Merge branch 'tracing-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[pandora-kernel.git] / kernel / trace / trace.c
1 /*
2  * ring buffer based function tracer
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally taken from the RT patch by:
8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code from the latency_tracer, that is:
11  *  Copyright (C) 2004-2006 Ingo Molnar
12  *  Copyright (C) 2004 William Lee Irwin III
13  */
14 #include <linux/ring_buffer.h>
15 #include <linux/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/ctype.h>
36 #include <linux/init.h>
37 #include <linux/poll.h>
38 #include <linux/gfp.h>
39 #include <linux/fs.h>
40
41 #include "trace.h"
42 #include "trace_output.h"
43
44 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
45
46 /*
47  * On boot up, the ring buffer is set to the minimum size, so that
48  * we do not waste memory on systems that are not using tracing.
49  */
50 int ring_buffer_expanded;
51
52 /*
53  * We need to change this state when a selftest is running.
54  * A selftest will lurk into the ring-buffer to count the
55  * entries inserted during the selftest although some concurrent
56  * insertions into the ring-buffer such as trace_printk could occurred
57  * at the same time, giving false positive or negative results.
58  */
59 static bool __read_mostly tracing_selftest_running;
60
61 /*
62  * If a tracer is running, we do not want to run SELFTEST.
63  */
64 bool __read_mostly tracing_selftest_disabled;
65
66 /* For tracers that don't implement custom flags */
67 static struct tracer_opt dummy_tracer_opt[] = {
68         { }
69 };
70
71 static struct tracer_flags dummy_tracer_flags = {
72         .val = 0,
73         .opts = dummy_tracer_opt
74 };
75
76 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
77 {
78         return 0;
79 }
80
81 /*
82  * Kill all tracing for good (never come back).
83  * It is initialized to 1 but will turn to zero if the initialization
84  * of the tracer is successful. But that is the only place that sets
85  * this back to zero.
86  */
87 static int tracing_disabled = 1;
88
89 DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
90
91 static inline void ftrace_disable_cpu(void)
92 {
93         preempt_disable();
94         local_inc(&__get_cpu_var(ftrace_cpu_disabled));
95 }
96
97 static inline void ftrace_enable_cpu(void)
98 {
99         local_dec(&__get_cpu_var(ftrace_cpu_disabled));
100         preempt_enable();
101 }
102
103 static cpumask_var_t __read_mostly      tracing_buffer_mask;
104
105 /* Define which cpu buffers are currently read in trace_pipe */
106 static cpumask_var_t                    tracing_reader_cpumask;
107
108 #define for_each_tracing_cpu(cpu)       \
109         for_each_cpu(cpu, tracing_buffer_mask)
110
111 /*
112  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
113  *
114  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
115  * is set, then ftrace_dump is called. This will output the contents
116  * of the ftrace buffers to the console.  This is very useful for
117  * capturing traces that lead to crashes and outputing it to a
118  * serial console.
119  *
120  * It is default off, but you can enable it with either specifying
121  * "ftrace_dump_on_oops" in the kernel command line, or setting
122  * /proc/sys/kernel/ftrace_dump_on_oops to true.
123  */
124 int ftrace_dump_on_oops;
125
126 static int tracing_set_tracer(const char *buf);
127
128 #define MAX_TRACER_SIZE         100
129 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
130 static char *default_bootup_tracer;
131
132 static int __init set_ftrace(char *str)
133 {
134         strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
135         default_bootup_tracer = bootup_tracer_buf;
136         /* We are using ftrace early, expand it */
137         ring_buffer_expanded = 1;
138         return 1;
139 }
140 __setup("ftrace=", set_ftrace);
141
142 static int __init set_ftrace_dump_on_oops(char *str)
143 {
144         ftrace_dump_on_oops = 1;
145         return 1;
146 }
147 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
148
149 unsigned long long ns2usecs(cycle_t nsec)
150 {
151         nsec += 500;
152         do_div(nsec, 1000);
153         return nsec;
154 }
155
156 /*
157  * The global_trace is the descriptor that holds the tracing
158  * buffers for the live tracing. For each CPU, it contains
159  * a link list of pages that will store trace entries. The
160  * page descriptor of the pages in the memory is used to hold
161  * the link list by linking the lru item in the page descriptor
162  * to each of the pages in the buffer per CPU.
163  *
164  * For each active CPU there is a data field that holds the
165  * pages for the buffer for that CPU. Each CPU has the same number
166  * of pages allocated for its buffer.
167  */
168 static struct trace_array       global_trace;
169
170 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
171
172 int filter_current_check_discard(struct ring_buffer *buffer,
173                                  struct ftrace_event_call *call, void *rec,
174                                  struct ring_buffer_event *event)
175 {
176         return filter_check_discard(call, rec, buffer, event);
177 }
178 EXPORT_SYMBOL_GPL(filter_current_check_discard);
179
180 cycle_t ftrace_now(int cpu)
181 {
182         u64 ts;
183
184         /* Early boot up does not have a buffer yet */
185         if (!global_trace.buffer)
186                 return trace_clock_local();
187
188         ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
189         ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
190
191         return ts;
192 }
193
194 /*
195  * The max_tr is used to snapshot the global_trace when a maximum
196  * latency is reached. Some tracers will use this to store a maximum
197  * trace while it continues examining live traces.
198  *
199  * The buffers for the max_tr are set up the same as the global_trace.
200  * When a snapshot is taken, the link list of the max_tr is swapped
201  * with the link list of the global_trace and the buffers are reset for
202  * the global_trace so the tracing can continue.
203  */
204 static struct trace_array       max_tr;
205
206 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
207
208 /* tracer_enabled is used to toggle activation of a tracer */
209 static int                      tracer_enabled = 1;
210
211 /**
212  * tracing_is_enabled - return tracer_enabled status
213  *
214  * This function is used by other tracers to know the status
215  * of the tracer_enabled flag.  Tracers may use this function
216  * to know if it should enable their features when starting
217  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
218  */
219 int tracing_is_enabled(void)
220 {
221         return tracer_enabled;
222 }
223
224 /*
225  * trace_buf_size is the size in bytes that is allocated
226  * for a buffer. Note, the number of bytes is always rounded
227  * to page size.
228  *
229  * This number is purposely set to a low number of 16384.
230  * If the dump on oops happens, it will be much appreciated
231  * to not have to wait for all that output. Anyway this can be
232  * boot time and run time configurable.
233  */
234 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
235
236 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
237
238 /* trace_types holds a link list of available tracers. */
239 static struct tracer            *trace_types __read_mostly;
240
241 /* current_trace points to the tracer that is currently active */
242 static struct tracer            *current_trace __read_mostly;
243
244 /*
245  * trace_types_lock is used to protect the trace_types list.
246  * This lock is also used to keep user access serialized.
247  * Accesses from userspace will grab this lock while userspace
248  * activities happen inside the kernel.
249  */
250 static DEFINE_MUTEX(trace_types_lock);
251
252 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
253 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
254
255 /* trace_flags holds trace_options default values */
256 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
257         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
258         TRACE_ITER_GRAPH_TIME;
259
260 static int trace_stop_count;
261 static DEFINE_SPINLOCK(tracing_start_lock);
262
263 /**
264  * trace_wake_up - wake up tasks waiting for trace input
265  *
266  * Simply wakes up any task that is blocked on the trace_wait
267  * queue. These is used with trace_poll for tasks polling the trace.
268  */
269 void trace_wake_up(void)
270 {
271         int cpu;
272
273         if (trace_flags & TRACE_ITER_BLOCK)
274                 return;
275         /*
276          * The runqueue_is_locked() can fail, but this is the best we
277          * have for now:
278          */
279         cpu = get_cpu();
280         if (!runqueue_is_locked(cpu))
281                 wake_up(&trace_wait);
282         put_cpu();
283 }
284
285 static int __init set_buf_size(char *str)
286 {
287         unsigned long buf_size;
288
289         if (!str)
290                 return 0;
291         buf_size = memparse(str, &str);
292         /* nr_entries can not be zero */
293         if (buf_size == 0)
294                 return 0;
295         trace_buf_size = buf_size;
296         return 1;
297 }
298 __setup("trace_buf_size=", set_buf_size);
299
300 unsigned long nsecs_to_usecs(unsigned long nsecs)
301 {
302         return nsecs / 1000;
303 }
304
305 /* These must match the bit postions in trace_iterator_flags */
306 static const char *trace_options[] = {
307         "print-parent",
308         "sym-offset",
309         "sym-addr",
310         "verbose",
311         "raw",
312         "hex",
313         "bin",
314         "block",
315         "stacktrace",
316         "sched-tree",
317         "trace_printk",
318         "ftrace_preempt",
319         "branch",
320         "annotate",
321         "userstacktrace",
322         "sym-userobj",
323         "printk-msg-only",
324         "context-info",
325         "latency-format",
326         "sleep-time",
327         "graph-time",
328         NULL
329 };
330
331 static struct {
332         u64 (*func)(void);
333         const char *name;
334 } trace_clocks[] = {
335         { trace_clock_local,    "local" },
336         { trace_clock_global,   "global" },
337 };
338
339 int trace_clock_id;
340
341 /*
342  * trace_parser_get_init - gets the buffer for trace parser
343  */
344 int trace_parser_get_init(struct trace_parser *parser, int size)
345 {
346         memset(parser, 0, sizeof(*parser));
347
348         parser->buffer = kmalloc(size, GFP_KERNEL);
349         if (!parser->buffer)
350                 return 1;
351
352         parser->size = size;
353         return 0;
354 }
355
356 /*
357  * trace_parser_put - frees the buffer for trace parser
358  */
359 void trace_parser_put(struct trace_parser *parser)
360 {
361         kfree(parser->buffer);
362 }
363
364 /*
365  * trace_get_user - reads the user input string separated by  space
366  * (matched by isspace(ch))
367  *
368  * For each string found the 'struct trace_parser' is updated,
369  * and the function returns.
370  *
371  * Returns number of bytes read.
372  *
373  * See kernel/trace/trace.h for 'struct trace_parser' details.
374  */
375 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
376         size_t cnt, loff_t *ppos)
377 {
378         char ch;
379         size_t read = 0;
380         ssize_t ret;
381
382         if (!*ppos)
383                 trace_parser_clear(parser);
384
385         ret = get_user(ch, ubuf++);
386         if (ret)
387                 goto out;
388
389         read++;
390         cnt--;
391
392         /*
393          * The parser is not finished with the last write,
394          * continue reading the user input without skipping spaces.
395          */
396         if (!parser->cont) {
397                 /* skip white space */
398                 while (cnt && isspace(ch)) {
399                         ret = get_user(ch, ubuf++);
400                         if (ret)
401                                 goto out;
402                         read++;
403                         cnt--;
404                 }
405
406                 /* only spaces were written */
407                 if (isspace(ch)) {
408                         *ppos += read;
409                         ret = read;
410                         goto out;
411                 }
412
413                 parser->idx = 0;
414         }
415
416         /* read the non-space input */
417         while (cnt && !isspace(ch)) {
418                 if (parser->idx < parser->size - 1)
419                         parser->buffer[parser->idx++] = ch;
420                 else {
421                         ret = -EINVAL;
422                         goto out;
423                 }
424                 ret = get_user(ch, ubuf++);
425                 if (ret)
426                         goto out;
427                 read++;
428                 cnt--;
429         }
430
431         /* We either got finished input or we have to wait for another call. */
432         if (isspace(ch)) {
433                 parser->buffer[parser->idx] = 0;
434                 parser->cont = false;
435         } else {
436                 parser->cont = true;
437                 parser->buffer[parser->idx++] = ch;
438         }
439
440         *ppos += read;
441         ret = read;
442
443 out:
444         return ret;
445 }
446
447 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
448 {
449         int len;
450         int ret;
451
452         if (!cnt)
453                 return 0;
454
455         if (s->len <= s->readpos)
456                 return -EBUSY;
457
458         len = s->len - s->readpos;
459         if (cnt > len)
460                 cnt = len;
461         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
462         if (ret == cnt)
463                 return -EFAULT;
464
465         cnt -= ret;
466
467         s->readpos += cnt;
468         return cnt;
469 }
470
471 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
472 {
473         int len;
474         void *ret;
475
476         if (s->len <= s->readpos)
477                 return -EBUSY;
478
479         len = s->len - s->readpos;
480         if (cnt > len)
481                 cnt = len;
482         ret = memcpy(buf, s->buffer + s->readpos, cnt);
483         if (!ret)
484                 return -EFAULT;
485
486         s->readpos += cnt;
487         return cnt;
488 }
489
490 /*
491  * ftrace_max_lock is used to protect the swapping of buffers
492  * when taking a max snapshot. The buffers themselves are
493  * protected by per_cpu spinlocks. But the action of the swap
494  * needs its own lock.
495  *
496  * This is defined as a raw_spinlock_t in order to help
497  * with performance when lockdep debugging is enabled.
498  *
499  * It is also used in other places outside the update_max_tr
500  * so it needs to be defined outside of the
501  * CONFIG_TRACER_MAX_TRACE.
502  */
503 static raw_spinlock_t ftrace_max_lock =
504         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
505
506 #ifdef CONFIG_TRACER_MAX_TRACE
507 unsigned long __read_mostly     tracing_max_latency;
508 unsigned long __read_mostly     tracing_thresh;
509
510 /*
511  * Copy the new maximum trace into the separate maximum-trace
512  * structure. (this way the maximum trace is permanently saved,
513  * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
514  */
515 static void
516 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
517 {
518         struct trace_array_cpu *data = tr->data[cpu];
519         struct trace_array_cpu *max_data = tr->data[cpu];
520
521         max_tr.cpu = cpu;
522         max_tr.time_start = data->preempt_timestamp;
523
524         max_data = max_tr.data[cpu];
525         max_data->saved_latency = tracing_max_latency;
526         max_data->critical_start = data->critical_start;
527         max_data->critical_end = data->critical_end;
528
529         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
530         max_data->pid = tsk->pid;
531         max_data->uid = task_uid(tsk);
532         max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
533         max_data->policy = tsk->policy;
534         max_data->rt_priority = tsk->rt_priority;
535
536         /* record this tasks comm */
537         tracing_record_cmdline(tsk);
538 }
539
540 /**
541  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
542  * @tr: tracer
543  * @tsk: the task with the latency
544  * @cpu: The cpu that initiated the trace.
545  *
546  * Flip the buffers between the @tr and the max_tr and record information
547  * about which task was the cause of this latency.
548  */
549 void
550 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
551 {
552         struct ring_buffer *buf = tr->buffer;
553
554         if (trace_stop_count)
555                 return;
556
557         WARN_ON_ONCE(!irqs_disabled());
558         __raw_spin_lock(&ftrace_max_lock);
559
560         tr->buffer = max_tr.buffer;
561         max_tr.buffer = buf;
562
563         __update_max_tr(tr, tsk, cpu);
564         __raw_spin_unlock(&ftrace_max_lock);
565 }
566
567 /**
568  * update_max_tr_single - only copy one trace over, and reset the rest
569  * @tr - tracer
570  * @tsk - task with the latency
571  * @cpu - the cpu of the buffer to copy.
572  *
573  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
574  */
575 void
576 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
577 {
578         int ret;
579
580         if (trace_stop_count)
581                 return;
582
583         WARN_ON_ONCE(!irqs_disabled());
584         __raw_spin_lock(&ftrace_max_lock);
585
586         ftrace_disable_cpu();
587
588         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
589
590         if (ret == -EBUSY) {
591                 /*
592                  * We failed to swap the buffer due to a commit taking
593                  * place on this CPU. We fail to record, but we reset
594                  * the max trace buffer (no one writes directly to it)
595                  * and flag that it failed.
596                  */
597                 trace_array_printk(&max_tr, _THIS_IP_,
598                         "Failed to swap buffers due to commit in progress\n");
599         }
600
601         ftrace_enable_cpu();
602
603         WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
604
605         __update_max_tr(tr, tsk, cpu);
606         __raw_spin_unlock(&ftrace_max_lock);
607 }
608 #endif /* CONFIG_TRACER_MAX_TRACE */
609
610 /**
611  * register_tracer - register a tracer with the ftrace system.
612  * @type - the plugin for the tracer
613  *
614  * Register a new plugin tracer.
615  */
616 int register_tracer(struct tracer *type)
617 __releases(kernel_lock)
618 __acquires(kernel_lock)
619 {
620         struct tracer *t;
621         int ret = 0;
622
623         if (!type->name) {
624                 pr_info("Tracer must have a name\n");
625                 return -1;
626         }
627
628         if (strlen(type->name) > MAX_TRACER_SIZE) {
629                 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
630                 return -1;
631         }
632
633         /*
634          * When this gets called we hold the BKL which means that
635          * preemption is disabled. Various trace selftests however
636          * need to disable and enable preemption for successful tests.
637          * So we drop the BKL here and grab it after the tests again.
638          */
639         unlock_kernel();
640         mutex_lock(&trace_types_lock);
641
642         tracing_selftest_running = true;
643
644         for (t = trace_types; t; t = t->next) {
645                 if (strcmp(type->name, t->name) == 0) {
646                         /* already found */
647                         pr_info("Tracer %s already registered\n",
648                                 type->name);
649                         ret = -1;
650                         goto out;
651                 }
652         }
653
654         if (!type->set_flag)
655                 type->set_flag = &dummy_set_flag;
656         if (!type->flags)
657                 type->flags = &dummy_tracer_flags;
658         else
659                 if (!type->flags->opts)
660                         type->flags->opts = dummy_tracer_opt;
661         if (!type->wait_pipe)
662                 type->wait_pipe = default_wait_pipe;
663
664
665 #ifdef CONFIG_FTRACE_STARTUP_TEST
666         if (type->selftest && !tracing_selftest_disabled) {
667                 struct tracer *saved_tracer = current_trace;
668                 struct trace_array *tr = &global_trace;
669
670                 /*
671                  * Run a selftest on this tracer.
672                  * Here we reset the trace buffer, and set the current
673                  * tracer to be this tracer. The tracer can then run some
674                  * internal tracing to verify that everything is in order.
675                  * If we fail, we do not register this tracer.
676                  */
677                 tracing_reset_online_cpus(tr);
678
679                 current_trace = type;
680                 /* the test is responsible for initializing and enabling */
681                 pr_info("Testing tracer %s: ", type->name);
682                 ret = type->selftest(type, tr);
683                 /* the test is responsible for resetting too */
684                 current_trace = saved_tracer;
685                 if (ret) {
686                         printk(KERN_CONT "FAILED!\n");
687                         goto out;
688                 }
689                 /* Only reset on passing, to avoid touching corrupted buffers */
690                 tracing_reset_online_cpus(tr);
691
692                 printk(KERN_CONT "PASSED\n");
693         }
694 #endif
695
696         type->next = trace_types;
697         trace_types = type;
698
699  out:
700         tracing_selftest_running = false;
701         mutex_unlock(&trace_types_lock);
702
703         if (ret || !default_bootup_tracer)
704                 goto out_unlock;
705
706         if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
707                 goto out_unlock;
708
709         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
710         /* Do we want this tracer to start on bootup? */
711         tracing_set_tracer(type->name);
712         default_bootup_tracer = NULL;
713         /* disable other selftests, since this will break it. */
714         tracing_selftest_disabled = 1;
715 #ifdef CONFIG_FTRACE_STARTUP_TEST
716         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
717                type->name);
718 #endif
719
720  out_unlock:
721         lock_kernel();
722         return ret;
723 }
724
725 void unregister_tracer(struct tracer *type)
726 {
727         struct tracer **t;
728
729         mutex_lock(&trace_types_lock);
730         for (t = &trace_types; *t; t = &(*t)->next) {
731                 if (*t == type)
732                         goto found;
733         }
734         pr_info("Tracer %s not registered\n", type->name);
735         goto out;
736
737  found:
738         *t = (*t)->next;
739
740         if (type == current_trace && tracer_enabled) {
741                 tracer_enabled = 0;
742                 tracing_stop();
743                 if (current_trace->stop)
744                         current_trace->stop(&global_trace);
745                 current_trace = &nop_trace;
746         }
747 out:
748         mutex_unlock(&trace_types_lock);
749 }
750
751 static void __tracing_reset(struct trace_array *tr, int cpu)
752 {
753         ftrace_disable_cpu();
754         ring_buffer_reset_cpu(tr->buffer, cpu);
755         ftrace_enable_cpu();
756 }
757
758 void tracing_reset(struct trace_array *tr, int cpu)
759 {
760         struct ring_buffer *buffer = tr->buffer;
761
762         ring_buffer_record_disable(buffer);
763
764         /* Make sure all commits have finished */
765         synchronize_sched();
766         __tracing_reset(tr, cpu);
767
768         ring_buffer_record_enable(buffer);
769 }
770
771 void tracing_reset_online_cpus(struct trace_array *tr)
772 {
773         struct ring_buffer *buffer = tr->buffer;
774         int cpu;
775
776         ring_buffer_record_disable(buffer);
777
778         /* Make sure all commits have finished */
779         synchronize_sched();
780
781         tr->time_start = ftrace_now(tr->cpu);
782
783         for_each_online_cpu(cpu)
784                 __tracing_reset(tr, cpu);
785
786         ring_buffer_record_enable(buffer);
787 }
788
789 void tracing_reset_current(int cpu)
790 {
791         tracing_reset(&global_trace, cpu);
792 }
793
794 void tracing_reset_current_online_cpus(void)
795 {
796         tracing_reset_online_cpus(&global_trace);
797 }
798
799 #define SAVED_CMDLINES 128
800 #define NO_CMDLINE_MAP UINT_MAX
801 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
802 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
803 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
804 static int cmdline_idx;
805 static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
806
807 /* temporary disable recording */
808 static atomic_t trace_record_cmdline_disabled __read_mostly;
809
810 static void trace_init_cmdlines(void)
811 {
812         memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
813         memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
814         cmdline_idx = 0;
815 }
816
817 int is_tracing_stopped(void)
818 {
819         return trace_stop_count;
820 }
821
822 /**
823  * ftrace_off_permanent - disable all ftrace code permanently
824  *
825  * This should only be called when a serious anomally has
826  * been detected.  This will turn off the function tracing,
827  * ring buffers, and other tracing utilites. It takes no
828  * locks and can be called from any context.
829  */
830 void ftrace_off_permanent(void)
831 {
832         tracing_disabled = 1;
833         ftrace_stop();
834         tracing_off_permanent();
835 }
836
837 /**
838  * tracing_start - quick start of the tracer
839  *
840  * If tracing is enabled but was stopped by tracing_stop,
841  * this will start the tracer back up.
842  */
843 void tracing_start(void)
844 {
845         struct ring_buffer *buffer;
846         unsigned long flags;
847
848         if (tracing_disabled)
849                 return;
850
851         spin_lock_irqsave(&tracing_start_lock, flags);
852         if (--trace_stop_count) {
853                 if (trace_stop_count < 0) {
854                         /* Someone screwed up their debugging */
855                         WARN_ON_ONCE(1);
856                         trace_stop_count = 0;
857                 }
858                 goto out;
859         }
860
861
862         buffer = global_trace.buffer;
863         if (buffer)
864                 ring_buffer_record_enable(buffer);
865
866         buffer = max_tr.buffer;
867         if (buffer)
868                 ring_buffer_record_enable(buffer);
869
870         ftrace_start();
871  out:
872         spin_unlock_irqrestore(&tracing_start_lock, flags);
873 }
874
875 /**
876  * tracing_stop - quick stop of the tracer
877  *
878  * Light weight way to stop tracing. Use in conjunction with
879  * tracing_start.
880  */
881 void tracing_stop(void)
882 {
883         struct ring_buffer *buffer;
884         unsigned long flags;
885
886         ftrace_stop();
887         spin_lock_irqsave(&tracing_start_lock, flags);
888         if (trace_stop_count++)
889                 goto out;
890
891         buffer = global_trace.buffer;
892         if (buffer)
893                 ring_buffer_record_disable(buffer);
894
895         buffer = max_tr.buffer;
896         if (buffer)
897                 ring_buffer_record_disable(buffer);
898
899  out:
900         spin_unlock_irqrestore(&tracing_start_lock, flags);
901 }
902
903 void trace_stop_cmdline_recording(void);
904
905 static void trace_save_cmdline(struct task_struct *tsk)
906 {
907         unsigned pid, idx;
908
909         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
910                 return;
911
912         /*
913          * It's not the end of the world if we don't get
914          * the lock, but we also don't want to spin
915          * nor do we want to disable interrupts,
916          * so if we miss here, then better luck next time.
917          */
918         if (!__raw_spin_trylock(&trace_cmdline_lock))
919                 return;
920
921         idx = map_pid_to_cmdline[tsk->pid];
922         if (idx == NO_CMDLINE_MAP) {
923                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
924
925                 /*
926                  * Check whether the cmdline buffer at idx has a pid
927                  * mapped. We are going to overwrite that entry so we
928                  * need to clear the map_pid_to_cmdline. Otherwise we
929                  * would read the new comm for the old pid.
930                  */
931                 pid = map_cmdline_to_pid[idx];
932                 if (pid != NO_CMDLINE_MAP)
933                         map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
934
935                 map_cmdline_to_pid[idx] = tsk->pid;
936                 map_pid_to_cmdline[tsk->pid] = idx;
937
938                 cmdline_idx = idx;
939         }
940
941         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
942
943         __raw_spin_unlock(&trace_cmdline_lock);
944 }
945
946 void trace_find_cmdline(int pid, char comm[])
947 {
948         unsigned map;
949
950         if (!pid) {
951                 strcpy(comm, "<idle>");
952                 return;
953         }
954
955         if (pid > PID_MAX_DEFAULT) {
956                 strcpy(comm, "<...>");
957                 return;
958         }
959
960         preempt_disable();
961         __raw_spin_lock(&trace_cmdline_lock);
962         map = map_pid_to_cmdline[pid];
963         if (map != NO_CMDLINE_MAP)
964                 strcpy(comm, saved_cmdlines[map]);
965         else
966                 strcpy(comm, "<...>");
967
968         __raw_spin_unlock(&trace_cmdline_lock);
969         preempt_enable();
970 }
971
972 void tracing_record_cmdline(struct task_struct *tsk)
973 {
974         if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
975             !tracing_is_on())
976                 return;
977
978         trace_save_cmdline(tsk);
979 }
980
981 void
982 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
983                              int pc)
984 {
985         struct task_struct *tsk = current;
986
987         entry->preempt_count            = pc & 0xff;
988         entry->pid                      = (tsk) ? tsk->pid : 0;
989         entry->lock_depth               = (tsk) ? tsk->lock_depth : 0;
990         entry->flags =
991 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
992                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
993 #else
994                 TRACE_FLAG_IRQS_NOSUPPORT |
995 #endif
996                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
997                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
998                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
999 }
1000 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1001
1002 struct ring_buffer_event *
1003 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1004                           int type,
1005                           unsigned long len,
1006                           unsigned long flags, int pc)
1007 {
1008         struct ring_buffer_event *event;
1009
1010         event = ring_buffer_lock_reserve(buffer, len);
1011         if (event != NULL) {
1012                 struct trace_entry *ent = ring_buffer_event_data(event);
1013
1014                 tracing_generic_entry_update(ent, flags, pc);
1015                 ent->type = type;
1016         }
1017
1018         return event;
1019 }
1020
1021 static inline void
1022 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1023                              struct ring_buffer_event *event,
1024                              unsigned long flags, int pc,
1025                              int wake)
1026 {
1027         ring_buffer_unlock_commit(buffer, event);
1028
1029         ftrace_trace_stack(buffer, flags, 6, pc);
1030         ftrace_trace_userstack(buffer, flags, pc);
1031
1032         if (wake)
1033                 trace_wake_up();
1034 }
1035
1036 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1037                                 struct ring_buffer_event *event,
1038                                 unsigned long flags, int pc)
1039 {
1040         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1041 }
1042
1043 struct ring_buffer_event *
1044 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1045                                   int type, unsigned long len,
1046                                   unsigned long flags, int pc)
1047 {
1048         *current_rb = global_trace.buffer;
1049         return trace_buffer_lock_reserve(*current_rb,
1050                                          type, len, flags, pc);
1051 }
1052 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1053
1054 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1055                                         struct ring_buffer_event *event,
1056                                         unsigned long flags, int pc)
1057 {
1058         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1059 }
1060 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1061
1062 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1063                                        struct ring_buffer_event *event,
1064                                        unsigned long flags, int pc)
1065 {
1066         __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1067 }
1068 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1069
1070 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1071                                          struct ring_buffer_event *event)
1072 {
1073         ring_buffer_discard_commit(buffer, event);
1074 }
1075 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1076
1077 void
1078 trace_function(struct trace_array *tr,
1079                unsigned long ip, unsigned long parent_ip, unsigned long flags,
1080                int pc)
1081 {
1082         struct ftrace_event_call *call = &event_function;
1083         struct ring_buffer *buffer = tr->buffer;
1084         struct ring_buffer_event *event;
1085         struct ftrace_entry *entry;
1086
1087         /* If we are reading the ring buffer, don't trace */
1088         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
1089                 return;
1090
1091         event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1092                                           flags, pc);
1093         if (!event)
1094                 return;
1095         entry   = ring_buffer_event_data(event);
1096         entry->ip                       = ip;
1097         entry->parent_ip                = parent_ip;
1098
1099         if (!filter_check_discard(call, entry, buffer, event))
1100                 ring_buffer_unlock_commit(buffer, event);
1101 }
1102
1103 void
1104 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1105        unsigned long ip, unsigned long parent_ip, unsigned long flags,
1106        int pc)
1107 {
1108         if (likely(!atomic_read(&data->disabled)))
1109                 trace_function(tr, ip, parent_ip, flags, pc);
1110 }
1111
1112 #ifdef CONFIG_STACKTRACE
1113 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1114                                  unsigned long flags,
1115                                  int skip, int pc)
1116 {
1117         struct ftrace_event_call *call = &event_kernel_stack;
1118         struct ring_buffer_event *event;
1119         struct stack_entry *entry;
1120         struct stack_trace trace;
1121
1122         event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1123                                           sizeof(*entry), flags, pc);
1124         if (!event)
1125                 return;
1126         entry   = ring_buffer_event_data(event);
1127         memset(&entry->caller, 0, sizeof(entry->caller));
1128
1129         trace.nr_entries        = 0;
1130         trace.max_entries       = FTRACE_STACK_ENTRIES;
1131         trace.skip              = skip;
1132         trace.entries           = entry->caller;
1133
1134         save_stack_trace(&trace);
1135         if (!filter_check_discard(call, entry, buffer, event))
1136                 ring_buffer_unlock_commit(buffer, event);
1137 }
1138
1139 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1140                         int skip, int pc)
1141 {
1142         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1143                 return;
1144
1145         __ftrace_trace_stack(buffer, flags, skip, pc);
1146 }
1147
1148 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1149                    int pc)
1150 {
1151         __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1152 }
1153
1154 void
1155 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1156 {
1157         struct ftrace_event_call *call = &event_user_stack;
1158         struct ring_buffer_event *event;
1159         struct userstack_entry *entry;
1160         struct stack_trace trace;
1161
1162         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1163                 return;
1164
1165         event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1166                                           sizeof(*entry), flags, pc);
1167         if (!event)
1168                 return;
1169         entry   = ring_buffer_event_data(event);
1170
1171         entry->tgid             = current->tgid;
1172         memset(&entry->caller, 0, sizeof(entry->caller));
1173
1174         trace.nr_entries        = 0;
1175         trace.max_entries       = FTRACE_STACK_ENTRIES;
1176         trace.skip              = 0;
1177         trace.entries           = entry->caller;
1178
1179         save_stack_trace_user(&trace);
1180         if (!filter_check_discard(call, entry, buffer, event))
1181                 ring_buffer_unlock_commit(buffer, event);
1182 }
1183
1184 #ifdef UNUSED
1185 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1186 {
1187         ftrace_trace_userstack(tr, flags, preempt_count());
1188 }
1189 #endif /* UNUSED */
1190
1191 #endif /* CONFIG_STACKTRACE */
1192
1193 static void
1194 ftrace_trace_special(void *__tr,
1195                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1196                      int pc)
1197 {
1198         struct ftrace_event_call *call = &event_special;
1199         struct ring_buffer_event *event;
1200         struct trace_array *tr = __tr;
1201         struct ring_buffer *buffer = tr->buffer;
1202         struct special_entry *entry;
1203
1204         event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1205                                           sizeof(*entry), 0, pc);
1206         if (!event)
1207                 return;
1208         entry   = ring_buffer_event_data(event);
1209         entry->arg1                     = arg1;
1210         entry->arg2                     = arg2;
1211         entry->arg3                     = arg3;
1212
1213         if (!filter_check_discard(call, entry, buffer, event))
1214                 trace_buffer_unlock_commit(buffer, event, 0, pc);
1215 }
1216
1217 void
1218 __trace_special(void *__tr, void *__data,
1219                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1220 {
1221         ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1222 }
1223
1224 void
1225 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1226 {
1227         struct trace_array *tr = &global_trace;
1228         struct trace_array_cpu *data;
1229         unsigned long flags;
1230         int cpu;
1231         int pc;
1232
1233         if (tracing_disabled)
1234                 return;
1235
1236         pc = preempt_count();
1237         local_irq_save(flags);
1238         cpu = raw_smp_processor_id();
1239         data = tr->data[cpu];
1240
1241         if (likely(atomic_inc_return(&data->disabled) == 1))
1242                 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1243
1244         atomic_dec(&data->disabled);
1245         local_irq_restore(flags);
1246 }
1247
1248 /**
1249  * trace_vbprintk - write binary msg to tracing buffer
1250  *
1251  */
1252 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1253 {
1254         static raw_spinlock_t trace_buf_lock =
1255                 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1256         static u32 trace_buf[TRACE_BUF_SIZE];
1257
1258         struct ftrace_event_call *call = &event_bprint;
1259         struct ring_buffer_event *event;
1260         struct ring_buffer *buffer;
1261         struct trace_array *tr = &global_trace;
1262         struct trace_array_cpu *data;
1263         struct bprint_entry *entry;
1264         unsigned long flags;
1265         int disable;
1266         int resched;
1267         int cpu, len = 0, size, pc;
1268
1269         if (unlikely(tracing_selftest_running || tracing_disabled))
1270                 return 0;
1271
1272         /* Don't pollute graph traces with trace_vprintk internals */
1273         pause_graph_tracing();
1274
1275         pc = preempt_count();
1276         resched = ftrace_preempt_disable();
1277         cpu = raw_smp_processor_id();
1278         data = tr->data[cpu];
1279
1280         disable = atomic_inc_return(&data->disabled);
1281         if (unlikely(disable != 1))
1282                 goto out;
1283
1284         /* Lockdep uses trace_printk for lock tracing */
1285         local_irq_save(flags);
1286         __raw_spin_lock(&trace_buf_lock);
1287         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1288
1289         if (len > TRACE_BUF_SIZE || len < 0)
1290                 goto out_unlock;
1291
1292         size = sizeof(*entry) + sizeof(u32) * len;
1293         buffer = tr->buffer;
1294         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1295                                           flags, pc);
1296         if (!event)
1297                 goto out_unlock;
1298         entry = ring_buffer_event_data(event);
1299         entry->ip                       = ip;
1300         entry->fmt                      = fmt;
1301
1302         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1303         if (!filter_check_discard(call, entry, buffer, event))
1304                 ring_buffer_unlock_commit(buffer, event);
1305
1306 out_unlock:
1307         __raw_spin_unlock(&trace_buf_lock);
1308         local_irq_restore(flags);
1309
1310 out:
1311         atomic_dec_return(&data->disabled);
1312         ftrace_preempt_enable(resched);
1313         unpause_graph_tracing();
1314
1315         return len;
1316 }
1317 EXPORT_SYMBOL_GPL(trace_vbprintk);
1318
1319 int trace_array_printk(struct trace_array *tr,
1320                        unsigned long ip, const char *fmt, ...)
1321 {
1322         int ret;
1323         va_list ap;
1324
1325         if (!(trace_flags & TRACE_ITER_PRINTK))
1326                 return 0;
1327
1328         va_start(ap, fmt);
1329         ret = trace_array_vprintk(tr, ip, fmt, ap);
1330         va_end(ap);
1331         return ret;
1332 }
1333
1334 int trace_array_vprintk(struct trace_array *tr,
1335                         unsigned long ip, const char *fmt, va_list args)
1336 {
1337         static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1338         static char trace_buf[TRACE_BUF_SIZE];
1339
1340         struct ftrace_event_call *call = &event_print;
1341         struct ring_buffer_event *event;
1342         struct ring_buffer *buffer;
1343         struct trace_array_cpu *data;
1344         int cpu, len = 0, size, pc;
1345         struct print_entry *entry;
1346         unsigned long irq_flags;
1347         int disable;
1348
1349         if (tracing_disabled || tracing_selftest_running)
1350                 return 0;
1351
1352         pc = preempt_count();
1353         preempt_disable_notrace();
1354         cpu = raw_smp_processor_id();
1355         data = tr->data[cpu];
1356
1357         disable = atomic_inc_return(&data->disabled);
1358         if (unlikely(disable != 1))
1359                 goto out;
1360
1361         pause_graph_tracing();
1362         raw_local_irq_save(irq_flags);
1363         __raw_spin_lock(&trace_buf_lock);
1364         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1365
1366         len = min(len, TRACE_BUF_SIZE-1);
1367         trace_buf[len] = 0;
1368
1369         size = sizeof(*entry) + len + 1;
1370         buffer = tr->buffer;
1371         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1372                                           irq_flags, pc);
1373         if (!event)
1374                 goto out_unlock;
1375         entry = ring_buffer_event_data(event);
1376         entry->ip                       = ip;
1377
1378         memcpy(&entry->buf, trace_buf, len);
1379         entry->buf[len] = 0;
1380         if (!filter_check_discard(call, entry, buffer, event))
1381                 ring_buffer_unlock_commit(buffer, event);
1382
1383  out_unlock:
1384         __raw_spin_unlock(&trace_buf_lock);
1385         raw_local_irq_restore(irq_flags);
1386         unpause_graph_tracing();
1387  out:
1388         atomic_dec_return(&data->disabled);
1389         preempt_enable_notrace();
1390
1391         return len;
1392 }
1393
1394 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1395 {
1396         return trace_array_printk(&global_trace, ip, fmt, args);
1397 }
1398 EXPORT_SYMBOL_GPL(trace_vprintk);
1399
1400 enum trace_file_type {
1401         TRACE_FILE_LAT_FMT      = 1,
1402         TRACE_FILE_ANNOTATE     = 2,
1403 };
1404
1405 static void trace_iterator_increment(struct trace_iterator *iter)
1406 {
1407         /* Don't allow ftrace to trace into the ring buffers */
1408         ftrace_disable_cpu();
1409
1410         iter->idx++;
1411         if (iter->buffer_iter[iter->cpu])
1412                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1413
1414         ftrace_enable_cpu();
1415 }
1416
1417 static struct trace_entry *
1418 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1419 {
1420         struct ring_buffer_event *event;
1421         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1422
1423         /* Don't allow ftrace to trace into the ring buffers */
1424         ftrace_disable_cpu();
1425
1426         if (buf_iter)
1427                 event = ring_buffer_iter_peek(buf_iter, ts);
1428         else
1429                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1430
1431         ftrace_enable_cpu();
1432
1433         return event ? ring_buffer_event_data(event) : NULL;
1434 }
1435
1436 static struct trace_entry *
1437 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1438 {
1439         struct ring_buffer *buffer = iter->tr->buffer;
1440         struct trace_entry *ent, *next = NULL;
1441         int cpu_file = iter->cpu_file;
1442         u64 next_ts = 0, ts;
1443         int next_cpu = -1;
1444         int cpu;
1445
1446         /*
1447          * If we are in a per_cpu trace file, don't bother by iterating over
1448          * all cpu and peek directly.
1449          */
1450         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1451                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1452                         return NULL;
1453                 ent = peek_next_entry(iter, cpu_file, ent_ts);
1454                 if (ent_cpu)
1455                         *ent_cpu = cpu_file;
1456
1457                 return ent;
1458         }
1459
1460         for_each_tracing_cpu(cpu) {
1461
1462                 if (ring_buffer_empty_cpu(buffer, cpu))
1463                         continue;
1464
1465                 ent = peek_next_entry(iter, cpu, &ts);
1466
1467                 /*
1468                  * Pick the entry with the smallest timestamp:
1469                  */
1470                 if (ent && (!next || ts < next_ts)) {
1471                         next = ent;
1472                         next_cpu = cpu;
1473                         next_ts = ts;
1474                 }
1475         }
1476
1477         if (ent_cpu)
1478                 *ent_cpu = next_cpu;
1479
1480         if (ent_ts)
1481                 *ent_ts = next_ts;
1482
1483         return next;
1484 }
1485
1486 /* Find the next real entry, without updating the iterator itself */
1487 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1488                                           int *ent_cpu, u64 *ent_ts)
1489 {
1490         return __find_next_entry(iter, ent_cpu, ent_ts);
1491 }
1492
1493 /* Find the next real entry, and increment the iterator to the next entry */
1494 static void *find_next_entry_inc(struct trace_iterator *iter)
1495 {
1496         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1497
1498         if (iter->ent)
1499                 trace_iterator_increment(iter);
1500
1501         return iter->ent ? iter : NULL;
1502 }
1503
1504 static void trace_consume(struct trace_iterator *iter)
1505 {
1506         /* Don't allow ftrace to trace into the ring buffers */
1507         ftrace_disable_cpu();
1508         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1509         ftrace_enable_cpu();
1510 }
1511
1512 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1513 {
1514         struct trace_iterator *iter = m->private;
1515         int i = (int)*pos;
1516         void *ent;
1517
1518         (*pos)++;
1519
1520         /* can't go backwards */
1521         if (iter->idx > i)
1522                 return NULL;
1523
1524         if (iter->idx < 0)
1525                 ent = find_next_entry_inc(iter);
1526         else
1527                 ent = iter;
1528
1529         while (ent && iter->idx < i)
1530                 ent = find_next_entry_inc(iter);
1531
1532         iter->pos = *pos;
1533
1534         return ent;
1535 }
1536
1537 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1538 {
1539         struct trace_array *tr = iter->tr;
1540         struct ring_buffer_event *event;
1541         struct ring_buffer_iter *buf_iter;
1542         unsigned long entries = 0;
1543         u64 ts;
1544
1545         tr->data[cpu]->skipped_entries = 0;
1546
1547         if (!iter->buffer_iter[cpu])
1548                 return;
1549
1550         buf_iter = iter->buffer_iter[cpu];
1551         ring_buffer_iter_reset(buf_iter);
1552
1553         /*
1554          * We could have the case with the max latency tracers
1555          * that a reset never took place on a cpu. This is evident
1556          * by the timestamp being before the start of the buffer.
1557          */
1558         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1559                 if (ts >= iter->tr->time_start)
1560                         break;
1561                 entries++;
1562                 ring_buffer_read(buf_iter, NULL);
1563         }
1564
1565         tr->data[cpu]->skipped_entries = entries;
1566 }
1567
1568 /*
1569  * No necessary locking here. The worst thing which can
1570  * happen is loosing events consumed at the same time
1571  * by a trace_pipe reader.
1572  * Other than that, we don't risk to crash the ring buffer
1573  * because it serializes the readers.
1574  *
1575  * The current tracer is copied to avoid a global locking
1576  * all around.
1577  */
1578 static void *s_start(struct seq_file *m, loff_t *pos)
1579 {
1580         struct trace_iterator *iter = m->private;
1581         static struct tracer *old_tracer;
1582         int cpu_file = iter->cpu_file;
1583         void *p = NULL;
1584         loff_t l = 0;
1585         int cpu;
1586
1587         /* copy the tracer to avoid using a global lock all around */
1588         mutex_lock(&trace_types_lock);
1589         if (unlikely(old_tracer != current_trace && current_trace)) {
1590                 old_tracer = current_trace;
1591                 *iter->trace = *current_trace;
1592         }
1593         mutex_unlock(&trace_types_lock);
1594
1595         atomic_inc(&trace_record_cmdline_disabled);
1596
1597         if (*pos != iter->pos) {
1598                 iter->ent = NULL;
1599                 iter->cpu = 0;
1600                 iter->idx = -1;
1601
1602                 ftrace_disable_cpu();
1603
1604                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1605                         for_each_tracing_cpu(cpu)
1606                                 tracing_iter_reset(iter, cpu);
1607                 } else
1608                         tracing_iter_reset(iter, cpu_file);
1609
1610                 ftrace_enable_cpu();
1611
1612                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1613                         ;
1614
1615         } else {
1616                 l = *pos - 1;
1617                 p = s_next(m, p, &l);
1618         }
1619
1620         trace_event_read_lock();
1621         return p;
1622 }
1623
1624 static void s_stop(struct seq_file *m, void *p)
1625 {
1626         atomic_dec(&trace_record_cmdline_disabled);
1627         trace_event_read_unlock();
1628 }
1629
1630 static void print_lat_help_header(struct seq_file *m)
1631 {
1632         seq_puts(m, "#                  _------=> CPU#            \n");
1633         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1634         seq_puts(m, "#                | / _----=> need-resched    \n");
1635         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1636         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1637         seq_puts(m, "#                |||| /_--=> lock-depth       \n");
1638         seq_puts(m, "#                |||||/     delay             \n");
1639         seq_puts(m, "#  cmd     pid   |||||| time  |   caller      \n");
1640         seq_puts(m, "#     \\   /      ||||||   \\   |   /           \n");
1641 }
1642
1643 static void print_func_help_header(struct seq_file *m)
1644 {
1645         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1646         seq_puts(m, "#              | |       |          |         |\n");
1647 }
1648
1649
1650 static void
1651 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1652 {
1653         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1654         struct trace_array *tr = iter->tr;
1655         struct trace_array_cpu *data = tr->data[tr->cpu];
1656         struct tracer *type = current_trace;
1657         unsigned long entries = 0;
1658         unsigned long total = 0;
1659         unsigned long count;
1660         const char *name = "preemption";
1661         int cpu;
1662
1663         if (type)
1664                 name = type->name;
1665
1666
1667         for_each_tracing_cpu(cpu) {
1668                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1669                 /*
1670                  * If this buffer has skipped entries, then we hold all
1671                  * entries for the trace and we need to ignore the
1672                  * ones before the time stamp.
1673                  */
1674                 if (tr->data[cpu]->skipped_entries) {
1675                         count -= tr->data[cpu]->skipped_entries;
1676                         /* total is the same as the entries */
1677                         total += count;
1678                 } else
1679                         total += count +
1680                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1681                 entries += count;
1682         }
1683
1684         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1685                    name, UTS_RELEASE);
1686         seq_puts(m, "# -----------------------------------"
1687                  "---------------------------------\n");
1688         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1689                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1690                    nsecs_to_usecs(data->saved_latency),
1691                    entries,
1692                    total,
1693                    tr->cpu,
1694 #if defined(CONFIG_PREEMPT_NONE)
1695                    "server",
1696 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1697                    "desktop",
1698 #elif defined(CONFIG_PREEMPT)
1699                    "preempt",
1700 #else
1701                    "unknown",
1702 #endif
1703                    /* These are reserved for later use */
1704                    0, 0, 0, 0);
1705 #ifdef CONFIG_SMP
1706         seq_printf(m, " #P:%d)\n", num_online_cpus());
1707 #else
1708         seq_puts(m, ")\n");
1709 #endif
1710         seq_puts(m, "#    -----------------\n");
1711         seq_printf(m, "#    | task: %.16s-%d "
1712                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1713                    data->comm, data->pid, data->uid, data->nice,
1714                    data->policy, data->rt_priority);
1715         seq_puts(m, "#    -----------------\n");
1716
1717         if (data->critical_start) {
1718                 seq_puts(m, "#  => started at: ");
1719                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1720                 trace_print_seq(m, &iter->seq);
1721                 seq_puts(m, "\n#  => ended at:   ");
1722                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1723                 trace_print_seq(m, &iter->seq);
1724                 seq_puts(m, "\n#\n");
1725         }
1726
1727         seq_puts(m, "#\n");
1728 }
1729
1730 static void test_cpu_buff_start(struct trace_iterator *iter)
1731 {
1732         struct trace_seq *s = &iter->seq;
1733
1734         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1735                 return;
1736
1737         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1738                 return;
1739
1740         if (cpumask_test_cpu(iter->cpu, iter->started))
1741                 return;
1742
1743         if (iter->tr->data[iter->cpu]->skipped_entries)
1744                 return;
1745
1746         cpumask_set_cpu(iter->cpu, iter->started);
1747
1748         /* Don't print started cpu buffer for the first entry of the trace */
1749         if (iter->idx > 1)
1750                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1751                                 iter->cpu);
1752 }
1753
1754 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1755 {
1756         struct trace_seq *s = &iter->seq;
1757         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1758         struct trace_entry *entry;
1759         struct trace_event *event;
1760
1761         entry = iter->ent;
1762
1763         test_cpu_buff_start(iter);
1764
1765         event = ftrace_find_event(entry->type);
1766
1767         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1768                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1769                         if (!trace_print_lat_context(iter))
1770                                 goto partial;
1771                 } else {
1772                         if (!trace_print_context(iter))
1773                                 goto partial;
1774                 }
1775         }
1776
1777         if (event)
1778                 return event->trace(iter, sym_flags);
1779
1780         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1781                 goto partial;
1782
1783         return TRACE_TYPE_HANDLED;
1784 partial:
1785         return TRACE_TYPE_PARTIAL_LINE;
1786 }
1787
1788 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1789 {
1790         struct trace_seq *s = &iter->seq;
1791         struct trace_entry *entry;
1792         struct trace_event *event;
1793
1794         entry = iter->ent;
1795
1796         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1797                 if (!trace_seq_printf(s, "%d %d %llu ",
1798                                       entry->pid, iter->cpu, iter->ts))
1799                         goto partial;
1800         }
1801
1802         event = ftrace_find_event(entry->type);
1803         if (event)
1804                 return event->raw(iter, 0);
1805
1806         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1807                 goto partial;
1808
1809         return TRACE_TYPE_HANDLED;
1810 partial:
1811         return TRACE_TYPE_PARTIAL_LINE;
1812 }
1813
1814 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1815 {
1816         struct trace_seq *s = &iter->seq;
1817         unsigned char newline = '\n';
1818         struct trace_entry *entry;
1819         struct trace_event *event;
1820
1821         entry = iter->ent;
1822
1823         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1824                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1825                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1826                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1827         }
1828
1829         event = ftrace_find_event(entry->type);
1830         if (event) {
1831                 enum print_line_t ret = event->hex(iter, 0);
1832                 if (ret != TRACE_TYPE_HANDLED)
1833                         return ret;
1834         }
1835
1836         SEQ_PUT_FIELD_RET(s, newline);
1837
1838         return TRACE_TYPE_HANDLED;
1839 }
1840
1841 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1842 {
1843         struct trace_seq *s = &iter->seq;
1844         struct trace_entry *entry;
1845         struct trace_event *event;
1846
1847         entry = iter->ent;
1848
1849         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1850                 SEQ_PUT_FIELD_RET(s, entry->pid);
1851                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1852                 SEQ_PUT_FIELD_RET(s, iter->ts);
1853         }
1854
1855         event = ftrace_find_event(entry->type);
1856         return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1857 }
1858
1859 static int trace_empty(struct trace_iterator *iter)
1860 {
1861         int cpu;
1862
1863         /* If we are looking at one CPU buffer, only check that one */
1864         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1865                 cpu = iter->cpu_file;
1866                 if (iter->buffer_iter[cpu]) {
1867                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1868                                 return 0;
1869                 } else {
1870                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1871                                 return 0;
1872                 }
1873                 return 1;
1874         }
1875
1876         for_each_tracing_cpu(cpu) {
1877                 if (iter->buffer_iter[cpu]) {
1878                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1879                                 return 0;
1880                 } else {
1881                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1882                                 return 0;
1883                 }
1884         }
1885
1886         return 1;
1887 }
1888
1889 /*  Called with trace_event_read_lock() held. */
1890 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1891 {
1892         enum print_line_t ret;
1893
1894         if (iter->trace && iter->trace->print_line) {
1895                 ret = iter->trace->print_line(iter);
1896                 if (ret != TRACE_TYPE_UNHANDLED)
1897                         return ret;
1898         }
1899
1900         if (iter->ent->type == TRACE_BPRINT &&
1901                         trace_flags & TRACE_ITER_PRINTK &&
1902                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1903                 return trace_print_bprintk_msg_only(iter);
1904
1905         if (iter->ent->type == TRACE_PRINT &&
1906                         trace_flags & TRACE_ITER_PRINTK &&
1907                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1908                 return trace_print_printk_msg_only(iter);
1909
1910         if (trace_flags & TRACE_ITER_BIN)
1911                 return print_bin_fmt(iter);
1912
1913         if (trace_flags & TRACE_ITER_HEX)
1914                 return print_hex_fmt(iter);
1915
1916         if (trace_flags & TRACE_ITER_RAW)
1917                 return print_raw_fmt(iter);
1918
1919         return print_trace_fmt(iter);
1920 }
1921
1922 static int s_show(struct seq_file *m, void *v)
1923 {
1924         struct trace_iterator *iter = v;
1925
1926         if (iter->ent == NULL) {
1927                 if (iter->tr) {
1928                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1929                         seq_puts(m, "#\n");
1930                 }
1931                 if (iter->trace && iter->trace->print_header)
1932                         iter->trace->print_header(m);
1933                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1934                         /* print nothing if the buffers are empty */
1935                         if (trace_empty(iter))
1936                                 return 0;
1937                         print_trace_header(m, iter);
1938                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1939                                 print_lat_help_header(m);
1940                 } else {
1941                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1942                                 print_func_help_header(m);
1943                 }
1944         } else {
1945                 print_trace_line(iter);
1946                 trace_print_seq(m, &iter->seq);
1947         }
1948
1949         return 0;
1950 }
1951
1952 static const struct seq_operations tracer_seq_ops = {
1953         .start          = s_start,
1954         .next           = s_next,
1955         .stop           = s_stop,
1956         .show           = s_show,
1957 };
1958
1959 static struct trace_iterator *
1960 __tracing_open(struct inode *inode, struct file *file)
1961 {
1962         long cpu_file = (long) inode->i_private;
1963         void *fail_ret = ERR_PTR(-ENOMEM);
1964         struct trace_iterator *iter;
1965         struct seq_file *m;
1966         int cpu, ret;
1967
1968         if (tracing_disabled)
1969                 return ERR_PTR(-ENODEV);
1970
1971         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1972         if (!iter)
1973                 return ERR_PTR(-ENOMEM);
1974
1975         /*
1976          * We make a copy of the current tracer to avoid concurrent
1977          * changes on it while we are reading.
1978          */
1979         mutex_lock(&trace_types_lock);
1980         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1981         if (!iter->trace)
1982                 goto fail;
1983
1984         if (current_trace)
1985                 *iter->trace = *current_trace;
1986
1987         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
1988                 goto fail;
1989
1990         if (current_trace && current_trace->print_max)
1991                 iter->tr = &max_tr;
1992         else
1993                 iter->tr = &global_trace;
1994         iter->pos = -1;
1995         mutex_init(&iter->mutex);
1996         iter->cpu_file = cpu_file;
1997
1998         /* Notify the tracer early; before we stop tracing. */
1999         if (iter->trace && iter->trace->open)
2000                 iter->trace->open(iter);
2001
2002         /* Annotate start of buffers if we had overruns */
2003         if (ring_buffer_overruns(iter->tr->buffer))
2004                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2005
2006         /* stop the trace while dumping */
2007         tracing_stop();
2008
2009         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2010                 for_each_tracing_cpu(cpu) {
2011
2012                         iter->buffer_iter[cpu] =
2013                                 ring_buffer_read_start(iter->tr->buffer, cpu);
2014                         tracing_iter_reset(iter, cpu);
2015                 }
2016         } else {
2017                 cpu = iter->cpu_file;
2018                 iter->buffer_iter[cpu] =
2019                                 ring_buffer_read_start(iter->tr->buffer, cpu);
2020                 tracing_iter_reset(iter, cpu);
2021         }
2022
2023         ret = seq_open(file, &tracer_seq_ops);
2024         if (ret < 0) {
2025                 fail_ret = ERR_PTR(ret);
2026                 goto fail_buffer;
2027         }
2028
2029         m = file->private_data;
2030         m->private = iter;
2031
2032         mutex_unlock(&trace_types_lock);
2033
2034         return iter;
2035
2036  fail_buffer:
2037         for_each_tracing_cpu(cpu) {
2038                 if (iter->buffer_iter[cpu])
2039                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2040         }
2041         free_cpumask_var(iter->started);
2042         tracing_start();
2043  fail:
2044         mutex_unlock(&trace_types_lock);
2045         kfree(iter->trace);
2046         kfree(iter);
2047
2048         return fail_ret;
2049 }
2050
2051 int tracing_open_generic(struct inode *inode, struct file *filp)
2052 {
2053         if (tracing_disabled)
2054                 return -ENODEV;
2055
2056         filp->private_data = inode->i_private;
2057         return 0;
2058 }
2059
2060 static int tracing_release(struct inode *inode, struct file *file)
2061 {
2062         struct seq_file *m = (struct seq_file *)file->private_data;
2063         struct trace_iterator *iter;
2064         int cpu;
2065
2066         if (!(file->f_mode & FMODE_READ))
2067                 return 0;
2068
2069         iter = m->private;
2070
2071         mutex_lock(&trace_types_lock);
2072         for_each_tracing_cpu(cpu) {
2073                 if (iter->buffer_iter[cpu])
2074                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2075         }
2076
2077         if (iter->trace && iter->trace->close)
2078                 iter->trace->close(iter);
2079
2080         /* reenable tracing if it was previously enabled */
2081         tracing_start();
2082         mutex_unlock(&trace_types_lock);
2083
2084         seq_release(inode, file);
2085         mutex_destroy(&iter->mutex);
2086         free_cpumask_var(iter->started);
2087         kfree(iter->trace);
2088         kfree(iter);
2089         return 0;
2090 }
2091
2092 static int tracing_open(struct inode *inode, struct file *file)
2093 {
2094         struct trace_iterator *iter;
2095         int ret = 0;
2096
2097         /* If this file was open for write, then erase contents */
2098         if ((file->f_mode & FMODE_WRITE) &&
2099             (file->f_flags & O_TRUNC)) {
2100                 long cpu = (long) inode->i_private;
2101
2102                 if (cpu == TRACE_PIPE_ALL_CPU)
2103                         tracing_reset_online_cpus(&global_trace);
2104                 else
2105                         tracing_reset(&global_trace, cpu);
2106         }
2107
2108         if (file->f_mode & FMODE_READ) {
2109                 iter = __tracing_open(inode, file);
2110                 if (IS_ERR(iter))
2111                         ret = PTR_ERR(iter);
2112                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2113                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2114         }
2115         return ret;
2116 }
2117
2118 static void *
2119 t_next(struct seq_file *m, void *v, loff_t *pos)
2120 {
2121         struct tracer *t = v;
2122
2123         (*pos)++;
2124
2125         if (t)
2126                 t = t->next;
2127
2128         return t;
2129 }
2130
2131 static void *t_start(struct seq_file *m, loff_t *pos)
2132 {
2133         struct tracer *t;
2134         loff_t l = 0;
2135
2136         mutex_lock(&trace_types_lock);
2137         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2138                 ;
2139
2140         return t;
2141 }
2142
2143 static void t_stop(struct seq_file *m, void *p)
2144 {
2145         mutex_unlock(&trace_types_lock);
2146 }
2147
2148 static int t_show(struct seq_file *m, void *v)
2149 {
2150         struct tracer *t = v;
2151
2152         if (!t)
2153                 return 0;
2154
2155         seq_printf(m, "%s", t->name);
2156         if (t->next)
2157                 seq_putc(m, ' ');
2158         else
2159                 seq_putc(m, '\n');
2160
2161         return 0;
2162 }
2163
2164 static const struct seq_operations show_traces_seq_ops = {
2165         .start          = t_start,
2166         .next           = t_next,
2167         .stop           = t_stop,
2168         .show           = t_show,
2169 };
2170
2171 static int show_traces_open(struct inode *inode, struct file *file)
2172 {
2173         if (tracing_disabled)
2174                 return -ENODEV;
2175
2176         return seq_open(file, &show_traces_seq_ops);
2177 }
2178
2179 static ssize_t
2180 tracing_write_stub(struct file *filp, const char __user *ubuf,
2181                    size_t count, loff_t *ppos)
2182 {
2183         return count;
2184 }
2185
2186 static const struct file_operations tracing_fops = {
2187         .open           = tracing_open,
2188         .read           = seq_read,
2189         .write          = tracing_write_stub,
2190         .llseek         = seq_lseek,
2191         .release        = tracing_release,
2192 };
2193
2194 static const struct file_operations show_traces_fops = {
2195         .open           = show_traces_open,
2196         .read           = seq_read,
2197         .release        = seq_release,
2198 };
2199
2200 /*
2201  * Only trace on a CPU if the bitmask is set:
2202  */
2203 static cpumask_var_t tracing_cpumask;
2204
2205 /*
2206  * The tracer itself will not take this lock, but still we want
2207  * to provide a consistent cpumask to user-space:
2208  */
2209 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2210
2211 /*
2212  * Temporary storage for the character representation of the
2213  * CPU bitmask (and one more byte for the newline):
2214  */
2215 static char mask_str[NR_CPUS + 1];
2216
2217 static ssize_t
2218 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2219                      size_t count, loff_t *ppos)
2220 {
2221         int len;
2222
2223         mutex_lock(&tracing_cpumask_update_lock);
2224
2225         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2226         if (count - len < 2) {
2227                 count = -EINVAL;
2228                 goto out_err;
2229         }
2230         len += sprintf(mask_str + len, "\n");
2231         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2232
2233 out_err:
2234         mutex_unlock(&tracing_cpumask_update_lock);
2235
2236         return count;
2237 }
2238
2239 static ssize_t
2240 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2241                       size_t count, loff_t *ppos)
2242 {
2243         int err, cpu;
2244         cpumask_var_t tracing_cpumask_new;
2245
2246         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2247                 return -ENOMEM;
2248
2249         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2250         if (err)
2251                 goto err_unlock;
2252
2253         mutex_lock(&tracing_cpumask_update_lock);
2254
2255         local_irq_disable();
2256         __raw_spin_lock(&ftrace_max_lock);
2257         for_each_tracing_cpu(cpu) {
2258                 /*
2259                  * Increase/decrease the disabled counter if we are
2260                  * about to flip a bit in the cpumask:
2261                  */
2262                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2263                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2264                         atomic_inc(&global_trace.data[cpu]->disabled);
2265                 }
2266                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2267                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2268                         atomic_dec(&global_trace.data[cpu]->disabled);
2269                 }
2270         }
2271         __raw_spin_unlock(&ftrace_max_lock);
2272         local_irq_enable();
2273
2274         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2275
2276         mutex_unlock(&tracing_cpumask_update_lock);
2277         free_cpumask_var(tracing_cpumask_new);
2278
2279         return count;
2280
2281 err_unlock:
2282         free_cpumask_var(tracing_cpumask_new);
2283
2284         return err;
2285 }
2286
2287 static const struct file_operations tracing_cpumask_fops = {
2288         .open           = tracing_open_generic,
2289         .read           = tracing_cpumask_read,
2290         .write          = tracing_cpumask_write,
2291 };
2292
2293 static ssize_t
2294 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2295                        size_t cnt, loff_t *ppos)
2296 {
2297         struct tracer_opt *trace_opts;
2298         u32 tracer_flags;
2299         int len = 0;
2300         char *buf;
2301         int r = 0;
2302         int i;
2303
2304
2305         /* calculate max size */
2306         for (i = 0; trace_options[i]; i++) {
2307                 len += strlen(trace_options[i]);
2308                 len += 3; /* "no" and newline */
2309         }
2310
2311         mutex_lock(&trace_types_lock);
2312         tracer_flags = current_trace->flags->val;
2313         trace_opts = current_trace->flags->opts;
2314
2315         /*
2316          * Increase the size with names of options specific
2317          * of the current tracer.
2318          */
2319         for (i = 0; trace_opts[i].name; i++) {
2320                 len += strlen(trace_opts[i].name);
2321                 len += 3; /* "no" and newline */
2322         }
2323
2324         /* +1 for \0 */
2325         buf = kmalloc(len + 1, GFP_KERNEL);
2326         if (!buf) {
2327                 mutex_unlock(&trace_types_lock);
2328                 return -ENOMEM;
2329         }
2330
2331         for (i = 0; trace_options[i]; i++) {
2332                 if (trace_flags & (1 << i))
2333                         r += sprintf(buf + r, "%s\n", trace_options[i]);
2334                 else
2335                         r += sprintf(buf + r, "no%s\n", trace_options[i]);
2336         }
2337
2338         for (i = 0; trace_opts[i].name; i++) {
2339                 if (tracer_flags & trace_opts[i].bit)
2340                         r += sprintf(buf + r, "%s\n",
2341                                 trace_opts[i].name);
2342                 else
2343                         r += sprintf(buf + r, "no%s\n",
2344                                 trace_opts[i].name);
2345         }
2346         mutex_unlock(&trace_types_lock);
2347
2348         WARN_ON(r >= len + 1);
2349
2350         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2351
2352         kfree(buf);
2353         return r;
2354 }
2355
2356 /* Try to assign a tracer specific option */
2357 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2358 {
2359         struct tracer_flags *tracer_flags = trace->flags;
2360         struct tracer_opt *opts = NULL;
2361         int ret = 0, i = 0;
2362         int len;
2363
2364         for (i = 0; tracer_flags->opts[i].name; i++) {
2365                 opts = &tracer_flags->opts[i];
2366                 len = strlen(opts->name);
2367
2368                 if (strncmp(cmp, opts->name, len) == 0) {
2369                         ret = trace->set_flag(tracer_flags->val,
2370                                 opts->bit, !neg);
2371                         break;
2372                 }
2373         }
2374         /* Not found */
2375         if (!tracer_flags->opts[i].name)
2376                 return -EINVAL;
2377
2378         /* Refused to handle */
2379         if (ret)
2380                 return ret;
2381
2382         if (neg)
2383                 tracer_flags->val &= ~opts->bit;
2384         else
2385                 tracer_flags->val |= opts->bit;
2386
2387         return 0;
2388 }
2389
2390 static void set_tracer_flags(unsigned int mask, int enabled)
2391 {
2392         /* do nothing if flag is already set */
2393         if (!!(trace_flags & mask) == !!enabled)
2394                 return;
2395
2396         if (enabled)
2397                 trace_flags |= mask;
2398         else
2399                 trace_flags &= ~mask;
2400 }
2401
2402 static ssize_t
2403 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2404                         size_t cnt, loff_t *ppos)
2405 {
2406         char buf[64];
2407         char *cmp = buf;
2408         int neg = 0;
2409         int ret;
2410         int i;
2411
2412         if (cnt >= sizeof(buf))
2413                 return -EINVAL;
2414
2415         if (copy_from_user(&buf, ubuf, cnt))
2416                 return -EFAULT;
2417
2418         buf[cnt] = 0;
2419
2420         if (strncmp(buf, "no", 2) == 0) {
2421                 neg = 1;
2422                 cmp += 2;
2423         }
2424
2425         for (i = 0; trace_options[i]; i++) {
2426                 int len = strlen(trace_options[i]);
2427
2428                 if (strncmp(cmp, trace_options[i], len) == 0) {
2429                         set_tracer_flags(1 << i, !neg);
2430                         break;
2431                 }
2432         }
2433
2434         /* If no option could be set, test the specific tracer options */
2435         if (!trace_options[i]) {
2436                 mutex_lock(&trace_types_lock);
2437                 ret = set_tracer_option(current_trace, cmp, neg);
2438                 mutex_unlock(&trace_types_lock);
2439                 if (ret)
2440                         return ret;
2441         }
2442
2443         filp->f_pos += cnt;
2444
2445         return cnt;
2446 }
2447
2448 static const struct file_operations tracing_iter_fops = {
2449         .open           = tracing_open_generic,
2450         .read           = tracing_trace_options_read,
2451         .write          = tracing_trace_options_write,
2452 };
2453
2454 static const char readme_msg[] =
2455         "tracing mini-HOWTO:\n\n"
2456         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2457         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2458         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2459         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2460         "nop\n"
2461         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2462         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2463         "sched_switch\n"
2464         "# cat /sys/kernel/debug/tracing/trace_options\n"
2465         "noprint-parent nosym-offset nosym-addr noverbose\n"
2466         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2467         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2468         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2469         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2470 ;
2471
2472 static ssize_t
2473 tracing_readme_read(struct file *filp, char __user *ubuf,
2474                        size_t cnt, loff_t *ppos)
2475 {
2476         return simple_read_from_buffer(ubuf, cnt, ppos,
2477                                         readme_msg, strlen(readme_msg));
2478 }
2479
2480 static const struct file_operations tracing_readme_fops = {
2481         .open           = tracing_open_generic,
2482         .read           = tracing_readme_read,
2483 };
2484
2485 static ssize_t
2486 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2487                                 size_t cnt, loff_t *ppos)
2488 {
2489         char *buf_comm;
2490         char *file_buf;
2491         char *buf;
2492         int len = 0;
2493         int pid;
2494         int i;
2495
2496         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2497         if (!file_buf)
2498                 return -ENOMEM;
2499
2500         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2501         if (!buf_comm) {
2502                 kfree(file_buf);
2503                 return -ENOMEM;
2504         }
2505
2506         buf = file_buf;
2507
2508         for (i = 0; i < SAVED_CMDLINES; i++) {
2509                 int r;
2510
2511                 pid = map_cmdline_to_pid[i];
2512                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2513                         continue;
2514
2515                 trace_find_cmdline(pid, buf_comm);
2516                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2517                 buf += r;
2518                 len += r;
2519         }
2520
2521         len = simple_read_from_buffer(ubuf, cnt, ppos,
2522                                       file_buf, len);
2523
2524         kfree(file_buf);
2525         kfree(buf_comm);
2526
2527         return len;
2528 }
2529
2530 static const struct file_operations tracing_saved_cmdlines_fops = {
2531     .open       = tracing_open_generic,
2532     .read       = tracing_saved_cmdlines_read,
2533 };
2534
2535 static ssize_t
2536 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2537                   size_t cnt, loff_t *ppos)
2538 {
2539         char buf[64];
2540         int r;
2541
2542         r = sprintf(buf, "%u\n", tracer_enabled);
2543         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2544 }
2545
2546 static ssize_t
2547 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2548                    size_t cnt, loff_t *ppos)
2549 {
2550         struct trace_array *tr = filp->private_data;
2551         char buf[64];
2552         unsigned long val;
2553         int ret;
2554
2555         if (cnt >= sizeof(buf))
2556                 return -EINVAL;
2557
2558         if (copy_from_user(&buf, ubuf, cnt))
2559                 return -EFAULT;
2560
2561         buf[cnt] = 0;
2562
2563         ret = strict_strtoul(buf, 10, &val);
2564         if (ret < 0)
2565                 return ret;
2566
2567         val = !!val;
2568
2569         mutex_lock(&trace_types_lock);
2570         if (tracer_enabled ^ val) {
2571                 if (val) {
2572                         tracer_enabled = 1;
2573                         if (current_trace->start)
2574                                 current_trace->start(tr);
2575                         tracing_start();
2576                 } else {
2577                         tracer_enabled = 0;
2578                         tracing_stop();
2579                         if (current_trace->stop)
2580                                 current_trace->stop(tr);
2581                 }
2582         }
2583         mutex_unlock(&trace_types_lock);
2584
2585         filp->f_pos += cnt;
2586
2587         return cnt;
2588 }
2589
2590 static ssize_t
2591 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2592                        size_t cnt, loff_t *ppos)
2593 {
2594         char buf[MAX_TRACER_SIZE+2];
2595         int r;
2596
2597         mutex_lock(&trace_types_lock);
2598         if (current_trace)
2599                 r = sprintf(buf, "%s\n", current_trace->name);
2600         else
2601                 r = sprintf(buf, "\n");
2602         mutex_unlock(&trace_types_lock);
2603
2604         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2605 }
2606
2607 int tracer_init(struct tracer *t, struct trace_array *tr)
2608 {
2609         tracing_reset_online_cpus(tr);
2610         return t->init(tr);
2611 }
2612
2613 static int tracing_resize_ring_buffer(unsigned long size)
2614 {
2615         int ret;
2616
2617         /*
2618          * If kernel or user changes the size of the ring buffer
2619          * we use the size that was given, and we can forget about
2620          * expanding it later.
2621          */
2622         ring_buffer_expanded = 1;
2623
2624         ret = ring_buffer_resize(global_trace.buffer, size);
2625         if (ret < 0)
2626                 return ret;
2627
2628         ret = ring_buffer_resize(max_tr.buffer, size);
2629         if (ret < 0) {
2630                 int r;
2631
2632                 r = ring_buffer_resize(global_trace.buffer,
2633                                        global_trace.entries);
2634                 if (r < 0) {
2635                         /*
2636                          * AARGH! We are left with different
2637                          * size max buffer!!!!
2638                          * The max buffer is our "snapshot" buffer.
2639                          * When a tracer needs a snapshot (one of the
2640                          * latency tracers), it swaps the max buffer
2641                          * with the saved snap shot. We succeeded to
2642                          * update the size of the main buffer, but failed to
2643                          * update the size of the max buffer. But when we tried
2644                          * to reset the main buffer to the original size, we
2645                          * failed there too. This is very unlikely to
2646                          * happen, but if it does, warn and kill all
2647                          * tracing.
2648                          */
2649                         WARN_ON(1);
2650                         tracing_disabled = 1;
2651                 }
2652                 return ret;
2653         }
2654
2655         global_trace.entries = size;
2656
2657         return ret;
2658 }
2659
2660 /**
2661  * tracing_update_buffers - used by tracing facility to expand ring buffers
2662  *
2663  * To save on memory when the tracing is never used on a system with it
2664  * configured in. The ring buffers are set to a minimum size. But once
2665  * a user starts to use the tracing facility, then they need to grow
2666  * to their default size.
2667  *
2668  * This function is to be called when a tracer is about to be used.
2669  */
2670 int tracing_update_buffers(void)
2671 {
2672         int ret = 0;
2673
2674         mutex_lock(&trace_types_lock);
2675         if (!ring_buffer_expanded)
2676                 ret = tracing_resize_ring_buffer(trace_buf_size);
2677         mutex_unlock(&trace_types_lock);
2678
2679         return ret;
2680 }
2681
2682 struct trace_option_dentry;
2683
2684 static struct trace_option_dentry *
2685 create_trace_option_files(struct tracer *tracer);
2686
2687 static void
2688 destroy_trace_option_files(struct trace_option_dentry *topts);
2689
2690 static int tracing_set_tracer(const char *buf)
2691 {
2692         static struct trace_option_dentry *topts;
2693         struct trace_array *tr = &global_trace;
2694         struct tracer *t;
2695         int ret = 0;
2696
2697         mutex_lock(&trace_types_lock);
2698
2699         if (!ring_buffer_expanded) {
2700                 ret = tracing_resize_ring_buffer(trace_buf_size);
2701                 if (ret < 0)
2702                         goto out;
2703                 ret = 0;
2704         }
2705
2706         for (t = trace_types; t; t = t->next) {
2707                 if (strcmp(t->name, buf) == 0)
2708                         break;
2709         }
2710         if (!t) {
2711                 ret = -EINVAL;
2712                 goto out;
2713         }
2714         if (t == current_trace)
2715                 goto out;
2716
2717         trace_branch_disable();
2718         if (current_trace && current_trace->reset)
2719                 current_trace->reset(tr);
2720
2721         destroy_trace_option_files(topts);
2722
2723         current_trace = t;
2724
2725         topts = create_trace_option_files(current_trace);
2726
2727         if (t->init) {
2728                 ret = tracer_init(t, tr);
2729                 if (ret)
2730                         goto out;
2731         }
2732
2733         trace_branch_enable(tr);
2734  out:
2735         mutex_unlock(&trace_types_lock);
2736
2737         return ret;
2738 }
2739
2740 static ssize_t
2741 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2742                         size_t cnt, loff_t *ppos)
2743 {
2744         char buf[MAX_TRACER_SIZE+1];
2745         int i;
2746         size_t ret;
2747         int err;
2748
2749         ret = cnt;
2750
2751         if (cnt > MAX_TRACER_SIZE)
2752                 cnt = MAX_TRACER_SIZE;
2753
2754         if (copy_from_user(&buf, ubuf, cnt))
2755                 return -EFAULT;
2756
2757         buf[cnt] = 0;
2758
2759         /* strip ending whitespace. */
2760         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2761                 buf[i] = 0;
2762
2763         err = tracing_set_tracer(buf);
2764         if (err)
2765                 return err;
2766
2767         filp->f_pos += ret;
2768
2769         return ret;
2770 }
2771
2772 static ssize_t
2773 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2774                      size_t cnt, loff_t *ppos)
2775 {
2776         unsigned long *ptr = filp->private_data;
2777         char buf[64];
2778         int r;
2779
2780         r = snprintf(buf, sizeof(buf), "%ld\n",
2781                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2782         if (r > sizeof(buf))
2783                 r = sizeof(buf);
2784         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2785 }
2786
2787 static ssize_t
2788 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2789                       size_t cnt, loff_t *ppos)
2790 {
2791         unsigned long *ptr = filp->private_data;
2792         char buf[64];
2793         unsigned long val;
2794         int ret;
2795
2796         if (cnt >= sizeof(buf))
2797                 return -EINVAL;
2798
2799         if (copy_from_user(&buf, ubuf, cnt))
2800                 return -EFAULT;
2801
2802         buf[cnt] = 0;
2803
2804         ret = strict_strtoul(buf, 10, &val);
2805         if (ret < 0)
2806                 return ret;
2807
2808         *ptr = val * 1000;
2809
2810         return cnt;
2811 }
2812
2813 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2814 {
2815         long cpu_file = (long) inode->i_private;
2816         struct trace_iterator *iter;
2817         int ret = 0;
2818
2819         if (tracing_disabled)
2820                 return -ENODEV;
2821
2822         mutex_lock(&trace_types_lock);
2823
2824         /* We only allow one reader per cpu */
2825         if (cpu_file == TRACE_PIPE_ALL_CPU) {
2826                 if (!cpumask_empty(tracing_reader_cpumask)) {
2827                         ret = -EBUSY;
2828                         goto out;
2829                 }
2830                 cpumask_setall(tracing_reader_cpumask);
2831         } else {
2832                 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2833                         cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2834                 else {
2835                         ret = -EBUSY;
2836                         goto out;
2837                 }
2838         }
2839
2840         /* create a buffer to store the information to pass to userspace */
2841         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2842         if (!iter) {
2843                 ret = -ENOMEM;
2844                 goto out;
2845         }
2846
2847         /*
2848          * We make a copy of the current tracer to avoid concurrent
2849          * changes on it while we are reading.
2850          */
2851         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2852         if (!iter->trace) {
2853                 ret = -ENOMEM;
2854                 goto fail;
2855         }
2856         if (current_trace)
2857                 *iter->trace = *current_trace;
2858
2859         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2860                 ret = -ENOMEM;
2861                 goto fail;
2862         }
2863
2864         /* trace pipe does not show start of buffer */
2865         cpumask_setall(iter->started);
2866
2867         if (trace_flags & TRACE_ITER_LATENCY_FMT)
2868                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2869
2870         iter->cpu_file = cpu_file;
2871         iter->tr = &global_trace;
2872         mutex_init(&iter->mutex);
2873         filp->private_data = iter;
2874
2875         if (iter->trace->pipe_open)
2876                 iter->trace->pipe_open(iter);
2877
2878 out:
2879         mutex_unlock(&trace_types_lock);
2880         return ret;
2881
2882 fail:
2883         kfree(iter->trace);
2884         kfree(iter);
2885         mutex_unlock(&trace_types_lock);
2886         return ret;
2887 }
2888
2889 static int tracing_release_pipe(struct inode *inode, struct file *file)
2890 {
2891         struct trace_iterator *iter = file->private_data;
2892
2893         mutex_lock(&trace_types_lock);
2894
2895         if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2896                 cpumask_clear(tracing_reader_cpumask);
2897         else
2898                 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2899
2900         mutex_unlock(&trace_types_lock);
2901
2902         free_cpumask_var(iter->started);
2903         mutex_destroy(&iter->mutex);
2904         kfree(iter->trace);
2905         kfree(iter);
2906
2907         return 0;
2908 }
2909
2910 static unsigned int
2911 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2912 {
2913         struct trace_iterator *iter = filp->private_data;
2914
2915         if (trace_flags & TRACE_ITER_BLOCK) {
2916                 /*
2917                  * Always select as readable when in blocking mode
2918                  */
2919                 return POLLIN | POLLRDNORM;
2920         } else {
2921                 if (!trace_empty(iter))
2922                         return POLLIN | POLLRDNORM;
2923                 poll_wait(filp, &trace_wait, poll_table);
2924                 if (!trace_empty(iter))
2925                         return POLLIN | POLLRDNORM;
2926
2927                 return 0;
2928         }
2929 }
2930
2931
2932 void default_wait_pipe(struct trace_iterator *iter)
2933 {
2934         DEFINE_WAIT(wait);
2935
2936         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2937
2938         if (trace_empty(iter))
2939                 schedule();
2940
2941         finish_wait(&trace_wait, &wait);
2942 }
2943
2944 /*
2945  * This is a make-shift waitqueue.
2946  * A tracer might use this callback on some rare cases:
2947  *
2948  *  1) the current tracer might hold the runqueue lock when it wakes up
2949  *     a reader, hence a deadlock (sched, function, and function graph tracers)
2950  *  2) the function tracers, trace all functions, we don't want
2951  *     the overhead of calling wake_up and friends
2952  *     (and tracing them too)
2953  *
2954  *     Anyway, this is really very primitive wakeup.
2955  */
2956 void poll_wait_pipe(struct trace_iterator *iter)
2957 {
2958         set_current_state(TASK_INTERRUPTIBLE);
2959         /* sleep for 100 msecs, and try again. */
2960         schedule_timeout(HZ / 10);
2961 }
2962
2963 /* Must be called with trace_types_lock mutex held. */
2964 static int tracing_wait_pipe(struct file *filp)
2965 {
2966         struct trace_iterator *iter = filp->private_data;
2967
2968         while (trace_empty(iter)) {
2969
2970                 if ((filp->f_flags & O_NONBLOCK)) {
2971                         return -EAGAIN;
2972                 }
2973
2974                 mutex_unlock(&iter->mutex);
2975
2976                 iter->trace->wait_pipe(iter);
2977
2978                 mutex_lock(&iter->mutex);
2979
2980                 if (signal_pending(current))
2981                         return -EINTR;
2982
2983                 /*
2984                  * We block until we read something and tracing is disabled.
2985                  * We still block if tracing is disabled, but we have never
2986                  * read anything. This allows a user to cat this file, and
2987                  * then enable tracing. But after we have read something,
2988                  * we give an EOF when tracing is again disabled.
2989                  *
2990                  * iter->pos will be 0 if we haven't read anything.
2991                  */
2992                 if (!tracer_enabled && iter->pos)
2993                         break;
2994         }
2995
2996         return 1;
2997 }
2998
2999 /*
3000  * Consumer reader.
3001  */
3002 static ssize_t
3003 tracing_read_pipe(struct file *filp, char __user *ubuf,
3004                   size_t cnt, loff_t *ppos)
3005 {
3006         struct trace_iterator *iter = filp->private_data;
3007         static struct tracer *old_tracer;
3008         ssize_t sret;
3009
3010         /* return any leftover data */
3011         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3012         if (sret != -EBUSY)
3013                 return sret;
3014
3015         trace_seq_init(&iter->seq);
3016
3017         /* copy the tracer to avoid using a global lock all around */
3018         mutex_lock(&trace_types_lock);
3019         if (unlikely(old_tracer != current_trace && current_trace)) {
3020                 old_tracer = current_trace;
3021                 *iter->trace = *current_trace;
3022         }
3023         mutex_unlock(&trace_types_lock);
3024
3025         /*
3026          * Avoid more than one consumer on a single file descriptor
3027          * This is just a matter of traces coherency, the ring buffer itself
3028          * is protected.
3029          */
3030         mutex_lock(&iter->mutex);
3031         if (iter->trace->read) {
3032                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3033                 if (sret)
3034                         goto out;
3035         }
3036
3037 waitagain:
3038         sret = tracing_wait_pipe(filp);
3039         if (sret <= 0)
3040                 goto out;
3041
3042         /* stop when tracing is finished */
3043         if (trace_empty(iter)) {
3044                 sret = 0;
3045                 goto out;
3046         }
3047
3048         if (cnt >= PAGE_SIZE)
3049                 cnt = PAGE_SIZE - 1;
3050
3051         /* reset all but tr, trace, and overruns */
3052         memset(&iter->seq, 0,
3053                sizeof(struct trace_iterator) -
3054                offsetof(struct trace_iterator, seq));
3055         iter->pos = -1;
3056
3057         trace_event_read_lock();
3058         while (find_next_entry_inc(iter) != NULL) {
3059                 enum print_line_t ret;
3060                 int len = iter->seq.len;
3061
3062                 ret = print_trace_line(iter);
3063                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3064                         /* don't print partial lines */
3065                         iter->seq.len = len;
3066                         break;
3067                 }
3068                 if (ret != TRACE_TYPE_NO_CONSUME)
3069                         trace_consume(iter);
3070
3071                 if (iter->seq.len >= cnt)
3072                         break;
3073         }
3074         trace_event_read_unlock();
3075
3076         /* Now copy what we have to the user */
3077         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3078         if (iter->seq.readpos >= iter->seq.len)
3079                 trace_seq_init(&iter->seq);
3080
3081         /*
3082          * If there was nothing to send to user, inspite of consuming trace
3083          * entries, go back to wait for more entries.
3084          */
3085         if (sret == -EBUSY)
3086                 goto waitagain;
3087
3088 out:
3089         mutex_unlock(&iter->mutex);
3090
3091         return sret;
3092 }
3093
3094 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3095                                      struct pipe_buffer *buf)
3096 {
3097         __free_page(buf->page);
3098 }
3099
3100 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3101                                      unsigned int idx)
3102 {
3103         __free_page(spd->pages[idx]);
3104 }
3105
3106 static struct pipe_buf_operations tracing_pipe_buf_ops = {
3107         .can_merge              = 0,
3108         .map                    = generic_pipe_buf_map,
3109         .unmap                  = generic_pipe_buf_unmap,
3110         .confirm                = generic_pipe_buf_confirm,
3111         .release                = tracing_pipe_buf_release,
3112         .steal                  = generic_pipe_buf_steal,
3113         .get                    = generic_pipe_buf_get,
3114 };
3115
3116 static size_t
3117 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3118 {
3119         size_t count;
3120         int ret;
3121
3122         /* Seq buffer is page-sized, exactly what we need. */
3123         for (;;) {
3124                 count = iter->seq.len;
3125                 ret = print_trace_line(iter);
3126                 count = iter->seq.len - count;
3127                 if (rem < count) {
3128                         rem = 0;
3129                         iter->seq.len -= count;
3130                         break;
3131                 }
3132                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3133                         iter->seq.len -= count;
3134                         break;
3135                 }
3136
3137                 if (ret != TRACE_TYPE_NO_CONSUME)
3138                         trace_consume(iter);
3139                 rem -= count;
3140                 if (!find_next_entry_inc(iter)) {
3141                         rem = 0;
3142                         iter->ent = NULL;
3143                         break;
3144                 }
3145         }
3146
3147         return rem;
3148 }
3149
3150 static ssize_t tracing_splice_read_pipe(struct file *filp,
3151                                         loff_t *ppos,
3152                                         struct pipe_inode_info *pipe,
3153                                         size_t len,
3154                                         unsigned int flags)
3155 {
3156         struct page *pages[PIPE_BUFFERS];
3157         struct partial_page partial[PIPE_BUFFERS];
3158         struct trace_iterator *iter = filp->private_data;
3159         struct splice_pipe_desc spd = {
3160                 .pages          = pages,
3161                 .partial        = partial,
3162                 .nr_pages       = 0, /* This gets updated below. */
3163                 .flags          = flags,
3164                 .ops            = &tracing_pipe_buf_ops,
3165                 .spd_release    = tracing_spd_release_pipe,
3166         };
3167         static struct tracer *old_tracer;
3168         ssize_t ret;
3169         size_t rem;
3170         unsigned int i;
3171
3172         /* copy the tracer to avoid using a global lock all around */
3173         mutex_lock(&trace_types_lock);
3174         if (unlikely(old_tracer != current_trace && current_trace)) {
3175                 old_tracer = current_trace;
3176                 *iter->trace = *current_trace;
3177         }
3178         mutex_unlock(&trace_types_lock);
3179
3180         mutex_lock(&iter->mutex);
3181
3182         if (iter->trace->splice_read) {
3183                 ret = iter->trace->splice_read(iter, filp,
3184                                                ppos, pipe, len, flags);
3185                 if (ret)
3186                         goto out_err;
3187         }
3188
3189         ret = tracing_wait_pipe(filp);
3190         if (ret <= 0)
3191                 goto out_err;
3192
3193         if (!iter->ent && !find_next_entry_inc(iter)) {
3194                 ret = -EFAULT;
3195                 goto out_err;
3196         }
3197
3198         trace_event_read_lock();
3199
3200         /* Fill as many pages as possible. */
3201         for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3202                 pages[i] = alloc_page(GFP_KERNEL);
3203                 if (!pages[i])
3204                         break;
3205
3206                 rem = tracing_fill_pipe_page(rem, iter);
3207
3208                 /* Copy the data into the page, so we can start over. */
3209                 ret = trace_seq_to_buffer(&iter->seq,
3210                                           page_address(pages[i]),
3211                                           iter->seq.len);
3212                 if (ret < 0) {
3213                         __free_page(pages[i]);
3214                         break;
3215                 }
3216                 partial[i].offset = 0;
3217                 partial[i].len = iter->seq.len;
3218
3219                 trace_seq_init(&iter->seq);
3220         }
3221
3222         trace_event_read_unlock();
3223         mutex_unlock(&iter->mutex);
3224
3225         spd.nr_pages = i;
3226
3227         return splice_to_pipe(pipe, &spd);
3228
3229 out_err:
3230         mutex_unlock(&iter->mutex);
3231
3232         return ret;
3233 }
3234
3235 static ssize_t
3236 tracing_entries_read(struct file *filp, char __user *ubuf,
3237                      size_t cnt, loff_t *ppos)
3238 {
3239         struct trace_array *tr = filp->private_data;
3240         char buf[96];
3241         int r;
3242
3243         mutex_lock(&trace_types_lock);
3244         if (!ring_buffer_expanded)
3245                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3246                             tr->entries >> 10,
3247                             trace_buf_size >> 10);
3248         else
3249                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3250         mutex_unlock(&trace_types_lock);
3251
3252         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3253 }
3254
3255 static ssize_t
3256 tracing_entries_write(struct file *filp, const char __user *ubuf,
3257                       size_t cnt, loff_t *ppos)
3258 {
3259         unsigned long val;
3260         char buf[64];
3261         int ret, cpu;
3262
3263         if (cnt >= sizeof(buf))
3264                 return -EINVAL;
3265
3266         if (copy_from_user(&buf, ubuf, cnt))
3267                 return -EFAULT;
3268
3269         buf[cnt] = 0;
3270
3271         ret = strict_strtoul(buf, 10, &val);
3272         if (ret < 0)
3273                 return ret;
3274
3275         /* must have at least 1 entry */
3276         if (!val)
3277                 return -EINVAL;
3278
3279         mutex_lock(&trace_types_lock);
3280
3281         tracing_stop();
3282
3283         /* disable all cpu buffers */
3284         for_each_tracing_cpu(cpu) {
3285                 if (global_trace.data[cpu])
3286                         atomic_inc(&global_trace.data[cpu]->disabled);
3287                 if (max_tr.data[cpu])
3288                         atomic_inc(&max_tr.data[cpu]->disabled);
3289         }
3290
3291         /* value is in KB */
3292         val <<= 10;
3293
3294         if (val != global_trace.entries) {
3295                 ret = tracing_resize_ring_buffer(val);
3296                 if (ret < 0) {
3297                         cnt = ret;
3298                         goto out;
3299                 }
3300         }
3301
3302         filp->f_pos += cnt;
3303
3304         /* If check pages failed, return ENOMEM */
3305         if (tracing_disabled)
3306                 cnt = -ENOMEM;
3307  out:
3308         for_each_tracing_cpu(cpu) {
3309                 if (global_trace.data[cpu])
3310                         atomic_dec(&global_trace.data[cpu]->disabled);
3311                 if (max_tr.data[cpu])
3312                         atomic_dec(&max_tr.data[cpu]->disabled);
3313         }
3314
3315         tracing_start();
3316         max_tr.entries = global_trace.entries;
3317         mutex_unlock(&trace_types_lock);
3318
3319         return cnt;
3320 }
3321
3322 static int mark_printk(const char *fmt, ...)
3323 {
3324         int ret;
3325         va_list args;
3326         va_start(args, fmt);
3327         ret = trace_vprintk(0, fmt, args);
3328         va_end(args);
3329         return ret;
3330 }
3331
3332 static ssize_t
3333 tracing_mark_write(struct file *filp, const char __user *ubuf,
3334                                         size_t cnt, loff_t *fpos)
3335 {
3336         char *buf;
3337         char *end;
3338
3339         if (tracing_disabled)
3340                 return -EINVAL;
3341
3342         if (cnt > TRACE_BUF_SIZE)
3343                 cnt = TRACE_BUF_SIZE;
3344
3345         buf = kmalloc(cnt + 1, GFP_KERNEL);
3346         if (buf == NULL)
3347                 return -ENOMEM;
3348
3349         if (copy_from_user(buf, ubuf, cnt)) {
3350                 kfree(buf);
3351                 return -EFAULT;
3352         }
3353
3354         /* Cut from the first nil or newline. */
3355         buf[cnt] = '\0';
3356         end = strchr(buf, '\n');
3357         if (end)
3358                 *end = '\0';
3359
3360         cnt = mark_printk("%s\n", buf);
3361         kfree(buf);
3362         *fpos += cnt;
3363
3364         return cnt;
3365 }
3366
3367 static ssize_t tracing_clock_read(struct file *filp, char __user *ubuf,
3368                                   size_t cnt, loff_t *ppos)
3369 {
3370         char buf[64];
3371         int bufiter = 0;
3372         int i;
3373
3374         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)