f8cf959bad456dead3a5dec2fdd1a21e384f22bb
[linux-2.6.git] / kernel / trace / trace.c
1 /*
2  * ring buffer based function tracer
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally taken from the RT patch by:
8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code from the latency_tracer, that is:
11  *  Copyright (C) 2004-2006 Ingo Molnar
12  *  Copyright (C) 2004 William Lee Irwin III
13  */
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/fs.h>
40
41 #include "trace.h"
42 #include "trace_output.h"
43
44 #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(ftrace_cpu_disabled);
95 }
96
97 static inline void ftrace_enable_cpu(void)
98 {
99         __this_cpu_dec(ftrace_cpu_disabled);
100         preempt_enable();
101 }
102
103 cpumask_var_t __read_mostly     tracing_buffer_mask;
104
105 /*
106  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
107  *
108  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
109  * is set, then ftrace_dump is called. This will output the contents
110  * of the ftrace buffers to the console.  This is very useful for
111  * capturing traces that lead to crashes and outputing it to a
112  * serial console.
113  *
114  * It is default off, but you can enable it with either specifying
115  * "ftrace_dump_on_oops" in the kernel command line, or setting
116  * /proc/sys/kernel/ftrace_dump_on_oops
117  * Set 1 if you want to dump buffers of all CPUs
118  * Set 2 if you want to dump the buffer of the CPU that triggered oops
119  */
120
121 enum ftrace_dump_mode ftrace_dump_on_oops;
122
123 static int tracing_set_tracer(const char *buf);
124
125 #define MAX_TRACER_SIZE         100
126 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
127 static char *default_bootup_tracer;
128
129 static int __init set_cmdline_ftrace(char *str)
130 {
131         strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
132         default_bootup_tracer = bootup_tracer_buf;
133         /* We are using ftrace early, expand it */
134         ring_buffer_expanded = 1;
135         return 1;
136 }
137 __setup("ftrace=", set_cmdline_ftrace);
138
139 static int __init set_ftrace_dump_on_oops(char *str)
140 {
141         if (*str++ != '=' || !*str) {
142                 ftrace_dump_on_oops = DUMP_ALL;
143                 return 1;
144         }
145
146         if (!strcmp("orig_cpu", str)) {
147                 ftrace_dump_on_oops = DUMP_ORIG;
148                 return 1;
149         }
150
151         return 0;
152 }
153 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
154
155 unsigned long long ns2usecs(cycle_t nsec)
156 {
157         nsec += 500;
158         do_div(nsec, 1000);
159         return nsec;
160 }
161
162 /*
163  * The global_trace is the descriptor that holds the tracing
164  * buffers for the live tracing. For each CPU, it contains
165  * a link list of pages that will store trace entries. The
166  * page descriptor of the pages in the memory is used to hold
167  * the link list by linking the lru item in the page descriptor
168  * to each of the pages in the buffer per CPU.
169  *
170  * For each active CPU there is a data field that holds the
171  * pages for the buffer for that CPU. Each CPU has the same number
172  * of pages allocated for its buffer.
173  */
174 static struct trace_array       global_trace;
175
176 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
177
178 int filter_current_check_discard(struct ring_buffer *buffer,
179                                  struct ftrace_event_call *call, void *rec,
180                                  struct ring_buffer_event *event)
181 {
182         return filter_check_discard(call, rec, buffer, event);
183 }
184 EXPORT_SYMBOL_GPL(filter_current_check_discard);
185
186 cycle_t ftrace_now(int cpu)
187 {
188         u64 ts;
189
190         /* Early boot up does not have a buffer yet */
191         if (!global_trace.buffer)
192                 return trace_clock_local();
193
194         ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
195         ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
196
197         return ts;
198 }
199
200 /*
201  * The max_tr is used to snapshot the global_trace when a maximum
202  * latency is reached. Some tracers will use this to store a maximum
203  * trace while it continues examining live traces.
204  *
205  * The buffers for the max_tr are set up the same as the global_trace.
206  * When a snapshot is taken, the link list of the max_tr is swapped
207  * with the link list of the global_trace and the buffers are reset for
208  * the global_trace so the tracing can continue.
209  */
210 static struct trace_array       max_tr;
211
212 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
213
214 /* tracer_enabled is used to toggle activation of a tracer */
215 static int                      tracer_enabled = 1;
216
217 /**
218  * tracing_is_enabled - return tracer_enabled status
219  *
220  * This function is used by other tracers to know the status
221  * of the tracer_enabled flag.  Tracers may use this function
222  * to know if it should enable their features when starting
223  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
224  */
225 int tracing_is_enabled(void)
226 {
227         return tracer_enabled;
228 }
229
230 /*
231  * trace_buf_size is the size in bytes that is allocated
232  * for a buffer. Note, the number of bytes is always rounded
233  * to page size.
234  *
235  * This number is purposely set to a low number of 16384.
236  * If the dump on oops happens, it will be much appreciated
237  * to not have to wait for all that output. Anyway this can be
238  * boot time and run time configurable.
239  */
240 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
241
242 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
243
244 /* trace_types holds a link list of available tracers. */
245 static struct tracer            *trace_types __read_mostly;
246
247 /* current_trace points to the tracer that is currently active */
248 static struct tracer            *current_trace __read_mostly;
249
250 /*
251  * trace_types_lock is used to protect the trace_types list.
252  */
253 static DEFINE_MUTEX(trace_types_lock);
254
255 /*
256  * serialize the access of the ring buffer
257  *
258  * ring buffer serializes readers, but it is low level protection.
259  * The validity of the events (which returns by ring_buffer_peek() ..etc)
260  * are not protected by ring buffer.
261  *
262  * The content of events may become garbage if we allow other process consumes
263  * these events concurrently:
264  *   A) the page of the consumed events may become a normal page
265  *      (not reader page) in ring buffer, and this page will be rewrited
266  *      by events producer.
267  *   B) The page of the consumed events may become a page for splice_read,
268  *      and this page will be returned to system.
269  *
270  * These primitives allow multi process access to different cpu ring buffer
271  * concurrently.
272  *
273  * These primitives don't distinguish read-only and read-consume access.
274  * Multi read-only access are also serialized.
275  */
276
277 #ifdef CONFIG_SMP
278 static DECLARE_RWSEM(all_cpu_access_lock);
279 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
280
281 static inline void trace_access_lock(int cpu)
282 {
283         if (cpu == TRACE_PIPE_ALL_CPU) {
284                 /* gain it for accessing the whole ring buffer. */
285                 down_write(&all_cpu_access_lock);
286         } else {
287                 /* gain it for accessing a cpu ring buffer. */
288
289                 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
290                 down_read(&all_cpu_access_lock);
291
292                 /* Secondly block other access to this @cpu ring buffer. */
293                 mutex_lock(&per_cpu(cpu_access_lock, cpu));
294         }
295 }
296
297 static inline void trace_access_unlock(int cpu)
298 {
299         if (cpu == TRACE_PIPE_ALL_CPU) {
300                 up_write(&all_cpu_access_lock);
301         } else {
302                 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
303                 up_read(&all_cpu_access_lock);
304         }
305 }
306
307 static inline void trace_access_lock_init(void)
308 {
309         int cpu;
310
311         for_each_possible_cpu(cpu)
312                 mutex_init(&per_cpu(cpu_access_lock, cpu));
313 }
314
315 #else
316
317 static DEFINE_MUTEX(access_lock);
318
319 static inline void trace_access_lock(int cpu)
320 {
321         (void)cpu;
322         mutex_lock(&access_lock);
323 }
324
325 static inline void trace_access_unlock(int cpu)
326 {
327         (void)cpu;
328         mutex_unlock(&access_lock);
329 }
330
331 static inline void trace_access_lock_init(void)
332 {
333 }
334
335 #endif
336
337 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
338 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
339
340 /* trace_flags holds trace_options default values */
341 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
342         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
343         TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD;
344
345 static int trace_stop_count;
346 static DEFINE_SPINLOCK(tracing_start_lock);
347
348 /**
349  * trace_wake_up - wake up tasks waiting for trace input
350  *
351  * Simply wakes up any task that is blocked on the trace_wait
352  * queue. These is used with trace_poll for tasks polling the trace.
353  */
354 void trace_wake_up(void)
355 {
356         int cpu;
357
358         if (trace_flags & TRACE_ITER_BLOCK)
359                 return;
360         /*
361          * The runqueue_is_locked() can fail, but this is the best we
362          * have for now:
363          */
364         cpu = get_cpu();
365         if (!runqueue_is_locked(cpu))
366                 wake_up(&trace_wait);
367         put_cpu();
368 }
369
370 static int __init set_buf_size(char *str)
371 {
372         unsigned long buf_size;
373
374         if (!str)
375                 return 0;
376         buf_size = memparse(str, &str);
377         /* nr_entries can not be zero */
378         if (buf_size == 0)
379                 return 0;
380         trace_buf_size = buf_size;
381         return 1;
382 }
383 __setup("trace_buf_size=", set_buf_size);
384
385 static int __init set_tracing_thresh(char *str)
386 {
387         unsigned long threshhold;
388         int ret;
389
390         if (!str)
391                 return 0;
392         ret = strict_strtoul(str, 0, &threshhold);
393         if (ret < 0)
394                 return 0;
395         tracing_thresh = threshhold * 1000;
396         return 1;
397 }
398 __setup("tracing_thresh=", set_tracing_thresh);
399
400 unsigned long nsecs_to_usecs(unsigned long nsecs)
401 {
402         return nsecs / 1000;
403 }
404
405 /* These must match the bit postions in trace_iterator_flags */
406 static const char *trace_options[] = {
407         "print-parent",
408         "sym-offset",
409         "sym-addr",
410         "verbose",
411         "raw",
412         "hex",
413         "bin",
414         "block",
415         "stacktrace",
416         "trace_printk",
417         "ftrace_preempt",
418         "branch",
419         "annotate",
420         "userstacktrace",
421         "sym-userobj",
422         "printk-msg-only",
423         "context-info",
424         "latency-format",
425         "sleep-time",
426         "graph-time",
427         "record-cmd",
428         NULL
429 };
430
431 static struct {
432         u64 (*func)(void);
433         const char *name;
434 } trace_clocks[] = {
435         { trace_clock_local,    "local" },
436         { trace_clock_global,   "global" },
437 };
438
439 int trace_clock_id;
440
441 /*
442  * trace_parser_get_init - gets the buffer for trace parser
443  */
444 int trace_parser_get_init(struct trace_parser *parser, int size)
445 {
446         memset(parser, 0, sizeof(*parser));
447
448         parser->buffer = kmalloc(size, GFP_KERNEL);
449         if (!parser->buffer)
450                 return 1;
451
452         parser->size = size;
453         return 0;
454 }
455
456 /*
457  * trace_parser_put - frees the buffer for trace parser
458  */
459 void trace_parser_put(struct trace_parser *parser)
460 {
461         kfree(parser->buffer);
462 }
463
464 /*
465  * trace_get_user - reads the user input string separated by  space
466  * (matched by isspace(ch))
467  *
468  * For each string found the 'struct trace_parser' is updated,
469  * and the function returns.
470  *
471  * Returns number of bytes read.
472  *
473  * See kernel/trace/trace.h for 'struct trace_parser' details.
474  */
475 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
476         size_t cnt, loff_t *ppos)
477 {
478         char ch;
479         size_t read = 0;
480         ssize_t ret;
481
482         if (!*ppos)
483                 trace_parser_clear(parser);
484
485         ret = get_user(ch, ubuf++);
486         if (ret)
487                 goto out;
488
489         read++;
490         cnt--;
491
492         /*
493          * The parser is not finished with the last write,
494          * continue reading the user input without skipping spaces.
495          */
496         if (!parser->cont) {
497                 /* skip white space */
498                 while (cnt && isspace(ch)) {
499                         ret = get_user(ch, ubuf++);
500                         if (ret)
501                                 goto out;
502                         read++;
503                         cnt--;
504                 }
505
506                 /* only spaces were written */
507                 if (isspace(ch)) {
508                         *ppos += read;
509                         ret = read;
510                         goto out;
511                 }
512
513                 parser->idx = 0;
514         }
515
516         /* read the non-space input */
517         while (cnt && !isspace(ch)) {
518                 if (parser->idx < parser->size - 1)
519                         parser->buffer[parser->idx++] = ch;
520                 else {
521                         ret = -EINVAL;
522                         goto out;
523                 }
524                 ret = get_user(ch, ubuf++);
525                 if (ret)
526                         goto out;
527                 read++;
528                 cnt--;
529         }
530
531         /* We either got finished input or we have to wait for another call. */
532         if (isspace(ch)) {
533                 parser->buffer[parser->idx] = 0;
534                 parser->cont = false;
535         } else {
536                 parser->cont = true;
537                 parser->buffer[parser->idx++] = ch;
538         }
539
540         *ppos += read;
541         ret = read;
542
543 out:
544         return ret;
545 }
546
547 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
548 {
549         int len;
550         int ret;
551
552         if (!cnt)
553                 return 0;
554
555         if (s->len <= s->readpos)
556                 return -EBUSY;
557
558         len = s->len - s->readpos;
559         if (cnt > len)
560                 cnt = len;
561         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
562         if (ret == cnt)
563                 return -EFAULT;
564
565         cnt -= ret;
566
567         s->readpos += cnt;
568         return cnt;
569 }
570
571 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
572 {
573         int len;
574         void *ret;
575
576         if (s->len <= s->readpos)
577                 return -EBUSY;
578
579         len = s->len - s->readpos;
580         if (cnt > len)
581                 cnt = len;
582         ret = memcpy(buf, s->buffer + s->readpos, cnt);
583         if (!ret)
584                 return -EFAULT;
585
586         s->readpos += cnt;
587         return cnt;
588 }
589
590 /*
591  * ftrace_max_lock is used to protect the swapping of buffers
592  * when taking a max snapshot. The buffers themselves are
593  * protected by per_cpu spinlocks. But the action of the swap
594  * needs its own lock.
595  *
596  * This is defined as a arch_spinlock_t in order to help
597  * with performance when lockdep debugging is enabled.
598  *
599  * It is also used in other places outside the update_max_tr
600  * so it needs to be defined outside of the
601  * CONFIG_TRACER_MAX_TRACE.
602  */
603 static arch_spinlock_t ftrace_max_lock =
604         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
605
606 unsigned long __read_mostly     tracing_thresh;
607
608 #ifdef CONFIG_TRACER_MAX_TRACE
609 unsigned long __read_mostly     tracing_max_latency;
610
611 /*
612  * Copy the new maximum trace into the separate maximum-trace
613  * structure. (this way the maximum trace is permanently saved,
614  * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
615  */
616 static void
617 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
618 {
619         struct trace_array_cpu *data = tr->data[cpu];
620         struct trace_array_cpu *max_data;
621
622         max_tr.cpu = cpu;
623         max_tr.time_start = data->preempt_timestamp;
624
625         max_data = max_tr.data[cpu];
626         max_data->saved_latency = tracing_max_latency;
627         max_data->critical_start = data->critical_start;
628         max_data->critical_end = data->critical_end;
629
630         memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
631         max_data->pid = tsk->pid;
632         max_data->uid = task_uid(tsk);
633         max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
634         max_data->policy = tsk->policy;
635         max_data->rt_priority = tsk->rt_priority;
636
637         /* record this tasks comm */
638         tracing_record_cmdline(tsk);
639 }
640
641 /**
642  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
643  * @tr: tracer
644  * @tsk: the task with the latency
645  * @cpu: The cpu that initiated the trace.
646  *
647  * Flip the buffers between the @tr and the max_tr and record information
648  * about which task was the cause of this latency.
649  */
650 void
651 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
652 {
653         struct ring_buffer *buf = tr->buffer;
654
655         if (trace_stop_count)
656                 return;
657
658         WARN_ON_ONCE(!irqs_disabled());
659         if (!current_trace->use_max_tr) {
660                 WARN_ON_ONCE(1);
661                 return;
662         }
663         arch_spin_lock(&ftrace_max_lock);
664
665         tr->buffer = max_tr.buffer;
666         max_tr.buffer = buf;
667
668         __update_max_tr(tr, tsk, cpu);
669         arch_spin_unlock(&ftrace_max_lock);
670 }
671
672 /**
673  * update_max_tr_single - only copy one trace over, and reset the rest
674  * @tr - tracer
675  * @tsk - task with the latency
676  * @cpu - the cpu of the buffer to copy.
677  *
678  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
679  */
680 void
681 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
682 {
683         int ret;
684
685         if (trace_stop_count)
686                 return;
687
688         WARN_ON_ONCE(!irqs_disabled());
689         if (!current_trace->use_max_tr) {
690                 WARN_ON_ONCE(1);
691                 return;
692         }
693
694         arch_spin_lock(&ftrace_max_lock);
695
696         ftrace_disable_cpu();
697
698         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
699
700         if (ret == -EBUSY) {
701                 /*
702                  * We failed to swap the buffer due to a commit taking
703                  * place on this CPU. We fail to record, but we reset
704                  * the max trace buffer (no one writes directly to it)
705                  * and flag that it failed.
706                  */
707                 trace_array_printk(&max_tr, _THIS_IP_,
708                         "Failed to swap buffers due to commit in progress\n");
709         }
710
711         ftrace_enable_cpu();
712
713         WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
714
715         __update_max_tr(tr, tsk, cpu);
716         arch_spin_unlock(&ftrace_max_lock);
717 }
718 #endif /* CONFIG_TRACER_MAX_TRACE */
719
720 /**
721  * register_tracer - register a tracer with the ftrace system.
722  * @type - the plugin for the tracer
723  *
724  * Register a new plugin tracer.
725  */
726 int register_tracer(struct tracer *type)
727 __releases(kernel_lock)
728 __acquires(kernel_lock)
729 {
730         struct tracer *t;
731         int ret = 0;
732
733         if (!type->name) {
734                 pr_info("Tracer must have a name\n");
735                 return -1;
736         }
737
738         if (strlen(type->name) >= MAX_TRACER_SIZE) {
739                 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
740                 return -1;
741         }
742
743         mutex_lock(&trace_types_lock);
744
745         tracing_selftest_running = true;
746
747         for (t = trace_types; t; t = t->next) {
748                 if (strcmp(type->name, t->name) == 0) {
749                         /* already found */
750                         pr_info("Tracer %s already registered\n",
751                                 type->name);
752                         ret = -1;
753                         goto out;
754                 }
755         }
756
757         if (!type->set_flag)
758                 type->set_flag = &dummy_set_flag;
759         if (!type->flags)
760                 type->flags = &dummy_tracer_flags;
761         else
762                 if (!type->flags->opts)
763                         type->flags->opts = dummy_tracer_opt;
764         if (!type->wait_pipe)
765                 type->wait_pipe = default_wait_pipe;
766
767
768 #ifdef CONFIG_FTRACE_STARTUP_TEST
769         if (type->selftest && !tracing_selftest_disabled) {
770                 struct tracer *saved_tracer = current_trace;
771                 struct trace_array *tr = &global_trace;
772
773                 /*
774                  * Run a selftest on this tracer.
775                  * Here we reset the trace buffer, and set the current
776                  * tracer to be this tracer. The tracer can then run some
777                  * internal tracing to verify that everything is in order.
778                  * If we fail, we do not register this tracer.
779                  */
780                 tracing_reset_online_cpus(tr);
781
782                 current_trace = type;
783                 /* the test is responsible for initializing and enabling */
784                 pr_info("Testing tracer %s: ", type->name);
785                 ret = type->selftest(type, tr);
786                 /* the test is responsible for resetting too */
787                 current_trace = saved_tracer;
788                 if (ret) {
789                         printk(KERN_CONT "FAILED!\n");
790                         goto out;
791                 }
792                 /* Only reset on passing, to avoid touching corrupted buffers */
793                 tracing_reset_online_cpus(tr);
794
795                 printk(KERN_CONT "PASSED\n");
796         }
797 #endif
798
799         type->next = trace_types;
800         trace_types = type;
801
802  out:
803         tracing_selftest_running = false;
804         mutex_unlock(&trace_types_lock);
805
806         if (ret || !default_bootup_tracer)
807                 goto out_unlock;
808
809         if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
810                 goto out_unlock;
811
812         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
813         /* Do we want this tracer to start on bootup? */
814         tracing_set_tracer(type->name);
815         default_bootup_tracer = NULL;
816         /* disable other selftests, since this will break it. */
817         tracing_selftest_disabled = 1;
818 #ifdef CONFIG_FTRACE_STARTUP_TEST
819         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
820                type->name);
821 #endif
822
823  out_unlock:
824         return ret;
825 }
826
827 void unregister_tracer(struct tracer *type)
828 {
829         struct tracer **t;
830
831         mutex_lock(&trace_types_lock);
832         for (t = &trace_types; *t; t = &(*t)->next) {
833                 if (*t == type)
834                         goto found;
835         }
836         pr_info("Tracer %s not registered\n", type->name);
837         goto out;
838
839  found:
840         *t = (*t)->next;
841
842         if (type == current_trace && tracer_enabled) {
843                 tracer_enabled = 0;
844                 tracing_stop();
845                 if (current_trace->stop)
846                         current_trace->stop(&global_trace);
847                 current_trace = &nop_trace;
848         }
849 out:
850         mutex_unlock(&trace_types_lock);
851 }
852
853 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
854 {
855         ftrace_disable_cpu();
856         ring_buffer_reset_cpu(buffer, cpu);
857         ftrace_enable_cpu();
858 }
859
860 void tracing_reset(struct trace_array *tr, int cpu)
861 {
862         struct ring_buffer *buffer = tr->buffer;
863
864         ring_buffer_record_disable(buffer);
865
866         /* Make sure all commits have finished */
867         synchronize_sched();
868         __tracing_reset(buffer, cpu);
869
870         ring_buffer_record_enable(buffer);
871 }
872
873 void tracing_reset_online_cpus(struct trace_array *tr)
874 {
875         struct ring_buffer *buffer = tr->buffer;
876         int cpu;
877
878         ring_buffer_record_disable(buffer);
879
880         /* Make sure all commits have finished */
881         synchronize_sched();
882
883         tr->time_start = ftrace_now(tr->cpu);
884
885         for_each_online_cpu(cpu)
886                 __tracing_reset(buffer, cpu);
887
888         ring_buffer_record_enable(buffer);
889 }
890
891 void tracing_reset_current(int cpu)
892 {
893         tracing_reset(&global_trace, cpu);
894 }
895
896 void tracing_reset_current_online_cpus(void)
897 {
898         tracing_reset_online_cpus(&global_trace);
899 }
900
901 #define SAVED_CMDLINES 128
902 #define NO_CMDLINE_MAP UINT_MAX
903 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
904 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
905 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
906 static int cmdline_idx;
907 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
908
909 /* temporary disable recording */
910 static atomic_t trace_record_cmdline_disabled __read_mostly;
911
912 static void trace_init_cmdlines(void)
913 {
914         memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
915         memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
916         cmdline_idx = 0;
917 }
918
919 int is_tracing_stopped(void)
920 {
921         return trace_stop_count;
922 }
923
924 /**
925  * ftrace_off_permanent - disable all ftrace code permanently
926  *
927  * This should only be called when a serious anomally has
928  * been detected.  This will turn off the function tracing,
929  * ring buffers, and other tracing utilites. It takes no
930  * locks and can be called from any context.
931  */
932 void ftrace_off_permanent(void)
933 {
934         tracing_disabled = 1;
935         ftrace_stop();
936         tracing_off_permanent();
937 }
938
939 /**
940  * tracing_start - quick start of the tracer
941  *
942  * If tracing is enabled but was stopped by tracing_stop,
943  * this will start the tracer back up.
944  */
945 void tracing_start(void)
946 {
947         struct ring_buffer *buffer;
948         unsigned long flags;
949
950         if (tracing_disabled)
951                 return;
952
953         spin_lock_irqsave(&tracing_start_lock, flags);
954         if (--trace_stop_count) {
955                 if (trace_stop_count < 0) {
956                         /* Someone screwed up their debugging */
957                         WARN_ON_ONCE(1);
958                         trace_stop_count = 0;
959                 }
960                 goto out;
961         }
962
963         /* Prevent the buffers from switching */
964         arch_spin_lock(&ftrace_max_lock);
965
966         buffer = global_trace.buffer;
967         if (buffer)
968                 ring_buffer_record_enable(buffer);
969
970         buffer = max_tr.buffer;
971         if (buffer)
972                 ring_buffer_record_enable(buffer);
973
974         arch_spin_unlock(&ftrace_max_lock);
975
976         ftrace_start();
977  out:
978         spin_unlock_irqrestore(&tracing_start_lock, flags);
979 }
980
981 /**
982  * tracing_stop - quick stop of the tracer
983  *
984  * Light weight way to stop tracing. Use in conjunction with
985  * tracing_start.
986  */
987 void tracing_stop(void)
988 {
989         struct ring_buffer *buffer;
990         unsigned long flags;
991
992         ftrace_stop();
993         spin_lock_irqsave(&tracing_start_lock, flags);
994         if (trace_stop_count++)
995                 goto out;
996
997         /* Prevent the buffers from switching */
998         arch_spin_lock(&ftrace_max_lock);
999
1000         buffer = global_trace.buffer;
1001         if (buffer)
1002                 ring_buffer_record_disable(buffer);
1003
1004         buffer = max_tr.buffer;
1005         if (buffer)
1006                 ring_buffer_record_disable(buffer);
1007
1008         arch_spin_unlock(&ftrace_max_lock);
1009
1010  out:
1011         spin_unlock_irqrestore(&tracing_start_lock, flags);
1012 }
1013
1014 void trace_stop_cmdline_recording(void);
1015
1016 static void trace_save_cmdline(struct task_struct *tsk)
1017 {
1018         unsigned pid, idx;
1019
1020         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1021                 return;
1022
1023         /*
1024          * It's not the end of the world if we don't get
1025          * the lock, but we also don't want to spin
1026          * nor do we want to disable interrupts,
1027          * so if we miss here, then better luck next time.
1028          */
1029         if (!arch_spin_trylock(&trace_cmdline_lock))
1030                 return;
1031
1032         idx = map_pid_to_cmdline[tsk->pid];
1033         if (idx == NO_CMDLINE_MAP) {
1034                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1035
1036                 /*
1037                  * Check whether the cmdline buffer at idx has a pid
1038                  * mapped. We are going to overwrite that entry so we
1039                  * need to clear the map_pid_to_cmdline. Otherwise we
1040                  * would read the new comm for the old pid.
1041                  */
1042                 pid = map_cmdline_to_pid[idx];
1043                 if (pid != NO_CMDLINE_MAP)
1044                         map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1045
1046                 map_cmdline_to_pid[idx] = tsk->pid;
1047                 map_pid_to_cmdline[tsk->pid] = idx;
1048
1049                 cmdline_idx = idx;
1050         }
1051
1052         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1053
1054         arch_spin_unlock(&trace_cmdline_lock);
1055 }
1056
1057 void trace_find_cmdline(int pid, char comm[])
1058 {
1059         unsigned map;
1060
1061         if (!pid) {
1062                 strcpy(comm, "<idle>");
1063                 return;
1064         }
1065
1066         if (WARN_ON_ONCE(pid < 0)) {
1067                 strcpy(comm, "<XXX>");
1068                 return;
1069         }
1070
1071         if (pid > PID_MAX_DEFAULT) {
1072                 strcpy(comm, "<...>");
1073                 return;
1074         }
1075
1076         preempt_disable();
1077         arch_spin_lock(&trace_cmdline_lock);
1078         map = map_pid_to_cmdline[pid];
1079         if (map != NO_CMDLINE_MAP)
1080                 strcpy(comm, saved_cmdlines[map]);
1081         else
1082                 strcpy(comm, "<...>");
1083
1084         arch_spin_unlock(&trace_cmdline_lock);
1085         preempt_enable();
1086 }
1087
1088 void tracing_record_cmdline(struct task_struct *tsk)
1089 {
1090         if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1091             !tracing_is_on())
1092                 return;
1093
1094         trace_save_cmdline(tsk);
1095 }
1096
1097 void
1098 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1099                              int pc)
1100 {
1101         struct task_struct *tsk = current;
1102
1103         entry->preempt_count            = pc & 0xff;
1104         entry->pid                      = (tsk) ? tsk->pid : 0;
1105         entry->lock_depth               = (tsk) ? tsk->lock_depth : 0;
1106         entry->flags =
1107 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1108                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1109 #else
1110                 TRACE_FLAG_IRQS_NOSUPPORT |
1111 #endif
1112                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1113                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1114                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1115 }
1116 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1117
1118 struct ring_buffer_event *
1119 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1120                           int type,
1121                           unsigned long len,
1122                           unsigned long flags, int pc)
1123 {
1124         struct ring_buffer_event *event;
1125
1126         event = ring_buffer_lock_reserve(buffer, len);
1127         if (event != NULL) {
1128                 struct trace_entry *ent = ring_buffer_event_data(event);
1129
1130                 tracing_generic_entry_update(ent, flags, pc);
1131                 ent->type = type;
1132         }
1133
1134         return event;
1135 }
1136
1137 static inline void
1138 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1139                              struct ring_buffer_event *event,
1140                              unsigned long flags, int pc,
1141                              int wake)
1142 {
1143         ring_buffer_unlock_commit(buffer, event);
1144
1145         ftrace_trace_stack(buffer, flags, 6, pc);
1146         ftrace_trace_userstack(buffer, flags, pc);
1147
1148         if (wake)
1149                 trace_wake_up();
1150 }
1151
1152 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1153                                 struct ring_buffer_event *event,
1154                                 unsigned long flags, int pc)
1155 {
1156         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1157 }
1158
1159 struct ring_buffer_event *
1160 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1161                                   int type, unsigned long len,
1162                                   unsigned long flags, int pc)
1163 {
1164         *current_rb = global_trace.buffer;
1165         return trace_buffer_lock_reserve(*current_rb,
1166                                          type, len, flags, pc);
1167 }
1168 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1169
1170 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1171                                         struct ring_buffer_event *event,
1172                                         unsigned long flags, int pc)
1173 {
1174         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1175 }
1176 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1177
1178 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1179                                        struct ring_buffer_event *event,
1180                                        unsigned long flags, int pc)
1181 {
1182         __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1183 }
1184 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1185
1186 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1187                                          struct ring_buffer_event *event)
1188 {
1189         ring_buffer_discard_commit(buffer, event);
1190 }
1191 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1192
1193 void
1194 trace_function(struct trace_array *tr,
1195                unsigned long ip, unsigned long parent_ip, unsigned long flags,
1196                int pc)
1197 {
1198         struct ftrace_event_call *call = &event_function;
1199         struct ring_buffer *buffer = tr->buffer;
1200         struct ring_buffer_event *event;
1201         struct ftrace_entry *entry;
1202
1203         /* If we are reading the ring buffer, don't trace */
1204         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1205                 return;
1206
1207         event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1208                                           flags, pc);
1209         if (!event)
1210                 return;
1211         entry   = ring_buffer_event_data(event);
1212         entry->ip                       = ip;
1213         entry->parent_ip                = parent_ip;
1214
1215         if (!filter_check_discard(call, entry, buffer, event))
1216                 ring_buffer_unlock_commit(buffer, event);
1217 }
1218
1219 void
1220 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1221        unsigned long ip, unsigned long parent_ip, unsigned long flags,
1222        int pc)
1223 {
1224         if (likely(!atomic_read(&data->disabled)))
1225                 trace_function(tr, ip, parent_ip, flags, pc);
1226 }
1227
1228 #ifdef CONFIG_STACKTRACE
1229 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1230                                  unsigned long flags,
1231                                  int skip, int pc)
1232 {
1233         struct ftrace_event_call *call = &event_kernel_stack;
1234         struct ring_buffer_event *event;
1235         struct stack_entry *entry;
1236         struct stack_trace trace;
1237
1238         event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1239                                           sizeof(*entry), flags, pc);
1240         if (!event)
1241                 return;
1242         entry   = ring_buffer_event_data(event);
1243         memset(&entry->caller, 0, sizeof(entry->caller));
1244
1245         trace.nr_entries        = 0;
1246         trace.max_entries       = FTRACE_STACK_ENTRIES;
1247         trace.skip              = skip;
1248         trace.entries           = entry->caller;
1249
1250         save_stack_trace(&trace);
1251         if (!filter_check_discard(call, entry, buffer, event))
1252                 ring_buffer_unlock_commit(buffer, event);
1253 }
1254
1255 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1256                         int skip, int pc)
1257 {
1258         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1259                 return;
1260
1261         __ftrace_trace_stack(buffer, flags, skip, pc);
1262 }
1263
1264 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1265                    int pc)
1266 {
1267         __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1268 }
1269
1270 /**
1271  * trace_dump_stack - record a stack back trace in the trace buffer
1272  */
1273 void trace_dump_stack(void)
1274 {
1275         unsigned long flags;
1276
1277         if (tracing_disabled || tracing_selftest_running)
1278                 return;
1279
1280         local_save_flags(flags);
1281
1282         /* skipping 3 traces, seems to get us at the caller of this function */
1283         __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1284 }
1285
1286 static DEFINE_PER_CPU(int, user_stack_count);
1287
1288 void
1289 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1290 {
1291         struct ftrace_event_call *call = &event_user_stack;
1292         struct ring_buffer_event *event;
1293         struct userstack_entry *entry;
1294         struct stack_trace trace;
1295
1296         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1297                 return;
1298
1299         /*
1300          * NMIs can not handle page faults, even with fix ups.
1301          * The save user stack can (and often does) fault.
1302          */
1303         if (unlikely(in_nmi()))
1304                 return;
1305
1306         /*
1307          * prevent recursion, since the user stack tracing may
1308          * trigger other kernel events.
1309          */
1310         preempt_disable();
1311         if (__this_cpu_read(user_stack_count))
1312                 goto out;
1313
1314         __this_cpu_inc(user_stack_count);
1315
1316
1317
1318         event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1319                                           sizeof(*entry), flags, pc);
1320         if (!event)
1321                 return;
1322         entry   = ring_buffer_event_data(event);
1323
1324         entry->tgid             = current->tgid;
1325         memset(&entry->caller, 0, sizeof(entry->caller));
1326
1327         trace.nr_entries        = 0;
1328         trace.max_entries       = FTRACE_STACK_ENTRIES;
1329         trace.skip              = 0;
1330         trace.entries           = entry->caller;
1331
1332         save_stack_trace_user(&trace);
1333         if (!filter_check_discard(call, entry, buffer, event))
1334                 ring_buffer_unlock_commit(buffer, event);
1335
1336         __this_cpu_dec(user_stack_count);
1337
1338  out:
1339         preempt_enable();
1340 }
1341
1342 #ifdef UNUSED
1343 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1344 {
1345         ftrace_trace_userstack(tr, flags, preempt_count());
1346 }
1347 #endif /* UNUSED */
1348
1349 #endif /* CONFIG_STACKTRACE */
1350
1351 /**
1352  * trace_vbprintk - write binary msg to tracing buffer
1353  *
1354  */
1355 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1356 {
1357         static arch_spinlock_t trace_buf_lock =
1358                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1359         static u32 trace_buf[TRACE_BUF_SIZE];
1360
1361         struct ftrace_event_call *call = &event_bprint;
1362         struct ring_buffer_event *event;
1363         struct ring_buffer *buffer;
1364         struct trace_array *tr = &global_trace;
1365         struct trace_array_cpu *data;
1366         struct bprint_entry *entry;
1367         unsigned long flags;
1368         int disable;
1369         int cpu, len = 0, size, pc;
1370
1371         if (unlikely(tracing_selftest_running || tracing_disabled))
1372                 return 0;
1373
1374         /* Don't pollute graph traces with trace_vprintk internals */
1375         pause_graph_tracing();
1376
1377         pc = preempt_count();
1378         preempt_disable_notrace();
1379         cpu = raw_smp_processor_id();
1380         data = tr->data[cpu];
1381
1382         disable = atomic_inc_return(&data->disabled);
1383         if (unlikely(disable != 1))
1384                 goto out;
1385
1386         /* Lockdep uses trace_printk for lock tracing */
1387         local_irq_save(flags);
1388         arch_spin_lock(&trace_buf_lock);
1389         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1390
1391         if (len > TRACE_BUF_SIZE || len < 0)
1392                 goto out_unlock;
1393
1394         size = sizeof(*entry) + sizeof(u32) * len;
1395         buffer = tr->buffer;
1396         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1397                                           flags, pc);
1398         if (!event)
1399                 goto out_unlock;
1400         entry = ring_buffer_event_data(event);
1401         entry->ip                       = ip;
1402         entry->fmt                      = fmt;
1403
1404         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1405         if (!filter_check_discard(call, entry, buffer, event)) {
1406                 ring_buffer_unlock_commit(buffer, event);
1407                 ftrace_trace_stack(buffer, flags, 6, pc);
1408         }
1409
1410 out_unlock:
1411         arch_spin_unlock(&trace_buf_lock);
1412         local_irq_restore(flags);
1413
1414 out:
1415         atomic_dec_return(&data->disabled);
1416         preempt_enable_notrace();
1417         unpause_graph_tracing();
1418
1419         return len;
1420 }
1421 EXPORT_SYMBOL_GPL(trace_vbprintk);
1422
1423 int trace_array_printk(struct trace_array *tr,
1424                        unsigned long ip, const char *fmt, ...)
1425 {
1426         int ret;
1427         va_list ap;
1428
1429         if (!(trace_flags & TRACE_ITER_PRINTK))
1430                 return 0;
1431
1432         va_start(ap, fmt);
1433         ret = trace_array_vprintk(tr, ip, fmt, ap);
1434         va_end(ap);
1435         return ret;
1436 }
1437
1438 int trace_array_vprintk(struct trace_array *tr,
1439                         unsigned long ip, const char *fmt, va_list args)
1440 {
1441         static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1442         static char trace_buf[TRACE_BUF_SIZE];
1443
1444         struct ftrace_event_call *call = &event_print;
1445         struct ring_buffer_event *event;
1446         struct ring_buffer *buffer;
1447         struct trace_array_cpu *data;
1448         int cpu, len = 0, size, pc;
1449         struct print_entry *entry;
1450         unsigned long irq_flags;
1451         int disable;
1452
1453         if (tracing_disabled || tracing_selftest_running)
1454                 return 0;
1455
1456         pc = preempt_count();
1457         preempt_disable_notrace();
1458         cpu = raw_smp_processor_id();
1459         data = tr->data[cpu];
1460
1461         disable = atomic_inc_return(&data->disabled);
1462         if (unlikely(disable != 1))
1463                 goto out;
1464
1465         pause_graph_tracing();
1466         raw_local_irq_save(irq_flags);
1467         arch_spin_lock(&trace_buf_lock);
1468         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1469
1470         size = sizeof(*entry) + len + 1;
1471         buffer = tr->buffer;
1472         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1473                                           irq_flags, pc);
1474         if (!event)
1475                 goto out_unlock;
1476         entry = ring_buffer_event_data(event);
1477         entry->ip = ip;
1478
1479         memcpy(&entry->buf, trace_buf, len);
1480         entry->buf[len] = '\0';
1481         if (!filter_check_discard(call, entry, buffer, event)) {
1482                 ring_buffer_unlock_commit(buffer, event);
1483                 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1484         }
1485
1486  out_unlock:
1487         arch_spin_unlock(&trace_buf_lock);
1488         raw_local_irq_restore(irq_flags);
1489         unpause_graph_tracing();
1490  out:
1491         atomic_dec_return(&data->disabled);
1492         preempt_enable_notrace();
1493
1494         return len;
1495 }
1496
1497 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1498 {
1499         return trace_array_vprintk(&global_trace, ip, fmt, args);
1500 }
1501 EXPORT_SYMBOL_GPL(trace_vprintk);
1502
1503 static void trace_iterator_increment(struct trace_iterator *iter)
1504 {
1505         /* Don't allow ftrace to trace into the ring buffers */
1506         ftrace_disable_cpu();
1507
1508         iter->idx++;
1509         if (iter->buffer_iter[iter->cpu])
1510                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1511
1512         ftrace_enable_cpu();
1513 }
1514
1515 static struct trace_entry *
1516 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1517                 unsigned long *lost_events)
1518 {
1519         struct ring_buffer_event *event;
1520         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1521
1522         /* Don't allow ftrace to trace into the ring buffers */
1523         ftrace_disable_cpu();
1524
1525         if (buf_iter)
1526                 event = ring_buffer_iter_peek(buf_iter, ts);
1527         else
1528                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1529                                          lost_events);
1530
1531         ftrace_enable_cpu();
1532
1533         return event ? ring_buffer_event_data(event) : NULL;
1534 }
1535
1536 static struct trace_entry *
1537 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1538                   unsigned long *missing_events, u64 *ent_ts)
1539 {
1540         struct ring_buffer *buffer = iter->tr->buffer;
1541         struct trace_entry *ent, *next = NULL;
1542         unsigned long lost_events = 0, next_lost = 0;
1543         int cpu_file = iter->cpu_file;
1544         u64 next_ts = 0, ts;
1545         int next_cpu = -1;
1546         int cpu;
1547
1548         /*
1549          * If we are in a per_cpu trace file, don't bother by iterating over
1550          * all cpu and peek directly.
1551          */
1552         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1553                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1554                         return NULL;
1555                 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1556                 if (ent_cpu)
1557                         *ent_cpu = cpu_file;
1558
1559                 return ent;
1560         }
1561
1562         for_each_tracing_cpu(cpu) {
1563
1564                 if (ring_buffer_empty_cpu(buffer, cpu))
1565                         continue;
1566
1567                 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1568
1569                 /*
1570                  * Pick the entry with the smallest timestamp:
1571                  */
1572                 if (ent && (!next || ts < next_ts)) {
1573                         next = ent;
1574                         next_cpu = cpu;
1575                         next_ts = ts;
1576                         next_lost = lost_events;
1577                 }
1578         }
1579
1580         if (ent_cpu)
1581                 *ent_cpu = next_cpu;
1582
1583         if (ent_ts)
1584                 *ent_ts = next_ts;
1585
1586         if (missing_events)
1587                 *missing_events = next_lost;
1588
1589         return next;
1590 }
1591
1592 /* Find the next real entry, without updating the iterator itself */
1593 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1594                                           int *ent_cpu, u64 *ent_ts)
1595 {
1596         return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1597 }
1598
1599 /* Find the next real entry, and increment the iterator to the next entry */
1600 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1601 {
1602         iter->ent = __find_next_entry(iter, &iter->cpu,
1603                                       &iter->lost_events, &iter->ts);
1604
1605         if (iter->ent)
1606                 trace_iterator_increment(iter);
1607
1608         return iter->ent ? iter : NULL;
1609 }
1610
1611 static void trace_consume(struct trace_iterator *iter)
1612 {
1613         /* Don't allow ftrace to trace into the ring buffers */
1614         ftrace_disable_cpu();
1615         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1616                             &iter->lost_events);
1617         ftrace_enable_cpu();
1618 }
1619
1620 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1621 {
1622         struct trace_iterator *iter = m->private;
1623         int i = (int)*pos;
1624         void *ent;
1625
1626         WARN_ON_ONCE(iter->leftover);
1627
1628         (*pos)++;
1629
1630         /* can't go backwards */
1631         if (iter->idx > i)
1632                 return NULL;
1633
1634         if (iter->idx < 0)
1635                 ent = trace_find_next_entry_inc(iter);
1636         else
1637                 ent = iter;
1638
1639         while (ent && iter->idx < i)
1640                 ent = trace_find_next_entry_inc(iter);
1641
1642         iter->pos = *pos;
1643
1644         return ent;
1645 }
1646
1647 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1648 {
1649         struct trace_array *tr = iter->tr;
1650         struct ring_buffer_event *event;
1651         struct ring_buffer_iter *buf_iter;
1652         unsigned long entries = 0;
1653         u64 ts;
1654
1655         tr->data[cpu]->skipped_entries = 0;
1656
1657         if (!iter->buffer_iter[cpu])
1658                 return;
1659
1660         buf_iter = iter->buffer_iter[cpu];
1661         ring_buffer_iter_reset(buf_iter);
1662
1663         /*
1664          * We could have the case with the max latency tracers
1665          * that a reset never took place on a cpu. This is evident
1666          * by the timestamp being before the start of the buffer.
1667          */
1668         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1669                 if (ts >= iter->tr->time_start)
1670                         break;
1671                 entries++;
1672                 ring_buffer_read(buf_iter, NULL);
1673         }
1674
1675         tr->data[cpu]->skipped_entries = entries;
1676 }
1677
1678 /*
1679  * The current tracer is copied to avoid a global locking
1680  * all around.
1681  */
1682 static void *s_start(struct seq_file *m, loff_t *pos)
1683 {
1684         struct trace_iterator *iter = m->private;
1685         static struct tracer *old_tracer;
1686         int cpu_file = iter->cpu_file;
1687         void *p = NULL;
1688         loff_t l = 0;
1689         int cpu;
1690
1691         /* copy the tracer to avoid using a global lock all around */
1692         mutex_lock(&trace_types_lock);
1693         if (unlikely(old_tracer != current_trace && current_trace)) {
1694                 old_tracer = current_trace;
1695                 *iter->trace = *current_trace;
1696         }
1697         mutex_unlock(&trace_types_lock);
1698
1699         atomic_inc(&trace_record_cmdline_disabled);
1700
1701         if (*pos != iter->pos) {
1702                 iter->ent = NULL;
1703                 iter->cpu = 0;
1704                 iter->idx = -1;
1705
1706                 ftrace_disable_cpu();
1707
1708                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1709                         for_each_tracing_cpu(cpu)
1710                                 tracing_iter_reset(iter, cpu);
1711                 } else
1712                         tracing_iter_reset(iter, cpu_file);
1713
1714                 ftrace_enable_cpu();
1715
1716                 iter->leftover = 0;
1717                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1718                         ;
1719
1720         } else {
1721                 /*
1722                  * If we overflowed the seq_file before, then we want
1723                  * to just reuse the trace_seq buffer again.
1724                  */
1725                 if (iter->leftover)
1726                         p = iter;
1727                 else {
1728                         l = *pos - 1;
1729                         p = s_next(m, p, &l);
1730                 }
1731         }
1732
1733         trace_event_read_lock();
1734         trace_access_lock(cpu_file);
1735         return p;
1736 }
1737
1738 static void s_stop(struct seq_file *m, void *p)
1739 {
1740         struct trace_iterator *iter = m->private;
1741
1742         atomic_dec(&trace_record_cmdline_disabled);
1743         trace_access_unlock(iter->cpu_file);
1744         trace_event_read_unlock();
1745 }
1746
1747 static void print_lat_help_header(struct seq_file *m)
1748 {
1749         seq_puts(m, "#                  _------=> CPU#            \n");
1750         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1751         seq_puts(m, "#                | / _----=> need-resched    \n");
1752         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1753         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1754         seq_puts(m, "#                |||| /_--=> lock-depth       \n");
1755         seq_puts(m, "#                |||||/     delay             \n");
1756         seq_puts(m, "#  cmd     pid   |||||| time  |   caller      \n");
1757         seq_puts(m, "#     \\   /      ||||||   \\   |   /           \n");
1758 }
1759
1760 static void print_func_help_header(struct seq_file *m)
1761 {
1762         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1763         seq_puts(m, "#              | |       |          |         |\n");
1764 }
1765
1766
1767 void
1768 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1769 {
1770         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1771         struct trace_array *tr = iter->tr;
1772         struct trace_array_cpu *data = tr->data[tr->cpu];
1773         struct tracer *type = current_trace;
1774         unsigned long entries = 0;
1775         unsigned long total = 0;
1776         unsigned long count;
1777         const char *name = "preemption";
1778         int cpu;
1779
1780         if (type)
1781                 name = type->name;
1782
1783
1784         for_each_tracing_cpu(cpu) {
1785                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1786                 /*
1787                  * If this buffer has skipped entries, then we hold all
1788                  * entries for the trace and we need to ignore the
1789                  * ones before the time stamp.
1790                  */
1791                 if (tr->data[cpu]->skipped_entries) {
1792                         count -= tr->data[cpu]->skipped_entries;
1793                         /* total is the same as the entries */
1794                         total += count;
1795                 } else
1796                         total += count +
1797                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1798                 entries += count;
1799         }
1800
1801         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1802                    name, UTS_RELEASE);
1803         seq_puts(m, "# -----------------------------------"
1804                  "---------------------------------\n");
1805         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1806                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1807                    nsecs_to_usecs(data->saved_latency),
1808                    entries,
1809                    total,
1810                    tr->cpu,
1811 #if defined(CONFIG_PREEMPT_NONE)
1812                    "server",
1813 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1814                    "desktop",
1815 #elif defined(CONFIG_PREEMPT)
1816                    "preempt",
1817 #else
1818                    "unknown",
1819 #endif
1820                    /* These are reserved for later use */
1821                    0, 0, 0, 0);
1822 #ifdef CONFIG_SMP
1823         seq_printf(m, " #P:%d)\n", num_online_cpus());
1824 #else
1825         seq_puts(m, ")\n");
1826 #endif
1827         seq_puts(m, "#    -----------------\n");
1828         seq_printf(m, "#    | task: %.16s-%d "
1829                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1830                    data->comm, data->pid, data->uid, data->nice,
1831                    data->policy, data->rt_priority);
1832         seq_puts(m, "#    -----------------\n");
1833
1834         if (data->critical_start) {
1835                 seq_puts(m, "#  => started at: ");
1836                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1837                 trace_print_seq(m, &iter->seq);
1838                 seq_puts(m, "\n#  => ended at:   ");
1839                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1840                 trace_print_seq(m, &iter->seq);
1841                 seq_puts(m, "\n#\n");
1842         }
1843
1844         seq_puts(m, "#\n");
1845 }
1846
1847 static void test_cpu_buff_start(struct trace_iterator *iter)
1848 {
1849         struct trace_seq *s = &iter->seq;
1850
1851         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1852                 return;
1853
1854         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1855                 return;
1856
1857         if (cpumask_test_cpu(iter->cpu, iter->started))
1858                 return;
1859
1860         if (iter->tr->data[iter->cpu]->skipped_entries)
1861                 return;
1862
1863         cpumask_set_cpu(iter->cpu, iter->started);
1864
1865         /* Don't print started cpu buffer for the first entry of the trace */
1866         if (iter->idx > 1)
1867                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1868                                 iter->cpu);
1869 }
1870
1871 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1872 {
1873         struct trace_seq *s = &iter->seq;
1874         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1875         struct trace_entry *entry;
1876         struct trace_event *event;
1877
1878         entry = iter->ent;
1879
1880         test_cpu_buff_start(iter);
1881
1882         event = ftrace_find_event(entry->type);
1883
1884         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1885                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1886                         if (!trace_print_lat_context(iter))
1887                                 goto partial;
1888                 } else {
1889                         if (!trace_print_context(iter))
1890                                 goto partial;
1891                 }
1892         }
1893
1894         if (event)
1895                 return event->funcs->trace(iter, sym_flags, event);
1896
1897         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1898                 goto partial;
1899
1900         return TRACE_TYPE_HANDLED;
1901 partial:
1902         return TRACE_TYPE_PARTIAL_LINE;
1903 }
1904
1905 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1906 {
1907         struct trace_seq *s = &iter->seq;
1908         struct trace_entry *entry;
1909         struct trace_event *event;
1910
1911         entry = iter->ent;
1912
1913         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1914                 if (!trace_seq_printf(s, "%d %d %llu ",
1915                                       entry->pid, iter->cpu, iter->ts))
1916                         goto partial;
1917         }
1918
1919         event = ftrace_find_event(entry->type);
1920         if (event)
1921                 return event->funcs->raw(iter, 0, event);
1922
1923         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1924                 goto partial;
1925
1926         return TRACE_TYPE_HANDLED;
1927 partial:
1928         return TRACE_TYPE_PARTIAL_LINE;
1929 }
1930
1931 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1932 {
1933         struct trace_seq *s = &iter->seq;
1934         unsigned char newline = '\n';
1935         struct trace_entry *entry;
1936         struct trace_event *event;
1937
1938         entry = iter->ent;
1939
1940         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1941                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1942                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1943                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1944         }
1945
1946         event = ftrace_find_event(entry->type);
1947         if (event) {
1948                 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1949                 if (ret != TRACE_TYPE_HANDLED)
1950                         return ret;
1951         }
1952
1953         SEQ_PUT_FIELD_RET(s, newline);
1954
1955         return TRACE_TYPE_HANDLED;
1956 }
1957
1958 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1959 {
1960         struct trace_seq *s = &iter->seq;
1961         struct trace_entry *entry;
1962         struct trace_event *event;
1963
1964         entry = iter->ent;
1965
1966         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1967                 SEQ_PUT_FIELD_RET(s, entry->pid);
1968                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1969                 SEQ_PUT_FIELD_RET(s, iter->ts);
1970         }
1971
1972         event = ftrace_find_event(entry->type);
1973         return event ? event->funcs->binary(iter, 0, event) :
1974                 TRACE_TYPE_HANDLED;
1975 }
1976
1977 int trace_empty(struct trace_iterator *iter)
1978 {
1979         int cpu;
1980
1981         /* If we are looking at one CPU buffer, only check that one */
1982         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1983                 cpu = iter->cpu_file;
1984                 if (iter->buffer_iter[cpu]) {
1985                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1986                                 return 0;
1987                 } else {
1988                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1989                                 return 0;
1990                 }
1991                 return 1;
1992         }
1993
1994         for_each_tracing_cpu(cpu) {
1995                 if (iter->buffer_iter[cpu]) {
1996                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1997                                 return 0;
1998                 } else {
1999                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2000                                 return 0;
2001                 }
2002         }
2003
2004         return 1;
2005 }
2006
2007 /*  Called with trace_event_read_lock() held. */
2008 enum print_line_t print_trace_line(struct trace_iterator *iter)
2009 {
2010         enum print_line_t ret;
2011
2012         if (iter->lost_events)
2013                 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2014                                  iter->cpu, iter->lost_events);
2015
2016         if (iter->trace && iter->trace->print_line) {
2017                 ret = iter->trace->print_line(iter);
2018                 if (ret != TRACE_TYPE_UNHANDLED)
2019                         return ret;
2020         }
2021
2022         if (iter->ent->type == TRACE_BPRINT &&
2023                         trace_flags & TRACE_ITER_PRINTK &&
2024                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2025                 return trace_print_bprintk_msg_only(iter);
2026
2027         if (iter->ent->type == TRACE_PRINT &&
2028                         trace_flags & TRACE_ITER_PRINTK &&
2029                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2030                 return trace_print_printk_msg_only(iter);
2031
2032         if (trace_flags & TRACE_ITER_BIN)
2033                 return print_bin_fmt(iter);
2034
2035         if (trace_flags & TRACE_ITER_HEX)
2036                 return print_hex_fmt(iter);
2037
2038         if (trace_flags & TRACE_ITER_RAW)
2039                 return print_raw_fmt(iter);
2040
2041         return print_trace_fmt(iter);
2042 }
2043
2044 void trace_default_header(struct seq_file *m)
2045 {
2046         struct trace_iterator *iter = m->private;
2047
2048         if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2049                 /* print nothing if the buffers are empty */
2050                 if (trace_empty(iter))
2051                         return;
2052                 print_trace_header(m, iter);
2053                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2054                         print_lat_help_header(m);
2055         } else {
2056                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2057                         print_func_help_header(m);
2058         }
2059 }
2060
2061 static int s_show(struct seq_file *m, void *v)
2062 {
2063         struct trace_iterator *iter = v;
2064         int ret;
2065
2066         if (iter->ent == NULL) {
2067                 if (iter->tr) {
2068                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2069                         seq_puts(m, "#\n");
2070                 }
2071                 if (iter->trace && iter->trace->print_header)
2072                         iter->trace->print_header(m);
2073                 else
2074                         trace_default_header(m);
2075
2076         } else if (iter->leftover) {
2077                 /*
2078                  * If we filled the seq_file buffer earlier, we
2079                  * want to just show it now.
2080                  */
2081                 ret = trace_print_seq(m, &iter->seq);
2082
2083                 /* ret should this time be zero, but you never know */
2084                 iter->leftover = ret;
2085
2086         } else {
2087                 print_trace_line(iter);
2088                 ret = trace_print_seq(m, &iter->seq);
2089                 /*
2090                  * If we overflow the seq_file buffer, then it will
2091                  * ask us for this data again at start up.
2092                  * Use that instead.
2093                  *  ret is 0 if seq_file write succeeded.
2094                  *        -1 otherwise.
2095                  */
2096                 iter->leftover = ret;
2097         }
2098
2099         return 0;
2100 }
2101
2102 static const struct seq_operations tracer_seq_ops = {
2103         .start          = s_start,
2104         .next           = s_next,
2105         .stop           = s_stop,
2106         .show           = s_show,
2107 };
2108
2109 static struct trace_iterator *
2110 __tracing_open(struct inode *inode, struct file *file)
2111 {
2112         long cpu_file = (long) inode->i_private;
2113         void *fail_ret = ERR_PTR(-ENOMEM);
2114         struct trace_iterator *iter;
2115         struct seq_file *m;
2116         int cpu, ret;
2117
2118         if (tracing_disabled)
2119                 return ERR_PTR(-ENODEV);
2120
2121         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2122         if (!iter)
2123                 return ERR_PTR(-ENOMEM);
2124
2125         /*
2126          * We make a copy of the current tracer to avoid concurrent
2127          * changes on it while we are reading.
2128          */
2129         mutex_lock(&trace_types_lock);
2130         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2131         if (!iter->trace)
2132                 goto fail;
2133
2134         if (current_trace)
2135                 *iter->trace = *current_trace;
2136
2137         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2138                 goto fail;
2139
2140         if (current_trace && current_trace->print_max)
2141                 iter->tr = &max_tr;
2142         else
2143                 iter->tr = &global_trace;
2144         iter->pos = -1;
2145         mutex_init(&iter->mutex);
2146         iter->cpu_file = cpu_file;
2147
2148         /* Notify the tracer early; before we stop tracing. */
2149         if (iter->trace && iter->trace->open)
2150                 iter->trace->open(iter);
2151
2152         /* Annotate start of buffers if we had overruns */
2153         if (ring_buffer_overruns(iter->tr->buffer))
2154                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2155
2156         /* stop the trace while dumping */
2157         tracing_stop();
2158
2159         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2160                 for_each_tracing_cpu(cpu) {
2161                         iter->buffer_iter[cpu] =
2162                                 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2163                 }
2164                 ring_buffer_read_prepare_sync();
2165                 for_each_tracing_cpu(cpu) {
2166                         ring_buffer_read_start(iter->buffer_iter[cpu]);
2167                         tracing_iter_reset(iter, cpu);
2168                 }
2169         } else {
2170                 cpu = iter->cpu_file;
2171                 iter->buffer_iter[cpu] =
2172                         ring_buffer_read_prepare(iter->tr->buffer, cpu);
2173                 ring_buffer_read_prepare_sync();
2174                 ring_buffer_read_start(iter->buffer_iter[cpu]);
2175                 tracing_iter_reset(iter, cpu);
2176         }
2177
2178         ret = seq_open(file, &tracer_seq_ops);
2179         if (ret < 0) {
2180                 fail_ret = ERR_PTR(ret);
2181                 goto fail_buffer;
2182         }
2183
2184         m = file->private_data;
2185         m->private = iter;
2186
2187         mutex_unlock(&trace_types_lock);
2188
2189         return iter;
2190
2191  fail_buffer:
2192         for_each_tracing_cpu(cpu) {
2193                 if (iter->buffer_iter[cpu])
2194                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2195         }
2196         free_cpumask_var(iter->started);
2197         tracing_start();
2198  fail:
2199         mutex_unlock(&trace_types_lock);
2200         kfree(iter->trace);
2201         kfree(iter);
2202
2203         return fail_ret;
2204 }
2205
2206 int tracing_open_generic(struct inode *inode, struct file *filp)
2207 {
2208         if (tracing_disabled)
2209                 return -ENODEV;
2210
2211         filp->private_data = inode->i_private;
2212         return 0;
2213 }
2214
2215 static int tracing_release(struct inode *inode, struct file *file)
2216 {
2217         struct seq_file *m = file->private_data;
2218         struct trace_iterator *iter;
2219         int cpu;
2220
2221         if (!(file->f_mode & FMODE_READ))
2222                 return 0;
2223
2224         iter = m->private;
2225
2226         mutex_lock(&trace_types_lock);
2227         for_each_tracing_cpu(cpu) {
2228                 if (iter->buffer_iter[cpu])
2229                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2230         }
2231
2232         if (iter->trace && iter->trace->close)
2233                 iter->trace->close(iter);
2234
2235         /* reenable tracing if it was previously enabled */
2236         tracing_start();
2237         mutex_unlock(&trace_types_lock);
2238
2239         seq_release(inode, file);
2240         mutex_destroy(&iter->mutex);
2241         free_cpumask_var(iter->started);
2242         kfree(iter->trace);
2243         kfree(iter);
2244         return 0;
2245 }
2246
2247 static int tracing_open(struct inode *inode, struct file *file)
2248 {
2249         struct trace_iterator *iter;
2250         int ret = 0;
2251
2252         /* If this file was open for write, then erase contents */
2253         if ((file->f_mode & FMODE_WRITE) &&
2254             (file->f_flags & O_TRUNC)) {
2255                 long cpu = (long) inode->i_private;
2256
2257                 if (cpu == TRACE_PIPE_ALL_CPU)
2258                         tracing_reset_online_cpus(&global_trace);
2259                 else
2260                         tracing_reset(&global_trace, cpu);
2261         }
2262
2263         if (file->f_mode & FMODE_READ) {
2264                 iter = __tracing_open(inode, file);
2265                 if (IS_ERR(iter))
2266                         ret = PTR_ERR(iter);
2267                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2268                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2269         }
2270         return ret;
2271 }
2272
2273 static void *
2274 t_next(struct seq_file *m, void *v, loff_t *pos)
2275 {
2276         struct tracer *t = v;
2277
2278         (*pos)++;
2279
2280         if (t)
2281                 t = t->next;
2282
2283         return t;
2284 }
2285
2286 static void *t_start(struct seq_file *m, loff_t *pos)
2287 {
2288         struct tracer *t;
2289         loff_t l = 0;
2290
2291         mutex_lock(&trace_types_lock);
2292         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2293                 ;
2294
2295         return t;
2296 }
2297
2298 static void t_stop(struct seq_file *m, void *p)
2299 {
2300         mutex_unlock(&trace_types_lock);
2301 }
2302
2303 static int t_show(struct seq_file *m, void *v)
2304 {
2305         struct tracer *t = v;
2306
2307         if (!t)
2308                 return 0;
2309
2310         seq_printf(m, "%s", t->name);
2311         if (t->next)
2312                 seq_putc(m, ' ');
2313         else
2314                 seq_putc(m, '\n');
2315
2316         return 0;
2317 }
2318
2319 static const struct seq_operations show_traces_seq_ops = {
2320         .start          = t_start,
2321         .next           = t_next,
2322         .stop           = t_stop,
2323         .show           = t_show,
2324 };
2325
2326 static int show_traces_open(struct inode *inode, struct file *file)
2327 {
2328         if (tracing_disabled)
2329                 return -ENODEV;
2330
2331         return seq_open(file, &show_traces_seq_ops);
2332 }
2333
2334 static ssize_t
2335 tracing_write_stub(struct file *filp, const char __user *ubuf,
2336                    size_t count, loff_t *ppos)
2337 {
2338         return count;
2339 }
2340
2341 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2342 {
2343         if (file->f_mode & FMODE_READ)
2344                 return seq_lseek(file, offset, origin);
2345         else
2346                 return 0;
2347 }
2348
2349 static const struct file_operations tracing_fops = {
2350         .open           = tracing_open,
2351         .read           = seq_read,
2352         .write          = tracing_write_stub,
2353         .llseek         = tracing_seek,
2354         .release        = tracing_release,
2355 };
2356
2357 static const struct file_operations show_traces_fops = {
2358         .open           = show_traces_open,
2359         .read           = seq_read,
2360         .release        = seq_release,
2361         .llseek         = seq_lseek,
2362 };
2363
2364 /*
2365  * Only trace on a CPU if the bitmask is set:
2366  */
2367 static cpumask_var_t tracing_cpumask;
2368
2369 /*
2370  * The tracer itself will not take this lock, but still we want
2371  * to provide a consistent cpumask to user-space:
2372  */
2373 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2374
2375 /*
2376  * Temporary storage for the character representation of the
2377  * CPU bitmask (and one more byte for the newline):
2378  */
2379 static char mask_str[NR_CPUS + 1];
2380
2381 static ssize_t
2382 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2383                      size_t count, loff_t *ppos)
2384 {
2385         int len;
2386
2387         mutex_lock(&tracing_cpumask_update_lock);
2388
2389         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2390         if (count - len < 2) {
2391                 count = -EINVAL;
2392                 goto out_err;
2393         }
2394         len += sprintf(mask_str + len, "\n");
2395         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2396
2397 out_err:
2398         mutex_unlock(&tracing_cpumask_update_lock);
2399
2400         return count;
2401 }
2402
2403 static ssize_t
2404 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2405                       size_t count, loff_t *ppos)
2406 {
2407         int err, cpu;
2408         cpumask_var_t tracing_cpumask_new;
2409
2410         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2411                 return -ENOMEM;
2412
2413         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2414         if (err)
2415                 goto err_unlock;
2416
2417         mutex_lock(&tracing_cpumask_update_lock);
2418
2419         local_irq_disable();
2420         arch_spin_lock(&ftrace_max_lock);
2421         for_each_tracing_cpu(cpu) {
2422                 /*
2423                  * Increase/decrease the disabled counter if we are
2424                  * about to flip a bit in the cpumask:
2425                  */
2426                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2427                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2428                         atomic_inc(&global_trace.data[cpu]->disabled);
2429                 }
2430                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2431                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2432                         atomic_dec(&global_trace.data[cpu]->disabled);
2433                 }
2434         }
2435         arch_spin_unlock(&ftrace_max_lock);
2436         local_irq_enable();
2437
2438         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2439
2440         mutex_unlock(&tracing_cpumask_update_lock);
2441         free_cpumask_var(tracing_cpumask_new);
2442
2443         return count;
2444
2445 err_unlock:
2446         free_cpumask_var(tracing_cpumask_new);
2447
2448         return err;
2449 }
2450
2451 static const struct file_operations tracing_cpumask_fops = {
2452         .open           = tracing_open_generic,
2453         .read           = tracing_cpumask_read,
2454         .write          = tracing_cpumask_write,
2455         .llseek         = generic_file_llseek,
2456 };
2457
2458 static int tracing_trace_options_show(struct seq_file *m, void *v)
2459 {
2460         struct tracer_opt *trace_opts;
2461         u32 tracer_flags;
2462         int i;
2463
2464         mutex_lock(&trace_types_lock);
2465         tracer_flags = current_trace->flags->val;
2466         trace_opts = current_trace->flags->opts;
2467
2468         for (i = 0; trace_options[i]; i++) {
2469                 if (trace_flags & (1 << i))
2470                         seq_printf(m, "%s\n", trace_options[i]);
2471                 else
2472                         seq_printf(m, "no%s\n", trace_options[i]);
2473         }
2474
2475         for (i = 0; trace_opts[i].name; i++) {
2476                 if (tracer_flags & trace_opts[i].bit)
2477                         seq_printf(m, "%s\n", trace_opts[i].name);
2478                 else
2479                         seq_printf(m, "no%s\n", trace_opts[i].name);
2480         }
2481         mutex_unlock(&trace_types_lock);
2482
2483         return 0;
2484 }
2485
2486 static int __set_tracer_option(struct tracer *trace,
2487                                struct tracer_flags *tracer_flags,
2488                                struct tracer_opt *opts, int neg)
2489 {
2490         int ret;
2491
2492         ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2493         if (ret)
2494                 return ret;
2495
2496         if (neg)
2497                 tracer_flags->val &= ~opts->bit;
2498         else
2499                 tracer_flags->val |= opts->bit;
2500         return 0;
2501 }
2502
2503 /* Try to assign a tracer specific option */
2504 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2505 {
2506         struct tracer_flags *tracer_flags = trace->flags;
2507         struct tracer_opt *opts = NULL;
2508         int i;
2509
2510         for (i = 0; tracer_flags->opts[i].name; i++) {
2511                 opts = &tracer_flags->opts[i];
2512
2513                 if (strcmp(cmp, opts->name) == 0)
2514                         return __set_tracer_option(trace, trace->flags,
2515                                                    opts, neg);
2516         }
2517
2518         return -EINVAL;
2519 }
2520
2521 static void set_tracer_flags(unsigned int mask, int enabled)
2522 {
2523         /* do nothing if flag is already set */
2524         if (!!(trace_flags & mask) == !!enabled)
2525                 return;
2526
2527         if (enabled)
2528                 trace_flags |= mask;
2529         else
2530                 trace_flags &= ~mask;
2531
2532         if (mask == TRACE_ITER_RECORD_CMD)
2533                 trace_event_enable_cmd_record(enabled);
2534 }
2535
2536 static ssize_t
2537 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2538                         size_t cnt, loff_t *ppos)
2539 {
2540         char buf[64];
2541         char *cmp;
2542         int neg = 0;
2543         int ret;
2544         int i;
2545
2546         if (cnt >= sizeof(buf))
2547                 return -EINVAL;
2548
2549         if (copy_from_user(&buf, ubuf, cnt))
2550                 return -EFAULT;
2551
2552         buf[cnt] = 0;
2553         cmp = strstrip(buf);
2554
2555         if (strncmp(cmp, "no", 2) == 0) {
2556                 neg = 1;
2557                 cmp += 2;
2558         }
2559
2560         for (i = 0; trace_options[i]; i++) {
2561                 if (strcmp(cmp, trace_options[i]) == 0) {
2562                         set_tracer_flags(1 << i, !neg);
2563                         break;
2564                 }
2565         }
2566
2567         /* If no option could be set, test the specific tracer options */
2568         if (!trace_options[i]) {
2569                 mutex_lock(&trace_types_lock);
2570                 ret = set_tracer_option(current_trace, cmp, neg);
2571                 mutex_unlock(&trace_types_lock);
2572                 if (ret)
2573                         return ret;
2574         }
2575
2576         *ppos += cnt;
2577
2578         return cnt;
2579 }
2580
2581 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2582 {
2583         if (tracing_disabled)
2584                 return -ENODEV;
2585         return single_open(file, tracing_trace_options_show, NULL);
2586 }
2587
2588 static const struct file_operations tracing_iter_fops = {
2589         .open           = tracing_trace_options_open,
2590         .read           = seq_read,
2591         .llseek         = seq_lseek,
2592         .release        = single_release,
2593         .write          = tracing_trace_options_write,
2594 };
2595
2596 static const char readme_msg[] =
2597         "tracing mini-HOWTO:\n\n"
2598         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2599         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2600         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2601         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2602         "nop\n"
2603         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2604         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2605         "sched_switch\n"
2606         "# cat /sys/kernel/debug/tracing/trace_options\n"
2607         "noprint-parent nosym-offset nosym-addr noverbose\n"
2608         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2609         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2610         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2611         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2612 ;
2613
2614 static ssize_t
2615 tracing_readme_read(struct file *filp, char __user *ubuf,
2616                        size_t cnt, loff_t *ppos)
2617 {
2618         return simple_read_from_buffer(ubuf, cnt, ppos,
2619                                         readme_msg, strlen(readme_msg));
2620 }
2621
2622 static const struct file_operations tracing_readme_fops = {
2623         .open           = tracing_open_generic,
2624         .read           = tracing_readme_read,
2625         .llseek         = generic_file_llseek,
2626 };
2627
2628 static ssize_t
2629 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2630                                 size_t cnt, loff_t *ppos)
2631 {
2632         char *buf_comm;
2633         char *file_buf;
2634         char *buf;
2635         int len = 0;
2636         int pid;
2637         int i;
2638
2639         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2640         if (!file_buf)
2641                 return -ENOMEM;
2642
2643         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2644         if (!buf_comm) {
2645                 kfree(file_buf);
2646                 return -ENOMEM;
2647         }
2648
2649         buf = file_buf;
2650
2651         for (i = 0; i < SAVED_CMDLINES; i++) {
2652                 int r;
2653
2654                 pid = map_cmdline_to_pid[i];
2655                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2656                         continue;
2657
2658                 trace_find_cmdline(pid, buf_comm);
2659                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2660                 buf += r;
2661                 len += r;
2662         }
2663
2664         len = simple_read_from_buffer(ubuf, cnt, ppos,
2665                                       file_buf, len);
2666
2667         kfree(file_buf);
2668         kfree(buf_comm);
2669
2670         return len;
2671 }
2672
2673 static const struct file_operations tracing_saved_cmdlines_fops = {
2674     .open       = tracing_open_generic,
2675     .read       = tracing_saved_cmdlines_read,
2676     .llseek     = generic_file_llseek,
2677 };
2678
2679 static ssize_t
2680 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2681                   size_t cnt, loff_t *ppos)
2682 {
2683         char buf[64];
2684         int r;
2685
2686         r = sprintf(buf, "%u\n", tracer_enabled);
2687         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2688 }
2689
2690 static ssize_t
2691 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2692                    size_t cnt, loff_t *ppos)
2693 {
2694         struct trace_array *tr = filp->private_data;
2695         char buf[64];
2696         unsigned long val;
2697         int ret;
2698
2699         if (cnt >= sizeof(buf))
2700                 return -EINVAL;
2701
2702         if (copy_from_user(&buf, ubuf, cnt))
2703                 return -EFAULT;
2704
2705         buf[cnt] = 0;
2706
2707         ret = strict_strtoul(buf, 10, &val);
2708         if (ret < 0)
2709                 return ret;
2710
2711         val = !!val;
2712
2713         mutex_lock(&trace_types_lock);
2714         if (tracer_enabled ^ val) {
2715                 if (val) {
2716                         tracer_enabled = 1;
2717                         if (current_trace->start)
2718                                 current_trace->start(tr);
2719                         tracing_start();
2720                 } else {
2721                         tracer_enabled = 0;
2722                         tracing_stop();
2723                         if (current_trace->stop)
2724                                 current_trace->stop(tr);
2725                 }
2726         }
2727         mutex_unlock(&trace_types_lock);
2728
2729         *ppos += cnt;
2730
2731         return cnt;
2732 }
2733
2734 static ssize_t
2735 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2736                        size_t cnt, loff_t *ppos)
2737 {
2738         char buf[MAX_TRACER_SIZE+2];
2739         int r;
2740
2741         mutex_lock(&trace_types_lock);
2742         if (current_trace)
2743                 r = sprintf(buf, "%s\n", current_trace->name);
2744         else
2745                 r = sprintf(buf, "\n");
2746         mutex_unlock(&trace_types_lock);
2747
2748         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2749 }
2750
2751 int tracer_init(struct tracer *t, struct trace_array *tr)
2752 {
2753         tracing_reset_online_cpus(tr);
2754         return t->init(tr);
2755 }
2756
2757 static int tracing_resize_ring_buffer(unsigned long size)
2758 {
2759         int ret;
2760
2761         /*
2762          * If kernel or user changes the size of the ring buffer
2763          * we use the size that was given, and we can forget about
2764          * expanding it later.
2765          */
2766         ring_buffer_expanded = 1;
2767
2768         ret = ring_buffer_resize(global_trace.buffer, size);
2769         if (ret < 0)
2770                 return ret;
2771
2772         if (!current_trace->use_max_tr)
2773                 goto out;
2774
2775         ret = ring_buffer_resize(max_tr.buffer, size);
2776         if (ret < 0) {
2777                 int r;
2778
2779                 r = ring_buffer_resize(global_trace.buffer,
2780                                        global_trace.entries);
2781                 if (r < 0) {
2782                         /*
2783                          * AARGH! We are left with different
2784                          * size max buffer!!!!
2785                          * The max buffer is our "snapshot" buffer.
2786                          * When a tracer needs a snapshot (one of the
2787                          * latency tracers), it swaps the max buffer
2788                          * with the saved snap shot. We succeeded to
2789                          * update the size of the main buffer, but failed to
2790                          * update the size of the max buffer. But when we tried
2791                          * to reset the main buffer to the original size, we
2792                          * failed there too. This is very unlikely to
2793                          * happen, but if it does, warn and kill all
2794                          * tracing.
2795                          */
2796                         WARN_ON(1);
2797                         tracing_disabled = 1;
2798                 }
2799                 return ret;
2800         }
2801
2802         max_tr.entries = size;
2803  out:
2804         global_trace.entries = size;
2805
2806         return ret;
2807 }
2808
2809
2810 /**
2811  * tracing_update_buffers - used by tracing facility to expand ring buffers
2812  *
2813  * To save on memory when the tracing is never used on a system with it
2814  * configured in. The ring buffers are set to a minimum size. But once
2815  * a user starts to use the tracing facility, then they need to grow
2816  * to their default size.
2817  *
2818  * This function is to be called when a tracer is about to be used.
2819  */
2820 int tracing_update_buffers(void)
2821 {
2822         int ret = 0;
2823
2824         mutex_lock(&trace_types_lock);
2825         if (!ring_buffer_expanded)
2826                 ret = tracing_resize_ring_buffer(trace_buf_size);
2827         mutex_unlock(&trace_types_lock);
2828
2829         return ret;
2830 }
2831
2832 struct trace_option_dentry;
2833
2834 static struct trace_option_dentry *
2835 create_trace_option_files(struct tracer *tracer);
2836
2837 static void
2838 destroy_trace_option_files(struct trace_option_dentry *topts);
2839
2840 static int tracing_set_tracer(const char *buf)
2841 {
2842         static struct trace_option_dentry *topts;
2843         struct trace_array *tr = &global_trace;
2844         struct tracer *t;
2845         int ret = 0;
2846
2847         mutex_lock(&trace_types_lock);
2848
2849         if (!ring_buffer_expanded) {
2850                 ret = tracing_resize_ring_buffer(trace_buf_size);
2851                 if (ret < 0)
2852                         goto out;
2853                 ret = 0;
2854         }
2855
2856         for (t = trace_types; t; t = t->next) {
2857                 if (strcmp(t->name, buf) == 0)
2858                         break;
2859         }
2860         if (!t) {
2861                 ret = -EINVAL;
2862                 goto out;
2863         }
2864         if (t == current_trace)
2865                 goto out;
2866
2867         trace_branch_disable();
2868         if (current_trace && current_trace->reset)
2869                 current_trace->reset(tr);
2870         if (current_trace && current_trace->use_max_tr) {
2871                 /*
2872                  * We don't free the ring buffer. instead, resize it because
2873                  * The max_tr ring buffer has some state (e.g. ring->clock) and
2874                  * we want preserve it.
2875                  */
2876                 ring_buffer_resize(max_tr.buffer, 1);
2877                 max_tr.entries = 1;
2878         }
2879         destroy_trace_option_files(topts);
2880
2881         current_trace = t;
2882
2883         topts = create_trace_option_files(current_trace);
2884         if (current_trace->use_max_tr) {
2885                 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
2886                 if (ret < 0)
2887                         goto out;
2888                 max_tr.entries = global_trace.entries;
2889         }
2890
2891         if (t->init) {
2892                 ret = tracer_init(t, tr);
2893                 if (ret)
2894                         goto out;
2895         }
2896
2897         trace_branch_enable(tr);
2898  out:
2899         mutex_unlock(&trace_types_lock);
2900
2901         return ret;
2902 }
2903
2904 static ssize_t
2905 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2906                         size_t cnt, loff_t *ppos)
2907 {
2908         char buf[MAX_TRACER_SIZE+1];
2909         int i;
2910         size_t ret;
2911         int err;
2912
2913         ret = cnt;
2914
2915         if (cnt > MAX_TRACER_SIZE)
2916                 cnt = MAX_TRACER_SIZE;
2917
2918         if (copy_from_user(&buf, ubuf, cnt))
2919                 return -EFAULT;
2920
2921         buf[cnt] = 0;
2922
2923         /* strip ending whitespace. */
2924         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2925                 buf[i] = 0;
2926
2927         err = tracing_set_tracer(buf);
2928         if (err)
2929                 return err;
2930
2931         *ppos += ret;
2932
2933         return ret;
2934 }
2935
2936 static ssize_t
2937 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2938                      size_t cnt, loff_t *ppos)
2939 {
2940         unsigned long *ptr = filp->private_data;
2941         char buf[64];
2942         int r;
2943
2944         r = snprintf(buf, sizeof(buf), "%ld\n",
2945                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2946         if (r > sizeof(buf))
2947                 r = sizeof(buf);
2948         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2949 }
2950
2951 static ssize_t
2952 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2953                       size_t cnt, loff_t *ppos)
2954 {
2955         unsigned long *ptr = filp->private_data;
2956         char buf[64];
2957         unsigned long val;
2958         int ret;
2959
2960         if (cnt >= sizeof(buf))
2961                 return -EINVAL;
2962
2963         if (copy_from_user(&buf, ubuf, cnt))
2964                 return -EFAULT;
2965
2966         buf[cnt] = 0;
2967
2968         ret = strict_strtoul(buf, 10, &val);
2969         if (ret < 0)
2970                 return ret;
2971
2972         *ptr = val * 1000;
2973
2974         return cnt;
2975 }
2976
2977 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2978 {
2979         long cpu_file = (long) inode->i_private;
2980         struct trace_iterator *iter;
2981         int ret = 0;
2982
2983         if (tracing_disabled)
2984                 return -ENODEV;
2985
2986         mutex_lock(&trace_types_lock);
2987
2988         /* create a buffer to store the information to pass to userspace */
2989         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2990         if (!iter) {
2991                 ret = -ENOMEM;
2992                 goto out;
2993         }
2994
2995         /*
2996          * We make a copy of the current tracer to avoid concurrent
2997          * changes on it while we are reading.
2998          */
2999         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3000         if (!iter->trace) {
3001                 ret = -ENOMEM;
3002                 goto fail;
3003         }
3004         if (current_trace)
3005                 *iter->trace = *current_trace;
3006
3007         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3008                 ret = -ENOMEM;
3009                 goto fail;
3010         }
3011
3012         /* trace pipe does not show start of buffer */
3013         cpumask_setall(iter->started);
3014
3015         if (trace_flags & TRACE_ITER_LATENCY_FMT)
3016                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3017
3018         iter->cpu_file = cpu_file;
3019         iter->tr = &global_trace;
3020         mutex_init(&iter->mutex);
3021         filp->private_data = iter;
3022
3023         if (iter->trace->pipe_open)
3024                 iter->trace->pipe_open(iter);
3025
3026         nonseekable_open(inode, filp);
3027 out:
3028         mutex_unlock(&trace_types_lock);
3029         return ret;
3030
3031 fail:
3032         kfree(iter->trace);
3033         kfree(iter);
3034         mutex_unlock(&trace_types_lock);
3035         return ret;
3036 }
3037
3038 static int tracing_release_pipe(struct inode *inode, struct file *file)
3039 {
3040         struct trace_iterator *iter = file->private_data;
3041
3042         mutex_lock(&trace_types_lock);
3043
3044         if (iter->trace->pipe_close)
3045                 iter->trace->pipe_close(iter);
3046
3047         mutex_unlock(&trace_types_lock);
3048
3049         free_cpumask_var(iter->started);
3050         mutex_destroy(&iter->mutex);
3051         kfree(iter->trace);
3052         kfree(iter);
3053
3054         return 0;
3055 }
3056
3057 static unsigned int
3058 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3059 {
3060         struct trace_iterator *iter = filp->private_data;
3061
3062         if (trace_flags & TRACE_ITER_BLOCK) {
3063                 /*
3064                  * Always select as readable when in blocking mode
3065                  */
3066                 return POLLIN | POLLRDNORM;
3067         } else {
3068                 if (!trace_empty(iter))
3069                         return POLLIN | POLLRDNORM;
3070                 poll_wait(filp, &trace_wait, poll_table);
3071                 if (!trace_empty(iter))
3072                         return POLLIN | POLLRDNORM;
3073
3074                 return 0;
3075         }
3076 }
3077
3078
3079 void default_wait_pipe(struct trace_iterator *iter)
3080 {
3081         DEFINE_WAIT(wait);
3082
3083         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3084
3085         if (trace_empty(iter))
3086                 schedule();
3087
3088         finish_wait(&trace_wait, &wait);
3089 }
3090
3091 /*
3092  * This is a make-shift waitqueue.
3093  * A tracer might use this callback on some rare cases:
3094  *
3095  *  1) the current tracer might hold the runqueue lock when it wakes up
3096  *     a reader, hence a deadlock (sched, function, and function graph tracers)
3097  *  2) the function tracers, trace all functions, we don't want
3098  *     the overhead of calling wake_up and friends
3099  *     (and tracing them too)
3100  *
3101  *     Anyway, this is really very primitive wakeup.
3102  */
3103 void poll_wait_pipe(struct trace_iterator *iter)
3104 {
3105         set_current_state(TASK_INTERRUPTIBLE);
3106         /* sleep for 100 msecs, and try again. */
3107         schedule_timeout(HZ / 10);
3108 }
3109
3110 /* Must be called with trace_types_lock mutex held. */
3111 static int tracing_wait_pipe(struct file *filp)
3112 {
3113         struct trace_iterator *iter = filp->private_data;
3114
3115         while (trace_empty(iter)) {
3116
3117                 if ((filp->f_flags & O_NONBLOCK)) {
3118                         return -EAGAIN;
3119                 }
3120
3121                 mutex_unlock(&iter->mutex);
3122
3123                 iter->trace->wait_pipe(iter);
3124
3125                 mutex_lock(&iter->mutex);
3126
3127                 if (signal_pending(current))
3128                         return -EINTR;
3129
3130                 /*
3131                  * We block until we read something and tracing is disabled.
3132                  * We still block if tracing is disabled, but we have never
3133                  * read anything. This allows a user to cat this file, and
3134                  * then enable tracing. But after we have read something,
3135                  * we give an EOF when tracing is again disabled.
3136                  *
3137                  * iter->pos will be 0 if we haven't read anything.
3138                  */
3139                 if (!tracer_enabled && iter->pos)
3140                         break;
3141         }
3142
3143         return 1;
3144 }
3145
3146 /*
3147  * Consumer reader.
3148  */
3149 static ssize_t
3150 tracing_read_pipe(struct file *filp, char __user *ubuf,
3151                   size_t cnt, loff_t *ppos)
3152 {
3153         struct trace_iterator *iter = filp->private_data;
3154         static struct tracer *old_tracer;
3155         ssize_t sret;
3156
3157         /* return any leftover data */
3158         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3159         if (sret != -EBUSY)
3160                 return sret;
3161
3162         trace_seq_init(&iter->seq);
3163
3164         /* copy the tracer to avoid using a global lock all around */
3165         mutex_lock(&trace_types_lock);
3166         if (unlikely(old_tracer != current_trace && current_trace)) {
3167                 old_tracer = current_trace;
3168                 *iter->trace = *current_trace;
3169         }
3170         mutex_unlock(&trace_types_lock);
3171
3172         /*
3173          * Avoid more than one consumer on a single file descriptor
3174          * This is just a matter of traces coherency, the ring buffer itself
3175          * is protected.
3176          */
3177         mutex_lock(&iter->mutex);
3178         if (iter->trace->read) {
3179                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3180                 if (sret)
3181                         goto out;
3182         }
3183
3184 waitagain:
3185         sret = tracing_wait_pipe(filp);
3186         if (sret <= 0)
3187                 goto out;
3188
3189         /* stop when tracing is finished */
3190         if (trace_empty(iter)) {
3191                 sret = 0;
3192                 goto out;
3193         }
3194
3195         if (cnt >= PAGE_SIZE)
3196                 cnt = PAGE_SIZE - 1;
3197
3198         /* reset all but tr, trace, and overruns */
3199         memset(&iter->seq, 0,
3200                sizeof(struct trace_iterator) -
3201                offsetof(struct trace_iterator, seq));
3202         iter->pos = -1;
3203
3204         trace_event_read_lock();
3205         trace_access_lock(iter->cpu_file);
3206         while (trace_find_next_entry_inc(iter) != NULL) {
3207                 enum print_line_t ret;
3208                 int len = iter->seq.len;
3209
3210                 ret = print_trace_line(iter);
3211                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3212                         /* don't print partial lines */
3213                         iter->seq.len = len;
3214                         break;
3215                 }
3216                 if (ret != TRACE_TYPE_NO_CONSUME)
3217                         trace_consume(iter);
3218
3219                 if (iter->seq.len >= cnt)
3220                         break;
3221         }
3222         trace_access_unlock(iter->cpu_file);
3223         trace_event_read_unlock();
3224
3225         /* Now copy what we have to the user */
3226         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3227         if (iter->seq.readpos >= iter->seq.len)
3228                 trace_seq_init(&iter->seq);
3229
3230         /*
3231          * If there was nothing to send to user, inspite of consuming trace
3232          * entries, go back to wait for more entries.
3233          */
3234         if (sret == -EBUSY)
3235                 goto waitagain;
3236
3237 out:
3238         mutex_unlock(&iter->mutex);
3239
3240         return sret;
3241 }
3242
3243 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3244                                      struct pipe_buffer *buf)
3245 {
3246         __free_page(buf->page);
3247 }
3248
3249 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3250                                      unsigned int idx)
3251 {
3252         __free_page(spd->pages[idx]);
3253 }
3254
3255 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3256         .can_merge              = 0,
3257         .map                    = generic_pipe_buf_map,
3258         .unmap                  = generic_pipe_buf_unmap,
3259         .confirm                = generic_pipe_buf_confirm,
3260         .release                = tracing_pipe_buf_release,
3261         .steal                  = generic_pipe_buf_steal,
3262         .get                    = generic_pipe_buf_get,
3263 };
3264
3265 static size_t
3266 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3267 {
3268         size_t count;
3269         int ret;
3270
3271         /* Seq buffer is page-sized, exactly what we need. */
3272         for (;;) {
3273                 count = iter->seq.len;
3274                 ret = print_trace_line(iter);
3275                 count = iter->seq.len - count;
3276                 if (rem < count) {
3277                         rem = 0;
3278                         iter->seq.len -= count;
3279                         break;
3280                 }
3281                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3282                         iter->seq.len -= count;
3283                         break;
3284                 }
3285
3286                 if (ret != TRACE_TYPE_NO_CONSUME)
3287                         trace_consume(iter);
3288                 rem -= count;
3289                 if (!trace_find_next_entry_inc(iter))   {
3290                         rem = 0;
3291                         iter->ent = NULL;
3292                         break;
3293                 }
3294         }
3295
3296         return rem;
3297 }
3298
3299 static ssize_t tracing_splice_read_pipe(struct file *filp,
3300                                         loff_t *ppos,
3301                                         struct pipe_inode_info *pipe,
3302                                         size_t len,
3303                                         unsigned int flags)
3304 {
3305         struct page *pages_def[PIPE_DEF_BUFFERS];
3306         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3307         struct trace_iterator *iter = filp->private_data;
3308         struct splice_pipe_desc spd = {
3309                 .pages          = pages_def,
3310                 .partial        = partial_def,
3311                 .nr_pages       = 0, /* This gets updated below. */
3312                 .flags          = flags,
3313                 .ops            = &tracing_pipe_buf_ops,
3314                 .spd_release    = tracing_spd_release_pipe,
3315         };
3316         static struct tracer *old_tracer;
3317         ssize_t ret;
3318         size_t rem;
3319         unsigned int i;
3320
3321         if (splice_grow_spd(pipe, &spd))
3322                 return -ENOMEM;
3323
3324         /* copy the tracer to avoid using a global lock all around */
3325         mutex_lock(&trace_types_lock);
3326         if (unlikely(old_tracer != current_trace && current_trace)) {
3327                 old_tracer = current_trace;
3328                 *iter->trace = *current_trace;
3329         }
3330         mutex_unlock(&trace_types_lock);
3331
3332         mutex_lock(&iter->mutex);
3333
3334         if (iter->trace->splice_read) {
3335                 ret = iter->trace->splice_read(iter, filp,
3336                                                ppos, pipe, len, flags);
3337                 if (ret)
3338                         goto out_err;
3339         }
3340
3341         ret = tracing_wait_pipe(filp);
3342         if (ret <= 0)
3343                 goto out_err;
3344
3345         if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3346                 ret = -EFAULT;
3347                 goto out_err;
3348         }
3349
3350         trace_event_read_lock();
3351         trace_access_lock(iter->cpu_file);
3352
3353         /* Fill as many pages as possible. */
3354         for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3355                 spd.pages[i] = alloc_page(GFP_KERNEL);
3356                 if (!spd.pages[i])
3357                         break;
3358
3359                 rem = tracing_fill_pipe_page(rem, iter);
3360
3361                 /* Copy the data into the page, so we can start over. */
3362                 ret = trace_seq_to_buffer(&iter->seq,
3363                                           page_address(spd.pages[i]),
3364                                           iter->seq.len);
3365                 if (ret < 0) {
3366                         __free_page(spd.pages[i]);
3367                         break;
3368                 }
3369                 spd.partial[i].offset = 0;
3370                 spd.partial[i].len = iter->seq.len;
3371
3372                 trace_seq_init(&iter->seq);
3373         }
3374
3375         trace_access_unlock(iter->cpu_file);
3376         trace_event_read_unlock();
3377         mutex_unlock(&iter->mutex);
3378
3379         spd.nr_pages = i;
3380
3381         ret = splice_to_pipe(pipe, &spd);
3382 out:
3383         splice_shrink_spd(pipe, &spd);
3384         return ret;
3385
3386 out_err:
3387         mutex_unlock(&iter->mutex);
3388         goto out;
3389 }
3390
3391 static ssize_t
3392 tracing_entries_read(struct file *filp, char __user *ubuf,
3393                      size_t cnt, loff_t *ppos)
3394 {
3395         struct trace_array *tr = filp->private_data;
3396         char buf[96];
3397         int r;
3398
3399         mutex_lock(&trace_types_lock);
3400         if (!ring_buffer_expanded)
3401                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3402                             tr->entries >> 10,
3403                             trace_buf_size >> 10);
3404         else
3405                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3406         mutex_unlock(&trace_types_lock);
3407
3408         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3409 }
3410
3411 static ssize_t
3412 tracing_entries_write(struct file *filp, const char __user *ubuf,
3413                       size_t cnt, loff_t *ppos)
3414 {
3415         unsigned long val;
3416         char buf[64];
3417         int ret, cpu;
3418
3419         if (cnt >= sizeof(buf))
3420                 return -EINVAL;
3421
3422         if (copy_from_user(&buf, ubuf, cnt))
3423                 return -EFAULT;
3424
3425         buf[cnt] = 0;
3426
3427         ret = strict_strtoul(buf, 10, &val);
3428         if (ret < 0)
3429                 return ret;
3430
3431         /* must have at least 1 entry */
3432         if (!val)
3433                 return -EINVAL;
3434
3435         mutex_lock(&trace_types_lock);
3436
3437         tracing_stop();
3438
3439         /* disable all cpu buffers */
3440         for_each_tracing_cpu(cpu) {
3441                 if (global_trace.data[cpu])
3442                         atomic_inc(&global_trace.data[cpu]->disabled);
3443                 if (max_tr.data[cpu])
3444                         atomic_inc(&max_tr.data[cpu]->disabled);
3445         }
3446
3447         /* value is in KB */
3448         val <<= 10;
3449
3450         if (val != global_trace.entries) {
3451                 ret = tracing_resize_ring_buffer(val);
3452                 if (ret < 0) {
3453                         cnt = ret;
3454                         goto out;
3455                 }
3456         }
3457
3458         *ppos += cnt;
3459
3460         /* If check pages failed, return ENOMEM */
3461         if (tracing_disabled)
3462                 cnt = -ENOMEM;
3463  out:
3464         for_each_tracing_cpu(cpu) {
3465                 if (global_trace.data[cpu])
3466                         atomic_dec(&global_trace.data[cpu]->disabled);
3467                 if (max_tr.data[cpu])
3468                         atomic_dec(&max_tr.data[cpu]->disabled);
3469         }
3470
3471         tracing_start();
3472         mutex_unlock(&trace_types_lock);
3473
3474         return cnt;
3475 }
3476
3477 static int mark_printk(const char *fmt, ...)
3478 {
3479         int ret;
3480         va_list args;
3481         va_start(args, fmt);
3482         ret = trace_vprintk(0, fmt, args);
3483         va_end(args);
3484         return ret;
3485 }
3486
3487 static ssize_t
3488 tracing_mark_write(struct file *filp, const char __user *ubuf,
3489                                         size_t cnt, loff_t *fpos)
3490 {
3491         char *buf;
3492         size_t written;
3493
3494         if (tracing_disabled)
3495                 return -EINVAL;
3496
3497         if (cnt > TRACE_BUF_SIZE)
3498                 cnt = TRACE_BUF_SIZE;
3499
3500         buf = kmalloc(cnt + 2, GFP_KERNEL);
3501         if (buf == NULL)
3502                 return -ENOMEM;
3503
3504         if (copy_from_user(buf, ubuf, cnt)) {
3505                 kfree(buf);
3506                 return -EFAULT;
3507         }
3508         if (buf[cnt-1] != '\n') {
3509                 buf[cnt] = '\n';
3510                 buf[cnt+1] = '\0';
3511         } else
3512                 buf[cnt] = '\0';
3513
3514         written = mark_printk("%s", buf);
3515         kfree(buf);
3516         *fpos += written;
3517
3518         /* don't tell userspace we wrote more - it might confuse them */
3519         if (written > cnt)
3520                 written = cnt;
3521
3522         return written;
3523 }
3524
3525 static int tracing_clock_show(struct seq_file *m, void *v)
3526 {
3527         int i;
3528
3529         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3530                 seq_printf(m,
3531                         "%s%s%s%s", i ? " " : "",
3532                         i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3533                         i == trace_clock_id ? "]" : "");
3534         seq_putc(m, '\n');
3535
3536         return 0;
3537 }
3538
3539 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3540                                    size_t cnt, loff_t *fpos)
3541 {
3542         char buf[64];
3543         const char *clockstr;
3544         int i;
3545
3546         if (cnt >= sizeof(buf))
3547                 return -EINVAL;
3548
3549         if (copy_from_user(&buf, ubuf, cnt))
3550                 return -EFAULT;
3551
3552         buf[cnt] = 0;
3553
3554         clockstr = strstrip(buf);
3555
3556         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3557                 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3558                         break;
3559         }
3560         if (i == ARRAY_SIZE(trace_clocks))
3561                 return -EINVAL;
3562
3563         trace_clock_id = i;
3564
3565         mutex_lock(&trace_types_lock);
3566
3567         ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3568         if (max_tr.buffer)
3569                 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3570
3571         mutex_unlock(&trace_types_lock);
3572
3573         *fpos += cnt;
3574
3575         return cnt;
3576 }
3577
3578 static int tracing_clock_open(struct inode *inode, struct file *file)
3579 {
3580         if (tracing_disabled)
3581                 return -ENODEV;
3582         return single_open(file, tracing_clock_show, NULL);
3583 }
3584
3585 static const struct file_operations tracing_max_lat_fops = {
3586         .open           = tracing_open_generic,
3587         .read           = tracing_max_lat_read,
3588         .write          = tracing_max_lat_write,
3589         .llseek         = generic_file_llseek,
3590 };
3591
3592 static const struct file_operations tracing_ctrl_fops = {
3593         .open           = tracing_open_generic,
3594         .read           = tracing_ctrl_read,
3595         .write          = tracing_ctrl_write,
3596         .llseek         = generic_file_llseek,
3597 };
3598
3599 static const struct file_operations set_tracer_fops = {
3600         .open           = tracing_open_generic,
3601         .read           = tracing_set_trace_read,
3602         .write          = tracing_set_trace_write,
3603         .llseek         = generic_file_llseek,
3604 };
3605
3606 static const struct file_operations tracing_pipe_fops = {
3607         .open           = tracing_open_pipe,
3608         .poll           = tracing_poll_pipe,
3609         .read           = tracing_read_pipe,
3610         .splice_read    = tracing_splice_read_pipe,
3611         .release        = tracing_release_pipe,
3612         .llseek         = no_llseek,
3613 };
3614
3615 static const struct file_operations tracing_entries_fops = {
3616         .open           = tracing_open_generic,
3617         .read           = tracing_entries_read,
3618         .write          = tracing_entries_write,
3619         .llseek         = generic_file_llseek,
3620 };
3621
3622 static const struct file_operations tracing_mark_fops = {
3623         .open           = tracing_open_generic,
3624         .write          = tracing_mark_write,
3625         .llseek         = generic_file_llseek,
3626 };
3627
3628 static const struct file_operations trace_clock_fops = {
3629         .open           = tracing_clock_open,
3630         .read           = seq_read,
3631         .llseek         = seq_lseek,
3632         .release        = single_release,
3633         .write          = tracing_clock_write,
3634 };
3635
3636 struct ftrace_buffer_info {
3637         struct trace_array      *tr;
3638         void                    *spare;
3639         int                     cpu;
3640         unsigned int            read;
3641 };
3642
3643 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3644 {
3645         int cpu = (int)(long)inode->i_private;
3646         struct ftrace_buffer_info *info;
3647
3648         if (tracing_disabled)
3649                 return -ENODEV;
3650
3651         info = kzalloc(sizeof(*info), GFP_KERNEL);
3652         if (!info)
3653                 return -ENOMEM;
3654
3655         info->tr        = &global_trace;
3656         info->cpu       = cpu;
3657         info->spare     = NULL;
3658         /* Force reading ring buffer for first read */
3659         info->read      = (unsigned int)-1;
3660
3661         filp->private_data = info;
3662
3663         return nonseekable_open(inode, filp);
3664 }
3665
3666 static ssize_t
3667 tracing_buffers_read(struct file *filp, char __user *ubuf,
3668                      size_t count, loff_t *ppos)
3669 {
3670         struct ftrace_buffer_info *info = filp->private_data;
3671         ssize_t ret;
3672         size_t size;
3673
3674         if (!count)
3675                 return 0;
3676
3677         if (!info->spare)
3678                 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3679         if (!info->spare)
3680                 return -ENOMEM;
3681
3682         /* Do we have previous read data to read? */
3683         if (info->read < PAGE_SIZE)
3684                 goto read;
3685
3686         info->read = 0;
3687
3688         trace_access_lock(info->cpu);
3689         ret = ring_buffer_read_page(info->tr->buffer,
3690                                     &info->spare,
3691                                     count,
3692                                     info->cpu, 0);
3693         trace_access_unlock(info->cpu);
3694         if (ret < 0)
3695                 return 0;
3696
3697 read:
3698         size = PAGE_SIZE - info->read;
3699         if (size > count)
3700                 size = count;
3701
3702         ret = copy_to_user(ubuf, info->spare + info->read, size);
3703         if (ret == size)
3704                 return -EFAULT;
3705         size -= ret;
3706
3707         *ppos += size;
3708         info->read += size;
3709
3710         return size;
3711 }
3712
3713 static int tracing_buffers_release(struct inode *inode, struct file *file)
3714 {
3715         struct ftrace_buffer_info *info = file->private_data;
3716
3717         if (info->spare)
3718                 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3719         kfree(info);
3720
3721         return 0;
3722 }
3723
3724 struct buffer_ref {
3725         struct ring_buffer      *buffer;
3726         void                    *page;
3727         int                     ref;
3728 };
3729
3730 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3731                                     struct pipe_buffer *buf)
3732 {
3733         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3734
3735         if (--ref->ref)
3736                 return;
3737
3738         ring_buffer_free_read_page(ref->buffer, ref->page);
3739         kfree(ref);
3740         buf->private = 0;
3741 }
3742
3743 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3744                                  struct pipe_buffer *buf)
3745 {
3746         return 1;
3747 }
3748
3749 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3750                                 struct pipe_buffer *buf)
3751 {
3752         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3753
3754         ref->ref++;
3755 }
3756
3757 /* Pipe buffer operations for a buffer. */
3758 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3759         .can_merge              = 0,
3760         .map                    = generic_pipe_buf_map,
3761         .unmap                  = generic_pipe_buf_unmap,
3762         .confirm                = generic_pipe_buf_confirm,
3763         .release                = buffer_pipe_buf_release,
3764         .steal                  = buffer_pipe_buf_steal,
3765         .get                    = buffer_pipe_buf_get,
3766 };
3767
3768 /*
3769  * Callback from splice_to_pipe(), if we need to release some pages
3770  * at the end of the spd in case we error'ed out in filling the pipe.
3771  */
3772 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3773 {
3774         struct buffer_ref *ref =
3775                 (struct buffer_ref *)spd->partial[i].private;
3776
3777         if (--ref->ref)
3778                 return;
3779
3780         ring_buffer_free_read_page(ref->buffer, ref->page);
3781         kfree(ref);
3782         spd->partial[i].private = 0;
3783 }
3784
3785 static ssize_t
3786 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3787                             struct pipe_inode_info *pipe, size_t len,
3788                             unsigned int flags)
3789 {
3790         struct ftrace_buffer_info *info = file->private_data;
3791         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3792         struct page *pages_def[PIPE_DEF_BUFFERS];
3793         struct splice_pipe_desc spd = {
3794                 .pages          = pages_def,
3795                 .partial        = partial_def,
3796                 .flags          = flags,
3797                 .ops            = &buffer_pipe_buf_ops,
3798                 .spd_release    = buffer_spd_release,
3799         };
3800         struct buffer_ref *ref;
3801         int entries, size, i;
3802         size_t ret;
3803
3804         if (splice_grow_spd(pipe, &spd))
3805                 return -ENOMEM;
3806
3807         if (*ppos & (PAGE_SIZE - 1)) {
3808                 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3809                 ret = -EINVAL;
3810                 goto out;
3811         }
3812
3813         if (len & (PAGE_SIZE - 1)) {
3814                 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3815                 if (len < PAGE_SIZE) {
3816                         ret = -EINVAL;
3817                         goto out;
3818                 }
3819                 len &= PAGE_MASK;
3820         }
3821
3822         trace_access_lock(info->cpu);
3823         entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3824
3825         for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3826                 struct page *page;
3827                 int r;
3828
3829                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3830                 if (!ref)
3831                         break;
3832
3833                 ref->ref = 1;
3834                 ref->buffer = info->tr->buffer;
3835                 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3836                 if (!ref->page) {
3837                         kfree(ref);
3838                         break;
3839                 }
3840
3841                 r = ring_buffer_read_page(ref->buffer, &ref->page,
3842                                           len, info->cpu, 1);
3843                 if (r < 0) {
3844                         ring_buffer_free_read_page(ref->buffer,
3845                                                    ref->page);
3846                         kfree(ref);
3847                         break;
3848                 }
3849
3850                 /*
3851                  * zero out any left over data, this is going to
3852                  * user land.
3853                  */
3854                 size = ring_buffer_page_len(ref->page);
3855                 if (size < PAGE_SIZE)
3856                         memset(ref->page + size, 0, PAGE_SIZE - size);
3857
3858                 page = virt_to_page(ref->page);
3859
3860                 spd.pages[i] = page;
3861                 spd.partial[i].len = PAGE_SIZE;
3862                 spd.partial[i].offset = 0;
3863                 spd.partial[i].private = (unsigned long)ref;
3864                 spd.nr_pages++;
3865                 *ppos += PAGE_SIZE;
3866
3867                 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3868         }
3869
3870         trace_access_unlock(info->cpu);
3871         spd.nr_pages = i;
3872
3873         /* did we read anything? */
3874         if (!spd.nr_pages) {
3875                 if (flags & SPLICE_F_NONBLOCK)
3876                         ret = -EAGAIN;
3877                 else
3878                         ret = 0;
3879                 /* TODO: block */
3880                 goto out;
3881         }
3882
3883         ret = splice_to_pipe(pipe, &spd);
3884         splice_shrink_spd(pipe, &spd);
3885 out:
3886         return ret;
3887 }
3888
3889 static const struct file_operations tracing_buffers_fops = {
3890         .open           = tracing_buffers_open,
3891         .read           = tracing_buffers_read,
3892         .release        = tracing_buffers_release,
3893         .splice_read    = tracing_buffers_splice_read,
3894         .llseek         = no_llseek,
3895 };
3896
3897 static ssize_t
3898 tracing_stats_read(struct file *filp, char __user *ubuf,
3899                    size_t count, loff_t *ppos)
3900 {
3901         unsigned long cpu = (unsigned long)filp->private_data;
3902         struct trace_array *tr = &global_trace;
3903         struct trace_seq *s;
3904         unsigned long cnt;
3905
3906         s = kmalloc(sizeof(*s), GFP_KERNEL);
3907         if (!s)
3908                 return -ENOMEM;
3909
3910         trace_seq_init(s);
3911
3912         cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3913         trace_seq_printf(s, "entries: %ld\n", cnt);
3914
3915         cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3916         trace_seq_printf(s, "overrun: %ld\n", cnt);
3917
3918         cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3919         trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3920
3921         count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3922
3923         kfree(s);
3924
3925         return count;
3926 }
3927
3928 static const struct file_operations tracing_stats_fops = {
3929         .open           = tracing_open_generic,
3930         .read           = tracing_stats_read,
3931         .llseek         = generic_file_llseek,
3932 };
3933
3934 #ifdef CONFIG_DYNAMIC_FTRACE
3935
3936 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3937 {
3938         return 0;
3939 }
3940
3941 static ssize_t
3942 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3943                   size_t cnt, loff_t *ppos)
3944 {
3945         static char ftrace_dyn_info_buffer[1024];
3946         static DEFINE_MUTEX(dyn_info_mutex);
3947         unsigned long *p = filp->private_data;
3948         char *buf = ftrace_dyn_info_buffer;
3949         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3950         int r;
3951
3952         mutex_lock(&dyn_info_mutex);
3953         r = sprintf(buf, "%ld ", *p);
3954
3955         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3956         buf[r++] = '\n';
3957
3958         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3959
3960         mutex_unlock(&dyn_info_mutex);
3961
3962         return r;
3963 }
3964
3965 static const struct file_operations tracing_dyn_info_fops = {
3966         .open           = tracing_open_generic,
3967         .read           = tracing_read_dyn_info,
3968         .llseek         = generic_file_llseek,
3969 };
3970 #endif
3971
3972 static struct dentry *d_tracer;
3973
3974 struct dentry *tracing_init_dentry(void)
3975 {
3976         static int once;
3977
3978         if (d_tracer)
3979                 return d_tracer;
3980
3981         if (!debugfs_initialized())
3982                 return NULL;
3983
3984         d_tracer = debugfs_create_dir("tracing", NULL);
3985
3986         if (!d_tracer && !once) {
3987                 once = 1;
3988                 pr_warning("Could not create debugfs directory 'tracing'\n");
3989                 return NULL;
3990         }
3991
3992         return d_tracer;
3993 }
3994
3995 static struct dentry *d_percpu;
3996
3997 struct dentry *tracing_dentry_percpu(void)
3998 {
3999         static int once;
4000         struct dentry *d_tracer;
4001
4002         if (d_percpu)
4003                 return d_percpu;
4004
4005         d_tracer = tracing_init_dentry();
4006
4007         if (!d_tracer)
4008                 return NULL;
4009
4010         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4011
4012         if (!d_percpu && !once) {
4013                 once = 1;
4014                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4015                 return NULL;
4016         }
4017
4018         return d_percpu;
4019 }
4020
4021 static void tracing_init_debugfs_percpu(long cpu)
4022 {
4023         struct dentry *d_percpu = tracing_dentry_percpu();
4024         struct dentry *d_cpu;
4025         char cpu_dir[30]; /* 30 characters should be more than enough */
4026
4027         snprintf(cpu_dir, 30, "cpu%ld", cpu);
4028         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4029         if (!d_cpu) {
4030                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4031                 return;
4032         }
4033
4034         /* per cpu trace_pipe */
4035         trace_create_file("trace_pipe", 0444, d_cpu,
4036                         (void *) cpu, &tracing_pipe_fops);
4037
4038         /* per cpu trace */
4039         trace_create_file("trace", 0644, d_cpu,
4040                         (void *) cpu, &tracing_fops);
4041
4042         trace_create_file("trace_pipe_raw", 0444, d_cpu,
4043                         (void *) cpu, &tracing_buffers_fops);
4044
4045         trace_create_file("stats", 0444, d_cpu,
4046                         (void *) cpu, &tracing_stats_fops);
4047 }
4048
4049 #ifdef CONFIG_FTRACE_SELFTEST
4050 /* Let selftest have access to static functions in this file */
4051 #include "trace_selftest.c"
4052 #endif
4053
4054 struct trace_option_dentry {
4055         struct tracer_opt               *opt;
4056         struct tracer_flags             *flags;
4057         struct dentry                   *entry;
4058 };
4059
4060 static ssize_t
4061 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4062                         loff_t *ppos)
4063 {
4064         struct trace_option_dentry *topt = filp->private_data;
4065         char *buf;
4066
4067         if (topt->flags->val & topt->opt->bit)
4068                 buf = "1\n";
4069         else
4070                 buf = "0\n";
4071
4072         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4073 }
4074
4075 static ssize_t
4076 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4077                          loff_t *ppos)
4078 {
4079         struct trace_option_dentry *topt = filp->private_data;
4080         unsigned long val;
4081         char buf[64];
4082         int ret;
4083
4084         if (cnt >= sizeof(buf))
4085                 return -EINVAL;
4086
4087         if (copy_from_user(&buf, ubuf, cnt))
4088                 return -EFAULT;
4089
4090         buf[cnt] = 0;
4091
4092         ret = strict_strtoul(buf, 10, &val);
4093         if (ret < 0)
4094                 return ret;
4095
4096         if (val != 0 && val != 1)
4097                 return -EINVAL;
4098
4099         if (!!(topt->flags->val & topt->opt->bit) != val) {
4100                 mutex_lock(&trace_types_lock);
4101                 ret = __set_tracer_option(current_trace, topt->flags,
4102                                           topt->opt, !val);
4103                 mutex_unlock(&trace_types_lock);
4104                 if (ret)
4105                         return ret;
4106         }
4107
4108         *ppos += cnt;
4109
4110         return cnt;
4111 }
4112
4113
4114 static const struct file_operations trace_options_fops = {
4115         .open = tracing_open_generic,
4116         .read = trace_options_read,
4117         .write = trace_options_write,
4118         .llseek = generic_file_llseek,
4119 };
4120
4121 static ssize_t
4122 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4123                         loff_t *ppos)
4124 {
4125         long index = (long)filp->private_data;
4126         char *buf;
4127
4128         if (trace_flags & (1 << index))
4129                 buf = "1\n";
4130         else
4131                 buf = "0\n";
4132
4133         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4134 }
4135
4136 static ssize_t
4137 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4138                          loff_t *ppos)
4139 {
4140         long index = (long)filp->private_data;
4141         char buf[64];
4142         unsigned long val;
4143         int ret;
4144
4145         if (cnt >= sizeof(buf))
4146                 return -EINVAL;
4147
4148         if (copy_from_user(&buf, ubuf, cnt))
4149                 return -EFAULT;
4150
4151         buf[cnt] = 0;
4152
4153         ret = strict_strtoul(buf, 10, &val);
4154         if (ret < 0)
4155                 return ret;
4156
4157         if (val != 0 && val != 1)
4158                 return -EINVAL;
4159         set_tracer_flags(1 << index, val);
4160
4161         *ppos += cnt;
4162
4163         return cnt;
4164 }
4165
4166 static const struct file_operations trace_options_core_fops = {
4167         .open = tracing_open_generic,
4168         .read = trace_options_core_read,
4169         .write = trace_options_core_write,
4170         .llseek = generic_file_llseek,
4171 };
4172
4173 struct dentry *trace_create_file(const char *name,
4174                                  mode_t mode,
4175                                  struct dentry *parent,
4176                                  void *data,
4177                                  const struct file_operations *fops)
4178 {
4179         struct dentry *ret;
4180
4181         ret = debugfs_create_file(name, mode, parent, data, fops);
4182         if (!ret)
4183                 pr_warning("Could not create debugfs '%s' entry\n", name);
4184
4185         return ret;
4186 }
4187
4188
4189 static struct dentry *trace_options_init_dentry(void)
4190 {
4191         struct dentry *d_tracer;
4192         static struct dentry *t_options;
4193
4194         if (t_options)
4195                 return t_options;
4196
4197         d_tracer = tracing_init_dentry();
4198         if (!d_tracer)
4199                 return NULL;
4200
4201         t_options = debugfs_create_dir("options", d_tracer);
4202         if (!t_options) {
4203                 pr_warning("Could not create debugfs directory 'options'\n");
4204                 return NULL;
4205         }
4206
4207         return t_options;
4208 }
4209
4210 static void
4211 create_trace_option_file(struct trace_option_dentry *topt,
4212                          struct tracer_flags *flags,
4213                          struct tracer_opt *opt)
4214 {
4215         struct dentry *t_options;
4216
4217         t_options = trace_options_init_dentry();
4218         if (!t_options)
4219                 return;
4220
4221         topt->flags = flags;
4222         topt->opt = opt;
4223
4224         topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4225                                     &trace_options_fops);
4226
4227 }
4228
4229 static struct trace_option_dentry *
4230 create_trace_option_files(struct tracer *tracer)
4231 {
4232         struct trace_option_dentry *topts;
4233         struct tracer_flags *flags;
4234         struct tracer_opt *opts;
4235         int cnt;
4236
4237         if (!tracer)
4238                 return NULL;
4239
4240         flags = tracer->flags;
4241
4242         if (!flags || !flags->opts)
4243                 return NULL;
4244
4245         opts = flags->opts;
4246
4247         for (cnt = 0; opts[cnt].name; cnt++)
4248                 ;
4249
4250         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4251         if (!topts)
4252                 return NULL;
4253
4254         for (cnt = 0; opts[cnt].name; cnt++)
4255                 create_trace_option_file(&topts[cnt], flags,
4256                                          &opts[cnt]);
4257
4258         return topts;
4259 }
4260
4261 static void
4262 destroy_trace_option_files(struct trace_option_dentry *topts)
4263 {
4264         int cnt;
4265
4266         if (!topts)
4267                 return;
4268
4269         for (cnt = 0; topts[cnt].opt; cnt++) {
4270                 if (topts[cnt].entry)
4271                         debugfs_remove(topts[cnt].entry);
4272         }
4273
4274         kfree(topts);
4275 }
4276
4277 static struct dentry *
4278 create_trace_option_core_file(const char *option, long index)
4279 {
4280         struct dentry *t_options;
4281
4282         t_options = trace_options_init_dentry();
4283         if (!t_options)
4284                 return NULL;
4285
4286         return trace_create_file(option, 0644, t_options, (void *)index,
4287                                     &trace_options_core_fops);
4288 }
4289
4290 static __init void create_trace_options_dir(void)
4291 {
4292         struct dentry *t_options;
4293         int i;
4294
4295         t_options = trace_options_init_dentry();
4296         if (!t_options)
4297                 return;
4298
4299         for (i = 0; trace_options[i]; i++)
4300                 create_trace_option_core_file(trace_options[i], i);
4301 }
4302
4303 static __init int tracer_init_debugfs(void)
4304 {
4305         struct dentry *d_tracer;
4306         int cpu;
4307
4308         trace_access_lock_init();
4309
4310         d_tracer = tracing_init_dentry();
4311
4312         trace_create_file("tracing_enabled", 0644, d_tracer,
4313                         &global_trace, &tracing_ctrl_fops);
4314
4315         trace_create_file("trace_options", 0644, d_tracer,
4316                         NULL, &tracing_iter_fops);
4317
4318         trace_create_file("tracing_cpumask", 0644, d_tracer,
4319                         NULL, &tracing_cpumask_fops);
4320
4321         trace_create_file("trace", 0644, d_tracer,
4322                         (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4323
4324         trace_create_file("available_tracers", 0444, d_tracer,
4325                         &global_trace, &show_traces_fops);
4326
4327         trace_create_file("current_tracer", 0644, d_tracer,
4328                         &global_trace, &set_tracer_fops);
4329
4330 #ifdef CONFIG_TRACER_MAX_TRACE
4331         trace_create_file("tracing_max_latency", 0644, d_tracer,
4332                         &tracing_max_latency, &tracing_max_lat_fops);
4333 #endif
4334
4335         trace_create_file("tracing_thresh", 0644, d_tracer,
4336                         &tracing_thresh, &tracing_max_lat_fops);
4337
4338         trace_create_file("README", 0444, d_tracer,
4339                         NULL, &tracing_readme_fops);
4340
4341         trace_create_file("trace_pipe", 0444, d_tracer,
4342                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4343
4344         trace_create_file("buffer_size_kb", 0644, d_tracer,
4345                         &global_trace, &tracing_entries_fops);
4346
4347         trace_create_file("trace_marker", 0220, d_tracer,
4348                         NULL, &tracing_mark_fops);
4349
4350         trace_create_file("saved_cmdlines", 0444, d_tracer,
4351                         NULL, &tracing_saved_cmdlines_fops);
4352
4353         trace_create_file("trace_clock", 0644, d_tracer, NULL,
4354                           &trace_clock_fops);
4355
4356 #ifdef CONFIG_DYNAMIC_FTRACE
4357         trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4358                         &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4359 #endif
4360
4361         create_trace_options_dir();
4362
4363         for_each_tracing_cpu(cpu)
4364                 tracing_init_debugfs_percpu(cpu);
4365
4366         return 0;
4367 }
4368
4369 static int trace_panic_handler(struct notifier_block *this,
4370                                unsigned long event, void *unused)
4371 {
4372         if (ftrace_dump_on_oops)
4373                 ftrace_dump(ftrace_dump_on_oops);
4374         return NOTIFY_OK;
4375 }
4376
4377 static struct notifier_block trace_panic_notifier = {
4378         .notifier_call  = trace_panic_handler,
4379         .next           = NULL,
4380         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
4381 };
4382
4383 static int trace_die_handler(struct notifier_block *self,
4384                              unsigned long val,
4385                              void *data)
4386 {
4387         switch (val) {
4388         case DIE_OOPS:
4389                 if (ftrace_dump_on_oops)
4390                         ftrace_dump(ftrace_dump_on_oops);
4391                 break;
4392         default:
4393                 break;
4394         }
4395         return NOTIFY_OK;
4396 }
4397
4398 static struct notifier_block trace_die_notifier = {
4399         .notifier_call = trace_die_handler,
4400         .priority = 200
4401 };
4402
4403 /*
4404  * printk is set to max of 1024, we really don't need it that big.
4405  * Nothing should be printing 1000 characters anyway.
4406  */
4407 #define TRACE_MAX_PRINT         1000
4408
4409 /*
4410  * Define here KERN_TRACE so that we have one place to modify
4411  * it if we decide to change what log level the ftrace dump
4412  * should be at.
4413  */
4414 #define KERN_TRACE              KERN_EMERG
4415
4416 void
4417 trace_printk_seq(struct trace_seq *s)
4418 {
4419         /* Probably should print a warning here. */
4420         if (s->len >= 1000)
4421                 s->len = 1000;
4422
4423         /* should be zero ended, but we are paranoid. */
4424         s->buffer[s->len] = 0;
4425
4426         printk(KERN_TRACE "%s", s->buffer);
4427
4428         trace_seq_init(s);
4429 }
4430
4431 void trace_init_global_iter(struct trace_iterator *iter)
4432 {
4433         iter->tr = &global_trace;
4434         iter->trace = current_trace;
4435         iter->cpu_file = TRACE_PIPE_ALL_CPU;
4436 }
4437
4438 static void
4439 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4440 {
4441         static arch_spinlock_t ftrace_dump_lock =
4442                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4443         /* use static because iter can be a bit big for the stack */
4444         static struct trace_iterator iter;
4445         unsigned int old_userobj;
4446         static int dump_ran;
4447         unsigned long flags;
4448         int cnt = 0, cpu;
4449
4450         /* only one dump */
4451         local_irq_save(flags);
4452         arch_spin_lock(&ftrace_dump_lock);
4453         if (dump_ran)
4454                 goto out;
4455
4456         dump_ran = 1;
4457
4458         tracing_off();
4459
4460         if (disable_tracing)
4461                 ftrace_kill();
4462
4463         trace_init_global_iter(&iter);
4464
4465         for_each_tracing_cpu(cpu) {
4466                 atomic_inc(&iter.tr->data[cpu]->disabled);
4467         }
4468
4469         old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4470
4471         /* don't look at user memory in panic mode */
4472         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4473
4474         /* Simulate the iterator */
4475         iter.tr = &global_trace;
4476         iter.trace = current_trace;
4477
4478         switch (oops_dump_mode) {
4479         case DUMP_ALL:
4480                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4481                 break;
4482         case DUMP_ORIG:
4483                 iter.cpu_file = raw_smp_processor_id();
4484                 break;
4485         case DUMP_NONE:
4486                 goto out_enable;
4487         default:
4488                 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4489                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4490         }
4491
4492         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4493
4494         /*
4495          * We need to stop all tracing on all CPUS to read the
4496          * the next buffer. This is a bit expensive, but is
4497          * not done often. We fill all what we can read,
4498          * and then release the locks again.
4499          */
4500
4501         while (!trace_empty(&iter)) {
4502
4503                 if (!cnt)
4504                         printk(KERN_TRACE "---------------------------------\n");
4505
4506                 cnt++;
4507
4508                 /* reset all but tr, trace, and overruns */
4509                 memset(&iter.seq, 0,
4510                        sizeof(struct trace_iterator) -
4511                        offsetof(struct trace_iterator, seq));
4512                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4513                 iter.pos = -1;
4514
4515                 if (trace_find_next_entry_inc(&iter) != NULL) {
4516                         int ret;
4517
4518                         ret = print_trace_line(&iter);
4519                         if (ret != TRACE_TYPE_NO_CONSUME)
4520                                 trace_consume(&iter);
4521                 }
4522
4523                 trace_printk_seq(&iter.seq);
4524         }
4525
4526         if (!cnt)
4527                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4528         else
4529                 printk(KERN_TRACE "---------------------------------\n");
4530
4531  out_enable:
4532         /* Re-enable tracing if requested */
4533         if (!disable_tracing) {
4534                 trace_flags |= old_userobj;
4535
4536                 for_each_tracing_cpu(cpu) {
4537                         atomic_dec(&iter.tr->data[cpu]->disabled);
4538                 }
4539                 tracing_on();
4540         }
4541
4542  out:
4543         arch_spin_unlock(&ftrace_dump_lock);
4544         local_irq_restore(flags);
4545 }
4546
4547 /* By default: disable tracing after the dump */
4548 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4549 {
4550         __ftrace_dump(true, oops_dump_mode);
4551 }
4552
4553 __init static int tracer_alloc_buffers(void)
4554 {
4555         int ring_buf_size;
4556         int i;
4557         int ret = -ENOMEM;
4558
4559         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4560                 goto out;
4561
4562         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4563                 goto out_free_buffer_mask;
4564
4565         /* To save memory, keep the ring buffer size to its minimum */
4566         if (ring_buffer_expanded)
4567                 ring_buf_size = trace_buf_size;
4568         else
4569                 ring_buf_size = 1;
4570
4571         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4572         cpumask_copy(tracing_cpumask, cpu_all_mask);
4573
4574         /* TODO: make the number of buffers hot pluggable with CPUS */
4575         global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4576                                                    TRACE_BUFFER_FLAGS);
4577         if (!global_trace.buffer) {
4578                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4579                 WARN_ON(1);
4580                 goto out_free_cpumask;
4581         }
4582         global_trace.entries = ring_buffer_size(global_trace.buffer);
4583
4584
4585 #ifdef CONFIG_TRACER_MAX_TRACE
4586         max_tr.buffer = ring_buffer_alloc(1, TRACE_BUFFER_FLAGS);
4587         if (!max_tr.buffer) {
4588                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4589                 WARN_ON(1);
4590                 ring_buffer_free(global_trace.buffer);
4591                 goto out_free_cpumask;
4592         }
4593         max_tr.entries = 1;
4594 #endif
4595
4596         /* Allocate the first page for all buffers */
4597         for_each_tracing_cpu(i) {
4598                 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4599                 max_tr.data[i] = &per_cpu(max_tr_data, i);
4600         }
4601
4602         trace_init_cmdlines();
4603
4604         register_tracer(&nop_trace);
4605         current_trace = &nop_trace;
4606         /* All seems OK, enable tracing */
4607         tracing_disabled = 0;
4608
4609         atomic_notifier_chain_register(&panic_notifier_list,
4610                                        &trace_panic_notifier);
4611
4612         register_die_notifier(&trace_die_notifier);
4613
4614         return 0;
4615
4616 out_free_cpumask:
4617         free_cpumask_var(tracing_cpumask);
4618 out_free_buffer_mask:
4619         free_cpumask_var(tracing_buffer_mask);
4620 out:
4621         return ret;
4622 }
4623
4624 __init static int clear_boot_tracer(void)
4625 {
4626         /*
4627          * The default tracer at boot buffer is an init section.
4628          * This function is called in lateinit. If we did not
4629          * find the boot tracer, then clear it out, to prevent
4630          * later registration from accessing the buffer that is
4631          * about to be freed.
4632          */
4633         if (!default_bootup_tracer)
4634                 return 0;
4635
4636         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4637                default_bootup_tracer);
4638         default_bootup_tracer = NULL;
4639
4640         return 0;
4641 }
4642
4643 early_initcall(tracer_alloc_buffers);
4644 fs_initcall(tracer_init_debugfs);
4645 late_initcall(clear_boot_tracer);