Merge branch 'for-rmk/samsung6' of git://git.fluff.org/bjdooks/linux into devel-stable
[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/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/ctype.h>
36 #include <linux/init.h>
37 #include <linux/poll.h>
38 #include <linux/gfp.h>
39 #include <linux/fs.h>
40
41 #include "trace.h"
42 #include "trace_output.h"
43
44 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
45
46 /*
47  * On boot up, the ring buffer is set to the minimum size, so that
48  * we do not waste memory on systems that are not using tracing.
49  */
50 int ring_buffer_expanded;
51
52 /*
53  * We need to change this state when a selftest is running.
54  * A selftest will lurk into the ring-buffer to count the
55  * entries inserted during the selftest although some concurrent
56  * insertions into the ring-buffer such as trace_printk could occurred
57  * at the same time, giving false positive or negative results.
58  */
59 static bool __read_mostly tracing_selftest_running;
60
61 /*
62  * If a tracer is running, we do not want to run SELFTEST.
63  */
64 bool __read_mostly tracing_selftest_disabled;
65
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
68         { }
69 };
70
71 static struct tracer_flags dummy_tracer_flags = {
72         .val = 0,
73         .opts = dummy_tracer_opt
74 };
75
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
77 {
78         return 0;
79 }
80
81 /*
82  * Kill all tracing for good (never come back).
83  * It is initialized to 1 but will turn to zero if the initialization
84  * of the tracer is successful. But that is the only place that sets
85  * this back to zero.
86  */
87 static int tracing_disabled = 1;
88
89 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
90
91 static inline void ftrace_disable_cpu(void)
92 {
93         preempt_disable();
94         __this_cpu_inc(per_cpu_var(ftrace_cpu_disabled));
95 }
96
97 static inline void ftrace_enable_cpu(void)
98 {
99         __this_cpu_dec(per_cpu_var(ftrace_cpu_disabled));
100         preempt_enable();
101 }
102
103 static cpumask_var_t __read_mostly      tracing_buffer_mask;
104
105 /* Define which cpu buffers are currently read in trace_pipe */
106 static cpumask_var_t                    tracing_reader_cpumask;
107
108 #define for_each_tracing_cpu(cpu)       \
109         for_each_cpu(cpu, tracing_buffer_mask)
110
111 /*
112  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
113  *
114  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
115  * is set, then ftrace_dump is called. This will output the contents
116  * of the ftrace buffers to the console.  This is very useful for
117  * capturing traces that lead to crashes and outputing it to a
118  * serial console.
119  *
120  * It is default off, but you can enable it with either specifying
121  * "ftrace_dump_on_oops" in the kernel command line, or setting
122  * /proc/sys/kernel/ftrace_dump_on_oops to true.
123  */
124 int ftrace_dump_on_oops;
125
126 static int tracing_set_tracer(const char *buf);
127
128 #define MAX_TRACER_SIZE         100
129 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
130 static char *default_bootup_tracer;
131
132 static int __init set_cmdline_ftrace(char *str)
133 {
134         strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
135         default_bootup_tracer = bootup_tracer_buf;
136         /* We are using ftrace early, expand it */
137         ring_buffer_expanded = 1;
138         return 1;
139 }
140 __setup("ftrace=", set_cmdline_ftrace);
141
142 static int __init set_ftrace_dump_on_oops(char *str)
143 {
144         ftrace_dump_on_oops = 1;
145         return 1;
146 }
147 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
148
149 unsigned long long ns2usecs(cycle_t nsec)
150 {
151         nsec += 500;
152         do_div(nsec, 1000);
153         return nsec;
154 }
155
156 /*
157  * The global_trace is the descriptor that holds the tracing
158  * buffers for the live tracing. For each CPU, it contains
159  * a link list of pages that will store trace entries. The
160  * page descriptor of the pages in the memory is used to hold
161  * the link list by linking the lru item in the page descriptor
162  * to each of the pages in the buffer per CPU.
163  *
164  * For each active CPU there is a data field that holds the
165  * pages for the buffer for that CPU. Each CPU has the same number
166  * of pages allocated for its buffer.
167  */
168 static struct trace_array       global_trace;
169
170 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
171
172 int filter_current_check_discard(struct ring_buffer *buffer,
173                                  struct ftrace_event_call *call, void *rec,
174                                  struct ring_buffer_event *event)
175 {
176         return filter_check_discard(call, rec, buffer, event);
177 }
178 EXPORT_SYMBOL_GPL(filter_current_check_discard);
179
180 cycle_t ftrace_now(int cpu)
181 {
182         u64 ts;
183
184         /* Early boot up does not have a buffer yet */
185         if (!global_trace.buffer)
186                 return trace_clock_local();
187
188         ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
189         ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
190
191         return ts;
192 }
193
194 /*
195  * The max_tr is used to snapshot the global_trace when a maximum
196  * latency is reached. Some tracers will use this to store a maximum
197  * trace while it continues examining live traces.
198  *
199  * The buffers for the max_tr are set up the same as the global_trace.
200  * When a snapshot is taken, the link list of the max_tr is swapped
201  * with the link list of the global_trace and the buffers are reset for
202  * the global_trace so the tracing can continue.
203  */
204 static struct trace_array       max_tr;
205
206 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
207
208 /* tracer_enabled is used to toggle activation of a tracer */
209 static int                      tracer_enabled = 1;
210
211 /**
212  * tracing_is_enabled - return tracer_enabled status
213  *
214  * This function is used by other tracers to know the status
215  * of the tracer_enabled flag.  Tracers may use this function
216  * to know if it should enable their features when starting
217  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
218  */
219 int tracing_is_enabled(void)
220 {
221         return tracer_enabled;
222 }
223
224 /*
225  * trace_buf_size is the size in bytes that is allocated
226  * for a buffer. Note, the number of bytes is always rounded
227  * to page size.
228  *
229  * This number is purposely set to a low number of 16384.
230  * If the dump on oops happens, it will be much appreciated
231  * to not have to wait for all that output. Anyway this can be
232  * boot time and run time configurable.
233  */
234 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
235
236 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
237
238 /* trace_types holds a link list of available tracers. */
239 static struct tracer            *trace_types __read_mostly;
240
241 /* current_trace points to the tracer that is currently active */
242 static struct tracer            *current_trace __read_mostly;
243
244 /*
245  * trace_types_lock is used to protect the trace_types list.
246  * This lock is also used to keep user access serialized.
247  * Accesses from userspace will grab this lock while userspace
248  * activities happen inside the kernel.
249  */
250 static DEFINE_MUTEX(trace_types_lock);
251
252 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
253 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
254
255 /* trace_flags holds trace_options default values */
256 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
257         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
258         TRACE_ITER_GRAPH_TIME;
259
260 static int trace_stop_count;
261 static DEFINE_SPINLOCK(tracing_start_lock);
262
263 /**
264  * trace_wake_up - wake up tasks waiting for trace input
265  *
266  * Simply wakes up any task that is blocked on the trace_wait
267  * queue. These is used with trace_poll for tasks polling the trace.
268  */
269 void trace_wake_up(void)
270 {
271         int cpu;
272
273         if (trace_flags & TRACE_ITER_BLOCK)
274                 return;
275         /*
276          * The runqueue_is_locked() can fail, but this is the best we
277          * have for now:
278          */
279         cpu = get_cpu();
280         if (!runqueue_is_locked(cpu))
281                 wake_up(&trace_wait);
282         put_cpu();
283 }
284
285 static int __init set_buf_size(char *str)
286 {
287         unsigned long buf_size;
288
289         if (!str)
290                 return 0;
291         buf_size = memparse(str, &str);
292         /* nr_entries can not be zero */
293         if (buf_size == 0)
294                 return 0;
295         trace_buf_size = buf_size;
296         return 1;
297 }
298 __setup("trace_buf_size=", set_buf_size);
299
300 unsigned long nsecs_to_usecs(unsigned long nsecs)
301 {
302         return nsecs / 1000;
303 }
304
305 /* These must match the bit postions in trace_iterator_flags */
306 static const char *trace_options[] = {
307         "print-parent",
308         "sym-offset",
309         "sym-addr",
310         "verbose",
311         "raw",
312         "hex",
313         "bin",
314         "block",
315         "stacktrace",
316         "trace_printk",
317         "ftrace_preempt",
318         "branch",
319         "annotate",
320         "userstacktrace",
321         "sym-userobj",
322         "printk-msg-only",
323         "context-info",
324         "latency-format",
325         "sleep-time",
326         "graph-time",
327         NULL
328 };
329
330 static struct {
331         u64 (*func)(void);
332         const char *name;
333 } trace_clocks[] = {
334         { trace_clock_local,    "local" },
335         { trace_clock_global,   "global" },
336 };
337
338 int trace_clock_id;
339
340 /*
341  * trace_parser_get_init - gets the buffer for trace parser
342  */
343 int trace_parser_get_init(struct trace_parser *parser, int size)
344 {
345         memset(parser, 0, sizeof(*parser));
346
347         parser->buffer = kmalloc(size, GFP_KERNEL);
348         if (!parser->buffer)
349                 return 1;
350
351         parser->size = size;
352         return 0;
353 }
354
355 /*
356  * trace_parser_put - frees the buffer for trace parser
357  */
358 void trace_parser_put(struct trace_parser *parser)
359 {
360         kfree(parser->buffer);
361 }
362
363 /*
364  * trace_get_user - reads the user input string separated by  space
365  * (matched by isspace(ch))
366  *
367  * For each string found the 'struct trace_parser' is updated,
368  * and the function returns.
369  *
370  * Returns number of bytes read.
371  *
372  * See kernel/trace/trace.h for 'struct trace_parser' details.
373  */
374 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
375         size_t cnt, loff_t *ppos)
376 {
377         char ch;
378         size_t read = 0;
379         ssize_t ret;
380
381         if (!*ppos)
382                 trace_parser_clear(parser);
383
384         ret = get_user(ch, ubuf++);
385         if (ret)
386                 goto out;
387
388         read++;
389         cnt--;
390
391         /*
392          * The parser is not finished with the last write,
393          * continue reading the user input without skipping spaces.
394          */
395         if (!parser->cont) {
396                 /* skip white space */
397                 while (cnt && isspace(ch)) {
398                         ret = get_user(ch, ubuf++);
399                         if (ret)
400                                 goto out;
401                         read++;
402                         cnt--;
403                 }
404
405                 /* only spaces were written */
406                 if (isspace(ch)) {
407                         *ppos += read;
408                         ret = read;
409                         goto out;
410                 }
411
412                 parser->idx = 0;
413         }
414
415         /* read the non-space input */
416         while (cnt && !isspace(ch)) {
417                 if (parser->idx < parser->size - 1)
418                         parser->buffer[parser->idx++] = ch;
419                 else {
420                         ret = -EINVAL;
421                         goto out;
422                 }
423                 ret = get_user(ch, ubuf++);
424                 if (ret)
425                         goto out;
426                 read++;
427                 cnt--;
428         }
429
430         /* We either got finished input or we have to wait for another call. */
431         if (isspace(ch)) {
432                 parser->buffer[parser->idx] = 0;
433                 parser->cont = false;
434         } else {
435                 parser->cont = true;
436                 parser->buffer[parser->idx++] = ch;
437         }
438
439         *ppos += read;
440         ret = read;
441
442 out:
443         return ret;
444 }
445
446 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
447 {
448         int len;
449         int ret;
450
451         if (!cnt)
452                 return 0;
453
454         if (s->len <= s->readpos)
455                 return -EBUSY;
456
457         len = s->len - s->readpos;
458         if (cnt > len)
459                 cnt = len;
460         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
461         if (ret == cnt)
462                 return -EFAULT;
463
464         cnt -= ret;
465
466         s->readpos += cnt;
467         return cnt;
468 }
469
470 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
471 {
472         int len;
473         void *ret;
474
475         if (s->len <= s->readpos)
476                 return -EBUSY;
477
478         len = s->len - s->readpos;
479         if (cnt > len)
480                 cnt = len;
481         ret = memcpy(buf, s->buffer + s->readpos, cnt);
482         if (!ret)
483                 return -EFAULT;
484
485         s->readpos += cnt;
486         return cnt;
487 }
488
489 /*
490  * ftrace_max_lock is used to protect the swapping of buffers
491  * when taking a max snapshot. The buffers themselves are
492  * protected by per_cpu spinlocks. But the action of the swap
493  * needs its own lock.
494  *
495  * This is defined as a arch_spinlock_t in order to help
496  * with performance when lockdep debugging is enabled.
497  *
498  * It is also used in other places outside the update_max_tr
499  * so it needs to be defined outside of the
500  * CONFIG_TRACER_MAX_TRACE.
501  */
502 static arch_spinlock_t ftrace_max_lock =
503         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
504
505 #ifdef CONFIG_TRACER_MAX_TRACE
506 unsigned long __read_mostly     tracing_max_latency;
507 unsigned long __read_mostly     tracing_thresh;
508
509 /*
510  * Copy the new maximum trace into the separate maximum-trace
511  * structure. (this way the maximum trace is permanently saved,
512  * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
513  */
514 static void
515 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
516 {
517         struct trace_array_cpu *data = tr->data[cpu];
518         struct trace_array_cpu *max_data = tr->data[cpu];
519
520         max_tr.cpu = cpu;
521         max_tr.time_start = data->preempt_timestamp;
522
523         max_data = max_tr.data[cpu];
524         max_data->saved_latency = tracing_max_latency;
525         max_data->critical_start = data->critical_start;
526         max_data->critical_end = data->critical_end;
527
528         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
529         max_data->pid = tsk->pid;
530         max_data->uid = task_uid(tsk);
531         max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
532         max_data->policy = tsk->policy;
533         max_data->rt_priority = tsk->rt_priority;
534
535         /* record this tasks comm */
536         tracing_record_cmdline(tsk);
537 }
538
539 /**
540  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
541  * @tr: tracer
542  * @tsk: the task with the latency
543  * @cpu: The cpu that initiated the trace.
544  *
545  * Flip the buffers between the @tr and the max_tr and record information
546  * about which task was the cause of this latency.
547  */
548 void
549 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
550 {
551         struct ring_buffer *buf = tr->buffer;
552
553         if (trace_stop_count)
554                 return;
555
556         WARN_ON_ONCE(!irqs_disabled());
557         arch_spin_lock(&ftrace_max_lock);
558
559         tr->buffer = max_tr.buffer;
560         max_tr.buffer = buf;
561
562         __update_max_tr(tr, tsk, cpu);
563         arch_spin_unlock(&ftrace_max_lock);
564 }
565
566 /**
567  * update_max_tr_single - only copy one trace over, and reset the rest
568  * @tr - tracer
569  * @tsk - task with the latency
570  * @cpu - the cpu of the buffer to copy.
571  *
572  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
573  */
574 void
575 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
576 {
577         int ret;
578
579         if (trace_stop_count)
580                 return;
581
582         WARN_ON_ONCE(!irqs_disabled());
583         arch_spin_lock(&ftrace_max_lock);
584
585         ftrace_disable_cpu();
586
587         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
588
589         if (ret == -EBUSY) {
590                 /*
591                  * We failed to swap the buffer due to a commit taking
592                  * place on this CPU. We fail to record, but we reset
593                  * the max trace buffer (no one writes directly to it)
594                  * and flag that it failed.
595                  */
596                 trace_array_printk(&max_tr, _THIS_IP_,
597                         "Failed to swap buffers due to commit in progress\n");
598         }
599
600         ftrace_enable_cpu();
601
602         WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
603
604         __update_max_tr(tr, tsk, cpu);
605         arch_spin_unlock(&ftrace_max_lock);
606 }
607 #endif /* CONFIG_TRACER_MAX_TRACE */
608
609 /**
610  * register_tracer - register a tracer with the ftrace system.
611  * @type - the plugin for the tracer
612  *
613  * Register a new plugin tracer.
614  */
615 int register_tracer(struct tracer *type)
616 __releases(kernel_lock)
617 __acquires(kernel_lock)
618 {
619         struct tracer *t;
620         int ret = 0;
621
622         if (!type->name) {
623                 pr_info("Tracer must have a name\n");
624                 return -1;
625         }
626
627         if (strlen(type->name) > MAX_TRACER_SIZE) {
628                 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
629                 return -1;
630         }
631
632         /*
633          * When this gets called we hold the BKL which means that
634          * preemption is disabled. Various trace selftests however
635          * need to disable and enable preemption for successful tests.
636          * So we drop the BKL here and grab it after the tests again.
637          */
638         unlock_kernel();
639         mutex_lock(&trace_types_lock);
640
641         tracing_selftest_running = true;
642
643         for (t = trace_types; t; t = t->next) {
644                 if (strcmp(type->name, t->name) == 0) {
645                         /* already found */
646                         pr_info("Tracer %s already registered\n",
647                                 type->name);
648                         ret = -1;
649                         goto out;
650                 }
651         }
652
653         if (!type->set_flag)
654                 type->set_flag = &dummy_set_flag;
655         if (!type->flags)
656                 type->flags = &dummy_tracer_flags;
657         else
658                 if (!type->flags->opts)
659                         type->flags->opts = dummy_tracer_opt;
660         if (!type->wait_pipe)
661                 type->wait_pipe = default_wait_pipe;
662
663
664 #ifdef CONFIG_FTRACE_STARTUP_TEST
665         if (type->selftest && !tracing_selftest_disabled) {
666                 struct tracer *saved_tracer = current_trace;
667                 struct trace_array *tr = &global_trace;
668
669                 /*
670                  * Run a selftest on this tracer.
671                  * Here we reset the trace buffer, and set the current
672                  * tracer to be this tracer. The tracer can then run some
673                  * internal tracing to verify that everything is in order.
674                  * If we fail, we do not register this tracer.
675                  */
676                 tracing_reset_online_cpus(tr);
677
678                 current_trace = type;
679                 /* the test is responsible for initializing and enabling */
680                 pr_info("Testing tracer %s: ", type->name);
681                 ret = type->selftest(type, tr);
682                 /* the test is responsible for resetting too */
683                 current_trace = saved_tracer;
684                 if (ret) {
685                         printk(KERN_CONT "FAILED!\n");
686                         goto out;
687                 }
688                 /* Only reset on passing, to avoid touching corrupted buffers */
689                 tracing_reset_online_cpus(tr);
690
691                 printk(KERN_CONT "PASSED\n");
692         }
693 #endif
694
695         type->next = trace_types;
696         trace_types = type;
697
698  out:
699         tracing_selftest_running = false;
700         mutex_unlock(&trace_types_lock);
701
702         if (ret || !default_bootup_tracer)
703                 goto out_unlock;
704
705         if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
706                 goto out_unlock;
707
708         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
709         /* Do we want this tracer to start on bootup? */
710         tracing_set_tracer(type->name);
711         default_bootup_tracer = NULL;
712         /* disable other selftests, since this will break it. */
713         tracing_selftest_disabled = 1;
714 #ifdef CONFIG_FTRACE_STARTUP_TEST
715         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
716                type->name);
717 #endif
718
719  out_unlock:
720         lock_kernel();
721         return ret;
722 }
723
724 void unregister_tracer(struct tracer *type)
725 {
726         struct tracer **t;
727
728         mutex_lock(&trace_types_lock);
729         for (t = &trace_types; *t; t = &(*t)->next) {
730                 if (*t == type)
731                         goto found;
732         }
733         pr_info("Tracer %s not registered\n", type->name);
734         goto out;
735
736  found:
737         *t = (*t)->next;
738
739         if (type == current_trace && tracer_enabled) {
740                 tracer_enabled = 0;
741                 tracing_stop();
742                 if (current_trace->stop)
743                         current_trace->stop(&global_trace);
744                 current_trace = &nop_trace;
745         }
746 out:
747         mutex_unlock(&trace_types_lock);
748 }
749
750 static void __tracing_reset(struct trace_array *tr, int cpu)
751 {
752         ftrace_disable_cpu();
753         ring_buffer_reset_cpu(tr->buffer, cpu);
754         ftrace_enable_cpu();
755 }
756
757 void tracing_reset(struct trace_array *tr, int cpu)
758 {
759         struct ring_buffer *buffer = tr->buffer;
760
761         ring_buffer_record_disable(buffer);
762
763         /* Make sure all commits have finished */
764         synchronize_sched();
765         __tracing_reset(tr, cpu);
766
767         ring_buffer_record_enable(buffer);
768 }
769
770 void tracing_reset_online_cpus(struct trace_array *tr)
771 {
772         struct ring_buffer *buffer = tr->buffer;
773         int cpu;
774
775         ring_buffer_record_disable(buffer);
776
777         /* Make sure all commits have finished */
778         synchronize_sched();
779
780         tr->time_start = ftrace_now(tr->cpu);
781
782         for_each_online_cpu(cpu)
783                 __tracing_reset(tr, cpu);
784
785         ring_buffer_record_enable(buffer);
786 }
787
788 void tracing_reset_current(int cpu)
789 {
790         tracing_reset(&global_trace, cpu);
791 }
792
793 void tracing_reset_current_online_cpus(void)
794 {
795         tracing_reset_online_cpus(&global_trace);
796 }
797
798 #define SAVED_CMDLINES 128
799 #define NO_CMDLINE_MAP UINT_MAX
800 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
801 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
802 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
803 static int cmdline_idx;
804 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
805
806 /* temporary disable recording */
807 static atomic_t trace_record_cmdline_disabled __read_mostly;
808
809 static void trace_init_cmdlines(void)
810 {
811         memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
812         memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
813         cmdline_idx = 0;
814 }
815
816 int is_tracing_stopped(void)
817 {
818         return trace_stop_count;
819 }
820
821 /**
822  * ftrace_off_permanent - disable all ftrace code permanently
823  *
824  * This should only be called when a serious anomally has
825  * been detected.  This will turn off the function tracing,
826  * ring buffers, and other tracing utilites. It takes no
827  * locks and can be called from any context.
828  */
829 void ftrace_off_permanent(void)
830 {
831         tracing_disabled = 1;
832         ftrace_stop();
833         tracing_off_permanent();
834 }
835
836 /**
837  * tracing_start - quick start of the tracer
838  *
839  * If tracing is enabled but was stopped by tracing_stop,
840  * this will start the tracer back up.
841  */
842 void tracing_start(void)
843 {
844         struct ring_buffer *buffer;
845         unsigned long flags;
846
847         if (tracing_disabled)
848                 return;
849
850         spin_lock_irqsave(&tracing_start_lock, flags);
851         if (--trace_stop_count) {
852                 if (trace_stop_count < 0) {
853                         /* Someone screwed up their debugging */
854                         WARN_ON_ONCE(1);
855                         trace_stop_count = 0;
856                 }
857                 goto out;
858         }
859
860
861         buffer = global_trace.buffer;
862         if (buffer)
863                 ring_buffer_record_enable(buffer);
864
865         buffer = max_tr.buffer;
866         if (buffer)
867                 ring_buffer_record_enable(buffer);
868
869         ftrace_start();
870  out:
871         spin_unlock_irqrestore(&tracing_start_lock, flags);
872 }
873
874 /**
875  * tracing_stop - quick stop of the tracer
876  *
877  * Light weight way to stop tracing. Use in conjunction with
878  * tracing_start.
879  */
880 void tracing_stop(void)
881 {
882         struct ring_buffer *buffer;
883         unsigned long flags;
884
885         ftrace_stop();
886         spin_lock_irqsave(&tracing_start_lock, flags);
887         if (trace_stop_count++)
888                 goto out;
889
890         buffer = global_trace.buffer;
891         if (buffer)
892                 ring_buffer_record_disable(buffer);
893
894         buffer = max_tr.buffer;
895         if (buffer)
896                 ring_buffer_record_disable(buffer);
897
898  out:
899         spin_unlock_irqrestore(&tracing_start_lock, flags);
900 }
901
902 void trace_stop_cmdline_recording(void);
903
904 static void trace_save_cmdline(struct task_struct *tsk)
905 {
906         unsigned pid, idx;
907
908         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
909                 return;
910
911         /*
912          * It's not the end of the world if we don't get
913          * the lock, but we also don't want to spin
914          * nor do we want to disable interrupts,
915          * so if we miss here, then better luck next time.
916          */
917         if (!arch_spin_trylock(&trace_cmdline_lock))
918                 return;
919
920         idx = map_pid_to_cmdline[tsk->pid];
921         if (idx == NO_CMDLINE_MAP) {
922                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
923
924                 /*
925                  * Check whether the cmdline buffer at idx has a pid
926                  * mapped. We are going to overwrite that entry so we
927                  * need to clear the map_pid_to_cmdline. Otherwise we
928                  * would read the new comm for the old pid.
929                  */
930                 pid = map_cmdline_to_pid[idx];
931                 if (pid != NO_CMDLINE_MAP)
932                         map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
933
934                 map_cmdline_to_pid[idx] = tsk->pid;
935                 map_pid_to_cmdline[tsk->pid] = idx;
936
937                 cmdline_idx = idx;
938         }
939
940         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
941
942         arch_spin_unlock(&trace_cmdline_lock);
943 }
944
945 void trace_find_cmdline(int pid, char comm[])
946 {
947         unsigned map;
948
949         if (!pid) {
950                 strcpy(comm, "<idle>");
951                 return;
952         }
953
954         if (WARN_ON_ONCE(pid < 0)) {
955                 strcpy(comm, "<XXX>");
956                 return;
957         }
958
959         if (pid > PID_MAX_DEFAULT) {
960                 strcpy(comm, "<...>");
961                 return;
962         }
963
964         preempt_disable();
965         arch_spin_lock(&trace_cmdline_lock);
966         map = map_pid_to_cmdline[pid];
967         if (map != NO_CMDLINE_MAP)
968                 strcpy(comm, saved_cmdlines[map]);
969         else
970                 strcpy(comm, "<...>");
971
972         arch_spin_unlock(&trace_cmdline_lock);
973         preempt_enable();
974 }
975
976 void tracing_record_cmdline(struct task_struct *tsk)
977 {
978         if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
979             !tracing_is_on())
980                 return;
981
982         trace_save_cmdline(tsk);
983 }
984
985 void
986 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
987                              int pc)
988 {
989         struct task_struct *tsk = current;
990
991         entry->preempt_count            = pc & 0xff;
992         entry->pid                      = (tsk) ? tsk->pid : 0;
993         entry->lock_depth               = (tsk) ? tsk->lock_depth : 0;
994         entry->flags =
995 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
996                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
997 #else
998                 TRACE_FLAG_IRQS_NOSUPPORT |
999 #endif
1000                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1001                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1002                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1003 }
1004 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1005
1006 struct ring_buffer_event *
1007 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1008                           int type,
1009                           unsigned long len,
1010                           unsigned long flags, int pc)
1011 {
1012         struct ring_buffer_event *event;
1013
1014         event = ring_buffer_lock_reserve(buffer, len);
1015         if (event != NULL) {
1016                 struct trace_entry *ent = ring_buffer_event_data(event);
1017
1018                 tracing_generic_entry_update(ent, flags, pc);
1019                 ent->type = type;
1020         }
1021
1022         return event;
1023 }
1024
1025 static inline void
1026 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1027                              struct ring_buffer_event *event,
1028                              unsigned long flags, int pc,
1029                              int wake)
1030 {
1031         ring_buffer_unlock_commit(buffer, event);
1032
1033         ftrace_trace_stack(buffer, flags, 6, pc);
1034         ftrace_trace_userstack(buffer, flags, pc);
1035
1036         if (wake)
1037                 trace_wake_up();
1038 }
1039
1040 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1041                                 struct ring_buffer_event *event,
1042                                 unsigned long flags, int pc)
1043 {
1044         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1045 }
1046
1047 struct ring_buffer_event *
1048 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1049                                   int type, unsigned long len,
1050                                   unsigned long flags, int pc)
1051 {
1052         *current_rb = global_trace.buffer;
1053         return trace_buffer_lock_reserve(*current_rb,
1054                                          type, len, flags, pc);
1055 }
1056 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1057
1058 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1059                                         struct ring_buffer_event *event,
1060                                         unsigned long flags, int pc)
1061 {
1062         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1063 }
1064 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1065
1066 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1067                                        struct ring_buffer_event *event,
1068                                        unsigned long flags, int pc)
1069 {
1070         __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1071 }
1072 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1073
1074 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1075                                          struct ring_buffer_event *event)
1076 {
1077         ring_buffer_discard_commit(buffer, event);
1078 }
1079 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1080
1081 void
1082 trace_function(struct trace_array *tr,
1083                unsigned long ip, unsigned long parent_ip, unsigned long flags,
1084                int pc)
1085 {
1086         struct ftrace_event_call *call = &event_function;
1087         struct ring_buffer *buffer = tr->buffer;
1088         struct ring_buffer_event *event;
1089         struct ftrace_entry *entry;
1090
1091         /* If we are reading the ring buffer, don't trace */
1092         if (unlikely(__this_cpu_read(per_cpu_var(ftrace_cpu_disabled))))
1093                 return;
1094
1095         event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1096                                           flags, pc);
1097         if (!event)
1098                 return;
1099         entry   = ring_buffer_event_data(event);
1100         entry->ip                       = ip;
1101         entry->parent_ip                = parent_ip;
1102
1103         if (!filter_check_discard(call, entry, buffer, event))
1104                 ring_buffer_unlock_commit(buffer, event);
1105 }
1106
1107 void
1108 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1109        unsigned long ip, unsigned long parent_ip, unsigned long flags,
1110        int pc)
1111 {
1112         if (likely(!atomic_read(&data->disabled)))
1113                 trace_function(tr, ip, parent_ip, flags, pc);
1114 }
1115
1116 #ifdef CONFIG_STACKTRACE
1117 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1118                                  unsigned long flags,
1119                                  int skip, int pc)
1120 {
1121         struct ftrace_event_call *call = &event_kernel_stack;
1122         struct ring_buffer_event *event;
1123         struct stack_entry *entry;
1124         struct stack_trace trace;
1125
1126         event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1127                                           sizeof(*entry), flags, pc);
1128         if (!event)
1129                 return;
1130         entry   = ring_buffer_event_data(event);
1131         memset(&entry->caller, 0, sizeof(entry->caller));
1132
1133         trace.nr_entries        = 0;
1134         trace.max_entries       = FTRACE_STACK_ENTRIES;
1135         trace.skip              = skip;
1136         trace.entries           = entry->caller;
1137
1138         save_stack_trace(&trace);
1139         if (!filter_check_discard(call, entry, buffer, event))
1140                 ring_buffer_unlock_commit(buffer, event);
1141 }
1142
1143 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1144                         int skip, int pc)
1145 {
1146         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1147                 return;
1148
1149         __ftrace_trace_stack(buffer, flags, skip, pc);
1150 }
1151
1152 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1153                    int pc)
1154 {
1155         __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1156 }
1157
1158 /**
1159  * trace_dump_stack - record a stack back trace in the trace buffer
1160  */
1161 void trace_dump_stack(void)
1162 {
1163         unsigned long flags;
1164
1165         if (tracing_disabled || tracing_selftest_running)
1166                 return;
1167
1168         local_save_flags(flags);
1169
1170         /* skipping 3 traces, seems to get us at the caller of this function */
1171         __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1172 }
1173
1174 void
1175 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1176 {
1177         struct ftrace_event_call *call = &event_user_stack;
1178         struct ring_buffer_event *event;
1179         struct userstack_entry *entry;
1180         struct stack_trace trace;
1181
1182         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1183                 return;
1184
1185         event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1186                                           sizeof(*entry), flags, pc);
1187         if (!event)
1188                 return;
1189         entry   = ring_buffer_event_data(event);
1190
1191         entry->tgid             = current->tgid;
1192         memset(&entry->caller, 0, sizeof(entry->caller));
1193
1194         trace.nr_entries        = 0;
1195         trace.max_entries       = FTRACE_STACK_ENTRIES;
1196         trace.skip              = 0;
1197         trace.entries           = entry->caller;
1198
1199         save_stack_trace_user(&trace);
1200         if (!filter_check_discard(call, entry, buffer, event))
1201                 ring_buffer_unlock_commit(buffer, event);
1202 }
1203
1204 #ifdef UNUSED
1205 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1206 {
1207         ftrace_trace_userstack(tr, flags, preempt_count());
1208 }
1209 #endif /* UNUSED */
1210
1211 #endif /* CONFIG_STACKTRACE */
1212
1213 static void
1214 ftrace_trace_special(void *__tr,
1215                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1216                      int pc)
1217 {
1218         struct ftrace_event_call *call = &event_special;
1219         struct ring_buffer_event *event;
1220         struct trace_array *tr = __tr;
1221         struct ring_buffer *buffer = tr->buffer;
1222         struct special_entry *entry;
1223
1224         event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1225                                           sizeof(*entry), 0, pc);
1226         if (!event)
1227                 return;
1228         entry   = ring_buffer_event_data(event);
1229         entry->arg1                     = arg1;
1230         entry->arg2                     = arg2;
1231         entry->arg3                     = arg3;
1232
1233         if (!filter_check_discard(call, entry, buffer, event))
1234                 trace_buffer_unlock_commit(buffer, event, 0, pc);
1235 }
1236
1237 void
1238 __trace_special(void *__tr, void *__data,
1239                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1240 {
1241         ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1242 }
1243
1244 void
1245 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1246 {
1247         struct trace_array *tr = &global_trace;
1248         struct trace_array_cpu *data;
1249         unsigned long flags;
1250         int cpu;
1251         int pc;
1252
1253         if (tracing_disabled)
1254                 return;
1255
1256         pc = preempt_count();
1257         local_irq_save(flags);
1258         cpu = raw_smp_processor_id();
1259         data = tr->data[cpu];
1260
1261         if (likely(atomic_inc_return(&data->disabled) == 1))
1262                 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1263
1264         atomic_dec(&data->disabled);
1265         local_irq_restore(flags);
1266 }
1267
1268 /**
1269  * trace_vbprintk - write binary msg to tracing buffer
1270  *
1271  */
1272 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1273 {
1274         static arch_spinlock_t trace_buf_lock =
1275                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1276         static u32 trace_buf[TRACE_BUF_SIZE];
1277
1278         struct ftrace_event_call *call = &event_bprint;
1279         struct ring_buffer_event *event;
1280         struct ring_buffer *buffer;
1281         struct trace_array *tr = &global_trace;
1282         struct trace_array_cpu *data;
1283         struct bprint_entry *entry;
1284         unsigned long flags;
1285         int disable;
1286         int resched;
1287         int cpu, len = 0, size, pc;
1288
1289         if (unlikely(tracing_selftest_running || tracing_disabled))
1290                 return 0;
1291
1292         /* Don't pollute graph traces with trace_vprintk internals */
1293         pause_graph_tracing();
1294
1295         pc = preempt_count();
1296         resched = ftrace_preempt_disable();
1297         cpu = raw_smp_processor_id();
1298         data = tr->data[cpu];
1299
1300         disable = atomic_inc_return(&data->disabled);
1301         if (unlikely(disable != 1))
1302                 goto out;
1303
1304         /* Lockdep uses trace_printk for lock tracing */
1305         local_irq_save(flags);
1306         arch_spin_lock(&trace_buf_lock);
1307         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1308
1309         if (len > TRACE_BUF_SIZE || len < 0)
1310                 goto out_unlock;
1311
1312         size = sizeof(*entry) + sizeof(u32) * len;
1313         buffer = tr->buffer;
1314         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1315                                           flags, pc);
1316         if (!event)
1317                 goto out_unlock;
1318         entry = ring_buffer_event_data(event);
1319         entry->ip                       = ip;
1320         entry->fmt                      = fmt;
1321
1322         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1323         if (!filter_check_discard(call, entry, buffer, event))
1324                 ring_buffer_unlock_commit(buffer, event);
1325
1326 out_unlock:
1327         arch_spin_unlock(&trace_buf_lock);
1328         local_irq_restore(flags);
1329
1330 out:
1331         atomic_dec_return(&data->disabled);
1332         ftrace_preempt_enable(resched);
1333         unpause_graph_tracing();
1334
1335         return len;
1336 }
1337 EXPORT_SYMBOL_GPL(trace_vbprintk);
1338
1339 int trace_array_printk(struct trace_array *tr,
1340                        unsigned long ip, const char *fmt, ...)
1341 {
1342         int ret;
1343         va_list ap;
1344
1345         if (!(trace_flags & TRACE_ITER_PRINTK))
1346                 return 0;
1347
1348         va_start(ap, fmt);
1349         ret = trace_array_vprintk(tr, ip, fmt, ap);
1350         va_end(ap);
1351         return ret;
1352 }
1353
1354 int trace_array_vprintk(struct trace_array *tr,
1355                         unsigned long ip, const char *fmt, va_list args)
1356 {
1357         static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1358         static char trace_buf[TRACE_BUF_SIZE];
1359
1360         struct ftrace_event_call *call = &event_print;
1361         struct ring_buffer_event *event;
1362         struct ring_buffer *buffer;
1363         struct trace_array_cpu *data;
1364         int cpu, len = 0, size, pc;
1365         struct print_entry *entry;
1366         unsigned long irq_flags;
1367         int disable;
1368
1369         if (tracing_disabled || tracing_selftest_running)
1370                 return 0;
1371
1372         pc = preempt_count();
1373         preempt_disable_notrace();
1374         cpu = raw_smp_processor_id();
1375         data = tr->data[cpu];
1376
1377         disable = atomic_inc_return(&data->disabled);
1378         if (unlikely(disable != 1))
1379                 goto out;
1380
1381         pause_graph_tracing();
1382         raw_local_irq_save(irq_flags);
1383         arch_spin_lock(&trace_buf_lock);
1384         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1385
1386         size = sizeof(*entry) + len + 1;
1387         buffer = tr->buffer;
1388         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1389                                           irq_flags, pc);
1390         if (!event)
1391                 goto out_unlock;
1392         entry = ring_buffer_event_data(event);
1393         entry->ip = ip;
1394
1395         memcpy(&entry->buf, trace_buf, len);
1396         entry->buf[len] = '\0';
1397         if (!filter_check_discard(call, entry, buffer, event))
1398                 ring_buffer_unlock_commit(buffer, event);
1399
1400  out_unlock:
1401         arch_spin_unlock(&trace_buf_lock);
1402         raw_local_irq_restore(irq_flags);
1403         unpause_graph_tracing();
1404  out:
1405         atomic_dec_return(&data->disabled);
1406         preempt_enable_notrace();
1407
1408         return len;
1409 }
1410
1411 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1412 {
1413         return trace_array_vprintk(&global_trace, ip, fmt, args);
1414 }
1415 EXPORT_SYMBOL_GPL(trace_vprintk);
1416
1417 enum trace_file_type {
1418         TRACE_FILE_LAT_FMT      = 1,
1419         TRACE_FILE_ANNOTATE     = 2,
1420 };
1421
1422 static void trace_iterator_increment(struct trace_iterator *iter)
1423 {
1424         /* Don't allow ftrace to trace into the ring buffers */
1425         ftrace_disable_cpu();
1426
1427         iter->idx++;
1428         if (iter->buffer_iter[iter->cpu])
1429                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1430
1431         ftrace_enable_cpu();
1432 }
1433
1434 static struct trace_entry *
1435 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1436 {
1437         struct ring_buffer_event *event;
1438         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1439
1440         /* Don't allow ftrace to trace into the ring buffers */
1441         ftrace_disable_cpu();
1442
1443         if (buf_iter)
1444                 event = ring_buffer_iter_peek(buf_iter, ts);
1445         else
1446                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1447
1448         ftrace_enable_cpu();
1449
1450         return event ? ring_buffer_event_data(event) : NULL;
1451 }
1452
1453 static struct trace_entry *
1454 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1455 {
1456         struct ring_buffer *buffer = iter->tr->buffer;
1457         struct trace_entry *ent, *next = NULL;
1458         int cpu_file = iter->cpu_file;
1459         u64 next_ts = 0, ts;
1460         int next_cpu = -1;
1461         int cpu;
1462
1463         /*
1464          * If we are in a per_cpu trace file, don't bother by iterating over
1465          * all cpu and peek directly.
1466          */
1467         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1468                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1469                         return NULL;
1470                 ent = peek_next_entry(iter, cpu_file, ent_ts);
1471                 if (ent_cpu)
1472                         *ent_cpu = cpu_file;
1473
1474                 return ent;
1475         }
1476
1477         for_each_tracing_cpu(cpu) {
1478
1479                 if (ring_buffer_empty_cpu(buffer, cpu))
1480                         continue;
1481
1482                 ent = peek_next_entry(iter, cpu, &ts);
1483
1484                 /*
1485                  * Pick the entry with the smallest timestamp:
1486                  */
1487                 if (ent && (!next || ts < next_ts)) {
1488                         next = ent;
1489                         next_cpu = cpu;
1490                         next_ts = ts;
1491                 }
1492         }
1493
1494         if (ent_cpu)
1495                 *ent_cpu = next_cpu;
1496
1497         if (ent_ts)
1498                 *ent_ts = next_ts;
1499
1500         return next;
1501 }
1502
1503 /* Find the next real entry, without updating the iterator itself */
1504 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1505                                           int *ent_cpu, u64 *ent_ts)
1506 {
1507         return __find_next_entry(iter, ent_cpu, ent_ts);
1508 }
1509
1510 /* Find the next real entry, and increment the iterator to the next entry */
1511 static void *find_next_entry_inc(struct trace_iterator *iter)
1512 {
1513         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1514
1515         if (iter->ent)
1516                 trace_iterator_increment(iter);
1517
1518         return iter->ent ? iter : NULL;
1519 }
1520
1521 static void trace_consume(struct trace_iterator *iter)
1522 {
1523         /* Don't allow ftrace to trace into the ring buffers */
1524         ftrace_disable_cpu();
1525         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1526         ftrace_enable_cpu();
1527 }
1528
1529 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1530 {
1531         struct trace_iterator *iter = m->private;
1532         int i = (int)*pos;
1533         void *ent;
1534
1535         WARN_ON_ONCE(iter->leftover);
1536
1537         (*pos)++;
1538
1539         /* can't go backwards */
1540         if (iter->idx > i)
1541                 return NULL;
1542
1543         if (iter->idx < 0)
1544                 ent = find_next_entry_inc(iter);
1545         else
1546                 ent = iter;
1547
1548         while (ent && iter->idx < i)
1549                 ent = find_next_entry_inc(iter);
1550
1551         iter->pos = *pos;
1552
1553         return ent;
1554 }
1555
1556 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1557 {
1558         struct trace_array *tr = iter->tr;
1559         struct ring_buffer_event *event;
1560         struct ring_buffer_iter *buf_iter;
1561         unsigned long entries = 0;
1562         u64 ts;
1563
1564         tr->data[cpu]->skipped_entries = 0;
1565
1566         if (!iter->buffer_iter[cpu])
1567                 return;
1568
1569         buf_iter = iter->buffer_iter[cpu];
1570         ring_buffer_iter_reset(buf_iter);
1571
1572         /*
1573          * We could have the case with the max latency tracers
1574          * that a reset never took place on a cpu. This is evident
1575          * by the timestamp being before the start of the buffer.
1576          */
1577         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1578                 if (ts >= iter->tr->time_start)
1579                         break;
1580                 entries++;
1581                 ring_buffer_read(buf_iter, NULL);
1582         }
1583
1584         tr->data[cpu]->skipped_entries = entries;
1585 }
1586
1587 /*
1588  * No necessary locking here. The worst thing which can
1589  * happen is loosing events consumed at the same time
1590  * by a trace_pipe reader.
1591  * Other than that, we don't risk to crash the ring buffer
1592  * because it serializes the readers.
1593  *
1594  * The current tracer is copied to avoid a global locking
1595  * all around.
1596  */
1597 static void *s_start(struct seq_file *m, loff_t *pos)
1598 {
1599         struct trace_iterator *iter = m->private;
1600         static struct tracer *old_tracer;
1601         int cpu_file = iter->cpu_file;
1602         void *p = NULL;
1603         loff_t l = 0;
1604         int cpu;
1605
1606         /* copy the tracer to avoid using a global lock all around */
1607         mutex_lock(&trace_types_lock);
1608         if (unlikely(old_tracer != current_trace && current_trace)) {
1609                 old_tracer = current_trace;
1610                 *iter->trace = *current_trace;
1611         }
1612         mutex_unlock(&trace_types_lock);
1613
1614         atomic_inc(&trace_record_cmdline_disabled);
1615
1616         if (*pos != iter->pos) {
1617                 iter->ent = NULL;
1618                 iter->cpu = 0;
1619                 iter->idx = -1;
1620
1621                 ftrace_disable_cpu();
1622
1623                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1624                         for_each_tracing_cpu(cpu)
1625                                 tracing_iter_reset(iter, cpu);
1626                 } else
1627                         tracing_iter_reset(iter, cpu_file);
1628
1629                 ftrace_enable_cpu();
1630
1631                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1632                         ;
1633
1634         } else {
1635                 /*
1636                  * If we overflowed the seq_file before, then we want
1637                  * to just reuse the trace_seq buffer again.
1638                  */
1639                 if (iter->leftover)
1640                         p = iter;
1641                 else {
1642                         l = *pos - 1;
1643                         p = s_next(m, p, &l);
1644                 }
1645         }
1646
1647         trace_event_read_lock();
1648         return p;
1649 }
1650
1651 static void s_stop(struct seq_file *m, void *p)
1652 {
1653         atomic_dec(&trace_record_cmdline_disabled);
1654         trace_event_read_unlock();
1655 }
1656
1657 static void print_lat_help_header(struct seq_file *m)
1658 {
1659         seq_puts(m, "#                  _------=> CPU#            \n");
1660         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1661         seq_puts(m, "#                | / _----=> need-resched    \n");
1662         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1663         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1664         seq_puts(m, "#                |||| /_--=> lock-depth       \n");
1665         seq_puts(m, "#                |||||/     delay             \n");
1666         seq_puts(m, "#  cmd     pid   |||||| time  |   caller      \n");
1667         seq_puts(m, "#     \\   /      ||||||   \\   |   /           \n");
1668 }
1669
1670 static void print_func_help_header(struct seq_file *m)
1671 {
1672         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1673         seq_puts(m, "#              | |       |          |         |\n");
1674 }
1675
1676
1677 static void
1678 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1679 {
1680         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1681         struct trace_array *tr = iter->tr;
1682         struct trace_array_cpu *data = tr->data[tr->cpu];
1683         struct tracer *type = current_trace;
1684         unsigned long entries = 0;
1685         unsigned long total = 0;
1686         unsigned long count;
1687         const char *name = "preemption";
1688         int cpu;
1689
1690         if (type)
1691                 name = type->name;
1692
1693
1694         for_each_tracing_cpu(cpu) {
1695                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1696                 /*
1697                  * If this buffer has skipped entries, then we hold all
1698                  * entries for the trace and we need to ignore the
1699                  * ones before the time stamp.
1700                  */
1701                 if (tr->data[cpu]->skipped_entries) {
1702                         count -= tr->data[cpu]->skipped_entries;
1703                         /* total is the same as the entries */
1704                         total += count;
1705                 } else
1706                         total += count +
1707                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1708                 entries += count;
1709         }
1710
1711         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1712                    name, UTS_RELEASE);
1713         seq_puts(m, "# -----------------------------------"
1714                  "---------------------------------\n");
1715         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1716                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1717                    nsecs_to_usecs(data->saved_latency),
1718                    entries,
1719                    total,
1720                    tr->cpu,
1721 #if defined(CONFIG_PREEMPT_NONE)
1722                    "server",
1723 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1724                    "desktop",
1725 #elif defined(CONFIG_PREEMPT)
1726                    "preempt",
1727 #else
1728                    "unknown",
1729 #endif
1730                    /* These are reserved for later use */
1731                    0, 0, 0, 0);
1732 #ifdef CONFIG_SMP
1733         seq_printf(m, " #P:%d)\n", num_online_cpus());
1734 #else
1735         seq_puts(m, ")\n");
1736 #endif
1737         seq_puts(m, "#    -----------------\n");
1738         seq_printf(m, "#    | task: %.16s-%d "
1739                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1740                    data->comm, data->pid, data->uid, data->nice,
1741                    data->policy, data->rt_priority);
1742         seq_puts(m, "#    -----------------\n");
1743
1744         if (data->critical_start) {
1745                 seq_puts(m, "#  => started at: ");
1746                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1747                 trace_print_seq(m, &iter->seq);
1748                 seq_puts(m, "\n#  => ended at:   ");
1749                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1750                 trace_print_seq(m, &iter->seq);
1751                 seq_puts(m, "\n#\n");
1752         }
1753
1754         seq_puts(m, "#\n");
1755 }
1756
1757 static void test_cpu_buff_start(struct trace_iterator *iter)
1758 {
1759         struct trace_seq *s = &iter->seq;
1760
1761         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1762                 return;
1763
1764         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1765                 return;
1766
1767         if (cpumask_test_cpu(iter->cpu, iter->started))
1768                 return;
1769
1770         if (iter->tr->data[iter->cpu]->skipped_entries)
1771                 return;
1772
1773         cpumask_set_cpu(iter->cpu, iter->started);
1774
1775         /* Don't print started cpu buffer for the first entry of the trace */
1776         if (iter->idx > 1)
1777                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1778                                 iter->cpu);
1779 }
1780
1781 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1782 {
1783         struct trace_seq *s = &iter->seq;
1784         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1785         struct trace_entry *entry;
1786         struct trace_event *event;
1787
1788         entry = iter->ent;
1789
1790         test_cpu_buff_start(iter);
1791
1792         event = ftrace_find_event(entry->type);
1793
1794         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1795                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1796                         if (!trace_print_lat_context(iter))
1797                                 goto partial;
1798                 } else {
1799                         if (!trace_print_context(iter))
1800                                 goto partial;
1801                 }
1802         }
1803
1804         if (event)
1805                 return event->trace(iter, sym_flags);
1806
1807         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1808                 goto partial;
1809
1810         return TRACE_TYPE_HANDLED;
1811 partial:
1812         return TRACE_TYPE_PARTIAL_LINE;
1813 }
1814
1815 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1816 {
1817         struct trace_seq *s = &iter->seq;
1818         struct trace_entry *entry;
1819         struct trace_event *event;
1820
1821         entry = iter->ent;
1822
1823         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1824                 if (!trace_seq_printf(s, "%d %d %llu ",
1825                                       entry->pid, iter->cpu, iter->ts))
1826                         goto partial;
1827         }
1828
1829         event = ftrace_find_event(entry->type);
1830         if (event)
1831                 return event->raw(iter, 0);
1832
1833         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1834                 goto partial;
1835
1836         return TRACE_TYPE_HANDLED;
1837 partial:
1838         return TRACE_TYPE_PARTIAL_LINE;
1839 }
1840
1841 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1842 {
1843         struct trace_seq *s = &iter->seq;
1844         unsigned char newline = '\n';
1845         struct trace_entry *entry;
1846         struct trace_event *event;
1847
1848         entry = iter->ent;
1849
1850         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1851                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1852                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1853                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1854         }
1855
1856         event = ftrace_find_event(entry->type);
1857         if (event) {
1858                 enum print_line_t ret = event->hex(iter, 0);
1859                 if (ret != TRACE_TYPE_HANDLED)
1860                         return ret;
1861         }
1862
1863         SEQ_PUT_FIELD_RET(s, newline);
1864
1865         return TRACE_TYPE_HANDLED;
1866 }
1867
1868 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1869 {
1870         struct trace_seq *s = &iter->seq;
1871         struct trace_entry *entry;
1872         struct trace_event *event;
1873
1874         entry = iter->ent;
1875
1876         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1877                 SEQ_PUT_FIELD_RET(s, entry->pid);
1878                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1879                 SEQ_PUT_FIELD_RET(s, iter->ts);
1880         }
1881
1882         event = ftrace_find_event(entry->type);
1883         return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1884 }
1885
1886 static int trace_empty(struct trace_iterator *iter)
1887 {
1888         int cpu;
1889
1890         /* If we are looking at one CPU buffer, only check that one */
1891         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1892                 cpu = iter->cpu_file;
1893                 if (iter->buffer_iter[cpu]) {
1894                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1895                                 return 0;
1896                 } else {
1897                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1898                                 return 0;
1899                 }
1900                 return 1;
1901         }
1902
1903         for_each_tracing_cpu(cpu) {
1904                 if (iter->buffer_iter[cpu]) {
1905                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1906                                 return 0;
1907                 } else {
1908                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1909                                 return 0;
1910                 }
1911         }
1912
1913         return 1;
1914 }
1915
1916 /*  Called with trace_event_read_lock() held. */
1917 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1918 {
1919         enum print_line_t ret;
1920
1921         if (iter->trace && iter->trace->print_line) {
1922                 ret = iter->trace->print_line(iter);
1923                 if (ret != TRACE_TYPE_UNHANDLED)
1924                         return ret;
1925         }
1926
1927         if (iter->ent->type == TRACE_BPRINT &&
1928                         trace_flags & TRACE_ITER_PRINTK &&
1929                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1930                 return trace_print_bprintk_msg_only(iter);
1931
1932         if (iter->ent->type == TRACE_PRINT &&
1933                         trace_flags & TRACE_ITER_PRINTK &&
1934                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1935                 return trace_print_printk_msg_only(iter);
1936
1937         if (trace_flags & TRACE_ITER_BIN)
1938                 return print_bin_fmt(iter);
1939
1940         if (trace_flags & TRACE_ITER_HEX)
1941                 return print_hex_fmt(iter);
1942
1943         if (trace_flags & TRACE_ITER_RAW)
1944                 return print_raw_fmt(iter);
1945
1946         return print_trace_fmt(iter);
1947 }
1948
1949 static int s_show(struct seq_file *m, void *v)
1950 {
1951         struct trace_iterator *iter = v;
1952         int ret;
1953
1954         if (iter->ent == NULL) {
1955                 if (iter->tr) {
1956                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1957                         seq_puts(m, "#\n");
1958                 }
1959                 if (iter->trace && iter->trace->print_header)
1960                         iter->trace->print_header(m);
1961                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1962                         /* print nothing if the buffers are empty */
1963                         if (trace_empty(iter))
1964                                 return 0;
1965                         print_trace_header(m, iter);
1966                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1967                                 print_lat_help_header(m);
1968                 } else {
1969                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1970                                 print_func_help_header(m);
1971                 }
1972         } else if (iter->leftover) {
1973                 /*
1974                  * If we filled the seq_file buffer earlier, we
1975                  * want to just show it now.
1976                  */
1977                 ret = trace_print_seq(m, &iter->seq);
1978
1979                 /* ret should this time be zero, but you never know */
1980                 iter->leftover = ret;
1981
1982         } else {
1983                 print_trace_line(iter);
1984                 ret = trace_print_seq(m, &iter->seq);
1985                 /*
1986                  * If we overflow the seq_file buffer, then it will
1987                  * ask us for this data again at start up.
1988                  * Use that instead.
1989                  *  ret is 0 if seq_file write succeeded.
1990                  *        -1 otherwise.
1991                  */
1992                 iter->leftover = ret;
1993         }
1994
1995         return 0;
1996 }
1997
1998 static const struct seq_operations tracer_seq_ops = {
1999         .start          = s_start,
2000         .next           = s_next,
2001         .stop           = s_stop,
2002         .show           = s_show,
2003 };
2004
2005 static struct trace_iterator *
2006 __tracing_open(struct inode *inode, struct file *file)
2007 {
2008         long cpu_file = (long) inode->i_private;
2009         void *fail_ret = ERR_PTR(-ENOMEM);
2010         struct trace_iterator *iter;
2011         struct seq_file *m;
2012         int cpu, ret;
2013
2014         if (tracing_disabled)
2015                 return ERR_PTR(-ENODEV);
2016
2017         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2018         if (!iter)
2019                 return ERR_PTR(-ENOMEM);
2020
2021         /*
2022          * We make a copy of the current tracer to avoid concurrent
2023          * changes on it while we are reading.
2024          */
2025         mutex_lock(&trace_types_lock);
2026         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2027         if (!iter->trace)
2028                 goto fail;
2029
2030         if (current_trace)
2031                 *iter->trace = *current_trace;
2032
2033         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2034                 goto fail;
2035
2036         if (current_trace && current_trace->print_max)
2037                 iter->tr = &max_tr;
2038         else
2039                 iter->tr = &global_trace;
2040         iter->pos = -1;
2041         mutex_init(&iter->mutex);
2042         iter->cpu_file = cpu_file;
2043
2044         /* Notify the tracer early; before we stop tracing. */
2045         if (iter->trace && iter->trace->open)
2046                 iter->trace->open(iter);
2047
2048         /* Annotate start of buffers if we had overruns */
2049         if (ring_buffer_overruns(iter->tr->buffer))
2050                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2051
2052         /* stop the trace while dumping */
2053         tracing_stop();
2054
2055         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2056                 for_each_tracing_cpu(cpu) {
2057
2058                         iter->buffer_iter[cpu] =
2059                                 ring_buffer_read_start(iter->tr->buffer, cpu);
2060                         tracing_iter_reset(iter, cpu);
2061                 }
2062         } else {
2063                 cpu = iter->cpu_file;
2064                 iter->buffer_iter[cpu] =
2065                                 ring_buffer_read_start(iter->tr->buffer, cpu);
2066                 tracing_iter_reset(iter, cpu);
2067         }
2068
2069         ret = seq_open(file, &tracer_seq_ops);
2070         if (ret < 0) {
2071                 fail_ret = ERR_PTR(ret);
2072                 goto fail_buffer;
2073         }
2074
2075         m = file->private_data;
2076         m->private = iter;
2077
2078         mutex_unlock(&trace_types_lock);
2079
2080         return iter;
2081
2082  fail_buffer:
2083         for_each_tracing_cpu(cpu) {
2084                 if (iter->buffer_iter[cpu])
2085                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2086         }
2087         free_cpumask_var(iter->started);
2088         tracing_start();
2089  fail:
2090         mutex_unlock(&trace_types_lock);
2091         kfree(iter->trace);
2092         kfree(iter);
2093
2094         return fail_ret;
2095 }
2096
2097 int tracing_open_generic(struct inode *inode, struct file *filp)
2098 {
2099         if (tracing_disabled)
2100                 return -ENODEV;
2101
2102         filp->private_data = inode->i_private;
2103         return 0;
2104 }
2105
2106 static int tracing_release(struct inode *inode, struct file *file)
2107 {
2108         struct seq_file *m = (struct seq_file *)file->private_data;
2109         struct trace_iterator *iter;
2110         int cpu;
2111
2112         if (!(file->f_mode & FMODE_READ))
2113                 return 0;
2114
2115         iter = m->private;
2116
2117         mutex_lock(&trace_types_lock);
2118         for_each_tracing_cpu(cpu) {
2119                 if (iter->buffer_iter[cpu])
2120                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2121         }
2122
2123         if (iter->trace && iter->trace->close)
2124                 iter->trace->close(iter);
2125
2126         /* reenable tracing if it was previously enabled */
2127         tracing_start();
2128         mutex_unlock(&trace_types_lock);
2129
2130         seq_release(inode, file);
2131         mutex_destroy(&iter->mutex);
2132         free_cpumask_var(iter->started);
2133         kfree(iter->trace);
2134         kfree(iter);
2135         return 0;
2136 }
2137
2138 static int tracing_open(struct inode *inode, struct file *file)
2139 {
2140         struct trace_iterator *iter;
2141         int ret = 0;
2142
2143         /* If this file was open for write, then erase contents */
2144         if ((file->f_mode & FMODE_WRITE) &&
2145             (file->f_flags & O_TRUNC)) {
2146                 long cpu = (long) inode->i_private;
2147
2148                 if (cpu == TRACE_PIPE_ALL_CPU)
2149                         tracing_reset_online_cpus(&global_trace);
2150                 else
2151                         tracing_reset(&global_trace, cpu);
2152         }
2153
2154         if (file->f_mode & FMODE_READ) {
2155                 iter = __tracing_open(inode, file);
2156                 if (IS_ERR(iter))
2157                         ret = PTR_ERR(iter);
2158                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2159                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2160         }
2161         return ret;
2162 }
2163
2164 static void *
2165 t_next(struct seq_file *m, void *v, loff_t *pos)
2166 {
2167         struct tracer *t = v;
2168
2169         (*pos)++;
2170
2171         if (t)
2172                 t = t->next;
2173
2174         return t;
2175 }
2176
2177 static void *t_start(struct seq_file *m, loff_t *pos)
2178 {
2179         struct tracer *t;
2180         loff_t l = 0;
2181
2182         mutex_lock(&trace_types_lock);
2183         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2184                 ;
2185
2186         return t;
2187 }
2188
2189 static void t_stop(struct seq_file *m, void *p)
2190 {
2191         mutex_unlock(&trace_types_lock);
2192 }
2193
2194 static int t_show(struct seq_file *m, void *v)
2195 {
2196         struct tracer *t = v;
2197
2198         if (!t)
2199                 return 0;
2200
2201         seq_printf(m, "%s", t->name);
2202         if (t->next)
2203                 seq_putc(m, ' ');
2204         else
2205                 seq_putc(m, '\n');
2206
2207         return 0;
2208 }
2209
2210 static const struct seq_operations show_traces_seq_ops = {
2211         .start          = t_start,
2212         .next           = t_next,
2213         .stop           = t_stop,
2214         .show           = t_show,
2215 };
2216
2217 static int show_traces_open(struct inode *inode, struct file *file)
2218 {
2219         if (tracing_disabled)
2220                 return -ENODEV;
2221
2222         return seq_open(file, &show_traces_seq_ops);
2223 }
2224
2225 static ssize_t
2226 tracing_write_stub(struct file *filp, const char __user *ubuf,
2227                    size_t count, loff_t *ppos)
2228 {
2229         return count;
2230 }
2231
2232 static const struct file_operations tracing_fops = {
2233         .open           = tracing_open,
2234         .read           = seq_read,
2235         .write          = tracing_write_stub,
2236         .llseek         = seq_lseek,
2237         .release        = tracing_release,
2238 };
2239
2240 static const struct file_operations show_traces_fops = {
2241         .open           = show_traces_open,
2242         .read           = seq_read,
2243         .release        = seq_release,
2244 };
2245
2246 /*
2247  * Only trace on a CPU if the bitmask is set:
2248  */
2249 static cpumask_var_t tracing_cpumask;
2250
2251 /*
2252  * The tracer itself will not take this lock, but still we want
2253  * to provide a consistent cpumask to user-space:
2254  */
2255 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2256
2257 /*
2258  * Temporary storage for the character representation of the
2259  * CPU bitmask (and one more byte for the newline):
2260  */
2261 static char mask_str[NR_CPUS + 1];
2262
2263 static ssize_t
2264 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2265                      size_t count, loff_t *ppos)
2266 {
2267         int len;
2268
2269         mutex_lock(&tracing_cpumask_update_lock);
2270
2271         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2272         if (count - len < 2) {
2273                 count = -EINVAL;
2274                 goto out_err;
2275         }
2276         len += sprintf(mask_str + len, "\n");
2277         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2278
2279 out_err:
2280         mutex_unlock(&tracing_cpumask_update_lock);
2281
2282         return count;
2283 }
2284
2285 static ssize_t
2286 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2287                       size_t count, loff_t *ppos)
2288 {
2289         int err, cpu;
2290         cpumask_var_t tracing_cpumask_new;
2291
2292         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2293                 return -ENOMEM;
2294
2295         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2296         if (err)
2297                 goto err_unlock;
2298
2299         mutex_lock(&tracing_cpumask_update_lock);
2300
2301         local_irq_disable();
2302         arch_spin_lock(&ftrace_max_lock);
2303         for_each_tracing_cpu(cpu) {
2304                 /*
2305                  * Increase/decrease the disabled counter if we are
2306                  * about to flip a bit in the cpumask:
2307                  */
2308                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2309                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2310                         atomic_inc(&global_trace.data[cpu]->disabled);
2311                 }
2312                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2313                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2314                         atomic_dec(&global_trace.data[cpu]->disabled);
2315                 }
2316         }
2317         arch_spin_unlock(&ftrace_max_lock);
2318         local_irq_enable();
2319
2320         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2321
2322         mutex_unlock(&tracing_cpumask_update_lock);
2323         free_cpumask_var(tracing_cpumask_new);
2324
2325         return count;
2326
2327 err_unlock:
2328         free_cpumask_var(tracing_cpumask_new);
2329
2330         return err;
2331 }
2332
2333 static const struct file_operations tracing_cpumask_fops = {
2334         .open           = tracing_open_generic,
2335         .read           = tracing_cpumask_read,
2336         .write          = tracing_cpumask_write,
2337 };
2338
2339 static int tracing_trace_options_show(struct seq_file *m, void *v)
2340 {
2341         struct tracer_opt *trace_opts;
2342         u32 tracer_flags;
2343         int i;
2344
2345         mutex_lock(&trace_types_lock);
2346         tracer_flags = current_trace->flags->val;
2347         trace_opts = current_trace->flags->opts;
2348
2349         for (i = 0; trace_options[i]; i++) {
2350                 if (trace_flags & (1 << i))
2351                         seq_printf(m, "%s\n", trace_options[i]);
2352                 else
2353                         seq_printf(m, "no%s\n", trace_options[i]);
2354         }
2355
2356         for (i = 0; trace_opts[i].name; i++) {
2357                 if (tracer_flags & trace_opts[i].bit)
2358                         seq_printf(m, "%s\n", trace_opts[i].name);
2359                 else
2360                         seq_printf(m, "no%s\n", trace_opts[i].name);
2361         }
2362         mutex_unlock(&trace_types_lock);
2363
2364         return 0;
2365 }
2366
2367 static int __set_tracer_option(struct tracer *trace,
2368                                struct tracer_flags *tracer_flags,
2369                                struct tracer_opt *opts, int neg)
2370 {
2371         int ret;
2372
2373         ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2374         if (ret)
2375                 return ret;
2376
2377         if (neg)
2378                 tracer_flags->val &= ~opts->bit;
2379         else
2380                 tracer_flags->val |= opts->bit;
2381         return 0;
2382 }
2383
2384 /* Try to assign a tracer specific option */
2385 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2386 {
2387         struct tracer_flags *tracer_flags = trace->flags;
2388         struct tracer_opt *opts = NULL;
2389         int i;
2390
2391         for (i = 0; tracer_flags->opts[i].name; i++) {
2392                 opts = &tracer_flags->opts[i];
2393
2394                 if (strcmp(cmp, opts->name) == 0)
2395                         return __set_tracer_option(trace, trace->flags,
2396                                                    opts, neg);
2397         }
2398
2399         return -EINVAL;
2400 }
2401
2402 static void set_tracer_flags(unsigned int mask, int enabled)
2403 {
2404         /* do nothing if flag is already set */
2405         if (!!(trace_flags & mask) == !!enabled)
2406                 return;
2407
2408         if (enabled)
2409                 trace_flags |= mask;
2410         else
2411                 trace_flags &= ~mask;
2412 }
2413
2414 static ssize_t
2415 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2416                         size_t cnt, loff_t *ppos)
2417 {
2418         char buf[64];
2419         char *cmp;
2420         int neg = 0;
2421         int ret;
2422         int i;
2423
2424         if (cnt >= sizeof(buf))
2425                 return -EINVAL;
2426
2427         if (copy_from_user(&buf, ubuf, cnt))
2428                 return -EFAULT;
2429
2430         buf[cnt] = 0;
2431         cmp = strstrip(buf);
2432
2433         if (strncmp(cmp, "no", 2) == 0) {
2434                 neg = 1;
2435                 cmp += 2;
2436         }
2437
2438         for (i = 0; trace_options[i]; i++) {
2439                 if (strcmp(cmp, trace_options[i]) == 0) {
2440                         set_tracer_flags(1 << i, !neg);
2441                         break;
2442                 }
2443         }
2444
2445         /* If no option could be set, test the specific tracer options */
2446         if (!trace_options[i]) {
2447                 mutex_lock(&trace_types_lock);
2448                 ret = set_tracer_option(current_trace, cmp, neg);
2449                 mutex_unlock(&trace_types_lock);
2450                 if (ret)
2451                         return ret;
2452         }
2453
2454         *ppos += cnt;
2455
2456         return cnt;
2457 }
2458
2459 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2460 {
2461         if (tracing_disabled)
2462                 return -ENODEV;
2463         return single_open(file, tracing_trace_options_show, NULL);
2464 }
2465
2466 static const struct file_operations tracing_iter_fops = {
2467         .open           = tracing_trace_options_open,
2468         .read           = seq_read,
2469         .llseek         = seq_lseek,
2470         .release        = single_release,
2471         .write          = tracing_trace_options_write,
2472 };
2473
2474 static const char readme_msg[] =
2475         "tracing mini-HOWTO:\n\n"
2476         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2477         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2478         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2479         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2480         "nop\n"
2481         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2482         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2483         "sched_switch\n"
2484         "# cat /sys/kernel/debug/tracing/trace_options\n"
2485         "noprint-parent nosym-offset nosym-addr noverbose\n"
2486         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2487         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2488         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2489         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2490 ;
2491
2492 static ssize_t
2493 tracing_readme_read(struct file *filp, char __user *ubuf,
2494                        size_t cnt, loff_t *ppos)
2495 {
2496         return simple_read_from_buffer(ubuf, cnt, ppos,
2497                                         readme_msg, strlen(readme_msg));
2498 }
2499
2500 static const struct file_operations tracing_readme_fops = {
2501         .open           = tracing_open_generic,
2502         .read           = tracing_readme_read,
2503 };
2504
2505 static ssize_t
2506 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2507                                 size_t cnt, loff_t *ppos)
2508 {
2509         char *buf_comm;
2510         char *file_buf;
2511         char *buf;
2512         int len = 0;
2513         int pid;
2514         int i;
2515
2516         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2517         if (!file_buf)
2518                 return -ENOMEM;
2519
2520         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2521         if (!buf_comm) {
2522                 kfree(file_buf);
2523                 return -ENOMEM;
2524         }
2525
2526         buf = file_buf;
2527
2528         for (i = 0; i < SAVED_CMDLINES; i++) {
2529                 int r;
2530
2531                 pid = map_cmdline_to_pid[i];
2532                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2533                         continue;
2534
2535                 trace_find_cmdline(pid, buf_comm);
2536                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2537                 buf += r;
2538                 len += r;
2539         }
2540
2541         len = simple_read_from_buffer(ubuf, cnt, ppos,
2542                                       file_buf, len);
2543
2544         kfree(file_buf);
2545         kfree(buf_comm);
2546
2547         return len;
2548 }
2549
2550 static const struct file_operations tracing_saved_cmdlines_fops = {
2551     .open       = tracing_open_generic,
2552     .read       = tracing_saved_cmdlines_read,
2553 };
2554
2555 static ssize_t
2556 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2557                   size_t cnt, loff_t *ppos)
2558 {
2559         char buf[64];
2560         int r;
2561
2562         r = sprintf(buf, "%u\n", tracer_enabled);
2563         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2564 }
2565
2566 static ssize_t
2567 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2568                    size_t cnt, loff_t *ppos)
2569 {
2570         struct trace_array *tr = filp->private_data;
2571         char buf[64];
2572         unsigned long val;
2573         int ret;
2574
2575         if (cnt >= sizeof(buf))
2576                 return -EINVAL;
2577
2578         if (copy_from_user(&buf, ubuf, cnt))
2579                 return -EFAULT;
2580
2581         buf[cnt] = 0;
2582
2583         ret = strict_strtoul(buf, 10, &val);
2584         if (ret < 0)
2585                 return ret;
2586
2587         val = !!val;
2588
2589         mutex_lock(&trace_types_lock);
2590         if (tracer_enabled ^ val) {
2591                 if (val) {
2592                         tracer_enabled = 1;
2593                         if (current_trace->start)
2594                                 current_trace->start(tr);
2595                         tracing_start();
2596                 } else {
2597                         tracer_enabled = 0;
2598                         tracing_stop();
2599                         if (current_trace->stop)
2600                                 current_trace->stop(tr);
2601                 }
2602         }
2603         mutex_unlock(&trace_types_lock);
2604
2605         *ppos += cnt;
2606
2607         return cnt;
2608 }
2609
2610 static ssize_t
2611 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2612                        size_t cnt, loff_t *ppos)
2613 {
2614         char buf[MAX_TRACER_SIZE+2];
2615         int r;
2616
2617         mutex_lock(&trace_types_lock);
2618         if (current_trace)
2619                 r = sprintf(buf, "%s\n", current_trace->name);
2620         else
2621                 r = sprintf(buf, "\n");
2622         mutex_unlock(&trace_types_lock);
2623
2624         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2625 }
2626
2627 int tracer_init(struct tracer *t, struct trace_array *tr)
2628 {
2629         tracing_reset_online_cpus(tr);
2630         return t->init(tr);
2631 }
2632
2633 static int tracing_resize_ring_buffer(unsigned long size)
2634 {
2635         int ret;
2636
2637         /*
2638          * If kernel or user changes the size of the ring buffer
2639          * we use the size that was given, and we can forget about
2640          * expanding it later.
2641          */
2642         ring_buffer_expanded = 1;
2643
2644         ret = ring_buffer_resize(global_trace.buffer, size);
2645         if (ret < 0)
2646                 return ret;
2647
2648         ret = ring_buffer_resize(max_tr.buffer, size);
2649         if (ret < 0) {
2650                 int r;
2651
2652                 r = ring_buffer_resize(global_trace.buffer,
2653                                        global_trace.entries);
2654                 if (r < 0) {
2655                         /*
2656                          * AARGH! We are left with different
2657                          * size max buffer!!!!
2658                          * The max buffer is our "snapshot" buffer.
2659                          * When a tracer needs a snapshot (one of the
2660                          * latency tracers), it swaps the max buffer
2661                          * with the saved snap shot. We succeeded to
2662                          * update the size of the main buffer, but failed to
2663                          * update the size of the max buffer. But when we tried
2664                          * to reset the main buffer to the original size, we
2665                          * failed there too. This is very unlikely to
2666                          * happen, but if it does, warn and kill all
2667                          * tracing.
2668                          */
2669                         WARN_ON(1);
2670                         tracing_disabled = 1;
2671                 }
2672                 return ret;
2673         }
2674
2675         global_trace.entries = size;
2676
2677         return ret;
2678 }
2679
2680 /**
2681  * tracing_update_buffers - used by tracing facility to expand ring buffers
2682  *
2683  * To save on memory when the tracing is never used on a system with it
2684  * configured in. The ring buffers are set to a minimum size. But once
2685  * a user starts to use the tracing facility, then they need to grow
2686  * to their default size.
2687  *
2688  * This function is to be called when a tracer is about to be used.
2689  */
2690 int tracing_update_buffers(void)
2691 {
2692         int ret = 0;
2693
2694         mutex_lock(&trace_types_lock);
2695         if (!ring_buffer_expanded)
2696                 ret = tracing_resize_ring_buffer(trace_buf_size);
2697         mutex_unlock(&trace_types_lock);
2698
2699         return ret;
2700 }
2701
2702 struct trace_option_dentry;
2703
2704 static struct trace_option_dentry *
2705 create_trace_option_files(struct tracer *tracer);
2706
2707 static void
2708 destroy_trace_option_files(struct trace_option_dentry *topts);
2709
2710 static int tracing_set_tracer(const char *buf)
2711 {
2712         static struct trace_option_dentry *topts;
2713         struct trace_array *tr = &global_trace;
2714         struct tracer *t;
2715         int ret = 0;
2716
2717         mutex_lock(&trace_types_lock);
2718
2719         if (!ring_buffer_expanded) {
2720                 ret = tracing_resize_ring_buffer(trace_buf_size);
2721                 if (ret < 0)
2722                         goto out;
2723                 ret = 0;
2724         }
2725
2726         for (t = trace_types; t; t = t->next) {
2727                 if (strcmp(t->name, buf) == 0)
2728                         break;
2729         }
2730         if (!t) {
2731                 ret = -EINVAL;
2732                 goto out;
2733         }
2734         if (t == current_trace)
2735                 goto out;
2736
2737         trace_branch_disable();
2738         if (current_trace && current_trace->reset)
2739                 current_trace->reset(tr);
2740
2741         destroy_trace_option_files(topts);
2742
2743         current_trace = t;
2744
2745         topts = create_trace_option_files(current_trace);
2746
2747         if (t->init) {
2748                 ret = tracer_init(t, tr);
2749                 if (ret)
2750                         goto out;
2751         }
2752
2753         trace_branch_enable(tr);
2754  out:
2755         mutex_unlock(&trace_types_lock);
2756
2757         return ret;
2758 }
2759
2760 static ssize_t
2761 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2762                         size_t cnt, loff_t *ppos)
2763 {
2764         char buf[MAX_TRACER_SIZE+1];
2765         int i;
2766         size_t ret;
2767         int err;
2768
2769         ret = cnt;
2770
2771         if (cnt > MAX_TRACER_SIZE)
2772                 cnt = MAX_TRACER_SIZE;
2773
2774         if (copy_from_user(&buf, ubuf, cnt))
2775                 return -EFAULT;
2776
2777         buf[cnt] = 0;
2778
2779         /* strip ending whitespace. */
2780         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2781                 buf[i] = 0;
2782
2783         err = tracing_set_tracer(buf);
2784         if (err)
2785                 return err;
2786
2787         *ppos += ret;
2788
2789         return ret;
2790 }
2791
2792 static ssize_t
2793 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2794                      size_t cnt, loff_t *ppos)
2795 {
2796         unsigned long *ptr = filp->private_data;
2797         char buf[64];
2798         int r;
2799
2800         r = snprintf(buf, sizeof(buf), "%ld\n",
2801                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2802         if (r > sizeof(buf))
2803                 r = sizeof(buf);
2804         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2805 }
2806
2807 static ssize_t
2808 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2809                       size_t cnt, loff_t *ppos)
2810 {
2811         unsigned long *ptr = filp->private_data;
2812         char buf[64];
2813         unsigned long val;
2814         int ret;
2815
2816         if (cnt >= sizeof(buf))
2817                 return -EINVAL;
2818
2819         if (copy_from_user(&buf, ubuf, cnt))
2820                 return -EFAULT;
2821
2822         buf[cnt] = 0;
2823
2824         ret = strict_strtoul(buf, 10, &val);
2825         if (ret < 0)
2826                 return ret;
2827
2828         *ptr = val * 1000;
2829
2830         return cnt;
2831 }
2832
2833 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2834 {
2835         long cpu_file = (long) inode->i_private;
2836         struct trace_iterator *iter;
2837         int ret = 0;
2838
2839         if (tracing_disabled)
2840                 return -ENODEV;
2841
2842         mutex_lock(&trace_types_lock);
2843
2844         /* We only allow one reader per cpu */
2845         if (cpu_file == TRACE_PIPE_ALL_CPU) {
2846                 if (!cpumask_empty(tracing_reader_cpumask)) {
2847                         ret = -EBUSY;
2848                         goto out;
2849                 }
2850                 cpumask_setall(tracing_reader_cpumask);
2851         } else {
2852                 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2853                         cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2854                 else {
2855                         ret = -EBUSY;
2856                         goto out;
2857                 }
2858         }
2859
2860         /* create a buffer to store the information to pass to userspace */
2861         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2862         if (!iter) {
2863                 ret = -ENOMEM;
2864                 goto out;
2865         }
2866
2867         /*
2868          * We make a copy of the current tracer to avoid concurrent
2869          * changes on it while we are reading.
2870          */
2871         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2872         if (!iter->trace) {
2873                 ret = -ENOMEM;
2874                 goto fail;
2875         }
2876         if (current_trace)
2877                 *iter->trace = *current_trace;
2878
2879         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2880                 ret = -ENOMEM;
2881                 goto fail;
2882         }
2883
2884         /* trace pipe does not show start of buffer */
2885         cpumask_setall(iter->started);
2886
2887         if (trace_flags & TRACE_ITER_LATENCY_FMT)
2888                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2889
2890         iter->cpu_file = cpu_file;
2891         iter->tr = &global_trace;
2892         mutex_init(&iter->mutex);
2893         filp->private_data = iter;
2894
2895         if (iter->trace->pipe_open)
2896                 iter->trace->pipe_open(iter);
2897
2898 out:
2899         mutex_unlock(&trace_types_lock);
2900         return ret;
2901
2902 fail:
2903         kfree(iter->trace);
2904         kfree(iter);
2905         mutex_unlock(&trace_types_lock);
2906         return ret;
2907 }
2908
2909 static int tracing_release_pipe(struct inode *inode, struct file *file)
2910 {
2911         struct trace_iterator *iter = file->private_data;
2912
2913         mutex_lock(&trace_types_lock);
2914
2915         if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2916                 cpumask_clear(tracing_reader_cpumask);
2917         else
2918                 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2919
2920
2921         if (iter->trace->pipe_close)
2922                 iter->trace->pipe_close(iter);
2923
2924         mutex_unlock(&trace_types_lock);
2925
2926         free_cpumask_var(iter->started);
2927         mutex_destroy(&iter->mutex);
2928         kfree(iter->trace);
2929         kfree(iter);
2930
2931         return 0;
2932 }
2933
2934 static unsigned int
2935 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2936 {
2937         struct trace_iterator *iter = filp->private_data;
2938
2939         if (trace_flags & TRACE_ITER_BLOCK) {
2940                 /*
2941                  * Always select as readable when in blocking mode
2942                  */
2943                 return POLLIN | POLLRDNORM;
2944         } else {
2945                 if (!trace_empty(iter))
2946                         return POLLIN | POLLRDNORM;
2947                 poll_wait(filp, &trace_wait, poll_table);
2948                 if (!trace_empty(iter))
2949                         return POLLIN | POLLRDNORM;
2950
2951                 return 0;
2952         }
2953 }
2954
2955
2956 void default_wait_pipe(struct trace_iterator *iter)
2957 {
2958         DEFINE_WAIT(wait);
2959
2960         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2961
2962         if (trace_empty(iter))
2963                 schedule();
2964
2965         finish_wait(&trace_wait, &wait);
2966 }
2967
2968 /*
2969  * This is a make-shift waitqueue.
2970  * A tracer might use this callback on some rare cases:
2971  *
2972  *  1) the current tracer might hold the runqueue lock when it wakes up
2973  *     a reader, hence a deadlock (sched, function, and function graph tracers)
2974  *  2) the function tracers, trace all functions, we don't want
2975  *     the overhead of calling wake_up and friends
2976  *     (and tracing them too)
2977  *
2978  *     Anyway, this is really very primitive wakeup.
2979  */
2980 void poll_wait_pipe(struct trace_iterator *iter)
2981 {
2982         set_current_state(TASK_INTERRUPTIBLE);
2983         /* sleep for 100 msecs, and try again. */
2984         schedule_timeout(HZ / 10);
2985 }
2986
2987 /* Must be called with trace_types_lock mutex held. */
2988 static int tracing_wait_pipe(struct file *filp)
2989 {
2990         struct trace_iterator *iter = filp->private_data;
2991
2992         while (trace_empty(iter)) {
2993
2994                 if ((filp->f_flags & O_NONBLOCK)) {
2995                         return -EAGAIN;
2996                 }
2997
2998                 mutex_unlock(&iter->mutex);
2999
3000                 iter->trace->wait_pipe(iter);
3001
3002                 mutex_lock(&iter->mutex);
3003
3004                 if (signal_pending(current))
3005                         return -EINTR;
3006
3007                 /*
3008                  * We block until we read something and tracing is disabled.
3009                  * We still block if tracing is disabled, but we have never
3010                  * read anything. This allows a user to cat this file, and
3011                  * then enable tracing. But after we have read something,
3012                  * we give an EOF when tracing is again disabled.
3013                  *
3014                  * iter->pos will be 0 if we haven't read anything.
3015                  */
3016                 if (!tracer_enabled && iter->pos)
3017                         break;
3018         }
3019
3020         return 1;
3021 }
3022
3023 /*
3024  * Consumer reader.
3025  */
3026 static ssize_t
3027 tracing_read_pipe(struct file *filp, char __user *ubuf,
3028                   size_t cnt, loff_t *ppos)
3029 {
3030         struct trace_iterator *iter = filp->private_data;
3031         static struct tracer *old_tracer;
3032         ssize_t sret;
3033
3034         /* return any leftover data */
3035         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3036         if (sret != -EBUSY)
3037                 return sret;
3038
3039         trace_seq_init(&iter->seq);
3040
3041         /* copy the tracer to avoid using a global lock all around */
3042         mutex_lock(&trace_types_lock);
3043         if (unlikely(old_tracer != current_trace && current_trace)) {
3044                 old_tracer = current_trace;
3045                 *iter->trace = *current_trace;
3046         }
3047         mutex_unlock(&trace_types_lock);
3048
3049         /*
3050          * Avoid more than one consumer on a single file descriptor
3051          * This is just a matter of traces coherency, the ring buffer itself
3052          * is protected.
3053          */
3054         mutex_lock(&iter->mutex);
3055         if (iter->trace->read) {
3056                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3057                 if (sret)
3058                         goto out;
3059         }
3060
3061 waitagain:
3062         sret = tracing_wait_pipe(filp);
3063         if (sret <= 0)
3064                 goto out;
3065
3066         /* stop when tracing is finished */
3067         if (trace_empty(iter)) {
3068                 sret = 0;
3069                 goto out;
3070         }
3071
3072         if (cnt >= PAGE_SIZE)
3073                 cnt = PAGE_SIZE - 1;
3074
3075         /* reset all but tr, trace, and overruns */
3076         memset(&iter->seq, 0,
3077                sizeof(struct trace_iterator) -
3078                offsetof(struct trace_iterator, seq));
3079         iter->pos = -1;
3080
3081         trace_event_read_lock();
3082         while (find_next_entry_inc(iter) != NULL) {
3083                 enum print_line_t ret;
3084                 int len = iter->seq.len;
3085
3086                 ret = print_trace_line(iter);
3087                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3088                         /* don't print partial lines */
3089                         iter->seq.len = len;
3090                         break;
3091                 }
3092                 if (ret != TRACE_TYPE_NO_CONSUME)
3093                         trace_consume(iter);
3094
3095                 if (iter->seq.len >= cnt)
3096                         break;
3097         }
3098         trace_event_read_unlock();
3099
3100         /* Now copy what we have to the user */
3101         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3102         if (iter->seq.readpos >= iter->seq.len)
3103                 trace_seq_init(&iter->seq);
3104
3105         /*
3106          * If there was nothing to send to user, inspite of consuming trace
3107          * entries, go back to wait for more entries.
3108          */
3109         if (sret == -EBUSY)
3110                 goto waitagain;
3111
3112 out:
3113         mutex_unlock(&iter->mutex);
3114
3115         return sret;
3116 }
3117
3118 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3119                                      struct pipe_buffer *buf)
3120 {
3121         __free_page(buf->page);
3122 }
3123
3124 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3125                                      unsigned int idx)
3126 {
3127         __free_page(spd->pages[idx]);
3128 }
3129
3130 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3131         .can_merge              = 0,
3132         .map                    = generic_pipe_buf_map,
3133         .unmap                  = generic_pipe_buf_unmap,
3134         .confirm                = generic_pipe_buf_confirm,
3135         .release                = tracing_pipe_buf_release,
3136         .steal                  = generic_pipe_buf_steal,
3137         .get                    = generic_pipe_buf_get,
3138 };
3139
3140 static size_t
3141 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3142 {
3143         size_t count;
3144         int ret;
3145
3146         /* Seq buffer is page-sized, exactly what we need. */
3147         for (;;) {
3148                 count = iter->seq.len;
3149                 ret = print_trace_line(iter);
3150                 count = iter->seq.len - count;
3151                 if (rem < count) {
3152                         rem = 0;
3153                         iter->seq.len -= count;
3154                         break;
3155                 }
3156                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3157                         iter->seq.len -= count;
3158                         break;
3159                 }
3160
3161                 if (ret != TRACE_TYPE_NO_CONSUME)
3162                         trace_consume(iter);
3163                 rem -= count;
3164                 if (!find_next_entry_inc(iter)) {
3165                         rem = 0;
3166                         iter->ent = NULL;
3167                         break;
3168                 }
3169         }
3170
3171         return rem;
3172 }
3173
3174 static ssize_t tracing_splice_read_pipe(struct file *filp,
3175                                         loff_t *ppos,
3176                                         struct pipe_inode_info *pipe,
3177                                         size_t len,
3178                                         unsigned int flags)
3179 {
3180         struct page *pages[PIPE_BUFFERS];
3181         struct partial_page partial[PIPE_BUFFERS];
3182         struct trace_iterator *iter = filp->private_data;
3183         struct splice_pipe_desc spd = {
3184                 .pages          = pages,
3185                 .partial        = partial,
3186                 .nr_pages       = 0, /* This gets updated below. */
3187                 .flags          = flags,
3188                 .ops            = &tracing_pipe_buf_ops,
3189                 .spd_release    = tracing_spd_release_pipe,
3190         };
3191         static struct tracer *old_tracer;
3192         ssize_t ret;
3193         size_t rem;
3194         unsigned int i;
3195
3196         /* copy the tracer to avoid using a global lock all around */
3197         mutex_lock(&trace_types_lock);
3198         if (unlikely(old_tracer != current_trace && current_trace)) {
3199                 old_tracer = current_trace;
3200                 *iter->trace = *current_trace;
3201         }
3202         mutex_unlock(&trace_types_lock);
3203
3204         mutex_lock(&iter->mutex);
3205
3206         if (iter->trace->splice_read) {
3207                 ret = iter->trace->splice_read(iter, filp,
3208                                                ppos, pipe, len, flags);
3209                 if (ret)
3210                         goto out_err;
3211         }
3212
3213         ret = tracing_wait_pipe(filp);
3214         if (ret <= 0)
3215                 goto out_err;
3216
3217         if (!iter->ent && !find_next_entry_inc(iter)) {
3218                 ret = -EFAULT;
3219                 goto out_err;
3220         }
3221
3222         trace_event_read_lock();
3223
3224         /* Fill as many pages as possible. */
3225         for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3226                 pages[i] = alloc_page(GFP_KERNEL);
3227                 if (!pages[i])
3228                         break;
3229
3230                 rem = tracing_fill_pipe_page(rem, iter);
3231
3232                 /* Copy the data into the page, so we can start over. */
3233                 ret = trace_seq_to_buffer(&iter->seq,
3234                                           page_address(pages[i]),
3235                                           iter->seq.len);
3236                 if (ret < 0) {
3237                         __free_page(pages[i]);
3238                         break;
3239                 }
3240                 partial[i].offset = 0;
3241                 partial[i].len = iter->seq.len;
3242
3243                 trace_seq_init(&iter->seq);
3244         }
3245
3246         trace_event_read_unlock();
3247         mutex_unlock(&iter->mutex);
3248
3249         spd.nr_pages = i;
3250
3251         return splice_to_pipe(pipe, &spd);
3252
3253 out_err:
3254         mutex_unlock(&iter->mutex);
3255
3256         return ret;
3257 }
3258
3259 static ssize_t
3260 tracing_entries_read(struct file *filp, char __user *ubuf,
3261                      size_t cnt, loff_t *ppos)
3262 {
3263         struct trace_array *tr = filp->private_data;
3264         char buf[96];
3265         int r;
3266
3267         mutex_lock(&trace_types_lock);
3268         if (!ring_buffer_expanded)
3269                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3270                             tr->entries >> 10,
3271                             trace_buf_size >> 10);
3272         else
3273                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3274         mutex_unlock(&trace_types_lock);
3275
3276         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3277 }
3278
3279 static ssize_t
3280 tracing_entries_write(struct file *filp, const char __user *ubuf,
3281                       size_t cnt, loff_t *ppos)
3282 {
3283         unsigned long val;
3284         char buf[64];
3285         int ret, cpu;
3286
3287         if (cnt >= sizeof(buf))
3288                 return -EINVAL;
3289
3290         if (copy_from_user(&buf, ubuf, cnt))
3291                 return -EFAULT;
3292
3293         buf[cnt] = 0;
3294
3295         ret = strict_strtoul(buf, 10, &val);
3296         if (ret < 0)
3297                 return ret;
3298
3299         /* must have at least 1 entry */
3300         if (!val)
3301                 return -EINVAL;
3302
3303         mutex_lock(&trace_types_lock);
3304
3305         tracing_stop();
3306
3307         /* disable all cpu buffers */
3308         for_each_tracing_cpu(cpu) {
3309                 if (global_trace.data[cpu])
3310                         atomic_inc(&global_trace.data[cpu]->disabled);
3311                 if (max_tr.data[cpu])
3312                         atomic_inc(&max_tr.data[cpu]->disabled);
3313         }
3314
3315         /* value is in KB */
3316         val <<= 10;
3317
3318         if (val != global_trace.entries) {
3319                 ret = tracing_resize_ring_buffer(val);
3320                 if (ret < 0) {
3321                         cnt = ret;
3322                         goto out;
3323                 }
3324         }
3325
3326         *ppos += cnt;
3327
3328         /* If check pages failed, return ENOMEM */
3329         if (tracing_disabled)
3330                 cnt = -ENOMEM;
3331  out:
3332         for_each_tracing_cpu(cpu) {
3333                 if (global_trace.data[cpu])
3334                         atomic_dec(&global_trace.data[cpu]->disabled);
3335                 if (max_tr.data[cpu])
3336                         atomic_dec(&max_tr.data[cpu]->disabled);
3337         }
3338
3339         tracing_start();
3340         max_tr.entries = global_trace.entries;
3341         mutex_unlock(&trace_types_lock);
3342
3343         return cnt;
3344 }
3345
3346 static int mark_printk(const char *fmt, ...)
3347 {
3348         int ret;
3349         va_list args;
3350         va_start(args, fmt);
3351         ret = trace_vprintk(0, fmt, args);
3352         va_end(args);
3353         return ret;
3354 }
3355
3356 static ssize_t
3357 tracing_mark_write(struct file *filp, const char __user *ubuf,
3358                                         size_t cnt, loff_t *fpos)
3359 {
3360         char *buf;
3361
3362         if (tracing_disabled)
3363                 return -EINVAL;
3364
3365         if (cnt > TRACE_BUF_SIZE)
3366                 cnt = TRACE_BUF_SIZE;
3367
3368         buf = kmalloc(cnt + 2, GFP_KERNEL);
3369         if (buf == NULL)
3370                 return -ENOMEM;
3371
3372         if (copy_from_user(buf, ubuf, cnt)) {
3373                 kfree(buf);
3374                 return -EFAULT;
3375         }
3376         if (buf[cnt-1] != '\n') {
3377                 buf[cnt] = '\n';
3378                 buf[cnt+1] = '\0';
3379         } else
3380                 buf[cnt] = '\0';
3381
3382         cnt = mark_printk("%s", buf);
3383         kfree(buf);
3384         *fpos += cnt;
3385
3386         return cnt;
3387 }
3388
3389 static int tracing_clock_show(struct seq_file *m, void *v)
3390 {
3391         int i;
3392
3393         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3394                 seq_printf(m,
3395                         "%s%s%s%s", i ? " " : "",
3396                         i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3397                         i == trace_clock_id ? "]" : "");
3398         seq_putc(m, '\n');
3399
3400         return 0;
3401 }
3402
3403 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3404                                    size_t cnt, loff_t *fpos)
3405 {
3406         char buf[64];
3407         const char *clockstr;
3408         int i;
3409
3410         if (cnt >= sizeof(buf))
3411                 return -EINVAL;
3412
3413         if (copy_from_user(&buf, ubuf, cnt))
3414                 return -EFAULT;
3415
3416         buf[cnt] = 0;
3417
3418         clockstr = strstrip(buf);
3419
3420         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3421                 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3422                         break;
3423         }
3424         if (i == ARRAY_SIZE(trace_clocks))
3425                 return -EINVAL;
3426
3427         trace_clock_id = i;
3428
3429         mutex_lock(&trace_types_lock);
3430
3431         ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3432         if (max_tr.buffer)
3433                 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3434
3435         mutex_unlock(&trace_types_lock);
3436
3437         *fpos += cnt;
3438
3439         return cnt;
3440 }
3441
3442 static int tracing_clock_open(struct inode *inode, struct file *file)
3443 {
3444         if (tracing_disabled)
3445                 return -ENODEV;
3446         return single_open(file, tracing_clock_show, NULL);
3447 }
3448
3449 static const struct file_operations tracing_max_lat_fops = {
3450         .open           = tracing_open_generic,
3451         .read           = tracing_max_lat_read,
3452         .write          = tracing_max_lat_write,
3453 };
3454
3455 static const struct file_operations tracing_ctrl_fops = {
3456         .open           = tracing_open_generic,
3457         .read           = tracing_ctrl_read,
3458         .write          = tracing_ctrl_write,
3459 };
3460
3461 static const struct file_operations set_tracer_fops = {
3462         .open           = tracing_open_generic,
3463         .read           = tracing_set_trace_read,
3464         .write          = tracing_set_trace_write,
3465 };
3466
3467 static const struct file_operations tracing_pipe_fops = {
3468         .open           = tracing_open_pipe,
3469         .poll           = tracing_poll_pipe,
3470         .read           = tracing_read_pipe,
3471         .splice_read    = tracing_splice_read_pipe,
3472         .release        = tracing_release_pipe,
3473 };
3474
3475 static const struct file_operations tracing_entries_fops = {
3476         .open           = tracing_open_generic,
3477         .read           = tracing_entries_read,
3478         .write          = tracing_entries_write,
3479 };
3480
3481 static const struct file_operations tracing_mark_fops = {
3482         .open           = tracing_open_generic,
3483         .write          = tracing_mark_write,
3484 };
3485
3486 static const struct file_operations trace_clock_fops = {
3487         .open           = tracing_clock_open,
3488         .read           = seq_read,
3489         .llseek         = seq_lseek,
3490         .release        = single_release,
3491         .write          = tracing_clock_write,
3492 };
3493
3494 struct ftrace_buffer_info {
3495         struct trace_array      *tr;
3496         void                    *spare;
3497         int                     cpu;
3498         unsigned int            read;
3499 };
3500
3501 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3502 {
3503         int cpu = (int)(long)inode->i_private;
3504         struct ftrace_buffer_info *info;
3505
3506         if (tracing_disabled)
3507                 return -ENODEV;
3508
3509         info = kzalloc(sizeof(*info), GFP_KERNEL);
3510         if (!info)
3511                 return -ENOMEM;
3512
3513         info->tr        = &global_trace;
3514         info->cpu       = cpu;
3515         info->spare     = NULL;
3516         /* Force reading ring buffer for first read */
3517         info->read      = (unsigned int)-1;
3518
3519         filp->private_data = info;
3520
3521         return nonseekable_open(inode, filp);
3522 }
3523
3524 static ssize_t
3525 tracing_buffers_read(struct file *filp, char __user *ubuf,
3526                      size_t count, loff_t *ppos)
3527 {
3528         struct ftrace_buffer_info *info = filp->private_data;
3529         unsigned int pos;
3530         ssize_t ret;
3531         size_t size;
3532
3533         if (!count)
3534                 return 0;
3535
3536         if (!info->spare)
3537                 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3538         if (!info->spare)
3539                 return -ENOMEM;
3540
3541         /* Do we have previous read data to read? */
3542         if (info->read < PAGE_SIZE)
3543                 goto read;
3544
3545         info->read = 0;
3546
3547         ret = ring_buffer_read_page(info->tr->buffer,
3548                                     &info->spare,
3549                                     count,
3550                                     info->cpu, 0);
3551         if (ret < 0)
3552                 return 0;
3553
3554         pos = ring_buffer_page_len(info->spare);
3555
3556         if (pos < PAGE_SIZE)
3557                 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3558
3559 read:
3560         size = PAGE_SIZE - info->read;
3561         if (size > count)
3562                 size = count;
3563
3564         ret = copy_to_user(ubuf, info->spare + info->read, size);
3565         if (ret == size)
3566                 return -EFAULT;
3567         size -= ret;
3568
3569         *ppos += size;
3570         info->read += size;
3571
3572         return size;
3573 }
3574
3575 static int tracing_buffers_release(struct inode *inode, struct file *file)
3576 {
3577         struct ftrace_buffer_info *info = file->private_data;
3578
3579         if (info->spare)
3580                 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3581         kfree(info);
3582
3583         return 0;
3584 }
3585
3586 struct buffer_ref {
3587         struct ring_buffer      *buffer;
3588         void                    *page;
3589         int                     ref;
3590 };
3591
3592 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3593                                     struct pipe_buffer *buf)
3594 {
3595         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3596
3597         if (--ref->ref)
3598                 return;
3599
3600         ring_buffer_free_read_page(ref->buffer, ref->page);
3601         kfree(ref);
3602         buf->private = 0;
3603 }
3604
3605 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3606                                  struct pipe_buffer *buf)
3607 {
3608         return 1;
3609 }
3610
3611 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3612                                 struct pipe_buffer *buf)
3613 {
3614         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3615
3616         ref->ref++;
3617 }
3618
3619 /* Pipe buffer operations for a buffer. */
3620 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3621         .can_merge              = 0,
3622         .map                    = generic_pipe_buf_map,
3623         .unmap                  = generic_pipe_buf_unmap,
3624         .confirm                = generic_pipe_buf_confirm,
3625         .release                = buffer_pipe_buf_release,
3626         .steal                  = buffer_pipe_buf_steal,
3627         .get                    = buffer_pipe_buf_get,
3628 };
3629
3630 /*
3631  * Callback from splice_to_pipe(), if we need to release some pages
3632  * at the end of the spd in case we error'ed out in filling the pipe.
3633  */
3634 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3635 {
3636         struct buffer_ref *ref =
3637                 (struct buffer_ref *)spd->partial[i].private;
3638
3639         if (--ref->ref)
3640                 return;
3641
3642         ring_buffer_free_read_page(ref->buffer, ref->page);
3643         kfree(ref);
3644         spd->partial[i].private = 0;
3645 }
3646
3647 static ssize_t
3648 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3649                             struct pipe_inode_info *pipe, size_t len,
3650                             unsigned int flags)
3651 {
3652         struct ftrace_buffer_info *info = file->private_data;
3653         struct partial_page partial[PIPE_BUFFERS];
3654         struct page *pages[PIPE_BUFFERS];
3655         struct splice_pipe_desc spd = {
3656                 .pages          = pages,
3657                 .partial        = partial,
3658                 .flags          = flags,
3659                 .ops            = &buffer_pipe_buf_ops,
3660                 .spd_release    = buffer_spd_release,
3661         };
3662         struct buffer_ref *ref;
3663         int entries, size, i;
3664         size_t ret;
3665
3666         if (*ppos & (PAGE_SIZE - 1)) {
3667                 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3668                 return -EINVAL;
3669         }
3670
3671         if (len & (PAGE_SIZE - 1)) {
3672                 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3673                 if (len < PAGE_SIZE)
3674                         return -EINVAL;
3675                 len &= PAGE_MASK;
3676         }
3677
3678         entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3679
3680         for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3681                 struct page *page;
3682                 int r;
3683
3684                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3685                 if (!ref)
3686                         break;
3687
3688                 ref->ref = 1;
3689                 ref->buffer = info->tr->buffer;
3690                 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3691                 if (!ref->page) {
3692                         kfree(ref);
3693                         break;
3694                 }
3695
3696                 r = ring_buffer_read_page(ref->buffer, &ref->page,
3697                                           len, info->cpu, 1);
3698                 if (r < 0) {
3699                         ring_buffer_free_read_page(ref->buffer,
3700                                                    ref->page);
3701                         kfree(ref);
3702                         break;
3703                 }
3704
3705                 /*
3706                  * zero out any left over data, this is going to
3707                  * user land.
3708                  */
3709                 size = ring_buffer_page_len(ref->page);
3710                 if (size < PAGE_SIZE)
3711                         memset(ref->page + size, 0, PAGE_SIZE - size);
3712
3713                 page = virt_to_page(ref->page);
3714
3715                 spd.pages[i] = page;
3716                 spd.partial[i].len = PAGE_SIZE;
3717                 spd.partial[i].offset = 0;
3718                 spd.partial[i].private = (unsigned long)ref;
3719                 spd.nr_pages++;
3720                 *ppos += PAGE_SIZE;
3721
3722                 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3723         }
3724
3725         spd.nr_pages = i;
3726
3727         /* did we read anything? */
3728         if (!spd.nr_pages) {
3729                 if (flags & SPLICE_F_NONBLOCK)
3730                         ret = -EAGAIN;
3731                 else
3732                         ret = 0;
3733                 /* TODO: block */
3734                 return ret;
3735         }
3736
3737         ret = splice_to_pipe(pipe, &spd);
3738
3739         return ret;
3740 }
3741
3742 static const struct file_operations tracing_buffers_fops = {
3743         .open           = tracing_buffers_open,
3744         .read           = tracing_buffers_read,
3745         .release        = tracing_buffers_release,
3746         .splice_read    = tracing_buffers_splice_read,
3747         .llseek         = no_llseek,
3748 };
3749
3750 static ssize_t
3751 tracing_stats_read(struct file *filp, char __user *ubuf,
3752                    size_t count, loff_t *ppos)
3753 {
3754         unsigned long cpu = (unsigned long)filp->private_data;
3755         struct trace_array *tr = &global_trace;
3756         struct trace_seq *s;
3757         unsigned long cnt;
3758
3759         s = kmalloc(sizeof(*s), GFP_KERNEL);
3760         if (!s)
3761                 return -ENOMEM;
3762
3763         trace_seq_init(s);
3764
3765         cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3766         trace_seq_printf(s, "entries: %ld\n", cnt);
3767
3768         cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3769         trace_seq_printf(s, "overrun: %ld\n", cnt);
3770
3771         cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3772         trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3773
3774         count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3775
3776         kfree(s);
3777
3778         return count;
3779 }
3780
3781 static const struct file_operations tracing_stats_fops = {
3782         .open           = tracing_open_generic,
3783         .read           = tracing_stats_read,
3784 };
3785
3786 #ifdef CONFIG_DYNAMIC_FTRACE
3787
3788 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3789 {
3790         return 0;
3791 }
3792
3793 static ssize_t
3794 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3795                   size_t cnt, loff_t *ppos)
3796 {
3797         static char ftrace_dyn_info_buffer[1024];
3798         static DEFINE_MUTEX(dyn_info_mutex);
3799         unsigned long *p = filp->private_data;
3800         char *buf = ftrace_dyn_info_buffer;
3801         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3802         int r;
3803
3804         mutex_lock(&dyn_info_mutex);
3805         r = sprintf(buf, "%ld ", *p);
3806
3807         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3808         buf[r++] = '\n';
3809
3810         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3811
3812         mutex_unlock(&dyn_info_mutex);
3813
3814         return r;
3815 }
3816
3817 static const struct file_operations tracing_dyn_info_fops = {
3818         .open           = tracing_open_generic,
3819         .read           = tracing_read_dyn_info,
3820 };
3821 #endif
3822
3823 static struct dentry *d_tracer;
3824
3825 struct dentry *tracing_init_dentry(void)
3826 {
3827         static int once;
3828
3829         if (d_tracer)
3830                 return d_tracer;
3831
3832         if (!debugfs_initialized())
3833                 return NULL;
3834
3835         d_tracer = debugfs_create_dir("tracing", NULL);
3836
3837         if (!d_tracer && !once) {
3838                 once = 1;
3839                 pr_warning("Could not create debugfs directory 'tracing'\n");
3840                 return NULL;
3841         }
3842
3843         return d_tracer;
3844 }
3845
3846 static struct dentry *d_percpu;
3847
3848 struct dentry *tracing_dentry_percpu(void)
3849 {
3850         static int once;
3851         struct dentry *d_tracer;
3852
3853         if (d_percpu)
3854                 return d_percpu;
3855
3856         d_tracer = tracing_init_dentry();
3857
3858         if (!d_tracer)
3859                 return NULL;
3860
3861         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3862
3863         if (!d_percpu && !once) {
3864                 once = 1;
3865                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3866                 return NULL;
3867         }
3868
3869         return d_percpu;
3870 }
3871
3872 static void tracing_init_debugfs_percpu(long cpu)
3873 {
3874         struct dentry *d_percpu = tracing_dentry_percpu();
3875         struct dentry *d_cpu;
3876         /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3877         char cpu_dir[7];
3878
3879         if (cpu > 999 || cpu < 0)
3880                 return;
3881
3882         sprintf(cpu_dir, "cpu%ld", cpu);
3883         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3884         if (!d_cpu) {
3885                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3886                 return;
3887         }
3888
3889         /* per cpu trace_pipe */
3890         trace_create_file("trace_pipe", 0444, d_cpu,
3891                         (void *) cpu, &tracing_pipe_fops);
3892
3893         /* per cpu trace */
3894         trace_create_file("trace", 0644, d_cpu,
3895                         (void *) cpu, &tracing_fops);
3896
3897         trace_create_file("trace_pipe_raw", 0444, d_cpu,
3898                         (void *) cpu, &tracing_buffers_fops);
3899
3900         trace_create_file("stats", 0444, d_cpu,
3901                         (void *) cpu, &tracing_stats_fops);
3902 }
3903
3904 #ifdef CONFIG_FTRACE_SELFTEST
3905 /* Let selftest have access to static functions in this file */
3906 #include "trace_selftest.c"
3907 #endif
3908
3909 struct trace_option_dentry {
3910         struct tracer_opt               *opt;
3911         struct tracer_flags             *flags;
3912         struct dentry                   *entry;
3913 };
3914
3915 static ssize_t
3916 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3917                         loff_t *ppos)
3918 {
3919         struct trace_option_dentry *topt = filp->private_data;
3920         char *buf;
3921
3922         if (topt->flags->val & topt->opt->bit)
3923                 buf = "1\n";
3924         else
3925                 buf = "0\n";
3926
3927         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3928 }
3929
3930 static ssize_t
3931 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3932                          loff_t *ppos)
3933 {
3934         struct trace_option_dentry *topt = filp->private_data;
3935         unsigned long val;
3936         char buf[64];
3937         int ret;
3938
3939         if (cnt >= sizeof(buf))
3940                 return -EINVAL;
3941
3942         if (copy_from_user(&buf, ubuf, cnt))
3943                 return -EFAULT;
3944
3945         buf[cnt] = 0;
3946
3947         ret = strict_strtoul(buf, 10, &val);
3948         if (ret < 0)
3949                 return ret;
3950
3951         if (val != 0 && val != 1)
3952                 return -EINVAL;
3953
3954         if (!!(topt->flags->val & topt->opt->bit) != val) {
3955                 mutex_lock(&trace_types_lock);
3956                 ret = __set_tracer_option(current_trace, topt->flags,
3957                                           topt->opt, !val);
3958                 mutex_unlock(&trace_types_lock);
3959                 if (ret)
3960                         return ret;
3961         }
3962
3963         *ppos += cnt;
3964
3965         return cnt;
3966 }
3967
3968
3969 static const struct file_operations trace_options_fops = {
3970         .open = tracing_open_generic,
3971         .read = trace_options_read,
3972         .write = trace_options_write,
3973 };
3974
3975 static ssize_t
3976 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3977                         loff_t *ppos)
3978 {
3979         long index = (long)filp->private_data;
3980         char *buf;
3981
3982         if (trace_flags & (1 << index))
3983                 buf = "1\n";
3984         else
3985                 buf = "0\n";
3986
3987         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3988 }
3989
3990 static ssize_t
3991 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3992                          loff_t *ppos)
3993 {
3994         long index = (long)filp->private_data;
3995         char buf[64];
3996         unsigned long val;
3997         int ret;
3998
3999         if (cnt >= sizeof(buf))
4000                 return -EINVAL;
4001
4002         if (copy_from_user(&buf, ubuf, cnt))
4003                 return -EFAULT;
4004
4005         buf[cnt] = 0;
4006
4007         ret = strict_strtoul(buf, 10, &val);
4008         if (ret < 0)
4009                 return ret;
4010
4011         if (val != 0 && val != 1)
4012                 return -EINVAL;
4013         set_tracer_flags(1 << index, val);
4014
4015         *ppos += cnt;
4016
4017         return cnt;
4018 }
4019
4020 static const struct file_operations trace_options_core_fops = {
4021         .open = tracing_open_generic,
4022         .read = trace_options_core_read,
4023         .write = trace_options_core_write,
4024 };
4025
4026 struct dentry *trace_create_file(const char *name,
4027                                  mode_t mode,
4028                                  struct dentry *parent,
4029                                  void *data,
4030                                  const struct file_operations *fops)
4031 {
4032         struct dentry *ret;
4033
4034         ret = debugfs_create_file(name, mode, parent, data, fops);
4035         if (!ret)
4036                 pr_warning("Could not create debugfs '%s' entry\n", name);
4037
4038         return ret;
4039 }
4040
4041
4042 static struct dentry *trace_options_init_dentry(void)
4043 {
4044         struct dentry *d_tracer;
4045         static struct dentry *t_options;
4046
4047         if (t_options)
4048                 return t_options;
4049
4050         d_tracer = tracing_init_dentry();
4051         if (!d_tracer)
4052                 return NULL;
4053
4054         t_options = debugfs_create_dir("options", d_tracer);
4055         if (!t_options) {
4056                 pr_warning("Could not create debugfs directory 'options'\n");
4057                 return NULL;
4058         }
4059
4060         return t_options;
4061 }
4062
4063 static void
4064 create_trace_option_file(struct trace_option_dentry *topt,
4065                          struct tracer_flags *flags,
4066                          struct tracer_opt *opt)
4067 {
4068         struct dentry *t_options;
4069
4070         t_options = trace_options_init_dentry();
4071         if (!t_options)
4072                 return;
4073
4074         topt->flags = flags;
4075         topt->opt = opt;
4076
4077         topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4078                                     &trace_options_fops);
4079
4080 }
4081
4082 static struct trace_option_dentry *
4083 create_trace_option_files(struct tracer *tracer)
4084 {
4085         struct trace_option_dentry *topts;
4086         struct tracer_flags *flags;
4087         struct tracer_opt *opts;
4088         int cnt;
4089
4090         if (!tracer)
4091                 return NULL;
4092
4093         flags = tracer->flags;
4094
4095         if (!flags || !flags->opts)
4096                 return NULL;
4097
4098         opts = flags->opts;
4099
4100         for (cnt = 0; opts[cnt].name; cnt++)
4101                 ;
4102
4103         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4104         if (!topts)
4105                 return NULL;
4106
4107         for (cnt = 0; opts[cnt].name; cnt++)
4108                 create_trace_option_file(&topts[cnt], flags,
4109                                          &opts[cnt]);
4110
4111         return topts;
4112 }
4113
4114 static void
4115 destroy_trace_option_files(struct trace_option_dentry *topts)
4116 {
4117         int cnt;
4118
4119         if (!topts)
4120                 return;
4121
4122         for (cnt = 0; topts[cnt].opt; cnt++) {
4123                 if (topts[cnt].entry)
4124                         debugfs_remove(topts[cnt].entry);
4125         }
4126
4127         kfree(topts);
4128 }
4129
4130 static struct dentry *
4131 create_trace_option_core_file(const char *option, long index)
4132 {
4133         struct dentry *t_options;
4134
4135         t_options = trace_options_init_dentry();
4136         if (!t_options)
4137                 return NULL;
4138
4139         return trace_create_file(option, 0644, t_options, (void *)index,
4140                                     &trace_options_core_fops);
4141 }
4142
4143 static __init void create_trace_options_dir(void)
4144 {
4145         struct dentry *t_options;
4146         int i;
4147
4148         t_options = trace_options_init_dentry();
4149         if (!t_options)
4150                 return;
4151
4152         for (i = 0; trace_options[i]; i++)
4153                 create_trace_option_core_file(trace_options[i], i);
4154 }
4155
4156 static __init int tracer_init_debugfs(void)
4157 {
4158         struct dentry *d_tracer;
4159         int cpu;
4160
4161         d_tracer = tracing_init_dentry();
4162
4163         trace_create_file("tracing_enabled", 0644, d_tracer,
4164                         &global_trace, &tracing_ctrl_fops);
4165
4166         trace_create_file("trace_options", 0644, d_tracer,
4167                         NULL, &tracing_iter_fops);
4168
4169         trace_create_file("tracing_cpumask", 0644, d_tracer,
4170                         NULL, &tracing_cpumask_fops);
4171
4172         trace_create_file("trace", 0644, d_tracer,
4173                         (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4174
4175         trace_create_file("available_tracers", 0444, d_tracer,
4176                         &global_trace, &show_traces_fops);
4177
4178         trace_create_file("current_tracer", 0644, d_tracer,
4179                         &global_trace, &set_tracer_fops);
4180
4181 #ifdef CONFIG_TRACER_MAX_TRACE
4182         trace_create_file("tracing_max_latency", 0644, d_tracer,
4183                         &tracing_max_latency, &tracing_max_lat_fops);
4184
4185         trace_create_file("tracing_thresh", 0644, d_tracer,
4186                         &tracing_thresh, &tracing_max_lat_fops);
4187 #endif
4188
4189         trace_create_file("README", 0444, d_tracer,
4190                         NULL, &tracing_readme_fops);
4191
4192         trace_create_file("trace_pipe", 0444, d_tracer,
4193                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4194
4195         trace_create_file("buffer_size_kb", 0644, d_tracer,
4196                         &global_trace, &tracing_entries_fops);
4197
4198         trace_create_file("trace_marker", 0220, d_tracer,
4199                         NULL, &tracing_mark_fops);
4200
4201         trace_create_file("saved_cmdlines", 0444, d_tracer,
4202                         NULL, &tracing_saved_cmdlines_fops);
4203
4204         trace_create_file("trace_clock", 0644, d_tracer, NULL,
4205                           &trace_clock_fops);
4206
4207 #ifdef CONFIG_DYNAMIC_FTRACE
4208         trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4209                         &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4210 #endif
4211 #ifdef CONFIG_SYSPROF_TRACER
4212         init_tracer_sysprof_debugfs(d_tracer);
4213 #endif
4214
4215         create_trace_options_dir();
4216
4217         for_each_tracing_cpu(cpu)
4218                 tracing_init_debugfs_percpu(cpu);
4219
4220         return 0;
4221 }
4222
4223 static int trace_panic_handler(struct notifier_block *this,
4224                                unsigned long event, void *unused)
4225 {
4226         if (ftrace_dump_on_oops)
4227                 ftrace_dump();
4228         return NOTIFY_OK;
4229 }
4230
4231 static struct notifier_block trace_panic_notifier = {
4232         .notifier_call  = trace_panic_handler,
4233         .next           = NULL,
4234         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
4235 };
4236
4237 static int trace_die_handler(struct notifier_block *self,
4238                              unsigned long val,
4239                              void *data)
4240 {
4241         switch (val) {
4242         case DIE_OOPS:
4243                 if (ftrace_dump_on_oops)
4244                         ftrace_dump();
4245                 break;
4246         default:
4247                 break;
4248         }
4249         return NOTIFY_OK;
4250 }
4251
4252 static struct notifier_block trace_die_notifier = {
4253         .notifier_call = trace_die_handler,
4254         .priority = 200
4255 };
4256
4257 /*
4258  * printk is set to max of 1024, we really don't need it that big.
4259  * Nothing should be printing 1000 characters anyway.
4260  */
4261 #define TRACE_MAX_PRINT         1000
4262
4263 /*
4264  * Define here KERN_TRACE so that we have one place to modify
4265  * it if we decide to change what log level the ftrace dump
4266  * should be at.
4267  */
4268 #define KERN_TRACE              KERN_EMERG
4269
4270 static void
4271 trace_printk_seq(struct trace_seq *s)
4272 {
4273         /* Probably should print a warning here. */
4274         if (s->len >= 1000)
4275                 s->len = 1000;
4276
4277         /* should be zero ended, but we are paranoid. */
4278         s->buffer[s->len] = 0;
4279
4280         printk(KERN_TRACE "%s", s->buffer);
4281
4282         trace_seq_init(s);
4283 }
4284
4285 static void __ftrace_dump(bool disable_tracing)
4286 {
4287         static arch_spinlock_t ftrace_dump_lock =
4288                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4289         /* use static because iter can be a bit big for the stack */
4290         static struct trace_iterator iter;
4291         unsigned int old_userobj;
4292         static int dump_ran;
4293         unsigned long flags;
4294         int cnt = 0, cpu;
4295
4296         /* only one dump */
4297         local_irq_save(flags);
4298         arch_spin_lock(&ftrace_dump_lock);
4299         if (dump_ran)
4300                 goto out;
4301
4302         dump_ran = 1;
4303
4304         tracing_off();
4305
4306         if (disable_tracing)
4307                 ftrace_kill();
4308
4309         for_each_tracing_cpu(cpu) {
4310                 atomic_inc(&global_trace.data[cpu]->disabled);
4311         }
4312
4313         old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4314
4315         /* don't look at user memory in panic mode */
4316         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4317
4318         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4319
4320         /* Simulate the iterator */
4321         iter.tr = &global_trace;
4322         iter.trace = current_trace;
4323         iter.cpu_file = TRACE_PIPE_ALL_CPU;
4324
4325         /*
4326          * We need to stop all tracing on all CPUS to read the
4327          * the next buffer. This is a bit expensive, but is
4328          * not done often. We fill all what we can read,
4329          * and then release the locks again.
4330          */
4331
4332         while (!trace_empty(&iter)) {
4333
4334                 if (!cnt)
4335                         printk(KERN_TRACE "---------------------------------\n");
4336
4337                 cnt++;
4338
4339                 /* reset all but tr, trace, and overruns */
4340                 memset(&iter.seq, 0,
4341                        sizeof(struct trace_iterator) -
4342                        offsetof(struct trace_iterator, seq));
4343                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4344                 iter.pos = -1;
4345
4346                 if (find_next_entry_inc(&iter) != NULL) {
4347                         int ret;
4348
4349                         ret = print_trace_line(&iter);
4350                         if (ret != TRACE_TYPE_NO_CONSUME)
4351                                 trace_consume(&iter);
4352                 }
4353
4354                 trace_printk_seq(&iter.seq);
4355         }
4356
4357         if (!cnt)
4358                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4359         else
4360                 printk(KERN_TRACE "---------------------------------\n");
4361
4362         /* Re-enable tracing if requested */
4363         if (!disable_tracing) {
4364                 trace_flags |= old_userobj;
4365
4366                 for_each_tracing_cpu(cpu) {
4367                         atomic_dec(&global_trace.data[cpu]->disabled);
4368                 }
4369                 tracing_on();
4370         }
4371
4372  out:
4373         arch_spin_unlock(&ftrace_dump_lock);
4374         local_irq_restore(flags);
4375 }
4376
4377 /* By default: disable tracing after the dump */
4378 void ftrace_dump(void)
4379 {
4380         __ftrace_dump(true);
4381 }
4382
4383 __init static int tracer_alloc_buffers(void)
4384 {
4385         int ring_buf_size;
4386         int i;
4387         int ret = -ENOMEM;
4388
4389         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4390                 goto out;
4391
4392         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4393                 goto out_free_buffer_mask;
4394
4395         if (!zalloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4396                 goto out_free_tracing_cpumask;
4397
4398         /* To save memory, keep the ring buffer size to its minimum */
4399         if (ring_buffer_expanded)
4400                 ring_buf_size = trace_buf_size;
4401         else
4402                 ring_buf_size = 1;
4403
4404         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4405         cpumask_copy(tracing_cpumask, cpu_all_mask);
4406
4407         /* TODO: make the number of buffers hot pluggable with CPUS */
4408         global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4409                                                    TRACE_BUFFER_FLAGS);
4410         if (!global_trace.buffer) {
4411                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4412                 WARN_ON(1);
4413                 goto out_free_cpumask;
4414         }
4415         global_trace.entries = ring_buffer_size(global_trace.buffer);
4416
4417
4418 #ifdef CONFIG_TRACER_MAX_TRACE
4419         max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4420                                              TRACE_BUFFER_FLAGS);
4421         if (!max_tr.buffer) {
4422                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4423                 WARN_ON(1);
4424                 ring_buffer_free(global_trace.buffer);
4425                 goto out_free_cpumask;
4426         }
4427         max_tr.entries = ring_buffer_size(max_tr.buffer);
4428         WARN_ON(max_tr.entries != global_trace.entries);
4429 #endif
4430
4431         /* Allocate the first page for all buffers */
4432         for_each_tracing_cpu(i) {
4433                 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4434                 max_tr.data[i] = &per_cpu(max_tr_data, i);
4435         }
4436
4437         trace_init_cmdlines();
4438
4439         register_tracer(&nop_trace);
4440         current_trace = &nop_trace;
4441 #ifdef CONFIG_BOOT_TRACER
4442         register_tracer(&boot_tracer);
4443 #endif
4444         /* All seems OK, enable tracing */
4445         tracing_disabled = 0;
4446
4447         atomic_notifier_chain_register(&panic_notifier_list,
4448                                        &trace_panic_notifier);
4449
4450         register_die_notifier(&trace_die_notifier);
4451
4452         return 0;
4453
4454 out_free_cpumask:
4455         free_cpumask_var(tracing_reader_cpumask);
4456 out_free_tracing_cpumask:
4457         free_cpumask_var(tracing_cpumask);
4458 out_free_buffer_mask:
4459         free_cpumask_var(tracing_buffer_mask);
4460 out:
4461         return ret;
4462 }
4463
4464 __init static int clear_boot_tracer(void)
4465 {
4466         /*
4467          * The default tracer at boot buffer is an init section.
4468          * This function is called in lateinit. If we did not
4469          * find the boot tracer, then clear it out, to prevent
4470          * later registration from accessing the buffer that is
4471          * about to be freed.
4472          */
4473         if (!default_bootup_tracer)
4474                 return 0;
4475
4476         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4477                default_bootup_tracer);
4478         default_bootup_tracer = NULL;
4479
4480         return 0;
4481 }
4482
4483 early_initcall(tracer_alloc_buffers);
4484 fs_initcall(tracer_init_debugfs);
4485 late_initcall(clear_boot_tracer);