tracing: Fix preempt count leak
[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         event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1317                                           sizeof(*entry), flags, pc);
1318         if (!event)
1319                 goto out_drop_count;
1320         entry   = ring_buffer_event_data(event);
1321
1322         entry->tgid             = current->tgid;
1323         memset(&entry->caller, 0, sizeof(entry->caller));
1324
1325         trace.nr_entries        = 0;
1326         trace.max_entries       = FTRACE_STACK_ENTRIES;
1327         trace.skip              = 0;
1328         trace.entries           = entry->caller;
1329
1330         save_stack_trace_user(&trace);
1331         if (!filter_check_discard(call, entry, buffer, event))
1332                 ring_buffer_unlock_commit(buffer, event);
1333
1334  out_drop_count:
1335         __this_cpu_dec(user_stack_count);
1336  out:
1337         preempt_enable();
1338 }
1339
1340 #ifdef UNUSED
1341 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1342 {
1343         ftrace_trace_userstack(tr, flags, preempt_count());
1344 }
1345 #endif /* UNUSED */
1346
1347 #endif /* CONFIG_STACKTRACE */
1348
1349 /**
1350  * trace_vbprintk - write binary msg to tracing buffer
1351  *
1352  */
1353 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1354 {
1355         static arch_spinlock_t trace_buf_lock =
1356                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1357         static u32 trace_buf[TRACE_BUF_SIZE];
1358
1359         struct ftrace_event_call *call = &event_bprint;
1360         struct ring_buffer_event *event;
1361         struct ring_buffer *buffer;
1362         struct trace_array *tr = &global_trace;
1363         struct trace_array_cpu *data;
1364         struct bprint_entry *entry;
1365         unsigned long flags;
1366         int disable;
1367         int cpu, len = 0, size, pc;
1368
1369         if (unlikely(tracing_selftest_running || tracing_disabled))
1370                 return 0;
1371
1372         /* Don't pollute graph traces with trace_vprintk internals */
1373         pause_graph_tracing();
1374
1375         pc = preempt_count();
1376         preempt_disable_notrace();
1377         cpu = raw_smp_processor_id();
1378         data = tr->data[cpu];
1379
1380         disable = atomic_inc_return(&data->disabled);
1381         if (unlikely(disable != 1))
1382                 goto out;
1383
1384         /* Lockdep uses trace_printk for lock tracing */
1385         local_irq_save(flags);
1386         arch_spin_lock(&trace_buf_lock);
1387         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1388
1389         if (len > TRACE_BUF_SIZE || len < 0)
1390                 goto out_unlock;
1391
1392         size = sizeof(*entry) + sizeof(u32) * len;
1393         buffer = tr->buffer;
1394         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1395                                           flags, pc);
1396         if (!event)
1397                 goto out_unlock;
1398         entry = ring_buffer_event_data(event);
1399         entry->ip                       = ip;
1400         entry->fmt                      = fmt;
1401
1402         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1403         if (!filter_check_discard(call, entry, buffer, event)) {
1404                 ring_buffer_unlock_commit(buffer, event);
1405                 ftrace_trace_stack(buffer, flags, 6, pc);
1406         }
1407
1408 out_unlock:
1409         arch_spin_unlock(&trace_buf_lock);
1410         local_irq_restore(flags);
1411
1412 out:
1413         atomic_dec_return(&data->disabled);
1414         preempt_enable_notrace();
1415         unpause_graph_tracing();
1416
1417         return len;
1418 }
1419 EXPORT_SYMBOL_GPL(trace_vbprintk);
1420
1421 int trace_array_printk(struct trace_array *tr,
1422                        unsigned long ip, const char *fmt, ...)
1423 {
1424         int ret;
1425         va_list ap;
1426
1427         if (!(trace_flags & TRACE_ITER_PRINTK))
1428                 return 0;
1429
1430         va_start(ap, fmt);
1431         ret = trace_array_vprintk(tr, ip, fmt, ap);
1432         va_end(ap);
1433         return ret;
1434 }
1435
1436 int trace_array_vprintk(struct trace_array *tr,
1437                         unsigned long ip, const char *fmt, va_list args)
1438 {
1439         static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1440         static char trace_buf[TRACE_BUF_SIZE];
1441
1442         struct ftrace_event_call *call = &event_print;
1443         struct ring_buffer_event *event;
1444         struct ring_buffer *buffer;
1445         struct trace_array_cpu *data;
1446         int cpu, len = 0, size, pc;
1447         struct print_entry *entry;
1448         unsigned long irq_flags;
1449         int disable;
1450
1451         if (tracing_disabled || tracing_selftest_running)
1452                 return 0;
1453
1454         pc = preempt_count();
1455         preempt_disable_notrace();
1456         cpu = raw_smp_processor_id();
1457         data = tr->data[cpu];
1458
1459         disable = atomic_inc_return(&data->disabled);
1460         if (unlikely(disable != 1))
1461                 goto out;
1462
1463         pause_graph_tracing();
1464         raw_local_irq_save(irq_flags);
1465         arch_spin_lock(&trace_buf_lock);
1466         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1467
1468         size = sizeof(*entry) + len + 1;
1469         buffer = tr->buffer;
1470         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1471                                           irq_flags, pc);
1472         if (!event)
1473                 goto out_unlock;
1474         entry = ring_buffer_event_data(event);
1475         entry->ip = ip;
1476
1477         memcpy(&entry->buf, trace_buf, len);
1478         entry->buf[len] = '\0';
1479         if (!filter_check_discard(call, entry, buffer, event)) {
1480                 ring_buffer_unlock_commit(buffer, event);
1481                 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1482         }
1483
1484  out_unlock:
1485         arch_spin_unlock(&trace_buf_lock);
1486         raw_local_irq_restore(irq_flags);
1487         unpause_graph_tracing();
1488  out:
1489         atomic_dec_return(&data->disabled);
1490         preempt_enable_notrace();
1491
1492         return len;
1493 }
1494
1495 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1496 {
1497         return trace_array_vprintk(&global_trace, ip, fmt, args);
1498 }
1499 EXPORT_SYMBOL_GPL(trace_vprintk);
1500
1501 static void trace_iterator_increment(struct trace_iterator *iter)
1502 {
1503         /* Don't allow ftrace to trace into the ring buffers */
1504         ftrace_disable_cpu();
1505
1506         iter->idx++;
1507         if (iter->buffer_iter[iter->cpu])
1508                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1509
1510         ftrace_enable_cpu();
1511 }
1512
1513 static struct trace_entry *
1514 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1515                 unsigned long *lost_events)
1516 {
1517         struct ring_buffer_event *event;
1518         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1519
1520         /* Don't allow ftrace to trace into the ring buffers */
1521         ftrace_disable_cpu();
1522
1523         if (buf_iter)
1524                 event = ring_buffer_iter_peek(buf_iter, ts);
1525         else
1526                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1527                                          lost_events);
1528
1529         ftrace_enable_cpu();
1530
1531         return event ? ring_buffer_event_data(event) : NULL;
1532 }
1533
1534 static struct trace_entry *
1535 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1536                   unsigned long *missing_events, u64 *ent_ts)
1537 {
1538         struct ring_buffer *buffer = iter->tr->buffer;
1539         struct trace_entry *ent, *next = NULL;
1540         unsigned long lost_events = 0, next_lost = 0;
1541         int cpu_file = iter->cpu_file;
1542         u64 next_ts = 0, ts;
1543         int next_cpu = -1;
1544         int cpu;
1545
1546         /*
1547          * If we are in a per_cpu trace file, don't bother by iterating over
1548          * all cpu and peek directly.
1549          */
1550         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1551                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1552                         return NULL;
1553                 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1554                 if (ent_cpu)
1555                         *ent_cpu = cpu_file;
1556
1557                 return ent;
1558         }
1559
1560         for_each_tracing_cpu(cpu) {
1561
1562                 if (ring_buffer_empty_cpu(buffer, cpu))
1563                         continue;
1564
1565                 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1566
1567                 /*
1568                  * Pick the entry with the smallest timestamp:
1569                  */
1570                 if (ent && (!next || ts < next_ts)) {
1571                         next = ent;
1572                         next_cpu = cpu;
1573                         next_ts = ts;
1574                         next_lost = lost_events;
1575                 }
1576         }
1577
1578         if (ent_cpu)
1579                 *ent_cpu = next_cpu;
1580
1581         if (ent_ts)
1582                 *ent_ts = next_ts;
1583
1584         if (missing_events)
1585                 *missing_events = next_lost;
1586
1587         return next;
1588 }
1589
1590 /* Find the next real entry, without updating the iterator itself */
1591 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1592                                           int *ent_cpu, u64 *ent_ts)
1593 {
1594         return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1595 }
1596
1597 /* Find the next real entry, and increment the iterator to the next entry */
1598 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1599 {
1600         iter->ent = __find_next_entry(iter, &iter->cpu,
1601                                       &iter->lost_events, &iter->ts);
1602
1603         if (iter->ent)
1604                 trace_iterator_increment(iter);
1605
1606         return iter->ent ? iter : NULL;
1607 }
1608
1609 static void trace_consume(struct trace_iterator *iter)
1610 {
1611         /* Don't allow ftrace to trace into the ring buffers */
1612         ftrace_disable_cpu();
1613         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1614                             &iter->lost_events);
1615         ftrace_enable_cpu();
1616 }
1617
1618 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1619 {
1620         struct trace_iterator *iter = m->private;
1621         int i = (int)*pos;
1622         void *ent;
1623
1624         WARN_ON_ONCE(iter->leftover);
1625
1626         (*pos)++;
1627
1628         /* can't go backwards */
1629         if (iter->idx > i)
1630                 return NULL;
1631
1632         if (iter->idx < 0)
1633                 ent = trace_find_next_entry_inc(iter);
1634         else
1635                 ent = iter;
1636
1637         while (ent && iter->idx < i)
1638                 ent = trace_find_next_entry_inc(iter);
1639
1640         iter->pos = *pos;
1641
1642         return ent;
1643 }
1644
1645 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1646 {
1647         struct trace_array *tr = iter->tr;
1648         struct ring_buffer_event *event;
1649         struct ring_buffer_iter *buf_iter;
1650         unsigned long entries = 0;
1651         u64 ts;
1652
1653         tr->data[cpu]->skipped_entries = 0;
1654
1655         if (!iter->buffer_iter[cpu])
1656                 return;
1657
1658         buf_iter = iter->buffer_iter[cpu];
1659         ring_buffer_iter_reset(buf_iter);
1660
1661         /*
1662          * We could have the case with the max latency tracers
1663          * that a reset never took place on a cpu. This is evident
1664          * by the timestamp being before the start of the buffer.
1665          */
1666         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1667                 if (ts >= iter->tr->time_start)
1668                         break;
1669                 entries++;
1670                 ring_buffer_read(buf_iter, NULL);
1671         }
1672
1673         tr->data[cpu]->skipped_entries = entries;
1674 }
1675
1676 /*
1677  * The current tracer is copied to avoid a global locking
1678  * all around.
1679  */
1680 static void *s_start(struct seq_file *m, loff_t *pos)
1681 {
1682         struct trace_iterator *iter = m->private;
1683         static struct tracer *old_tracer;
1684         int cpu_file = iter->cpu_file;
1685         void *p = NULL;
1686         loff_t l = 0;
1687         int cpu;
1688
1689         /* copy the tracer to avoid using a global lock all around */
1690         mutex_lock(&trace_types_lock);
1691         if (unlikely(old_tracer != current_trace && current_trace)) {
1692                 old_tracer = current_trace;
1693                 *iter->trace = *current_trace;
1694         }
1695         mutex_unlock(&trace_types_lock);
1696
1697         atomic_inc(&trace_record_cmdline_disabled);
1698
1699         if (*pos != iter->pos) {
1700                 iter->ent = NULL;
1701                 iter->cpu = 0;
1702                 iter->idx = -1;
1703
1704                 ftrace_disable_cpu();
1705
1706                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1707                         for_each_tracing_cpu(cpu)
1708                                 tracing_iter_reset(iter, cpu);
1709                 } else
1710                         tracing_iter_reset(iter, cpu_file);
1711
1712                 ftrace_enable_cpu();
1713
1714                 iter->leftover = 0;
1715                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1716                         ;
1717
1718         } else {
1719                 /*
1720                  * If we overflowed the seq_file before, then we want
1721                  * to just reuse the trace_seq buffer again.
1722                  */
1723                 if (iter->leftover)
1724                         p = iter;
1725                 else {
1726                         l = *pos - 1;
1727                         p = s_next(m, p, &l);
1728                 }
1729         }
1730
1731         trace_event_read_lock();
1732         trace_access_lock(cpu_file);
1733         return p;
1734 }
1735
1736 static void s_stop(struct seq_file *m, void *p)
1737 {
1738         struct trace_iterator *iter = m->private;
1739
1740         atomic_dec(&trace_record_cmdline_disabled);
1741         trace_access_unlock(iter->cpu_file);
1742         trace_event_read_unlock();
1743 }
1744
1745 static void print_lat_help_header(struct seq_file *m)
1746 {
1747         seq_puts(m, "#                  _------=> CPU#            \n");
1748         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1749         seq_puts(m, "#                | / _----=> need-resched    \n");
1750         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1751         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1752         seq_puts(m, "#                |||| /_--=> lock-depth       \n");
1753         seq_puts(m, "#                |||||/     delay             \n");
1754         seq_puts(m, "#  cmd     pid   |||||| time  |   caller      \n");
1755         seq_puts(m, "#     \\   /      ||||||   \\   |   /           \n");
1756 }
1757
1758 static void print_func_help_header(struct seq_file *m)
1759 {
1760         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1761         seq_puts(m, "#              | |       |          |         |\n");
1762 }
1763
1764
1765 void
1766 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1767 {
1768         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1769         struct trace_array *tr = iter->tr;
1770         struct trace_array_cpu *data = tr->data[tr->cpu];
1771         struct tracer *type = current_trace;
1772         unsigned long entries = 0;
1773         unsigned long total = 0;
1774         unsigned long count;
1775         const char *name = "preemption";
1776         int cpu;
1777
1778         if (type)
1779                 name = type->name;
1780
1781
1782         for_each_tracing_cpu(cpu) {
1783                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1784                 /*
1785                  * If this buffer has skipped entries, then we hold all
1786                  * entries for the trace and we need to ignore the
1787                  * ones before the time stamp.
1788                  */
1789                 if (tr->data[cpu]->skipped_entries) {
1790                         count -= tr->data[cpu]->skipped_entries;
1791                         /* total is the same as the entries */
1792                         total += count;
1793                 } else
1794                         total += count +
1795                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1796                 entries += count;
1797         }
1798
1799         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1800                    name, UTS_RELEASE);
1801         seq_puts(m, "# -----------------------------------"
1802                  "---------------------------------\n");
1803         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1804                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1805                    nsecs_to_usecs(data->saved_latency),
1806                    entries,
1807                    total,
1808                    tr->cpu,
1809 #if defined(CONFIG_PREEMPT_NONE)
1810                    "server",
1811 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1812                    "desktop",
1813 #elif defined(CONFIG_PREEMPT)
1814                    "preempt",
1815 #else
1816                    "unknown",
1817 #endif
1818                    /* These are reserved for later use */
1819                    0, 0, 0, 0);
1820 #ifdef CONFIG_SMP
1821         seq_printf(m, " #P:%d)\n", num_online_cpus());
1822 #else
1823         seq_puts(m, ")\n");
1824 #endif
1825         seq_puts(m, "#    -----------------\n");
1826         seq_printf(m, "#    | task: %.16s-%d "
1827                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1828                    data->comm, data->pid, data->uid, data->nice,
1829                    data->policy, data->rt_priority);
1830         seq_puts(m, "#    -----------------\n");
1831
1832         if (data->critical_start) {
1833                 seq_puts(m, "#  => started at: ");
1834                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1835                 trace_print_seq(m, &iter->seq);
1836                 seq_puts(m, "\n#  => ended at:   ");
1837                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1838                 trace_print_seq(m, &iter->seq);
1839                 seq_puts(m, "\n#\n");
1840         }
1841
1842         seq_puts(m, "#\n");
1843 }
1844
1845 static void test_cpu_buff_start(struct trace_iterator *iter)
1846 {
1847         struct trace_seq *s = &iter->seq;
1848
1849         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1850                 return;
1851
1852         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1853                 return;
1854
1855         if (cpumask_test_cpu(iter->cpu, iter->started))
1856                 return;
1857
1858         if (iter->tr->data[iter->cpu]->skipped_entries)
1859                 return;
1860
1861         cpumask_set_cpu(iter->cpu, iter->started);
1862
1863         /* Don't print started cpu buffer for the first entry of the trace */
1864         if (iter->idx > 1)
1865                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1866                                 iter->cpu);
1867 }
1868
1869 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1870 {
1871         struct trace_seq *s = &iter->seq;
1872         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1873         struct trace_entry *entry;
1874         struct trace_event *event;
1875
1876         entry = iter->ent;
1877
1878         test_cpu_buff_start(iter);
1879
1880         event = ftrace_find_event(entry->type);
1881
1882         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1883                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1884                         if (!trace_print_lat_context(iter))
1885                                 goto partial;
1886                 } else {
1887                         if (!trace_print_context(iter))
1888                                 goto partial;
1889                 }
1890         }
1891
1892         if (event)
1893                 return event->funcs->trace(iter, sym_flags, event);
1894
1895         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1896                 goto partial;
1897
1898         return TRACE_TYPE_HANDLED;
1899 partial:
1900         return TRACE_TYPE_PARTIAL_LINE;
1901 }
1902
1903 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1904 {
1905         struct trace_seq *s = &iter->seq;
1906         struct trace_entry *entry;
1907         struct trace_event *event;
1908
1909         entry = iter->ent;
1910
1911         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1912                 if (!trace_seq_printf(s, "%d %d %llu ",
1913                                       entry->pid, iter->cpu, iter->ts))
1914                         goto partial;
1915         }
1916
1917         event = ftrace_find_event(entry->type);
1918         if (event)
1919                 return event->funcs->raw(iter, 0, event);
1920
1921         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1922                 goto partial;
1923
1924         return TRACE_TYPE_HANDLED;
1925 partial:
1926         return TRACE_TYPE_PARTIAL_LINE;
1927 }
1928
1929 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1930 {
1931         struct trace_seq *s = &iter->seq;
1932         unsigned char newline = '\n';
1933         struct trace_entry *entry;
1934         struct trace_event *event;
1935
1936         entry = iter->ent;
1937
1938         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1939                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1940                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1941                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1942         }
1943
1944         event = ftrace_find_event(entry->type);
1945         if (event) {
1946                 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1947                 if (ret != TRACE_TYPE_HANDLED)
1948                         return ret;
1949         }
1950
1951         SEQ_PUT_FIELD_RET(s, newline);
1952
1953         return TRACE_TYPE_HANDLED;
1954 }
1955
1956 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1957 {
1958         struct trace_seq *s = &iter->seq;
1959         struct trace_entry *entry;
1960         struct trace_event *event;
1961
1962         entry = iter->ent;
1963
1964         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1965                 SEQ_PUT_FIELD_RET(s, entry->pid);
1966                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1967                 SEQ_PUT_FIELD_RET(s, iter->ts);
1968         }
1969
1970         event = ftrace_find_event(entry->type);
1971         return event ? event->funcs->binary(iter, 0, event) :
1972                 TRACE_TYPE_HANDLED;
1973 }
1974
1975 int trace_empty(struct trace_iterator *iter)
1976 {
1977         int cpu;
1978
1979         /* If we are looking at one CPU buffer, only check that one */
1980         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1981                 cpu = iter->cpu_file;
1982                 if (iter->buffer_iter[cpu]) {
1983                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1984                                 return 0;
1985                 } else {
1986                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1987                                 return 0;
1988                 }
1989                 return 1;
1990         }
1991
1992         for_each_tracing_cpu(cpu) {
1993                 if (iter->buffer_iter[cpu]) {
1994                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1995                                 return 0;
1996                 } else {
1997                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1998                                 return 0;
1999                 }
2000         }
2001
2002         return 1;
2003 }
2004
2005 /*  Called with trace_event_read_lock() held. */
2006 enum print_line_t print_trace_line(struct trace_iterator *iter)
2007 {
2008         enum print_line_t ret;
2009
2010         if (iter->lost_events)
2011                 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2012                                  iter->cpu, iter->lost_events);
2013
2014         if (iter->trace && iter->trace->print_line) {
2015                 ret = iter->trace->print_line(iter);
2016                 if (ret != TRACE_TYPE_UNHANDLED)
2017                         return ret;
2018         }
2019
2020         if (iter->ent->type == TRACE_BPRINT &&
2021                         trace_flags & TRACE_ITER_PRINTK &&
2022                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2023                 return trace_print_bprintk_msg_only(iter);
2024
2025         if (iter->ent->type == TRACE_PRINT &&
2026                         trace_flags & TRACE_ITER_PRINTK &&
2027                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2028                 return trace_print_printk_msg_only(iter);
2029
2030         if (trace_flags & TRACE_ITER_BIN)
2031                 return print_bin_fmt(iter);
2032
2033         if (trace_flags & TRACE_ITER_HEX)
2034                 return print_hex_fmt(iter);
2035
2036         if (trace_flags & TRACE_ITER_RAW)
2037                 return print_raw_fmt(iter);
2038
2039         return print_trace_fmt(iter);
2040 }
2041
2042 void trace_default_header(struct seq_file *m)
2043 {
2044         struct trace_iterator *iter = m->private;
2045
2046         if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2047                 /* print nothing if the buffers are empty */
2048                 if (trace_empty(iter))
2049                         return;
2050                 print_trace_header(m, iter);
2051                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2052                         print_lat_help_header(m);
2053         } else {
2054                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2055                         print_func_help_header(m);
2056         }
2057 }
2058
2059 static int s_show(struct seq_file *m, void *v)
2060 {
2061         struct trace_iterator *iter = v;
2062         int ret;
2063
2064         if (iter->ent == NULL) {
2065                 if (iter->tr) {
2066                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2067                         seq_puts(m, "#\n");
2068                 }
2069                 if (iter->trace && iter->trace->print_header)
2070                         iter->trace->print_header(m);
2071                 else
2072                         trace_default_header(m);
2073
2074         } else if (iter->leftover) {
2075                 /*
2076                  * If we filled the seq_file buffer earlier, we
2077                  * want to just show it now.
2078                  */
2079                 ret = trace_print_seq(m, &iter->seq);
2080
2081                 /* ret should this time be zero, but you never know */
2082                 iter->leftover = ret;
2083
2084         } else {
2085                 print_trace_line(iter);
2086                 ret = trace_print_seq(m, &iter->seq);
2087                 /*
2088                  * If we overflow the seq_file buffer, then it will
2089                  * ask us for this data again at start up.
2090                  * Use that instead.
2091                  *  ret is 0 if seq_file write succeeded.
2092                  *        -1 otherwise.
2093                  */
2094                 iter->leftover = ret;
2095         }
2096
2097         return 0;
2098 }
2099
2100 static const struct seq_operations tracer_seq_ops = {
2101         .start          = s_start,
2102         .next           = s_next,
2103         .stop           = s_stop,
2104         .show           = s_show,
2105 };
2106
2107 static struct trace_iterator *
2108 __tracing_open(struct inode *inode, struct file *file)
2109 {
2110         long cpu_file = (long) inode->i_private;
2111         void *fail_ret = ERR_PTR(-ENOMEM);
2112         struct trace_iterator *iter;
2113         struct seq_file *m;
2114         int cpu, ret;
2115
2116         if (tracing_disabled)
2117                 return ERR_PTR(-ENODEV);
2118
2119         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2120         if (!iter)
2121                 return ERR_PTR(-ENOMEM);
2122
2123         /*
2124          * We make a copy of the current tracer to avoid concurrent
2125          * changes on it while we are reading.
2126          */
2127         mutex_lock(&trace_types_lock);
2128         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2129         if (!iter->trace)
2130                 goto fail;
2131
2132         if (current_trace)
2133                 *iter->trace = *current_trace;
2134
2135         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2136                 goto fail;
2137
2138         if (current_trace && current_trace->print_max)
2139                 iter->tr = &max_tr;
2140         else
2141                 iter->tr = &global_trace;
2142         iter->pos = -1;
2143         mutex_init(&iter->mutex);
2144         iter->cpu_file = cpu_file;
2145
2146         /* Notify the tracer early; before we stop tracing. */
2147         if (iter->trace && iter->trace->open)
2148                 iter->trace->open(iter);
2149
2150         /* Annotate start of buffers if we had overruns */
2151         if (ring_buffer_overruns(iter->tr->buffer))
2152                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2153
2154         /* stop the trace while dumping */
2155         tracing_stop();
2156
2157         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2158                 for_each_tracing_cpu(cpu) {
2159                         iter->buffer_iter[cpu] =
2160                                 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2161                 }
2162                 ring_buffer_read_prepare_sync();
2163                 for_each_tracing_cpu(cpu) {
2164                         ring_buffer_read_start(iter->buffer_iter[cpu]);
2165                         tracing_iter_reset(iter, cpu);
2166                 }
2167         } else {
2168                 cpu = iter->cpu_file;
2169                 iter->buffer_iter[cpu] =
2170                         ring_buffer_read_prepare(iter->tr->buffer, cpu);
2171                 ring_buffer_read_prepare_sync();
2172                 ring_buffer_read_start(iter->buffer_iter[cpu]);
2173                 tracing_iter_reset(iter, cpu);
2174         }
2175
2176         ret = seq_open(file, &tracer_seq_ops);
2177         if (ret < 0) {
2178                 fail_ret = ERR_PTR(ret);
2179                 goto fail_buffer;
2180         }
2181
2182         m = file->private_data;
2183         m->private = iter;
2184
2185         mutex_unlock(&trace_types_lock);
2186
2187         return iter;
2188
2189  fail_buffer:
2190         for_each_tracing_cpu(cpu) {
2191                 if (iter->buffer_iter[cpu])
2192                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2193         }
2194         free_cpumask_var(iter->started);
2195         tracing_start();
2196  fail:
2197         mutex_unlock(&trace_types_lock);
2198         kfree(iter->trace);
2199         kfree(iter);
2200
2201         return fail_ret;
2202 }
2203
2204 int tracing_open_generic(struct inode *inode, struct file *filp)
2205 {
2206         if (tracing_disabled)
2207                 return -ENODEV;
2208
2209         filp->private_data = inode->i_private;
2210         return 0;
2211 }
2212
2213 static int tracing_release(struct inode *inode, struct file *file)
2214 {
2215         struct seq_file *m = file->private_data;
2216         struct trace_iterator *iter;
2217         int cpu;
2218
2219         if (!(file->f_mode & FMODE_READ))
2220                 return 0;
2221
2222         iter = m->private;
2223
2224         mutex_lock(&trace_types_lock);
2225         for_each_tracing_cpu(cpu) {
2226                 if (iter->buffer_iter[cpu])
2227                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2228         }
2229
2230         if (iter->trace && iter->trace->close)
2231                 iter->trace->close(iter);
2232
2233         /* reenable tracing if it was previously enabled */
2234         tracing_start();
2235         mutex_unlock(&trace_types_lock);
2236
2237         seq_release(inode, file);
2238         mutex_destroy(&iter->mutex);
2239         free_cpumask_var(iter->started);
2240         kfree(iter->trace);
2241         kfree(iter);
2242         return 0;
2243 }
2244
2245 static int tracing_open(struct inode *inode, struct file *file)
2246 {
2247         struct trace_iterator *iter;
2248         int ret = 0;
2249
2250         /* If this file was open for write, then erase contents */
2251         if ((file->f_mode & FMODE_WRITE) &&
2252             (file->f_flags & O_TRUNC)) {
2253                 long cpu = (long) inode->i_private;
2254
2255                 if (cpu == TRACE_PIPE_ALL_CPU)
2256                         tracing_reset_online_cpus(&global_trace);
2257                 else
2258                         tracing_reset(&global_trace, cpu);
2259         }
2260
2261         if (file->f_mode & FMODE_READ) {
2262                 iter = __tracing_open(inode, file);
2263                 if (IS_ERR(iter))
2264                         ret = PTR_ERR(iter);
2265                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2266                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2267         }
2268         return ret;
2269 }
2270
2271 static void *
2272 t_next(struct seq_file *m, void *v, loff_t *pos)
2273 {
2274         struct tracer *t = v;
2275
2276         (*pos)++;
2277
2278         if (t)
2279                 t = t->next;
2280
2281         return t;
2282 }
2283
2284 static void *t_start(struct seq_file *m, loff_t *pos)
2285 {
2286         struct tracer *t;
2287         loff_t l = 0;
2288
2289         mutex_lock(&trace_types_lock);
2290         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2291                 ;
2292
2293         return t;
2294 }
2295
2296 static void t_stop(struct seq_file *m, void *p)
2297 {
2298         mutex_unlock(&trace_types_lock);
2299 }
2300
2301 static int t_show(struct seq_file *m, void *v)
2302 {
2303         struct tracer *t = v;
2304
2305         if (!t)
2306                 return 0;
2307
2308         seq_printf(m, "%s", t->name);
2309         if (t->next)
2310                 seq_putc(m, ' ');
2311         else
2312                 seq_putc(m, '\n');
2313
2314         return 0;
2315 }
2316
2317 static const struct seq_operations show_traces_seq_ops = {
2318         .start          = t_start,
2319         .next           = t_next,
2320         .stop           = t_stop,
2321         .show           = t_show,
2322 };
2323
2324 static int show_traces_open(struct inode *inode, struct file *file)
2325 {
2326         if (tracing_disabled)
2327                 return -ENODEV;
2328
2329         return seq_open(file, &show_traces_seq_ops);
2330 }
2331
2332 static ssize_t
2333 tracing_write_stub(struct file *filp, const char __user *ubuf,
2334                    size_t count, loff_t *ppos)
2335 {
2336         return count;
2337 }
2338
2339 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2340 {
2341         if (file->f_mode & FMODE_READ)
2342                 return seq_lseek(file, offset, origin);
2343         else
2344                 return 0;
2345 }
2346
2347 static const struct file_operations tracing_fops = {
2348         .open           = tracing_open,
2349         .read           = seq_read,
2350         .write          = tracing_write_stub,
2351         .llseek         = tracing_seek,
2352         .release        = tracing_release,
2353 };
2354
2355 static const struct file_operations show_traces_fops = {
2356         .open           = show_traces_open,
2357         .read           = seq_read,
2358         .release        = seq_release,
2359         .llseek         = seq_lseek,
2360 };
2361
2362 /*
2363  * Only trace on a CPU if the bitmask is set:
2364  */
2365 static cpumask_var_t tracing_cpumask;
2366
2367 /*
2368  * The tracer itself will not take this lock, but still we want
2369  * to provide a consistent cpumask to user-space:
2370  */
2371 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2372
2373 /*
2374  * Temporary storage for the character representation of the
2375  * CPU bitmask (and one more byte for the newline):
2376  */
2377 static char mask_str[NR_CPUS + 1];
2378
2379 static ssize_t
2380 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2381                      size_t count, loff_t *ppos)
2382 {
2383         int len;
2384
2385         mutex_lock(&tracing_cpumask_update_lock);
2386
2387         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2388         if (count - len < 2) {
2389                 count = -EINVAL;
2390                 goto out_err;
2391         }
2392         len += sprintf(mask_str + len, "\n");
2393         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2394
2395 out_err:
2396         mutex_unlock(&tracing_cpumask_update_lock);
2397
2398         return count;
2399 }
2400
2401 static ssize_t
2402 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2403                       size_t count, loff_t *ppos)
2404 {
2405         int err, cpu;
2406         cpumask_var_t tracing_cpumask_new;
2407
2408         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2409                 return -ENOMEM;
2410
2411         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2412         if (err)
2413                 goto err_unlock;
2414
2415         mutex_lock(&tracing_cpumask_update_lock);
2416
2417         local_irq_disable();
2418         arch_spin_lock(&ftrace_max_lock);
2419         for_each_tracing_cpu(cpu) {
2420                 /*
2421                  * Increase/decrease the disabled counter if we are
2422                  * about to flip a bit in the cpumask:
2423                  */
2424                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2425                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2426                         atomic_inc(&global_trace.data[cpu]->disabled);
2427                 }
2428                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2429                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2430                         atomic_dec(&global_trace.data[cpu]->disabled);
2431                 }
2432         }
2433         arch_spin_unlock(&ftrace_max_lock);
2434         local_irq_enable();
2435
2436         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2437
2438         mutex_unlock(&tracing_cpumask_update_lock);
2439         free_cpumask_var(tracing_cpumask_new);
2440
2441         return count;
2442
2443 err_unlock:
2444         free_cpumask_var(tracing_cpumask_new);
2445
2446         return err;
2447 }
2448
2449 static const struct file_operations tracing_cpumask_fops = {
2450         .open           = tracing_open_generic,
2451         .read           = tracing_cpumask_read,
2452         .write          = tracing_cpumask_write,
2453         .llseek         = generic_file_llseek,
2454 };
2455
2456 static int tracing_trace_options_show(struct seq_file *m, void *v)
2457 {
2458         struct tracer_opt *trace_opts;
2459         u32 tracer_flags;
2460         int i;
2461
2462         mutex_lock(&trace_types_lock);
2463         tracer_flags = current_trace->flags->val;
2464         trace_opts = current_trace->flags->opts;
2465
2466         for (i = 0; trace_options[i]; i++) {
2467                 if (trace_flags & (1 << i))
2468                         seq_printf(m, "%s\n", trace_options[i]);
2469                 else
2470                         seq_printf(m, "no%s\n", trace_options[i]);
2471         }
2472
2473         for (i = 0; trace_opts[i].name; i++) {
2474                 if (tracer_flags & trace_opts[i].bit)
2475                         seq_printf(m, "%s\n", trace_opts[i].name);
2476                 else
2477                         seq_printf(m, "no%s\n", trace_opts[i].name);
2478         }
2479         mutex_unlock(&trace_types_lock);
2480
2481         return 0;
2482 }
2483
2484 static int __set_tracer_option(struct tracer *trace,
2485                                struct tracer_flags *tracer_flags,
2486                                struct tracer_opt *opts, int neg)
2487 {
2488         int ret;
2489
2490         ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2491         if (ret)
2492                 return ret;
2493
2494         if (neg)
2495                 tracer_flags->val &= ~opts->bit;
2496         else
2497                 tracer_flags->val |= opts->bit;
2498         return 0;
2499 }
2500
2501 /* Try to assign a tracer specific option */
2502 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2503 {
2504         struct tracer_flags *tracer_flags = trace->flags;
2505         struct tracer_opt *opts = NULL;
2506         int i;
2507
2508         for (i = 0; tracer_flags->opts[i].name; i++) {
2509                 opts = &tracer_flags->opts[i];
2510
2511                 if (strcmp(cmp, opts->name) == 0)
2512                         return __set_tracer_option(trace, trace->flags,
2513                                                    opts, neg);
2514         }
2515
2516         return -EINVAL;
2517 }
2518
2519 static void set_tracer_flags(unsigned int mask, int enabled)
2520 {
2521         /* do nothing if flag is already set */
2522         if (!!(trace_flags & mask) == !!enabled)
2523                 return;
2524
2525         if (enabled)
2526                 trace_flags |= mask;
2527         else
2528                 trace_flags &= ~mask;
2529
2530         if (mask == TRACE_ITER_RECORD_CMD)
2531                 trace_event_enable_cmd_record(enabled);
2532 }
2533
2534 static ssize_t
2535 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2536                         size_t cnt, loff_t *ppos)
2537 {
2538         char buf[64];
2539         char *cmp;
2540         int neg = 0;
2541         int ret;
2542         int i;
2543
2544         if (cnt >= sizeof(buf))
2545                 return -EINVAL;
2546
2547         if (copy_from_user(&buf, ubuf, cnt))
2548                 return -EFAULT;
2549
2550         buf[cnt] = 0;
2551         cmp = strstrip(buf);
2552
2553         if (strncmp(cmp, "no", 2) == 0) {
2554                 neg = 1;
2555                 cmp += 2;
2556         }
2557
2558         for (i = 0; trace_options[i]; i++) {
2559                 if (strcmp(cmp, trace_options[i]) == 0) {
2560                         set_tracer_flags(1 << i, !neg);
2561                         break;
2562                 }
2563         }
2564
2565         /* If no option could be set, test the specific tracer options */
2566         if (!trace_options[i]) {
2567                 mutex_lock(&trace_types_lock);
2568                 ret = set_tracer_option(current_trace, cmp, neg);
2569                 mutex_unlock(&trace_types_lock);
2570                 if (ret)
2571                         return ret;
2572         }
2573
2574         *ppos += cnt;
2575
2576         return cnt;
2577 }
2578
2579 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2580 {
2581         if (tracing_disabled)
2582                 return -ENODEV;
2583         return single_open(file, tracing_trace_options_show, NULL);
2584 }
2585
2586 static const struct file_operations tracing_iter_fops = {
2587         .open           = tracing_trace_options_open,
2588         .read           = seq_read,
2589         .llseek         = seq_lseek,
2590         .release        = single_release,
2591         .write          = tracing_trace_options_write,
2592 };
2593
2594 static const char readme_msg[] =
2595         "tracing mini-HOWTO:\n\n"
2596         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2597         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2598         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2599         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2600         "nop\n"
2601         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2602         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2603         "sched_switch\n"
2604         "# cat /sys/kernel/debug/tracing/trace_options\n"
2605         "noprint-parent nosym-offset nosym-addr noverbose\n"
2606         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2607         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2608         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2609         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2610 ;
2611
2612 static ssize_t
2613 tracing_readme_read(struct file *filp, char __user *ubuf,
2614                        size_t cnt, loff_t *ppos)
2615 {
2616         return simple_read_from_buffer(ubuf, cnt, ppos,
2617                                         readme_msg, strlen(readme_msg));
2618 }
2619
2620 static const struct file_operations tracing_readme_fops = {
2621         .open           = tracing_open_generic,
2622         .read           = tracing_readme_read,
2623         .llseek         = generic_file_llseek,
2624 };
2625
2626 static ssize_t
2627 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2628                                 size_t cnt, loff_t *ppos)
2629 {
2630         char *buf_comm;
2631         char *file_buf;
2632         char *buf;
2633         int len = 0;
2634         int pid;
2635         int i;
2636
2637         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2638         if (!file_buf)
2639                 return -ENOMEM;
2640
2641         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2642         if (!buf_comm) {
2643                 kfree(file_buf);
2644                 return -ENOMEM;
2645         }
2646
2647         buf = file_buf;
2648
2649         for (i = 0; i < SAVED_CMDLINES; i++) {
2650                 int r;
2651
2652                 pid = map_cmdline_to_pid[i];
2653                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2654                         continue;
2655
2656                 trace_find_cmdline(pid, buf_comm);
2657                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2658                 buf += r;
2659                 len += r;
2660         }
2661
2662         len = simple_read_from_buffer(ubuf, cnt, ppos,
2663                                       file_buf, len);
2664
2665         kfree(file_buf);
2666         kfree(buf_comm);
2667
2668         return len;
2669 }
2670
2671 static const struct file_operations tracing_saved_cmdlines_fops = {
2672     .open       = tracing_open_generic,
2673     .read       = tracing_saved_cmdlines_read,
2674     .llseek     = generic_file_llseek,
2675 };
2676
2677 static ssize_t
2678 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2679                   size_t cnt, loff_t *ppos)
2680 {
2681         char buf[64];
2682         int r;
2683
2684         r = sprintf(buf, "%u\n", tracer_enabled);
2685         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2686 }
2687
2688 static ssize_t
2689 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2690                    size_t cnt, loff_t *ppos)
2691 {
2692         struct trace_array *tr = filp->private_data;
2693         char buf[64];
2694         unsigned long val;
2695         int ret;
2696
2697         if (cnt >= sizeof(buf))
2698                 return -EINVAL;
2699
2700         if (copy_from_user(&buf, ubuf, cnt))
2701                 return -EFAULT;
2702
2703         buf[cnt] = 0;
2704
2705         ret = strict_strtoul(buf, 10, &val);
2706         if (ret < 0)
2707                 return ret;
2708
2709         val = !!val;
2710
2711         mutex_lock(&trace_types_lock);
2712         if (tracer_enabled ^ val) {
2713                 if (val) {
2714                         tracer_enabled = 1;
2715                         if (current_trace->start)
2716                                 current_trace->start(tr);
2717                         tracing_start();
2718                 } else {
2719                         tracer_enabled = 0;
2720                         tracing_stop();
2721                         if (current_trace->stop)
2722                                 current_trace->stop(tr);
2723                 }
2724         }
2725         mutex_unlock(&trace_types_lock);
2726
2727         *ppos += cnt;
2728
2729         return cnt;
2730 }
2731
2732 static ssize_t
2733 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2734                        size_t cnt, loff_t *ppos)
2735 {
2736         char buf[MAX_TRACER_SIZE+2];
2737         int r;
2738
2739         mutex_lock(&trace_types_lock);
2740         if (current_trace)
2741                 r = sprintf(buf, "%s\n", current_trace->name);
2742         else
2743                 r = sprintf(buf, "\n");
2744         mutex_unlock(&trace_types_lock);
2745
2746         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2747 }
2748
2749 int tracer_init(struct tracer *t, struct trace_array *tr)
2750 {
2751         tracing_reset_online_cpus(tr);
2752         return t->init(tr);
2753 }
2754
2755 static int tracing_resize_ring_buffer(unsigned long size)
2756 {
2757         int ret;
2758
2759         /*
2760          * If kernel or user changes the size of the ring buffer
2761          * we use the size that was given, and we can forget about
2762          * expanding it later.
2763          */
2764         ring_buffer_expanded = 1;
2765
2766         ret = ring_buffer_resize(global_trace.buffer, size);
2767         if (ret < 0)
2768                 return ret;
2769
2770         if (!current_trace->use_max_tr)
2771                 goto out;
2772
2773         ret = ring_buffer_resize(max_tr.buffer, size);
2774         if (ret < 0) {
2775                 int r;
2776
2777                 r = ring_buffer_resize(global_trace.buffer,
2778                                        global_trace.entries);
2779                 if (r < 0) {
2780                         /*
2781                          * AARGH! We are left with different
2782                          * size max buffer!!!!
2783                          * The max buffer is our "snapshot" buffer.
2784                          * When a tracer needs a snapshot (one of the
2785                          * latency tracers), it swaps the max buffer
2786                          * with the saved snap shot. We succeeded to
2787                          * update the size of the main buffer, but failed to
2788                          * update the size of the max buffer. But when we tried
2789                          * to reset the main buffer to the original size, we
2790                          * failed there too. This is very unlikely to
2791                          * happen, but if it does, warn and kill all
2792                          * tracing.
2793                          */
2794                         WARN_ON(1);
2795                         tracing_disabled = 1;
2796                 }
2797                 return ret;
2798         }
2799
2800         max_tr.entries = size;
2801  out:
2802         global_trace.entries = size;
2803
2804         return ret;
2805 }
2806
2807
2808 /**
2809  * tracing_update_buffers - used by tracing facility to expand ring buffers
2810  *
2811  * To save on memory when the tracing is never used on a system with it
2812  * configured in. The ring buffers are set to a minimum size. But once
2813  * a user starts to use the tracing facility, then they need to grow
2814  * to their default size.
2815  *
2816  * This function is to be called when a tracer is about to be used.
2817  */
2818 int tracing_update_buffers(void)
2819 {
2820         int ret = 0;
2821
2822         mutex_lock(&trace_types_lock);
2823         if (!ring_buffer_expanded)
2824                 ret = tracing_resize_ring_buffer(trace_buf_size);
2825         mutex_unlock(&trace_types_lock);
2826
2827         return ret;
2828 }
2829
2830 struct trace_option_dentry;
2831
2832 static struct trace_option_dentry *
2833 create_trace_option_files(struct tracer *tracer);
2834
2835 static void
2836 destroy_trace_option_files(struct trace_option_dentry *topts);
2837
2838 static int tracing_set_tracer(const char *buf)
2839 {
2840         static struct trace_option_dentry *topts;
2841         struct trace_array *tr = &global_trace;
2842         struct tracer *t;
2843         int ret = 0;
2844
2845         mutex_lock(&trace_types_lock);
2846
2847         if (!ring_buffer_expanded) {
2848                 ret = tracing_resize_ring_buffer(trace_buf_size);
2849                 if (ret < 0)
2850                         goto out;
2851                 ret = 0;
2852         }
2853
2854         for (t = trace_types; t; t = t->next) {
2855                 if (strcmp(t->name, buf) == 0)
2856                         break;
2857         }
2858         if (!t) {
2859                 ret = -EINVAL;
2860                 goto out;
2861         }
2862         if (t == current_trace)
2863                 goto out;
2864
2865         trace_branch_disable();
2866         if (current_trace && current_trace->reset)
2867                 current_trace->reset(tr);
2868         if (current_trace && current_trace->use_max_tr) {
2869                 /*
2870                  * We don't free the ring buffer. instead, resize it because
2871                  * The max_tr ring buffer has some state (e.g. ring->clock) and
2872                  * we want preserve it.
2873                  */
2874                 ring_buffer_resize(max_tr.buffer, 1);
2875                 max_tr.entries = 1;
2876         }
2877         destroy_trace_option_files(topts);
2878
2879         current_trace = t;
2880
2881         topts = create_trace_option_files(current_trace);
2882         if (current_trace->use_max_tr) {
2883                 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
2884                 if (ret < 0)
2885                         goto out;
2886                 max_tr.entries = global_trace.entries;
2887         }
2888
2889         if (t->init) {
2890                 ret = tracer_init(t, tr);
2891                 if (ret)
2892                         goto out;
2893         }
2894
2895         trace_branch_enable(tr);
2896  out:
2897         mutex_unlock(&trace_types_lock);
2898
2899         return ret;
2900 }
2901
2902 static ssize_t
2903 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2904                         size_t cnt, loff_t *ppos)
2905 {
2906         char buf[MAX_TRACER_SIZE+1];
2907         int i;
2908         size_t ret;
2909         int err;
2910
2911         ret = cnt;
2912
2913         if (cnt > MAX_TRACER_SIZE)
2914                 cnt = MAX_TRACER_SIZE;
2915
2916         if (copy_from_user(&buf, ubuf, cnt))
2917                 return -EFAULT;
2918
2919         buf[cnt] = 0;
2920
2921         /* strip ending whitespace. */
2922         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2923                 buf[i] = 0;
2924
2925         err = tracing_set_tracer(buf);
2926         if (err)
2927                 return err;
2928
2929         *ppos += ret;
2930
2931         return ret;
2932 }
2933
2934 static ssize_t
2935 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2936                      size_t cnt, loff_t *ppos)
2937 {
2938         unsigned long *ptr = filp->private_data;
2939         char buf[64];
2940         int r;
2941
2942         r = snprintf(buf, sizeof(buf), "%ld\n",
2943                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2944         if (r > sizeof(buf))
2945                 r = sizeof(buf);
2946         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2947 }
2948
2949 static ssize_t
2950 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2951                       size_t cnt, loff_t *ppos)
2952 {
2953         unsigned long *ptr = filp->private_data;
2954         char buf[64];
2955         unsigned long val;
2956         int ret;
2957
2958         if (cnt >= sizeof(buf))
2959                 return -EINVAL;
2960
2961         if (copy_from_user(&buf, ubuf, cnt))
2962                 return -EFAULT;
2963
2964         buf[cnt] = 0;
2965
2966         ret = strict_strtoul(buf, 10, &val);
2967         if (ret < 0)
2968                 return ret;
2969
2970         *ptr = val * 1000;
2971
2972         return cnt;
2973 }
2974
2975 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2976 {
2977         long cpu_file = (long) inode->i_private;
2978         struct trace_iterator *iter;
2979         int ret = 0;
2980
2981         if (tracing_disabled)
2982                 return -ENODEV;
2983
2984         mutex_lock(&trace_types_lock);
2985
2986         /* create a buffer to store the information to pass to userspace */
2987         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2988         if (!iter) {
2989                 ret = -ENOMEM;
2990                 goto out;
2991         }
2992
2993         /*
2994          * We make a copy of the current tracer to avoid concurrent
2995          * changes on it while we are reading.
2996          */
2997         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2998         if (!iter->trace) {
2999                 ret = -ENOMEM;
3000                 goto fail;
3001         }
3002         if (current_trace)
3003                 *iter->trace = *current_trace;
3004
3005         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3006                 ret = -ENOMEM;
3007                 goto fail;
3008         }
3009
3010         /* trace pipe does not show start of buffer */
3011         cpumask_setall(iter->started);
3012
3013         if (trace_flags & TRACE_ITER_LATENCY_FMT)
3014                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3015
3016         iter->cpu_file = cpu_file;
3017         iter->tr = &global_trace;
3018         mutex_init(&iter->mutex);
3019         filp->private_data = iter;
3020
3021         if (iter->trace->pipe_open)
3022                 iter->trace->pipe_open(iter);
3023
3024         nonseekable_open(inode, filp);
3025 out:
3026         mutex_unlock(&trace_types_lock);
3027         return ret;
3028
3029 fail:
3030         kfree(iter->trace);
3031         kfree(iter);
3032         mutex_unlock(&trace_types_lock);
3033         return ret;
3034 }
3035
3036 static int tracing_release_pipe(struct inode *inode, struct file *file)
3037 {
3038         struct trace_iterator *iter = file->private_data;
3039
3040         mutex_lock(&trace_types_lock);
3041
3042         if (iter->trace->pipe_close)
3043                 iter->trace->pipe_close(iter);
3044
3045         mutex_unlock(&trace_types_lock);
3046
3047         free_cpumask_var(iter->started);
3048         mutex_destroy(&iter->mutex);
3049         kfree(iter->trace);
3050         kfree(iter);
3051
3052         return 0;
3053 }
3054
3055 static unsigned int
3056 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3057 {
3058         struct trace_iterator *iter = filp->private_data;
3059
3060         if (trace_flags & TRACE_ITER_BLOCK) {
3061                 /*
3062                  * Always select as readable when in blocking mode
3063                  */
3064                 return POLLIN | POLLRDNORM;
3065         } else {
3066                 if (!trace_empty(iter))
3067                         return POLLIN | POLLRDNORM;
3068                 poll_wait(filp, &trace_wait, poll_table);
3069                 if (!trace_empty(iter))
3070                         return POLLIN | POLLRDNORM;
3071
3072                 return 0;
3073         }
3074 }
3075
3076
3077 void default_wait_pipe(struct trace_iterator *iter)
3078 {
3079         DEFINE_WAIT(wait);
3080
3081         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3082
3083         if (trace_empty(iter))
3084                 schedule();
3085
3086         finish_wait(&trace_wait, &wait);
3087 }
3088
3089 /*
3090  * This is a make-shift waitqueue.
3091  * A tracer might use this callback on some rare cases:
3092  *
3093  *  1) the current tracer might hold the runqueue lock when it wakes up
3094  *     a reader, hence a deadlock (sched, function, and function graph tracers)
3095  *  2) the function tracers, trace all functions, we don't want
3096  *     the overhead of calling wake_up and friends
3097  *     (and tracing them too)
3098  *
3099  *     Anyway, this is really very primitive wakeup.
3100  */
3101 void poll_wait_pipe(struct trace_iterator *iter)
3102 {
3103         set_current_state(TASK_INTERRUPTIBLE);
3104         /* sleep for 100 msecs, and try again. */
3105         schedule_timeout(HZ / 10);
3106 }
3107
3108 /* Must be called with trace_types_lock mutex held. */
3109 static int tracing_wait_pipe(struct file *filp)
3110 {
3111         struct trace_iterator *iter = filp->private_data;
3112
3113         while (trace_empty(iter)) {
3114
3115                 if ((filp->f_flags & O_NONBLOCK)) {
3116                         return -EAGAIN;
3117                 }
3118
3119                 mutex_unlock(&iter->mutex);
3120
3121                 iter->trace->wait_pipe(iter);
3122
3123                 mutex_lock(&iter->mutex);
3124
3125                 if (signal_pending(current))
3126                         return -EINTR;
3127
3128                 /*
3129                  * We block until we read something and tracing is disabled.
3130                  * We still block if tracing is disabled, but we have never
3131                  * read anything. This allows a user to cat this file, and
3132                  * then enable tracing. But after we have read something,
3133                  * we give an EOF when tracing is again disabled.
3134                  *
3135                  * iter->pos will be 0 if we haven't read anything.
3136                  */
3137                 if (!tracer_enabled && iter->pos)
3138                         break;
3139         }
3140
3141         return 1;
3142 }
3143
3144 /*
3145  * Consumer reader.
3146  */
3147 static ssize_t
3148 tracing_read_pipe(struct file *filp, char __user *ubuf,
3149                   size_t cnt, loff_t *ppos)
3150 {
3151         struct trace_iterator *iter = filp->private_data;
3152         static struct tracer *old_tracer;
3153         ssize_t sret;
3154
3155         /* return any leftover data */
3156         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3157         if (sret != -EBUSY)
3158                 return sret;
3159
3160         trace_seq_init(&iter->seq);
3161
3162         /* copy the tracer to avoid using a global lock all around */
3163         mutex_lock(&trace_types_lock);
3164         if (unlikely(old_tracer != current_trace && current_trace)) {
3165                 old_tracer = current_trace;
3166                 *iter->trace = *current_trace;
3167         }
3168         mutex_unlock(&trace_types_lock);
3169
3170         /*
3171          * Avoid more than one consumer on a single file descriptor
3172          * This is just a matter of traces coherency, the ring buffer itself
3173          * is protected.
3174          */
3175         mutex_lock(&iter->mutex);
3176         if (iter->trace->read) {
3177                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3178                 if (sret)
3179                         goto out;
3180         }
3181
3182 waitagain:
3183         sret = tracing_wait_pipe(filp);
3184         if (sret <= 0)
3185                 goto out;
3186
3187         /* stop when tracing is finished */
3188         if (trace_empty(iter)) {
3189                 sret = 0;
3190                 goto out;
3191         }
3192
3193         if (cnt >= PAGE_SIZE)
3194                 cnt = PAGE_SIZE - 1;
3195
3196         /* reset all but tr, trace, and overruns */
3197         memset(&iter->seq, 0,
3198                sizeof(struct trace_iterator) -
3199                offsetof(struct trace_iterator, seq));
3200         iter->pos = -1;
3201
3202         trace_event_read_lock();
3203         trace_access_lock(iter->cpu_file);
3204         while (trace_find_next_entry_inc(iter) != NULL) {
3205                 enum print_line_t ret;
3206                 int len = iter->seq.len;
3207
3208                 ret = print_trace_line(iter);
3209                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3210                         /* don't print partial lines */
3211                         iter->seq.len = len;
3212                         break;
3213                 }
3214                 if (ret != TRACE_TYPE_NO_CONSUME)
3215                         trace_consume(iter);
3216
3217                 if (iter->seq.len >= cnt)
3218                         break;
3219         }
3220         trace_access_unlock(iter->cpu_file);
3221         trace_event_read_unlock();
3222
3223         /* Now copy what we have to the user */
3224         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3225         if (iter->seq.readpos >= iter->seq.len)
3226                 trace_seq_init(&iter->seq);
3227
3228         /*
3229          * If there was nothing to send to user, inspite of consuming trace
3230          * entries, go back to wait for more entries.
3231          */
3232         if (sret == -EBUSY)
3233                 goto waitagain;
3234
3235 out:
3236         mutex_unlock(&iter->mutex);
3237
3238         return sret;
3239 }
3240
3241 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3242                                      struct pipe_buffer *buf)
3243 {
3244         __free_page(buf->page);
3245 }
3246
3247 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3248                                      unsigned int idx)
3249 {
3250         __free_page(spd->pages[idx]);
3251 }
3252
3253 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3254         .can_merge              = 0,
3255         .map                    = generic_pipe_buf_map,
3256         .unmap                  = generic_pipe_buf_unmap,
3257         .confirm                = generic_pipe_buf_confirm,
3258         .release                = tracing_pipe_buf_release,
3259         .steal                  = generic_pipe_buf_steal,
3260         .get                    = generic_pipe_buf_get,
3261 };
3262
3263 static size_t
3264 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3265 {
3266         size_t count;
3267         int ret;
3268
3269         /* Seq buffer is page-sized, exactly what we need. */
3270         for (;;) {
3271                 count = iter->seq.len;
3272                 ret = print_trace_line(iter);
3273                 count = iter->seq.len - count;
3274                 if (rem < count) {
3275                         rem = 0;
3276                         iter->seq.len -= count;
3277                         break;
3278                 }
3279                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3280                         iter->seq.len -= count;
3281                         break;
3282                 }
3283
3284                 if (ret != TRACE_TYPE_NO_CONSUME)
3285                         trace_consume(iter);
3286                 rem -= count;
3287                 if (!trace_find_next_entry_inc(iter))   {
3288                         rem = 0;
3289                         iter->ent = NULL;
3290                         break;
3291                 }
3292         }
3293
3294         return rem;
3295 }
3296
3297 static ssize_t tracing_splice_read_pipe(struct file *filp,
3298                                         loff_t *ppos,
3299                                         struct pipe_inode_info *pipe,
3300                                         size_t len,
3301                                         unsigned int flags)
3302 {
3303         struct page *pages_def[PIPE_DEF_BUFFERS];
3304         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3305         struct trace_iterator *iter = filp->private_data;
3306         struct splice_pipe_desc spd = {
3307                 .pages          = pages_def,
3308                 .partial        = partial_def,
3309                 .nr_pages       = 0, /* This gets updated below. */
3310                 .flags          = flags,
3311                 .ops            = &tracing_pipe_buf_ops,
3312                 .spd_release    = tracing_spd_release_pipe,
3313         };
3314         static struct tracer *old_tracer;
3315         ssize_t ret;
3316         size_t rem;
3317         unsigned int i;
3318
3319         if (splice_grow_spd(pipe, &spd))
3320                 return -ENOMEM;
3321
3322         /* copy the tracer to avoid using a global lock all around */
3323         mutex_lock(&trace_types_lock);
3324         if (unlikely(old_tracer != current_trace && current_trace)) {
3325                 old_tracer = current_trace;
3326                 *iter->trace = *current_trace;
3327         }
3328         mutex_unlock(&trace_types_lock);
3329
3330         mutex_lock(&iter->mutex);
3331
3332         if (iter->trace->splice_read) {
3333                 ret = iter->trace->splice_read(iter, filp,
3334                                                ppos, pipe, len, flags);
3335                 if (ret)
3336                         goto out_err;
3337         }
3338
3339         ret = tracing_wait_pipe(filp);
3340         if (ret <= 0)
3341                 goto out_err;
3342
3343         if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3344                 ret = -EFAULT;
3345                 goto out_err;
3346         }
3347
3348         trace_event_read_lock();
3349         trace_access_lock(iter->cpu_file);
3350
3351         /* Fill as many pages as possible. */
3352         for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3353                 spd.pages[i] = alloc_page(GFP_KERNEL);
3354                 if (!spd.pages[i])
3355                         break;
3356
3357                 rem = tracing_fill_pipe_page(rem, iter);
3358
3359                 /* Copy the data into the page, so we can start over. */
3360                 ret = trace_seq_to_buffer(&iter->seq,
3361                                           page_address(spd.pages[i]),
3362                                           iter->seq.len);
3363                 if (ret < 0) {
3364                         __free_page(spd.pages[i]);
3365                         break;
3366                 }
3367                 spd.partial[i].offset = 0;
3368                 spd.partial[i].len = iter->seq.len;
3369
3370                 trace_seq_init(&iter->seq);
3371         }
3372
3373         trace_access_unlock(iter->cpu_file);
3374         trace_event_read_unlock();
3375         mutex_unlock(&iter->mutex);
3376
3377         spd.nr_pages = i;
3378
3379         ret = splice_to_pipe(pipe, &spd);
3380 out:
3381         splice_shrink_spd(pipe, &spd);
3382         return ret;
3383
3384 out_err:
3385         mutex_unlock(&iter->mutex);
3386         goto out;
3387 }
3388
3389 static ssize_t
3390 tracing_entries_read(struct file *filp, char __user *ubuf,
3391                      size_t cnt, loff_t *ppos)
3392 {
3393         struct trace_array *tr = filp->private_data;
3394         char buf[96];
3395         int r;
3396
3397         mutex_lock(&trace_types_lock);
3398         if (!ring_buffer_expanded)
3399                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3400                             tr->entries >> 10,
3401                             trace_buf_size >> 10);
3402         else
3403                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3404         mutex_unlock(&trace_types_lock);
3405
3406         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3407 }
3408
3409 static ssize_t
3410 tracing_entries_write(struct file *filp, const char __user *ubuf,
3411                       size_t cnt, loff_t *ppos)
3412 {
3413         unsigned long val;
3414         char buf[64];
3415         int ret, cpu;
3416
3417         if (cnt >= sizeof(buf))
3418                 return -EINVAL;
3419
3420         if (copy_from_user(&buf, ubuf, cnt))
3421                 return -EFAULT;
3422
3423         buf[cnt] = 0;
3424
3425         ret = strict_strtoul(buf, 10, &val);
3426         if (ret < 0)
3427                 return ret;
3428
3429         /* must have at least 1 entry */
3430         if (!val)
3431                 return -EINVAL;
3432
3433         mutex_lock(&trace_types_lock);
3434
3435         tracing_stop();
3436
3437         /* disable all cpu buffers */
3438         for_each_tracing_cpu(cpu) {
3439                 if (global_trace.data[cpu])
3440                         atomic_inc(&global_trace.data[cpu]->disabled);
3441                 if (max_tr.data[cpu])
3442                         atomic_inc(&max_tr.data[cpu]->disabled);
3443         }
3444
3445         /* value is in KB */
3446         val <<= 10;
3447
3448         if (val != global_trace.entries) {
3449                 ret = tracing_resize_ring_buffer(val);
3450                 if (ret < 0) {
3451                         cnt = ret;
3452                         goto out;
3453                 }
3454         }
3455
3456         *ppos += cnt;
3457
3458         /* If check pages failed, return ENOMEM */
3459         if (tracing_disabled)
3460                 cnt = -ENOMEM;
3461  out:
3462         for_each_tracing_cpu(cpu) {
3463                 if (global_trace.data[cpu])
3464                         atomic_dec(&global_trace.data[cpu]->disabled);
3465                 if (max_tr.data[cpu])
3466                         atomic_dec(&max_tr.data[cpu]->disabled);
3467         }
3468
3469         tracing_start();
3470         mutex_unlock(&trace_types_lock);
3471
3472         return cnt;
3473 }
3474
3475 static int mark_printk(const char *fmt, ...)
3476 {
3477         int ret;
3478         va_list args;
3479         va_start(args, fmt);
3480         ret = trace_vprintk(0, fmt, args);
3481         va_end(args);
3482         return ret;
3483 }
3484
3485 static ssize_t
3486 tracing_mark_write(struct file *filp, const char __user *ubuf,
3487                                         size_t cnt, loff_t *fpos)
3488 {
3489         char *buf;
3490         size_t written;
3491
3492         if (tracing_disabled)
3493                 return -EINVAL;
3494
3495         if (cnt > TRACE_BUF_SIZE)
3496                 cnt = TRACE_BUF_SIZE;
3497
3498         buf = kmalloc(cnt + 2, GFP_KERNEL);
3499         if (buf == NULL)
3500                 return -ENOMEM;
3501
3502         if (copy_from_user(buf, ubuf, cnt)) {
3503                 kfree(buf);
3504                 return -EFAULT;
3505         }
3506         if (buf[cnt-1] != '\n') {
3507                 buf[cnt] = '\n';
3508                 buf[cnt+1] = '\0';
3509         } else
3510                 buf[cnt] = '\0';
3511
3512         written = mark_printk("%s", buf);
3513         kfree(buf);
3514         *fpos += written;
3515
3516         /* don't tell userspace we wrote more - it might confuse them */
3517         if (written > cnt)
3518                 written = cnt;
3519
3520         return written;
3521 }
3522
3523 static int tracing_clock_show(struct seq_file *m, void *v)
3524 {
3525         int i;
3526
3527         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3528                 seq_printf(m,
3529                         "%s%s%s%s", i ? " " : "",
3530                         i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3531                         i == trace_clock_id ? "]" : "");
3532         seq_putc(m, '\n');
3533
3534         return 0;
3535 }
3536
3537 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3538                                    size_t cnt, loff_t *fpos)
3539 {
3540         char buf[64];
3541         const char *clockstr;
3542         int i;
3543
3544         if (cnt >= sizeof(buf))
3545                 return -EINVAL;
3546
3547         if (copy_from_user(&buf, ubuf, cnt))
3548                 return -EFAULT;
3549
3550         buf[cnt] = 0;
3551
3552         clockstr = strstrip(buf);
3553
3554         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3555                 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3556                         break;
3557         }
3558         if (i == ARRAY_SIZE(trace_clocks))
3559                 return -EINVAL;
3560
3561         trace_clock_id = i;
3562
3563         mutex_lock(&trace_types_lock);
3564
3565         ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3566         if (max_tr.buffer)
3567                 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3568
3569         mutex_unlock(&trace_types_lock);
3570
3571         *fpos += cnt;
3572
3573         return cnt;
3574 }
3575
3576 static int tracing_clock_open(struct inode *inode, struct file *file)
3577 {
3578         if (tracing_disabled)
3579                 return -ENODEV;
3580         return single_open(file, tracing_clock_show, NULL);
3581 }
3582
3583 static const struct file_operations tracing_max_lat_fops = {
3584         .open           = tracing_open_generic,
3585         .read           = tracing_max_lat_read,
3586         .write          = tracing_max_lat_write,
3587         .llseek         = generic_file_llseek,
3588 };
3589
3590 static const struct file_operations tracing_ctrl_fops = {
3591         .open           = tracing_open_generic,
3592         .read           = tracing_ctrl_read,
3593         .write          = tracing_ctrl_write,
3594         .llseek         = generic_file_llseek,
3595 };
3596
3597 static const struct file_operations set_tracer_fops = {
3598         .open           = tracing_open_generic,
3599         .read           = tracing_set_trace_read,
3600         .write          = tracing_set_trace_write,
3601         .llseek         = generic_file_llseek,
3602 };
3603
3604 static const struct file_operations tracing_pipe_fops = {
3605         .open           = tracing_open_pipe,
3606         .poll           = tracing_poll_pipe,
3607         .read           = tracing_read_pipe,
3608         .splice_read    = tracing_splice_read_pipe,
3609         .release        = tracing_release_pipe,
3610         .llseek         = no_llseek,
3611 };
3612
3613 static const struct file_operations tracing_entries_fops = {
3614         .open           = tracing_open_generic,
3615         .read           = tracing_entries_read,
3616         .write          = tracing_entries_write,
3617         .llseek         = generic_file_llseek,
3618 };
3619
3620 static const struct file_operations tracing_mark_fops = {
3621         .open           = tracing_open_generic,
3622         .write          = tracing_mark_write,
3623         .llseek         = generic_file_llseek,
3624 };
3625
3626 static const struct file_operations trace_clock_fops = {
3627         .open           = tracing_clock_open,
3628         .read           = seq_read,
3629         .llseek         = seq_lseek,
3630         .release        = single_release,
3631         .write          = tracing_clock_write,
3632 };
3633
3634 struct ftrace_buffer_info {
3635         struct trace_array      *tr;
3636         void                    *spare;
3637         int                     cpu;
3638         unsigned int            read;
3639 };
3640
3641 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3642 {
3643         int cpu = (int)(long)inode->i_private;
3644         struct ftrace_buffer_info *info;
3645
3646         if (tracing_disabled)
3647                 return -ENODEV;
3648
3649         info = kzalloc(sizeof(*info), GFP_KERNEL);
3650         if (!info)
3651                 return -ENOMEM;
3652
3653         info->tr        = &global_trace;
3654         info->cpu       = cpu;
3655         info->spare     = NULL;
3656         /* Force reading ring buffer for first read */
3657         info->read      = (unsigned int)-1;
3658
3659         filp->private_data = info;
3660
3661         return nonseekable_open(inode, filp);
3662 }
3663
3664 static ssize_t
3665 tracing_buffers_read(struct file *filp, char __user *ubuf,
3666                      size_t count, loff_t *ppos)
3667 {
3668         struct ftrace_buffer_info *info = filp->private_data;
3669         ssize_t ret;
3670         size_t size;
3671
3672         if (!count)
3673                 return 0;
3674
3675         if (!info->spare)
3676                 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3677         if (!info->spare)
3678                 return -ENOMEM;
3679
3680         /* Do we have previous read data to read? */
3681         if (info->read < PAGE_SIZE)
3682                 goto read;
3683
3684         info->read = 0;
3685
3686         trace_access_lock(info->cpu);
3687         ret = ring_buffer_read_page(info->tr->buffer,
3688                                     &info->spare,
3689                                     count,
3690                                     info->cpu, 0);
3691         trace_access_unlock(info->cpu);
3692         if (ret < 0)
3693                 return 0;
3694
3695 read:
3696         size = PAGE_SIZE - info->read;
3697         if (size > count)
3698                 size = count;
3699
3700         ret = copy_to_user(ubuf, info->spare + info->read, size);
3701         if (ret == size)
3702                 return -EFAULT;
3703         size -= ret;
3704
3705         *ppos += size;
3706         info->read += size;
3707
3708         return size;
3709 }
3710
3711 static int tracing_buffers_release(struct inode *inode, struct file *file)
3712 {
3713         struct ftrace_buffer_info *info = file->private_data;
3714
3715         if (info->spare)
3716                 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3717         kfree(info);
3718
3719         return 0;
3720 }
3721
3722 struct buffer_ref {
3723         struct ring_buffer      *buffer;
3724         void                    *page;
3725         int                     ref;
3726 };
3727
3728 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3729                                     struct pipe_buffer *buf)
3730 {
3731         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3732
3733         if (--ref->ref)
3734                 return;
3735
3736         ring_buffer_free_read_page(ref->buffer, ref->page);
3737         kfree(ref);
3738         buf->private = 0;
3739 }
3740
3741 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3742                                  struct pipe_buffer *buf)
3743 {
3744         return 1;
3745 }
3746
3747 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3748                                 struct pipe_buffer *buf)
3749 {
3750         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3751
3752         ref->ref++;
3753 }
3754
3755 /* Pipe buffer operations for a buffer. */
3756 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3757         .can_merge              = 0,
3758         .map                    = generic_pipe_buf_map,
3759         .unmap                  = generic_pipe_buf_unmap,
3760         .confirm                = generic_pipe_buf_confirm,
3761         .release                = buffer_pipe_buf_release,
3762         .steal                  = buffer_pipe_buf_steal,
3763         .get                    = buffer_pipe_buf_get,
3764 };
3765
3766 /*
3767  * Callback from splice_to_pipe(), if we need to release some pages
3768  * at the end of the spd in case we error'ed out in filling the pipe.
3769  */
3770 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3771 {
3772         struct buffer_ref *ref =
3773                 (struct buffer_ref *)spd->partial[i].private;
3774
3775         if (--ref->ref)
3776                 return;
3777
3778         ring_buffer_free_read_page(ref->buffer, ref->page);
3779         kfree(ref);
3780         spd->partial[i].private = 0;
3781 }
3782
3783 static ssize_t
3784 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3785                             struct pipe_inode_info *pipe, size_t len,
3786                             unsigned int flags)
3787 {
3788         struct ftrace_buffer_info *info = file->private_data;
3789         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3790         struct page *pages_def[PIPE_DEF_BUFFERS];
3791         struct splice_pipe_desc spd = {
3792                 .pages          = pages_def,
3793                 .partial        = partial_def,
3794                 .flags          = flags,
3795                 .ops            = &buffer_pipe_buf_ops,
3796                 .spd_release    = buffer_spd_release,
3797         };
3798         struct buffer_ref *ref;
3799         int entries, size, i;
3800         size_t ret;
3801
3802         if (splice_grow_spd(pipe, &spd))
3803                 return -ENOMEM;
3804
3805         if (*ppos & (PAGE_SIZE - 1)) {
3806                 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3807                 ret = -EINVAL;
3808                 goto out;
3809         }
3810
3811         if (len & (PAGE_SIZE - 1)) {
3812                 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3813                 if (len < PAGE_SIZE) {
3814                         ret = -EINVAL;
3815                         goto out;
3816                 }
3817                 len &= PAGE_MASK;
3818         }
3819
3820         trace_access_lock(info->cpu);
3821         entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3822
3823         for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3824                 struct page *page;
3825                 int r;
3826
3827                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3828                 if (!ref)
3829                         break;
3830
3831                 ref->ref = 1;
3832                 ref->buffer = info->tr->buffer;
3833                 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3834                 if (!ref->page) {
3835                         kfree(ref);
3836                         break;
3837                 }
3838
3839                 r = ring_buffer_read_page(ref->buffer, &ref->page,
3840                                           len, info->cpu, 1);
3841                 if (r < 0) {
3842                         ring_buffer_free_read_page(ref->buffer,
3843                                                    ref->page);
3844                         kfree(ref);
3845                         break;
3846                 }
3847
3848                 /*
3849                  * zero out any left over data, this is going to
3850                  * user land.
3851                  */
3852                 size = ring_buffer_page_len(ref->page);
3853                 if (size < PAGE_SIZE)
3854                         memset(ref->page + size, 0, PAGE_SIZE - size);
3855
3856                 page = virt_to_page(ref->page);
3857
3858                 spd.pages[i] = page;
3859                 spd.partial[i].len = PAGE_SIZE;
3860                 spd.partial[i].offset = 0;
3861                 spd.partial[i].private = (unsigned long)ref;
3862                 spd.nr_pages++;
3863                 *ppos += PAGE_SIZE;
3864
3865                 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3866         }
3867
3868         trace_access_unlock(info->cpu);
3869         spd.nr_pages = i;
3870
3871         /* did we read anything? */
3872         if (!spd.nr_pages) {
3873                 if (flags & SPLICE_F_NONBLOCK)
3874                         ret = -EAGAIN;
3875                 else
3876                         ret = 0;
3877                 /* TODO: block */
3878                 goto out;
3879         }
3880
3881         ret = splice_to_pipe(pipe, &spd);
3882         splice_shrink_spd(pipe, &spd);
3883 out:
3884         return ret;
3885 }
3886
3887 static const struct file_operations tracing_buffers_fops = {
3888         .open           = tracing_buffers_open,
3889         .read           = tracing_buffers_read,
3890         .release        = tracing_buffers_release,
3891         .splice_read    = tracing_buffers_splice_read,
3892         .llseek         = no_llseek,
3893 };
3894
3895 static ssize_t
3896 tracing_stats_read(struct file *filp, char __user *ubuf,
3897                    size_t count, loff_t *ppos)
3898 {
3899         unsigned long cpu = (unsigned long)filp->private_data;
3900         struct trace_array *tr = &global_trace;
3901         struct trace_seq *s;
3902         unsigned long cnt;
3903
3904         s = kmalloc(sizeof(*s), GFP_KERNEL);
3905         if (!s)
3906                 return -ENOMEM;
3907
3908         trace_seq_init(s);
3909
3910         cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3911         trace_seq_printf(s, "entries: %ld\n", cnt);
3912
3913         cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3914         trace_seq_printf(s, "overrun: %ld\n", cnt);
3915
3916         cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3917         trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3918
3919         count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3920
3921         kfree(s);
3922
3923         return count;
3924 }
3925
3926 static const struct file_operations tracing_stats_fops = {
3927         .open           = tracing_open_generic,
3928         .read           = tracing_stats_read,
3929         .llseek         = generic_file_llseek,
3930 };
3931
3932 #ifdef CONFIG_DYNAMIC_FTRACE
3933
3934 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3935 {
3936         return 0;
3937 }
3938
3939 static ssize_t
3940 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3941                   size_t cnt, loff_t *ppos)
3942 {
3943         static char ftrace_dyn_info_buffer[1024];
3944         static DEFINE_MUTEX(dyn_info_mutex);
3945         unsigned long *p = filp->private_data;
3946         char *buf = ftrace_dyn_info_buffer;
3947         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3948         int r;
3949
3950         mutex_lock(&dyn_info_mutex);
3951         r = sprintf(buf, "%ld ", *p);
3952
3953         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3954         buf[r++] = '\n';
3955
3956         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3957
3958         mutex_unlock(&dyn_info_mutex);
3959
3960         return r;
3961 }
3962
3963 static const struct file_operations tracing_dyn_info_fops = {
3964         .open           = tracing_open_generic,
3965         .read           = tracing_read_dyn_info,
3966         .llseek         = generic_file_llseek,
3967 };
3968 #endif
3969
3970 static struct dentry *d_tracer;
3971
3972 struct dentry *tracing_init_dentry(void)
3973 {
3974         static int once;
3975
3976         if (d_tracer)
3977                 return d_tracer;
3978
3979         if (!debugfs_initialized())
3980                 return NULL;
3981
3982         d_tracer = debugfs_create_dir("tracing", NULL);
3983
3984         if (!d_tracer && !once) {
3985                 once = 1;
3986                 pr_warning("Could not create debugfs directory 'tracing'\n");
3987                 return NULL;
3988         }
3989
3990         return d_tracer;
3991 }
3992
3993 static struct dentry *d_percpu;
3994
3995 struct dentry *tracing_dentry_percpu(void)
3996 {
3997         static int once;
3998         struct dentry *d_tracer;
3999
4000         if (d_percpu)
4001                 return d_percpu;
4002
4003         d_tracer = tracing_init_dentry();
4004
4005         if (!d_tracer)
4006                 return NULL;
4007
4008         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4009
4010         if (!d_percpu && !once) {
4011                 once = 1;
4012                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4013                 return NULL;
4014         }
4015
4016         return d_percpu;
4017 }
4018
4019 static void tracing_init_debugfs_percpu(long cpu)
4020 {
4021         struct dentry *d_percpu = tracing_dentry_percpu();
4022         struct dentry *d_cpu;
4023         char cpu_dir[30]; /* 30 characters should be more than enough */
4024
4025         snprintf(cpu_dir, 30, "cpu%ld", cpu);
4026         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4027         if (!d_cpu) {
4028                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4029                 return;
4030         }
4031
4032         /* per cpu trace_pipe */
4033         trace_create_file("trace_pipe", 0444, d_cpu,
4034                         (void *) cpu, &tracing_pipe_fops);
4035
4036         /* per cpu trace */
4037         trace_create_file("trace", 0644, d_cpu,
4038                         (void *) cpu, &tracing_fops);
4039
4040         trace_create_file("trace_pipe_raw", 0444, d_cpu,
4041                         (void *) cpu, &tracing_buffers_fops);
4042
4043         trace_create_file("stats", 0444, d_cpu,
4044                         (void *) cpu, &tracing_stats_fops);
4045 }
4046
4047 #ifdef CONFIG_FTRACE_SELFTEST
4048 /* Let selftest have access to static functions in this file */
4049 #include "trace_selftest.c"
4050 #endif
4051
4052 struct trace_option_dentry {
4053         struct tracer_opt               *opt;
4054         struct tracer_flags             *flags;
4055         struct dentry                   *entry;
4056 };
4057
4058 static ssize_t
4059 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4060                         loff_t *ppos)
4061 {
4062         struct trace_option_dentry *topt = filp->private_data;
4063         char *buf;
4064
4065         if (topt->flags->val & topt->opt->bit)
4066                 buf = "1\n";
4067         else
4068                 buf = "0\n";
4069
4070         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4071 }
4072
4073 static ssize_t
4074 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4075                          loff_t *ppos)
4076 {
4077         struct trace_option_dentry *topt = filp->private_data;
4078         unsigned long val;
4079         char buf[64];
4080         int ret;
4081
4082         if (cnt >= sizeof(buf))
4083                 return -EINVAL;
4084
4085         if (copy_from_user(&buf, ubuf, cnt))
4086                 return -EFAULT;
4087
4088         buf[cnt] = 0;
4089
4090         ret = strict_strtoul(buf, 10, &val);
4091         if (ret < 0)
4092                 return ret;
4093
4094         if (val != 0 && val != 1)
4095                 return -EINVAL;
4096
4097         if (!!(topt->flags->val & topt->opt->bit) != val) {
4098                 mutex_lock(&trace_types_lock);
4099                 ret = __set_tracer_option(current_trace, topt->flags,
4100                                           topt->opt, !val);
4101                 mutex_unlock(&trace_types_lock);
4102                 if (ret)
4103                         return ret;
4104         }
4105
4106         *ppos += cnt;
4107
4108         return cnt;
4109 }
4110
4111
4112 static const struct file_operations trace_options_fops = {
4113         .open = tracing_open_generic,
4114         .read = trace_options_read,
4115         .write = trace_options_write,
4116         .llseek = generic_file_llseek,
4117 };
4118
4119 static ssize_t
4120 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4121                         loff_t *ppos)
4122 {
4123         long index = (long)filp->private_data;
4124         char *buf;
4125
4126         if (trace_flags & (1 << index))
4127                 buf = "1\n";
4128         else
4129                 buf = "0\n";
4130
4131         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4132 }
4133
4134 static ssize_t
4135 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4136                          loff_t *ppos)
4137 {
4138         long index = (long)filp->private_data;
4139         char buf[64];
4140         unsigned long val;
4141         int ret;
4142
4143         if (cnt >= sizeof(buf))
4144                 return -EINVAL;
4145
4146         if (copy_from_user(&buf, ubuf, cnt))
4147                 return -EFAULT;
4148
4149         buf[cnt] = 0;
4150
4151         ret = strict_strtoul(buf, 10, &val);
4152         if (ret < 0)
4153                 return ret;
4154
4155         if (val != 0 && val != 1)
4156                 return -EINVAL;
4157         set_tracer_flags(1 << index, val);
4158
4159         *ppos += cnt;
4160
4161         return cnt;
4162 }
4163
4164 static const struct file_operations trace_options_core_fops = {
4165         .open = tracing_open_generic,
4166         .read = trace_options_core_read,
4167         .write = trace_options_core_write,
4168         .llseek = generic_file_llseek,
4169 };
4170
4171 struct dentry *trace_create_file(const char *name,
4172                                  mode_t mode,
4173                                  struct dentry *parent,
4174                                  void *data,
4175                                  const struct file_operations *fops)
4176 {
4177         struct dentry *ret;
4178
4179         ret = debugfs_create_file(name, mode, parent, data, fops);
4180         if (!ret)
4181                 pr_warning("Could not create debugfs '%s' entry\n", name);
4182
4183         return ret;
4184 }
4185
4186
4187 static struct dentry *trace_options_init_dentry(void)
4188 {
4189         struct dentry *d_tracer;
4190         static struct dentry *t_options;
4191
4192         if (t_options)
4193                 return t_options;
4194
4195         d_tracer = tracing_init_dentry();
4196         if (!d_tracer)
4197                 return NULL;
4198
4199         t_options = debugfs_create_dir("options", d_tracer);
4200         if (!t_options) {
4201                 pr_warning("Could not create debugfs directory 'options'\n");
4202                 return NULL;
4203         }
4204
4205         return t_options;
4206 }
4207
4208 static void
4209 create_trace_option_file(struct trace_option_dentry *topt,
4210                          struct tracer_flags *flags,
4211                          struct tracer_opt *opt)
4212 {
4213         struct dentry *t_options;
4214
4215         t_options = trace_options_init_dentry();
4216         if (!t_options)
4217                 return;
4218
4219         topt->flags = flags;
4220         topt->opt = opt;
4221
4222         topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4223                                     &trace_options_fops);
4224
4225 }
4226
4227 static struct trace_option_dentry *
4228 create_trace_option_files(struct tracer *tracer)
4229 {
4230         struct trace_option_dentry *topts;
4231         struct tracer_flags *flags;
4232         struct tracer_opt *opts;
4233         int cnt;
4234
4235         if (!tracer)
4236                 return NULL;
4237
4238         flags = tracer->flags;
4239
4240         if (!flags || !flags->opts)
4241                 return NULL;
4242
4243         opts = flags->opts;
4244
4245         for (cnt = 0; opts[cnt].name; cnt++)
4246                 ;
4247
4248         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4249         if (!topts)
4250                 return NULL;
4251
4252         for (cnt = 0; opts[cnt].name; cnt++)
4253                 create_trace_option_file(&topts[cnt], flags,
4254                                          &opts[cnt]);
4255
4256         return topts;
4257 }
4258
4259 static void
4260 destroy_trace_option_files(struct trace_option_dentry *topts)
4261 {
4262         int cnt;
4263
4264         if (!topts)
4265                 return;
4266
4267         for (cnt = 0; topts[cnt].opt; cnt++) {
4268                 if (topts[cnt].entry)
4269                         debugfs_remove(topts[cnt].entry);
4270         }
4271
4272         kfree(topts);
4273 }
4274
4275 static struct dentry *
4276 create_trace_option_core_file(const char *option, long index)
4277 {
4278         struct dentry *t_options;
4279
4280         t_options = trace_options_init_dentry();
4281         if (!t_options)
4282                 return NULL;
4283
4284         return trace_create_file(option, 0644, t_options, (void *)index,
4285                                     &trace_options_core_fops);
4286 }
4287
4288 static __init void create_trace_options_dir(void)
4289 {
4290         struct dentry *t_options;
4291         int i;
4292
4293         t_options = trace_options_init_dentry();
4294         if (!t_options)
4295                 return;
4296
4297         for (i = 0; trace_options[i]; i++)
4298                 create_trace_option_core_file(trace_options[i], i);
4299 }
4300
4301 static __init int tracer_init_debugfs(void)
4302 {
4303         struct dentry *d_tracer;
4304         int cpu;
4305
4306         trace_access_lock_init();
4307
4308         d_tracer = tracing_init_dentry();
4309
4310         trace_create_file("tracing_enabled", 0644, d_tracer,
4311                         &global_trace, &tracing_ctrl_fops);
4312
4313         trace_create_file("trace_options", 0644, d_tracer,
4314                         NULL, &tracing_iter_fops);
4315
4316         trace_create_file("tracing_cpumask", 0644, d_tracer,
4317                         NULL, &tracing_cpumask_fops);
4318
4319         trace_create_file("trace", 0644, d_tracer,
4320                         (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4321
4322         trace_create_file("available_tracers", 0444, d_tracer,
4323                         &global_trace, &show_traces_fops);
4324
4325         trace_create_file("current_tracer", 0644, d_tracer,
4326                         &global_trace, &set_tracer_fops);
4327
4328 #ifdef CONFIG_TRACER_MAX_TRACE
4329         trace_create_file("tracing_max_latency", 0644, d_tracer,
4330                         &tracing_max_latency, &tracing_max_lat_fops);
4331 #endif
4332
4333         trace_create_file("tracing_thresh", 0644, d_tracer,
4334                         &tracing_thresh, &tracing_max_lat_fops);
4335
4336         trace_create_file("README", 0444, d_tracer,
4337                         NULL, &tracing_readme_fops);
4338
4339         trace_create_file("trace_pipe", 0444, d_tracer,
4340                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4341
4342         trace_create_file("buffer_size_kb", 0644, d_tracer,
4343                         &global_trace, &tracing_entries_fops);
4344
4345         trace_create_file("trace_marker", 0220, d_tracer,
4346                         NULL, &tracing_mark_fops);
4347
4348         trace_create_file("saved_cmdlines", 0444, d_tracer,
4349                         NULL, &tracing_saved_cmdlines_fops);
4350
4351         trace_create_file("trace_clock", 0644, d_tracer, NULL,
4352                           &trace_clock_fops);
4353
4354 #ifdef CONFIG_DYNAMIC_FTRACE
4355         trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4356                         &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4357 #endif
4358
4359         create_trace_options_dir();
4360
4361         for_each_tracing_cpu(cpu)
4362                 tracing_init_debugfs_percpu(cpu);
4363
4364         return 0;
4365 }
4366
4367 static int trace_panic_handler(struct notifier_block *this,
4368                                unsigned long event, void *unused)
4369 {
4370         if (ftrace_dump_on_oops)
4371                 ftrace_dump(ftrace_dump_on_oops);
4372         return NOTIFY_OK;
4373 }
4374
4375 static struct notifier_block trace_panic_notifier = {
4376         .notifier_call  = trace_panic_handler,
4377         .next           = NULL,
4378         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
4379 };
4380
4381 static int trace_die_handler(struct notifier_block *self,
4382                              unsigned long val,
4383                              void *data)
4384 {
4385         switch (val) {
4386         case DIE_OOPS:
4387                 if (ftrace_dump_on_oops)
4388                         ftrace_dump(ftrace_dump_on_oops);
4389                 break;
4390         default:
4391                 break;
4392         }
4393         return NOTIFY_OK;
4394 }
4395
4396 static struct notifier_block trace_die_notifier = {
4397         .notifier_call = trace_die_handler,
4398         .priority = 200
4399 };
4400
4401 /*
4402  * printk is set to max of 1024, we really don't need it that big.
4403  * Nothing should be printing 1000 characters anyway.
4404  */
4405 #define TRACE_MAX_PRINT         1000
4406
4407 /*
4408  * Define here KERN_TRACE so that we have one place to modify
4409  * it if we decide to change what log level the ftrace dump
4410  * should be at.
4411  */
4412 #define KERN_TRACE              KERN_EMERG
4413
4414 void
4415 trace_printk_seq(struct trace_seq *s)
4416 {
4417         /* Probably should print a warning here. */
4418         if (s->len >= 1000)
4419                 s->len = 1000;
4420
4421         /* should be zero ended, but we are paranoid. */
4422         s->buffer[s->len] = 0;
4423
4424         printk(KERN_TRACE "%s", s->buffer);
4425
4426         trace_seq_init(s);
4427 }
4428
4429 void trace_init_global_iter(struct trace_iterator *iter)
4430 {
4431         iter->tr = &global_trace;
4432         iter->trace = current_trace;
4433         iter->cpu_file = TRACE_PIPE_ALL_CPU;
4434 }
4435
4436 static void
4437 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4438 {
4439         static arch_spinlock_t ftrace_dump_lock =
4440                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4441         /* use static because iter can be a bit big for the stack */
4442         static struct trace_iterator iter;
4443         unsigned int old_userobj;
4444         static int dump_ran;
4445         unsigned long flags;
4446         int cnt = 0, cpu;
4447
4448         /* only one dump */
4449         local_irq_save(flags);
4450         arch_spin_lock(&ftrace_dump_lock);
4451         if (dump_ran)
4452                 goto out;
4453
4454         dump_ran = 1;
4455
4456         tracing_off();
4457
4458         if (disable_tracing)
4459                 ftrace_kill();
4460
4461         trace_init_global_iter(&iter);
4462
4463         for_each_tracing_cpu(cpu) {
4464                 atomic_inc(&iter.tr->data[cpu]->disabled);
4465         }
4466
4467         old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4468
4469         /* don't look at user memory in panic mode */
4470         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4471
4472         /* Simulate the iterator */
4473         iter.tr = &global_trace;
4474         iter.trace = current_trace;
4475
4476         switch (oops_dump_mode) {
4477         case DUMP_ALL:
4478                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4479                 break;
4480         case DUMP_ORIG:
4481                 iter.cpu_file = raw_smp_processor_id();
4482                 break;
4483         case DUMP_NONE:
4484                 goto out_enable;
4485         default:
4486                 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4487                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4488         }
4489
4490         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4491
4492         /*
4493          * We need to stop all tracing on all CPUS to read the
4494          * the next buffer. This is a bit expensive, but is
4495          * not done often. We fill all what we can read,
4496          * and then release the locks again.
4497          */
4498
4499         while (!trace_empty(&iter)) {
4500
4501                 if (!cnt)
4502                         printk(KERN_TRACE "---------------------------------\n");
4503
4504                 cnt++;
4505
4506                 /* reset all but tr, trace, and overruns */
4507                 memset(&iter.seq, 0,
4508                        sizeof(struct trace_iterator) -
4509                        offsetof(struct trace_iterator, seq));
4510                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4511                 iter.pos = -1;
4512
4513                 if (trace_find_next_entry_inc(&iter) != NULL) {
4514                         int ret;
4515
4516                         ret = print_trace_line(&iter);
4517                         if (ret != TRACE_TYPE_NO_CONSUME)
4518                                 trace_consume(&iter);
4519                 }
4520
4521                 trace_printk_seq(&iter.seq);
4522         }
4523
4524         if (!cnt)
4525                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4526         else
4527                 printk(KERN_TRACE "---------------------------------\n");
4528
4529  out_enable:
4530         /* Re-enable tracing if requested */
4531         if (!disable_tracing) {
4532                 trace_flags |= old_userobj;
4533
4534                 for_each_tracing_cpu(cpu) {
4535                         atomic_dec(&iter.tr->data[cpu]->disabled);
4536                 }
4537                 tracing_on();
4538         }
4539
4540  out:
4541         arch_spin_unlock(&ftrace_dump_lock);
4542         local_irq_restore(flags);
4543 }
4544
4545 /* By default: disable tracing after the dump */
4546 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4547 {
4548         __ftrace_dump(true, oops_dump_mode);
4549 }
4550
4551 __init static int tracer_alloc_buffers(void)
4552 {
4553         int ring_buf_size;
4554         int i;
4555         int ret = -ENOMEM;
4556
4557         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4558                 goto out;
4559
4560         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4561                 goto out_free_buffer_mask;
4562
4563         /* To save memory, keep the ring buffer size to its minimum */
4564         if (ring_buffer_expanded)
4565                 ring_buf_size = trace_buf_size;
4566         else
4567                 ring_buf_size = 1;
4568
4569         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4570         cpumask_copy(tracing_cpumask, cpu_all_mask);
4571
4572         /* TODO: make the number of buffers hot pluggable with CPUS */
4573         global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4574                                                    TRACE_BUFFER_FLAGS);
4575         if (!global_trace.buffer) {
4576                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4577                 WARN_ON(1);
4578                 goto out_free_cpumask;
4579         }
4580         global_trace.entries = ring_buffer_size(global_trace.buffer);
4581
4582
4583 #ifdef CONFIG_TRACER_MAX_TRACE
4584         max_tr.buffer = ring_buffer_alloc(1, TRACE_BUFFER_FLAGS);
4585         if (!max_tr.buffer) {
4586                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4587                 WARN_ON(1);
4588                 ring_buffer_free(global_trace.buffer);
4589                 goto out_free_cpumask;
4590         }
4591         max_tr.entries = 1;
4592 #endif
4593
4594         /* Allocate the first page for all buffers */
4595         for_each_tracing_cpu(i) {
4596                 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4597                 max_tr.data[i] = &per_cpu(max_tr_data, i);
4598         }
4599
4600         trace_init_cmdlines();
4601
4602         register_tracer(&nop_trace);
4603         current_trace = &nop_trace;
4604         /* All seems OK, enable tracing */
4605         tracing_disabled = 0;
4606
4607         atomic_notifier_chain_register(&panic_notifier_list,
4608                                        &trace_panic_notifier);
4609
4610         register_die_notifier(&trace_die_notifier);
4611
4612         return 0;
4613
4614 out_free_cpumask:
4615         free_cpumask_var(tracing_cpumask);
4616 out_free_buffer_mask:
4617         free_cpumask_var(tracing_buffer_mask);
4618 out:
4619         return ret;
4620 }
4621
4622 __init static int clear_boot_tracer(void)
4623 {
4624         /*
4625          * The default tracer at boot buffer is an init section.
4626          * This function is called in lateinit. If we did not
4627          * find the boot tracer, then clear it out, to prevent
4628          * later registration from accessing the buffer that is
4629          * about to be freed.
4630          */
4631         if (!default_bootup_tracer)
4632                 return 0;
4633
4634         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4635                default_bootup_tracer);
4636         default_bootup_tracer = NULL;
4637
4638         return 0;
4639 }
4640
4641 early_initcall(tracer_alloc_buffers);
4642 fs_initcall(tracer_init_debugfs);
4643 late_initcall(clear_boot_tracer);