trace, documentation: Fix branch profiling location in debugfs
[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 <generated/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/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/fs.h>
40
41 #include "trace.h"
42 #include "trace_output.h"
43
44 /*
45  * On boot up, the ring buffer is set to the minimum size, so that
46  * we do not waste memory on systems that are not using tracing.
47  */
48 int ring_buffer_expanded;
49
50 /*
51  * We need to change this state when a selftest is running.
52  * A selftest will lurk into the ring-buffer to count the
53  * entries inserted during the selftest although some concurrent
54  * insertions into the ring-buffer such as trace_printk could occurred
55  * at the same time, giving false positive or negative results.
56  */
57 static bool __read_mostly tracing_selftest_running;
58
59 /*
60  * If a tracer is running, we do not want to run SELFTEST.
61  */
62 bool __read_mostly tracing_selftest_disabled;
63
64 /* For tracers that don't implement custom flags */
65 static struct tracer_opt dummy_tracer_opt[] = {
66         { }
67 };
68
69 static struct tracer_flags dummy_tracer_flags = {
70         .val = 0,
71         .opts = dummy_tracer_opt
72 };
73
74 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
75 {
76         return 0;
77 }
78
79 /*
80  * Kill all tracing for good (never come back).
81  * It is initialized to 1 but will turn to zero if the initialization
82  * of the tracer is successful. But that is the only place that sets
83  * this back to zero.
84  */
85 static int tracing_disabled = 1;
86
87 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
88
89 static inline void ftrace_disable_cpu(void)
90 {
91         preempt_disable();
92         __this_cpu_inc(ftrace_cpu_disabled);
93 }
94
95 static inline void ftrace_enable_cpu(void)
96 {
97         __this_cpu_dec(ftrace_cpu_disabled);
98         preempt_enable();
99 }
100
101 cpumask_var_t __read_mostly     tracing_buffer_mask;
102
103 /*
104  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
105  *
106  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
107  * is set, then ftrace_dump is called. This will output the contents
108  * of the ftrace buffers to the console.  This is very useful for
109  * capturing traces that lead to crashes and outputing it to a
110  * serial console.
111  *
112  * It is default off, but you can enable it with either specifying
113  * "ftrace_dump_on_oops" in the kernel command line, or setting
114  * /proc/sys/kernel/ftrace_dump_on_oops
115  * Set 1 if you want to dump buffers of all CPUs
116  * Set 2 if you want to dump the buffer of the CPU that triggered oops
117  */
118
119 enum ftrace_dump_mode ftrace_dump_on_oops;
120
121 static int tracing_set_tracer(const char *buf);
122
123 #define MAX_TRACER_SIZE         100
124 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
125 static char *default_bootup_tracer;
126
127 static int __init set_cmdline_ftrace(char *str)
128 {
129         strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
130         default_bootup_tracer = bootup_tracer_buf;
131         /* We are using ftrace early, expand it */
132         ring_buffer_expanded = 1;
133         return 1;
134 }
135 __setup("ftrace=", set_cmdline_ftrace);
136
137 static int __init set_ftrace_dump_on_oops(char *str)
138 {
139         if (*str++ != '=' || !*str) {
140                 ftrace_dump_on_oops = DUMP_ALL;
141                 return 1;
142         }
143
144         if (!strcmp("orig_cpu", str)) {
145                 ftrace_dump_on_oops = DUMP_ORIG;
146                 return 1;
147         }
148
149         return 0;
150 }
151 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
152
153 unsigned long long ns2usecs(cycle_t nsec)
154 {
155         nsec += 500;
156         do_div(nsec, 1000);
157         return nsec;
158 }
159
160 /*
161  * The global_trace is the descriptor that holds the tracing
162  * buffers for the live tracing. For each CPU, it contains
163  * a link list of pages that will store trace entries. The
164  * page descriptor of the pages in the memory is used to hold
165  * the link list by linking the lru item in the page descriptor
166  * to each of the pages in the buffer per CPU.
167  *
168  * For each active CPU there is a data field that holds the
169  * pages for the buffer for that CPU. Each CPU has the same number
170  * of pages allocated for its buffer.
171  */
172 static struct trace_array       global_trace;
173
174 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
175
176 int filter_current_check_discard(struct ring_buffer *buffer,
177                                  struct ftrace_event_call *call, void *rec,
178                                  struct ring_buffer_event *event)
179 {
180         return filter_check_discard(call, rec, buffer, event);
181 }
182 EXPORT_SYMBOL_GPL(filter_current_check_discard);
183
184 cycle_t ftrace_now(int cpu)
185 {
186         u64 ts;
187
188         /* Early boot up does not have a buffer yet */
189         if (!global_trace.buffer)
190                 return trace_clock_local();
191
192         ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
193         ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
194
195         return ts;
196 }
197
198 /*
199  * The max_tr is used to snapshot the global_trace when a maximum
200  * latency is reached. Some tracers will use this to store a maximum
201  * trace while it continues examining live traces.
202  *
203  * The buffers for the max_tr are set up the same as the global_trace.
204  * When a snapshot is taken, the link list of the max_tr is swapped
205  * with the link list of the global_trace and the buffers are reset for
206  * the global_trace so the tracing can continue.
207  */
208 static struct trace_array       max_tr;
209
210 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
211
212 /* tracer_enabled is used to toggle activation of a tracer */
213 static int                      tracer_enabled = 1;
214
215 /**
216  * tracing_is_enabled - return tracer_enabled status
217  *
218  * This function is used by other tracers to know the status
219  * of the tracer_enabled flag.  Tracers may use this function
220  * to know if it should enable their features when starting
221  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
222  */
223 int tracing_is_enabled(void)
224 {
225         return tracer_enabled;
226 }
227
228 /*
229  * trace_buf_size is the size in bytes that is allocated
230  * for a buffer. Note, the number of bytes is always rounded
231  * to page size.
232  *
233  * This number is purposely set to a low number of 16384.
234  * If the dump on oops happens, it will be much appreciated
235  * to not have to wait for all that output. Anyway this can be
236  * boot time and run time configurable.
237  */
238 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
239
240 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
241
242 /* trace_types holds a link list of available tracers. */
243 static struct tracer            *trace_types __read_mostly;
244
245 /* current_trace points to the tracer that is currently active */
246 static struct tracer            *current_trace __read_mostly;
247
248 /*
249  * trace_types_lock is used to protect the trace_types list.
250  */
251 static DEFINE_MUTEX(trace_types_lock);
252
253 /*
254  * serialize the access of the ring buffer
255  *
256  * ring buffer serializes readers, but it is low level protection.
257  * The validity of the events (which returns by ring_buffer_peek() ..etc)
258  * are not protected by ring buffer.
259  *
260  * The content of events may become garbage if we allow other process consumes
261  * these events concurrently:
262  *   A) the page of the consumed events may become a normal page
263  *      (not reader page) in ring buffer, and this page will be rewrited
264  *      by events producer.
265  *   B) The page of the consumed events may become a page for splice_read,
266  *      and this page will be returned to system.
267  *
268  * These primitives allow multi process access to different cpu ring buffer
269  * concurrently.
270  *
271  * These primitives don't distinguish read-only and read-consume access.
272  * Multi read-only access are also serialized.
273  */
274
275 #ifdef CONFIG_SMP
276 static DECLARE_RWSEM(all_cpu_access_lock);
277 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
278
279 static inline void trace_access_lock(int cpu)
280 {
281         if (cpu == TRACE_PIPE_ALL_CPU) {
282                 /* gain it for accessing the whole ring buffer. */
283                 down_write(&all_cpu_access_lock);
284         } else {
285                 /* gain it for accessing a cpu ring buffer. */
286
287                 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
288                 down_read(&all_cpu_access_lock);
289
290                 /* Secondly block other access to this @cpu ring buffer. */
291                 mutex_lock(&per_cpu(cpu_access_lock, cpu));
292         }
293 }
294
295 static inline void trace_access_unlock(int cpu)
296 {
297         if (cpu == TRACE_PIPE_ALL_CPU) {
298                 up_write(&all_cpu_access_lock);
299         } else {
300                 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
301                 up_read(&all_cpu_access_lock);
302         }
303 }
304
305 static inline void trace_access_lock_init(void)
306 {
307         int cpu;
308
309         for_each_possible_cpu(cpu)
310                 mutex_init(&per_cpu(cpu_access_lock, cpu));
311 }
312
313 #else
314
315 static DEFINE_MUTEX(access_lock);
316
317 static inline void trace_access_lock(int cpu)
318 {
319         (void)cpu;
320         mutex_lock(&access_lock);
321 }
322
323 static inline void trace_access_unlock(int cpu)
324 {
325         (void)cpu;
326         mutex_unlock(&access_lock);
327 }
328
329 static inline void trace_access_lock_init(void)
330 {
331 }
332
333 #endif
334
335 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
336 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
337
338 /* trace_flags holds trace_options default values */
339 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
340         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
341         TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE;
342
343 static int trace_stop_count;
344 static DEFINE_SPINLOCK(tracing_start_lock);
345
346 /**
347  * trace_wake_up - wake up tasks waiting for trace input
348  *
349  * Simply wakes up any task that is blocked on the trace_wait
350  * queue. These is used with trace_poll for tasks polling the trace.
351  */
352 void trace_wake_up(void)
353 {
354         int cpu;
355
356         if (trace_flags & TRACE_ITER_BLOCK)
357                 return;
358         /*
359          * The runqueue_is_locked() can fail, but this is the best we
360          * have for now:
361          */
362         cpu = get_cpu();
363         if (!runqueue_is_locked(cpu))
364                 wake_up(&trace_wait);
365         put_cpu();
366 }
367
368 static int __init set_buf_size(char *str)
369 {
370         unsigned long buf_size;
371
372         if (!str)
373                 return 0;
374         buf_size = memparse(str, &str);
375         /* nr_entries can not be zero */
376         if (buf_size == 0)
377                 return 0;
378         trace_buf_size = buf_size;
379         return 1;
380 }
381 __setup("trace_buf_size=", set_buf_size);
382
383 static int __init set_tracing_thresh(char *str)
384 {
385         unsigned long threshhold;
386         int ret;
387
388         if (!str)
389                 return 0;
390         ret = strict_strtoul(str, 0, &threshhold);
391         if (ret < 0)
392                 return 0;
393         tracing_thresh = threshhold * 1000;
394         return 1;
395 }
396 __setup("tracing_thresh=", set_tracing_thresh);
397
398 unsigned long nsecs_to_usecs(unsigned long nsecs)
399 {
400         return nsecs / 1000;
401 }
402
403 /* These must match the bit postions in trace_iterator_flags */
404 static const char *trace_options[] = {
405         "print-parent",
406         "sym-offset",
407         "sym-addr",
408         "verbose",
409         "raw",
410         "hex",
411         "bin",
412         "block",
413         "stacktrace",
414         "trace_printk",
415         "ftrace_preempt",
416         "branch",
417         "annotate",
418         "userstacktrace",
419         "sym-userobj",
420         "printk-msg-only",
421         "context-info",
422         "latency-format",
423         "sleep-time",
424         "graph-time",
425         "record-cmd",
426         "overwrite",
427         NULL
428 };
429
430 static struct {
431         u64 (*func)(void);
432         const char *name;
433 } trace_clocks[] = {
434         { trace_clock_local,    "local" },
435         { trace_clock_global,   "global" },
436 };
437
438 int trace_clock_id;
439
440 /*
441  * trace_parser_get_init - gets the buffer for trace parser
442  */
443 int trace_parser_get_init(struct trace_parser *parser, int size)
444 {
445         memset(parser, 0, sizeof(*parser));
446
447         parser->buffer = kmalloc(size, GFP_KERNEL);
448         if (!parser->buffer)
449                 return 1;
450
451         parser->size = size;
452         return 0;
453 }
454
455 /*
456  * trace_parser_put - frees the buffer for trace parser
457  */
458 void trace_parser_put(struct trace_parser *parser)
459 {
460         kfree(parser->buffer);
461 }
462
463 /*
464  * trace_get_user - reads the user input string separated by  space
465  * (matched by isspace(ch))
466  *
467  * For each string found the 'struct trace_parser' is updated,
468  * and the function returns.
469  *
470  * Returns number of bytes read.
471  *
472  * See kernel/trace/trace.h for 'struct trace_parser' details.
473  */
474 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
475         size_t cnt, loff_t *ppos)
476 {
477         char ch;
478         size_t read = 0;
479         ssize_t ret;
480
481         if (!*ppos)
482                 trace_parser_clear(parser);
483
484         ret = get_user(ch, ubuf++);
485         if (ret)
486                 goto out;
487
488         read++;
489         cnt--;
490
491         /*
492          * The parser is not finished with the last write,
493          * continue reading the user input without skipping spaces.
494          */
495         if (!parser->cont) {
496                 /* skip white space */
497                 while (cnt && isspace(ch)) {
498                         ret = get_user(ch, ubuf++);
499                         if (ret)
500                                 goto out;
501                         read++;
502                         cnt--;
503                 }
504
505                 /* only spaces were written */
506                 if (isspace(ch)) {
507                         *ppos += read;
508                         ret = read;
509                         goto out;
510                 }
511
512                 parser->idx = 0;
513         }
514
515         /* read the non-space input */
516         while (cnt && !isspace(ch)) {
517                 if (parser->idx < parser->size - 1)
518                         parser->buffer[parser->idx++] = ch;
519                 else {
520                         ret = -EINVAL;
521                         goto out;
522                 }
523                 ret = get_user(ch, ubuf++);
524                 if (ret)
525                         goto out;
526                 read++;
527                 cnt--;
528         }
529
530         /* We either got finished input or we have to wait for another call. */
531         if (isspace(ch)) {
532                 parser->buffer[parser->idx] = 0;
533                 parser->cont = false;
534         } else {
535                 parser->cont = true;
536                 parser->buffer[parser->idx++] = ch;
537         }
538
539         *ppos += read;
540         ret = read;
541
542 out:
543         return ret;
544 }
545
546 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
547 {
548         int len;
549         int ret;
550
551         if (!cnt)
552                 return 0;
553
554         if (s->len <= s->readpos)
555                 return -EBUSY;
556
557         len = s->len - s->readpos;
558         if (cnt > len)
559                 cnt = len;
560         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
561         if (ret == cnt)
562                 return -EFAULT;
563
564         cnt -= ret;
565
566         s->readpos += cnt;
567         return cnt;
568 }
569
570 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
571 {
572         int len;
573         void *ret;
574
575         if (s->len <= s->readpos)
576                 return -EBUSY;
577
578         len = s->len - s->readpos;
579         if (cnt > len)
580                 cnt = len;
581         ret = memcpy(buf, s->buffer + s->readpos, cnt);
582         if (!ret)
583                 return -EFAULT;
584
585         s->readpos += cnt;
586         return cnt;
587 }
588
589 /*
590  * ftrace_max_lock is used to protect the swapping of buffers
591  * when taking a max snapshot. The buffers themselves are
592  * protected by per_cpu spinlocks. But the action of the swap
593  * needs its own lock.
594  *
595  * This is defined as a arch_spinlock_t in order to help
596  * with performance when lockdep debugging is enabled.
597  *
598  * It is also used in other places outside the update_max_tr
599  * so it needs to be defined outside of the
600  * CONFIG_TRACER_MAX_TRACE.
601  */
602 static arch_spinlock_t ftrace_max_lock =
603         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
604
605 unsigned long __read_mostly     tracing_thresh;
606
607 #ifdef CONFIG_TRACER_MAX_TRACE
608 unsigned long __read_mostly     tracing_max_latency;
609
610 /*
611  * Copy the new maximum trace into the separate maximum-trace
612  * structure. (this way the maximum trace is permanently saved,
613  * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
614  */
615 static void
616 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
617 {
618         struct trace_array_cpu *data = tr->data[cpu];
619         struct trace_array_cpu *max_data;
620
621         max_tr.cpu = cpu;
622         max_tr.time_start = data->preempt_timestamp;
623
624         max_data = max_tr.data[cpu];
625         max_data->saved_latency = tracing_max_latency;
626         max_data->critical_start = data->critical_start;
627         max_data->critical_end = data->critical_end;
628
629         memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
630         max_data->pid = tsk->pid;
631         max_data->uid = task_uid(tsk);
632         max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
633         max_data->policy = tsk->policy;
634         max_data->rt_priority = tsk->rt_priority;
635
636         /* record this tasks comm */
637         tracing_record_cmdline(tsk);
638 }
639
640 /**
641  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
642  * @tr: tracer
643  * @tsk: the task with the latency
644  * @cpu: The cpu that initiated the trace.
645  *
646  * Flip the buffers between the @tr and the max_tr and record information
647  * about which task was the cause of this latency.
648  */
649 void
650 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
651 {
652         struct ring_buffer *buf = tr->buffer;
653
654         if (trace_stop_count)
655                 return;
656
657         WARN_ON_ONCE(!irqs_disabled());
658         if (!current_trace->use_max_tr) {
659                 WARN_ON_ONCE(1);
660                 return;
661         }
662         arch_spin_lock(&ftrace_max_lock);
663
664         tr->buffer = max_tr.buffer;
665         max_tr.buffer = buf;
666
667         __update_max_tr(tr, tsk, cpu);
668         arch_spin_unlock(&ftrace_max_lock);
669 }
670
671 /**
672  * update_max_tr_single - only copy one trace over, and reset the rest
673  * @tr - tracer
674  * @tsk - task with the latency
675  * @cpu - the cpu of the buffer to copy.
676  *
677  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
678  */
679 void
680 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
681 {
682         int ret;
683
684         if (trace_stop_count)
685                 return;
686
687         WARN_ON_ONCE(!irqs_disabled());
688         if (!current_trace->use_max_tr) {
689                 WARN_ON_ONCE(1);
690                 return;
691         }
692
693         arch_spin_lock(&ftrace_max_lock);
694
695         ftrace_disable_cpu();
696
697         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
698
699         if (ret == -EBUSY) {
700                 /*
701                  * We failed to swap the buffer due to a commit taking
702                  * place on this CPU. We fail to record, but we reset
703                  * the max trace buffer (no one writes directly to it)
704                  * and flag that it failed.
705                  */
706                 trace_array_printk(&max_tr, _THIS_IP_,
707                         "Failed to swap buffers due to commit in progress\n");
708         }
709
710         ftrace_enable_cpu();
711
712         WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
713
714         __update_max_tr(tr, tsk, cpu);
715         arch_spin_unlock(&ftrace_max_lock);
716 }
717 #endif /* CONFIG_TRACER_MAX_TRACE */
718
719 /**
720  * register_tracer - register a tracer with the ftrace system.
721  * @type - the plugin for the tracer
722  *
723  * Register a new plugin tracer.
724  */
725 int register_tracer(struct tracer *type)
726 __releases(kernel_lock)
727 __acquires(kernel_lock)
728 {
729         struct tracer *t;
730         int ret = 0;
731
732         if (!type->name) {
733                 pr_info("Tracer must have a name\n");
734                 return -1;
735         }
736
737         if (strlen(type->name) >= MAX_TRACER_SIZE) {
738                 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
739                 return -1;
740         }
741
742         mutex_lock(&trace_types_lock);
743
744         tracing_selftest_running = true;
745
746         for (t = trace_types; t; t = t->next) {
747                 if (strcmp(type->name, t->name) == 0) {
748                         /* already found */
749                         pr_info("Tracer %s already registered\n",
750                                 type->name);
751                         ret = -1;
752                         goto out;
753                 }
754         }
755
756         if (!type->set_flag)
757                 type->set_flag = &dummy_set_flag;
758         if (!type->flags)
759                 type->flags = &dummy_tracer_flags;
760         else
761                 if (!type->flags->opts)
762                         type->flags->opts = dummy_tracer_opt;
763         if (!type->wait_pipe)
764                 type->wait_pipe = default_wait_pipe;
765
766
767 #ifdef CONFIG_FTRACE_STARTUP_TEST
768         if (type->selftest && !tracing_selftest_disabled) {
769                 struct tracer *saved_tracer = current_trace;
770                 struct trace_array *tr = &global_trace;
771
772                 /*
773                  * Run a selftest on this tracer.
774                  * Here we reset the trace buffer, and set the current
775                  * tracer to be this tracer. The tracer can then run some
776                  * internal tracing to verify that everything is in order.
777                  * If we fail, we do not register this tracer.
778                  */
779                 tracing_reset_online_cpus(tr);
780
781                 current_trace = type;
782
783                 /* If we expanded the buffers, make sure the max is expanded too */
784                 if (ring_buffer_expanded && type->use_max_tr)
785                         ring_buffer_resize(max_tr.buffer, trace_buf_size);
786
787                 /* the test is responsible for initializing and enabling */
788                 pr_info("Testing tracer %s: ", type->name);
789                 ret = type->selftest(type, tr);
790                 /* the test is responsible for resetting too */
791                 current_trace = saved_tracer;
792                 if (ret) {
793                         printk(KERN_CONT "FAILED!\n");
794                         goto out;
795                 }
796                 /* Only reset on passing, to avoid touching corrupted buffers */
797                 tracing_reset_online_cpus(tr);
798
799                 /* Shrink the max buffer again */
800                 if (ring_buffer_expanded && type->use_max_tr)
801                         ring_buffer_resize(max_tr.buffer, 1);
802
803                 printk(KERN_CONT "PASSED\n");
804         }
805 #endif
806
807         type->next = trace_types;
808         trace_types = type;
809
810  out:
811         tracing_selftest_running = false;
812         mutex_unlock(&trace_types_lock);
813
814         if (ret || !default_bootup_tracer)
815                 goto out_unlock;
816
817         if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
818                 goto out_unlock;
819
820         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
821         /* Do we want this tracer to start on bootup? */
822         tracing_set_tracer(type->name);
823         default_bootup_tracer = NULL;
824         /* disable other selftests, since this will break it. */
825         tracing_selftest_disabled = 1;
826 #ifdef CONFIG_FTRACE_STARTUP_TEST
827         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
828                type->name);
829 #endif
830
831  out_unlock:
832         return ret;
833 }
834
835 void unregister_tracer(struct tracer *type)
836 {
837         struct tracer **t;
838
839         mutex_lock(&trace_types_lock);
840         for (t = &trace_types; *t; t = &(*t)->next) {
841                 if (*t == type)
842                         goto found;
843         }
844         pr_info("Tracer %s not registered\n", type->name);
845         goto out;
846
847  found:
848         *t = (*t)->next;
849
850         if (type == current_trace && tracer_enabled) {
851                 tracer_enabled = 0;
852                 tracing_stop();
853                 if (current_trace->stop)
854                         current_trace->stop(&global_trace);
855                 current_trace = &nop_trace;
856         }
857 out:
858         mutex_unlock(&trace_types_lock);
859 }
860
861 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
862 {
863         ftrace_disable_cpu();
864         ring_buffer_reset_cpu(buffer, cpu);
865         ftrace_enable_cpu();
866 }
867
868 void tracing_reset(struct trace_array *tr, int cpu)
869 {
870         struct ring_buffer *buffer = tr->buffer;
871
872         ring_buffer_record_disable(buffer);
873
874         /* Make sure all commits have finished */
875         synchronize_sched();
876         __tracing_reset(buffer, cpu);
877
878         ring_buffer_record_enable(buffer);
879 }
880
881 void tracing_reset_online_cpus(struct trace_array *tr)
882 {
883         struct ring_buffer *buffer = tr->buffer;
884         int cpu;
885
886         ring_buffer_record_disable(buffer);
887
888         /* Make sure all commits have finished */
889         synchronize_sched();
890
891         tr->time_start = ftrace_now(tr->cpu);
892
893         for_each_online_cpu(cpu)
894                 __tracing_reset(buffer, cpu);
895
896         ring_buffer_record_enable(buffer);
897 }
898
899 void tracing_reset_current(int cpu)
900 {
901         tracing_reset(&global_trace, cpu);
902 }
903
904 void tracing_reset_current_online_cpus(void)
905 {
906         tracing_reset_online_cpus(&global_trace);
907 }
908
909 #define SAVED_CMDLINES 128
910 #define NO_CMDLINE_MAP UINT_MAX
911 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
912 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
913 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
914 static int cmdline_idx;
915 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
916
917 /* temporary disable recording */
918 static atomic_t trace_record_cmdline_disabled __read_mostly;
919
920 static void trace_init_cmdlines(void)
921 {
922         memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
923         memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
924         cmdline_idx = 0;
925 }
926
927 int is_tracing_stopped(void)
928 {
929         return trace_stop_count;
930 }
931
932 /**
933  * ftrace_off_permanent - disable all ftrace code permanently
934  *
935  * This should only be called when a serious anomally has
936  * been detected.  This will turn off the function tracing,
937  * ring buffers, and other tracing utilites. It takes no
938  * locks and can be called from any context.
939  */
940 void ftrace_off_permanent(void)
941 {
942         tracing_disabled = 1;
943         ftrace_stop();
944         tracing_off_permanent();
945 }
946
947 /**
948  * tracing_start - quick start of the tracer
949  *
950  * If tracing is enabled but was stopped by tracing_stop,
951  * this will start the tracer back up.
952  */
953 void tracing_start(void)
954 {
955         struct ring_buffer *buffer;
956         unsigned long flags;
957
958         if (tracing_disabled)
959                 return;
960
961         spin_lock_irqsave(&tracing_start_lock, flags);
962         if (--trace_stop_count) {
963                 if (trace_stop_count < 0) {
964                         /* Someone screwed up their debugging */
965                         WARN_ON_ONCE(1);
966                         trace_stop_count = 0;
967                 }
968                 goto out;
969         }
970
971         /* Prevent the buffers from switching */
972         arch_spin_lock(&ftrace_max_lock);
973
974         buffer = global_trace.buffer;
975         if (buffer)
976                 ring_buffer_record_enable(buffer);
977
978         buffer = max_tr.buffer;
979         if (buffer)
980                 ring_buffer_record_enable(buffer);
981
982         arch_spin_unlock(&ftrace_max_lock);
983
984         ftrace_start();
985  out:
986         spin_unlock_irqrestore(&tracing_start_lock, flags);
987 }
988
989 /**
990  * tracing_stop - quick stop of the tracer
991  *
992  * Light weight way to stop tracing. Use in conjunction with
993  * tracing_start.
994  */
995 void tracing_stop(void)
996 {
997         struct ring_buffer *buffer;
998         unsigned long flags;
999
1000         ftrace_stop();
1001         spin_lock_irqsave(&tracing_start_lock, flags);
1002         if (trace_stop_count++)
1003                 goto out;
1004
1005         /* Prevent the buffers from switching */
1006         arch_spin_lock(&ftrace_max_lock);
1007
1008         buffer = global_trace.buffer;
1009         if (buffer)
1010                 ring_buffer_record_disable(buffer);
1011
1012         buffer = max_tr.buffer;
1013         if (buffer)
1014                 ring_buffer_record_disable(buffer);
1015
1016         arch_spin_unlock(&ftrace_max_lock);
1017
1018  out:
1019         spin_unlock_irqrestore(&tracing_start_lock, flags);
1020 }
1021
1022 void trace_stop_cmdline_recording(void);
1023
1024 static void trace_save_cmdline(struct task_struct *tsk)
1025 {
1026         unsigned pid, idx;
1027
1028         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1029                 return;
1030
1031         /*
1032          * It's not the end of the world if we don't get
1033          * the lock, but we also don't want to spin
1034          * nor do we want to disable interrupts,
1035          * so if we miss here, then better luck next time.
1036          */
1037         if (!arch_spin_trylock(&trace_cmdline_lock))
1038                 return;
1039
1040         idx = map_pid_to_cmdline[tsk->pid];
1041         if (idx == NO_CMDLINE_MAP) {
1042                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1043
1044                 /*
1045                  * Check whether the cmdline buffer at idx has a pid
1046                  * mapped. We are going to overwrite that entry so we
1047                  * need to clear the map_pid_to_cmdline. Otherwise we
1048                  * would read the new comm for the old pid.
1049                  */
1050                 pid = map_cmdline_to_pid[idx];
1051                 if (pid != NO_CMDLINE_MAP)
1052                         map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1053
1054                 map_cmdline_to_pid[idx] = tsk->pid;
1055                 map_pid_to_cmdline[tsk->pid] = idx;
1056
1057                 cmdline_idx = idx;
1058         }
1059
1060         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1061
1062         arch_spin_unlock(&trace_cmdline_lock);
1063 }
1064
1065 void trace_find_cmdline(int pid, char comm[])
1066 {
1067         unsigned map;
1068
1069         if (!pid) {
1070                 strcpy(comm, "<idle>");
1071                 return;
1072         }
1073
1074         if (WARN_ON_ONCE(pid < 0)) {
1075                 strcpy(comm, "<XXX>");
1076                 return;
1077         }
1078
1079         if (pid > PID_MAX_DEFAULT) {
1080                 strcpy(comm, "<...>");
1081                 return;
1082         }
1083
1084         preempt_disable();
1085         arch_spin_lock(&trace_cmdline_lock);
1086         map = map_pid_to_cmdline[pid];
1087         if (map != NO_CMDLINE_MAP)
1088                 strcpy(comm, saved_cmdlines[map]);
1089         else
1090                 strcpy(comm, "<...>");
1091
1092         arch_spin_unlock(&trace_cmdline_lock);
1093         preempt_enable();
1094 }
1095
1096 void tracing_record_cmdline(struct task_struct *tsk)
1097 {
1098         if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1099             !tracing_is_on())
1100                 return;
1101
1102         trace_save_cmdline(tsk);
1103 }
1104
1105 void
1106 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1107                              int pc)
1108 {
1109         struct task_struct *tsk = current;
1110
1111         entry->preempt_count            = pc & 0xff;
1112         entry->pid                      = (tsk) ? tsk->pid : 0;
1113         entry->flags =
1114 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1115                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1116 #else
1117                 TRACE_FLAG_IRQS_NOSUPPORT |
1118 #endif
1119                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1120                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1121                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1122 }
1123 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1124
1125 struct ring_buffer_event *
1126 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1127                           int type,
1128                           unsigned long len,
1129                           unsigned long flags, int pc)
1130 {
1131         struct ring_buffer_event *event;
1132
1133         event = ring_buffer_lock_reserve(buffer, len);
1134         if (event != NULL) {
1135                 struct trace_entry *ent = ring_buffer_event_data(event);
1136
1137                 tracing_generic_entry_update(ent, flags, pc);
1138                 ent->type = type;
1139         }
1140
1141         return event;
1142 }
1143
1144 static inline void
1145 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1146                              struct ring_buffer_event *event,
1147                              unsigned long flags, int pc,
1148                              int wake)
1149 {
1150         ring_buffer_unlock_commit(buffer, event);
1151
1152         ftrace_trace_stack(buffer, flags, 6, pc);
1153         ftrace_trace_userstack(buffer, flags, pc);
1154
1155         if (wake)
1156                 trace_wake_up();
1157 }
1158
1159 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1160                                 struct ring_buffer_event *event,
1161                                 unsigned long flags, int pc)
1162 {
1163         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1164 }
1165
1166 struct ring_buffer_event *
1167 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1168                                   int type, unsigned long len,
1169                                   unsigned long flags, int pc)
1170 {
1171         *current_rb = global_trace.buffer;
1172         return trace_buffer_lock_reserve(*current_rb,
1173                                          type, len, flags, pc);
1174 }
1175 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1176
1177 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1178                                         struct ring_buffer_event *event,
1179                                         unsigned long flags, int pc)
1180 {
1181         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1182 }
1183 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1184
1185 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1186                                        struct ring_buffer_event *event,
1187                                        unsigned long flags, int pc)
1188 {
1189         __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1190 }
1191 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1192
1193 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1194                                          struct ring_buffer_event *event)
1195 {
1196         ring_buffer_discard_commit(buffer, event);
1197 }
1198 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1199
1200 void
1201 trace_function(struct trace_array *tr,
1202                unsigned long ip, unsigned long parent_ip, unsigned long flags,
1203                int pc)
1204 {
1205         struct ftrace_event_call *call = &event_function;
1206         struct ring_buffer *buffer = tr->buffer;
1207         struct ring_buffer_event *event;
1208         struct ftrace_entry *entry;
1209
1210         /* If we are reading the ring buffer, don't trace */
1211         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1212                 return;
1213
1214         event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1215                                           flags, pc);
1216         if (!event)
1217                 return;
1218         entry   = ring_buffer_event_data(event);
1219         entry->ip                       = ip;
1220         entry->parent_ip                = parent_ip;
1221
1222         if (!filter_check_discard(call, entry, buffer, event))
1223                 ring_buffer_unlock_commit(buffer, event);
1224 }
1225
1226 void
1227 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1228        unsigned long ip, unsigned long parent_ip, unsigned long flags,
1229        int pc)
1230 {
1231         if (likely(!atomic_read(&data->disabled)))
1232                 trace_function(tr, ip, parent_ip, flags, pc);
1233 }
1234
1235 #ifdef CONFIG_STACKTRACE
1236 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1237                                  unsigned long flags,
1238                                  int skip, int pc)
1239 {
1240         struct ftrace_event_call *call = &event_kernel_stack;
1241         struct ring_buffer_event *event;
1242         struct stack_entry *entry;
1243         struct stack_trace trace;
1244
1245         event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1246                                           sizeof(*entry), flags, pc);
1247         if (!event)
1248                 return;
1249         entry   = ring_buffer_event_data(event);
1250         memset(&entry->caller, 0, sizeof(entry->caller));
1251
1252         trace.nr_entries        = 0;
1253         trace.max_entries       = FTRACE_STACK_ENTRIES;
1254         trace.skip              = skip;
1255         trace.entries           = entry->caller;
1256
1257         save_stack_trace(&trace);
1258         if (!filter_check_discard(call, entry, buffer, event))
1259                 ring_buffer_unlock_commit(buffer, event);
1260 }
1261
1262 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1263                         int skip, int pc)
1264 {
1265         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1266                 return;
1267
1268         __ftrace_trace_stack(buffer, flags, skip, pc);
1269 }
1270
1271 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1272                    int pc)
1273 {
1274         __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1275 }
1276
1277 /**
1278  * trace_dump_stack - record a stack back trace in the trace buffer
1279  */
1280 void trace_dump_stack(void)
1281 {
1282         unsigned long flags;
1283
1284         if (tracing_disabled || tracing_selftest_running)
1285                 return;
1286
1287         local_save_flags(flags);
1288
1289         /* skipping 3 traces, seems to get us at the caller of this function */
1290         __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1291 }
1292
1293 static DEFINE_PER_CPU(int, user_stack_count);
1294
1295 void
1296 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1297 {
1298         struct ftrace_event_call *call = &event_user_stack;
1299         struct ring_buffer_event *event;
1300         struct userstack_entry *entry;
1301         struct stack_trace trace;
1302
1303         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1304                 return;
1305
1306         /*
1307          * NMIs can not handle page faults, even with fix ups.
1308          * The save user stack can (and often does) fault.
1309          */
1310         if (unlikely(in_nmi()))
1311                 return;
1312
1313         /*
1314          * prevent recursion, since the user stack tracing may
1315          * trigger other kernel events.
1316          */
1317         preempt_disable();
1318         if (__this_cpu_read(user_stack_count))
1319                 goto out;
1320
1321         __this_cpu_inc(user_stack_count);
1322
1323         event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1324                                           sizeof(*entry), flags, pc);
1325         if (!event)
1326                 goto out_drop_count;
1327         entry   = ring_buffer_event_data(event);
1328
1329         entry->tgid             = current->tgid;
1330         memset(&entry->caller, 0, sizeof(entry->caller));
1331
1332         trace.nr_entries        = 0;
1333         trace.max_entries       = FTRACE_STACK_ENTRIES;
1334         trace.skip              = 0;
1335         trace.entries           = entry->caller;
1336
1337         save_stack_trace_user(&trace);
1338         if (!filter_check_discard(call, entry, buffer, event))
1339                 ring_buffer_unlock_commit(buffer, event);
1340
1341  out_drop_count:
1342         __this_cpu_dec(user_stack_count);
1343  out:
1344         preempt_enable();
1345 }
1346
1347 #ifdef UNUSED
1348 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1349 {
1350         ftrace_trace_userstack(tr, flags, preempt_count());
1351 }
1352 #endif /* UNUSED */
1353
1354 #endif /* CONFIG_STACKTRACE */
1355
1356 /**
1357  * trace_vbprintk - write binary msg to tracing buffer
1358  *
1359  */
1360 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1361 {
1362         static arch_spinlock_t trace_buf_lock =
1363                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1364         static u32 trace_buf[TRACE_BUF_SIZE];
1365
1366         struct ftrace_event_call *call = &event_bprint;
1367         struct ring_buffer_event *event;
1368         struct ring_buffer *buffer;
1369         struct trace_array *tr = &global_trace;
1370         struct trace_array_cpu *data;
1371         struct bprint_entry *entry;
1372         unsigned long flags;
1373         int disable;
1374         int cpu, len = 0, size, pc;
1375
1376         if (unlikely(tracing_selftest_running || tracing_disabled))
1377                 return 0;
1378
1379         /* Don't pollute graph traces with trace_vprintk internals */
1380         pause_graph_tracing();
1381
1382         pc = preempt_count();
1383         preempt_disable_notrace();
1384         cpu = raw_smp_processor_id();
1385         data = tr->data[cpu];
1386
1387         disable = atomic_inc_return(&data->disabled);
1388         if (unlikely(disable != 1))
1389                 goto out;
1390
1391         /* Lockdep uses trace_printk for lock tracing */
1392         local_irq_save(flags);
1393         arch_spin_lock(&trace_buf_lock);
1394         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1395
1396         if (len > TRACE_BUF_SIZE || len < 0)
1397                 goto out_unlock;
1398
1399         size = sizeof(*entry) + sizeof(u32) * len;
1400         buffer = tr->buffer;
1401         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1402                                           flags, pc);
1403         if (!event)
1404                 goto out_unlock;
1405         entry = ring_buffer_event_data(event);
1406         entry->ip                       = ip;
1407         entry->fmt                      = fmt;
1408
1409         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1410         if (!filter_check_discard(call, entry, buffer, event)) {
1411                 ring_buffer_unlock_commit(buffer, event);
1412                 ftrace_trace_stack(buffer, flags, 6, pc);
1413         }
1414
1415 out_unlock:
1416         arch_spin_unlock(&trace_buf_lock);
1417         local_irq_restore(flags);
1418
1419 out:
1420         atomic_dec_return(&data->disabled);
1421         preempt_enable_notrace();
1422         unpause_graph_tracing();
1423
1424         return len;
1425 }
1426 EXPORT_SYMBOL_GPL(trace_vbprintk);
1427
1428 int trace_array_printk(struct trace_array *tr,
1429                        unsigned long ip, const char *fmt, ...)
1430 {
1431         int ret;
1432         va_list ap;
1433
1434         if (!(trace_flags & TRACE_ITER_PRINTK))
1435                 return 0;
1436
1437         va_start(ap, fmt);
1438         ret = trace_array_vprintk(tr, ip, fmt, ap);
1439         va_end(ap);
1440         return ret;
1441 }
1442
1443 int trace_array_vprintk(struct trace_array *tr,
1444                         unsigned long ip, const char *fmt, va_list args)
1445 {
1446         static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1447         static char trace_buf[TRACE_BUF_SIZE];
1448
1449         struct ftrace_event_call *call = &event_print;
1450         struct ring_buffer_event *event;
1451         struct ring_buffer *buffer;
1452         struct trace_array_cpu *data;
1453         int cpu, len = 0, size, pc;
1454         struct print_entry *entry;
1455         unsigned long irq_flags;
1456         int disable;
1457
1458         if (tracing_disabled || tracing_selftest_running)
1459                 return 0;
1460
1461         pc = preempt_count();
1462         preempt_disable_notrace();
1463         cpu = raw_smp_processor_id();
1464         data = tr->data[cpu];
1465
1466         disable = atomic_inc_return(&data->disabled);
1467         if (unlikely(disable != 1))
1468                 goto out;
1469
1470         pause_graph_tracing();
1471         raw_local_irq_save(irq_flags);
1472         arch_spin_lock(&trace_buf_lock);
1473         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1474
1475         size = sizeof(*entry) + len + 1;
1476         buffer = tr->buffer;
1477         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1478                                           irq_flags, pc);
1479         if (!event)
1480                 goto out_unlock;
1481         entry = ring_buffer_event_data(event);
1482         entry->ip = ip;
1483
1484         memcpy(&entry->buf, trace_buf, len);
1485         entry->buf[len] = '\0';
1486         if (!filter_check_discard(call, entry, buffer, event)) {
1487                 ring_buffer_unlock_commit(buffer, event);
1488                 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1489         }
1490
1491  out_unlock:
1492         arch_spin_unlock(&trace_buf_lock);
1493         raw_local_irq_restore(irq_flags);
1494         unpause_graph_tracing();
1495  out:
1496         atomic_dec_return(&data->disabled);
1497         preempt_enable_notrace();
1498
1499         return len;
1500 }
1501
1502 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1503 {
1504         return trace_array_vprintk(&global_trace, ip, fmt, args);
1505 }
1506 EXPORT_SYMBOL_GPL(trace_vprintk);
1507
1508 static void trace_iterator_increment(struct trace_iterator *iter)
1509 {
1510         /* Don't allow ftrace to trace into the ring buffers */
1511         ftrace_disable_cpu();
1512
1513         iter->idx++;
1514         if (iter->buffer_iter[iter->cpu])
1515                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1516
1517         ftrace_enable_cpu();
1518 }
1519
1520 static struct trace_entry *
1521 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1522                 unsigned long *lost_events)
1523 {
1524         struct ring_buffer_event *event;
1525         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1526
1527         /* Don't allow ftrace to trace into the ring buffers */
1528         ftrace_disable_cpu();
1529
1530         if (buf_iter)
1531                 event = ring_buffer_iter_peek(buf_iter, ts);
1532         else
1533                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1534                                          lost_events);
1535
1536         ftrace_enable_cpu();
1537
1538         return event ? ring_buffer_event_data(event) : NULL;
1539 }
1540
1541 static struct trace_entry *
1542 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1543                   unsigned long *missing_events, u64 *ent_ts)
1544 {
1545         struct ring_buffer *buffer = iter->tr->buffer;
1546         struct trace_entry *ent, *next = NULL;
1547         unsigned long lost_events = 0, next_lost = 0;
1548         int cpu_file = iter->cpu_file;
1549         u64 next_ts = 0, ts;
1550         int next_cpu = -1;
1551         int cpu;
1552
1553         /*
1554          * If we are in a per_cpu trace file, don't bother by iterating over
1555          * all cpu and peek directly.
1556          */
1557         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1558                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1559                         return NULL;
1560                 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1561                 if (ent_cpu)
1562                         *ent_cpu = cpu_file;
1563
1564                 return ent;
1565         }
1566
1567         for_each_tracing_cpu(cpu) {
1568
1569                 if (ring_buffer_empty_cpu(buffer, cpu))
1570                         continue;
1571
1572                 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1573
1574                 /*
1575                  * Pick the entry with the smallest timestamp:
1576                  */
1577                 if (ent && (!next || ts < next_ts)) {
1578                         next = ent;
1579                         next_cpu = cpu;
1580                         next_ts = ts;
1581                         next_lost = lost_events;
1582                 }
1583         }
1584
1585         if (ent_cpu)
1586                 *ent_cpu = next_cpu;
1587
1588         if (ent_ts)
1589                 *ent_ts = next_ts;
1590
1591         if (missing_events)
1592                 *missing_events = next_lost;
1593
1594         return next;
1595 }
1596
1597 /* Find the next real entry, without updating the iterator itself */
1598 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1599                                           int *ent_cpu, u64 *ent_ts)
1600 {
1601         return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1602 }
1603
1604 /* Find the next real entry, and increment the iterator to the next entry */
1605 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1606 {
1607         iter->ent = __find_next_entry(iter, &iter->cpu,
1608                                       &iter->lost_events, &iter->ts);
1609
1610         if (iter->ent)
1611                 trace_iterator_increment(iter);
1612
1613         return iter->ent ? iter : NULL;
1614 }
1615
1616 static void trace_consume(struct trace_iterator *iter)
1617 {
1618         /* Don't allow ftrace to trace into the ring buffers */
1619         ftrace_disable_cpu();
1620         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1621                             &iter->lost_events);
1622         ftrace_enable_cpu();
1623 }
1624
1625 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1626 {
1627         struct trace_iterator *iter = m->private;
1628         int i = (int)*pos;
1629         void *ent;
1630
1631         WARN_ON_ONCE(iter->leftover);
1632
1633         (*pos)++;
1634
1635         /* can't go backwards */
1636         if (iter->idx > i)
1637                 return NULL;
1638
1639         if (iter->idx < 0)
1640                 ent = trace_find_next_entry_inc(iter);
1641         else
1642                 ent = iter;
1643
1644         while (ent && iter->idx < i)
1645                 ent = trace_find_next_entry_inc(iter);
1646
1647         iter->pos = *pos;
1648
1649         return ent;
1650 }
1651
1652 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1653 {
1654         struct trace_array *tr = iter->tr;
1655         struct ring_buffer_event *event;
1656         struct ring_buffer_iter *buf_iter;
1657         unsigned long entries = 0;
1658         u64 ts;
1659
1660         tr->data[cpu]->skipped_entries = 0;
1661
1662         if (!iter->buffer_iter[cpu])
1663                 return;
1664
1665         buf_iter = iter->buffer_iter[cpu];
1666         ring_buffer_iter_reset(buf_iter);
1667
1668         /*
1669          * We could have the case with the max latency tracers
1670          * that a reset never took place on a cpu. This is evident
1671          * by the timestamp being before the start of the buffer.
1672          */
1673         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1674                 if (ts >= iter->tr->time_start)
1675                         break;
1676                 entries++;
1677                 ring_buffer_read(buf_iter, NULL);
1678         }
1679
1680         tr->data[cpu]->skipped_entries = entries;
1681 }
1682
1683 /*
1684  * The current tracer is copied to avoid a global locking
1685  * all around.
1686  */
1687 static void *s_start(struct seq_file *m, loff_t *pos)
1688 {
1689         struct trace_iterator *iter = m->private;
1690         static struct tracer *old_tracer;
1691         int cpu_file = iter->cpu_file;
1692         void *p = NULL;
1693         loff_t l = 0;
1694         int cpu;
1695
1696         /* copy the tracer to avoid using a global lock all around */
1697         mutex_lock(&trace_types_lock);
1698         if (unlikely(old_tracer != current_trace && current_trace)) {
1699                 old_tracer = current_trace;
1700                 *iter->trace = *current_trace;
1701         }
1702         mutex_unlock(&trace_types_lock);
1703
1704         atomic_inc(&trace_record_cmdline_disabled);
1705
1706         if (*pos != iter->pos) {
1707                 iter->ent = NULL;
1708                 iter->cpu = 0;
1709                 iter->idx = -1;
1710
1711                 ftrace_disable_cpu();
1712
1713                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1714                         for_each_tracing_cpu(cpu)
1715                                 tracing_iter_reset(iter, cpu);
1716                 } else
1717                         tracing_iter_reset(iter, cpu_file);
1718
1719                 ftrace_enable_cpu();
1720
1721                 iter->leftover = 0;
1722                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1723                         ;
1724
1725         } else {
1726                 /*
1727                  * If we overflowed the seq_file before, then we want
1728                  * to just reuse the trace_seq buffer again.
1729                  */
1730                 if (iter->leftover)
1731                         p = iter;
1732                 else {
1733                         l = *pos - 1;
1734                         p = s_next(m, p, &l);
1735                 }
1736         }
1737
1738         trace_event_read_lock();
1739         trace_access_lock(cpu_file);
1740         return p;
1741 }
1742
1743 static void s_stop(struct seq_file *m, void *p)
1744 {
1745         struct trace_iterator *iter = m->private;
1746
1747         atomic_dec(&trace_record_cmdline_disabled);
1748         trace_access_unlock(iter->cpu_file);
1749         trace_event_read_unlock();
1750 }
1751
1752 static void print_lat_help_header(struct seq_file *m)
1753 {
1754         seq_puts(m, "#                  _------=> CPU#            \n");
1755         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1756         seq_puts(m, "#                | / _----=> need-resched    \n");
1757         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1758         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1759         seq_puts(m, "#                |||| /     delay             \n");
1760         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1761         seq_puts(m, "#     \\   /      |||||  \\    |   /           \n");
1762 }
1763
1764 static void print_func_help_header(struct seq_file *m)
1765 {
1766         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1767         seq_puts(m, "#              | |       |          |         |\n");
1768 }
1769
1770
1771 void
1772 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1773 {
1774         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1775         struct trace_array *tr = iter->tr;
1776         struct trace_array_cpu *data = tr->data[tr->cpu];
1777         struct tracer *type = current_trace;
1778         unsigned long entries = 0;
1779         unsigned long total = 0;
1780         unsigned long count;
1781         const char *name = "preemption";
1782         int cpu;
1783
1784         if (type)
1785                 name = type->name;
1786
1787
1788         for_each_tracing_cpu(cpu) {
1789                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1790                 /*
1791                  * If this buffer has skipped entries, then we hold all
1792                  * entries for the trace and we need to ignore the
1793                  * ones before the time stamp.
1794                  */
1795                 if (tr->data[cpu]->skipped_entries) {
1796                         count -= tr->data[cpu]->skipped_entries;
1797                         /* total is the same as the entries */
1798                         total += count;
1799                 } else
1800                         total += count +
1801                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1802                 entries += count;
1803         }
1804
1805         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1806                    name, UTS_RELEASE);
1807         seq_puts(m, "# -----------------------------------"
1808                  "---------------------------------\n");
1809         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1810                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1811                    nsecs_to_usecs(data->saved_latency),
1812                    entries,
1813                    total,
1814                    tr->cpu,
1815 #if defined(CONFIG_PREEMPT_NONE)
1816                    "server",
1817 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1818                    "desktop",
1819 #elif defined(CONFIG_PREEMPT)
1820                    "preempt",
1821 #else
1822                    "unknown",
1823 #endif
1824                    /* These are reserved for later use */
1825                    0, 0, 0, 0);
1826 #ifdef CONFIG_SMP
1827         seq_printf(m, " #P:%d)\n", num_online_cpus());
1828 #else
1829         seq_puts(m, ")\n");
1830 #endif
1831         seq_puts(m, "#    -----------------\n");
1832         seq_printf(m, "#    | task: %.16s-%d "
1833                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1834                    data->comm, data->pid, data->uid, data->nice,
1835                    data->policy, data->rt_priority);
1836         seq_puts(m, "#    -----------------\n");
1837
1838         if (data->critical_start) {
1839                 seq_puts(m, "#  => started at: ");
1840                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1841                 trace_print_seq(m, &iter->seq);
1842                 seq_puts(m, "\n#  => ended at:   ");
1843                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1844                 trace_print_seq(m, &iter->seq);
1845                 seq_puts(m, "\n#\n");
1846         }
1847
1848         seq_puts(m, "#\n");
1849 }
1850
1851 static void test_cpu_buff_start(struct trace_iterator *iter)
1852 {
1853         struct trace_seq *s = &iter->seq;
1854
1855         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1856                 return;
1857
1858         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1859                 return;
1860
1861         if (cpumask_test_cpu(iter->cpu, iter->started))
1862                 return;
1863
1864         if (iter->tr->data[iter->cpu]->skipped_entries)
1865                 return;
1866
1867         cpumask_set_cpu(iter->cpu, iter->started);
1868
1869         /* Don't print started cpu buffer for the first entry of the trace */
1870         if (iter->idx > 1)
1871                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1872                                 iter->cpu);
1873 }
1874
1875 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1876 {
1877         struct trace_seq *s = &iter->seq;
1878         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1879         struct trace_entry *entry;
1880         struct trace_event *event;
1881
1882         entry = iter->ent;
1883
1884         test_cpu_buff_start(iter);
1885
1886         event = ftrace_find_event(entry->type);
1887
1888         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1889                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1890                         if (!trace_print_lat_context(iter))
1891                                 goto partial;
1892                 } else {
1893                         if (!trace_print_context(iter))
1894                                 goto partial;
1895                 }
1896         }
1897
1898         if (event)
1899                 return event->funcs->trace(iter, sym_flags, event);
1900
1901         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1902                 goto partial;
1903
1904         return TRACE_TYPE_HANDLED;
1905 partial:
1906         return TRACE_TYPE_PARTIAL_LINE;
1907 }
1908
1909 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1910 {
1911         struct trace_seq *s = &iter->seq;
1912         struct trace_entry *entry;
1913         struct trace_event *event;
1914
1915         entry = iter->ent;
1916
1917         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1918                 if (!trace_seq_printf(s, "%d %d %llu ",
1919                                       entry->pid, iter->cpu, iter->ts))
1920                         goto partial;
1921         }
1922
1923         event = ftrace_find_event(entry->type);
1924         if (event)
1925                 return event->funcs->raw(iter, 0, event);
1926
1927         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1928                 goto partial;
1929
1930         return TRACE_TYPE_HANDLED;
1931 partial:
1932         return TRACE_TYPE_PARTIAL_LINE;
1933 }
1934
1935 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1936 {
1937         struct trace_seq *s = &iter->seq;
1938         unsigned char newline = '\n';
1939         struct trace_entry *entry;
1940         struct trace_event *event;
1941
1942         entry = iter->ent;
1943
1944         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1945                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1946                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1947                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1948         }
1949
1950         event = ftrace_find_event(entry->type);
1951         if (event) {
1952                 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1953                 if (ret != TRACE_TYPE_HANDLED)
1954                         return ret;
1955         }
1956
1957         SEQ_PUT_FIELD_RET(s, newline);
1958
1959         return TRACE_TYPE_HANDLED;
1960 }
1961
1962 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1963 {
1964         struct trace_seq *s = &iter->seq;
1965         struct trace_entry *entry;
1966         struct trace_event *event;
1967
1968         entry = iter->ent;
1969
1970         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1971                 SEQ_PUT_FIELD_RET(s, entry->pid);
1972                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1973                 SEQ_PUT_FIELD_RET(s, iter->ts);
1974         }
1975
1976         event = ftrace_find_event(entry->type);
1977         return event ? event->funcs->binary(iter, 0, event) :
1978                 TRACE_TYPE_HANDLED;
1979 }
1980
1981 int trace_empty(struct trace_iterator *iter)
1982 {
1983         int cpu;
1984
1985         /* If we are looking at one CPU buffer, only check that one */
1986         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1987                 cpu = iter->cpu_file;
1988                 if (iter->buffer_iter[cpu]) {
1989                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1990                                 return 0;
1991                 } else {
1992                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1993                                 return 0;
1994                 }
1995                 return 1;
1996         }
1997
1998         for_each_tracing_cpu(cpu) {
1999                 if (iter->buffer_iter[cpu]) {
2000                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2001                                 return 0;
2002                 } else {
2003                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2004                                 return 0;
2005                 }
2006         }
2007
2008         return 1;
2009 }
2010
2011 /*  Called with trace_event_read_lock() held. */
2012 enum print_line_t print_trace_line(struct trace_iterator *iter)
2013 {
2014         enum print_line_t ret;
2015
2016         if (iter->lost_events)
2017                 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2018                                  iter->cpu, iter->lost_events);
2019
2020         if (iter->trace && iter->trace->print_line) {
2021                 ret = iter->trace->print_line(iter);
2022                 if (ret != TRACE_TYPE_UNHANDLED)
2023                         return ret;
2024         }
2025
2026         if (iter->ent->type == TRACE_BPRINT &&
2027                         trace_flags & TRACE_ITER_PRINTK &&
2028                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2029                 return trace_print_bprintk_msg_only(iter);
2030
2031         if (iter->ent->type == TRACE_PRINT &&
2032                         trace_flags & TRACE_ITER_PRINTK &&
2033                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2034                 return trace_print_printk_msg_only(iter);
2035
2036         if (trace_flags & TRACE_ITER_BIN)
2037                 return print_bin_fmt(iter);
2038
2039         if (trace_flags & TRACE_ITER_HEX)
2040                 return print_hex_fmt(iter);
2041
2042         if (trace_flags & TRACE_ITER_RAW)
2043                 return print_raw_fmt(iter);
2044
2045         return print_trace_fmt(iter);
2046 }
2047
2048 void trace_default_header(struct seq_file *m)
2049 {
2050         struct trace_iterator *iter = m->private;
2051
2052         if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2053                 /* print nothing if the buffers are empty */
2054                 if (trace_empty(iter))
2055                         return;
2056                 print_trace_header(m, iter);
2057                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2058                         print_lat_help_header(m);
2059         } else {
2060                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2061                         print_func_help_header(m);
2062         }
2063 }
2064
2065 static int s_show(struct seq_file *m, void *v)
2066 {
2067         struct trace_iterator *iter = v;
2068         int ret;
2069
2070         if (iter->ent == NULL) {
2071                 if (iter->tr) {
2072                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2073                         seq_puts(m, "#\n");
2074                 }
2075                 if (iter->trace && iter->trace->print_header)
2076                         iter->trace->print_header(m);
2077                 else
2078                         trace_default_header(m);
2079
2080         } else if (iter->leftover) {
2081                 /*
2082                  * If we filled the seq_file buffer earlier, we
2083                  * want to just show it now.
2084                  */
2085                 ret = trace_print_seq(m, &iter->seq);
2086
2087                 /* ret should this time be zero, but you never know */
2088                 iter->leftover = ret;
2089
2090         } else {
2091                 print_trace_line(iter);
2092                 ret = trace_print_seq(m, &iter->seq);
2093                 /*
2094                  * If we overflow the seq_file buffer, then it will
2095                  * ask us for this data again at start up.
2096                  * Use that instead.
2097                  *  ret is 0 if seq_file write succeeded.
2098                  *        -1 otherwise.
2099                  */
2100                 iter->leftover = ret;
2101         }
2102
2103         return 0;
2104 }
2105
2106 static const struct seq_operations tracer_seq_ops = {
2107         .start          = s_start,
2108         .next           = s_next,
2109         .stop           = s_stop,
2110         .show           = s_show,
2111 };
2112
2113 static struct trace_iterator *
2114 __tracing_open(struct inode *inode, struct file *file)
2115 {
2116         long cpu_file = (long) inode->i_private;
2117         void *fail_ret = ERR_PTR(-ENOMEM);
2118         struct trace_iterator *iter;
2119         struct seq_file *m;
2120         int cpu, ret;
2121
2122         if (tracing_disabled)
2123                 return ERR_PTR(-ENODEV);
2124
2125         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2126         if (!iter)
2127                 return ERR_PTR(-ENOMEM);
2128
2129         /*
2130          * We make a copy of the current tracer to avoid concurrent
2131          * changes on it while we are reading.
2132          */
2133         mutex_lock(&trace_types_lock);
2134         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2135         if (!iter->trace)
2136                 goto fail;
2137
2138         if (current_trace)
2139                 *iter->trace = *current_trace;
2140
2141         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2142                 goto fail;
2143
2144         if (current_trace && current_trace->print_max)
2145                 iter->tr = &max_tr;
2146         else
2147                 iter->tr = &global_trace;
2148         iter->pos = -1;
2149         mutex_init(&iter->mutex);
2150         iter->cpu_file = cpu_file;
2151
2152         /* Notify the tracer early; before we stop tracing. */
2153         if (iter->trace && iter->trace->open)
2154                 iter->trace->open(iter);
2155
2156         /* Annotate start of buffers if we had overruns */
2157         if (ring_buffer_overruns(iter->tr->buffer))
2158                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2159
2160         /* stop the trace while dumping */
2161         tracing_stop();
2162
2163         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2164                 for_each_tracing_cpu(cpu) {
2165                         iter->buffer_iter[cpu] =
2166                                 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2167                 }
2168                 ring_buffer_read_prepare_sync();
2169                 for_each_tracing_cpu(cpu) {
2170                         ring_buffer_read_start(iter->buffer_iter[cpu]);
2171                         tracing_iter_reset(iter, cpu);
2172                 }
2173         } else {
2174                 cpu = iter->cpu_file;
2175                 iter->buffer_iter[cpu] =
2176                         ring_buffer_read_prepare(iter->tr->buffer, cpu);
2177                 ring_buffer_read_prepare_sync();
2178                 ring_buffer_read_start(iter->buffer_iter[cpu]);
2179                 tracing_iter_reset(iter, cpu);
2180         }
2181
2182         ret = seq_open(file, &tracer_seq_ops);
2183         if (ret < 0) {
2184                 fail_ret = ERR_PTR(ret);
2185                 goto fail_buffer;
2186         }
2187
2188         m = file->private_data;
2189         m->private = iter;
2190
2191         mutex_unlock(&trace_types_lock);
2192
2193         return iter;
2194
2195  fail_buffer:
2196         for_each_tracing_cpu(cpu) {
2197                 if (iter->buffer_iter[cpu])
2198                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2199         }
2200         free_cpumask_var(iter->started);
2201         tracing_start();
2202  fail:
2203         mutex_unlock(&trace_types_lock);
2204         kfree(iter->trace);
2205         kfree(iter);
2206
2207         return fail_ret;
2208 }
2209
2210 int tracing_open_generic(struct inode *inode, struct file *filp)
2211 {
2212         if (tracing_disabled)
2213                 return -ENODEV;
2214
2215         filp->private_data = inode->i_private;
2216         return 0;
2217 }
2218
2219 static int tracing_release(struct inode *inode, struct file *file)
2220 {
2221         struct seq_file *m = file->private_data;
2222         struct trace_iterator *iter;
2223         int cpu;
2224
2225         if (!(file->f_mode & FMODE_READ))
2226                 return 0;
2227
2228         iter = m->private;
2229
2230         mutex_lock(&trace_types_lock);
2231         for_each_tracing_cpu(cpu) {
2232                 if (iter->buffer_iter[cpu])
2233                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2234         }
2235
2236         if (iter->trace && iter->trace->close)
2237                 iter->trace->close(iter);
2238
2239         /* reenable tracing if it was previously enabled */
2240         tracing_start();
2241         mutex_unlock(&trace_types_lock);
2242
2243         seq_release(inode, file);
2244         mutex_destroy(&iter->mutex);
2245         free_cpumask_var(iter->started);
2246         kfree(iter->trace);
2247         kfree(iter);
2248         return 0;
2249 }
2250
2251 static int tracing_open(struct inode *inode, struct file *file)
2252 {
2253         struct trace_iterator *iter;
2254         int ret = 0;
2255
2256         /* If this file was open for write, then erase contents */
2257         if ((file->f_mode & FMODE_WRITE) &&
2258             (file->f_flags & O_TRUNC)) {
2259                 long cpu = (long) inode->i_private;
2260
2261                 if (cpu == TRACE_PIPE_ALL_CPU)
2262                         tracing_reset_online_cpus(&global_trace);
2263                 else
2264                         tracing_reset(&global_trace, cpu);
2265         }
2266
2267         if (file->f_mode & FMODE_READ) {
2268                 iter = __tracing_open(inode, file);
2269                 if (IS_ERR(iter))
2270                         ret = PTR_ERR(iter);
2271                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2272                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2273         }
2274         return ret;
2275 }
2276
2277 static void *
2278 t_next(struct seq_file *m, void *v, loff_t *pos)
2279 {
2280         struct tracer *t = v;
2281
2282         (*pos)++;
2283
2284         if (t)
2285                 t = t->next;
2286
2287         return t;
2288 }
2289
2290 static void *t_start(struct seq_file *m, loff_t *pos)
2291 {
2292         struct tracer *t;
2293         loff_t l = 0;
2294
2295         mutex_lock(&trace_types_lock);
2296         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2297                 ;
2298
2299         return t;
2300 }
2301
2302 static void t_stop(struct seq_file *m, void *p)
2303 {
2304         mutex_unlock(&trace_types_lock);
2305 }
2306
2307 static int t_show(struct seq_file *m, void *v)
2308 {
2309         struct tracer *t = v;
2310
2311         if (!t)
2312                 return 0;
2313
2314         seq_printf(m, "%s", t->name);
2315         if (t->next)
2316                 seq_putc(m, ' ');
2317         else
2318                 seq_putc(m, '\n');
2319
2320         return 0;
2321 }
2322
2323 static const struct seq_operations show_traces_seq_ops = {
2324         .start          = t_start,
2325         .next           = t_next,
2326         .stop           = t_stop,
2327         .show           = t_show,
2328 };
2329
2330 static int show_traces_open(struct inode *inode, struct file *file)
2331 {
2332         if (tracing_disabled)
2333                 return -ENODEV;
2334
2335         return seq_open(file, &show_traces_seq_ops);
2336 }
2337
2338 static ssize_t
2339 tracing_write_stub(struct file *filp, const char __user *ubuf,
2340                    size_t count, loff_t *ppos)
2341 {
2342         return count;
2343 }
2344
2345 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2346 {
2347         if (file->f_mode & FMODE_READ)
2348                 return seq_lseek(file, offset, origin);
2349         else
2350                 return 0;
2351 }
2352
2353 static const struct file_operations tracing_fops = {
2354         .open           = tracing_open,
2355         .read           = seq_read,
2356         .write          = tracing_write_stub,
2357         .llseek         = tracing_seek,
2358         .release        = tracing_release,
2359 };
2360
2361 static const struct file_operations show_traces_fops = {
2362         .open           = show_traces_open,
2363         .read           = seq_read,
2364         .release        = seq_release,
2365         .llseek         = seq_lseek,
2366 };
2367
2368 /*
2369  * Only trace on a CPU if the bitmask is set:
2370  */
2371 static cpumask_var_t tracing_cpumask;
2372
2373 /*
2374  * The tracer itself will not take this lock, but still we want
2375  * to provide a consistent cpumask to user-space:
2376  */
2377 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2378
2379 /*
2380  * Temporary storage for the character representation of the
2381  * CPU bitmask (and one more byte for the newline):
2382  */
2383 static char mask_str[NR_CPUS + 1];
2384
2385 static ssize_t
2386 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2387                      size_t count, loff_t *ppos)
2388 {
2389         int len;
2390
2391         mutex_lock(&tracing_cpumask_update_lock);
2392
2393         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2394         if (count - len < 2) {
2395                 count = -EINVAL;
2396                 goto out_err;
2397         }
2398         len += sprintf(mask_str + len, "\n");
2399         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2400
2401 out_err:
2402         mutex_unlock(&tracing_cpumask_update_lock);
2403
2404         return count;
2405 }
2406
2407 static ssize_t
2408 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2409                       size_t count, loff_t *ppos)
2410 {
2411         int err, cpu;
2412         cpumask_var_t tracing_cpumask_new;
2413
2414         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2415                 return -ENOMEM;
2416
2417         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2418         if (err)
2419                 goto err_unlock;
2420
2421         mutex_lock(&tracing_cpumask_update_lock);
2422
2423         local_irq_disable();
2424         arch_spin_lock(&ftrace_max_lock);
2425         for_each_tracing_cpu(cpu) {
2426                 /*
2427                  * Increase/decrease the disabled counter if we are
2428                  * about to flip a bit in the cpumask:
2429                  */
2430                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2431                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2432                         atomic_inc(&global_trace.data[cpu]->disabled);
2433                 }
2434                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2435                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2436                         atomic_dec(&global_trace.data[cpu]->disabled);
2437                 }
2438         }
2439         arch_spin_unlock(&ftrace_max_lock);
2440         local_irq_enable();
2441
2442         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2443
2444         mutex_unlock(&tracing_cpumask_update_lock);
2445         free_cpumask_var(tracing_cpumask_new);
2446
2447         return count;
2448
2449 err_unlock:
2450         free_cpumask_var(tracing_cpumask_new);
2451
2452         return err;
2453 }
2454
2455 static const struct file_operations tracing_cpumask_fops = {
2456         .open           = tracing_open_generic,
2457         .read           = tracing_cpumask_read,
2458         .write          = tracing_cpumask_write,
2459         .llseek         = generic_file_llseek,
2460 };
2461
2462 static int tracing_trace_options_show(struct seq_file *m, void *v)
2463 {
2464         struct tracer_opt *trace_opts;
2465         u32 tracer_flags;
2466         int i;
2467
2468         mutex_lock(&trace_types_lock);
2469         tracer_flags = current_trace->flags->val;
2470         trace_opts = current_trace->flags->opts;
2471
2472         for (i = 0; trace_options[i]; i++) {
2473                 if (trace_flags & (1 << i))
2474                         seq_printf(m, "%s\n", trace_options[i]);
2475                 else
2476                         seq_printf(m, "no%s\n", trace_options[i]);
2477         }
2478
2479         for (i = 0; trace_opts[i].name; i++) {
2480                 if (tracer_flags & trace_opts[i].bit)
2481                         seq_printf(m, "%s\n", trace_opts[i].name);
2482                 else
2483                         seq_printf(m, "no%s\n", trace_opts[i].name);
2484         }
2485         mutex_unlock(&trace_types_lock);
2486
2487         return 0;
2488 }
2489
2490 static int __set_tracer_option(struct tracer *trace,
2491                                struct tracer_flags *tracer_flags,
2492                                struct tracer_opt *opts, int neg)
2493 {
2494         int ret;
2495
2496         ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2497         if (ret)
2498                 return ret;
2499
2500         if (neg)
2501                 tracer_flags->val &= ~opts->bit;
2502         else
2503                 tracer_flags->val |= opts->bit;
2504         return 0;
2505 }
2506
2507 /* Try to assign a tracer specific option */
2508 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2509 {
2510         struct tracer_flags *tracer_flags = trace->flags;
2511         struct tracer_opt *opts = NULL;
2512         int i;
2513
2514         for (i = 0; tracer_flags->opts[i].name; i++) {
2515                 opts = &tracer_flags->opts[i];
2516
2517                 if (strcmp(cmp, opts->name) == 0)
2518                         return __set_tracer_option(trace, trace->flags,
2519                                                    opts, neg);
2520         }
2521
2522         return -EINVAL;
2523 }
2524
2525 static void set_tracer_flags(unsigned int mask, int enabled)
2526 {
2527         /* do nothing if flag is already set */
2528         if (!!(trace_flags & mask) == !!enabled)
2529                 return;
2530
2531         if (enabled)
2532                 trace_flags |= mask;
2533         else
2534                 trace_flags &= ~mask;
2535
2536         if (mask == TRACE_ITER_RECORD_CMD)
2537                 trace_event_enable_cmd_record(enabled);
2538
2539         if (mask == TRACE_ITER_OVERWRITE)
2540                 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2541 }
2542
2543 static ssize_t
2544 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2545                         size_t cnt, loff_t *ppos)
2546 {
2547         char buf[64];
2548         char *cmp;
2549         int neg = 0;
2550         int ret;
2551         int i;
2552
2553         if (cnt >= sizeof(buf))
2554                 return -EINVAL;
2555
2556         if (copy_from_user(&buf, ubuf, cnt))
2557                 return -EFAULT;
2558
2559         buf[cnt] = 0;
2560         cmp = strstrip(buf);
2561
2562         if (strncmp(cmp, "no", 2) == 0) {
2563                 neg = 1;
2564                 cmp += 2;
2565         }
2566
2567         for (i = 0; trace_options[i]; i++) {
2568                 if (strcmp(cmp, trace_options[i]) == 0) {
2569                         set_tracer_flags(1 << i, !neg);
2570                         break;
2571                 }
2572         }
2573
2574         /* If no option could be set, test the specific tracer options */
2575         if (!trace_options[i]) {
2576                 mutex_lock(&trace_types_lock);
2577                 ret = set_tracer_option(current_trace, cmp, neg);
2578                 mutex_unlock(&trace_types_lock);
2579                 if (ret)
2580                         return ret;
2581         }
2582
2583         *ppos += cnt;
2584
2585         return cnt;
2586 }
2587
2588 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2589 {
2590         if (tracing_disabled)
2591                 return -ENODEV;
2592         return single_open(file, tracing_trace_options_show, NULL);
2593 }
2594
2595 static const struct file_operations tracing_iter_fops = {
2596         .open           = tracing_trace_options_open,
2597         .read           = seq_read,
2598         .llseek         = seq_lseek,
2599         .release        = single_release,
2600         .write          = tracing_trace_options_write,
2601 };
2602
2603 static const char readme_msg[] =
2604         "tracing mini-HOWTO:\n\n"
2605         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2606         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2607         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2608         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2609         "nop\n"
2610         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2611         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2612         "sched_switch\n"
2613         "# cat /sys/kernel/debug/tracing/trace_options\n"
2614         "noprint-parent nosym-offset nosym-addr noverbose\n"
2615         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2616         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2617         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2618         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2619 ;
2620
2621 static ssize_t
2622 tracing_readme_read(struct file *filp, char __user *ubuf,
2623                        size_t cnt, loff_t *ppos)
2624 {
2625         return simple_read_from_buffer(ubuf, cnt, ppos,
2626                                         readme_msg, strlen(readme_msg));
2627 }
2628
2629 static const struct file_operations tracing_readme_fops = {
2630         .open           = tracing_open_generic,
2631         .read           = tracing_readme_read,
2632         .llseek         = generic_file_llseek,
2633 };
2634
2635 static ssize_t
2636 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2637                                 size_t cnt, loff_t *ppos)
2638 {
2639         char *buf_comm;
2640         char *file_buf;
2641         char *buf;
2642         int len = 0;
2643         int pid;
2644         int i;
2645
2646         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2647         if (!file_buf)
2648                 return -ENOMEM;
2649
2650         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2651         if (!buf_comm) {
2652                 kfree(file_buf);
2653                 return -ENOMEM;
2654         }
2655
2656         buf = file_buf;
2657
2658         for (i = 0; i < SAVED_CMDLINES; i++) {
2659                 int r;
2660
2661                 pid = map_cmdline_to_pid[i];
2662                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2663                         continue;
2664
2665                 trace_find_cmdline(pid, buf_comm);
2666                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2667                 buf += r;
2668                 len += r;
2669         }
2670
2671         len = simple_read_from_buffer(ubuf, cnt, ppos,
2672                                       file_buf, len);
2673
2674         kfree(file_buf);
2675         kfree(buf_comm);
2676
2677         return len;
2678 }
2679
2680 static const struct file_operations tracing_saved_cmdlines_fops = {
2681     .open       = tracing_open_generic,
2682     .read       = tracing_saved_cmdlines_read,
2683     .llseek     = generic_file_llseek,
2684 };
2685
2686 static ssize_t
2687 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2688                   size_t cnt, loff_t *ppos)
2689 {
2690         char buf[64];
2691         int r;
2692
2693         r = sprintf(buf, "%u\n", tracer_enabled);
2694         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2695 }
2696
2697 static ssize_t
2698 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2699                    size_t cnt, loff_t *ppos)
2700 {
2701         struct trace_array *tr = filp->private_data;
2702         char buf[64];
2703         unsigned long val;
2704         int ret;
2705
2706         if (cnt >= sizeof(buf))
2707                 return -EINVAL;
2708
2709         if (copy_from_user(&buf, ubuf, cnt))
2710                 return -EFAULT;
2711
2712         buf[cnt] = 0;
2713
2714         ret = strict_strtoul(buf, 10, &val);
2715         if (ret < 0)
2716                 return ret;
2717
2718         val = !!val;
2719
2720         mutex_lock(&trace_types_lock);
2721         if (tracer_enabled ^ val) {
2722
2723                 /* Only need to warn if this is used to change the state */
2724                 WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2725
2726                 if (val) {
2727                         tracer_enabled = 1;
2728                         if (current_trace->start)
2729                                 current_trace->start(tr);
2730                         tracing_start();
2731                 } else {
2732                         tracer_enabled = 0;
2733                         tracing_stop();
2734                         if (current_trace->stop)
2735                                 current_trace->stop(tr);
2736                 }
2737         }
2738         mutex_unlock(&trace_types_lock);
2739
2740         *ppos += cnt;
2741
2742         return cnt;
2743 }
2744
2745 static ssize_t
2746 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2747                        size_t cnt, loff_t *ppos)
2748 {
2749         char buf[MAX_TRACER_SIZE+2];
2750         int r;
2751
2752         mutex_lock(&trace_types_lock);
2753         if (current_trace)
2754                 r = sprintf(buf, "%s\n", current_trace->name);
2755         else
2756                 r = sprintf(buf, "\n");
2757         mutex_unlock(&trace_types_lock);
2758
2759         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2760 }
2761
2762 int tracer_init(struct tracer *t, struct trace_array *tr)
2763 {
2764         tracing_reset_online_cpus(tr);
2765         return t->init(tr);
2766 }
2767
2768 static int tracing_resize_ring_buffer(unsigned long size)
2769 {
2770         int ret;
2771
2772         /*
2773          * If kernel or user changes the size of the ring buffer
2774          * we use the size that was given, and we can forget about
2775          * expanding it later.
2776          */
2777         ring_buffer_expanded = 1;
2778
2779         ret = ring_buffer_resize(global_trace.buffer, size);
2780         if (ret < 0)
2781                 return ret;
2782
2783         if (!current_trace->use_max_tr)
2784                 goto out;
2785
2786         ret = ring_buffer_resize(max_tr.buffer, size);
2787         if (ret < 0) {
2788                 int r;
2789
2790                 r = ring_buffer_resize(global_trace.buffer,
2791                                        global_trace.entries);
2792                 if (r < 0) {
2793                         /*
2794                          * AARGH! We are left with different
2795                          * size max buffer!!!!
2796                          * The max buffer is our "snapshot" buffer.
2797                          * When a tracer needs a snapshot (one of the
2798                          * latency tracers), it swaps the max buffer
2799                          * with the saved snap shot. We succeeded to
2800                          * update the size of the main buffer, but failed to
2801                          * update the size of the max buffer. But when we tried
2802                          * to reset the main buffer to the original size, we
2803                          * failed there too. This is very unlikely to
2804                          * happen, but if it does, warn and kill all
2805                          * tracing.
2806                          */
2807                         WARN_ON(1);
2808                         tracing_disabled = 1;
2809                 }
2810                 return ret;
2811         }
2812
2813         max_tr.entries = size;
2814  out:
2815         global_trace.entries = size;
2816
2817         return ret;
2818 }
2819
2820
2821 /**
2822  * tracing_update_buffers - used by tracing facility to expand ring buffers
2823  *
2824  * To save on memory when the tracing is never used on a system with it
2825  * configured in. The ring buffers are set to a minimum size. But once
2826  * a user starts to use the tracing facility, then they need to grow
2827  * to their default size.
2828  *
2829  * This function is to be called when a tracer is about to be used.
2830  */
2831 int tracing_update_buffers(void)
2832 {
2833         int ret = 0;
2834
2835         mutex_lock(&trace_types_lock);
2836         if (!ring_buffer_expanded)
2837                 ret = tracing_resize_ring_buffer(trace_buf_size);
2838         mutex_unlock(&trace_types_lock);
2839
2840         return ret;
2841 }
2842
2843 struct trace_option_dentry;
2844
2845 static struct trace_option_dentry *
2846 create_trace_option_files(struct tracer *tracer);
2847
2848 static void
2849 destroy_trace_option_files(struct trace_option_dentry *topts);
2850
2851 static int tracing_set_tracer(const char *buf)
2852 {
2853         static struct trace_option_dentry *topts;
2854         struct trace_array *tr = &global_trace;
2855         struct tracer *t;
2856         int ret = 0;
2857
2858         mutex_lock(&trace_types_lock);
2859
2860         if (!ring_buffer_expanded) {
2861                 ret = tracing_resize_ring_buffer(trace_buf_size);
2862                 if (ret < 0)
2863                         goto out;
2864                 ret = 0;
2865         }
2866
2867         for (t = trace_types; t; t = t->next) {
2868                 if (strcmp(t->name, buf) == 0)
2869                         break;
2870         }
2871         if (!t) {
2872                 ret = -EINVAL;
2873                 goto out;
2874         }
2875         if (t == current_trace)
2876                 goto out;
2877
2878         trace_branch_disable();
2879         if (current_trace && current_trace->reset)
2880                 current_trace->reset(tr);
2881         if (current_trace && current_trace->use_max_tr) {
2882                 /*
2883                  * We don't free the ring buffer. instead, resize it because
2884                  * The max_tr ring buffer has some state (e.g. ring->clock) and
2885                  * we want preserve it.
2886                  */
2887                 ring_buffer_resize(max_tr.buffer, 1);
2888                 max_tr.entries = 1;
2889         }
2890         destroy_trace_option_files(topts);
2891
2892         current_trace = t;
2893
2894         topts = create_trace_option_files(current_trace);
2895         if (current_trace->use_max_tr) {
2896                 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
2897                 if (ret < 0)
2898                         goto out;
2899                 max_tr.entries = global_trace.entries;
2900         }
2901
2902         if (t->init) {
2903                 ret = tracer_init(t, tr);
2904                 if (ret)
2905                         goto out;
2906         }
2907
2908         trace_branch_enable(tr);
2909  out:
2910         mutex_unlock(&trace_types_lock);
2911
2912         return ret;
2913 }
2914
2915 static ssize_t
2916 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2917                         size_t cnt, loff_t *ppos)
2918 {
2919         char buf[MAX_TRACER_SIZE+1];
2920         int i;
2921         size_t ret;
2922         int err;
2923
2924         ret = cnt;
2925
2926         if (cnt > MAX_TRACER_SIZE)
2927                 cnt = MAX_TRACER_SIZE;
2928
2929         if (copy_from_user(&buf, ubuf, cnt))
2930                 return -EFAULT;
2931
2932         buf[cnt] = 0;
2933
2934         /* strip ending whitespace. */
2935         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2936                 buf[i] = 0;
2937
2938         err = tracing_set_tracer(buf);
2939         if (err)
2940                 return err;
2941
2942         *ppos += ret;
2943
2944         return ret;
2945 }
2946
2947 static ssize_t
2948 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2949                      size_t cnt, loff_t *ppos)
2950 {
2951         unsigned long *ptr = filp->private_data;
2952         char buf[64];
2953         int r;
2954
2955         r = snprintf(buf, sizeof(buf), "%ld\n",
2956                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2957         if (r > sizeof(buf))
2958                 r = sizeof(buf);
2959         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2960 }
2961
2962 static ssize_t
2963 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2964                       size_t cnt, loff_t *ppos)
2965 {
2966         unsigned long *ptr = filp->private_data;
2967         char buf[64];
2968         unsigned long val;
2969         int ret;
2970
2971         if (cnt >= sizeof(buf))
2972                 return -EINVAL;
2973
2974         if (copy_from_user(&buf, ubuf, cnt))
2975                 return -EFAULT;
2976
2977         buf[cnt] = 0;
2978
2979         ret = strict_strtoul(buf, 10, &val);
2980         if (ret < 0)
2981                 return ret;
2982
2983         *ptr = val * 1000;
2984
2985         return cnt;
2986 }
2987
2988 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2989 {
2990         long cpu_file = (long) inode->i_private;
2991         struct trace_iterator *iter;
2992         int ret = 0;
2993
2994         if (tracing_disabled)
2995                 return -ENODEV;
2996
2997         mutex_lock(&trace_types_lock);
2998
2999         /* create a buffer to store the information to pass to userspace */
3000         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3001         if (!iter) {
3002                 ret = -ENOMEM;
3003                 goto out;
3004         }
3005
3006         /*
3007          * We make a copy of the current tracer to avoid concurrent
3008          * changes on it while we are reading.
3009          */
3010         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3011         if (!iter->trace) {
3012                 ret = -ENOMEM;
3013                 goto fail;
3014         }
3015         if (current_trace)
3016                 *iter->trace = *current_trace;
3017
3018         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3019                 ret = -ENOMEM;
3020                 goto fail;
3021         }
3022
3023         /* trace pipe does not show start of buffer */
3024         cpumask_setall(iter->started);
3025
3026         if (trace_flags & TRACE_ITER_LATENCY_FMT)
3027                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3028
3029         iter->cpu_file = cpu_file;
3030         iter->tr = &global_trace;
3031         mutex_init(&iter->mutex);
3032         filp->private_data = iter;
3033
3034         if (iter->trace->pipe_open)
3035                 iter->trace->pipe_open(iter);
3036
3037         nonseekable_open(inode, filp);
3038 out:
3039         mutex_unlock(&trace_types_lock);
3040         return ret;
3041
3042 fail:
3043         kfree(iter->trace);
3044         kfree(iter);
3045         mutex_unlock(&trace_types_lock);
3046         return ret;
3047 }
3048
3049 static int tracing_release_pipe(struct inode *inode, struct file *file)
3050 {
3051         struct trace_iterator *iter = file->private_data;
3052
3053         mutex_lock(&trace_types_lock);
3054
3055         if (iter->trace->pipe_close)
3056                 iter->trace->pipe_close(iter);
3057
3058         mutex_unlock(&trace_types_lock);
3059
3060         free_cpumask_var(iter->started);
3061         mutex_destroy(&iter->mutex);
3062         kfree(iter->trace);
3063         kfree(iter);
3064
3065         return 0;
3066 }
3067
3068 static unsigned int
3069 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3070 {
3071         struct trace_iterator *iter = filp->private_data;
3072
3073         if (trace_flags & TRACE_ITER_BLOCK) {
3074                 /*
3075                  * Always select as readable when in blocking mode
3076                  */
3077                 return POLLIN | POLLRDNORM;
3078         } else {
3079                 if (!trace_empty(iter))
3080                         return POLLIN | POLLRDNORM;
3081                 poll_wait(filp, &trace_wait, poll_table);
3082                 if (!trace_empty(iter))
3083                         return POLLIN | POLLRDNORM;
3084
3085                 return 0;
3086         }
3087 }
3088
3089
3090 void default_wait_pipe(struct trace_iterator *iter)
3091 {
3092         DEFINE_WAIT(wait);
3093
3094         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3095
3096         if (trace_empty(iter))
3097                 schedule();
3098
3099         finish_wait(&trace_wait, &wait);
3100 }
3101
3102 /*
3103  * This is a make-shift waitqueue.
3104  * A tracer might use this callback on some rare cases:
3105  *
3106  *  1) the current tracer might hold the runqueue lock when it wakes up
3107  *     a reader, hence a deadlock (sched, function, and function graph tracers)
3108  *  2) the function tracers, trace all functions, we don't want
3109  *     the overhead of calling wake_up and friends
3110  *     (and tracing them too)
3111  *
3112  *     Anyway, this is really very primitive wakeup.
3113  */
3114 void poll_wait_pipe(struct trace_iterator *iter)
3115 {
3116         set_current_state(TASK_INTERRUPTIBLE);
3117         /* sleep for 100 msecs, and try again. */
3118         schedule_timeout(HZ / 10);
3119 }
3120
3121 /* Must be called with trace_types_lock mutex held. */
3122 static int tracing_wait_pipe(struct file *filp)
3123 {
3124         struct trace_iterator *iter = filp->private_data;
3125
3126         while (trace_empty(iter)) {
3127
3128                 if ((filp->f_flags & O_NONBLOCK)) {
3129                         return -EAGAIN;
3130                 }
3131
3132                 mutex_unlock(&iter->mutex);
3133
3134                 iter->trace->wait_pipe(iter);
3135
3136                 mutex_lock(&iter->mutex);
3137
3138                 if (signal_pending(current))
3139                         return -EINTR;
3140
3141                 /*
3142                  * We block until we read something and tracing is disabled.
3143                  * We still block if tracing is disabled, but we have never
3144                  * read anything. This allows a user to cat this file, and
3145                  * then enable tracing. But after we have read something,
3146                  * we give an EOF when tracing is again disabled.
3147                  *
3148                  * iter->pos will be 0 if we haven't read anything.
3149                  */
3150                 if (!tracer_enabled && iter->pos)
3151                         break;
3152         }
3153
3154         return 1;
3155 }
3156
3157 /*
3158  * Consumer reader.
3159  */
3160 static ssize_t
3161 tracing_read_pipe(struct file *filp, char __user *ubuf,
3162                   size_t cnt, loff_t *ppos)
3163 {
3164         struct trace_iterator *iter = filp->private_data;
3165         static struct tracer *old_tracer;
3166         ssize_t sret;
3167
3168         /* return any leftover data */
3169         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3170         if (sret != -EBUSY)
3171                 return sret;
3172
3173         trace_seq_init(&iter->seq);
3174
3175         /* copy the tracer to avoid using a global lock all around */
3176         mutex_lock(&trace_types_lock);
3177         if (unlikely(old_tracer != current_trace && current_trace)) {
3178                 old_tracer = current_trace;
3179                 *iter->trace = *current_trace;
3180         }
3181         mutex_unlock(&trace_types_lock);
3182
3183         /*
3184          * Avoid more than one consumer on a single file descriptor
3185          * This is just a matter of traces coherency, the ring buffer itself
3186          * is protected.
3187          */
3188         mutex_lock(&iter->mutex);
3189         if (iter->trace->read) {
3190                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3191                 if (sret)
3192                         goto out;
3193         }
3194
3195 waitagain:
3196         sret = tracing_wait_pipe(filp);
3197         if (sret <= 0)
3198                 goto out;
3199
3200         /* stop when tracing is finished */
3201         if (trace_empty(iter)) {
3202                 sret = 0;
3203                 goto out;
3204         }
3205
3206         if (cnt >= PAGE_SIZE)
3207                 cnt = PAGE_SIZE - 1;
3208
3209         /* reset all but tr, trace, and overruns */
3210         memset(&iter->seq, 0,
3211                sizeof(struct trace_iterator) -
3212                offsetof(struct trace_iterator, seq));
3213         iter->pos = -1;
3214
3215         trace_event_read_lock();
3216         trace_access_lock(iter->cpu_file);
3217         while (trace_find_next_entry_inc(iter) != NULL) {
3218                 enum print_line_t ret;
3219                 int len = iter->seq.len;
3220
3221                 ret = print_trace_line(iter);
3222                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3223                         /* don't print partial lines */
3224                         iter->seq.len = len;
3225                         break;
3226                 }
3227                 if (ret != TRACE_TYPE_NO_CONSUME)
3228                         trace_consume(iter);
3229
3230                 if (iter->seq.len >= cnt)
3231                         break;
3232         }
3233         trace_access_unlock(iter->cpu_file);
3234         trace_event_read_unlock();
3235
3236         /* Now copy what we have to the user */
3237         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3238         if (iter->seq.readpos >= iter->seq.len)
3239                 trace_seq_init(&iter->seq);
3240
3241         /*
3242          * If there was nothing to send to user, inspite of consuming trace
3243          * entries, go back to wait for more entries.
3244          */
3245         if (sret == -EBUSY)
3246                 goto waitagain;
3247
3248 out:
3249         mutex_unlock(&iter->mutex);
3250
3251         return sret;
3252 }
3253
3254 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3255                                      struct pipe_buffer *buf)
3256 {
3257         __free_page(buf->page);
3258 }
3259
3260 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3261                                      unsigned int idx)
3262 {
3263         __free_page(spd->pages[idx]);
3264 }
3265
3266 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3267         .can_merge              = 0,
3268         .map                    = generic_pipe_buf_map,
3269         .unmap                  = generic_pipe_buf_unmap,
3270         .confirm                = generic_pipe_buf_confirm,
3271         .release                = tracing_pipe_buf_release,
3272         .steal                  = generic_pipe_buf_steal,
3273         .get                    = generic_pipe_buf_get,
3274 };
3275
3276 static size_t
3277 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3278 {
3279         size_t count;
3280         int ret;
3281
3282         /* Seq buffer is page-sized, exactly what we need. */
3283         for (;;) {
3284                 count = iter->seq.len;
3285                 ret = print_trace_line(iter);
3286                 count = iter->seq.len - count;
3287                 if (rem < count) {
3288                         rem = 0;
3289                         iter->seq.len -= count;
3290                         break;
3291                 }
3292                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3293                         iter->seq.len -= count;
3294                         break;
3295                 }
3296
3297                 if (ret != TRACE_TYPE_NO_CONSUME)
3298                         trace_consume(iter);
3299                 rem -= count;
3300                 if (!trace_find_next_entry_inc(iter))   {
3301                         rem = 0;
3302                         iter->ent = NULL;
3303                         break;
3304                 }
3305         }
3306
3307         return rem;
3308 }
3309
3310 static ssize_t tracing_splice_read_pipe(struct file *filp,
3311                                         loff_t *ppos,
3312                                         struct pipe_inode_info *pipe,
3313                                         size_t len,
3314                                         unsigned int flags)
3315 {
3316         struct page *pages_def[PIPE_DEF_BUFFERS];
3317         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3318         struct trace_iterator *iter = filp->private_data;
3319         struct splice_pipe_desc spd = {
3320                 .pages          = pages_def,
3321                 .partial        = partial_def,
3322                 .nr_pages       = 0, /* This gets updated below. */
3323                 .flags          = flags,
3324                 .ops            = &tracing_pipe_buf_ops,
3325                 .spd_release    = tracing_spd_release_pipe,
3326         };
3327         static struct tracer *old_tracer;
3328         ssize_t ret;
3329         size_t rem;
3330         unsigned int i;
3331
3332         if (splice_grow_spd(pipe, &spd))
3333                 return -ENOMEM;
3334
3335         /* copy the tracer to avoid using a global lock all around */
3336         mutex_lock(&trace_types_lock);
3337         if (unlikely(old_tracer != current_trace && current_trace)) {
3338                 old_tracer = current_trace;
3339                 *iter->trace = *current_trace;
3340         }
3341         mutex_unlock(&trace_types_lock);
3342
3343         mutex_lock(&iter->mutex);
3344
3345         if (iter->trace->splice_read) {
3346                 ret = iter->trace->splice_read(iter, filp,
3347                                                ppos, pipe, len, flags);
3348                 if (ret)
3349                         goto out_err;
3350         }
3351
3352         ret = tracing_wait_pipe(filp);
3353         if (ret <= 0)
3354                 goto out_err;
3355
3356         if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3357                 ret = -EFAULT;
3358                 goto out_err;
3359         }
3360
3361         trace_event_read_lock();
3362         trace_access_lock(iter->cpu_file);
3363
3364         /* Fill as many pages as possible. */
3365         for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3366                 spd.pages[i] = alloc_page(GFP_KERNEL);
3367                 if (!spd.pages[i])
3368                         break;
3369
3370                 rem = tracing_fill_pipe_page(rem, iter);
3371
3372                 /* Copy the data into the page, so we can start over. */
3373                 ret = trace_seq_to_buffer(&iter->seq,
3374                                           page_address(spd.pages[i]),
3375                                           iter->seq.len);
3376                 if (ret < 0) {
3377                         __free_page(spd.pages[i]);
3378                         break;
3379                 }
3380                 spd.partial[i].offset = 0;
3381                 spd.partial[i].len = iter->seq.len;
3382
3383                 trace_seq_init(&iter->seq);
3384         }
3385
3386         trace_access_unlock(iter->cpu_file);
3387         trace_event_read_unlock();
3388         mutex_unlock(&iter->mutex);
3389
3390         spd.nr_pages = i;
3391
3392         ret = splice_to_pipe(pipe, &spd);
3393 out:
3394         splice_shrink_spd(pipe, &spd);
3395         return ret;
3396
3397 out_err:
3398         mutex_unlock(&iter->mutex);
3399         goto out;
3400 }
3401
3402 static ssize_t
3403 tracing_entries_read(struct file *filp, char __user *ubuf,
3404                      size_t cnt, loff_t *ppos)
3405 {
3406         struct trace_array *tr = filp->private_data;
3407         char buf[96];
3408         int r;
3409
3410         mutex_lock(&trace_types_lock);
3411         if (!ring_buffer_expanded)
3412                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3413                             tr->entries >> 10,
3414                             trace_buf_size >> 10);
3415         else
3416                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3417         mutex_unlock(&trace_types_lock);
3418
3419         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3420 }
3421
3422 static ssize_t
3423 tracing_entries_write(struct file *filp, const char __user *ubuf,
3424                       size_t cnt, loff_t *ppos)
3425 {
3426         unsigned long val;
3427         char buf[64];
3428         int ret, cpu;
3429
3430         if (cnt >= sizeof(buf))
3431                 return -EINVAL;
3432
3433         if (copy_from_user(&buf, ubuf, cnt))
3434                 return -EFAULT;
3435
3436         buf[cnt] = 0;
3437
3438         ret = strict_strtoul(buf, 10, &val);
3439         if (ret < 0)
3440                 return ret;
3441
3442         /* must have at least 1 entry */
3443         if (!val)
3444                 return -EINVAL;
3445
3446         mutex_lock(&trace_types_lock);
3447
3448         tracing_stop();
3449
3450         /* disable all cpu buffers */
3451         for_each_tracing_cpu(cpu) {
3452                 if (global_trace.data[cpu])
3453                         atomic_inc(&global_trace.data[cpu]->disabled);
3454                 if (max_tr.data[cpu])
3455                         atomic_inc(&max_tr.data[cpu]->disabled);
3456         }
3457
3458         /* value is in KB */
3459         val <<= 10;
3460
3461         if (val != global_trace.entries) {
3462                 ret = tracing_resize_ring_buffer(val);
3463                 if (ret < 0) {
3464                         cnt = ret;
3465                         goto out;
3466                 }
3467         }
3468
3469         *ppos += cnt;
3470
3471         /* If check pages failed, return ENOMEM */
3472         if (tracing_disabled)
3473                 cnt = -ENOMEM;
3474  out:
3475         for_each_tracing_cpu(cpu) {
3476                 if (global_trace.data[cpu])
3477                         atomic_dec(&global_trace.data[cpu]->disabled);
3478                 if (max_tr.data[cpu])
3479                         atomic_dec(&max_tr.data[cpu]->disabled);
3480         }
3481
3482         tracing_start();
3483         mutex_unlock(&trace_types_lock);
3484
3485         return cnt;
3486 }
3487
3488 static int mark_printk(const char *fmt, ...)
3489 {
3490         int ret;
3491         va_list args;
3492         va_start(args, fmt);
3493         ret = trace_vprintk(0, fmt, args);
3494         va_end(args);
3495         return ret;
3496 }
3497
3498 static ssize_t
3499 tracing_mark_write(struct file *filp, const char __user *ubuf,
3500                                         size_t cnt, loff_t *fpos)
3501 {
3502         char *buf;
3503         size_t written;
3504
3505         if (tracing_disabled)
3506                 return -EINVAL;
3507
3508         if (cnt > TRACE_BUF_SIZE)
3509                 cnt = TRACE_BUF_SIZE;
3510
3511         buf = kmalloc(cnt + 2, GFP_KERNEL);
3512         if (buf == NULL)
3513                 return -ENOMEM;
3514
3515         if (copy_from_user(buf, ubuf, cnt)) {
3516                 kfree(buf);
3517                 return -EFAULT;
3518         }
3519         if (buf[cnt-1] != '\n') {
3520                 buf[cnt] = '\n';
3521                 buf[cnt+1] = '\0';
3522         } else
3523                 buf[cnt] = '\0';
3524
3525         written = mark_printk("%s", buf);
3526         kfree(buf);
3527         *fpos += written;
3528
3529         /* don't tell userspace we wrote more - it might confuse them */
3530         if (written > cnt)
3531                 written = cnt;
3532
3533         return written;
3534 }
3535
3536 static int tracing_clock_show(struct seq_file *m, void *v)
3537 {
3538         int i;
3539
3540         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3541                 seq_printf(m,
3542                         "%s%s%s%s", i ? " " : "",
3543                         i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3544                         i == trace_clock_id ? "]" : "");
3545         seq_putc(m, '\n');
3546
3547         return 0;
3548 }
3549
3550 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3551                                    size_t cnt, loff_t *fpos)
3552 {
3553         char buf[64];
3554         const char *clockstr;
3555         int i;
3556
3557         if (cnt >= sizeof(buf))
3558                 return -EINVAL;
3559
3560         if (copy_from_user(&buf, ubuf, cnt))
3561                 return -EFAULT;
3562
3563         buf[cnt] = 0;
3564
3565         clockstr = strstrip(buf);
3566
3567         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3568                 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3569                         break;
3570         }
3571         if (i == ARRAY_SIZE(trace_clocks))
3572                 return -EINVAL;
3573
3574         trace_clock_id = i;
3575
3576         mutex_lock(&trace_types_lock);
3577
3578         ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3579         if (max_tr.buffer)
3580                 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3581
3582         mutex_unlock(&trace_types_lock);
3583
3584         *fpos += cnt;
3585
3586         return cnt;
3587 }
3588
3589 static int tracing_clock_open(struct inode *inode, struct file *file)
3590 {
3591         if (tracing_disabled)
3592                 return -ENODEV;
3593         return single_open(file, tracing_clock_show, NULL);
3594 }
3595
3596 static const struct file_operations tracing_max_lat_fops = {
3597         .open           = tracing_open_generic,
3598         .read           = tracing_max_lat_read,
3599         .write          = tracing_max_lat_write,
3600         .llseek         = generic_file_llseek,
3601 };
3602
3603 static const struct file_operations tracing_ctrl_fops = {
3604         .open           = tracing_open_generic,
3605         .read           = tracing_ctrl_read,
3606         .write          = tracing_ctrl_write,
3607         .llseek         = generic_file_llseek,
3608 };
3609
3610 static const struct file_operations set_tracer_fops = {
3611         .open           = tracing_open_generic,
3612         .read           = tracing_set_trace_read,
3613         .write          = tracing_set_trace_write,
3614         .llseek         = generic_file_llseek,
3615 };
3616
3617 static const struct file_operations tracing_pipe_fops = {
3618         .open           = tracing_open_pipe,
3619         .poll           = tracing_poll_pipe,
3620         .read           = tracing_read_pipe,
3621         .splice_read    = tracing_splice_read_pipe,
3622         .release        = tracing_release_pipe,
3623         .llseek         = no_llseek,
3624 };
3625
3626 static const struct file_operations tracing_entries_fops = {
3627         .open           = tracing_open_generic,
3628         .read           = tracing_entries_read,
3629         .write          = tracing_entries_write,
3630         .llseek         = generic_file_llseek,
3631 };
3632
3633 static const struct file_operations tracing_mark_fops = {
3634         .open           = tracing_open_generic,
3635         .write          = tracing_mark_write,
3636         .llseek         = generic_file_llseek,
3637 };
3638
3639 static const struct file_operations trace_clock_fops = {
3640         .open           = tracing_clock_open,
3641         .read           = seq_read,
3642         .llseek         = seq_lseek,
3643         .release        = single_release,
3644         .write          = tracing_clock_write,
3645 };
3646
3647 struct ftrace_buffer_info {
3648         struct trace_array      *tr;
3649         void                    *spare;
3650         int                     cpu;
3651         unsigned int            read;
3652 };
3653
3654 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3655 {
3656         int cpu = (int)(long)inode->i_private;
3657         struct ftrace_buffer_info *info;
3658
3659         if (tracing_disabled)
3660                 return -ENODEV;
3661
3662         info = kzalloc(sizeof(*info), GFP_KERNEL);
3663         if (!info)
3664                 return -ENOMEM;
3665
3666         info->tr        = &global_trace;
3667         info->cpu       = cpu;
3668         info->spare     = NULL;
3669         /* Force reading ring buffer for first read */
3670         info->read      = (unsigned int)-1;
3671
3672         filp->private_data = info;
3673
3674         return nonseekable_open(inode, filp);
3675 }
3676
3677 static ssize_t
3678 tracing_buffers_read(struct file *filp, char __user *ubuf,
3679                      size_t count, loff_t *ppos)
3680 {
3681         struct ftrace_buffer_info *info = filp->private_data;
3682         ssize_t ret;
3683         size_t size;
3684
3685         if (!count)
3686                 return 0;
3687
3688         if (!info->spare)
3689                 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3690         if (!info->spare)
3691                 return -ENOMEM;
3692
3693         /* Do we have previous read data to read? */
3694         if (info->read < PAGE_SIZE)
3695                 goto read;
3696
3697         info->read = 0;
3698
3699         trace_access_lock(info->cpu);
3700         ret = ring_buffer_read_page(info->tr->buffer,
3701                                     &info->spare,
3702                                     count,
3703                                     info->cpu, 0);
3704         trace_access_unlock(info->cpu);
3705         if (ret < 0)
3706                 return 0;
3707
3708 read:
3709         size = PAGE_SIZE - info->read;
3710         if (size > count)
3711                 size = count;
3712
3713         ret = copy_to_user(ubuf, info->spare + info->read, size);
3714         if (ret == size)
3715                 return -EFAULT;
3716         size -= ret;
3717
3718         *ppos += size;
3719         info->read += size;
3720
3721         return size;
3722 }
3723
3724 static int tracing_buffers_release(struct inode *inode, struct file *file)
3725 {
3726         struct ftrace_buffer_info *info = file->private_data;
3727
3728         if (info->spare)
3729                 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3730         kfree(info);
3731
3732         return 0;
3733 }
3734
3735 struct buffer_ref {
3736         struct ring_buffer      *buffer;
3737         void                    *page;
3738         int                     ref;
3739 };
3740
3741 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3742                                     struct pipe_buffer *buf)
3743 {
3744         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3745
3746         if (--ref->ref)
3747                 return;
3748
3749         ring_buffer_free_read_page(ref->buffer, ref->page);
3750         kfree(ref);
3751         buf->private = 0;
3752 }
3753
3754 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3755                                  struct pipe_buffer *buf)
3756 {
3757         return 1;
3758 }
3759
3760 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3761                                 struct pipe_buffer *buf)
3762 {
3763         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3764
3765         ref->ref++;
3766 }
3767
3768 /* Pipe buffer operations for a buffer. */
3769 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3770         .can_merge              = 0,
3771         .map                    = generic_pipe_buf_map,
3772         .unmap                  = generic_pipe_buf_unmap,
3773         .confirm                = generic_pipe_buf_confirm,
3774         .release                = buffer_pipe_buf_release,
3775         .steal                  = buffer_pipe_buf_steal,
3776         .get                    = buffer_pipe_buf_get,
3777 };
3778
3779 /*
3780  * Callback from splice_to_pipe(), if we need to release some pages
3781  * at the end of the spd in case we error'ed out in filling the pipe.
3782  */
3783 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3784 {
3785         struct buffer_ref *ref =
3786                 (struct buffer_ref *)spd->partial[i].private;
3787
3788         if (--ref->ref)
3789                 return;
3790
3791         ring_buffer_free_read_page(ref->buffer, ref->page);
3792         kfree(ref);
3793         spd->partial[i].private = 0;
3794 }
3795
3796 static ssize_t
3797 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3798                             struct pipe_inode_info *pipe, size_t len,
3799                             unsigned int flags)
3800 {
3801         struct ftrace_buffer_info *info = file->private_data;
3802         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3803         struct page *pages_def[PIPE_DEF_BUFFERS];
3804         struct splice_pipe_desc spd = {
3805                 .pages          = pages_def,
3806                 .partial        = partial_def,
3807                 .flags          = flags,
3808                 .ops            = &buffer_pipe_buf_ops,
3809                 .spd_release    = buffer_spd_release,
3810         };
3811         struct buffer_ref *ref;
3812         int entries, size, i;
3813         size_t ret;
3814
3815         if (splice_grow_spd(pipe, &spd))
3816                 return -ENOMEM;
3817
3818         if (*ppos & (PAGE_SIZE - 1)) {
3819                 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3820                 ret = -EINVAL;
3821                 goto out;
3822         }
3823
3824         if (len & (PAGE_SIZE - 1)) {
3825                 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3826                 if (len < PAGE_SIZE) {
3827                         ret = -EINVAL;
3828                         goto out;
3829                 }
3830                 len &= PAGE_MASK;
3831         }
3832
3833         trace_access_lock(info->cpu);
3834         entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3835
3836         for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3837                 struct page *page;
3838                 int r;
3839
3840                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3841                 if (!ref)
3842                         break;
3843
3844                 ref->ref = 1;
3845                 ref->buffer = info->tr->buffer;
3846                 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3847                 if (!ref->page) {
3848                         kfree(ref);
3849                         break;
3850                 }
3851
3852                 r = ring_buffer_read_page(ref->buffer, &ref->page,
3853                                           len, info->cpu, 1);
3854                 if (r < 0) {
3855                         ring_buffer_free_read_page(ref->buffer,
3856                                                    ref->page);
3857                         kfree(ref);
3858                         break;
3859                 }
3860
3861                 /*
3862                  * zero out any left over data, this is going to
3863                  * user land.
3864                  */
3865                 size = ring_buffer_page_len(ref->page);
3866                 if (size < PAGE_SIZE)
3867                         memset(ref->page + size, 0, PAGE_SIZE - size);
3868
3869                 page = virt_to_page(ref->page);
3870
3871                 spd.pages[i] = page;
3872                 spd.partial[i].len = PAGE_SIZE;
3873                 spd.partial[i].offset = 0;
3874                 spd.partial[i].private = (unsigned long)ref;
3875                 spd.nr_pages++;
3876                 *ppos += PAGE_SIZE;
3877
3878                 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3879         }
3880
3881         trace_access_unlock(info->cpu);
3882         spd.nr_pages = i;
3883
3884         /* did we read anything? */
3885         if (!spd.nr_pages) {
3886                 if (flags & SPLICE_F_NONBLOCK)
3887                         ret = -EAGAIN;
3888                 else
3889                         ret = 0;
3890                 /* TODO: block */
3891                 goto out;
3892         }
3893
3894         ret = splice_to_pipe(pipe, &spd);
3895         splice_shrink_spd(pipe, &spd);
3896 out:
3897         return ret;
3898 }
3899
3900 static const struct file_operations tracing_buffers_fops = {
3901         .open           = tracing_buffers_open,
3902         .read           = tracing_buffers_read,
3903         .release        = tracing_buffers_release,
3904         .splice_read    = tracing_buffers_splice_read,
3905         .llseek         = no_llseek,
3906 };
3907
3908 static ssize_t
3909 tracing_stats_read(struct file *filp, char __user *ubuf,
3910                    size_t count, loff_t *ppos)
3911 {
3912         unsigned long cpu = (unsigned long)filp->private_data;
3913         struct trace_array *tr = &global_trace;
3914         struct trace_seq *s;
3915         unsigned long cnt;
3916
3917         s = kmalloc(sizeof(*s), GFP_KERNEL);
3918         if (!s)
3919                 return -ENOMEM;
3920
3921         trace_seq_init(s);
3922
3923         cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3924         trace_seq_printf(s, "entries: %ld\n", cnt);
3925
3926         cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3927         trace_seq_printf(s, "overrun: %ld\n", cnt);
3928
3929         cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3930         trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3931
3932         count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3933
3934         kfree(s);
3935
3936         return count;
3937 }
3938
3939 static const struct file_operations tracing_stats_fops = {
3940         .open           = tracing_open_generic,
3941         .read           = tracing_stats_read,
3942         .llseek         = generic_file_llseek,
3943 };
3944
3945 #ifdef CONFIG_DYNAMIC_FTRACE
3946
3947 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3948 {
3949         return 0;
3950 }
3951
3952 static ssize_t
3953 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3954                   size_t cnt, loff_t *ppos)
3955 {
3956         static char ftrace_dyn_info_buffer[1024];
3957         static DEFINE_MUTEX(dyn_info_mutex);
3958         unsigned long *p = filp->private_data;
3959         char *buf = ftrace_dyn_info_buffer;
3960         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3961         int r;
3962
3963         mutex_lock(&dyn_info_mutex);
3964         r = sprintf(buf, "%ld ", *p);
3965
3966         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3967         buf[r++] = '\n';
3968
3969         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3970
3971         mutex_unlock(&dyn_info_mutex);
3972
3973         return r;
3974 }
3975
3976 static const struct file_operations tracing_dyn_info_fops = {
3977         .open           = tracing_open_generic,
3978         .read           = tracing_read_dyn_info,
3979         .llseek         = generic_file_llseek,
3980 };
3981 #endif
3982
3983 static struct dentry *d_tracer;
3984
3985 struct dentry *tracing_init_dentry(void)
3986 {
3987         static int once;
3988
3989         if (d_tracer)
3990                 return d_tracer;
3991
3992         if (!debugfs_initialized())
3993                 return NULL;
3994
3995         d_tracer = debugfs_create_dir("tracing", NULL);
3996
3997         if (!d_tracer && !once) {
3998                 once = 1;
3999                 pr_warning("Could not create debugfs directory 'tracing'\n");
4000                 return NULL;
4001         }
4002
4003         return d_tracer;
4004 }
4005
4006 static struct dentry *d_percpu;
4007
4008 struct dentry *tracing_dentry_percpu(void)
4009 {
4010         static int once;
4011         struct dentry *d_tracer;
4012
4013         if (d_percpu)
4014                 return d_percpu;
4015
4016         d_tracer = tracing_init_dentry();
4017
4018         if (!d_tracer)
4019                 return NULL;
4020
4021         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4022
4023         if (!d_percpu && !once) {
4024                 once = 1;
4025                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4026                 return NULL;
4027         }
4028
4029         return d_percpu;
4030 }
4031
4032 static void tracing_init_debugfs_percpu(long cpu)
4033 {
4034         struct dentry *d_percpu = tracing_dentry_percpu();
4035         struct dentry *d_cpu;
4036         char cpu_dir[30]; /* 30 characters should be more than enough */
4037
4038         snprintf(cpu_dir, 30, "cpu%ld", cpu);
4039         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4040         if (!d_cpu) {
4041                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4042                 return;
4043         }
4044
4045         /* per cpu trace_pipe */
4046         trace_create_file("trace_pipe", 0444, d_cpu,
4047                         (void *) cpu, &tracing_pipe_fops);
4048
4049         /* per cpu trace */
4050         trace_create_file("trace", 0644, d_cpu,
4051                         (void *) cpu, &tracing_fops);
4052
4053         trace_create_file("trace_pipe_raw", 0444, d_cpu,
4054                         (void *) cpu, &tracing_buffers_fops);
4055
4056         trace_create_file("stats", 0444, d_cpu,
4057                         (void *) cpu, &tracing_stats_fops);
4058 }
4059
4060 #ifdef CONFIG_FTRACE_SELFTEST
4061 /* Let selftest have access to static functions in this file */
4062 #include "trace_selftest.c"
4063 #endif
4064
4065 struct trace_option_dentry {
4066         struct tracer_opt               *opt;
4067         struct tracer_flags             *flags;
4068         struct dentry                   *entry;
4069 };
4070
4071 static ssize_t
4072 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4073                         loff_t *ppos)
4074 {
4075         struct trace_option_dentry *topt = filp->private_data;
4076         char *buf;
4077
4078         if (topt->flags->val & topt->opt->bit)
4079                 buf = "1\n";
4080         else
4081                 buf = "0\n";
4082
4083         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4084 }
4085
4086 static ssize_t
4087 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4088                          loff_t *ppos)
4089 {
4090         struct trace_option_dentry *topt = filp->private_data;
4091         unsigned long val;
4092         char buf[64];
4093         int ret;
4094
4095         if (cnt >= sizeof(buf))
4096                 return -EINVAL;
4097
4098         if (copy_from_user(&buf, ubuf, cnt))
4099                 return -EFAULT;
4100
4101         buf[cnt] = 0;
4102
4103         ret = strict_strtoul(buf, 10, &val);
4104         if (ret < 0)
4105                 return ret;
4106
4107         if (val != 0 && val != 1)
4108                 return -EINVAL;
4109
4110         if (!!(topt->flags->val & topt->opt->bit) != val) {
4111                 mutex_lock(&trace_types_lock);
4112                 ret = __set_tracer_option(current_trace, topt->flags,
4113                                           topt->opt, !val);
4114                 mutex_unlock(&trace_types_lock);
4115                 if (ret)
4116                         return ret;
4117         }
4118
4119         *ppos += cnt;
4120
4121         return cnt;
4122 }
4123
4124
4125 static const struct file_operations trace_options_fops = {
4126         .open = tracing_open_generic,
4127         .read = trace_options_read,
4128         .write = trace_options_write,
4129         .llseek = generic_file_llseek,
4130 };
4131
4132 static ssize_t
4133 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4134                         loff_t *ppos)
4135 {
4136         long index = (long)filp->private_data;
4137         char *buf;
4138
4139         if (trace_flags & (1 << index))
4140                 buf = "1\n";
4141         else
4142                 buf = "0\n";
4143
4144         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4145 }
4146
4147 static ssize_t
4148 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4149                          loff_t *ppos)
4150 {
4151         long index = (long)filp->private_data;
4152         char buf[64];
4153         unsigned long val;
4154         int ret;
4155
4156         if (cnt >= sizeof(buf))
4157                 return -EINVAL;
4158
4159         if (copy_from_user(&buf, ubuf, cnt))
4160                 return -EFAULT;
4161
4162         buf[cnt] = 0;
4163
4164         ret = strict_strtoul(buf, 10, &val);
4165         if (ret < 0)
4166                 return ret;
4167
4168         if (val != 0 && val != 1)
4169                 return -EINVAL;
4170         set_tracer_flags(1 << index, val);
4171
4172         *ppos += cnt;
4173
4174         return cnt;
4175 }
4176
4177 static const struct file_operations trace_options_core_fops = {
4178         .open = tracing_open_generic,
4179         .read = trace_options_core_read,
4180         .write = trace_options_core_write,
4181         .llseek = generic_file_llseek,
4182 };
4183
4184 struct dentry *trace_create_file(const char *name,
4185                                  mode_t mode,
4186                                  struct dentry *parent,
4187                                  void *data,
4188                                  const struct file_operations *fops)
4189 {
4190         struct dentry *ret;
4191
4192         ret = debugfs_create_file(name, mode, parent, data, fops);
4193         if (!ret)
4194                 pr_warning("Could not create debugfs '%s' entry\n", name);
4195
4196         return ret;
4197 }
4198
4199
4200 static struct dentry *trace_options_init_dentry(void)
4201 {
4202         struct dentry *d_tracer;
4203         static struct dentry *t_options;
4204
4205         if (t_options)
4206                 return t_options;
4207
4208         d_tracer = tracing_init_dentry();
4209         if (!d_tracer)
4210                 return NULL;
4211
4212         t_options = debugfs_create_dir("options", d_tracer);
4213         if (!t_options) {
4214                 pr_warning("Could not create debugfs directory 'options'\n");
4215                 return NULL;
4216         }
4217
4218         return t_options;
4219 }
4220
4221 static void
4222 create_trace_option_file(struct trace_option_dentry *topt,
4223                          struct tracer_flags *flags,
4224                          struct tracer_opt *opt)
4225 {
4226         struct dentry *t_options;
4227
4228         t_options = trace_options_init_dentry();
4229         if (!t_options)
4230                 return;
4231
4232         topt->flags = flags;
4233         topt->opt = opt;
4234
4235         topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4236                                     &trace_options_fops);
4237
4238 }
4239
4240 static struct trace_option_dentry *
4241 create_trace_option_files(struct tracer *tracer)
4242 {
4243         struct trace_option_dentry *topts;
4244         struct tracer_flags *flags;
4245         struct tracer_opt *opts;
4246         int cnt;
4247
4248         if (!tracer)
4249                 return NULL;
4250
4251         flags = tracer->flags;
4252
4253         if (!flags || !flags->opts)
4254                 return NULL;
4255
4256         opts = flags->opts;
4257
4258         for (cnt = 0; opts[cnt].name; cnt++)
4259                 ;
4260
4261         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4262         if (!topts)
4263                 return NULL;
4264
4265         for (cnt = 0; opts[cnt].name; cnt++)
4266                 create_trace_option_file(&topts[cnt], flags,
4267                                          &opts[cnt]);
4268
4269         return topts;
4270 }
4271
4272 static void
4273 destroy_trace_option_files(struct trace_option_dentry *topts)
4274 {
4275         int cnt;
4276
4277         if (!topts)
4278                 return;
4279
4280         for (cnt = 0; topts[cnt].opt; cnt++) {
4281                 if (topts[cnt].entry)
4282                         debugfs_remove(topts[cnt].entry);
4283         }
4284
4285         kfree(topts);
4286 }
4287
4288 static struct dentry *
4289 create_trace_option_core_file(const char *option, long index)
4290 {
4291         struct dentry *t_options;
4292
4293         t_options = trace_options_init_dentry();
4294         if (!t_options)
4295                 return NULL;
4296
4297         return trace_create_file(option, 0644, t_options, (void *)index,
4298                                     &trace_options_core_fops);
4299 }
4300
4301 static __init void create_trace_options_dir(void)
4302 {
4303         struct dentry *t_options;
4304         int i;
4305
4306         t_options = trace_options_init_dentry();
4307         if (!t_options)
4308                 return;
4309
4310         for (i = 0; trace_options[i]; i++)
4311                 create_trace_option_core_file(trace_options[i], i);
4312 }
4313
4314 static __init int tracer_init_debugfs(void)
4315 {
4316         struct dentry *d_tracer;
4317         int cpu;
4318
4319         trace_access_lock_init();
4320
4321         d_tracer = tracing_init_dentry();
4322
4323         trace_create_file("tracing_enabled", 0644, d_tracer,
4324                         &global_trace, &tracing_ctrl_fops);
4325
4326         trace_create_file("trace_options", 0644, d_tracer,
4327                         NULL, &tracing_iter_fops);
4328
4329         trace_create_file("tracing_cpumask", 0644, d_tracer,
4330                         NULL, &tracing_cpumask_fops);
4331
4332         trace_create_file("trace", 0644, d_tracer,
4333                         (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4334
4335         trace_create_file("available_tracers", 0444, d_tracer,
4336                         &global_trace, &show_traces_fops);
4337
4338         trace_create_file("current_tracer", 0644, d_tracer,
4339                         &global_trace, &set_tracer_fops);
4340
4341 #ifdef CONFIG_TRACER_MAX_TRACE
4342         trace_create_file("tracing_max_latency", 0644, d_tracer,
4343                         &tracing_max_latency, &tracing_max_lat_fops);
4344 #endif
4345
4346         trace_create_file("tracing_thresh", 0644, d_tracer,
4347                         &tracing_thresh, &tracing_max_lat_fops);
4348
4349         trace_create_file("README", 0444, d_tracer,
4350                         NULL, &tracing_readme_fops);
4351
4352         trace_create_file("trace_pipe", 0444, d_tracer,
4353                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4354
4355         trace_create_file("buffer_size_kb", 0644, d_tracer,
4356                         &global_trace, &tracing_entries_fops);
4357
4358         trace_create_file("trace_marker", 0220, d_tracer,
4359                         NULL, &tracing_mark_fops);
4360
4361         trace_create_file("saved_cmdlines", 0444, d_tracer,
4362                         NULL, &tracing_saved_cmdlines_fops);
4363
4364         trace_create_file("trace_clock", 0644, d_tracer, NULL,
4365                           &trace_clock_fops);
4366
4367 #ifdef CONFIG_DYNAMIC_FTRACE
4368         trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4369                         &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4370 #endif
4371
4372         create_trace_options_dir();
4373
4374         for_each_tracing_cpu(cpu)
4375                 tracing_init_debugfs_percpu(cpu);
4376
4377         return 0;
4378 }
4379
4380 static int trace_panic_handler(struct notifier_block *this,
4381                                unsigned long event, void *unused)
4382 {
4383         if (ftrace_dump_on_oops)
4384                 ftrace_dump(ftrace_dump_on_oops);
4385         return NOTIFY_OK;
4386 }
4387
4388 static struct notifier_block trace_panic_notifier = {
4389         .notifier_call  = trace_panic_handler,
4390         .next           = NULL,
4391         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
4392 };
4393
4394 static int trace_die_handler(struct notifier_block *self,
4395                              unsigned long val,
4396                              void *data)
4397 {
4398         switch (val) {
4399         case DIE_OOPS:
4400                 if (ftrace_dump_on_oops)
4401                         ftrace_dump(ftrace_dump_on_oops);
4402                 break;
4403         default:
4404                 break;
4405         }
4406         return NOTIFY_OK;
4407 }
4408
4409 static struct notifier_block trace_die_notifier = {
4410         .notifier_call = trace_die_handler,
4411         .priority = 200
4412 };
4413
4414 /*
4415  * printk is set to max of 1024, we really don't need it that big.
4416  * Nothing should be printing 1000 characters anyway.
4417  */
4418 #define TRACE_MAX_PRINT         1000
4419
4420 /*
4421  * Define here KERN_TRACE so that we have one place to modify
4422  * it if we decide to change what log level the ftrace dump
4423  * should be at.
4424  */
4425 #define KERN_TRACE              KERN_EMERG
4426
4427 void
4428 trace_printk_seq(struct trace_seq *s)
4429 {
4430         /* Probably should print a warning here. */
4431         if (s->len >= 1000)
4432                 s->len = 1000;
4433
4434         /* should be zero ended, but we are paranoid. */
4435         s->buffer[s->len] = 0;
4436
4437         printk(KERN_TRACE "%s", s->buffer);
4438
4439         trace_seq_init(s);
4440 }
4441
4442 void trace_init_global_iter(struct trace_iterator *iter)
4443 {
4444         iter->tr = &global_trace;
4445         iter->trace = current_trace;
4446         iter->cpu_file = TRACE_PIPE_ALL_CPU;
4447 }
4448
4449 static void
4450 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4451 {
4452         static arch_spinlock_t ftrace_dump_lock =
4453                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4454         /* use static because iter can be a bit big for the stack */
4455         static struct trace_iterator iter;
4456         unsigned int old_userobj;
4457         static int dump_ran;
4458         unsigned long flags;
4459         int cnt = 0, cpu;
4460
4461         /* only one dump */
4462         local_irq_save(flags);
4463         arch_spin_lock(&ftrace_dump_lock);
4464         if (dump_ran)
4465                 goto out;
4466
4467         dump_ran = 1;
4468
4469         tracing_off();
4470
4471         if (disable_tracing)
4472                 ftrace_kill();
4473
4474         trace_init_global_iter(&iter);
4475
4476         for_each_tracing_cpu(cpu) {
4477                 atomic_inc(&iter.tr->data[cpu]->disabled);
4478         }
4479
4480         old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4481
4482         /* don't look at user memory in panic mode */
4483         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4484
4485         /* Simulate the iterator */
4486         iter.tr = &global_trace;
4487         iter.trace = current_trace;
4488
4489         switch (oops_dump_mode) {
4490         case DUMP_ALL:
4491                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4492                 break;
4493         case DUMP_ORIG:
4494                 iter.cpu_file = raw_smp_processor_id();
4495                 break;
4496         case DUMP_NONE:
4497                 goto out_enable;
4498         default:
4499                 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4500                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4501         }
4502
4503         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4504
4505         /*
4506          * We need to stop all tracing on all CPUS to read the
4507          * the next buffer. This is a bit expensive, but is
4508          * not done often. We fill all what we can read,
4509          * and then release the locks again.
4510          */
4511
4512         while (!trace_empty(&iter)) {
4513
4514                 if (!cnt)
4515                         printk(KERN_TRACE "---------------------------------\n");
4516
4517                 cnt++;
4518
4519                 /* reset all but tr, trace, and overruns */
4520                 memset(&iter.seq, 0,
4521                        sizeof(struct trace_iterator) -
4522                        offsetof(struct trace_iterator, seq));
4523                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4524                 iter.pos = -1;
4525
4526                 if (trace_find_next_entry_inc(&iter) != NULL) {
4527                         int ret;
4528
4529                         ret = print_trace_line(&iter);
4530                         if (ret != TRACE_TYPE_NO_CONSUME)
4531                                 trace_consume(&iter);
4532                 }
4533
4534                 trace_printk_seq(&iter.seq);
4535         }
4536
4537         if (!cnt)
4538                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4539         else
4540                 printk(KERN_TRACE "---------------------------------\n");
4541
4542  out_enable:
4543         /* Re-enable tracing if requested */
4544         if (!disable_tracing) {
4545                 trace_flags |= old_userobj;
4546
4547                 for_each_tracing_cpu(cpu) {
4548                         atomic_dec(&iter.tr->data[cpu]->disabled);
4549                 }
4550                 tracing_on();
4551         }
4552
4553  out:
4554         arch_spin_unlock(&ftrace_dump_lock);
4555         local_irq_restore(flags);
4556 }
4557
4558 /* By default: disable tracing after the dump */
4559 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4560 {
4561         __ftrace_dump(true, oops_dump_mode);
4562 }
4563
4564 __init static int tracer_alloc_buffers(void)
4565 {
4566         int ring_buf_size;
4567         enum ring_buffer_flags rb_flags;
4568         int i;
4569         int ret = -ENOMEM;
4570
4571
4572         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4573                 goto out;
4574
4575         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4576                 goto out_free_buffer_mask;
4577
4578         /* To save memory, keep the ring buffer size to its minimum */
4579         if (ring_buffer_expanded)
4580                 ring_buf_size = trace_buf_size;
4581         else
4582                 ring_buf_size = 1;
4583
4584         rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
4585
4586         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4587         cpumask_copy(tracing_cpumask, cpu_all_mask);
4588
4589         /* TODO: make the number of buffers hot pluggable with CPUS */
4590         global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
4591         if (!global_trace.buffer) {
4592                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4593                 WARN_ON(1);
4594                 goto out_free_cpumask;
4595         }
4596         global_trace.entries = ring_buffer_size(global_trace.buffer);
4597
4598
4599 #ifdef CONFIG_TRACER_MAX_TRACE
4600         max_tr.buffer = ring_buffer_alloc(1, rb_flags);
4601         if (!max_tr.buffer) {
4602                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4603                 WARN_ON(1);
4604                 ring_buffer_free(global_trace.buffer);
4605                 goto out_free_cpumask;
4606         }
4607         max_tr.entries = 1;
4608 #endif
4609
4610         /* Allocate the first page for all buffers */
4611         for_each_tracing_cpu(i) {
4612                 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4613                 max_tr.data[i] = &per_cpu(max_tr_data, i);
4614         }
4615
4616         trace_init_cmdlines();
4617
4618         register_tracer(&nop_trace);
4619         current_trace = &nop_trace;
4620         /* All seems OK, enable tracing */
4621         tracing_disabled = 0;
4622
4623         atomic_notifier_chain_register(&panic_notifier_list,
4624                                        &trace_panic_notifier);
4625
4626         register_die_notifier(&trace_die_notifier);
4627
4628         return 0;
4629
4630 out_free_cpumask:
4631         free_cpumask_var(tracing_cpumask);
4632 out_free_buffer_mask:
4633         free_cpumask_var(tracing_buffer_mask);
4634 out:
4635         return ret;
4636 }
4637
4638 __init static int clear_boot_tracer(void)
4639 {
4640         /*
4641          * The default tracer at boot buffer is an init section.
4642          * This function is called in lateinit. If we did not
4643          * find the boot tracer, then clear it out, to prevent
4644          * later registration from accessing the buffer that is
4645          * about to be freed.
4646          */
4647         if (!default_bootup_tracer)
4648                 return 0;
4649
4650         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4651                default_bootup_tracer);
4652         default_bootup_tracer = NULL;
4653
4654         return 0;
4655 }
4656
4657 early_initcall(tracer_alloc_buffers);
4658 fs_initcall(tracer_init_debugfs);
4659 late_initcall(clear_boot_tracer);