (iter->ent->type == TRACE_PRINT &&
1633                         trace_flags & TRACE_ITER_PRINTK &&
1634                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1635                 return print_printk_msg_only(iter);
1636
1637         if (trace_flags & TRACE_ITER_BIN)
1638                 return print_bin_fmt(iter);
1639
1640         if (trace_flags & TRACE_ITER_HEX)
1641                 return print_hex_fmt(iter);
1642
1643         if (trace_flags & TRACE_ITER_RAW)
1644                 return print_raw_fmt(iter);
1645
1646         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1647                 return print_lat_fmt(iter, iter->idx, iter->cpu);
1648
1649         return print_trace_fmt(iter);
1650 }
1651
1652 static int s_show(struct seq_file *m, void *v)
1653 {
1654         struct trace_iterator *iter = v;
1655
1656         if (iter->ent == NULL) {
1657                 if (iter->tr) {
1658                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1659                         seq_puts(m, "#\n");
1660                 }
1661                 if (iter->trace && iter->trace->print_header)
1662                         iter->trace->print_header(m);
1663                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1664                         /* print nothing if the buffers are empty */
1665                         if (trace_empty(iter))
1666                                 return 0;
1667                         print_trace_header(m, iter);
1668                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1669                                 print_lat_help_header(m);
1670                 } else {
1671                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1672                                 print_func_help_header(m);
1673                 }
1674         } else {
1675                 print_trace_line(iter);
1676                 trace_print_seq(m, &iter->seq);
1677         }
1678
1679         return 0;
1680 }
1681
1682 static struct seq_operations tracer_seq_ops = {
1683         .start          = s_start,
1684         .next           = s_next,
1685         .stop           = s_stop,
1686         .show           = s_show,
1687 };
1688
1689 static struct trace_iterator *
1690 __tracing_open(struct inode *inode, struct file *file, int *ret)
1691 {
1692         struct trace_iterator *iter;
1693         struct seq_file *m;
1694         int cpu;
1695
1696         if (tracing_disabled) {
1697                 *ret = -ENODEV;
1698                 return NULL;
1699         }
1700
1701         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1702         if (!iter) {
1703                 *ret = -ENOMEM;
1704                 goto out;
1705         }
1706
1707         mutex_lock(&trace_types_lock);
1708         if (current_trace && current_trace->print_max)
1709                 iter->tr = &max_tr;
1710         else
1711                 iter->tr = inode->i_private;
1712         iter->trace = current_trace;
1713         iter->pos = -1;
1714
1715         /* Notify the tracer early; before we stop tracing. */
1716         if (iter->trace && iter->trace->open)
1717                 iter->trace->open(iter);
1718
1719         /* Annotate start of buffers if we had overruns */
1720         if (ring_buffer_overruns(iter->tr->buffer))
1721                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1722
1723
1724         for_each_tracing_cpu(cpu) {
1725
1726                 iter->buffer_iter[cpu] =
1727                         ring_buffer_read_start(iter->tr->buffer, cpu);
1728
1729                 if (!iter->buffer_iter[cpu])
1730                         goto fail_buffer;
1731         }
1732
1733         /* TODO stop tracer */
1734         *ret = seq_open(file, &tracer_seq_ops);
1735         if (*ret)
1736                 goto fail_buffer;
1737
1738         m = file->private_data;
1739         m->private = iter;
1740
1741         /* stop the trace while dumping */
1742         tracing_stop();
1743
1744         mutex_unlock(&trace_types_lock);
1745
1746  out:
1747         return iter;
1748
1749  fail_buffer:
1750         for_each_tracing_cpu(cpu) {
1751                 if (iter->buffer_iter[cpu])
1752                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
1753         }
1754         mutex_unlock(&trace_types_lock);
1755         kfree(iter);
1756
1757         return ERR_PTR(-ENOMEM);
1758 }
1759
1760 int tracing_open_generic(struct inode *inode, struct file *filp)
1761 {
1762         if (tracing_disabled)
1763                 return -ENODEV;
1764
1765         filp->private_data = inode->i_private;
1766         return 0;
1767 }
1768
1769 int tracing_release(struct inode *inode, struct file *file)
1770 {
1771         struct seq_file *m = (struct seq_file *)file->private_data;
1772         struct trace_iterator *iter = m->private;
1773         int cpu;
1774
1775         mutex_lock(&trace_types_lock);
1776         for_each_tracing_cpu(cpu) {
1777                 if (iter->buffer_iter[cpu])
1778                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
1779         }
1780
1781         if (iter->trace && iter->trace->close)
1782                 iter->trace->close(iter);
1783
1784         /* reenable tracing if it was previously enabled */
1785         tracing_start();
1786         mutex_unlock(&trace_types_lock);
1787
1788         seq_release(inode, file);
1789         kfree(iter);
1790         return 0;
1791 }
1792
1793 static int tracing_open(struct inode *inode, struct file *file)
1794 {
1795         int ret;
1796
1797         __tracing_open(inode, file, &ret);
1798
1799         return ret;
1800 }
1801
1802 static int tracing_lt_open(struct inode *inode, struct file *file)
1803 {
1804         struct trace_iterator *iter;
1805         int ret;
1806
1807         iter = __tracing_open(inode, file, &ret);
1808
1809         if (!ret)
1810                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1811
1812         return ret;
1813 }
1814
1815
1816 static void *
1817 t_next(struct seq_file *m, void *v, loff_t *pos)
1818 {
1819         struct tracer *t = m->private;
1820
1821         (*pos)++;
1822
1823         if (t)
1824                 t = t->next;
1825
1826         m->private = t;
1827
1828         return t;
1829 }
1830
1831 static void *t_start(struct seq_file *m, loff_t *pos)
1832 {
1833         struct tracer *t = m->private;
1834         loff_t l = 0;
1835
1836         mutex_lock(&trace_types_lock);
1837         for (; t && l < *pos; t = t_next(m, t, &l))
1838                 ;
1839
1840         return t;
1841 }
1842
1843 static void t_stop(struct seq_file *m, void *p)
1844 {
1845         mutex_unlock(&trace_types_lock);
1846 }
1847
1848 static int t_show(struct seq_file *m, void *v)
1849 {
1850         struct tracer *t = v;
1851
1852         if (!t)
1853                 return 0;
1854
1855         seq_printf(m, "%s", t->name);
1856         if (t->next)
1857                 seq_putc(m, ' ');
1858         else
1859                 seq_putc(m, '\n');
1860
1861         return 0;
1862 }
1863
1864 static struct seq_operations show_traces_seq_ops = {
1865         .start          = t_start,
1866         .next           = t_next,
1867         .stop           = t_stop,
1868         .show           = t_show,
1869 };
1870
1871 static int show_traces_open(struct inode *inode, struct file *file)
1872 {
1873         int ret;
1874
1875         if (tracing_disabled)
1876                 return -ENODEV;
1877
1878         ret = seq_open(file, &show_traces_seq_ops);
1879         if (!ret) {
1880                 struct seq_file *m = file->private_data;
1881                 m->private = trace_types;
1882         }
1883
1884         return ret;
1885 }
1886
1887 static struct file_operations tracing_fops = {
1888         .open           = tracing_open,
1889         .read           = seq_read,
1890         .llseek         = seq_lseek,
1891         .release        = tracing_release,
1892 };
1893
1894 static struct file_operations tracing_lt_fops = {
1895         .open           = tracing_lt_open,
1896         .read           = seq_read,
1897         .llseek         = seq_lseek,
1898         .release        = tracing_release,
1899 };
1900
1901 static struct file_operations show_traces_fops = {
1902         .open           = show_traces_open,
1903         .read           = seq_read,
1904         .release        = seq_release,
1905 };
1906
1907 /*
1908  * Only trace on a CPU if the bitmask is set:
1909  */
1910 static cpumask_var_t tracing_cpumask;
1911
1912 /*
1913  * The tracer itself will not take this lock, but still we want
1914  * to provide a consistent cpumask to user-space:
1915  */
1916 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1917
1918 /*
1919  * Temporary storage for the character representation of the
1920  * CPU bitmask (and one more byte for the newline):
1921  */
1922 static char mask_str[NR_CPUS + 1];
1923
1924 static ssize_t
1925 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1926                      size_t count, loff_t *ppos)
1927 {
1928         int len;
1929
1930         mutex_lock(&tracing_cpumask_update_lock);
1931
1932         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1933         if (count - len < 2) {
1934                 count = -EINVAL;
1935                 goto out_err;
1936         }
1937         len += sprintf(mask_str + len, "\n");
1938         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1939
1940 out_err:
1941         mutex_unlock(&tracing_cpumask_update_lock);
1942
1943         return count;
1944 }
1945
1946 static ssize_t
1947 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1948                       size_t count, loff_t *ppos)
1949 {
1950         int err, cpu;
1951         cpumask_var_t tracing_cpumask_new;
1952
1953         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
1954                 return -ENOMEM;
1955
1956         mutex_lock(&tracing_cpumask_update_lock);
1957         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
1958         if (err)
1959                 goto err_unlock;
1960
1961         local_irq_disable();
1962         __raw_spin_lock(&ftrace_max_lock);
1963         for_each_tracing_cpu(cpu) {
1964                 /*
1965                  * Increase/decrease the disabled counter if we are
1966                  * about to flip a bit in the cpumask:
1967                  */
1968                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
1969                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
1970                         atomic_inc(&global_trace.data[cpu]->disabled);
1971                 }
1972                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
1973                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
1974                         atomic_dec(&global_trace.data[cpu]->disabled);
1975                 }
1976         }
1977         __raw_spin_unlock(&ftrace_max_lock);
1978         local_irq_enable();
1979
1980         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
1981
1982         mutex_unlock(&tracing_cpumask_update_lock);
1983         free_cpumask_var(tracing_cpumask_new);
1984
1985         return count;
1986
1987 err_unlock:
1988         mutex_unlock(&tracing_cpumask_update_lock);
1989         free_cpumask_var(tracing_cpumask);
1990
1991         return err;
1992 }
1993
1994 static struct file_operations tracing_cpumask_fops = {
1995         .open           = tracing_open_generic,
1996         .read           = tracing_cpumask_read,
1997         .write          = tracing_cpumask_write,
1998 };
1999
2000 static ssize_t
2001 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2002                        size_t cnt, loff_t *ppos)
2003 {
2004         int i;
2005         char *buf;
2006         int r = 0;
2007         int len = 0;
2008         u32 tracer_flags = current_trace->flags->val;
2009         struct tracer_opt *trace_opts = current_trace->flags->opts;
2010
2011
2012         /* calulate max size */
2013         for (i = 0; trace_options[i]; i++) {
2014                 len += strlen(trace_options[i]);
2015                 len += 3; /* "no" and space */
2016         }
2017
2018         /*
2019          * Increase the size with names of options specific
2020          * of the current tracer.
2021          */
2022         for (i = 0; trace_opts[i].name; i++) {
2023                 len += strlen(trace_opts[i].name);
2024                 len += 3; /* "no" and space */
2025         }
2026
2027         /* +2 for \n and \0 */
2028         buf = kmalloc(len + 2, GFP_KERNEL);
2029         if (!buf)
2030                 return -ENOMEM;
2031
2032         for (i = 0; trace_options[i]; i++) {
2033                 if (trace_flags & (1 << i))
2034                         r += sprintf(buf + r, "%s ", trace_options[i]);
2035                 else
2036                         r += sprintf(buf + r, "no%s ", trace_options[i]);
2037         }
2038
2039         for (i = 0; trace_opts[i].name; i++) {
2040                 if (tracer_flags & trace_opts[i].bit)
2041                         r += sprintf(buf + r, "%s ",
2042                                 trace_opts[i].name);
2043                 else
2044                         r += sprintf(buf + r, "no%s ",
2045                                 trace_opts[i].name);
2046         }
2047
2048         r += sprintf(buf + r, "\n");
2049         WARN_ON(r >= len + 2);
2050
2051         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2052
2053         kfree(buf);
2054
2055         return r;
2056 }
2057
2058 /* Try to assign a tracer specific option */
2059 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2060 {
2061         struct tracer_flags *trace_flags = trace->flags;
2062         struct tracer_opt *opts = NULL;
2063         int ret = 0, i = 0;
2064         int len;
2065
2066         for (i = 0; trace_flags->opts[i].name; i++) {
2067                 opts = &trace_flags->opts[i];
2068                 len = strlen(opts->name);
2069
2070                 if (strncmp(cmp, opts->name, len) == 0) {
2071                         ret = trace->set_flag(trace_flags->val,
2072                                 opts->bit, !neg);
2073                         break;
2074                 }
2075         }
2076         /* Not found */
2077         if (!trace_flags->opts[i].name)
2078                 return -EINVAL;
2079
2080         /* Refused to handle */
2081         if (ret)
2082                 return ret;
2083
2084         if (neg)
2085                 trace_flags->val &= ~opts->bit;
2086         else
2087                 trace_flags->val |= opts->bit;
2088
2089         return 0;
2090 }
2091
2092 static ssize_t
2093 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2094                         size_t cnt, loff_t *ppos)
2095 {
2096         char buf[64];
2097         char *cmp = buf;
2098         int neg = 0;
2099         int ret;
2100         int i;
2101
2102         if (cnt >= sizeof(buf))
2103                 return -EINVAL;
2104
2105         if (copy_from_user(&buf, ubuf, cnt))
2106                 return -EFAULT;
2107
2108         buf[cnt] = 0;
2109
2110         if (strncmp(buf, "no", 2) == 0) {
2111                 neg = 1;
2112                 cmp += 2;
2113         }
2114
2115         for (i = 0; trace_options[i]; i++) {
2116                 int len = strlen(trace_options[i]);
2117
2118                 if (strncmp(cmp, trace_options[i], len) == 0) {
2119                         if (neg)
2120                                 trace_flags &= ~(1 << i);
2121                         else
2122                                 trace_flags |= (1 << i);
2123                         break;
2124                 }
2125         }
2126
2127         /* If no option could be set, test the specific tracer options */
2128         if (!trace_options[i]) {
2129                 ret = set_tracer_option(current_trace, cmp, neg);
2130                 if (ret)
2131                         return ret;
2132         }
2133
2134         filp->f_pos += cnt;
2135
2136         return cnt;
2137 }
2138
2139 static struct file_operations tracing_iter_fops = {
2140         .open           = tracing_open_generic,
2141         .read           = tracing_trace_options_read,
2142         .write          = tracing_trace_options_write,
2143 };
2144
2145 static const char readme_msg[] =
2146         "tracing mini-HOWTO:\n\n"
2147         "# mkdir /debug\n"
2148         "# mount -t debugfs nodev /debug\n\n"
2149         "# cat /debug/tracing/available_tracers\n"
2150         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2151         "# cat /debug/tracing/current_tracer\n"
2152         "none\n"
2153         "# echo sched_switch > /debug/tracing/current_tracer\n"
2154         "# cat /debug/tracing/current_tracer\n"
2155         "sched_switch\n"
2156         "# cat /debug/tracing/trace_options\n"
2157         "noprint-parent nosym-offset nosym-addr noverbose\n"
2158         "# echo print-parent > /debug/tracing/trace_options\n"
2159         "# echo 1 > /debug/tracing/tracing_enabled\n"
2160         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2161         "echo 0 > /debug/tracing/tracing_enabled\n"
2162 ;
2163
2164 static ssize_t
2165 tracing_readme_read(struct file *filp, char __user *ubuf,
2166                        size_t cnt, loff_t *ppos)
2167 {
2168         return simple_read_from_buffer(ubuf, cnt, ppos,
2169                                         readme_msg, strlen(readme_msg));
2170 }
2171
2172 static struct file_operations tracing_readme_fops = {
2173         .open           = tracing_open_generic,
2174         .read           = tracing_readme_read,
2175 };
2176
2177 static ssize_t
2178 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2179                   size_t cnt, loff_t *ppos)
2180 {
2181         char buf[64];
2182         int r;
2183
2184         r = sprintf(buf, "%u\n", tracer_enabled);
2185         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2186 }
2187
2188 static ssize_t
2189 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2190                    size_t cnt, loff_t *ppos)
2191 {
2192         struct trace_array *tr = filp->private_data;
2193         char buf[64];
2194         long val;
2195         int ret;
2196
2197         if (cnt >= sizeof(buf))
2198                 return -EINVAL;
2199
2200         if (copy_from_user(&buf, ubuf, cnt))
2201                 return -EFAULT;
2202
2203         buf[cnt] = 0;
2204
2205         ret = strict_strtoul(buf, 10, &val);
2206         if (ret < 0)
2207                 return ret;
2208
2209         val = !!val;
2210
2211         mutex_lock(&trace_types_lock);
2212         if (tracer_enabled ^ val) {
2213                 if (val) {
2214                         tracer_enabled = 1;
2215                         if (current_trace->start)
2216                                 current_trace->start(tr);
2217                         tracing_start();
2218                 } else {
2219                         tracer_enabled = 0;
2220                         tracing_stop();
2221                         if (current_trace->stop)
2222                                 current_trace->stop(tr);
2223                 }
2224         }
2225         mutex_unlock(&trace_types_lock);
2226
2227         filp->f_pos += cnt;
2228
2229         return cnt;
2230 }
2231
2232 static ssize_t
2233 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2234                        size_t cnt, loff_t *ppos)
2235 {
2236         char buf[max_tracer_type_len+2];
2237         int r;
2238
2239         mutex_lock(&trace_types_lock);
2240         if (current_trace)
2241                 r = sprintf(buf, "%s\n", current_trace->name);
2242         else
2243                 r = sprintf(buf, "\n");
2244         mutex_unlock(&trace_types_lock);
2245
2246         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2247 }
2248
2249 static int tracing_set_tracer(char *buf)
2250 {
2251         struct trace_array *tr = &global_trace;
2252         struct tracer *t;
2253         int ret = 0;
2254
2255         mutex_lock(&trace_types_lock);
2256         for (t = trace_types; t; t = t->next) {
2257                 if (strcmp(t->name, buf) == 0)
2258                         break;
2259         }
2260         if (!t) {
2261                 ret = -EINVAL;
2262                 goto out;
2263         }
2264         if (t == current_trace)
2265                 goto out;
2266
2267         trace_branch_disable();
2268         if (current_trace && current_trace->reset)
2269                 current_trace->reset(tr);
2270
2271         current_trace = t;
2272         if (t->init) {
2273                 ret = t->init(tr);
2274                 if (ret)
2275                         goto out;
2276         }
2277
2278         trace_branch_enable(tr);
2279  out:
2280         mutex_unlock(&trace_types_lock);
2281
2282         return ret;
2283 }
2284
2285 static ssize_t
2286 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2287                         size_t cnt, loff_t *ppos)
2288 {
2289         char buf[max_tracer_type_len+1];
2290         int i;
2291         size_t ret;
2292         int err;
2293
2294         ret = cnt;
2295
2296         if (cnt > max_tracer_type_len)
2297                 cnt = max_tracer_type_len;
2298
2299         if (copy_from_user(&buf, ubuf, cnt))
2300                 return -EFAULT;
2301
2302         buf[cnt] = 0;
2303
2304         /* strip ending whitespace. */
2305         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2306                 buf[i] = 0;
2307
2308         err = tracing_set_tracer(buf);
2309         if (err)
2310                 return err;
2311
2312         filp->f_pos += ret;
2313
2314         return ret;
2315 }
2316
2317 static ssize_t
2318 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2319                      size_t cnt, loff_t *ppos)
2320 {
2321         unsigned long *ptr = filp->private_data;
2322         char buf[64];
2323         int r;
2324
2325         r = snprintf(buf, sizeof(buf), "%ld\n",
2326                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2327         if (r > sizeof(buf))
2328                 r = sizeof(buf);
2329         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2330 }
2331
2332 static ssize_t
2333 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2334                       size_t cnt, loff_t *ppos)
2335 {
2336         long *ptr = filp->private_data;
2337         char buf[64];
2338         long val;
2339         int ret;
2340
2341         if (cnt >= sizeof(buf))
2342                 return -EINVAL;
2343
2344         if (copy_from_user(&buf, ubuf, cnt))
2345                 return -EFAULT;
2346
2347         buf[cnt] = 0;
2348
2349         ret = strict_strtoul(buf, 10, &val);
2350         if (ret < 0)
2351                 return ret;
2352
2353         *ptr = val * 1000;
2354
2355         return cnt;
2356 }
2357
2358 static atomic_t tracing_reader;
2359
2360 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2361 {
2362         struct trace_iterator *iter;
2363
2364         if (tracing_disabled)
2365                 return -ENODEV;
2366
2367         /* We only allow for reader of the pipe */
2368         if (atomic_inc_return(&tracing_reader) != 1) {
2369                 atomic_dec(&tracing_reader);
2370                 return -EBUSY;
2371         }
2372
2373         /* create a buffer to store the information to pass to userspace */
2374         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2375         if (!iter)
2376                 return -ENOMEM;
2377
2378         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2379                 kfree(iter);
2380                 return -ENOMEM;
2381         }
2382
2383         mutex_lock(&trace_types_lock);
2384
2385         /* trace pipe does not show start of buffer */
2386         cpumask_setall(iter->started);
2387
2388         iter->tr = &global_trace;
2389         iter->trace = current_trace;
2390         filp->private_data = iter;
2391
2392         if (iter->trace->pipe_open)
2393                 iter->trace->pipe_open(iter);
2394         mutex_unlock(&trace_types_lock);
2395
2396         return 0;
2397 }
2398
2399 static int tracing_release_pipe(struct inode *inode, struct file *file)
2400 {
2401         struct trace_iterator *iter = file->private_data;
2402
2403         free_cpumask_var(iter->started);
2404         kfree(iter);
2405         atomic_dec(&tracing_reader);
2406
2407         return 0;
2408 }
2409
2410 static unsigned int
2411 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2412 {
2413         struct trace_iterator *iter = filp->private_data;
2414
2415         if (trace_flags & TRACE_ITER_BLOCK) {
2416                 /*
2417                  * Always select as readable when in blocking mode
2418                  */
2419                 return POLLIN | POLLRDNORM;
2420         } else {
2421                 if (!trace_empty(iter))
2422                         return POLLIN | POLLRDNORM;
2423                 poll_wait(filp, &trace_wait, poll_table);
2424                 if (!trace_empty(iter))
2425                         return POLLIN | POLLRDNORM;
2426
2427                 return 0;
2428         }
2429 }
2430
2431 /*
2432  * Consumer reader.
2433  */
2434 static ssize_t
2435 tracing_read_pipe(struct file *filp, char __user *ubuf,
2436                   size_t cnt, loff_t *ppos)
2437 {
2438         struct trace_iterator *iter = filp->private_data;
2439         ssize_t sret;
2440
2441         /* return any leftover data */
2442         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2443         if (sret != -EBUSY)
2444                 return sret;
2445
2446         trace_seq_reset(&iter->seq);
2447
2448         mutex_lock(&trace_types_lock);
2449         if (iter->trace->read) {
2450                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2451                 if (sret)
2452                         goto out;
2453         }
2454
2455 waitagain:
2456         sret = 0;
2457         while (trace_empty(iter)) {
2458
2459                 if ((filp->f_flags & O_NONBLOCK)) {
2460                         sret = -EAGAIN;
2461                         goto out;
2462                 }
2463
2464                 /*
2465                  * This is a make-shift waitqueue. The reason we don't use
2466                  * an actual wait queue is because:
2467                  *  1) we only ever have one waiter
2468                  *  2) the tracing, traces all functions, we don't want
2469                  *     the overhead of calling wake_up and friends
2470                  *     (and tracing them too)
2471                  *     Anyway, this is really very primitive wakeup.
2472                  */
2473                 set_current_state(TASK_INTERRUPTIBLE);
2474                 iter->tr->waiter = current;
2475
2476                 mutex_unlock(&trace_types_lock);
2477
2478                 /* sleep for 100 msecs, and try again. */
2479                 schedule_timeout(HZ/10);
2480
2481                 mutex_lock(&trace_types_lock);
2482
2483                 iter->tr->waiter = NULL;
2484
2485                 if (signal_pending(current)) {
2486                         sret = -EINTR;
2487                         goto out;
2488                 }
2489
2490                 if (iter->trace != current_trace)
2491                         goto out;
2492
2493                 /*
2494                  * We block until we read something and tracing is disabled.
2495                  * We still block if tracing is disabled, but we have never
2496                  * read anything. This allows a user to cat this file, and
2497                  * then enable tracing. But after we have read something,
2498                  * we give an EOF when tracing is again disabled.
2499                  *
2500                  * iter->pos will be 0 if we haven't read anything.
2501                  */
2502                 if (!tracer_enabled && iter->pos)
2503                         break;
2504
2505                 continue;
2506         }
2507
2508         /* stop when tracing is finished */
2509         if (trace_empty(iter))
2510                 goto out;
2511
2512         if (cnt >= PAGE_SIZE)
2513                 cnt = PAGE_SIZE - 1;
2514
2515         /* reset all but tr, trace, and overruns */
2516         memset(&iter->seq, 0,
2517                sizeof(struct trace_iterator) -
2518                offsetof(struct trace_iterator, seq));
2519         iter->pos = -1;
2520
2521         while (find_next_entry_inc(iter) != NULL) {
2522                 enum print_line_t ret;
2523                 int len = iter->seq.len;
2524
2525                 ret = print_trace_line(iter);
2526                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2527                         /* don't print partial lines */
2528                         iter->seq.len = len;
2529                         break;
2530                 }
2531
2532                 trace_consume(iter);
2533
2534                 if (iter->seq.len >= cnt)
2535                         break;
2536         }
2537
2538         /* Now copy what we have to the user */
2539         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2540         if (iter->seq.readpos >= iter->seq.len)
2541                 trace_seq_reset(&iter->seq);
2542
2543         /*
2544          * If there was nothing to send to user, inspite of consuming trace
2545          * entries, go back to wait for more entries.
2546          */
2547         if (sret == -EBUSY)
2548                 goto waitagain;
2549
2550 out:
2551         mutex_unlock(&trace_types_lock);
2552
2553         return sret;
2554 }
2555
2556 static ssize_t
2557 tracing_entries_read(struct file *filp, char __user *ubuf,
2558                      size_t cnt, loff_t *ppos)
2559 {
2560         struct trace_array *tr = filp->private_data;
2561         char buf[64];
2562         int r;
2563
2564         r = sprintf(buf, "%lu\n", tr->entries >> 10);
2565         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2566 }
2567
2568 static ssize_t
2569 tracing_entries_write(struct file *filp, const char __user *ubuf,
2570                       size_t cnt, loff_t *ppos)
2571 {
2572         unsigned long val;
2573         char buf[64];
2574         int ret, cpu;
2575
2576         if (cnt >= sizeof(buf))
2577                 return -EINVAL;
2578
2579         if (copy_from_user(&buf, ubuf, cnt))
2580                 return -EFAULT;
2581
2582         buf[cnt] = 0;
2583
2584         ret = strict_strtoul(buf, 10, &val);
2585         if (ret < 0)
2586                 return ret;
2587
2588         /* must have at least 1 entry */
2589         if (!val)
2590                 return -EINVAL;
2591
2592         mutex_lock(&trace_types_lock);
2593
2594         tracing_stop();
2595
2596         /* disable all cpu buffers */
2597         for_each_tracing_cpu(cpu) {
2598                 if (global_trace.data[cpu])
2599                         atomic_inc(&global_trace.data[cpu]->disabled);
2600                 if (max_tr.data[cpu])
2601                         atomic_inc(&max_tr.data[cpu]->disabled);
2602         }
2603
2604         /* value is in KB */
2605         val <<= 10;
2606
2607         if (val != global_trace.entries) {
2608                 ret = ring_buffer_resize(global_trace.buffer, val);
2609                 if (ret < 0) {
2610                         cnt = ret;
2611                         goto out;
2612                 }
2613
2614                 ret = ring_buffer_resize(max_tr.buffer, val);
2615                 if (ret < 0) {
2616                         int r;
2617                         cnt = ret;
2618                         r = ring_buffer_resize(global_trace.buffer,
2619                                                global_trace.entries);
2620                         if (r < 0) {
2621                                 /* AARGH! We are left with different
2622                                  * size max buffer!!!! */
2623                                 WARN_ON(1);
2624                                 tracing_disabled = 1;
2625                         }
2626                         goto out;
2627                 }
2628
2629                 global_trace.entries = val;
2630         }
2631
2632         filp->f_pos += cnt;
2633
2634         /* If check pages failed, return ENOMEM */
2635         if (tracing_disabled)
2636                 cnt = -ENOMEM;
2637  out:
2638         for_each_tracing_cpu(cpu) {
2639                 if (global_trace.data[cpu])
2640                         atomic_dec(&global_trace.data[cpu]->disabled);
2641                 if (max_tr.data[cpu])
2642                         atomic_dec(&max_tr.data[cpu]->disabled);
2643         }
2644
2645         tracing_start();
2646         max_tr.entries = global_trace.entries;
2647         mutex_unlock(&trace_types_lock);
2648
2649         return cnt;
2650 }
2651
2652 static int mark_printk(const char *fmt, ...)
2653 {
2654         int ret;
2655         va_list args;
2656         va_start(args, fmt);
2657         ret = trace_vprintk(0, -1, fmt, args);
2658         va_end(args);
2659         return ret;
2660 }
2661
2662 static ssize_t
2663 tracing_mark_write(struct file *filp, const char __user *ubuf,
2664                                         size_t cnt, loff_t *fpos)
2665 {
2666         char *buf;
2667         char *end;
2668
2669         if (tracing_disabled)
2670                 return -EINVAL;
2671
2672         if (cnt > TRACE_BUF_SIZE)
2673                 cnt = TRACE_BUF_SIZE;
2674
2675         buf = kmalloc(cnt + 1, GFP_KERNEL);
2676         if (buf == NULL)
2677                 return -ENOMEM;
2678
2679         if (copy_from_user(buf, ubuf, cnt)) {
2680                 kfree(buf);
2681                 return -EFAULT;
2682         }
2683
2684         /* Cut from the first nil or newline. */
2685         buf[cnt] = '\0';
2686         end = strchr(buf, '\n');
2687         if (end)
2688                 *end = '\0';
2689
2690         cnt = mark_printk("%s\n", buf);
2691         kfree(buf);
2692         *fpos += cnt;
2693
2694         return cnt;
2695 }
2696
2697 static struct file_operations tracing_max_lat_fops = {
2698         .open           = tracing_open_generic,
2699         .read           = tracing_max_lat_read,
2700         .write          = tracing_max_lat_write,
2701 };
2702
2703 static struct file_operations tracing_ctrl_fops = {
2704         .open           = tracing_open_generic,
2705         .read           = tracing_ctrl_read,
2706         .write          = tracing_ctrl_write,
2707 };
2708
2709 static struct file_operations set_tracer_fops = {
2710         .open           = tracing_open_generic,
2711         .read           = tracing_set_trace_read,
2712         .write          = tracing_set_trace_write,
2713 };
2714
2715 static struct file_operations tracing_pipe_fops = {
2716         .open           = tracing_open_pipe,
2717         .poll           = tracing_poll_pipe,
2718         .read           = tracing_read_pipe,
2719         .release        = tracing_release_pipe,
2720 };
2721
2722 static struct file_operations tracing_entries_fops = {
2723         .open           = tracing_open_generic,
2724         .read           = tracing_entries_read,
2725         .write          = tracing_entries_write,
2726 };
2727
2728 static struct file_operations tracing_mark_fops = {
2729         .open           = tracing_open_generic,
2730         .write          = tracing_mark_write,
2731 };
2732
2733 #ifdef CONFIG_DYNAMIC_FTRACE
2734
2735 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
2736 {
2737         return 0;
2738 }
2739
2740 static ssize_t
2741 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
2742                   size_t cnt, loff_t *ppos)
2743 {
2744         static char ftrace_dyn_info_buffer[1024];
2745         static DEFINE_MUTEX(dyn_info_mutex);
2746         unsigned long *p = filp->private_data;
2747         char *buf = ftrace_dyn_info_buffer;
2748         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
2749         int r;
2750
2751         mutex_lock(&dyn_info_mutex);
2752         r = sprintf(buf, "%ld ", *p);
2753
2754         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
2755         buf[r++] = '\n';
2756
2757         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2758
2759         mutex_unlock(&dyn_info_mutex);
2760
2761         return r;
2762 }
2763
2764 static struct file_operations tracing_dyn_info_fops = {
2765         .open           = tracing_open_generic,
2766         .read           = tracing_read_dyn_info,
2767 };
2768 #endif
2769
2770 static struct dentry *d_tracer;
2771
2772 struct dentry *tracing_init_dentry(void)
2773 {
2774         static int once;
2775
2776         if (d_tracer)
2777                 return d_tracer;
2778
2779         d_tracer = debugfs_create_dir("tracing", NULL);
2780
2781         if (!d_tracer && !once) {
2782                 once = 1;
2783                 pr_warning("Could not create debugfs directory 'tracing'\n");
2784                 return NULL;
2785         }
2786
2787         return d_tracer;
2788 }
2789
2790 #ifdef CONFIG_FTRACE_SELFTEST
2791 /* Let selftest have access to static functions in this file */
2792 #include "trace_selftest.c"
2793 #endif
2794
2795 static __init int tracer_init_debugfs(void)
2796 {
2797         struct dentry *d_tracer;
2798         struct dentry *entry;
2799
2800         d_tracer = tracing_init_dentry();
2801
2802         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
2803                                     &global_trace, &tracing_ctrl_fops);
2804         if (!entry)
2805                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
2806
2807         entry = debugfs_create_file("trace_options", 0644, d_tracer,
2808                                     NULL, &tracing_iter_fops);
2809         if (!entry)
2810                 pr_warning("Could not create debugfs 'trace_options' entry\n");
2811
2812         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
2813                                     NULL, &tracing_cpumask_fops);
2814         if (!entry)
2815                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
2816
2817         entry = debugfs_create_file("latency_trace", 0444, d_tracer,
2818                                     &global_trace, &tracing_lt_fops);
2819         if (!entry)
2820                 pr_warning("Could not create debugfs 'latency_trace' entry\n");
2821
2822         entry = debugfs_create_file("trace", 0444, d_tracer,
2823                                     &global_trace, &tracing_fops);
2824         if (!entry)
2825                 pr_warning("Could not create debugfs 'trace' entry\n");
2826
2827         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
2828                                     &global_trace, &show_traces_fops);
2829         if (!entry)
2830                 pr_warning("Could not create debugfs 'available_tracers' entry\n");
2831
2832         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
2833                                     &global_trace, &set_tracer_fops);
2834         if (!entry)
2835                 pr_warning("Could not create debugfs 'current_tracer' entry\n");
2836
2837         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
2838                                     &tracing_max_latency,
2839                                     &tracing_max_lat_fops);
2840         if (!entry)
2841                 pr_warning("Could not create debugfs "
2842                            "'tracing_max_latency' entry\n");
2843
2844         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
2845                                     &tracing_thresh, &tracing_max_lat_fops);
2846         if (!entry)
2847                 pr_warning("Could not create debugfs "
2848                            "'tracing_thresh' entry\n");
2849         entry = debugfs_create_file("README", 0644, d_tracer,
2850                                     NULL, &tracing_readme_fops);
2851         if (!entry)
2852                 pr_warning("Could not create debugfs 'README' entry\n");
2853
2854         entry = debugfs_create_file("trace_pipe", 0644, d_tracer,
2855                                     NULL, &tracing_pipe_fops);
2856         if (!entry)
2857                 pr_warning("Could not create debugfs "
2858                            "'trace_pipe' entry\n");
2859
2860         entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
2861                                     &global_trace, &tracing_entries_fops);
2862         if (!entry)
2863                 pr_warning("Could not create debugfs "
2864                            "'buffer_size_kb' entry\n");
2865
2866         entry = debugfs_create_file("trace_marker", 0220, d_tracer,
2867                                     NULL, &tracing_mark_fops);
2868         if (!entry)
2869                 pr_warning("Could not create debugfs "
2870                            "'trace_marker' entry\n");
2871
2872 #ifdef CONFIG_DYNAMIC_FTRACE
2873         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
2874                                     &ftrace_update_tot_cnt,
2875                                     &tracing_dyn_info_fops);
2876         if (!entry)
2877                 pr_warning("Could not create debugfs "
2878                            "'dyn_ftrace_total_info' entry\n");
2879 #endif
2880 #ifdef CONFIG_SYSPROF_TRACER
2881         init_tracer_sysprof_debugfs(d_tracer);
2882 #endif
2883         return 0;
2884 }
2885
2886 int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
2887 {
2888         static DEFINE_SPINLOCK(trace_buf_lock);
2889         static char trace_buf[TRACE_BUF_SIZE];
2890
2891         struct ring_buffer_event *event;
2892         struct trace_array *tr = &global_trace;
2893         struct trace_array_cpu *data;
2894         int cpu, len = 0, size, pc;
2895         struct print_entry *entry;
2896         unsigned long irq_flags;
2897
2898         if (tracing_disabled || tracing_selftest_running)
2899                 return 0;
2900
2901         pc = preempt_count();
2902         preempt_disable_notrace();
2903         cpu = raw_smp_processor_id();
2904         data = tr->data[cpu];
2905
2906         if (unlikely(atomic_read(&data->disabled)))
2907                 goto out;
2908
2909         pause_graph_tracing();
2910         spin_lock_irqsave(&trace_buf_lock, irq_flags);
2911         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
2912
2913         len = min(len, TRACE_BUF_SIZE-1);
2914         trace_buf[len] = 0;
2915
2916         size = sizeof(*entry) + len + 1;
2917         event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
2918         if (!event)
2919                 goto out_unlock;
2920         entry = ring_buffer_event_data(event);
2921         tracing_generic_entry_update(&entry->ent, irq_flags, pc);
2922         entry->ent.type                 = TRACE_PRINT;
2923         entry->ip                       = ip;
2924         entry->depth                    = depth;
2925
2926         memcpy(&entry->buf, trace_buf, len);
2927         entry->buf[len] = 0;
2928         ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
2929
2930  out_unlock:
2931         spin_unlock_irqrestore(&trace_buf_lock, irq_flags);
2932         unpause_graph_tracing();
2933  out:
2934         preempt_enable_notrace();
2935
2936         return len;
2937 }
2938 EXPORT_SYMBOL_GPL(trace_vprintk);
2939
2940 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
2941 {
2942         int ret;
2943         va_list ap;
2944
2945         if (!(trace_flags & TRACE_ITER_PRINTK))
2946                 return 0;
2947
2948         va_start(ap, fmt);
2949         ret = trace_vprintk(ip, task_curr_ret_stack(current), fmt, ap);
2950         va_end(ap);
2951         return ret;
2952 }
2953 EXPORT_SYMBOL_GPL(__ftrace_printk);
2954
2955 static int trace_panic_handler(struct notifier_block *this,
2956                                unsigned long event, void *unused)
2957 {
2958         if (ftrace_dump_on_oops)
2959                 ftrace_dump();
2960         return NOTIFY_OK;
2961 }
2962
2963 static struct notifier_block trace_panic_notifier = {
2964         .notifier_call  = trace_panic_handler,
2965         .next           = NULL,
2966         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
2967 };
2968
2969 static int trace_die_handler(struct notifier_block *self,
2970                              unsigned long val,
2971                              void *data)
2972 {
2973         switch (val) {
2974         case DIE_OOPS:
2975                 if (ftrace_dump_on_oops)
2976                         ftrace_dump();
2977                 break;
2978         default:
2979                 break;
2980         }
2981         return NOTIFY_OK;
2982 }
2983
2984 static struct notifier_block trace_die_notifier = {
2985         .notifier_call = trace_die_handler,
2986         .priority = 200
2987 };
2988
2989 /*
2990  * printk is set to max of 1024, we really don't need it that big.
2991  * Nothing should be printing 1000 characters anyway.
2992  */
2993 #define TRACE_MAX_PRINT         1000
2994
2995 /*
2996  * Define here KERN_TRACE so that we have one place to modify
2997  * it if we decide to change what log level the ftrace dump
2998  * should be at.
2999  */
3000 #define KERN_TRACE              KERN_EMERG
3001
3002 static void
3003 trace_printk_seq(struct trace_seq *s)
3004 {
3005         /* Probably should print a warning here. */
3006         if (s->len >= 1000)
3007                 s->len = 1000;
3008
3009         /* should be zero ended, but we are paranoid. */
3010         s->buffer[s->len] = 0;
3011
3012         printk(KERN_TRACE "%s", s->buffer);
3013
3014         trace_seq_reset(s);
3015 }
3016
3017 void ftrace_dump(void)
3018 {
3019         static DEFINE_SPINLOCK(ftrace_dump_lock);
3020         /* use static because iter can be a bit big for the stack */
3021         static struct trace_iterator iter;
3022         static int dump_ran;
3023         unsigned long flags;
3024         int cnt = 0, cpu;
3025
3026         /* only one dump */
3027         spin_lock_irqsave(&ftrace_dump_lock, flags);
3028         if (dump_ran)
3029                 goto out;
3030
3031         dump_ran = 1;
3032
3033         /* No turning back! */
3034         tracing_off();
3035         ftrace_kill();
3036
3037         for_each_tracing_cpu(cpu) {
3038                 atomic_inc(&global_trace.data[cpu]->disabled);
3039         }
3040
3041         /* don't look at user memory in panic mode */
3042         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3043
3044         printk(KERN_TRACE "Dumping ftrace buffer:\n");
3045
3046         iter.tr = &global_trace;
3047         iter.trace = current_trace;
3048
3049         /*
3050          * We need to stop all tracing on all CPUS to read the
3051          * the next buffer. This is a bit expensive, but is
3052          * not done often. We fill all what we can read,
3053          * and then release the locks again.
3054          */
3055
3056         while (!trace_empty(&iter)) {
3057
3058                 if (!cnt)
3059                         printk(KERN_TRACE "---------------------------------\n");
3060
3061                 cnt++;
3062
3063                 /* reset all but tr, trace, and overruns */
3064                 memset(&iter.seq, 0,
3065                        sizeof(struct trace_iterator) -
3066                        offsetof(struct trace_iterator, seq));
3067                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3068                 iter.pos = -1;
3069
3070                 if (find_next_entry_inc(&iter) != NULL) {
3071                         print_trace_line(&iter);
3072                         trace_consume(&iter);
3073                 }
3074
3075                 trace_printk_seq(&iter.seq);
3076         }
3077
3078         if (!cnt)
3079                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
3080         else
3081                 printk(KERN_TRACE "---------------------------------\n");
3082
3083  out:
3084         spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3085 }
3086
3087 __init static int tracer_alloc_buffers(void)
3088 {
3089         struct trace_array_cpu *data;
3090         int i;
3091         int ret = -ENOMEM;
3092
3093         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
3094                 goto out;
3095
3096         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
3097                 goto out_free_buffer_mask;
3098
3099         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
3100         cpumask_copy(tracing_cpumask, cpu_all_mask);
3101
3102         /* TODO: make the number of buffers hot pluggable with CPUS */
3103         global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3104                                                    TRACE_BUFFER_FLAGS);
3105         if (!global_trace.buffer) {
3106                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3107                 WARN_ON(1);
3108                 goto out_free_cpumask;
3109         }
3110         global_trace.entries = ring_buffer_size(global_trace.buffer);
3111
3112
3113 #ifdef CONFIG_TRACER_MAX_TRACE
3114         max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3115                                              TRACE_BUFFER_FLAGS);
3116         if (!max_tr.buffer) {
3117                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3118                 WARN_ON(1);
3119                 ring_buffer_free(global_trace.buffer);
3120                 goto out_free_cpumask;
3121         }
3122         max_tr.entries = ring_buffer_size(max_tr.buffer);
3123         WARN_ON(max_tr.entries != global_trace.entries);
3124 #endif
3125
3126         /* Allocate the first page for all buffers */
3127         for_each_tracing_cpu(i) {
3128                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3129                 max_tr.data[i] = &per_cpu(max_data, i);
3130         }
3131
3132         trace_init_cmdlines();
3133
3134         register_tracer(&nop_trace);
3135 #ifdef CONFIG_BOOT_TRACER
3136         register_tracer(&boot_tracer);
3137         current_trace = &boot_tracer;
3138         current_trace->init(&global_trace);
3139 #else
3140         current_trace = &nop_trace;
3141 #endif
3142         /* All seems OK, enable tracing */
3143         tracing_disabled = 0;
3144
3145         atomic_notifier_chain_register(&panic_notifier_list,
3146                                        &trace_panic_notifier);
3147
3148         register_die_notifier(&trace_die_notifier);
3149         ret = 0;
3150
3151 out_free_cpumask:
3152         free_cpumask_var(tracing_cpumask);
3153 out_free_buffer_mask:
3154         free_cpumask_var(tracing_buffer_mask);
3155 out:
3156         return ret;
3157 }
3158 early_initcall(tracer_alloc_buffers);
3159 fs_initcall(tracer_init_debugfs);