Merge branch 'tracing-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6.git] / 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/ring_buffer.h>
15 #include <linux/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/ctype.h>
36 #include <linux/init.h>
37 #include <linux/poll.h>
38 #include <linux/gfp.h>
39 #include <linux/fs.h>
40
41 #include "trace.h"
42 #include "trace_output.h"
43
44 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
45
46 /*
47  * On boot up, the ring buffer is set to the minimum size, so that
48  * we do not waste memory on systems that are not using tracing.
49  */
50 int ring_buffer_expanded;
51
52 /*
53  * We need to change this state when a selftest is running.
54  * A selftest will lurk into the ring-buffer to count the
55  * entries inserted during the selftest although some concurrent
56  * insertions into the ring-buffer such as trace_printk could occurred
57  * at the same time, giving false positive or negative results.
58  */
59 static bool __read_mostly tracing_selftest_running;
60
61 /*
62  * If a tracer is running, we do not want to run SELFTEST.
63  */
64 bool __read_mostly tracing_selftest_disabled;
65
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
68         { }
69 };
70
71 static struct tracer_flags dummy_tracer_flags = {
72         .val = 0,
73         .opts = dummy_tracer_opt
74 };
75
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
77 {
78         return 0;
79 }
80
81 /*
82  * Kill all tracing for good (never come back).
83  * It is initialized to 1 but will turn to zero if the initialization
84  * of the tracer is successful. But that is the only place that sets
85  * this back to zero.
86  */
87 static int tracing_disabled = 1;
88
89 DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
90
91 static inline void ftrace_disable_cpu(void)
92 {
93         preempt_disable();
94         local_inc(&__get_cpu_var(ftrace_cpu_disabled));
95 }
96
97 static inline void ftrace_enable_cpu(void)
98 {
99         local_dec(&__get_cpu_var(ftrace_cpu_disabled));
100         preempt_enable();
101 }
102
103 static cpumask_var_t __read_mostly      tracing_buffer_mask;
104
105 /* Define which cpu buffers are currently read in trace_pipe */
106 static cpumask_var_t                    tracing_reader_cpumask;
107
108 #define for_each_tracing_cpu(cpu)       \
109         for_each_cpu(cpu, tracing_buffer_mask)
110
111 /*
112  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
113  *
114  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
115  * is set, then ftrace_dump is called. This will output the contents
116  * of the ftrace buffers to the console.  This is very useful for
117  * capturing traces that lead to crashes and outputing it to a
118  * serial console.
119  *
120  * It is default off, but you can enable it with either specifying
121  * "ftrace_dump_on_oops" in the kernel command line, or setting
122  * /proc/sys/kernel/ftrace_dump_on_oops to true.
123  */
124 int ftrace_dump_on_oops;
125
126 static int tracing_set_tracer(const char *buf);
127
128 #define MAX_TRACER_SIZE         100
129 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
130 static char *default_bootup_tracer;
131
132 static int __init set_cmdline_ftrace(char *str)
133 {
134         strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
135         default_bootup_tracer = bootup_tracer_buf;
136         /* We are using ftrace early, expand it */
137         ring_buffer_expanded = 1;
138         return 1;
139 }
140 __setup("ftrace=", set_cmdline_ftrace);
141
142 static int __init set_ftrace_dump_on_oops(char *str)
143 {
144         ftrace_dump_on_oops = 1;
145         return 1;
146 }
147 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
148
149 unsigned long long ns2usecs(cycle_t nsec)
150 {
151         nsec += 500;
152         do_div(nsec, 1000);
153         return nsec;
154 }
155
156 /*
157  * The global_trace is the descriptor that holds the tracing
158  * buffers for the live tracing. For each CPU, it contains
159  * a link list of pages that will store trace entries. The
160  * page descriptor of the pages in the memory is used to hold
161  * the link list by linking the lru item in the page descriptor
162  * to each of the pages in the buffer per CPU.
163  *
164  * For each active CPU there is a data field that holds the
165  * pages for the buffer for that CPU. Each CPU has the same number
166  * of pages allocated for its buffer.
167  */
168 static struct trace_array       global_trace;
169
170 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
171
172 int filter_current_check_discard(struct ring_buffer *buffer,
173                                  struct ftrace_event_call *call, void *rec,
174                                  struct ring_buffer_event *event)
175 {
176         return filter_check_discard(call, rec, buffer, event);
177 }
178 EXPORT_SYMBOL_GPL(filter_current_check_discard);
179
180 cycle_t ftrace_now(int cpu)
181 {
182         u64 ts;
183
184         /* Early boot up does not have a buffer yet */
185         if (!global_trace.buffer)
186                 return trace_clock_local();
187
188         ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
189         ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
190
191         return ts;
192 }
193
194 /*
195  * The max_tr is used to snapshot the global_trace when a maximum
196  * latency is reached. Some tracers will use this to store a maximum
197  * trace while it continues examining live traces.
198  *
199  * The buffers for the max_tr are set up the same as the global_trace.
200  * When a snapshot is taken, the link list of the max_tr is swapped
201  * with the link list of the global_trace and the buffers are reset for
202  * the global_trace so the tracing can continue.
203  */
204 static struct trace_array       max_tr;
205
206 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
207
208 /* tracer_enabled is used to toggle activation of a tracer */
209 static int                      tracer_enabled = 1;
210
211 /**
212  * tracing_is_enabled - return tracer_enabled status
213  *
214  * This function is used by other tracers to know the status
215  * of the tracer_enabled flag.  Tracers may use this function
216  * to know if it should enable their features when starting
217  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
218  */
219 int tracing_is_enabled(void)
220 {
221         return tracer_enabled;
222 }
223
224 /*
225  * trace_buf_size is the size in bytes that is allocated
226  * for a buffer. Note, the number of bytes is always rounded
227  * to page size.
228  *
229  * This number is purposely set to a low number of 16384.
230  * If the dump on oops happens, it will be much appreciated
231  * to not have to wait for all that output. Anyway this can be
232  * boot time and run time configurable.
233  */
234 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
235
236 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
237
238 /* trace_types holds a link list of available tracers. */
239 static struct tracer            *trace_types __read_mostly;
240
241 /* current_trace points to the tracer that is currently active */
242 static struct tracer            *current_trace __read_mostly;
243
244 /*
245  * trace_types_lock is used to protect the trace_types list.
246  * This lock is also used to keep user access serialized.
247  * Accesses from userspace will grab this lock while userspace
248  * activities happen inside the kernel.
249  */
250 static DEFINE_MUTEX(trace_types_lock);
251
252 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
253 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
254
255 /* trace_flags holds trace_options default values */
256 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
257         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
258         TRACE_ITER_GRAPH_TIME;
259
260 static int trace_stop_count;
261 static DEFINE_SPINLOCK(tracing_start_lock);
262
263 /**
264  * trace_wake_up - wake up tasks waiting for trace input
265  *
266  * Simply wakes up any task that is blocked on the trace_wait
267  * queue. These is used with trace_poll for tasks polling the trace.
268  */
269 void trace_wake_up(void)
270 {
271         int cpu;
272
273         if (trace_flags & TRACE_ITER_BLOCK)
274                 return;
275         /*
276          * The runqueue_is_locked() can fail, but this is the best we
277          * have for now:
278          */
279         cpu = get_cpu();
280         if (!runqueue_is_locked(cpu))
281                 wake_up(&trace_wait);
282         put_cpu();
283 }
284
285 static int __init set_buf_size(char *str)
286 {
287         unsigned long buf_size;
288
289         if (!str)
290                 return 0;
291         buf_size = memparse(str, &str);
292         /* nr_entries can not be zero */
293         if (buf_size == 0)
294                 return 0;
295         trace_buf_size = buf_size;
296         return 1;
297 }
298 __setup("trace_buf_size=", set_buf_size);
299
300 unsigned long nsecs_to_usecs(unsigned long nsecs)
301 {
302         return nsecs / 1000;
303 }
304
305 /* These must match the bit postions in trace_iterator_flags */
306 static const char *trace_options[] = {
307         "print-parent",
308         "sym-offset",
309         "sym-addr",
310         "verbose",
311         "raw",
312         "hex",
313         "bin",
314         "block",
315         "stacktrace",
316         "sched-tree",
317         "trace_printk",
318         "ftrace_preempt",
319         "branch",
320         "annotate",
321         "userstacktrace",
322         "sym-userobj",
323         "printk-msg-only",
324         "context-info",
325         "latency-format",
326         "sleep-time",
327         "graph-time",
328         NULL
329 };
330
331 static struct {
332         u64 (*func)(void);
333         const char *name;
334 } trace_clocks[] = {
335         { trace_clock_local,    "local" },
336         { trace_clock_global,   "global" },
337 };
338
339 int trace_clock_id;
340
341 /*
342  * trace_parser_get_init - gets the buffer for trace parser
343  */
344 int trace_parser_get_init(struct trace_parser *parser, int size)
345 {
346         memset(parser, 0, sizeof(*parser));
347
348         parser->buffer = kmalloc(size, GFP_KERNEL);
349         if (!parser->buffer)
350                 return 1;
351
352         parser->size = size;
353         return 0;
354 }
355
356 /*
357  * trace_parser_put - frees the buffer for trace parser
358  */
359 void trace_parser_put(struct trace_parser *parser)
360 {
361         kfree(parser->buffer);
362 }
363
364 /*
365  * trace_get_user - reads the user input string separated by  space
366  * (matched by isspace(ch))
367  *
368  * For each string found the 'struct trace_parser' is updated,
369  * and the function returns.
370  *
371  * Returns number of bytes read.
372  *
373  * See kernel/trace/trace.h for 'struct trace_parser' details.
374  */
375 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
376         size_t cnt, loff_t *ppos)
377 {
378         char ch;
379         size_t read = 0;
380         ssize_t ret;
381
382         if (!*ppos)
383                 trace_parser_clear(parser);
384
385         ret = get_user(ch, ubuf++);
386         if (ret)
387                 goto out;
388
389         read++;
390         cnt--;
391
392         /*
393          * The parser is not finished with the last write,
394          * continue reading the user input without skipping spaces.
395          */
396         if (!parser->cont) {
397                 /* skip white space */
398                 while (cnt && isspace(ch)) {
399                         ret = get_user(ch, ubuf++);
400                         if (ret)
401                                 goto out;
402                         read++;
403                         cnt--;
404                 }
405
406                 /* only spaces were written */
407                 if (isspace(ch)) {
408                         *ppos += read;
409                         ret = read;
410                         goto out;
411                 }
412
413                 parser->idx = 0;
414         }
415
416         /* read the non-space input */
417         while (cnt && !isspace(ch)) {
418                 if (parser->idx < parser->size - 1)
419                         parser->buffer[parser->idx++] = ch;
420                 else {
421                         ret = -EINVAL;
422                         goto out;
423                 }
424                 ret = get_user(ch, ubuf++);
425                 if (ret)
426                         goto out;
427                 read++;
428                 cnt--;
429         }
430
431         /* We either got finished input or we have to wait for another call. */
432         if (isspace(ch)) {
433                 parser->buffer[parser->idx] = 0;
434                 parser->cont = false;
435         } else {
436                 parser->cont = true;
437                 parser->buffer[parser->idx++] = ch;
438         }
439
440         *ppos += read;
441         ret = read;
442
443 out:
444         return ret;
445 }
446
447 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
448 {
449         int len;
450         int ret;
451
452         if (!cnt)
453                 return 0;
454
455         if (s->len <= s->readpos)
456                 return -EBUSY;
457
458         len = s->len - s->readpos;
459         if (cnt > len)
460                 cnt = len;
461         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
462         if (ret == cnt)
463                 return -EFAULT;
464
465         cnt -= ret;
466
467         s->readpos += cnt;
468         return cnt;
469 }
470
471 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
472 {
473         int len;
474         void *ret;
475
476         if (s->len <= s->readpos)
477                 return -EBUSY;
478
479         len = s->len - s->readpos;
480         if (cnt > len)
481                 cnt = len;
482         ret = memcpy(buf, s->buffer + s->readpos, cnt);
483         if (!ret)
484                 return -EFAULT;
485
486         s->readpos += cnt;
487         return cnt;
488 }
489
490 /*
491  * ftrace_max_lock is used to protect the swapping of buffers
492  * when taking a max snapshot. The buffers themselves are
493  * protected by per_cpu spinlocks. But the action of the swap
494  * needs its own lock.
495  *
496  * This is defined as a raw_spinlock_t in order to help
497  * with performance when lockdep debugging is enabled.
498  *
499  * It is also used in other places outside the update_max_tr
500  * so it needs to be defined outside of the
501  * CONFIG_TRACER_MAX_TRACE.
502  */
503 static raw_spinlock_t ftrace_max_lock =
504         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
505
506 #ifdef CONFIG_TRACER_MAX_TRACE
507 unsigned long __read_mostly     tracing_max_latency;
508 unsigned long __read_mostly     tracing_thresh;
509
510 /*
511  * Copy the new maximum trace into the separate maximum-trace
512  * structure. (this way the maximum trace is permanently saved,
513  * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
514  */
515 static void
516 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
517 {
518         struct trace_array_cpu *data = tr->data[cpu];
519         struct trace_array_cpu *max_data = tr->data[cpu];
520
521         max_tr.cpu = cpu;
522         max_tr.time_start = data->preempt_timestamp;
523
524         max_data = max_tr.data[cpu];
525         max_data->saved_latency = tracing_max_latency;
526         max_data->critical_start = data->critical_start;
527         max_data->critical_end = data->critical_end;
528
529         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
530         max_data->pid = tsk->pid;
531         max_data->uid = task_uid(tsk);
532         max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
533         max_data->policy = tsk->policy;
534         max_data->rt_priority = tsk->rt_priority;
535
536         /* record this tasks comm */
537         tracing_record_cmdline(tsk);
538 }
539
540 /**
541  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
542  * @tr: tracer
543  * @tsk: the task with the latency
544  * @cpu: The cpu that initiated the trace.
545  *
546  * Flip the buffers between the @tr and the max_tr and record information
547  * about which task was the cause of this latency.
548  */
549 void
550 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
551 {
552         struct ring_buffer *buf = tr->buffer;
553
554         if (trace_stop_count)
555                 return;
556
557         WARN_ON_ONCE(!irqs_disabled());
558         __raw_spin_lock(&ftrace_max_lock);
559
560         tr->buffer = max_tr.buffer;
561         max_tr.buffer = buf;
562
563         __update_max_tr(tr, tsk, cpu);
564         __raw_spin_unlock(&ftrace_max_lock);
565 }
566
567 /**
568  * update_max_tr_single - only copy one trace over, and reset the rest
569  * @tr - tracer
570  * @tsk - task with the latency
571  * @cpu - the cpu of the buffer to copy.
572  *
573  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
574  */
575 void
576 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
577 {
578         int ret;
579
580         if (trace_stop_count)
581                 return;
582
583         WARN_ON_ONCE(!irqs_disabled());
584         __raw_spin_lock(&ftrace_max_lock);
585
586         ftrace_disable_cpu();
587
588         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
589
590         if (ret == -EBUSY) {
591                 /*
592                  * We failed to swap the buffer due to a commit taking
593                  * place on this CPU. We fail to record, but we reset
594                  * the max trace buffer (no one writes directly to it)
595                  * and flag that it failed.
596                  */
597                 trace_array_printk(&max_tr, _THIS_IP_,
598                         "Failed to swap buffers due to commit in progress\n");
599         }
600
601         ftrace_enable_cpu();
602
603         WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
604
605         __update_max_tr(tr, tsk, cpu);
606         __raw_spin_unlock(&ftrace_max_lock);
607 }
608 #endif /* CONFIG_TRACER_MAX_TRACE */
609
610 /**
611  * register_tracer - register a tracer with the ftrace system.
612  * @type - the plugin for the tracer
613  *
614  * Register a new plugin tracer.
615  */
616 int register_tracer(struct tracer *type)
617 __releases(kernel_lock)
618 __acquires(kernel_lock)
619 {
620         struct tracer *t;
621         int ret = 0;
622
623         if (!type->name) {
624                 pr_info("Tracer must have a name\n");
625                 return -1;
626         }
627
628         if (strlen(type->name) > MAX_TRACER_SIZE) {
629                 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
630                 return -1;
631         }
632
633         /*
634          * When this gets called we hold the BKL which means that
635          * preemption is disabled. Various trace selftests however
636          * need to disable and enable preemption for successful tests.
637          * So we drop the BKL here and grab it after the tests again.
638          */
639         unlock_kernel();
640         mutex_lock(&trace_types_lock);
641
642         tracing_selftest_running = true;
643
644         for (t = trace_types; t; t = t->next) {
645                 if (strcmp(type->name, t->name) == 0) {
646                         /* already found */
647                         pr_info("Tracer %s already registered\n",
648                                 type->name);
649                         ret = -1;
650                         goto out;
651                 }
652         }
653
654         if (!type->set_flag)
655                 type->set_flag = &dummy_set_flag;
656         if (!type->flags)
657                 type->flags = &dummy_tracer_flags;
658         else
659                 if (!type->flags->opts)
660                         type->flags->opts = dummy_tracer_opt;
661         if (!type->wait_pipe)
662                 type->wait_pipe = default_wait_pipe;
663
664
665 #ifdef CONFIG_FTRACE_STARTUP_TEST
666         if (type->selftest && !tracing_selftest_disabled) {
667                 struct tracer *saved_tracer = current_trace;
668                 struct trace_array *tr = &global_trace;
669
670                 /*
671                  * Run a selftest on this tracer.
672                  * Here we reset the trace buffer, and set the current
673                  * tracer to be this tracer. The tracer can then run some
674                  * internal tracing to verify that everything is in order.
675                  * If we fail, we do not register this tracer.
676                  */
677                 tracing_reset_online_cpus(tr);
678
679                 current_trace = type;
680                 /* the test is responsible for initializing and enabling */
681                 pr_info("Testing tracer %s: ", type->name);
682                 ret = type->selftest(type, tr);
683                 /* the test is responsible for resetting too */
684                 current_trace = saved_tracer;
685                 if (ret) {
686                         printk(KERN_CONT "FAILED!\n");
687                         goto out;
688                 }
689                 /* Only reset on passing, to avoid touching corrupted buffers */
690                 tracing_reset_online_cpus(tr);
691
692                 printk(KERN_CONT "PASSED\n");
693         }
694 #endif
695
696         type->next = trace_types;
697         trace_types = type;
698
699  out:
700         tracing_selftest_running = false;
701         mutex_unlock(&trace_types_lock);
702
703         if (ret || !default_bootup_tracer)
704                 goto out_unlock;
705
706         if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
707                 goto out_unlock;
708
709         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
710         /* Do we want this tracer to start on bootup? */
711         tracing_set_tracer(type->name);
712         default_bootup_tracer = NULL;
713         /* disable other selftests, since this will break it. */
714         tracing_selftest_disabled = 1;
715 #ifdef CONFIG_FTRACE_STARTUP_TEST
716         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
717                type->name);
718 #endif
719
720  out_unlock:
721         lock_kernel();
722         return ret;
723 }
724
725 void unregister_tracer(struct tracer *type)
726 {
727         struct tracer **t;
728
729         mutex_lock(&trace_types_lock);
730         for (t = &trace_types; *t; t = &(*t)->next) {
731                 if (*t == type)
732                         goto found;
733         }
734         pr_info("Tracer %s not registered\n", type->name);
735         goto out;
736
737  found:
738         *t = (*t)->next;
739
740         if (type == current_trace && tracer_enabled) {
741                 tracer_enabled = 0;
742                 tracing_stop();
743                 if (current_trace->stop)
744                         current_trace->stop(&global_trace);
745                 current_trace = &nop_trace;
746         }
747 out:
748         mutex_unlock(&trace_types_lock);
749 }
750
751 static void __tracing_reset(struct trace_array *tr, int cpu)
752 {
753         ftrace_disable_cpu();
754         ring_buffer_reset_cpu(tr->buffer, cpu);
755         ftrace_enable_cpu();
756 }
757
758 void tracing_reset(struct trace_array *tr, int cpu)
759 {
760         struct ring_buffer *buffer = tr->buffer;
761
762         ring_buffer_record_disable(buffer);
763
764         /* Make sure all commits have finished */
765         synchronize_sched();
766         __tracing_reset(tr, cpu);
767
768         ring_buffer_record_enable(buffer);
769 }
770
771 void tracing_reset_online_cpus(struct trace_array *tr)
772 {
773         struct ring_buffer *buffer = tr->buffer;
774         int cpu;
775
776         ring_buffer_record_disable(buffer);
777
778         /* Make sure all commits have finished */
779         synchronize_sched();
780
781         tr->time_start = ftrace_now(tr->cpu);
782
783         for_each_online_cpu(cpu)
784                 __tracing_reset(tr, cpu);
785
786         ring_buffer_record_enable(buffer);
787 }
788
789 void tracing_reset_current(int cpu)
790 {
791         tracing_reset(&global_trace, cpu);
792 }
793
794 void tracing_reset_current_online_cpus(void)
795 {
796         tracing_reset_online_cpus(&global_trace);
797 }
798
799 #define SAVED_CMDLINES 128
800 #define NO_CMDLINE_MAP UINT_MAX
801 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
802 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
803 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
804 static int cmdline_idx;
805 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
806
807 /* temporary disable recording */
808 static atomic_t trace_record_cmdline_disabled __read_mostly;
809
810 static void trace_init_cmdlines(void)
811 {
812         memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
813         memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
814         cmdline_idx = 0;
815 }
816
817 int is_tracing_stopped(void)
818 {
819         return trace_stop_count;
820 }
821
822 /**
823  * ftrace_off_permanent - disable all ftrace code permanently
824  *
825  * This should only be called when a serious anomally has
826  * been detected.  This will turn off the function tracing,
827  * ring buffers, and other tracing utilites. It takes no
828  * locks and can be called from any context.
829  */
830 void ftrace_off_permanent(void)
831 {
832         tracing_disabled = 1;
833         ftrace_stop();
834         tracing_off_permanent();
835 }
836
837 /**
838  * tracing_start - quick start of the tracer
839  *
840  * If tracing is enabled but was stopped by tracing_stop,
841  * this will start the tracer back up.
842  */
843 void tracing_start(void)
844 {
845         struct ring_buffer *buffer;
846         unsigned long flags;
847
848         if (tracing_disabled)
849                 return;
850
851         spin_lock_irqsave(&tracing_start_lock, flags);
852         if (--trace_stop_count) {
853                 if (trace_stop_count < 0) {
854                         /* Someone screwed up their debugging */
855                         WARN_ON_ONCE(1);
856                         trace_stop_count = 0;
857                 }
858                 goto out;
859         }
860
861
862         buffer = global_trace.buffer;
863         if (buffer)
864                 ring_buffer_record_enable(buffer);
865
866         buffer = max_tr.buffer;
867         if (buffer)
868                 ring_buffer_record_enable(buffer);
869
870         ftrace_start();
871  out:
872         spin_unlock_irqrestore(&tracing_start_lock, flags);
873 }
874
875 /**
876  * tracing_stop - quick stop of the tracer
877  *
878  * Light weight way to stop tracing. Use in conjunction with
879  * tracing_start.
880  */
881 void tracing_stop(void)
882 {
883         struct ring_buffer *buffer;
884         unsigned long flags;
885
886         ftrace_stop();
887         spin_lock_irqsave(&tracing_start_lock, flags);
888         if (trace_stop_count++)
889                 goto out;
890
891         buffer = global_trace.buffer;
892         if (buffer)
893                 ring_buffer_record_disable(buffer);
894
895         buffer = max_tr.buffer;
896         if (buffer)
897                 ring_buffer_record_disable(buffer);
898
899  out:
900         spin_unlock_irqrestore(&tracing_start_lock, flags);
901 }
902
903 void trace_stop_cmdline_recording(void);
904
905 static void trace_save_cmdline(struct task_struct *tsk)
906 {
907         unsigned pid, idx;
908
909         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
910                 return;
911
912         /*
913          * It's not the end of the world if we don't get
914          * the lock, but we also don't want to spin
915          * nor do we want to disable interrupts,
916          * so if we miss here, then better luck next time.
917          */
918         if (!__raw_spin_trylock(&trace_cmdline_lock))
919                 return;
920
921         idx = map_pid_to_cmdline[tsk->pid];
922         if (idx == NO_CMDLINE_MAP) {
923                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
924
925                 /*
926                  * Check whether the cmdline buffer at idx has a pid
927                  * mapped. We are going to overwrite that entry so we
928                  * need to clear the map_pid_to_cmdline. Otherwise we
929                  * would read the new comm for the old pid.
930                  */
931                 pid = map_cmdline_to_pid[idx];
932                 if (pid != NO_CMDLINE_MAP)
933                         map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
934
935                 map_cmdline_to_pid[idx] = tsk->pid;
936                 map_pid_to_cmdline[tsk->pid] = idx;
937
938                 cmdline_idx = idx;
939         }
940
941         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
942
943         __raw_spin_unlock(&trace_cmdline_lock);
944 }
945
946 void trace_find_cmdline(int pid, char comm[])
947 {
948         unsigned map;
949
950         if (!pid) {
951                 strcpy(comm, "<idle>");
952                 return;
953         }
954
955         if (pid > PID_MAX_DEFAULT) {
956                 strcpy(comm, "<...>");
957                 return;
958         }
959
960         preempt_disable();
961         __raw_spin_lock(&trace_cmdline_lock);
962         map = map_pid_to_cmdline[pid];
963         if (map != NO_CMDLINE_MAP)
964                 strcpy(comm, saved_cmdlines[map]);
965         else
966                 strcpy(comm, "<...>");
967
968         __raw_spin_unlock(&trace_cmdline_lock);
969         preempt_enable();
970 }
971
972 void tracing_record_cmdline(struct task_struct *tsk)
973 {
974         if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
975             !tracing_is_on())
976                 return;
977
978         trace_save_cmdline(tsk);
979 }
980
981 void
982 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
983                              int pc)
984 {
985         struct task_struct *tsk = current;
986
987         entry->preempt_count            = pc & 0xff;
988         entry->pid                      = (tsk) ? tsk->pid : 0;
989         entry->lock_depth               = (tsk) ? tsk->lock_depth : 0;
990         entry->flags =
991 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
992                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
993 #else
994                 TRACE_FLAG_IRQS_NOSUPPORT |
995 #endif
996                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
997                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
998                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
999 }
1000 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1001
1002 struct ring_buffer_event *
1003 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1004                           int type,
1005                           unsigned long len,
1006                           unsigned long flags, int pc)
1007 {
1008         struct ring_buffer_event *event;
1009
1010         event = ring_buffer_lock_reserve(buffer, len);
1011         if (event != NULL) {
1012                 struct trace_entry *ent = ring_buffer_event_data(event);
1013
1014                 tracing_generic_entry_update(ent, flags, pc);
1015                 ent->type = type;
1016         }
1017
1018         return event;
1019 }
1020
1021 static inline void
1022 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1023                              struct ring_buffer_event *event,
1024                              unsigned long flags, int pc,
1025                              int wake)
1026 {
1027         ring_buffer_unlock_commit(buffer, event);
1028
1029         ftrace_trace_stack(buffer, flags, 6, pc);
1030         ftrace_trace_userstack(buffer, flags, pc);
1031
1032         if (wake)
1033                 trace_wake_up();
1034 }
1035
1036 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1037                                 struct ring_buffer_event *event,
1038                                 unsigned long flags, int pc)
1039 {
1040         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1041 }
1042
1043 struct ring_buffer_event *
1044 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1045                                   int type, unsigned long len,
1046                                   unsigned long flags, int pc)
1047 {
1048         *current_rb = global_trace.buffer;
1049         return trace_buffer_lock_reserve(*current_rb,
1050                                          type, len, flags, pc);
1051 }
1052 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1053
1054 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1055                                         struct ring_buffer_event *event,
1056                                         unsigned long flags, int pc)
1057 {
1058         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1059 }
1060 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1061
1062 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1063                                        struct ring_buffer_event *event,
1064                                        unsigned long flags, int pc)
1065 {
1066         __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1067 }
1068 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1069
1070 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1071                                          struct ring_buffer_event *event)
1072 {
1073         ring_buffer_discard_commit(buffer, event);
1074 }
1075 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1076
1077 void
1078 trace_function(struct trace_array *tr,
1079                unsigned long ip, unsigned long parent_ip, unsigned long flags,
1080                int pc)
1081 {
1082         struct ftrace_event_call *call = &event_function;
1083         struct ring_buffer *buffer = tr->buffer;
1084         struct ring_buffer_event *event;
1085         struct ftrace_entry *entry;
1086
1087         /* If we are reading the ring buffer, don't trace */
1088         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
1089                 return;
1090
1091         event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1092                                           flags, pc);
1093         if (!event)
1094                 return;
1095         entry   = ring_buffer_event_data(event);
1096         entry->ip                       = ip;
1097         entry->parent_ip                = parent_ip;
1098
1099         if (!filter_check_discard(call, entry, buffer, event))
1100                 ring_buffer_unlock_commit(buffer, event);
1101 }
1102
1103 void
1104 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1105        unsigned long ip, unsigned long parent_ip, unsigned long flags,
1106        int pc)
1107 {
1108         if (likely(!atomic_read(&data->disabled)))
1109                 trace_function(tr, ip, parent_ip, flags, pc);
1110 }
1111
1112 #ifdef CONFIG_STACKTRACE
1113 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1114                                  unsigned long flags,
1115                                  int skip, int pc)
1116 {
1117         struct ftrace_event_call *call = &event_kernel_stack;
1118         struct ring_buffer_event *event;
1119         struct stack_entry *entry;
1120         struct stack_trace trace;
1121
1122         event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1123                                           sizeof(*entry), flags, pc);
1124         if (!event)
1125                 return;
1126         entry   = ring_buffer_event_data(event);
1127         memset(&entry->caller, 0, sizeof(entry->caller));
1128
1129         trace.nr_entries        = 0;
1130         trace.max_entries       = FTRACE_STACK_ENTRIES;
1131         trace.skip              = skip;
1132         trace.entries           = entry->caller;
1133
1134         save_stack_trace(&trace);
1135         if (!filter_check_discard(call, entry, buffer, event))
1136                 ring_buffer_unlock_commit(buffer, event);
1137 }
1138
1139 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1140                         int skip, int pc)
1141 {
1142         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1143                 return;
1144
1145         __ftrace_trace_stack(buffer, flags, skip, pc);
1146 }
1147
1148 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1149                    int pc)
1150 {
1151         __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1152 }
1153
1154 void
1155 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1156 {
1157         struct ftrace_event_call *call = &event_user_stack;
1158         struct ring_buffer_event *event;
1159         struct userstack_entry *entry;
1160         struct stack_trace trace;
1161
1162         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1163                 return;
1164
1165         event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1166                                           sizeof(*entry), flags, pc);
1167         if (!event)
1168                 return;
1169         entry   = ring_buffer_event_data(event);
1170
1171         entry->tgid             = current->tgid;
1172         memset(&entry->caller, 0, sizeof(entry->caller));
1173
1174         trace.nr_entries        = 0;
1175         trace.max_entries       = FTRACE_STACK_ENTRIES;
1176         trace.skip              = 0;
1177         trace.entries           = entry->caller;
1178
1179         save_stack_trace_user(&trace);
1180         if (!filter_check_discard(call, entry, buffer, event))
1181                 ring_buffer_unlock_commit(buffer, event);
1182 }
1183
1184 #ifdef UNUSED
1185 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1186 {
1187         ftrace_trace_userstack(tr, flags, preempt_count());
1188 }
1189 #endif /* UNUSED */
1190
1191 #endif /* CONFIG_STACKTRACE */
1192
1193 static void
1194 ftrace_trace_special(void *__tr,
1195                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1196                      int pc)
1197 {
1198         struct ftrace_event_call *call = &event_special;
1199         struct ring_buffer_event *event;
1200         struct trace_array *tr = __tr;
1201         struct ring_buffer *buffer = tr->buffer;
1202         struct special_entry *entry;
1203
1204         event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1205                                           sizeof(*entry), 0, pc);
1206         if (!event)
1207                 return;
1208         entry   = ring_buffer_event_data(event);
1209         entry->arg1                     = arg1;
1210         entry->arg2                     = arg2;
1211         entry->arg3                     = arg3;
1212
1213         if (!filter_check_discard(call, entry, buffer, event))
1214                 trace_buffer_unlock_commit(buffer, event, 0, pc);
1215 }
1216
1217 void
1218 __trace_special(void *__tr, void *__data,
1219                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1220 {
1221         ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1222 }
1223
1224 void
1225 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1226 {
1227         struct trace_array *tr = &global_trace;
1228         struct trace_array_cpu *data;
1229         unsigned long flags;
1230         int cpu;
1231         int pc;
1232
1233         if (tracing_disabled)
1234                 return;
1235
1236         pc = preempt_count();
1237         local_irq_save(flags);
1238         cpu = raw_smp_processor_id();
1239         data = tr->data[cpu];
1240
1241         if (likely(atomic_inc_return(&data->disabled) == 1))
1242                 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1243
1244         atomic_dec(&data->disabled);
1245         local_irq_restore(flags);
1246 }
1247
1248 /**
1249  * trace_vbprintk - write binary msg to tracing buffer
1250  *
1251  */
1252 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1253 {
1254         static raw_spinlock_t trace_buf_lock =
1255                 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1256         static u32 trace_buf[TRACE_BUF_SIZE];
1257
1258         struct ftrace_event_call *call = &event_bprint;
1259         struct ring_buffer_event *event;
1260         struct ring_buffer *buffer;
1261         struct trace_array *tr = &global_trace;
1262         struct trace_array_cpu *data;
1263         struct bprint_entry *entry;
1264         unsigned long flags;
1265         int disable;
1266         int resched;
1267         int cpu, len = 0, size, pc;
1268
1269         if (unlikely(tracing_selftest_running || tracing_disabled))
1270                 return 0;
1271
1272         /* Don't pollute graph traces with trace_vprintk internals */
1273         pause_graph_tracing();
1274
1275         pc = preempt_count();
1276         resched = ftrace_preempt_disable();
1277         cpu = raw_smp_processor_id();
1278         data = tr->data[cpu];
1279
1280         disable = atomic_inc_return(&data->disabled);
1281         if (unlikely(disable != 1))
1282                 goto out;
1283
1284         /* Lockdep uses trace_printk for lock tracing */
1285         local_irq_save(flags);
1286         __raw_spin_lock(&trace_buf_lock);
1287         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1288
1289         if (len > TRACE_BUF_SIZE || len < 0)
1290                 goto out_unlock;
1291
1292         size = sizeof(*entry) + sizeof(u32) * len;
1293         buffer = tr->buffer;
1294         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1295                                           flags, pc);
1296         if (!event)
1297                 goto out_unlock;
1298         entry = ring_buffer_event_data(event);
1299         entry->ip                       = ip;
1300         entry->fmt                      = fmt;
1301
1302         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1303         if (!filter_check_discard(call, entry, buffer, event))
1304                 ring_buffer_unlock_commit(buffer, event);
1305
1306 out_unlock:
1307         __raw_spin_unlock(&trace_buf_lock);
1308         local_irq_restore(flags);
1309
1310 out:
1311         atomic_dec_return(&data->disabled);
1312         ftrace_preempt_enable(resched);
1313         unpause_graph_tracing();
1314
1315         return len;
1316 }
1317 EXPORT_SYMBOL_GPL(trace_vbprintk);
1318
1319 int trace_array_printk(struct trace_array *tr,
1320                        unsigned long ip, const char *fmt, ...)
1321 {
1322         int ret;
1323         va_list ap;
1324
1325         if (!(trace_flags & TRACE_ITER_PRINTK))
1326                 return 0;
1327
1328         va_start(ap, fmt);
1329         ret = trace_array_vprintk(tr, ip, fmt, ap);
1330         va_end(ap);
1331         return ret;
1332 }
1333
1334 int trace_array_vprintk(struct trace_array *tr,
1335                         unsigned long ip, const char *fmt, va_list args)
1336 {
1337         static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1338         static char trace_buf[TRACE_BUF_SIZE];
1339
1340         struct ftrace_event_call *call = &event_print;
1341         struct ring_buffer_event *event;
1342         struct ring_buffer *buffer;
1343         struct trace_array_cpu *data;
1344         int cpu, len = 0, size, pc;
1345         struct print_entry *entry;
1346         unsigned long irq_flags;
1347         int disable;
1348
1349         if (tracing_disabled || tracing_selftest_running)
1350                 return 0;
1351
1352         pc = preempt_count();
1353         preempt_disable_notrace();
1354         cpu = raw_smp_processor_id();
1355         data = tr->data[cpu];
1356
1357         disable = atomic_inc_return(&data->disabled);
1358         if (unlikely(disable != 1))
1359                 goto out;
1360
1361         pause_graph_tracing();
1362         raw_local_irq_save(irq_flags);
1363         __raw_spin_lock(&trace_buf_lock);
1364         if (args == NULL) {
1365                 strncpy(trace_buf, fmt, TRACE_BUF_SIZE);
1366                 len = strlen(trace_buf);
1367         } else
1368                 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1369
1370         size = sizeof(*entry) + len + 1;
1371         buffer = tr->buffer;
1372         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1373                                           irq_flags, pc);
1374         if (!event)
1375                 goto out_unlock;
1376         entry = ring_buffer_event_data(event);
1377         entry->ip = ip;
1378
1379         memcpy(&entry->buf, trace_buf, len);
1380         entry->buf[len] = '\0';
1381         if (!filter_check_discard(call, entry, buffer, event))
1382                 ring_buffer_unlock_commit(buffer, event);
1383
1384  out_unlock:
1385         __raw_spin_unlock(&trace_buf_lock);
1386         raw_local_irq_restore(irq_flags);
1387         unpause_graph_tracing();
1388  out:
1389         atomic_dec_return(&data->disabled);
1390         preempt_enable_notrace();
1391
1392         return len;
1393 }
1394
1395 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1396 {
1397         return trace_array_vprintk(&global_trace, ip, fmt, args);
1398 }
1399 EXPORT_SYMBOL_GPL(trace_vprintk);
1400
1401 enum trace_file_type {
1402         TRACE_FILE_LAT_FMT      = 1,
1403         TRACE_FILE_ANNOTATE     = 2,
1404 };
1405
1406 static void trace_iterator_increment(struct trace_iterator *iter)
1407 {
1408         /* Don't allow ftrace to trace into the ring buffers */
1409         ftrace_disable_cpu();
1410
1411         iter->idx++;
1412         if (iter->buffer_iter[iter->cpu])
1413                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1414
1415         ftrace_enable_cpu();
1416 }
1417
1418 static struct trace_entry *
1419 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1420 {
1421         struct ring_buffer_event *event;
1422         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1423
1424         /* Don't allow ftrace to trace into the ring buffers */
1425         ftrace_disable_cpu();
1426
1427         if (buf_iter)
1428                 event = ring_buffer_iter_peek(buf_iter, ts);
1429         else
1430                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1431
1432         ftrace_enable_cpu();
1433
1434         return event ? ring_buffer_event_data(event) : NULL;
1435 }
1436
1437 static struct trace_entry *
1438 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1439 {
1440         struct ring_buffer *buffer = iter->tr->buffer;
1441         struct trace_entry *ent, *next = NULL;
1442         int cpu_file = iter->cpu_file;
1443         u64 next_ts = 0, ts;
1444         int next_cpu = -1;
1445         int cpu;
1446
1447         /*
1448          * If we are in a per_cpu trace file, don't bother by iterating over
1449          * all cpu and peek directly.
1450          */
1451         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1452                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1453                         return NULL;
1454                 ent = peek_next_entry(iter, cpu_file, ent_ts);
1455                 if (ent_cpu)
1456                         *ent_cpu = cpu_file;
1457
1458                 return ent;
1459         }
1460
1461         for_each_tracing_cpu(cpu) {
1462
1463                 if (ring_buffer_empty_cpu(buffer, cpu))
1464                         continue;
1465
1466                 ent = peek_next_entry(iter, cpu, &ts);
1467
1468                 /*
1469                  * Pick the entry with the smallest timestamp:
1470                  */
1471                 if (ent && (!next || ts < next_ts)) {
1472                         next = ent;
1473                         next_cpu = cpu;
1474                         next_ts = ts;
1475                 }
1476         }
1477
1478         if (ent_cpu)
1479                 *ent_cpu = next_cpu;
1480
1481         if (ent_ts)
1482                 *ent_ts = next_ts;
1483
1484         return next;
1485 }
1486
1487 /* Find the next real entry, without updating the iterator itself */
1488 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1489                                           int *ent_cpu, u64 *ent_ts)
1490 {
1491         return __find_next_entry(iter, ent_cpu, ent_ts);
1492 }
1493
1494 /* Find the next real entry, and increment the iterator to the next entry */
1495 static void *find_next_entry_inc(struct trace_iterator *iter)
1496 {
1497         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1498
1499         if (iter->ent)
1500                 trace_iterator_increment(iter);
1501
1502         return iter->ent ? iter : NULL;
1503 }
1504
1505 static void trace_consume(struct trace_iterator *iter)
1506 {
1507         /* Don't allow ftrace to trace into the ring buffers */
1508         ftrace_disable_cpu();
1509         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1510         ftrace_enable_cpu();
1511 }
1512
1513 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1514 {
1515         struct trace_iterator *iter = m->private;
1516         int i = (int)*pos;
1517         void *ent;
1518
1519         (*pos)++;
1520
1521         /* can't go backwards */
1522         if (iter->idx > i)
1523                 return NULL;
1524
1525         if (iter->idx < 0)
1526                 ent = find_next_entry_inc(iter);
1527         else
1528                 ent = iter;
1529
1530         while (ent && iter->idx < i)
1531                 ent = find_next_entry_inc(iter);
1532
1533         iter->pos = *pos;
1534
1535         return ent;
1536 }
1537
1538 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1539 {
1540         struct trace_array *tr = iter->tr;
1541         struct ring_buffer_event *event;
1542         struct ring_buffer_iter *buf_iter;
1543         unsigned long entries = 0;
1544         u64 ts;
1545
1546         tr->data[cpu]->skipped_entries = 0;
1547
1548         if (!iter->buffer_iter[cpu])
1549                 return;
1550
1551         buf_iter = iter->buffer_iter[cpu];
1552         ring_buffer_iter_reset(buf_iter);
1553
1554         /*
1555          * We could have the case with the max latency tracers
1556          * that a reset never took place on a cpu. This is evident
1557          * by the timestamp being before the start of the buffer.
1558          */
1559         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1560                 if (ts >= iter->tr->time_start)
1561                         break;
1562                 entries++;
1563                 ring_buffer_read(buf_iter, NULL);
1564         }
1565
1566         tr->data[cpu]->skipped_entries = entries;
1567 }
1568
1569 /*
1570  * No necessary locking here. The worst thing which can
1571  * happen is loosing events consumed at the same time
1572  * by a trace_pipe reader.
1573  * Other than that, we don't risk to crash the ring buffer
1574  * because it serializes the readers.
1575  *
1576  * The current tracer is copied to avoid a global locking
1577  * all around.
1578  */
1579 static void *s_start(struct seq_file *m, loff_t *pos)
1580 {
1581         struct trace_iterator *iter = m->private;
1582         static struct tracer *old_tracer;
1583         int cpu_file = iter->cpu_file;
1584         void *p = NULL;
1585         loff_t l = 0;
1586         int cpu;
1587
1588         /* copy the tracer to avoid using a global lock all around */
1589         mutex_lock(&trace_types_lock);
1590         if (unlikely(old_tracer != current_trace && current_trace)) {
1591                 old_tracer = current_trace;
1592                 *iter->trace = *current_trace;
1593         }
1594         mutex_unlock(&trace_types_lock);
1595
1596         atomic_inc(&trace_record_cmdline_disabled);
1597
1598         if (*pos != iter->pos) {
1599                 iter->ent = NULL;
1600                 iter->cpu = 0;
1601                 iter->idx = -1;
1602
1603                 ftrace_disable_cpu();
1604
1605                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1606                         for_each_tracing_cpu(cpu)
1607                                 tracing_iter_reset(iter, cpu);
1608                 } else
1609                         tracing_iter_reset(iter, cpu_file);
1610
1611                 ftrace_enable_cpu();
1612
1613                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1614                         ;
1615
1616         } else {
1617                 l = *pos - 1;
1618                 p = s_next(m, p, &l);
1619         }
1620
1621         trace_event_read_lock();
1622         return p;
1623 }
1624
1625 static void s_stop(struct seq_file *m, void *p)
1626 {
1627         atomic_dec(&trace_record_cmdline_disabled);
1628         trace_event_read_unlock();
1629 }
1630
1631 static void print_lat_help_header(struct seq_file *m)
1632 {
1633         seq_puts(m, "#                  _------=> CPU#            \n");
1634         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1635         seq_puts(m, "#                | / _----=> need-resched    \n");
1636         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1637         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1638         seq_puts(m, "#                |||| /_--=> lock-depth       \n");
1639         seq_puts(m, "#                |||||/     delay             \n");
1640         seq_puts(m, "#  cmd     pid   |||||| time  |   caller      \n");
1641         seq_puts(m, "#     \\   /      ||||||   \\   |   /           \n");
1642 }
1643
1644 static void print_func_help_header(struct seq_file *m)
1645 {
1646         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1647         seq_puts(m, "#              | |       |          |         |\n");
1648 }
1649
1650
1651 static void
1652 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1653 {
1654         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1655         struct trace_array *tr = iter->tr;
1656         struct trace_array_cpu *data = tr->data[tr->cpu];
1657         struct tracer *type = current_trace;
1658         unsigned long entries = 0;
1659         unsigned long total = 0;
1660         unsigned long count;
1661         const char *name = "preemption";
1662         int cpu;
1663
1664         if (type)
1665                 name = type->name;
1666
1667
1668         for_each_tracing_cpu(cpu) {
1669                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1670                 /*
1671                  * If this buffer has skipped entries, then we hold all
1672                  * entries for the trace and we need to ignore the
1673                  * ones before the time stamp.
1674                  */
1675                 if (tr->data[cpu]->skipped_entries) {
1676                         count -= tr->data[cpu]->skipped_entries;
1677                         /* total is the same as the entries */
1678                         total += count;
1679                 } else
1680                         total += count +
1681                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1682                 entries += count;
1683         }
1684
1685         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1686                    name, UTS_RELEASE);
1687         seq_puts(m, "# -----------------------------------"
1688                  "---------------------------------\n");
1689         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1690                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1691                    nsecs_to_usecs(data->saved_latency),
1692                    entries,
1693                    total,
1694                    tr->cpu,
1695 #if defined(CONFIG_PREEMPT_NONE)
1696                    "server",
1697 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1698                    "desktop",
1699 #elif defined(CONFIG_PREEMPT)
1700                    "preempt",
1701 #else
1702                    "unknown",
1703 #endif
1704                    /* These are reserved for later use */
1705                    0, 0, 0, 0);
1706 #ifdef CONFIG_SMP
1707         seq_printf(m, " #P:%d)\n", num_online_cpus());
1708 #else
1709         seq_puts(m, ")\n");
1710 #endif
1711         seq_puts(m, "#    -----------------\n");
1712         seq_printf(m, "#    | task: %.16s-%d "
1713                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1714                    data->comm, data->pid, data->uid, data->nice,
1715                    data->policy, data->rt_priority);
1716         seq_puts(m, "#    -----------------\n");
1717
1718         if (data->critical_start) {
1719                 seq_puts(m, "#  => started at: ");
1720                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1721                 trace_print_seq(m, &iter->seq);
1722                 seq_puts(m, "\n#  => ended at:   ");
1723                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1724                 trace_print_seq(m, &iter->seq);
1725                 seq_puts(m, "\n#\n");
1726         }
1727
1728         seq_puts(m, "#\n");
1729 }
1730
1731 static void test_cpu_buff_start(struct trace_iterator *iter)
1732 {
1733         struct trace_seq *s = &iter->seq;
1734
1735         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1736                 return;
1737
1738         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1739                 return;
1740
1741         if (cpumask_test_cpu(iter->cpu, iter->started))
1742                 return;
1743
1744         if (iter->tr->data[iter->cpu]->skipped_entries)
1745                 return;
1746
1747         cpumask_set_cpu(iter->cpu, iter->started);
1748
1749         /* Don't print started cpu buffer for the first entry of the trace */
1750         if (iter->idx > 1)
1751                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1752                                 iter->cpu);
1753 }
1754
1755 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1756 {
1757         struct trace_seq *s = &iter->seq;
1758         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1759         struct trace_entry *entry;
1760         struct trace_event *event;
1761
1762         entry = iter->ent;
1763
1764         test_cpu_buff_start(iter);
1765
1766         event = ftrace_find_event(entry->type);
1767
1768         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1769                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1770                         if (!trace_print_lat_context(iter))
1771                                 goto partial;
1772                 } else {
1773                         if (!trace_print_context(iter))
1774                                 goto partial;
1775                 }
1776         }
1777
1778         if (event)
1779                 return event->trace(iter, sym_flags);
1780
1781         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1782                 goto partial;
1783
1784         return TRACE_TYPE_HANDLED;
1785 partial:
1786         return TRACE_TYPE_PARTIAL_LINE;
1787 }
1788
1789 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1790 {
1791         struct trace_seq *s = &iter->seq;
1792         struct trace_entry *entry;
1793         struct trace_event *event;
1794
1795         entry = iter->ent;
1796
1797         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1798                 if (!trace_seq_printf(s, "%d %d %llu ",
1799                                       entry->pid, iter->cpu, iter->ts))
1800                         goto partial;
1801         }
1802
1803         event = ftrace_find_event(entry->type);
1804         if (event)
1805                 return event->raw(iter, 0);
1806
1807         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1808                 goto partial;
1809
1810         return TRACE_TYPE_HANDLED;
1811 partial:
1812         return TRACE_TYPE_PARTIAL_LINE;
1813 }
1814
1815 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1816 {
1817         struct trace_seq *s = &iter->seq;
1818         unsigned char newline = '\n';
1819         struct trace_entry *entry;
1820         struct trace_event *event;
1821
1822         entry = iter->ent;
1823
1824         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1825                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1826                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1827                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1828         }
1829
1830         event = ftrace_find_event(entry->type);
1831         if (event) {
1832                 enum print_line_t ret = event->hex(iter, 0);
1833                 if (ret != TRACE_TYPE_HANDLED)
1834                         return ret;
1835         }
1836
1837         SEQ_PUT_FIELD_RET(s, newline);
1838
1839         return TRACE_TYPE_HANDLED;
1840 }
1841
1842 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1843 {
1844         struct trace_seq *s = &iter->seq;
1845         struct trace_entry *entry;
1846         struct trace_event *event;
1847
1848         entry = iter->ent;
1849
1850         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1851                 SEQ_PUT_FIELD_RET(s, entry->pid);
1852                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1853                 SEQ_PUT_FIELD_RET(s, iter->ts);
1854         }
1855
1856         event = ftrace_find_event(entry->type);
1857         return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1858 }
1859
1860 static int trace_empty(struct trace_iterator *iter)
1861 {
1862         int cpu;
1863
1864         /* If we are looking at one CPU buffer, only check that one */
1865         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1866                 cpu = iter->cpu_file;
1867                 if (iter->buffer_iter[cpu]) {
1868                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1869                                 return 0;
1870                 } else {
1871                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1872                                 return 0;
1873                 }
1874                 return 1;
1875         }
1876
1877         for_each_tracing_cpu(cpu) {
1878                 if (iter->buffer_iter[cpu]) {
1879                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1880                                 return 0;
1881                 } else {
1882                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1883                                 return 0;
1884                 }
1885         }
1886
1887         return 1;
1888 }
1889
1890 /*  Called with trace_event_read_lock() held. */
1891 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1892 {
1893         enum print_line_t ret;
1894
1895         if (iter->trace && iter->trace->print_line) {
1896                 ret = iter->trace->print_line(iter);
1897                 if (ret != TRACE_TYPE_UNHANDLED)
1898                         return ret;
1899         }
1900
1901         if (iter->ent->type == TRACE_BPRINT &&
1902                         trace_flags & TRACE_ITER_PRINTK &&
1903                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1904                 return trace_print_bprintk_msg_only(iter);
1905
1906         if (iter->ent->type == TRACE_PRINT &&
1907                         trace_flags & TRACE_ITER_PRINTK &&
1908                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1909                 return trace_print_printk_msg_only(iter);
1910
1911         if (trace_flags & TRACE_ITER_BIN)
1912                 return print_bin_fmt(iter);
1913
1914         if (trace_flags & TRACE_ITER_HEX)
1915                 return print_hex_fmt(iter);
1916
1917         if (trace_flags & TRACE_ITER_RAW)
1918                 return print_raw_fmt(iter);
1919
1920         return print_trace_fmt(iter);
1921 }
1922
1923 static int s_show(struct seq_file *m, void *v)
1924 {
1925         struct trace_iterator *iter = v;
1926
1927         if (iter->ent == NULL) {
1928                 if (iter->tr) {
1929                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1930                         seq_puts(m, "#\n");
1931                 }
1932                 if (iter->trace && iter->trace->print_header)
1933                         iter->trace->print_header(m);
1934                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1935                         /* print nothing if the buffers are empty */
1936                         if (trace_empty(iter))
1937                                 return 0;
1938                         print_trace_header(m, iter);
1939                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1940                                 print_lat_help_header(m);
1941                 } else {
1942                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1943                                 print_func_help_header(m);
1944                 }
1945         } else {
1946                 print_trace_line(iter);
1947                 trace_print_seq(m, &iter->seq);
1948         }
1949
1950         return 0;
1951 }
1952
1953 static const struct seq_operations tracer_seq_ops = {
1954         .start          = s_start,
1955         .next           = s_next,
1956         .stop           = s_stop,
1957         .show           = s_show,
1958 };
1959
1960 static struct trace_iterator *
1961 __tracing_open(struct inode *inode, struct file *file)
1962 {
1963         long cpu_file = (long) inode->i_private;
1964         void *fail_ret = ERR_PTR(-ENOMEM);
1965         struct trace_iterator *iter;
1966         struct seq_file *m;
1967         int cpu, ret;
1968
1969         if (tracing_disabled)
1970                 return ERR_PTR(-ENODEV);
1971
1972         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1973         if (!iter)
1974                 return ERR_PTR(-ENOMEM);
1975
1976         /*
1977          * We make a copy of the current tracer to avoid concurrent
1978          * changes on it while we are reading.
1979          */
1980         mutex_lock(&trace_types_lock);
1981         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1982         if (!iter->trace)
1983                 goto fail;
1984
1985         if (current_trace)
1986                 *iter->trace = *current_trace;
1987
1988         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
1989                 goto fail;
1990
1991         if (current_trace && current_trace->print_max)
1992                 iter->tr = &max_tr;
1993         else
1994                 iter->tr = &global_trace;
1995         iter->pos = -1;
1996         mutex_init(&iter->mutex);
1997         iter->cpu_file = cpu_file;
1998
1999         /* Notify the tracer early; before we stop tracing. */
2000         if (iter->trace && iter->trace->open)
2001                 iter->trace->open(iter);
2002
2003         /* Annotate start of buffers if we had overruns */
2004         if (ring_buffer_overruns(iter->tr->buffer))
2005                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2006
2007         /* stop the trace while dumping */
2008         tracing_stop();
2009
2010         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2011                 for_each_tracing_cpu(cpu) {
2012
2013                         iter->buffer_iter[cpu] =
2014                                 ring_buffer_read_start(iter->tr->buffer, cpu);
2015                         tracing_iter_reset(iter, cpu);
2016                 }
2017         } else {
2018                 cpu = iter->cpu_file;
2019                 iter->buffer_iter[cpu] =
2020                                 ring_buffer_read_start(iter->tr->buffer, cpu);
2021                 tracing_iter_reset(iter, cpu);
2022         }
2023
2024         ret = seq_open(file, &tracer_seq_ops);
2025         if (ret < 0) {
2026                 fail_ret = ERR_PTR(ret);
2027                 goto fail_buffer;
2028         }
2029
2030         m = file->private_data;
2031         m->private = iter;
2032
2033         mutex_unlock(&trace_types_lock);
2034
2035         return iter;
2036
2037  fail_buffer:
2038         for_each_tracing_cpu(cpu) {
2039                 if (iter->buffer_iter[cpu])
2040                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2041         }
2042         free_cpumask_var(iter->started);
2043         tracing_start();
2044  fail:
2045         mutex_unlock(&trace_types_lock);
2046         kfree(iter->trace);
2047         kfree(iter);
2048
2049         return fail_ret;
2050 }
2051
2052 int tracing_open_generic(struct inode *inode, struct file *filp)
2053 {
2054         if (tracing_disabled)
2055                 return -ENODEV;
2056
2057         filp->private_data = inode->i_private;
2058         return 0;
2059 }
2060
2061 static int tracing_release(struct inode *inode, struct file *file)
2062 {
2063         struct seq_file *m = (struct seq_file *)file->private_data;
2064         struct trace_iterator *iter;
2065         int cpu;
2066
2067         if (!(file->f_mode & FMODE_READ))
2068                 return 0;
2069
2070         iter = m->private;
2071
2072         mutex_lock(&trace_types_lock);
2073         for_each_tracing_cpu(cpu) {
2074                 if (iter->buffer_iter[cpu])
2075                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2076         }
2077
2078         if (iter->trace && iter->trace->close)
2079                 iter->trace->close(iter);
2080
2081         /* reenable tracing if it was previously enabled */
2082         tracing_start();
2083         mutex_unlock(&trace_types_lock);
2084
2085         seq_release(inode, file);
2086         mutex_destroy(&iter->mutex);
2087         free_cpumask_var(iter->started);
2088         kfree(iter->trace);
2089         kfree(iter);
2090         return 0;
2091 }
2092
2093 static int tracing_open(struct inode *inode, struct file *file)
2094 {
2095         struct trace_iterator *iter;
2096         int ret = 0;
2097
2098         /* If this file was open for write, then erase contents */
2099         if ((file->f_mode & FMODE_WRITE) &&
2100             (file->f_flags & O_TRUNC)) {
2101                 long cpu = (long) inode->i_private;
2102
2103                 if (cpu == TRACE_PIPE_ALL_CPU)
2104                         tracing_reset_online_cpus(&global_trace);
2105                 else
2106                         tracing_reset(&global_trace, cpu);
2107         }
2108
2109         if (file->f_mode & FMODE_READ) {
2110                 iter = __tracing_open(inode, file);
2111                 if (IS_ERR(iter))
2112                         ret = PTR_ERR(iter);
2113                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2114                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2115         }
2116         return ret;
2117 }
2118
2119 static void *
2120 t_next(struct seq_file *m, void *v, loff_t *pos)
2121 {
2122         struct tracer *t = v;
2123
2124         (*pos)++;
2125
2126         if (t)
2127                 t = t->next;
2128
2129         return t;
2130 }
2131
2132 static void *t_start(struct seq_file *m, loff_t *pos)
2133 {
2134         struct tracer *t;
2135         loff_t l = 0;
2136
2137         mutex_lock(&trace_types_lock);
2138         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2139                 ;
2140
2141         return t;
2142 }
2143
2144 static void t_stop(struct seq_file *m, void *p)
2145 {
2146         mutex_unlock(&trace_types_lock);
2147 }
2148
2149 static int t_show(struct seq_file *m, void *v)
2150 {
2151         struct tracer *t = v;
2152
2153         if (!t)
2154                 return 0;
2155
2156         seq_printf(m, "%s", t->name);
2157         if (t->next)
2158                 seq_putc(m, ' ');
2159         else
2160                 seq_putc(m, '\n');
2161
2162         return 0;
2163 }
2164
2165 static const struct seq_operations show_traces_seq_ops = {
2166         .start          = t_start,
2167         .next           = t_next,
2168         .stop           = t_stop,
2169         .show           = t_show,
2170 };
2171
2172 static int show_traces_open(struct inode *inode, struct file *file)
2173 {
2174         if (tracing_disabled)
2175                 return -ENODEV;
2176
2177         return seq_open(file, &show_traces_seq_ops);
2178 }
2179
2180 static ssize_t
2181 tracing_write_stub(struct file *filp, const char __user *ubuf,
2182                    size_t count, loff_t *ppos)
2183 {
2184         return count;
2185 }
2186
2187 static const struct file_operations tracing_fops = {
2188         .open           = tracing_open,
2189         .read           = seq_read,
2190         .write          = tracing_write_stub,
2191         .llseek         = seq_lseek,
2192         .release        = tracing_release,
2193 };
2194
2195 static const struct file_operations show_traces_fops = {
2196         .open           = show_traces_open,
2197         .read           = seq_read,
2198         .release        = seq_release,
2199 };
2200
2201 /*
2202  * Only trace on a CPU if the bitmask is set:
2203  */
2204 static cpumask_var_t tracing_cpumask;
2205
2206 /*
2207  * The tracer itself will not take this lock, but still we want
2208  * to provide a consistent cpumask to user-space:
2209  */
2210 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2211
2212 /*
2213  * Temporary storage for the character representation of the
2214  * CPU bitmask (and one more byte for the newline):
2215  */
2216 static char mask_str[NR_CPUS + 1];
2217
2218 static ssize_t
2219 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2220                      size_t count, loff_t *ppos)
2221 {
2222         int len;
2223
2224         mutex_lock(&tracing_cpumask_update_lock);
2225
2226         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2227         if (count - len < 2) {
2228                 count = -EINVAL;
2229                 goto out_err;
2230         }
2231         len += sprintf(mask_str + len, "\n");
2232         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2233
2234 out_err:
2235         mutex_unlock(&tracing_cpumask_update_lock);
2236
2237         return count;
2238 }
2239
2240 static ssize_t
2241 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2242                       size_t count, loff_t *ppos)
2243 {
2244         int err, cpu;
2245         cpumask_var_t tracing_cpumask_new;
2246
2247         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2248                 return -ENOMEM;
2249
2250         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2251         if (err)
2252                 goto err_unlock;
2253
2254         mutex_lock(&tracing_cpumask_update_lock);
2255
2256         local_irq_disable();
2257         __raw_spin_lock(&ftrace_max_lock);
2258         for_each_tracing_cpu(cpu) {
2259                 /*
2260                  * Increase/decrease the disabled counter if we are
2261                  * about to flip a bit in the cpumask:
2262                  */
2263                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2264                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2265                         atomic_inc(&global_trace.data[cpu]->disabled);
2266                 }
2267                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2268                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2269                         atomic_dec(&global_trace.data[cpu]->disabled);
2270                 }
2271         }
2272         __raw_spin_unlock(&ftrace_max_lock);
2273         local_irq_enable();
2274
2275         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2276
2277         mutex_unlock(&tracing_cpumask_update_lock);
2278         free_cpumask_var(tracing_cpumask_new);
2279
2280         return count;
2281
2282 err_unlock:
2283         free_cpumask_var(tracing_cpumask_new);
2284
2285         return err;
2286 }
2287
2288 static const struct file_operations tracing_cpumask_fops = {
2289         .open           = tracing_open_generic,
2290         .read           = tracing_cpumask_read,
2291         .write          = tracing_cpumask_write,
2292 };
2293
2294 static ssize_t
2295 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2296                        size_t cnt, loff_t *ppos)
2297 {
2298         struct tracer_opt *trace_opts;
2299         u32 tracer_flags;
2300         int len = 0;
2301         char *buf;
2302         int r = 0;
2303         int i;
2304
2305
2306         /* calculate max size */
2307         for (i = 0; trace_options[i]; i++) {
2308                 len += strlen(trace_options[i]);
2309                 len += 3; /* "no" and newline */
2310         }
2311
2312         mutex_lock(&trace_types_lock);
2313         tracer_flags = current_trace->flags->val;
2314         trace_opts = current_trace->flags->opts;
2315
2316         /*
2317          * Increase the size with names of options specific
2318          * of the current tracer.
2319          */
2320         for (i = 0; trace_opts[i].name; i++) {
2321                 len += strlen(trace_opts[i].name);
2322                 len += 3; /* "no" and newline */
2323         }
2324
2325         /* +1 for \0 */
2326         buf = kmalloc(len + 1, GFP_KERNEL);
2327         if (!buf) {
2328                 mutex_unlock(&trace_types_lock);
2329                 return -ENOMEM;
2330         }
2331
2332         for (i = 0; trace_options[i]; i++) {
2333                 if (trace_flags & (1 << i))
2334                         r += sprintf(buf + r, "%s\n", trace_options[i]);
2335                 else
2336                         r += sprintf(buf + r, "no%s\n", trace_options[i]);
2337         }
2338
2339         for (i = 0; trace_opts[i].name; i++) {
2340                 if (tracer_flags & trace_opts[i].bit)
2341                         r += sprintf(buf + r, "%s\n",
2342                                 trace_opts[i].name);
2343                 else
2344                         r += sprintf(buf + r, "no%s\n",
2345                                 trace_opts[i].name);
2346         }
2347         mutex_unlock(&trace_types_lock);
2348
2349         WARN_ON(r >= len + 1);
2350
2351         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2352
2353         kfree(buf);
2354         return r;
2355 }
2356
2357 /* Try to assign a tracer specific option */
2358 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2359 {
2360         struct tracer_flags *tracer_flags = trace->flags;
2361         struct tracer_opt *opts = NULL;
2362         int ret = 0, i = 0;
2363         int len;
2364
2365         for (i = 0; tracer_flags->opts[i].name; i++) {
2366                 opts = &tracer_flags->opts[i];
2367                 len = strlen(opts->name);
2368
2369                 if (strncmp(cmp, opts->name, len) == 0) {
2370                         ret = trace->set_flag(tracer_flags->val,
2371                                 opts->bit, !neg);
2372                         break;
2373                 }
2374         }
2375         /* Not found */
2376         if (!tracer_flags->opts[i].name)
2377                 return -EINVAL;
2378
2379         /* Refused to handle */
2380         if (ret)
2381                 return ret;
2382
2383         if (neg)
2384                 tracer_flags->val &= ~opts->bit;
2385         else
2386                 tracer_flags->val |= opts->bit;
2387
2388         return 0;
2389 }
2390
2391 static void set_tracer_flags(unsigned int mask, int enabled)
2392 {
2393         /* do nothing if flag is already set */
2394         if (!!(trace_flags & mask) == !!enabled)
2395                 return;
2396
2397         if (enabled)
2398                 trace_flags |= mask;
2399         else
2400                 trace_flags &= ~mask;
2401 }
2402
2403 static ssize_t
2404 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2405                         size_t cnt, loff_t *ppos)
2406 {
2407         char buf[64];
2408         char *cmp = buf;
2409         int neg = 0;
2410         int ret;
2411         int i;
2412
2413         if (cnt >= sizeof(buf))
2414                 return -EINVAL;
2415
2416         if (copy_from_user(&buf, ubuf, cnt))
2417                 return -EFAULT;
2418
2419         buf[cnt] = 0;
2420
2421         if (strncmp(buf, "no", 2) == 0) {
2422                 neg = 1;
2423                 cmp += 2;
2424         }
2425
2426         for (i = 0; trace_options[i]; i++) {
2427                 int len = strlen(trace_options[i]);
2428
2429                 if (strncmp(cmp, trace_options[i], len) == 0) {
2430                         set_tracer_flags(1 << i, !neg);
2431                         break;
2432                 }
2433         }
2434
2435         /* If no option could be set, test the specific tracer options */
2436         if (!trace_options[i]) {
2437                 mutex_lock(&trace_types_lock);
2438                 ret = set_tracer_option(current_trace, cmp, neg);
2439                 mutex_unlock(&trace_types_lock);
2440                 if (ret)
2441                         return ret;
2442         }
2443
2444         *ppos += cnt;
2445
2446         return cnt;
2447 }
2448
2449 static const struct file_operations tracing_iter_fops = {
2450         .open           = tracing_open_generic,
2451         .read           = tracing_trace_options_read,
2452         .write          = tracing_trace_options_write,
2453 };
2454
2455 static const char readme_msg[] =
2456         "tracing mini-HOWTO:\n\n"
2457         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2458         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2459         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2460         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2461         "nop\n"
2462         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2463         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2464         "sched_switch\n"
2465         "# cat /sys/kernel/debug/tracing/trace_options\n"
2466         "noprint-parent nosym-offset nosym-addr noverbose\n"
2467         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2468         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2469         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2470         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2471 ;
2472
2473 static ssize_t
2474 tracing_readme_read(struct file *filp, char __user *ubuf,
2475                        size_t cnt, loff_t *ppos)
2476 {
2477         return simple_read_from_buffer(ubuf, cnt, ppos,
2478                                         readme_msg, strlen(readme_msg));
2479 }
2480
2481 static const struct file_operations tracing_readme_fops = {
2482         .open           = tracing_open_generic,
2483         .read           = tracing_readme_read,
2484 };
2485
2486 static ssize_t
2487 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2488                                 size_t cnt, loff_t *ppos)
2489 {
2490         char *buf_comm;
2491         char *file_buf;
2492         char *buf;
2493         int len = 0;
2494         int pid;
2495         int i;
2496
2497         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2498         if (!file_buf)
2499                 return -ENOMEM;
2500
2501         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2502         if (!buf_comm) {
2503                 kfree(file_buf);
2504                 return -ENOMEM;
2505         }
2506
2507         buf = file_buf;
2508
2509         for (i = 0; i < SAVED_CMDLINES; i++) {
2510                 int r;
2511
2512                 pid = map_cmdline_to_pid[i];
2513                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2514                         continue;
2515
2516                 trace_find_cmdline(pid, buf_comm);
2517                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2518                 buf += r;
2519                 len += r;
2520         }
2521
2522         len = simple_read_from_buffer(ubuf, cnt, ppos,
2523                                       file_buf, len);
2524
2525         kfree(file_buf);
2526         kfree(buf_comm);
2527
2528         return len;
2529 }
2530
2531 static const struct file_operations tracing_saved_cmdlines_fops = {
2532     .open       = tracing_open_generic,
2533     .read       = tracing_saved_cmdlines_read,
2534 };
2535
2536 static ssize_t
2537 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2538                   size_t cnt, loff_t *ppos)
2539 {
2540         char buf[64];
2541         int r;
2542
2543         r = sprintf(buf, "%u\n", tracer_enabled);
2544         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2545 }
2546
2547 static ssize_t
2548 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2549                    size_t cnt, loff_t *ppos)
2550 {
2551         struct trace_array *tr = filp->private_data;
2552         char buf[64];
2553         unsigned long val;
2554         int ret;
2555
2556         if (cnt >= sizeof(buf))
2557                 return -EINVAL;
2558
2559         if (copy_from_user(&buf, ubuf, cnt))
2560                 return -EFAULT;
2561
2562         buf[cnt] = 0;
2563
2564         ret = strict_strtoul(buf, 10, &val);
2565         if (ret < 0)
2566                 return ret;
2567
2568         val = !!val;
2569
2570         mutex_lock(&trace_types_lock);
2571         if (tracer_enabled ^ val) {
2572                 if (val) {
2573                         tracer_enabled = 1;
2574                         if (current_trace->start)
2575                                 current_trace->start(tr);
2576                         tracing_start();
2577                 } else {
2578                         tracer_enabled = 0;
2579                         tracing_stop();
2580                         if (current_trace->stop)
2581                                 current_trace->stop(tr);
2582                 }
2583         }
2584         mutex_unlock(&trace_types_lock);
2585
2586         *ppos += cnt;
2587
2588         return cnt;
2589 }
2590
2591 static ssize_t
2592 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2593                        size_t cnt, loff_t *ppos)
2594 {
2595         char buf[MAX_TRACER_SIZE+2];
2596         int r;
2597
2598         mutex_lock(&trace_types_lock);
2599         if (current_trace)
2600                 r = sprintf(buf, "%s\n", current_trace->name);
2601         else
2602                 r = sprintf(buf, "\n");
2603         mutex_unlock(&trace_types_lock);
2604
2605         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2606 }
2607
2608 int tracer_init(struct tracer *t, struct trace_array *tr)
2609 {
2610         tracing_reset_online_cpus(tr);
2611         return t->init(tr);
2612 }
2613
2614 static int tracing_resize_ring_buffer(unsigned long size)
2615 {
2616         int ret;
2617
2618         /*
2619          * If kernel or user changes the size of the ring buffer
2620          * we use the size that was given, and we can forget about
2621          * expanding it later.
2622          */
2623         ring_buffer_expanded = 1;
2624
2625         ret = ring_buffer_resize(global_trace.buffer, size);
2626         if (ret < 0)
2627                 return ret;
2628
2629         ret = ring_buffer_resize(max_tr.buffer, size);
2630         if (ret < 0) {
2631                 int r;
2632
2633                 r = ring_buffer_resize(global_trace.buffer,
2634                                        global_trace.entries);
2635                 if (r < 0) {
2636                         /*
2637                          * AARGH! We are left with different
2638                          * size max buffer!!!!
2639                          * The max buffer is our "snapshot" buffer.
2640                          * When a tracer needs a snapshot (one of the
2641                          * latency tracers), it swaps the max buffer
2642                          * with the saved snap shot. We succeeded to
2643                          * update the size of the main buffer, but failed to
2644                          * update the size of the max buffer. But when we tried
2645                          * to reset the main buffer to the original size, we
2646                          * failed there too. This is very unlikely to
2647                          * happen, but if it does, warn and kill all
2648                          * tracing.
2649                          */
2650                         WARN_ON(1);
2651                         tracing_disabled = 1;
2652                 }
2653                 return ret;
2654         }
2655
2656         global_trace.entries = size;
2657
2658         return ret;
2659 }
2660
2661 /**
2662  * tracing_update_buffers - used by tracing facility to expand ring buffers
2663  *
2664  * To save on memory when the tracing is never used on a system with it
2665  * configured in. The ring buffers are set to a minimum size. But once
2666  * a user starts to use the tracing facility, then they need to grow
2667  * to their default size.
2668  *
2669  * This function is to be called when a tracer is about to be used.
2670  */
2671 int tracing_update_buffers(void)
2672 {
2673         int ret = 0;
2674
2675         mutex_lock(&trace_types_lock);
2676         if (!ring_buffer_expanded)
2677                 ret = tracing_resize_ring_buffer(trace_buf_size);
2678         mutex_unlock(&trace_types_lock);
2679
2680         return ret;
2681 }
2682
2683 struct trace_option_dentry;
2684
2685 static struct trace_option_dentry *
2686 create_trace_option_files(struct tracer *tracer);
2687
2688 static void
2689 destroy_trace_option_files(struct trace_option_dentry *topts);
2690
2691 static int tracing_set_tracer(const char *buf)
2692 {
2693         static struct trace_option_dentry *topts;
2694         struct trace_array *tr = &global_trace;
2695         struct tracer *t;
2696         int ret = 0;
2697
2698         mutex_lock(&trace_types_lock);
2699
2700         if (!ring_buffer_expanded) {
2701                 ret = tracing_resize_ring_buffer(trace_buf_size);
2702                 if (ret < 0)
2703                         goto out;
2704                 ret = 0;
2705         }
2706
2707         for (t = trace_types; t; t = t->next) {
2708                 if (strcmp(t->name, buf) == 0)
2709                         break;
2710         }
2711         if (!t) {
2712                 ret = -EINVAL;
2713                 goto out;
2714         }
2715         if (t == current_trace)
2716                 goto out;
2717
2718         trace_branch_disable();
2719         if (current_trace && current_trace->reset)
2720                 current_trace->reset(tr);
2721
2722         destroy_trace_option_files(topts);
2723
2724         current_trace = t;
2725
2726         topts = create_trace_option_files(current_trace);
2727
2728         if (t->init) {
2729                 ret = tracer_init(t, tr);
2730                 if (ret)
2731                         goto out;
2732         }
2733
2734         trace_branch_enable(tr);
2735  out:
2736         mutex_unlock(&trace_types_lock);
2737
2738         return ret;
2739 }
2740
2741 static ssize_t
2742 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2743                         size_t cnt, loff_t *ppos)
2744 {
2745         char buf[MAX_TRACER_SIZE+1];
2746         int i;
2747         size_t ret;
2748         int err;
2749
2750         ret = cnt;
2751
2752         if (cnt > MAX_TRACER_SIZE)
2753                 cnt = MAX_TRACER_SIZE;
2754
2755         if (copy_from_user(&buf, ubuf, cnt))
2756                 return -EFAULT;
2757
2758         buf[cnt] = 0;
2759
2760         /* strip ending whitespace. */
2761         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2762                 buf[i] = 0;
2763
2764         err = tracing_set_tracer(buf);
2765         if (err)
2766                 return err;
2767
2768         *ppos += ret;
2769
2770         return ret;
2771 }
2772
2773 static ssize_t
2774 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2775                      size_t cnt, loff_t *ppos)
2776 {
2777         unsigned long *ptr = filp->private_data;
2778         char buf[64];
2779         int r;
2780
2781         r = snprintf(buf, sizeof(buf), "%ld\n",
2782                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2783         if (r > sizeof(buf))
2784                 r = sizeof(buf);
2785         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2786 }
2787
2788 static ssize_t
2789 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2790                       size_t cnt, loff_t *ppos)
2791 {
2792         unsigned long *ptr = filp->private_data;
2793         char buf[64];
2794         unsigned long val;
2795         int ret;
2796
2797         if (cnt >= sizeof(buf))
2798                 return -EINVAL;
2799
2800         if (copy_from_user(&buf, ubuf, cnt))
2801                 return -EFAULT;
2802
2803         buf[cnt] = 0;
2804
2805         ret = strict_strtoul(buf, 10, &val);
2806         if (ret < 0)
2807                 return ret;
2808
2809         *ptr = val * 1000;
2810
2811         return cnt;
2812 }
2813
2814 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2815 {
2816         long cpu_file = (long) inode->i_private;
2817         struct trace_iterator *iter;
2818         int ret = 0;
2819
2820         if (tracing_disabled)
2821                 return -ENODEV;
2822
2823         mutex_lock(&trace_types_lock);
2824
2825         /* We only allow one reader per cpu */
2826         if (cpu_file == TRACE_PIPE_ALL_CPU) {
2827                 if (!cpumask_empty(tracing_reader_cpumask)) {
2828                         ret = -EBUSY;
2829                         goto out;
2830                 }
2831                 cpumask_setall(tracing_reader_cpumask);
2832         } else {
2833                 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2834                         cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2835                 else {
2836                         ret = -EBUSY;
2837                         goto out;
2838                 }
2839         }
2840
2841         /* create a buffer to store the information to pass to userspace */
2842         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2843         if (!iter) {
2844                 ret = -ENOMEM;
2845                 goto out;
2846         }
2847
2848         /*
2849          * We make a copy of the current tracer to avoid concurrent
2850          * changes on it while we are reading.
2851          */
2852         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2853         if (!iter->trace) {
2854                 ret = -ENOMEM;
2855                 goto fail;
2856         }
2857         if (current_trace)
2858                 *iter->trace = *current_trace;
2859
2860         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2861                 ret = -ENOMEM;
2862                 goto fail;
2863         }
2864
2865         /* trace pipe does not show start of buffer */
2866         cpumask_setall(iter->started);
2867
2868         if (trace_flags & TRACE_ITER_LATENCY_FMT)
2869                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2870
2871         iter->cpu_file = cpu_file;
2872         iter->tr = &global_trace;
2873         mutex_init(&iter->mutex);
2874         filp->private_data = iter;
2875
2876         if (iter->trace->pipe_open)
2877                 iter->trace->pipe_open(iter);
2878
2879 out:
2880         mutex_unlock(&trace_types_lock);
2881         return ret;
2882
2883 fail:
2884         kfree(iter->trace);
2885         kfree(iter);
2886         mutex_unlock(&trace_types_lock);
2887         return ret;
2888 }
2889
2890 static int tracing_release_pipe(struct inode *inode, struct file *file)
2891 {
2892         struct trace_iterator *iter = file->private_data;
2893
2894         mutex_lock(&trace_types_lock);
2895
2896         if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2897                 cpumask_clear(tracing_reader_cpumask);
2898         else
2899                 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2900
2901         mutex_unlock(&trace_types_lock);
2902
2903         free_cpumask_var(iter->started);
2904         mutex_destroy(&iter->mutex);
2905         kfree(iter->trace);
2906         kfree(iter);
2907
2908         return 0;
2909 }
2910
2911 static unsigned int
2912 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2913 {
2914         struct trace_iterator *iter = filp->private_data;
2915
2916         if (trace_flags & TRACE_ITER_BLOCK) {
2917                 /*
2918                  * Always select as readable when in blocking mode
2919                  */
2920                 return POLLIN | POLLRDNORM;
2921         } else {
2922                 if (!trace_empty(iter))
2923                         return POLLIN | POLLRDNORM;
2924                 poll_wait(filp, &trace_wait, poll_table);
2925                 if (!trace_empty(iter))
2926                         return POLLIN | POLLRDNORM;
2927
2928                 return 0;
2929         }
2930 }
2931
2932
2933 void default_wait_pipe(struct trace_iterator *iter)
2934 {
2935         DEFINE_WAIT(wait);
2936
2937         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2938
2939         if (trace_empty(iter))
2940                 schedule();
2941
2942         finish_wait(&trace_wait, &wait);
2943 }
2944
2945 /*
2946  * This is a make-shift waitqueue.
2947  * A tracer might use this callback on some rare cases:
2948  *
2949  *  1) the current tracer might hold the runqueue lock when it wakes up
2950  *     a reader, hence a deadlock (sched, function, and function graph tracers)
2951  *  2) the function tracers, trace all functions, we don't want
2952  *     the overhead of calling wake_up and friends
2953  *     (and tracing them too)
2954  *
2955  *     Anyway, this is really very primitive wakeup.
2956  */
2957 void poll_wait_pipe(struct trace_iterator *iter)
2958 {
2959         set_current_state(TASK_INTERRUPTIBLE);
2960         /* sleep for 100 msecs, and try again. */
2961         schedule_timeout(HZ / 10);
2962 }
2963
2964 /* Must be called with trace_types_lock mutex held. */
2965 static int tracing_wait_pipe(struct file *filp)
2966 {
2967         struct trace_iterator *iter = filp->private_data;
2968
2969         while (trace_empty(iter)) {
2970
2971                 if ((filp->f_flags & O_NONBLOCK)) {
2972                         return -EAGAIN;
2973                 }
2974
2975                 mutex_unlock(&iter->mutex);
2976
2977                 iter->trace->wait_pipe(iter);
2978
2979                 mutex_lock(&iter->mutex);
2980
2981                 if (signal_pending(current))
2982                         return -EINTR;
2983
2984                 /*
2985                  * We block until we read something and tracing is disabled.
2986                  * We still block if tracing is disabled, but we have never
2987                  * read anything. This allows a user to cat this file, and
2988                  * then enable tracing. But after we have read something,
2989                  * we give an EOF when tracing is again disabled.
2990                  *
2991                  * iter->pos will be 0 if we haven't read anything.
2992                  */
2993                 if (!tracer_enabled && iter->pos)
2994                         break;
2995         }
2996
2997         return 1;
2998 }
2999
3000 /*
3001  * Consumer reader.
3002  */
3003 static ssize_t
3004 tracing_read_pipe(struct file *filp, char __user *ubuf,
3005                   size_t cnt, loff_t *ppos)
3006 {
3007         struct trace_iterator *iter = filp->private_data;
3008         static struct tracer *old_tracer;
3009         ssize_t sret;
3010
3011         /* return any leftover data */
3012         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3013         if (sret != -EBUSY)
3014                 return sret;
3015
3016         trace_seq_init(&iter->seq);
3017
3018         /* copy the tracer to avoid using a global lock all around */
3019         mutex_lock(&trace_types_lock);
3020         if (unlikely(old_tracer != current_trace && current_trace)) {
3021                 old_tracer = current_trace;
3022                 *iter->trace = *current_trace;
3023         }
3024         mutex_unlock(&trace_types_lock);
3025
3026         /*
3027          * Avoid more than one consumer on a single file descriptor
3028          * This is just a matter of traces coherency, the ring buffer itself
3029          * is protected.
3030          */
3031         mutex_lock(&iter->mutex);
3032         if (iter->trace->read) {
3033                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3034                 if (sret)
3035                         goto out;
3036         }
3037
3038 waitagain:
3039         sret = tracing_wait_pipe(filp);
3040         if (sret <= 0)
3041                 goto out;
3042
3043         /* stop when tracing is finished */
3044         if (trace_empty(iter)) {
3045                 sret = 0;
3046                 goto out;
3047         }
3048
3049         if (cnt >= PAGE_SIZE)
3050                 cnt = PAGE_SIZE - 1;
3051
3052         /* reset all but tr, trace, and overruns */
3053         memset(&iter->seq, 0,
3054                sizeof(struct trace_iterator) -
3055                offsetof(struct trace_iterator, seq));
3056         iter->pos = -1;
3057
3058         trace_event_read_lock();
3059         while (find_next_entry_inc(iter) != NULL) {
3060                 enum print_line_t ret;
3061                 int len = iter->seq.len;
3062
3063                 ret = print_trace_line(iter);
3064                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3065                         /* don't print partial lines */
3066                         iter->seq.len = len;
3067                         break;
3068                 }
3069                 if (ret != TRACE_TYPE_NO_CONSUME)
3070                         trace_consume(iter);
3071
3072                 if (iter->seq.len >= cnt)
3073                         break;
3074         }
3075         trace_event_read_unlock();
3076
3077         /* Now copy what we have to the user */
3078         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3079         if (iter->seq.readpos >= iter->seq.len)
3080                 trace_seq_init(&iter->seq);
3081
3082         /*
3083          * If there was nothing to send to user, inspite of consuming trace
3084          * entries, go back to wait for more entries.
3085          */
3086         if (sret == -EBUSY)
3087                 goto waitagain;
3088
3089 out:
3090         mutex_unlock(&iter->mutex);
3091
3092         return sret;
3093 }
3094
3095 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3096                                      struct pipe_buffer *buf)
3097 {
3098         __free_page(buf->page);
3099 }
3100
3101 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3102                                      unsigned int idx)
3103 {
3104         __free_page(spd->pages[idx]);
3105 }
3106
3107 static struct pipe_buf_operations tracing_pipe_buf_ops = {
3108         .can_merge              = 0,
3109         .map                    = generic_pipe_buf_map,
3110         .unmap                  = generic_pipe_buf_unmap,
3111         .confirm                = generic_pipe_buf_confirm,
3112         .release                = tracing_pipe_buf_release,
3113         .steal                  = generic_pipe_buf_steal,
3114         .get                    = generic_pipe_buf_get,
3115 };
3116
3117 static size_t
3118 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3119 {
3120         size_t count;
3121         int ret;
3122
3123         /* Seq buffer is page-sized, exactly what we need. */
3124         for (;;) {
3125                 count = iter->seq.len;
3126                 ret = print_trace_line(iter);
3127                 count = iter->seq.len - count;
3128                 if (rem < count) {
3129                         rem = 0;
3130                         iter->seq.len -= count;
3131                         break;
3132                 }
3133                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3134                         iter->seq.len -= count;
3135                         break;
3136                 }
3137
3138                 if (ret != TRACE_TYPE_NO_CONSUME)
3139                         trace_consume(iter);
3140                 rem -= count;
3141                 if (!find_next_entry_inc(iter)) {
3142                         rem = 0;
3143                         iter->ent = NULL;
3144                         break;
3145                 }
3146         }
3147
3148         return rem;
3149 }
3150
3151 static ssize_t tracing_splice_read_pipe(struct file *filp,
3152                                         loff_t *ppos,
3153                                         struct pipe_inode_info *pipe,
3154                                         size_t len,
3155                                         unsigned int flags)
3156 {
3157         struct page *pages[PIPE_BUFFERS];
3158         struct partial_page partial[PIPE_BUFFERS];
3159         struct trace_iterator *iter = filp->private_data;
3160         struct splice_pipe_desc spd = {
3161                 .pages          = pages,
3162                 .partial        = partial,
3163                 .nr_pages       = 0, /* This gets updated below. */
3164                 .flags          = flags,
3165                 .ops            = &tracing_pipe_buf_ops,
3166                 .spd_release    = tracing_spd_release_pipe,
3167         };
3168         static struct tracer *old_tracer;
3169         ssize_t ret;
3170         size_t rem;
3171         unsigned int i;
3172
3173         /* copy the tracer to avoid using a global lock all around */
3174         mutex_lock(&trace_types_lock);
3175         if (unlikely(old_tracer != current_trace && current_trace)) {
3176                 old_tracer = current_trace;
3177                 *iter->trace = *current_trace;
3178         }
3179         mutex_unlock(&trace_types_lock);
3180
3181         mutex_lock(&iter->mutex);
3182
3183         if (iter->trace->splice_read) {
3184                 ret = iter->trace->splice_read(iter, filp,
3185                                                ppos, pipe, len, flags);
3186                 if (ret)
3187                         goto out_err;
3188         }
3189
3190         ret = tracing_wait_pipe(filp);
3191         if (ret <= 0)
3192                 goto out_err;
3193
3194         if (!iter->ent && !find_next_entry_inc(iter)) {
3195                 ret = -EFAULT;
3196                 goto out_err;
3197         }
3198
3199         trace_event_read_lock();
3200
3201         /* Fill as many pages as possible. */
3202         for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3203                 pages[i] = alloc_page(GFP_KERNEL);
3204                 if (!pages[i])
3205                         break;
3206
3207                 rem = tracing_fill_pipe_page(rem, iter);
3208
3209                 /* Copy the data into the page, so we can start over. */
3210                 ret = trace_seq_to_buffer(&iter->seq,
3211                                           page_address(pages[i]),
3212                                           iter->seq.len);
3213                 if (ret < 0) {
3214                         __free_page(pages[i]);
3215                         break;
3216                 }
3217                 partial[i].offset = 0;
3218                 partial[i].len = iter->seq.len;
3219
3220                 trace_seq_init(&iter->seq);
3221         }
3222
3223         trace_event_read_unlock();
3224         mutex_unlock(&iter->mutex);
3225
3226         spd.nr_pages = i;
3227
3228         return splice_to_pipe(pipe, &spd);
3229
3230 out_err:
3231         mutex_unlock(&iter->mutex);
3232
3233         return ret;
3234 }
3235
3236 static ssize_t
3237 tracing_entries_read(struct file *filp, char __user *ubuf,
3238                      size_t cnt, loff_t *ppos)
3239 {
3240         struct trace_array *tr = filp->private_data;
3241         char buf[96];
3242         int r;
3243
3244         mutex_lock(&trace_types_lock);
3245         if (!ring_buffer_expanded)
3246                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3247                             tr->entries >> 10,
3248                             trace_buf_size >> 10);
3249         else
3250                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3251         mutex_unlock(&trace_types_lock);
3252
3253         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3254 }
3255
3256 static ssize_t
3257 tracing_entries_write(struct file *filp, const char __user *ubuf,
3258                       size_t cnt, loff_t *ppos)
3259 {
3260         unsigned long val;
3261         char buf[64];
3262         int ret, cpu;
3263
3264         if (cnt >= sizeof(buf))
3265                 return -EINVAL;
3266
3267         if (copy_from_user(&buf, ubuf, cnt))
3268                 return -EFAULT;
3269
3270         buf[cnt] = 0;
3271
3272         ret = strict_strtoul(buf, 10, &val);
3273         if (ret < 0)
3274                 return ret;
3275
3276         /* must have at least 1 entry */
3277         if (!val)
3278                 return -EINVAL;
3279
3280         mutex_lock(&trace_types_lock);
3281
3282         tracing_stop();
3283
3284         /* disable all cpu buffers */
3285         for_each_tracing_cpu(cpu) {
3286                 if (global_trace.data[cpu])
3287                         atomic_inc(&global_trace.data[cpu]->disabled);
3288                 if (max_tr.data[cpu])
3289                         atomic_inc(&max_tr.data[cpu]->disabled);
3290         }
3291
3292         /* value is in KB */
3293         val <<= 10;
3294
3295         if (val != global_trace.entries) {
3296                 ret = tracing_resize_ring_buffer(val);
3297                 if (ret < 0) {
3298                         cnt = ret;
3299                         goto out;
3300                 }
3301         }
3302
3303         *ppos += cnt;
3304
3305         /* If check pages failed, return ENOMEM */
3306         if (tracing_disabled)
3307                 cnt = -ENOMEM;
3308  out:
3309         for_each_tracing_cpu(cpu) {
3310                 if (global_trace.data[cpu])
3311                         atomic_dec(&global_trace.data[cpu]->disabled);
3312                 if (max_tr.data[cpu])
3313                         atomic_dec(&max_tr.data[cpu]->disabled);
3314         }
3315
3316         tracing_start();
3317         max_tr.entries = global_trace.entries;
3318         mutex_unlock(&trace_types_lock);
3319
3320         return cnt;
3321 }
3322
3323 static ssize_t
3324 tracing_mark_write(struct file *filp, const char __user *ubuf,
3325                                         size_t cnt, loff_t *fpos)
3326 {
3327         char *buf;
3328
3329         if (tracing_disabled)
3330                 return -EINVAL;
3331
3332         if (cnt > TRACE_BUF_SIZE)
3333                 cnt = TRACE_BUF_SIZE;
3334
3335         buf = kmalloc(cnt + 2, GFP_KERNEL);
3336         if (buf == NULL)
3337                 return -ENOMEM;
3338
3339         if (copy_from_user(buf, ubuf, cnt)) {
3340                 kfree(buf);
3341                 return -EFAULT;
3342         }
3343         if (buf[cnt-1] != '\n') {
3344                 buf[cnt] = '\n';
3345                 buf[cnt+1] = '\0';
3346         } else
3347                 buf[cnt] = '\0';
3348
3349         cnt = trace_vprintk(0, buf, NULL);
3350         kfree(buf);
3351         *fpos += cnt;
3352
3353         return cnt;
3354 }
3355
3356 static ssize_t tracing_clock_read(struct file *filp, char __user *ubuf,
3357                                   size_t cnt, loff_t *ppos)
3358 {
3359         char buf[64];
3360         int bufiter = 0;
3361         int i;
3362
3363         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3364                 bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter,
3365                         "%s%s%s%s", i ? " " : "",
3366                         i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3367                         i == trace_clock_id ? "]" : "");
3368         bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter, "\n");
3369
3370         return simple_read_from_buffer(ubuf, cnt, ppos, buf, bufiter);
3371 }
3372
3373 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3374                                    size_t cnt, loff_t *fpos)
3375 {
3376         char buf[64];
3377         const char *clockstr;
3378         int i;
3379
3380         if (cnt >= sizeof(buf))
3381                 return -EINVAL;
3382
3383         if (copy_from_user(&buf, ubuf, cnt))
3384                 return -EFAULT;
3385
3386         buf[cnt] = 0;
3387
3388         clockstr = strstrip(buf);
3389
3390         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3391                 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3392                         break;
3393         }
3394         if (i == ARRAY_SIZE(trace_clocks))
3395                 return -EINVAL;
3396
3397         trace_clock_id = i;
3398
3399         mutex_lock(&trace_types_lock);
3400
3401         ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3402         if (max_tr.buffer)
3403                 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3404
3405         mutex_unlock(&trace_types_lock);
3406
3407         *fpos += cnt;
3408
3409         return cnt;
3410 }
3411
3412 static const struct file_operations tracing_max_lat_fops = {
3413         .open           = tracing_open_generic,
3414         .read           = tracing_max_lat_read,
3415         .write          = tracing_max_lat_write,
3416 };
3417
3418 static const struct file_operations tracing_ctrl_fops = {
3419         .open           = tracing_open_generic,
3420         .read           = tracing_ctrl_read,
3421         .write          = tracing_ctrl_write,
3422 };
3423
3424 static const struct file_operations set_tracer_fops = {
3425         .open           = tracing_open_generic,
3426         .read           = tracing_set_trace_read,
3427         .write          = tracing_set_trace_write,
3428 };
3429
3430 static const struct file_operations tracing_pipe_fops = {
3431         .open           = tracing_open_pipe,
3432         .poll           = tracing_poll_pipe,
3433         .read           = tracing_read_pipe,
3434         .splice_read    = tracing_splice_read_pipe,
3435         .release        = tracing_release_pipe,
3436 };
3437
3438 static const struct file_operations tracing_entries_fops = {
3439         .open           = tracing_open_generic,
3440         .read           = tracing_entries_read,
3441         .write          = tracing_entries_write,
3442 };
3443
3444 static const struct file_operations tracing_mark_fops = {
3445         .open           = tracing_open_generic,
3446         .write          = tracing_mark_write,
3447 };
3448
3449 static const struct file_operations trace_clock_fops = {
3450         .open           = tracing_open_generic,
3451         .read           = tracing_clock_read,
3452         .write          = tracing_clock_write,
3453 };
3454
3455 struct ftrace_buffer_info {
3456         struct trace_array      *tr;
3457         void                    *spare;
3458         int                     cpu;
3459         unsigned int            read;
3460 };
3461
3462 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3463 {
3464         int cpu = (int)(long)inode->i_private;
3465         struct ftrace_buffer_info *info;
3466
3467         if (tracing_disabled)
3468                 return -ENODEV;
3469
3470         info = kzalloc(sizeof(*info), GFP_KERNEL);
3471         if (!info)
3472                 return -ENOMEM;
3473
3474         info->tr        = &global_trace;
3475         info->cpu       = cpu;
3476         info->spare     = NULL;
3477         /* Force reading ring buffer for first read */
3478         info->read      = (unsigned int)-1;
3479
3480         filp->private_data = info;
3481
3482         return nonseekable_open(inode, filp);
3483 }
3484
3485 static ssize_t
3486 tracing_buffers_read(struct file *filp, char __user *ubuf,
3487                      size_t count, loff_t *ppos)
3488 {
3489         struct ftrace_buffer_info *info = filp->private_data;
3490         unsigned int pos;
3491         ssize_t ret;
3492         size_t size;
3493
3494         if (!count)
3495                 return 0;
3496
3497         if (!info->spare)
3498                 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3499         if (!info->spare)
3500                 return -ENOMEM;
3501
3502         /* Do we have previous read data to read? */
3503         if (info->read < PAGE_SIZE)
3504                 goto read;
3505
3506         info->read = 0;
3507
3508         ret = ring_buffer_read_page(info->tr->buffer,
3509                                     &info->spare,
3510                                     count,
3511                                     info->cpu, 0);
3512         if (ret < 0)
3513                 return 0;
3514
3515         pos = ring_buffer_page_len(info->spare);
3516
3517         if (pos < PAGE_SIZE)
3518                 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3519
3520 read:
3521         size = PAGE_SIZE - info->read;
3522         if (size > count)
3523                 size = count;
3524
3525         ret = copy_to_user(ubuf, info->spare + info->read, size);
3526         if (ret == size)
3527                 return -EFAULT;
3528         size -= ret;
3529
3530         *ppos += size;
3531         info->read += size;
3532
3533         return size;
3534 }
3535
3536 static int tracing_buffers_release(struct inode *inode, struct file *file)
3537 {
3538         struct ftrace_buffer_info *info = file->private_data;
3539
3540         if (info->spare)
3541                 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3542         kfree(info);
3543
3544         return 0;
3545 }
3546
3547 struct buffer_ref {
3548         struct ring_buffer      *buffer;
3549         void                    *page;
3550         int                     ref;
3551 };
3552
3553 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3554                                     struct pipe_buffer *buf)
3555 {
3556         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3557
3558         if (--ref->ref)
3559                 return;
3560
3561         ring_buffer_free_read_page(ref->buffer, ref->page);
3562         kfree(ref);
3563         buf->private = 0;
3564 }
3565
3566 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3567                                  struct pipe_buffer *buf)
3568 {
3569         return 1;
3570 }
3571
3572 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3573                                 struct pipe_buffer *buf)
3574 {
3575         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3576
3577         ref->ref++;
3578 }
3579
3580 /* Pipe buffer operations for a buffer. */
3581 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3582         .can_merge              = 0,
3583         .map                    = generic_pipe_buf_map,
3584         .unmap                  = generic_pipe_buf_unmap,
3585         .confirm                = generic_pipe_buf_confirm,
3586         .release                = buffer_pipe_buf_release,
3587         .steal                  = buffer_pipe_buf_steal,
3588         .get                    = buffer_pipe_buf_get,
3589 };
3590
3591 /*
3592  * Callback from splice_to_pipe(), if we need to release some pages
3593  * at the end of the spd in case we error'ed out in filling the pipe.
3594  */
3595 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3596 {
3597         struct buffer_ref *ref =
3598                 (struct buffer_ref *)spd->partial[i].private;
3599
3600         if (--ref->ref)
3601                 return;
3602
3603         ring_buffer_free_read_page(ref->buffer, ref->page);
3604         kfree(ref);
3605         spd->partial[i].private = 0;
3606 }
3607
3608 static ssize_t
3609 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3610                             struct pipe_inode_info *pipe, size_t len,
3611                             unsigned int flags)
3612 {
3613         struct ftrace_buffer_info *info = file->private_data;
3614         struct partial_page partial[PIPE_BUFFERS];
3615         struct page *pages[PIPE_BUFFERS];
3616         struct splice_pipe_desc spd = {
3617                 .pages          = pages,
3618                 .partial        = partial,
3619                 .flags          = flags,
3620                 .ops            = &buffer_pipe_buf_ops,
3621                 .spd_release    = buffer_spd_release,
3622         };
3623         struct buffer_ref *ref;
3624         int entries, size, i;
3625         size_t ret;
3626
3627         if (*ppos & (PAGE_SIZE - 1)) {
3628                 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3629                 return -EINVAL;
3630         }
3631
3632         if (len & (PAGE_SIZE - 1)) {
3633                 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3634                 if (len < PAGE_SIZE)
3635                         return -EINVAL;
3636                 len &= PAGE_MASK;
3637         }
3638
3639         entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3640
3641         for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3642                 struct page *page;
3643                 int r;
3644
3645                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3646                 if (!ref)
3647                         break;
3648
3649                 ref->ref = 1;
3650                 ref->buffer = info->tr->buffer;
3651                 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3652                 if (!ref->page) {
3653                         kfree(ref);
3654                         break;
3655                 }
3656
3657                 r = ring_buffer_read_page(ref->buffer, &ref->page,
3658                                           len, info->cpu, 1);
3659                 if (r < 0) {
3660                         ring_buffer_free_read_page(ref->buffer,
3661                                                    ref->page);
3662                         kfree(ref);
3663                         break;
3664                 }
3665
3666                 /*
3667                  * zero out any left over data, this is going to
3668                  * user land.
3669                  */
3670                 size = ring_buffer_page_len(ref->page);
3671                 if (size < PAGE_SIZE)
3672                         memset(ref->page + size, 0, PAGE_SIZE - size);
3673
3674                 page = virt_to_page(ref->page);
3675
3676                 spd.pages[i] = page;
3677                 spd.partial[i].len = PAGE_SIZE;
3678                 spd.partial[i].offset = 0;
3679                 spd.partial[i].private = (unsigned long)ref;
3680                 spd.nr_pages++;
3681                 *ppos += PAGE_SIZE;
3682
3683                 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3684         }
3685
3686         spd.nr_pages = i;
3687
3688         /* did we read anything? */
3689         if (!spd.nr_pages) {
3690                 if (flags & SPLICE_F_NONBLOCK)
3691                         ret = -EAGAIN;
3692                 else
3693                         ret = 0;
3694                 /* TODO: block */
3695                 return ret;
3696         }
3697
3698         ret = splice_to_pipe(pipe, &spd);
3699
3700         return ret;
3701 }
3702
3703 static const struct file_operations tracing_buffers_fops = {
3704         .open           = tracing_buffers_open,
3705         .read           = tracing_buffers_read,
3706         .release        = tracing_buffers_release,
3707         .splice_read    = tracing_buffers_splice_read,
3708         .llseek         = no_llseek,
3709 };
3710
3711 static ssize_t
3712 tracing_stats_read(struct file *filp, char __user *ubuf,
3713                    size_t count, loff_t *ppos)
3714 {
3715         unsigned long cpu = (unsigned long)filp->private_data;
3716         struct trace_array *tr = &global_trace;
3717         struct trace_seq *s;
3718         unsigned long cnt;
3719
3720         s = kmalloc(sizeof(*s), GFP_KERNEL);
3721         if (!s)
3722                 return -ENOMEM;
3723
3724         trace_seq_init(s);
3725
3726         cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3727         trace_seq_printf(s, "entries: %ld\n", cnt);
3728
3729         cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3730         trace_seq_printf(s, "overrun: %ld\n", cnt);
3731
3732         cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3733         trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3734
3735         count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3736
3737         kfree(s);
3738
3739         return count;
3740 }
3741
3742 static const struct file_operations tracing_stats_fops = {
3743         .open           = tracing_open_generic,
3744         .read           = tracing_stats_read,
3745 };
3746
3747 #ifdef CONFIG_DYNAMIC_FTRACE
3748
3749 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3750 {
3751         return 0;
3752 }
3753
3754 static ssize_t
3755 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3756                   size_t cnt, loff_t *ppos)
3757 {
3758         static char ftrace_dyn_info_buffer[1024];
3759         static DEFINE_MUTEX(dyn_info_mutex);
3760         unsigned long *p = filp->private_data;
3761         char *buf = ftrace_dyn_info_buffer;
3762         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3763         int r;
3764
3765         mutex_lock(&dyn_info_mutex);
3766         r = sprintf(buf, "%ld ", *p);
3767
3768         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3769         buf[r++] = '\n';
3770
3771         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3772
3773         mutex_unlock(&dyn_info_mutex);
3774
3775         return r;
3776 }
3777
3778 static const struct file_operations tracing_dyn_info_fops = {
3779         .open           = tracing_open_generic,
3780         .read           = tracing_read_dyn_info,
3781 };
3782 #endif
3783
3784 static struct dentry *d_tracer;
3785
3786 struct dentry *tracing_init_dentry(void)
3787 {
3788         static int once;
3789
3790         if (d_tracer)
3791                 return d_tracer;
3792
3793         if (!debugfs_initialized())
3794                 return NULL;
3795
3796         d_tracer = debugfs_create_dir("tracing", NULL);
3797
3798         if (!d_tracer && !once) {
3799                 once = 1;
3800                 pr_warning("Could not create debugfs directory 'tracing'\n");
3801                 return NULL;
3802         }
3803
3804         return d_tracer;
3805 }
3806
3807 static struct dentry *d_percpu;
3808
3809 struct dentry *tracing_dentry_percpu(void)
3810 {
3811         static int once;
3812         struct dentry *d_tracer;
3813
3814         if (d_percpu)
3815                 return d_percpu;
3816
3817         d_tracer = tracing_init_dentry();
3818
3819         if (!d_tracer)
3820                 return NULL;
3821
3822         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3823
3824         if (!d_percpu && !once) {
3825                 once = 1;
3826                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3827                 return NULL;
3828         }
3829
3830         return d_percpu;
3831 }
3832
3833 static void tracing_init_debugfs_percpu(long cpu)
3834 {
3835         struct dentry *d_percpu = tracing_dentry_percpu();
3836         struct dentry *d_cpu;
3837         /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3838         char cpu_dir[7];
3839
3840         if (cpu > 999 || cpu < 0)
3841                 return;
3842
3843         sprintf(cpu_dir, "cpu%ld", cpu);
3844         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3845         if (!d_cpu) {
3846                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3847                 return;
3848         }
3849
3850         /* per cpu trace_pipe */
3851         trace_create_file("trace_pipe", 0444, d_cpu,
3852                         (void *) cpu, &tracing_pipe_fops);
3853
3854         /* per cpu trace */
3855         trace_create_file("trace", 0644, d_cpu,
3856                         (void *) cpu, &tracing_fops);
3857
3858         trace_create_file("trace_pipe_raw", 0444, d_cpu,
3859                         (void *) cpu, &tracing_buffers_fops);
3860
3861         trace_create_file("stats", 0444, d_cpu,
3862                         (void *) cpu, &tracing_stats_fops);
3863 }
3864
3865 #ifdef CONFIG_FTRACE_SELFTEST
3866 /* Let selftest have access to static functions in this file */
3867 #include "trace_selftest.c"
3868 #endif
3869
3870 struct trace_option_dentry {
3871         struct tracer_opt               *opt;
3872         struct tracer_flags             *flags;
3873         struct dentry                   *entry;
3874 };
3875
3876 static ssize_t
3877 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3878                         loff_t *ppos)
3879 {
3880         struct trace_option_dentry *topt = filp->private_data;
3881         char *buf;
3882
3883         if (topt->flags->val & topt->opt->bit)
3884                 buf = "1\n";
3885         else
3886                 buf = "0\n";
3887
3888         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3889 }
3890
3891 static ssize_t
3892 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3893                          loff_t *ppos)
3894 {
3895         struct trace_option_dentry *topt = filp->private_data;
3896         unsigned long val;
3897         char buf[64];
3898         int ret;
3899
3900         if (cnt >= sizeof(buf))
3901                 return -EINVAL;
3902
3903         if (copy_from_user(&buf, ubuf, cnt))
3904                 return -EFAULT;
3905
3906         buf[cnt] = 0;
3907
3908         ret = strict_strtoul(buf, 10, &val);
3909         if (ret < 0)
3910                 return ret;
3911
3912         ret = 0;
3913         switch (val) {
3914         case 0:
3915                 /* do nothing if already cleared */
3916                 if (!(topt->flags->val & topt->opt->bit))
3917                         break;
3918
3919                 mutex_lock(&trace_types_lock);
3920                 if (current_trace->set_flag)
3921                         ret = current_trace->set_flag(topt->flags->val,
3922                                                       topt->opt->bit, 0);
3923                 mutex_unlock(&trace_types_lock);
3924                 if (ret)
3925                         return ret;
3926                 topt->flags->val &= ~topt->opt->bit;
3927                 break;
3928         case 1:
3929                 /* do nothing if already set */
3930                 if (topt->flags->val & topt->opt->bit)
3931                         break;
3932
3933                 mutex_lock(&trace_types_lock);
3934                 if (current_trace->set_flag)
3935                         ret = current_trace->set_flag(topt->flags->val,
3936                                                       topt->opt->bit, 1);
3937                 mutex_unlock(&trace_types_lock);
3938                 if (ret)
3939                         return ret;
3940                 topt->flags->val |= topt->opt->bit;
3941                 break;
3942
3943         default:
3944                 return -EINVAL;
3945         }
3946
3947         *ppos += cnt;
3948
3949         return cnt;
3950 }
3951
3952
3953 static const struct file_operations trace_options_fops = {
3954         .open = tracing_open_generic,
3955         .read = trace_options_read,
3956         .write = trace_options_write,
3957 };
3958
3959 static ssize_t
3960 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3961                         loff_t *ppos)
3962 {
3963         long index = (long)filp->private_data;
3964         char *buf;
3965
3966         if (trace_flags & (1 << index))
3967                 buf = "1\n";
3968         else
3969                 buf = "0\n";
3970
3971         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3972 }
3973
3974 static ssize_t
3975 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3976                          loff_t *ppos)
3977 {
3978         long index = (long)filp->private_data;
3979         char buf[64];
3980         unsigned long val;
3981         int ret;
3982
3983         if (cnt >= sizeof(buf))
3984                 return -EINVAL;
3985
3986         if (copy_from_user(&buf, ubuf, cnt))
3987                 return -EFAULT;
3988
3989         buf[cnt] = 0;
3990
3991         ret = strict_strtoul(buf, 10, &val);
3992         if (ret < 0)
3993                 return ret;
3994
3995         if (val != 0 && val != 1)
3996                 return -EINVAL;
3997         set_tracer_flags(1 << index, val);
3998
3999         *ppos += cnt;
4000
4001         return cnt;
4002 }
4003
4004 static const struct file_operations trace_options_core_fops = {
4005         .open = tracing_open_generic,
4006         .read = trace_options_core_read,
4007         .write = trace_options_core_write,
4008 };
4009
4010 struct dentry *trace_create_file(const char *name,
4011                                  mode_t mode,
4012                                  struct dentry *parent,
4013                                  void *data,
4014                                  const struct file_operations *fops)
4015 {
4016         struct dentry *ret;
4017
4018         ret = debugfs_create_file(name, mode, parent, data, fops);
4019         if (!ret)
4020                 pr_warning("Could not create debugfs '%s' entry\n", name);
4021
4022         return ret;
4023 }
4024
4025
4026 static struct dentry *trace_options_init_dentry(void)
4027 {
4028         struct dentry *d_tracer;
4029         static struct dentry *t_options;
4030
4031         if (t_options)
4032                 return t_options;
4033
4034         d_tracer = tracing_init_dentry();
4035         if (!d_tracer)
4036                 return NULL;
4037
4038         t_options = debugfs_create_dir("options", d_tracer);
4039         if (!t_options) {
4040                 pr_warning("Could not create debugfs directory 'options'\n");
4041                 return NULL;
4042         }
4043
4044         return t_options;
4045 }
4046
4047 static void
4048 create_trace_option_file(struct trace_option_dentry *topt,
4049                          struct tracer_flags *flags,
4050                          struct tracer_opt *opt)
4051 {
4052         struct dentry *t_options;
4053
4054         t_options = trace_options_init_dentry();
4055         if (!t_options)
4056                 return;
4057
4058         topt->flags = flags;
4059         topt->opt = opt;
4060
4061         topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4062                                     &trace_options_fops);
4063
4064 }
4065
4066 static struct trace_option_dentry *
4067 create_trace_option_files(struct tracer *tracer)
4068 {
4069         struct trace_option_dentry *topts;
4070         struct tracer_flags *flags;
4071         struct tracer_opt *opts;
4072         int cnt;
4073
4074         if (!tracer)
4075                 return NULL;
4076
4077         flags = tracer->flags;
4078
4079         if (!flags || !flags->opts)
4080                 return NULL;
4081
4082         opts = flags->opts;
4083
4084         for (cnt = 0; opts[cnt].name; cnt++)
4085                 ;
4086
4087         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4088         if (!topts)
4089                 return NULL;
4090
4091         for (cnt = 0; opts[cnt].name; cnt++)
4092                 create_trace_option_file(&topts[cnt], flags,
4093                                          &opts[cnt]);
4094
4095         return topts;
4096 }
4097
4098 static void
4099 destroy_trace_option_files(struct trace_option_dentry *topts)
4100 {
4101         int cnt;
4102
4103         if (!topts)
4104                 return;
4105
4106         for (cnt = 0; topts[cnt].opt; cnt++) {
4107                 if (topts[cnt].entry)
4108                         debugfs_remove(topts[cnt].entry);
4109         }
4110
4111         kfree(topts);
4112 }
4113
4114 static struct dentry *
4115 create_trace_option_core_file(const char *option, long index)
4116 {
4117         struct dentry *t_options;
4118
4119         t_options = trace_options_init_dentry();
4120         if (!t_options)
4121                 return NULL;
4122
4123         return trace_create_file(option, 0644, t_options, (void *)index,
4124                                     &trace_options_core_fops);
4125 }
4126
4127 static __init void create_trace_options_dir(void)
4128 {
4129         struct dentry *t_options;
4130         int i;
4131
4132         t_options = trace_options_init_dentry();
4133         if (!t_options)
4134                 return;
4135
4136         for (i = 0; trace_options[i]; i++)
4137                 create_trace_option_core_file(trace_options[i], i);
4138 }
4139
4140 static __init int tracer_init_debugfs(void)
4141 {
4142         struct dentry *d_tracer;
4143         int cpu;
4144
4145         d_tracer = tracing_init_dentry();
4146
4147         trace_create_file("tracing_enabled", 0644, d_tracer,
4148                         &global_trace, &tracing_ctrl_fops);
4149
4150         trace_create_file("trace_options", 0644, d_tracer,
4151                         NULL, &tracing_iter_fops);
4152
4153         trace_create_file("tracing_cpumask", 0644, d_tracer,
4154                         NULL, &tracing_cpumask_fops);
4155
4156         trace_create_file("trace", 0644, d_tracer,
4157                         (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4158
4159         trace_create_file("available_tracers", 0444, d_tracer,
4160                         &global_trace, &show_traces_fops);
4161
4162         trace_create_file("current_tracer", 0644, d_tracer,
4163                         &global_trace, &set_tracer_fops);
4164
4165 #ifdef CONFIG_TRACER_MAX_TRACE
4166         trace_create_file("tracing_max_latency", 0644, d_tracer,
4167                         &tracing_max_latency, &tracing_max_lat_fops);
4168
4169         trace_create_file("tracing_thresh", 0644, d_tracer,
4170                         &tracing_thresh, &tracing_max_lat_fops);
4171 #endif
4172
4173         trace_create_file("README", 0444, d_tracer,
4174                         NULL, &tracing_readme_fops);
4175
4176         trace_create_file("trace_pipe", 0444, d_tracer,
4177                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4178
4179         trace_create_file("buffer_size_kb", 0644, d_tracer,
4180                         &global_trace, &tracing_entries_fops);
4181
4182         trace_create_file("trace_marker", 0220, d_tracer,
4183                         NULL, &tracing_mark_fops);
4184
4185         trace_create_file("saved_cmdlines", 0444, d_tracer,
4186                         NULL, &tracing_saved_cmdlines_fops);
4187
4188         trace_create_file("trace_clock", 0644, d_tracer, NULL,
4189                           &trace_clock_fops);
4190
4191 #ifdef CONFIG_DYNAMIC_FTRACE
4192         trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4193                         &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4194 #endif
4195 #ifdef CONFIG_SYSPROF_TRACER
4196         init_tracer_sysprof_debugfs(d_tracer);
4197 #endif
4198
4199         create_trace_options_dir();
4200
4201         for_each_tracing_cpu(cpu)
4202                 tracing_init_debugfs_percpu(cpu);
4203
4204         return 0;
4205 }
4206
4207 static int trace_panic_handler(struct notifier_block *this,
4208                                unsigned long event, void *unused)
4209 {
4210         if (ftrace_dump_on_oops)
4211                 ftrace_dump();
4212         return NOTIFY_OK;
4213 }
4214
4215 static struct notifier_block trace_panic_notifier = {
4216         .notifier_call  = trace_panic_handler,
4217         .next           = NULL,
4218         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
4219 };
4220
4221 static int trace_die_handler(struct notifier_block *self,
4222                              unsigned long val,
4223                              void *data)
4224 {
4225         switch (val) {
4226         case DIE_OOPS:
4227                 if (ftrace_dump_on_oops)
4228                         ftrace_dump();
4229                 break;
4230         default:
4231                 break;
4232         }
4233         return NOTIFY_OK;
4234 }
4235
4236 static struct notifier_block trace_die_notifier = {
4237         .notifier_call = trace_die_handler,
4238         .priority = 200
4239 };
4240
4241 /*
4242  * printk is set to max of 1024, we really don't need it that big.
4243  * Nothing should be printing 1000 characters anyway.
4244  */
4245 #define TRACE_MAX_PRINT         1000
4246
4247 /*
4248  * Define here KERN_TRACE so that we have one place to modify
4249  * it if we decide to change what log level the ftrace dump
4250  * should be at.
4251  */
4252 #define KERN_TRACE              KERN_EMERG
4253
4254 static void
4255 trace_printk_seq(struct trace_seq *s)
4256 {
4257         /* Probably should print a warning here. */
4258         if (s->len >= 1000)
4259                 s->len = 1000;
4260
4261         /* should be zero ended, but we are paranoid. */
4262         s->buffer[s->len] = 0;
4263
4264         printk(KERN_TRACE "%s", s->buffer);
4265
4266         trace_seq_init(s);
4267 }
4268
4269 static void __ftrace_dump(bool disable_tracing)
4270 {
4271         static raw_spinlock_t ftrace_dump_lock =
4272                 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
4273         /* use static because iter can be a bit big for the stack */
4274         static struct trace_iterator iter;
4275         unsigned int old_userobj;
4276         static int dump_ran;
4277         unsigned long flags;
4278         int cnt = 0, cpu;
4279
4280         /* only one dump */
4281         local_irq_save(flags);
4282         __raw_spin_lock(&ftrace_dump_lock);
4283         if (dump_ran)
4284                 goto out;
4285
4286         dump_ran = 1;
4287
4288         tracing_off();
4289
4290         if (disable_tracing)
4291                 ftrace_kill();
4292
4293         for_each_tracing_cpu(cpu) {
4294                 atomic_inc(&global_trace.data[cpu]->disabled);
4295         }
4296
4297         old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4298
4299         /* don't look at user memory in panic mode */
4300         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4301
4302         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4303
4304         /* Simulate the iterator */
4305         iter.tr = &global_trace;
4306         iter.trace = current_trace;
4307         iter.cpu_file = TRACE_PIPE_ALL_CPU;
4308
4309         /*
4310          * We need to stop all tracing on all CPUS to read the
4311          * the next buffer. This is a bit expensive, but is
4312          * not done often. We fill all what we can read,
4313          * and then release the locks again.
4314          */
4315
4316         while (!trace_empty(&iter)) {
4317
4318                 if (!cnt)
4319                         printk(KERN_TRACE "---------------------------------\n");
4320
4321                 cnt++;
4322
4323                 /* reset all but tr, trace, and overruns */
4324                 memset(&iter.seq, 0,
4325                        sizeof(struct trace_iterator) -
4326                        offsetof(struct trace_iterator, seq));
4327                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4328                 iter.pos = -1;
4329
4330                 if (find_next_entry_inc(&iter) != NULL) {
4331                         int ret;
4332
4333                         ret = print_trace_line(&iter);
4334                         if (ret != TRACE_TYPE_NO_CONSUME)
4335                                 trace_consume(&iter);
4336                 }
4337
4338                 trace_printk_seq(&iter.seq);
4339         }
4340
4341         if (!cnt)
4342                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4343         else
4344                 printk(KERN_TRACE "---------------------------------\n");
4345
4346         /* Re-enable tracing if requested */
4347         if (!disable_tracing) {
4348                 trace_flags |= old_userobj;
4349
4350                 for_each_tracing_cpu(cpu) {
4351                         atomic_dec(&global_trace.data[cpu]->disabled);
4352                 }
4353                 tracing_on();
4354         }
4355
4356  out:
4357         __raw_spin_unlock(&ftrace_dump_lock);
4358         local_irq_restore(flags);
4359 }
4360
4361 /* By default: disable tracing after the dump */
4362 void ftrace_dump(void)
4363 {
4364         __ftrace_dump(true);
4365 }
4366
4367 __init static int tracer_alloc_buffers(void)
4368 {
4369         int ring_buf_size;
4370         int i;
4371         int ret = -ENOMEM;
4372
4373         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4374                 goto out;
4375
4376         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4377                 goto out_free_buffer_mask;
4378
4379         if (!zalloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4380                 goto out_free_tracing_cpumask;
4381
4382         /* To save memory, keep the ring buffer size to its minimum */
4383         if (ring_buffer_expanded)
4384                 ring_buf_size = trace_buf_size;
4385         else
4386                 ring_buf_size = 1;
4387
4388         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4389         cpumask_copy(tracing_cpumask, cpu_all_mask);
4390
4391         /* TODO: make the number of buffers hot pluggable with CPUS */
4392         global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4393                                                    TRACE_BUFFER_FLAGS);
4394         if (!global_trace.buffer) {
4395                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4396                 WARN_ON(1);
4397                 goto out_free_cpumask;
4398         }
4399         global_trace.entries = ring_buffer_size(global_trace.buffer);
4400
4401
4402 #ifdef CONFIG_TRACER_MAX_TRACE
4403         max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4404                                              TRACE_BUFFER_FLAGS);
4405         if (!max_tr.buffer) {
4406                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4407                 WARN_ON(1);
4408                 ring_buffer_free(global_trace.buffer);
4409                 goto out_free_cpumask;
4410         }
4411         max_tr.entries = ring_buffer_size(max_tr.buffer);
4412         WARN_ON(max_tr.entries != global_trace.entries);
4413 #endif
4414
4415         /* Allocate the first page for all buffers */
4416         for_each_tracing_cpu(i) {
4417                 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4418                 max_tr.data[i] = &per_cpu(max_data, i);
4419         }
4420
4421         trace_init_cmdlines();
4422
4423         register_tracer(&nop_trace);
4424         current_trace = &nop_trace;
4425 #ifdef CONFIG_BOOT_TRACER
4426         register_tracer(&boot_tracer);
4427 #endif
4428         /* All seems OK, enable tracing */
4429         tracing_disabled = 0;
4430
4431         atomic_notifier_chain_register(&panic_notifier_list,
4432                                        &trace_panic_notifier);
4433
4434         register_die_notifier(&trace_die_notifier);
4435
4436         return 0;
4437
4438 out_free_cpumask:
4439         free_cpumask_var(tracing_reader_cpumask);
4440 out_free_tracing_cpumask:
4441         free_cpumask_var(tracing_cpumask);
4442 out_free_buffer_mask:
4443         free_cpumask_var(tracing_buffer_mask);
4444 out:
4445         return ret;
4446 }
4447
4448 __init static int clear_boot_tracer(void)
4449 {
4450         /*
4451          * The default tracer at boot buffer is an init section.
4452          * This function is called in lateinit. If we did not
4453          * find the boot tracer, then clear it out, to prevent
4454          * later registration from accessing the buffer that is
4455          * about to be freed.
4456          */
4457         if (!default_bootup_tracer)
4458                 return 0;
4459
4460         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4461                default_bootup_tracer);
4462         default_bootup_tracer = NULL;
4463
4464         return 0;
4465 }
4466
4467 early_initcall(tracer_alloc_buffers);
4468 fs_initcall(tracer_init_debugfs);
4469 late_initcall(clear_boot_tracer);