7d56cecc2c6ea8bba5bc87d5d1e86d4e2c393c6a
[linux-2.6.git] / kernel / trace / trace.c
1 /*
2  * ring buffer based function tracer
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally taken from the RT patch by:
8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code from the latency_tracer, that is:
11  *  Copyright (C) 2004-2006 Ingo Molnar
12  *  Copyright (C) 2004 William Lee Irwin III
13  */
14 #include <linux/ring_buffer.h>
15 #include <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(int, ftrace_cpu_disabled);
90
91 static inline void ftrace_disable_cpu(void)
92 {
93         preempt_disable();
94         __this_cpu_inc(per_cpu_var(ftrace_cpu_disabled));
95 }
96
97 static inline void ftrace_enable_cpu(void)
98 {
99         __this_cpu_dec(per_cpu_var(ftrace_cpu_disabled));
100         preempt_enable();
101 }
102
103 static cpumask_var_t __read_mostly      tracing_buffer_mask;
104
105 /* Define which cpu buffers are currently read in trace_pipe */
106 static cpumask_var_t                    tracing_reader_cpumask;
107
108 #define for_each_tracing_cpu(cpu)       \
109         for_each_cpu(cpu, tracing_buffer_mask)
110
111 /*
112  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
113  *
114  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
115  * is set, then ftrace_dump is called. This will output the contents
116  * of the ftrace buffers to the console.  This is very useful for
117  * capturing traces that lead to crashes and outputing it to a
118  * serial console.
119  *
120  * It is default off, but you can enable it with either specifying
121  * "ftrace_dump_on_oops" in the kernel command line, or setting
122  * /proc/sys/kernel/ftrace_dump_on_oops to true.
123  */
124 int ftrace_dump_on_oops;
125
126 static int tracing_set_tracer(const char *buf);
127
128 #define MAX_TRACER_SIZE         100
129 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
130 static char *default_bootup_tracer;
131
132 static int __init set_cmdline_ftrace(char *str)
133 {
134         strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
135         default_bootup_tracer = bootup_tracer_buf;
136         /* We are using ftrace early, expand it */
137         ring_buffer_expanded = 1;
138         return 1;
139 }
140 __setup("ftrace=", set_cmdline_ftrace);
141
142 static int __init set_ftrace_dump_on_oops(char *str)
143 {
144         ftrace_dump_on_oops = 1;
145         return 1;
146 }
147 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
148
149 unsigned long long ns2usecs(cycle_t nsec)
150 {
151         nsec += 500;
152         do_div(nsec, 1000);
153         return nsec;
154 }
155
156 /*
157  * The global_trace is the descriptor that holds the tracing
158  * buffers for the live tracing. For each CPU, it contains
159  * a link list of pages that will store trace entries. The
160  * page descriptor of the pages in the memory is used to hold
161  * the link list by linking the lru item in the page descriptor
162  * to each of the pages in the buffer per CPU.
163  *
164  * For each active CPU there is a data field that holds the
165  * pages for the buffer for that CPU. Each CPU has the same number
166  * of pages allocated for its buffer.
167  */
168 static struct trace_array       global_trace;
169
170 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
171
172 int filter_current_check_discard(struct ring_buffer *buffer,
173                                  struct ftrace_event_call *call, void *rec,
174                                  struct ring_buffer_event *event)
175 {
176         return filter_check_discard(call, rec, buffer, event);
177 }
178 EXPORT_SYMBOL_GPL(filter_current_check_discard);
179
180 cycle_t ftrace_now(int cpu)
181 {
182         u64 ts;
183
184         /* Early boot up does not have a buffer yet */
185         if (!global_trace.buffer)
186                 return trace_clock_local();
187
188         ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
189         ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
190
191         return ts;
192 }
193
194 /*
195  * The max_tr is used to snapshot the global_trace when a maximum
196  * latency is reached. Some tracers will use this to store a maximum
197  * trace while it continues examining live traces.
198  *
199  * The buffers for the max_tr are set up the same as the global_trace.
200  * When a snapshot is taken, the link list of the max_tr is swapped
201  * with the link list of the global_trace and the buffers are reset for
202  * the global_trace so the tracing can continue.
203  */
204 static struct trace_array       max_tr;
205
206 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
207
208 /* tracer_enabled is used to toggle activation of a tracer */
209 static int                      tracer_enabled = 1;
210
211 /**
212  * tracing_is_enabled - return tracer_enabled status
213  *
214  * This function is used by other tracers to know the status
215  * of the tracer_enabled flag.  Tracers may use this function
216  * to know if it should enable their features when starting
217  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
218  */
219 int tracing_is_enabled(void)
220 {
221         return tracer_enabled;
222 }
223
224 /*
225  * trace_buf_size is the size in bytes that is allocated
226  * for a buffer. Note, the number of bytes is always rounded
227  * to page size.
228  *
229  * This number is purposely set to a low number of 16384.
230  * If the dump on oops happens, it will be much appreciated
231  * to not have to wait for all that output. Anyway this can be
232  * boot time and run time configurable.
233  */
234 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
235
236 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
237
238 /* trace_types holds a link list of available tracers. */
239 static struct tracer            *trace_types __read_mostly;
240
241 /* current_trace points to the tracer that is currently active */
242 static struct tracer            *current_trace __read_mostly;
243
244 /*
245  * trace_types_lock is used to protect the trace_types list.
246  * This lock is also used to keep user access serialized.
247  * Accesses from userspace will grab this lock while userspace
248  * activities happen inside the kernel.
249  */
250 static DEFINE_MUTEX(trace_types_lock);
251
252 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
253 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
254
255 /* trace_flags holds trace_options default values */
256 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
257         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
258         TRACE_ITER_GRAPH_TIME;
259
260 static int trace_stop_count;
261 static DEFINE_SPINLOCK(tracing_start_lock);
262
263 /**
264  * trace_wake_up - wake up tasks waiting for trace input
265  *
266  * Simply wakes up any task that is blocked on the trace_wait
267  * queue. These is used with trace_poll for tasks polling the trace.
268  */
269 void trace_wake_up(void)
270 {
271         int cpu;
272
273         if (trace_flags & TRACE_ITER_BLOCK)
274                 return;
275         /*
276          * The runqueue_is_locked() can fail, but this is the best we
277          * have for now:
278          */
279         cpu = get_cpu();
280         if (!runqueue_is_locked(cpu))
281                 wake_up(&trace_wait);
282         put_cpu();
283 }
284
285 static int __init set_buf_size(char *str)
286 {
287         unsigned long buf_size;
288
289         if (!str)
290                 return 0;
291         buf_size = memparse(str, &str);
292         /* nr_entries can not be zero */
293         if (buf_size == 0)
294                 return 0;
295         trace_buf_size = buf_size;
296         return 1;
297 }
298 __setup("trace_buf_size=", set_buf_size);
299
300 unsigned long nsecs_to_usecs(unsigned long nsecs)
301 {
302         return nsecs / 1000;
303 }
304
305 /* These must match the bit postions in trace_iterator_flags */
306 static const char *trace_options[] = {
307         "print-parent",
308         "sym-offset",
309         "sym-addr",
310         "verbose",
311         "raw",
312         "hex",
313         "bin",
314         "block",
315         "stacktrace",
316         "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 arch_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 arch_spinlock_t ftrace_max_lock =
504         (arch_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 arch_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(__this_cpu_read(per_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 arch_spinlock_t trace_buf_lock =
1255                 (arch_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 arch_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         size = sizeof(*entry) + len + 1;
1367         buffer = tr->buffer;
1368         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1369                                           irq_flags, pc);
1370         if (!event)
1371                 goto out_unlock;
1372         entry = ring_buffer_event_data(event);
1373         entry->ip = ip;
1374
1375         memcpy(&entry->buf, trace_buf, len);
1376         entry->buf[len] = '\0';
1377         if (!filter_check_discard(call, entry, buffer, event))
1378                 ring_buffer_unlock_commit(buffer, event);
1379
1380  out_unlock:
1381         __raw_spin_unlock(&trace_buf_lock);
1382         raw_local_irq_restore(irq_flags);
1383         unpause_graph_tracing();
1384  out:
1385         atomic_dec_return(&data->disabled);
1386         preempt_enable_notrace();
1387
1388         return len;
1389 }
1390
1391 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1392 {
1393         return trace_array_vprintk(&global_trace, ip, fmt, args);
1394 }
1395 EXPORT_SYMBOL_GPL(trace_vprintk);
1396
1397 enum trace_file_type {
1398         TRACE_FILE_LAT_FMT      = 1,
1399         TRACE_FILE_ANNOTATE     = 2,
1400 };
1401
1402 static void trace_iterator_increment(struct trace_iterator *iter)
1403 {
1404         /* Don't allow ftrace to trace into the ring buffers */
1405         ftrace_disable_cpu();
1406
1407         iter->idx++;
1408         if (iter->buffer_iter[iter->cpu])
1409                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1410
1411         ftrace_enable_cpu();
1412 }
1413
1414 static struct trace_entry *
1415 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1416 {
1417         struct ring_buffer_event *event;
1418         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1419
1420         /* Don't allow ftrace to trace into the ring buffers */
1421         ftrace_disable_cpu();
1422
1423         if (buf_iter)
1424                 event = ring_buffer_iter_peek(buf_iter, ts);
1425         else
1426                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1427
1428         ftrace_enable_cpu();
1429
1430         return event ? ring_buffer_event_data(event) : NULL;
1431 }
1432
1433 static struct trace_entry *
1434 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1435 {
1436         struct ring_buffer *buffer = iter->tr->buffer;
1437         struct trace_entry *ent, *next = NULL;
1438         int cpu_file = iter->cpu_file;
1439         u64 next_ts = 0, ts;
1440         int next_cpu = -1;
1441         int cpu;
1442
1443         /*
1444          * If we are in a per_cpu trace file, don't bother by iterating over
1445          * all cpu and peek directly.
1446          */
1447         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1448                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1449                         return NULL;
1450                 ent = peek_next_entry(iter, cpu_file, ent_ts);
1451                 if (ent_cpu)
1452                         *ent_cpu = cpu_file;
1453
1454                 return ent;
1455         }
1456
1457         for_each_tracing_cpu(cpu) {
1458
1459                 if (ring_buffer_empty_cpu(buffer, cpu))
1460                         continue;
1461
1462                 ent = peek_next_entry(iter, cpu, &ts);
1463
1464                 /*
1465                  * Pick the entry with the smallest timestamp:
1466                  */
1467                 if (ent && (!next || ts < next_ts)) {
1468                         next = ent;
1469                         next_cpu = cpu;
1470                         next_ts = ts;
1471                 }
1472         }
1473
1474         if (ent_cpu)
1475                 *ent_cpu = next_cpu;
1476
1477         if (ent_ts)
1478                 *ent_ts = next_ts;
1479
1480         return next;
1481 }
1482
1483 /* Find the next real entry, without updating the iterator itself */
1484 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1485                                           int *ent_cpu, u64 *ent_ts)
1486 {
1487         return __find_next_entry(iter, ent_cpu, ent_ts);
1488 }
1489
1490 /* Find the next real entry, and increment the iterator to the next entry */
1491 static void *find_next_entry_inc(struct trace_iterator *iter)
1492 {
1493         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1494
1495         if (iter->ent)
1496                 trace_iterator_increment(iter);
1497
1498         return iter->ent ? iter : NULL;
1499 }
1500
1501 static void trace_consume(struct trace_iterator *iter)
1502 {
1503         /* Don't allow ftrace to trace into the ring buffers */
1504         ftrace_disable_cpu();
1505         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1506         ftrace_enable_cpu();
1507 }
1508
1509 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1510 {
1511         struct trace_iterator *iter = m->private;
1512         int i = (int)*pos;
1513         void *ent;
1514
1515         WARN_ON_ONCE(iter->leftover);
1516
1517         (*pos)++;
1518
1519         /* can't go backwards */
1520         if (iter->idx > i)
1521                 return NULL;
1522
1523         if (iter->idx < 0)
1524                 ent = find_next_entry_inc(iter);
1525         else
1526                 ent = iter;
1527
1528         while (ent && iter->idx < i)
1529                 ent = find_next_entry_inc(iter);
1530
1531         iter->pos = *pos;
1532
1533         return ent;
1534 }
1535
1536 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1537 {
1538         struct trace_array *tr = iter->tr;
1539         struct ring_buffer_event *event;
1540         struct ring_buffer_iter *buf_iter;
1541         unsigned long entries = 0;
1542         u64 ts;
1543
1544         tr->data[cpu]->skipped_entries = 0;
1545
1546         if (!iter->buffer_iter[cpu])
1547                 return;
1548
1549         buf_iter = iter->buffer_iter[cpu];
1550         ring_buffer_iter_reset(buf_iter);
1551
1552         /*
1553          * We could have the case with the max latency tracers
1554          * that a reset never took place on a cpu. This is evident
1555          * by the timestamp being before the start of the buffer.
1556          */
1557         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1558                 if (ts >= iter->tr->time_start)
1559                         break;
1560                 entries++;
1561                 ring_buffer_read(buf_iter, NULL);
1562         }
1563
1564         tr->data[cpu]->skipped_entries = entries;
1565 }
1566
1567 /*
1568  * No necessary locking here. The worst thing which can
1569  * happen is loosing events consumed at the same time
1570  * by a trace_pipe reader.
1571  * Other than that, we don't risk to crash the ring buffer
1572  * because it serializes the readers.
1573  *
1574  * The current tracer is copied to avoid a global locking
1575  * all around.
1576  */
1577 static void *s_start(struct seq_file *m, loff_t *pos)
1578 {
1579         struct trace_iterator *iter = m->private;
1580         static struct tracer *old_tracer;
1581         int cpu_file = iter->cpu_file;
1582         void *p = NULL;
1583         loff_t l = 0;
1584         int cpu;
1585
1586         /* copy the tracer to avoid using a global lock all around */
1587         mutex_lock(&trace_types_lock);
1588         if (unlikely(old_tracer != current_trace && current_trace)) {
1589                 old_tracer = current_trace;
1590                 *iter->trace = *current_trace;
1591         }
1592         mutex_unlock(&trace_types_lock);
1593
1594         atomic_inc(&trace_record_cmdline_disabled);
1595
1596         if (*pos != iter->pos) {
1597                 iter->ent = NULL;
1598                 iter->cpu = 0;
1599                 iter->idx = -1;
1600
1601                 ftrace_disable_cpu();
1602
1603                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1604                         for_each_tracing_cpu(cpu)
1605                                 tracing_iter_reset(iter, cpu);
1606                 } else
1607                         tracing_iter_reset(iter, cpu_file);
1608
1609                 ftrace_enable_cpu();
1610
1611                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1612                         ;
1613
1614         } else {
1615                 /*
1616                  * If we overflowed the seq_file before, then we want
1617                  * to just reuse the trace_seq buffer again.
1618                  */
1619                 if (iter->leftover)
1620                         p = iter;
1621                 else {
1622                         l = *pos - 1;
1623                         p = s_next(m, p, &l);
1624                 }
1625         }
1626
1627         trace_event_read_lock();
1628         return p;
1629 }
1630
1631 static void s_stop(struct seq_file *m, void *p)
1632 {
1633         atomic_dec(&trace_record_cmdline_disabled);
1634         trace_event_read_unlock();
1635 }
1636
1637 static void print_lat_help_header(struct seq_file *m)
1638 {
1639         seq_puts(m, "#                  _------=> CPU#            \n");
1640         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1641         seq_puts(m, "#                | / _----=> need-resched    \n");
1642         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1643         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1644         seq_puts(m, "#                |||| /_--=> lock-depth       \n");
1645         seq_puts(m, "#                |||||/     delay             \n");
1646         seq_puts(m, "#  cmd     pid   |||||| time  |   caller      \n");
1647         seq_puts(m, "#     \\   /      ||||||   \\   |   /           \n");
1648 }
1649
1650 static void print_func_help_header(struct seq_file *m)
1651 {
1652         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1653         seq_puts(m, "#              | |       |          |         |\n");
1654 }
1655
1656
1657 static void
1658 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1659 {
1660         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1661         struct trace_array *tr = iter->tr;
1662         struct trace_array_cpu *data = tr->data[tr->cpu];
1663         struct tracer *type = current_trace;
1664         unsigned long entries = 0;
1665         unsigned long total = 0;
1666         unsigned long count;
1667         const char *name = "preemption";
1668         int cpu;
1669
1670         if (type)
1671                 name = type->name;
1672
1673
1674         for_each_tracing_cpu(cpu) {
1675                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1676                 /*
1677                  * If this buffer has skipped entries, then we hold all
1678                  * entries for the trace and we need to ignore the
1679                  * ones before the time stamp.
1680                  */
1681                 if (tr->data[cpu]->skipped_entries) {
1682                         count -= tr->data[cpu]->skipped_entries;
1683                         /* total is the same as the entries */
1684                         total += count;
1685                 } else
1686                         total += count +
1687                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1688                 entries += count;
1689         }
1690
1691         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1692                    name, UTS_RELEASE);
1693         seq_puts(m, "# -----------------------------------"
1694                  "---------------------------------\n");
1695         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1696                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1697                    nsecs_to_usecs(data->saved_latency),
1698                    entries,
1699                    total,
1700                    tr->cpu,
1701 #if defined(CONFIG_PREEMPT_NONE)
1702                    "server",
1703 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1704                    "desktop",
1705 #elif defined(CONFIG_PREEMPT)
1706                    "preempt",
1707 #else
1708                    "unknown",
1709 #endif
1710                    /* These are reserved for later use */
1711                    0, 0, 0, 0);
1712 #ifdef CONFIG_SMP
1713         seq_printf(m, " #P:%d)\n", num_online_cpus());
1714 #else
1715         seq_puts(m, ")\n");
1716 #endif
1717         seq_puts(m, "#    -----------------\n");
1718         seq_printf(m, "#    | task: %.16s-%d "
1719                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1720                    data->comm, data->pid, data->uid, data->nice,
1721                    data->policy, data->rt_priority);
1722         seq_puts(m, "#    -----------------\n");
1723
1724         if (data->critical_start) {
1725                 seq_puts(m, "#  => started at: ");
1726                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1727                 trace_print_seq(m, &iter->seq);
1728                 seq_puts(m, "\n#  => ended at:   ");
1729                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1730                 trace_print_seq(m, &iter->seq);
1731                 seq_puts(m, "\n#\n");
1732         }
1733
1734         seq_puts(m, "#\n");
1735 }
1736
1737 static void test_cpu_buff_start(struct trace_iterator *iter)
1738 {
1739         struct trace_seq *s = &iter->seq;
1740
1741         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1742                 return;
1743
1744         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1745                 return;
1746
1747         if (cpumask_test_cpu(iter->cpu, iter->started))
1748                 return;
1749
1750         if (iter->tr->data[iter->cpu]->skipped_entries)
1751                 return;
1752
1753         cpumask_set_cpu(iter->cpu, iter->started);
1754
1755         /* Don't print started cpu buffer for the first entry of the trace */
1756         if (iter->idx > 1)
1757                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1758                                 iter->cpu);
1759 }
1760
1761 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1762 {
1763         struct trace_seq *s = &iter->seq;
1764         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1765         struct trace_entry *entry;
1766         struct trace_event *event;
1767
1768         entry = iter->ent;
1769
1770         test_cpu_buff_start(iter);
1771
1772         event = ftrace_find_event(entry->type);
1773
1774         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1775                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1776                         if (!trace_print_lat_context(iter))
1777                                 goto partial;
1778                 } else {
1779                         if (!trace_print_context(iter))
1780                                 goto partial;
1781                 }
1782         }
1783
1784         if (event)
1785                 return event->trace(iter, sym_flags);
1786
1787         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1788                 goto partial;
1789
1790         return TRACE_TYPE_HANDLED;
1791 partial:
1792         return TRACE_TYPE_PARTIAL_LINE;
1793 }
1794
1795 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1796 {
1797         struct trace_seq *s = &iter->seq;
1798         struct trace_entry *entry;
1799         struct trace_event *event;
1800
1801         entry = iter->ent;
1802
1803         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1804                 if (!trace_seq_printf(s, "%d %d %llu ",
1805                                       entry->pid, iter->cpu, iter->ts))
1806                         goto partial;
1807         }
1808
1809         event = ftrace_find_event(entry->type);
1810         if (event)
1811                 return event->raw(iter, 0);
1812
1813         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1814                 goto partial;
1815
1816         return TRACE_TYPE_HANDLED;
1817 partial:
1818         return TRACE_TYPE_PARTIAL_LINE;
1819 }
1820
1821 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1822 {
1823         struct trace_seq *s = &iter->seq;
1824         unsigned char newline = '\n';
1825         struct trace_entry *entry;
1826         struct trace_event *event;
1827
1828         entry = iter->ent;
1829
1830         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1831                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1832                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1833                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1834         }
1835
1836         event = ftrace_find_event(entry->type);
1837         if (event) {
1838                 enum print_line_t ret = event->hex(iter, 0);
1839                 if (ret != TRACE_TYPE_HANDLED)
1840                         return ret;
1841         }
1842
1843         SEQ_PUT_FIELD_RET(s, newline);
1844
1845         return TRACE_TYPE_HANDLED;
1846 }
1847
1848 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1849 {
1850         struct trace_seq *s = &iter->seq;
1851         struct trace_entry *entry;
1852         struct trace_event *event;
1853
1854         entry = iter->ent;
1855
1856         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1857                 SEQ_PUT_FIELD_RET(s, entry->pid);
1858                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1859                 SEQ_PUT_FIELD_RET(s, iter->ts);
1860         }
1861
1862         event = ftrace_find_event(entry->type);
1863         return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1864 }
1865
1866 static int trace_empty(struct trace_iterator *iter)
1867 {
1868         int cpu;
1869
1870         /* If we are looking at one CPU buffer, only check that one */
1871         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1872                 cpu = iter->cpu_file;
1873                 if (iter->buffer_iter[cpu]) {
1874                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1875                                 return 0;
1876                 } else {
1877                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1878                                 return 0;
1879                 }
1880                 return 1;
1881         }
1882
1883         for_each_tracing_cpu(cpu) {
1884                 if (iter->buffer_iter[cpu]) {
1885                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1886                                 return 0;
1887                 } else {
1888                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1889                                 return 0;
1890                 }
1891         }
1892
1893         return 1;
1894 }
1895
1896 /*  Called with trace_event_read_lock() held. */
1897 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1898 {
1899         enum print_line_t ret;
1900
1901         if (iter->trace && iter->trace->print_line) {
1902                 ret = iter->trace->print_line(iter);
1903                 if (ret != TRACE_TYPE_UNHANDLED)
1904                         return ret;
1905         }
1906
1907         if (iter->ent->type == TRACE_BPRINT &&
1908                         trace_flags & TRACE_ITER_PRINTK &&
1909                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1910                 return trace_print_bprintk_msg_only(iter);
1911
1912         if (iter->ent->type == TRACE_PRINT &&
1913                         trace_flags & TRACE_ITER_PRINTK &&
1914                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1915                 return trace_print_printk_msg_only(iter);
1916
1917         if (trace_flags & TRACE_ITER_BIN)
1918                 return print_bin_fmt(iter);
1919
1920         if (trace_flags & TRACE_ITER_HEX)
1921                 return print_hex_fmt(iter);
1922
1923         if (trace_flags & TRACE_ITER_RAW)
1924                 return print_raw_fmt(iter);
1925
1926         return print_trace_fmt(iter);
1927 }
1928
1929 static int s_show(struct seq_file *m, void *v)
1930 {
1931         struct trace_iterator *iter = v;
1932         int ret;
1933
1934         if (iter->ent == NULL) {
1935                 if (iter->tr) {
1936                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1937                         seq_puts(m, "#\n");
1938                 }
1939                 if (iter->trace && iter->trace->print_header)
1940                         iter->trace->print_header(m);
1941                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1942                         /* print nothing if the buffers are empty */
1943                         if (trace_empty(iter))
1944                                 return 0;
1945                         print_trace_header(m, iter);
1946                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1947                                 print_lat_help_header(m);
1948                 } else {
1949                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1950                                 print_func_help_header(m);
1951                 }
1952         } else if (iter->leftover) {
1953                 /*
1954                  * If we filled the seq_file buffer earlier, we
1955                  * want to just show it now.
1956                  */
1957                 ret = trace_print_seq(m, &iter->seq);
1958
1959                 /* ret should this time be zero, but you never know */
1960                 iter->leftover = ret;
1961
1962         } else {
1963                 print_trace_line(iter);
1964                 ret = trace_print_seq(m, &iter->seq);
1965                 /*
1966                  * If we overflow the seq_file buffer, then it will
1967                  * ask us for this data again at start up.
1968                  * Use that instead.
1969                  *  ret is 0 if seq_file write succeeded.
1970                  *        -1 otherwise.
1971                  */
1972                 iter->leftover = ret;
1973         }
1974
1975         return 0;
1976 }
1977
1978 static const struct seq_operations tracer_seq_ops = {
1979         .start          = s_start,
1980         .next           = s_next,
1981         .stop           = s_stop,
1982         .show           = s_show,
1983 };
1984
1985 static struct trace_iterator *
1986 __tracing_open(struct inode *inode, struct file *file)
1987 {
1988         long cpu_file = (long) inode->i_private;
1989         void *fail_ret = ERR_PTR(-ENOMEM);
1990         struct trace_iterator *iter;
1991         struct seq_file *m;
1992         int cpu, ret;
1993
1994         if (tracing_disabled)
1995                 return ERR_PTR(-ENODEV);
1996
1997         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1998         if (!iter)
1999                 return ERR_PTR(-ENOMEM);
2000
2001         /*
2002          * We make a copy of the current tracer to avoid concurrent
2003          * changes on it while we are reading.
2004          */
2005         mutex_lock(&trace_types_lock);
2006         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2007         if (!iter->trace)
2008                 goto fail;
2009
2010         if (current_trace)
2011                 *iter->trace = *current_trace;
2012
2013         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2014                 goto fail;
2015
2016         if (current_trace && current_trace->print_max)
2017                 iter->tr = &max_tr;
2018         else
2019                 iter->tr = &global_trace;
2020         iter->pos = -1;
2021         mutex_init(&iter->mutex);
2022         iter->cpu_file = cpu_file;
2023
2024         /* Notify the tracer early; before we stop tracing. */
2025         if (iter->trace && iter->trace->open)
2026                 iter->trace->open(iter);
2027
2028         /* Annotate start of buffers if we had overruns */
2029         if (ring_buffer_overruns(iter->tr->buffer))
2030                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2031
2032         /* stop the trace while dumping */
2033         tracing_stop();
2034
2035         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2036                 for_each_tracing_cpu(cpu) {
2037
2038                         iter->buffer_iter[cpu] =
2039                                 ring_buffer_read_start(iter->tr->buffer, cpu);
2040                         tracing_iter_reset(iter, cpu);
2041                 }
2042         } else {
2043                 cpu = iter->cpu_file;
2044                 iter->buffer_iter[cpu] =
2045                                 ring_buffer_read_start(iter->tr->buffer, cpu);
2046                 tracing_iter_reset(iter, cpu);
2047         }
2048
2049         ret = seq_open(file, &tracer_seq_ops);
2050         if (ret < 0) {
2051                 fail_ret = ERR_PTR(ret);
2052                 goto fail_buffer;
2053         }
2054
2055         m = file->private_data;
2056         m->private = iter;
2057
2058         mutex_unlock(&trace_types_lock);
2059
2060         return iter;
2061
2062  fail_buffer:
2063         for_each_tracing_cpu(cpu) {
2064                 if (iter->buffer_iter[cpu])
2065                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2066         }
2067         free_cpumask_var(iter->started);
2068         tracing_start();
2069  fail:
2070         mutex_unlock(&trace_types_lock);
2071         kfree(iter->trace);
2072         kfree(iter);
2073
2074         return fail_ret;
2075 }
2076
2077 int tracing_open_generic(struct inode *inode, struct file *filp)
2078 {
2079         if (tracing_disabled)
2080                 return -ENODEV;
2081
2082         filp->private_data = inode->i_private;
2083         return 0;
2084 }
2085
2086 static int tracing_release(struct inode *inode, struct file *file)
2087 {
2088         struct seq_file *m = (struct seq_file *)file->private_data;
2089         struct trace_iterator *iter;
2090         int cpu;
2091
2092         if (!(file->f_mode & FMODE_READ))
2093                 return 0;
2094
2095         iter = m->private;
2096
2097         mutex_lock(&trace_types_lock);
2098         for_each_tracing_cpu(cpu) {
2099                 if (iter->buffer_iter[cpu])
2100                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2101         }
2102
2103         if (iter->trace && iter->trace->close)
2104                 iter->trace->close(iter);
2105
2106         /* reenable tracing if it was previously enabled */
2107         tracing_start();
2108         mutex_unlock(&trace_types_lock);
2109
2110         seq_release(inode, file);
2111         mutex_destroy(&iter->mutex);
2112         free_cpumask_var(iter->started);
2113         kfree(iter->trace);
2114         kfree(iter);
2115         return 0;
2116 }
2117
2118 static int tracing_open(struct inode *inode, struct file *file)
2119 {
2120         struct trace_iterator *iter;
2121         int ret = 0;
2122
2123         /* If this file was open for write, then erase contents */
2124         if ((file->f_mode & FMODE_WRITE) &&
2125             (file->f_flags & O_TRUNC)) {
2126                 long cpu = (long) inode->i_private;
2127
2128                 if (cpu == TRACE_PIPE_ALL_CPU)
2129                         tracing_reset_online_cpus(&global_trace);
2130                 else
2131                         tracing_reset(&global_trace, cpu);
2132         }
2133
2134         if (file->f_mode & FMODE_READ) {
2135                 iter = __tracing_open(inode, file);
2136                 if (IS_ERR(iter))
2137                         ret = PTR_ERR(iter);
2138                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2139                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2140         }
2141         return ret;
2142 }
2143
2144 static void *
2145 t_next(struct seq_file *m, void *v, loff_t *pos)
2146 {
2147         struct tracer *t = v;
2148
2149         (*pos)++;
2150
2151         if (t)
2152                 t = t->next;
2153
2154         return t;
2155 }
2156
2157 static void *t_start(struct seq_file *m, loff_t *pos)
2158 {
2159         struct tracer *t;
2160         loff_t l = 0;
2161
2162         mutex_lock(&trace_types_lock);
2163         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2164                 ;
2165
2166         return t;
2167 }
2168
2169 static void t_stop(struct seq_file *m, void *p)
2170 {
2171         mutex_unlock(&trace_types_lock);
2172 }
2173
2174 static int t_show(struct seq_file *m, void *v)
2175 {
2176         struct tracer *t = v;
2177
2178         if (!t)
2179                 return 0;
2180
2181         seq_printf(m, "%s", t->name);
2182         if (t->next)
2183                 seq_putc(m, ' ');
2184         else
2185                 seq_putc(m, '\n');
2186
2187         return 0;
2188 }
2189
2190 static const struct seq_operations show_traces_seq_ops = {
2191         .start          = t_start,
2192         .next           = t_next,
2193         .stop           = t_stop,
2194         .show           = t_show,
2195 };
2196
2197 static int show_traces_open(struct inode *inode, struct file *file)
2198 {
2199         if (tracing_disabled)
2200                 return -ENODEV;
2201
2202         return seq_open(file, &show_traces_seq_ops);
2203 }
2204
2205 static ssize_t
2206 tracing_write_stub(struct file *filp, const char __user *ubuf,
2207                    size_t count, loff_t *ppos)
2208 {
2209         return count;
2210 }
2211
2212 static const struct file_operations tracing_fops = {
2213         .open           = tracing_open,
2214         .read           = seq_read,
2215         .write          = tracing_write_stub,
2216         .llseek         = seq_lseek,
2217         .release        = tracing_release,
2218 };
2219
2220 static const struct file_operations show_traces_fops = {
2221         .open           = show_traces_open,
2222         .read           = seq_read,
2223         .release        = seq_release,
2224 };
2225
2226 /*
2227  * Only trace on a CPU if the bitmask is set:
2228  */
2229 static cpumask_var_t tracing_cpumask;
2230
2231 /*
2232  * The tracer itself will not take this lock, but still we want
2233  * to provide a consistent cpumask to user-space:
2234  */
2235 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2236
2237 /*
2238  * Temporary storage for the character representation of the
2239  * CPU bitmask (and one more byte for the newline):
2240  */
2241 static char mask_str[NR_CPUS + 1];
2242
2243 static ssize_t
2244 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2245                      size_t count, loff_t *ppos)
2246 {
2247         int len;
2248
2249         mutex_lock(&tracing_cpumask_update_lock);
2250
2251         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2252         if (count - len < 2) {
2253                 count = -EINVAL;
2254                 goto out_err;
2255         }
2256         len += sprintf(mask_str + len, "\n");
2257         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2258
2259 out_err:
2260         mutex_unlock(&tracing_cpumask_update_lock);
2261
2262         return count;
2263 }
2264
2265 static ssize_t
2266 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2267                       size_t count, loff_t *ppos)
2268 {
2269         int err, cpu;
2270         cpumask_var_t tracing_cpumask_new;
2271
2272         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2273                 return -ENOMEM;
2274
2275         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2276         if (err)
2277                 goto err_unlock;
2278
2279         mutex_lock(&tracing_cpumask_update_lock);
2280
2281         local_irq_disable();
2282         __raw_spin_lock(&ftrace_max_lock);
2283         for_each_tracing_cpu(cpu) {
2284                 /*
2285                  * Increase/decrease the disabled counter if we are
2286                  * about to flip a bit in the cpumask:
2287                  */
2288                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2289                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2290                         atomic_inc(&global_trace.data[cpu]->disabled);
2291                 }
2292                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2293                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2294                         atomic_dec(&global_trace.data[cpu]->disabled);
2295                 }
2296         }
2297         __raw_spin_unlock(&ftrace_max_lock);
2298         local_irq_enable();
2299
2300         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2301
2302         mutex_unlock(&tracing_cpumask_update_lock);
2303         free_cpumask_var(tracing_cpumask_new);
2304
2305         return count;
2306
2307 err_unlock:
2308         free_cpumask_var(tracing_cpumask_new);
2309
2310         return err;
2311 }
2312
2313 static const struct file_operations tracing_cpumask_fops = {
2314         .open           = tracing_open_generic,
2315         .read           = tracing_cpumask_read,
2316         .write          = tracing_cpumask_write,
2317 };
2318
2319 static ssize_t
2320 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2321                        size_t cnt, loff_t *ppos)
2322 {
2323         struct tracer_opt *trace_opts;
2324         u32 tracer_flags;
2325         int len = 0;
2326         char *buf;
2327         int r = 0;
2328         int i;
2329
2330
2331         /* calculate max size */
2332         for (i = 0; trace_options[i]; i++) {
2333                 len += strlen(trace_options[i]);
2334                 len += 3; /* "no" and newline */
2335         }
2336
2337         mutex_lock(&trace_types_lock);
2338         tracer_flags = current_trace->flags->val;
2339         trace_opts = current_trace->flags->opts;
2340
2341         /*
2342          * Increase the size with names of options specific
2343          * of the current tracer.
2344          */
2345         for (i = 0; trace_opts[i].name; i++) {
2346                 len += strlen(trace_opts[i].name);
2347                 len += 3; /* "no" and newline */
2348         }
2349
2350         /* +1 for \0 */
2351         buf = kmalloc(len + 1, GFP_KERNEL);
2352         if (!buf) {
2353                 mutex_unlock(&trace_types_lock);
2354                 return -ENOMEM;
2355         }
2356
2357         for (i = 0; trace_options[i]; i++) {
2358                 if (trace_flags & (1 << i))
2359                         r += sprintf(buf + r, "%s\n", trace_options[i]);
2360                 else
2361                         r += sprintf(buf + r, "no%s\n", trace_options[i]);
2362         }
2363
2364         for (i = 0; trace_opts[i].name; i++) {
2365                 if (tracer_flags & trace_opts[i].bit)
2366                         r += sprintf(buf + r, "%s\n",
2367                                 trace_opts[i].name);
2368                 else
2369                         r += sprintf(buf + r, "no%s\n",
2370                                 trace_opts[i].name);
2371         }
2372         mutex_unlock(&trace_types_lock);
2373
2374         WARN_ON(r >= len + 1);
2375
2376         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2377
2378         kfree(buf);
2379         return r;
2380 }
2381
2382 /* Try to assign a tracer specific option */
2383 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2384 {
2385         struct tracer_flags *tracer_flags = trace->flags;
2386         struct tracer_opt *opts = NULL;
2387         int ret = 0, i = 0;
2388         int len;
2389
2390         for (i = 0; tracer_flags->opts[i].name; i++) {
2391                 opts = &tracer_flags->opts[i];
2392                 len = strlen(opts->name);
2393
2394                 if (strncmp(cmp, opts->name, len) == 0) {
2395                         ret = trace->set_flag(tracer_flags->val,
2396                                 opts->bit, !neg);
2397                         break;
2398                 }
2399         }
2400         /* Not found */
2401         if (!tracer_flags->opts[i].name)
2402                 return -EINVAL;
2403
2404         /* Refused to handle */
2405         if (ret)
2406                 return ret;
2407
2408         if (neg)
2409                 tracer_flags->val &= ~opts->bit;
2410         else
2411                 tracer_flags->val |= opts->bit;
2412
2413         return 0;
2414 }
2415
2416 static void set_tracer_flags(unsigned int mask, int enabled)
2417 {
2418         /* do nothing if flag is already set */
2419         if (!!(trace_flags & mask) == !!enabled)
2420                 return;
2421
2422         if (enabled)
2423                 trace_flags |= mask;
2424         else
2425                 trace_flags &= ~mask;
2426 }
2427
2428 static ssize_t
2429 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2430                         size_t cnt, loff_t *ppos)
2431 {
2432         char buf[64];
2433         char *cmp = buf;
2434         int neg = 0;
2435         int ret;
2436         int i;
2437
2438         if (cnt >= sizeof(buf))
2439                 return -EINVAL;
2440
2441         if (copy_from_user(&buf, ubuf, cnt))
2442                 return -EFAULT;
2443
2444         buf[cnt] = 0;
2445
2446         if (strncmp(buf, "no", 2) == 0) {
2447                 neg = 1;
2448                 cmp += 2;
2449         }
2450
2451         for (i = 0; trace_options[i]; i++) {
2452                 int len = strlen(trace_options[i]);
2453
2454                 if (strncmp(cmp, trace_options[i], len) == 0) {
2455                         set_tracer_flags(1 << i, !neg);
2456                         break;
2457                 }
2458         }
2459
2460         /* If no option could be set, test the specific tracer options */
2461         if (!trace_options[i]) {
2462                 mutex_lock(&trace_types_lock);
2463                 ret = set_tracer_option(current_trace, cmp, neg);
2464                 mutex_unlock(&trace_types_lock);
2465                 if (ret)
2466                         return ret;
2467         }
2468
2469         *ppos += cnt;
2470
2471         return cnt;
2472 }
2473
2474 static const struct file_operations tracing_iter_fops = {
2475         .open           = tracing_open_generic,
2476         .read           = tracing_trace_options_read,
2477         .write          = tracing_trace_options_write,
2478 };
2479
2480 static const char readme_msg[] =
2481         "tracing mini-HOWTO:\n\n"
2482         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2483         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2484         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2485         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2486         "nop\n"
2487         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2488         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2489         "sched_switch\n"
2490         "# cat /sys/kernel/debug/tracing/trace_options\n"
2491         "noprint-parent nosym-offset nosym-addr noverbose\n"
2492         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2493         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2494         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2495         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2496 ;
2497
2498 static ssize_t
2499 tracing_readme_read(struct file *filp, char __user *ubuf,
2500                        size_t cnt, loff_t *ppos)
2501 {
2502         return simple_read_from_buffer(ubuf, cnt, ppos,
2503                                         readme_msg, strlen(readme_msg));
2504 }
2505
2506 static const struct file_operations tracing_readme_fops = {
2507         .open           = tracing_open_generic,
2508         .read           = tracing_readme_read,
2509 };
2510
2511 static ssize_t
2512 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2513                                 size_t cnt, loff_t *ppos)
2514 {
2515         char *buf_comm;
2516         char *file_buf;
2517         char *buf;
2518         int len = 0;
2519         int pid;
2520         int i;
2521
2522         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2523         if (!file_buf)
2524                 return -ENOMEM;
2525
2526         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2527         if (!buf_comm) {
2528                 kfree(file_buf);
2529                 return -ENOMEM;
2530         }
2531
2532         buf = file_buf;
2533
2534         for (i = 0; i < SAVED_CMDLINES; i++) {
2535                 int r;
2536
2537                 pid = map_cmdline_to_pid[i];
2538                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2539                         continue;
2540
2541                 trace_find_cmdline(pid, buf_comm);
2542                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2543                 buf += r;
2544                 len += r;
2545         }
2546
2547         len = simple_read_from_buffer(ubuf, cnt, ppos,
2548                                       file_buf, len);
2549
2550         kfree(file_buf);
2551         kfree(buf_comm);
2552
2553         return len;
2554 }
2555
2556 static const struct file_operations tracing_saved_cmdlines_fops = {
2557     .open       = tracing_open_generic,
2558     .read       = tracing_saved_cmdlines_read,
2559 };
2560
2561 static ssize_t
2562 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2563                   size_t cnt, loff_t *ppos)
2564 {
2565         char buf[64];
2566         int r;
2567
2568         r = sprintf(buf, "%u\n", tracer_enabled);
2569         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2570 }
2571
2572 static ssize_t
2573 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2574                    size_t cnt, loff_t *ppos)
2575 {
2576         struct trace_array *tr = filp->private_data;
2577         char buf[64];
2578         unsigned long val;
2579         int ret;
2580
2581         if (cnt >= sizeof(buf))
2582                 return -EINVAL;
2583
2584         if (copy_from_user(&buf, ubuf, cnt))
2585                 return -EFAULT;
2586
2587         buf[cnt] = 0;
2588
2589         ret = strict_strtoul(buf, 10, &val);
2590         if (ret < 0)
2591                 return ret;
2592
2593         val = !!val;
2594
2595         mutex_lock(&trace_types_lock);
2596         if (tracer_enabled ^ val) {
2597                 if (val) {
2598                         tracer_enabled = 1;
2599                         if (current_trace->start)
2600                                 current_trace->start(tr);
2601                         tracing_start();
2602                 } else {
2603                         tracer_enabled = 0;
2604                         tracing_stop();
2605                         if (current_trace->stop)
2606                                 current_trace->stop(tr);
2607                 }
2608         }
2609         mutex_unlock(&trace_types_lock);
2610
2611         *ppos += cnt;
2612
2613         return cnt;
2614 }
2615
2616 static ssize_t
2617 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2618                        size_t cnt, loff_t *ppos)
2619 {
2620         char buf[MAX_TRACER_SIZE+2];
2621         int r;
2622
2623         mutex_lock(&trace_types_lock);
2624         if (current_trace)
2625                 r = sprintf(buf, "%s\n", current_trace->name);
2626         else
2627                 r = sprintf(buf, "\n");
2628         mutex_unlock(&trace_types_lock);
2629
2630         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2631 }
2632
2633 int tracer_init(struct tracer *t, struct trace_array *tr)
2634 {
2635         tracing_reset_online_cpus(tr);
2636         return t->init(tr);
2637 }
2638
2639 static int tracing_resize_ring_buffer(unsigned long size)
2640 {
2641         int ret;
2642
2643         /*
2644          * If kernel or user changes the size of the ring buffer
2645          * we use the size that was given, and we can forget about
2646          * expanding it later.
2647          */
2648         ring_buffer_expanded = 1;
2649
2650         ret = ring_buffer_resize(global_trace.buffer, size);
2651         if (ret < 0)
2652                 return ret;
2653
2654         ret = ring_buffer_resize(max_tr.buffer, size);
2655         if (ret < 0) {
2656                 int r;
2657
2658                 r = ring_buffer_resize(global_trace.buffer,
2659                                        global_trace.entries);
2660                 if (r < 0) {
2661                         /*
2662                          * AARGH! We are left with different
2663                          * size max buffer!!!!
2664                          * The max buffer is our "snapshot" buffer.
2665                          * When a tracer needs a snapshot (one of the
2666                          * latency tracers), it swaps the max buffer
2667                          * with the saved snap shot. We succeeded to
2668                          * update the size of the main buffer, but failed to
2669                          * update the size of the max buffer. But when we tried
2670                          * to reset the main buffer to the original size, we
2671                          * failed there too. This is very unlikely to
2672                          * happen, but if it does, warn and kill all
2673                          * tracing.
2674                          */
2675                         WARN_ON(1);
2676                         tracing_disabled = 1;
2677                 }
2678                 return ret;
2679         }
2680
2681         global_trace.entries = size;
2682
2683         return ret;
2684 }
2685
2686 /**
2687  * tracing_update_buffers - used by tracing facility to expand ring buffers
2688  *
2689  * To save on memory when the tracing is never used on a system with it
2690  * configured in. The ring buffers are set to a minimum size. But once
2691  * a user starts to use the tracing facility, then they need to grow
2692  * to their default size.
2693  *
2694  * This function is to be called when a tracer is about to be used.
2695  */
2696 int tracing_update_buffers(void)
2697 {
2698         int ret = 0;
2699
2700         mutex_lock(&trace_types_lock);
2701         if (!ring_buffer_expanded)
2702                 ret = tracing_resize_ring_buffer(trace_buf_size);
2703         mutex_unlock(&trace_types_lock);
2704
2705         return ret;
2706 }
2707
2708 struct trace_option_dentry;
2709
2710 static struct trace_option_dentry *
2711 create_trace_option_files(struct tracer *tracer);
2712
2713 static void
2714 destroy_trace_option_files(struct trace_option_dentry *topts);
2715
2716 static int tracing_set_tracer(const char *buf)
2717 {
2718         static struct trace_option_dentry *topts;
2719         struct trace_array *tr = &global_trace;
2720         struct tracer *t;
2721         int ret = 0;
2722
2723         mutex_lock(&trace_types_lock);
2724
2725         if (!ring_buffer_expanded) {
2726                 ret = tracing_resize_ring_buffer(trace_buf_size);
2727                 if (ret < 0)
2728                         goto out;
2729                 ret = 0;
2730         }
2731
2732         for (t = trace_types; t; t = t->next) {
2733                 if (strcmp(t->name, buf) == 0)
2734                         break;
2735         }
2736         if (!t) {
2737                 ret = -EINVAL;
2738                 goto out;
2739         }
2740         if (t == current_trace)
2741                 goto out;
2742
2743         trace_branch_disable();
2744         if (current_trace && current_trace->reset)
2745                 current_trace->reset(tr);
2746
2747         destroy_trace_option_files(topts);
2748
2749         current_trace = t;
2750
2751         topts = create_trace_option_files(current_trace);
2752
2753         if (t->init) {
2754                 ret = tracer_init(t, tr);
2755                 if (ret)
2756                         goto out;
2757         }
2758
2759         trace_branch_enable(tr);
2760  out:
2761         mutex_unlock(&trace_types_lock);
2762
2763         return ret;
2764 }
2765
2766 static ssize_t
2767 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2768                         size_t cnt, loff_t *ppos)
2769 {
2770         char buf[MAX_TRACER_SIZE+1];
2771         int i;
2772         size_t ret;
2773         int err;
2774
2775         ret = cnt;
2776
2777         if (cnt > MAX_TRACER_SIZE)
2778                 cnt = MAX_TRACER_SIZE;
2779
2780         if (copy_from_user(&buf, ubuf, cnt))
2781                 return -EFAULT;
2782
2783         buf[cnt] = 0;
2784
2785         /* strip ending whitespace. */
2786         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2787                 buf[i] = 0;
2788
2789         err = tracing_set_tracer(buf);
2790         if (err)
2791                 return err;
2792
2793         *ppos += ret;
2794
2795         return ret;
2796 }
2797
2798 static ssize_t
2799 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2800                      size_t cnt, loff_t *ppos)
2801 {
2802         unsigned long *ptr = filp->private_data;
2803         char buf[64];
2804         int r;
2805
2806         r = snprintf(buf, sizeof(buf), "%ld\n",
2807                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2808         if (r > sizeof(buf))
2809                 r = sizeof(buf);
2810         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2811 }
2812
2813 static ssize_t
2814 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2815                       size_t cnt, loff_t *ppos)
2816 {
2817         unsigned long *ptr = filp->private_data;
2818         char buf[64];
2819         unsigned long val;
2820         int ret;
2821
2822         if (cnt >= sizeof(buf))
2823                 return -EINVAL;
2824
2825         if (copy_from_user(&buf, ubuf, cnt))
2826                 return -EFAULT;
2827
2828         buf[cnt] = 0;
2829
2830         ret = strict_strtoul(buf, 10, &val);
2831         if (ret < 0)
2832                 return ret;
2833
2834         *ptr = val * 1000;
2835
2836         return cnt;
2837 }
2838
2839 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2840 {
2841         long cpu_file = (long) inode->i_private;
2842         struct trace_iterator *iter;
2843         int ret = 0;
2844
2845         if (tracing_disabled)
2846                 return -ENODEV;
2847
2848         mutex_lock(&trace_types_lock);
2849
2850         /* We only allow one reader per cpu */
2851         if (cpu_file == TRACE_PIPE_ALL_CPU) {
2852                 if (!cpumask_empty(tracing_reader_cpumask)) {
2853                         ret = -EBUSY;
2854                         goto out;
2855                 }
2856                 cpumask_setall(tracing_reader_cpumask);
2857         } else {
2858                 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2859                         cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2860                 else {
2861                         ret = -EBUSY;
2862                         goto out;
2863                 }
2864         }
2865
2866         /* create a buffer to store the information to pass to userspace */
2867         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2868         if (!iter) {
2869                 ret = -ENOMEM;
2870                 goto out;
2871         }
2872
2873         /*
2874          * We make a copy of the current tracer to avoid concurrent
2875          * changes on it while we are reading.
2876          */
2877         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2878         if (!iter->trace) {
2879                 ret = -ENOMEM;
2880                 goto fail;
2881         }
2882         if (current_trace)
2883                 *iter->trace = *current_trace;
2884
2885         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2886                 ret = -ENOMEM;
2887                 goto fail;
2888         }
2889
2890         /* trace pipe does not show start of buffer */
2891         cpumask_setall(iter->started);
2892
2893         if (trace_flags & TRACE_ITER_LATENCY_FMT)
2894                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
2895
2896         iter->cpu_file = cpu_file;
2897         iter->tr = &global_trace;
2898         mutex_init(&iter->mutex);
2899         filp->private_data = iter;
2900
2901         if (iter->trace->pipe_open)
2902                 iter->trace->pipe_open(iter);
2903
2904 out:
2905         mutex_unlock(&trace_types_lock);
2906         return ret;
2907
2908 fail:
2909         kfree(iter->trace);
2910         kfree(iter);
2911         mutex_unlock(&trace_types_lock);
2912         return ret;
2913 }
2914
2915 static int tracing_release_pipe(struct inode *inode, struct file *file)
2916 {
2917         struct trace_iterator *iter = file->private_data;
2918
2919         mutex_lock(&trace_types_lock);
2920
2921         if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2922                 cpumask_clear(tracing_reader_cpumask);
2923         else
2924                 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2925
2926
2927         if (iter->trace->pipe_close)
2928                 iter->trace->pipe_close(iter);
2929
2930         mutex_unlock(&trace_types_lock);
2931
2932         free_cpumask_var(iter->started);
2933         mutex_destroy(&iter->mutex);
2934         kfree(iter->trace);
2935         kfree(iter);
2936
2937         return 0;
2938 }
2939
2940 static unsigned int
2941 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2942 {
2943         struct trace_iterator *iter = filp->private_data;
2944
2945         if (trace_flags & TRACE_ITER_BLOCK) {
2946                 /*
2947                  * Always select as readable when in blocking mode
2948                  */
2949                 return POLLIN | POLLRDNORM;
2950         } else {
2951                 if (!trace_empty(iter))
2952                         return POLLIN | POLLRDNORM;
2953                 poll_wait(filp, &trace_wait, poll_table);
2954                 if (!trace_empty(iter))
2955                         return POLLIN | POLLRDNORM;
2956
2957                 return 0;
2958         }
2959 }
2960
2961
2962 void default_wait_pipe(struct trace_iterator *iter)
2963 {
2964         DEFINE_WAIT(wait);
2965
2966         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2967
2968         if (trace_empty(iter))
2969                 schedule();
2970
2971         finish_wait(&trace_wait, &wait);
2972 }
2973
2974 /*
2975  * This is a make-shift waitqueue.
2976  * A tracer might use this callback on some rare cases:
2977  *
2978  *  1) the current tracer might hold the runqueue lock when it wakes up
2979  *     a reader, hence a deadlock (sched, function, and function graph tracers)
2980  *  2) the function tracers, trace all functions, we don't want
2981  *     the overhead of calling wake_up and friends
2982  *     (and tracing them too)
2983  *
2984  *     Anyway, this is really very primitive wakeup.
2985  */
2986 void poll_wait_pipe(struct trace_iterator *iter)
2987 {
2988         set_current_state(TASK_INTERRUPTIBLE);
2989         /* sleep for 100 msecs, and try again. */
2990         schedule_timeout(HZ / 10);
2991 }
2992
2993 /* Must be called with trace_types_lock mutex held. */
2994 static int tracing_wait_pipe(struct file *filp)
2995 {
2996         struct trace_iterator *iter = filp->private_data;
2997
2998         while (trace_empty(iter)) {
2999
3000                 if ((filp->f_flags & O_NONBLOCK)) {
3001                         return -EAGAIN;
3002                 }
3003
3004                 mutex_unlock(&iter->mutex);
3005
3006                 iter->trace->wait_pipe(iter);
3007
3008                 mutex_lock(&iter->mutex);
3009
3010                 if (signal_pending(current))
3011                         return -EINTR;
3012
3013                 /*
3014                  * We block until we read something and tracing is disabled.
3015                  * We still block if tracing is disabled, but we have never
3016                  * read anything. This allows a user to cat this file, and
3017                  * then enable tracing. But after we have read something,
3018                  * we give an EOF when tracing is again disabled.
3019                  *
3020                  * iter->pos will be 0 if we haven't read anything.
3021                  */
3022                 if (!tracer_enabled && iter->pos)
3023                         break;
3024         }
3025
3026         return 1;
3027 }
3028
3029 /*
3030  * Consumer reader.
3031  */
3032 static ssize_t
3033 tracing_read_pipe(struct file *filp, char __user *ubuf,
3034                   size_t cnt, loff_t *ppos)
3035 {
3036         struct trace_iterator *iter = filp->private_data;
3037         static struct tracer *old_tracer;
3038         ssize_t sret;
3039
3040         /* return any leftover data */
3041         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3042         if (sret != -EBUSY)
3043                 return sret;
3044
3045         trace_seq_init(&iter->seq);
3046
3047         /* copy the tracer to avoid using a global lock all around */
3048         mutex_lock(&trace_types_lock);
3049         if (unlikely(old_tracer != current_trace && current_trace)) {
3050                 old_tracer = current_trace;
3051                 *iter->trace = *current_trace;
3052         }
3053         mutex_unlock(&trace_types_lock);
3054
3055         /*
3056          * Avoid more than one consumer on a single file descriptor
3057          * This is just a matter of traces coherency, the ring buffer itself
3058          * is protected.
3059          */
3060         mutex_lock(&iter->mutex);
3061         if (iter->trace->read) {
3062                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3063                 if (sret)
3064                         goto out;
3065         }
3066
3067 waitagain:
3068         sret = tracing_wait_pipe(filp);
3069         if (sret <= 0)
3070                 goto out;
3071
3072         /* stop when tracing is finished */
3073         if (trace_empty(iter)) {
3074                 sret = 0;
3075                 goto out;
3076         }
3077
3078         if (cnt >= PAGE_SIZE)
3079                 cnt = PAGE_SIZE - 1;
3080
3081         /* reset all but tr, trace, and overruns */
3082         memset(&iter->seq, 0,
3083                sizeof(struct trace_iterator) -
3084                offsetof(struct trace_iterator, seq));
3085         iter->pos = -1;
3086
3087         trace_event_read_lock();
3088         while (find_next_entry_inc(iter) != NULL) {
3089                 enum print_line_t ret;
3090                 int len = iter->seq.len;
3091
3092                 ret = print_trace_line(iter);
3093                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3094                         /* don't print partial lines */
3095                         iter->seq.len = len;
3096                         break;
3097                 }
3098                 if (ret != TRACE_TYPE_NO_CONSUME)
3099                         trace_consume(iter);
3100
3101                 if (iter->seq.len >= cnt)
3102                         break;
3103         }
3104         trace_event_read_unlock();
3105
3106         /* Now copy what we have to the user */
3107         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3108         if (iter->seq.readpos >= iter->seq.len)
3109                 trace_seq_init(&iter->seq);
3110
3111         /*
3112          * If there was nothing to send to user, inspite of consuming trace
3113          * entries, go back to wait for more entries.
3114          */
3115         if (sret == -EBUSY)
3116                 goto waitagain;
3117
3118 out:
3119         mutex_unlock(&iter->mutex);
3120
3121         return sret;
3122 }
3123
3124 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3125                                      struct pipe_buffer *buf)
3126 {
3127         __free_page(buf->page);
3128 }
3129
3130 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3131                                      unsigned int idx)
3132 {
3133         __free_page(spd->pages[idx]);
3134 }
3135
3136 static struct pipe_buf_operations tracing_pipe_buf_ops = {
3137         .can_merge              = 0,
3138         .map                    = generic_pipe_buf_map,
3139         .unmap                  = generic_pipe_buf_unmap,
3140         .confirm                = generic_pipe_buf_confirm,
3141         .release                = tracing_pipe_buf_release,
3142         .steal                  = generic_pipe_buf_steal,
3143         .get                    = generic_pipe_buf_get,
3144 };
3145
3146 static size_t
3147 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3148 {
3149         size_t count;
3150         int ret;
3151
3152         /* Seq buffer is page-sized, exactly what we need. */
3153         for (;;) {
3154                 count = iter->seq.len;
3155                 ret = print_trace_line(iter);
3156                 count = iter->seq.len - count;
3157                 if (rem < count) {
3158                         rem = 0;
3159                         iter->seq.len -= count;
3160                         break;
3161                 }
3162                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3163                         iter->seq.len -= count;
3164                         break;
3165                 }
3166
3167                 if (ret != TRACE_TYPE_NO_CONSUME)
3168                         trace_consume(iter);
3169                 rem -= count;
3170                 if (!find_next_entry_inc(iter)) {
3171                         rem = 0;
3172                         iter->ent = NULL;
3173                         break;
3174                 }
3175         }
3176
3177         return rem;
3178 }
3179
3180 static ssize_t tracing_splice_read_pipe(struct file *filp,
3181                                         loff_t *ppos,
3182                                         struct pipe_inode_info *pipe,
3183                                         size_t len,
3184                                         unsigned int flags)
3185 {
3186         struct page *pages[PIPE_BUFFERS];
3187         struct partial_page partial[PIPE_BUFFERS];
3188         struct trace_iterator *iter = filp->private_data;
3189         struct splice_pipe_desc spd = {
3190                 .pages          = pages,
3191                 .partial        = partial,
3192                 .nr_pages       = 0, /* This gets updated below. */
3193                 .flags          = flags,
3194                 .ops            = &tracing_pipe_buf_ops,
3195                 .spd_release    = tracing_spd_release_pipe,
3196         };
3197         static struct tracer *old_tracer;
3198         ssize_t ret;
3199         size_t rem;
3200         unsigned int i;
3201
3202         /* copy the tracer to avoid using a global lock all around */
3203         mutex_lock(&trace_types_lock);
3204         if (unlikely(old_tracer != current_trace && current_trace)) {
3205                 old_tracer = current_trace;
3206                 *iter->trace = *current_trace;
3207         }
3208         mutex_unlock(&trace_types_lock);
3209
3210         mutex_lock(&iter->mutex);
3211
3212         if (iter->trace->splice_read) {
3213                 ret = iter->trace->splice_read(iter, filp,
3214                                                ppos, pipe, len, flags);
3215                 if (ret)
3216                         goto out_err;
3217         }
3218
3219         ret = tracing_wait_pipe(filp);
3220         if (ret <= 0)
3221                 goto out_err;
3222
3223         if (!iter->ent && !find_next_entry_inc(iter)) {
3224                 ret = -EFAULT;
3225                 goto out_err;
3226         }
3227
3228         trace_event_read_lock();
3229
3230         /* Fill as many pages as possible. */
3231         for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3232                 pages[i] = alloc_page(GFP_KERNEL);
3233                 if (!pages[i])
3234                         break;
3235
3236                 rem = tracing_fill_pipe_page(rem, iter);
3237
3238                 /* Copy the data into the page, so we can start over. */
3239                 ret = trace_seq_to_buffer(&iter->seq,
3240                                           page_address(pages[i]),
3241                                           iter->seq.len);
3242                 if (ret < 0) {
3243                         __free_page(pages[i]);
3244                         break;
3245                 }
3246                 partial[i].offset = 0;
3247                 partial[i].len = iter->seq.len;
3248
3249                 trace_seq_init(&iter->seq);
3250         }
3251
3252         trace_event_read_unlock();
3253         mutex_unlock(&iter->mutex);
3254
3255         spd.nr_pages = i;
3256
3257         return splice_to_pipe(pipe, &spd);
3258
3259 out_err:
3260         mutex_unlock(&iter->mutex);
3261
3262         return ret;
3263 }
3264
3265 static ssize_t
3266 tracing_entries_read(struct file *filp, char __user *ubuf,
3267                      size_t cnt, loff_t *ppos)
3268 {
3269         struct trace_array *tr = filp->private_data;
3270         char buf[96];
3271         int r;
3272
3273         mutex_lock(&trace_types_lock);
3274         if (!ring_buffer_expanded)
3275                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3276                             tr->entries >> 10,
3277                             trace_buf_size >> 10);
3278         else
3279                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3280         mutex_unlock(&trace_types_lock);
3281
3282         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3283 }
3284
3285 static ssize_t
3286 tracing_entries_write(struct file *filp, const char __user *ubuf,
3287                       size_t cnt, loff_t *ppos)
3288 {
3289         unsigned long val;
3290         char buf[64];
3291         int ret, cpu;
3292
3293         if (cnt >= sizeof(buf))
3294                 return -EINVAL;
3295
3296         if (copy_from_user(&buf, ubuf, cnt))
3297                 return -EFAULT;
3298
3299         buf[cnt] = 0;
3300
3301         ret = strict_strtoul(buf, 10, &val);
3302         if (ret < 0)
3303                 return ret;
3304
3305         /* must have at least 1 entry */
3306         if (!val)
3307                 return -EINVAL;
3308
3309         mutex_lock(&trace_types_lock);
3310
3311         tracing_stop();
3312
3313         /* disable all cpu buffers */
3314         for_each_tracing_cpu(cpu) {
3315                 if (global_trace.data[cpu])
3316                         atomic_inc(&global_trace.data[cpu]->disabled);
3317                 if (max_tr.data[cpu])
3318                         atomic_inc(&max_tr.data[cpu]->disabled);
3319         }
3320
3321         /* value is in KB */
3322         val <<= 10;
3323
3324         if (val != global_trace.entries) {
3325                 ret = tracing_resize_ring_buffer(val);
3326                 if (ret < 0) {
3327                         cnt = ret;
3328                         goto out;
3329                 }
3330         }
3331
3332         *ppos += cnt;
3333
3334         /* If check pages failed, return ENOMEM */
3335         if (tracing_disabled)
3336                 cnt = -ENOMEM;
3337  out:
3338         for_each_tracing_cpu(cpu) {
3339                 if (global_trace.data[cpu])
3340                         atomic_dec(&global_trace.data[cpu]->disabled);
3341                 if (max_tr.data[cpu])
3342                         atomic_dec(&max_tr.data[cpu]->disabled);
3343         }
3344
3345         tracing_start();
3346         max_tr.entries = global_trace.entries;
3347         mutex_unlock(&trace_types_lock);
3348
3349         return cnt;
3350 }
3351
3352 static int mark_printk(const char *fmt, ...)
3353 {
3354         int ret;
3355         va_list args;
3356         va_start(args, fmt);
3357         ret = trace_vprintk(0, fmt, args);
3358         va_end(args);
3359         return ret;
3360 }
3361
3362 static ssize_t
3363 tracing_mark_write(struct file *filp, const char __user *ubuf,
3364                                         size_t cnt, loff_t *fpos)
3365 {
3366         char *buf;
3367
3368         if (tracing_disabled)
3369                 return -EINVAL;
3370
3371         if (cnt > TRACE_BUF_SIZE)
3372                 cnt = TRACE_BUF_SIZE;
3373
3374         buf = kmalloc(cnt + 2, GFP_KERNEL);
3375         if (buf == NULL)
3376                 return -ENOMEM;
3377
3378         if (copy_from_user(buf, ubuf, cnt)) {
3379                 kfree(buf);
3380                 return -EFAULT;
3381         }
3382         if (buf[cnt-1] != '\n') {
3383                 buf[cnt] = '\n';
3384                 buf[cnt+1] = '\0';
3385         } else
3386                 buf[cnt] = '\0';
3387
3388         cnt = mark_printk("%s", buf);
3389         kfree(buf);
3390         *fpos += cnt;
3391
3392         return cnt;
3393 }
3394
3395 static ssize_t tracing_clock_read(struct file *filp, char __user *ubuf,
3396                                   size_t cnt, loff_t *ppos)
3397 {
3398         char buf[64];
3399         int bufiter = 0;
3400         int i;
3401
3402         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3403                 bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter,
3404                         "%s%s%s%s", i ? " " : "",
3405                         i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3406                         i == trace_clock_id ? "]" : "");
3407         bufiter += snprintf(buf + bufiter, sizeof(buf) - bufiter, "\n");
3408
3409         return simple_read_from_buffer(ubuf, cnt, ppos, buf, bufiter);
3410 }
3411
3412 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3413                                    size_t cnt, loff_t *fpos)
3414 {
3415         char buf[64];
3416         const char *clockstr;
3417         int i;
3418
3419         if (cnt >= sizeof(buf))
3420                 return -EINVAL;
3421
3422         if (copy_from_user(&buf, ubuf, cnt))
3423                 return -EFAULT;
3424
3425         buf[cnt] = 0;
3426
3427         clockstr = strstrip(buf);
3428
3429         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3430                 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3431                         break;
3432         }
3433         if (i == ARRAY_SIZE(trace_clocks))
3434                 return -EINVAL;
3435
3436         trace_clock_id = i;
3437
3438         mutex_lock(&trace_types_lock);
3439
3440         ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3441         if (max_tr.buffer)
3442                 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3443
3444         mutex_unlock(&trace_types_lock);
3445
3446         *fpos += cnt;
3447
3448         return cnt;
3449 }
3450
3451 static const struct file_operations tracing_max_lat_fops = {
3452         .open           = tracing_open_generic,
3453         .read           = tracing_max_lat_read,
3454         .write          = tracing_max_lat_write,
3455 };
3456
3457 static const struct file_operations tracing_ctrl_fops = {
3458         .open           = tracing_open_generic,
3459         .read           = tracing_ctrl_read,
3460         .write          = tracing_ctrl_write,
3461 };
3462
3463 static const struct file_operations set_tracer_fops = {
3464         .open           = tracing_open_generic,
3465         .read           = tracing_set_trace_read,
3466         .write          = tracing_set_trace_write,
3467 };
3468
3469 static const struct file_operations tracing_pipe_fops = {
3470         .open           = tracing_open_pipe,
3471         .poll           = tracing_poll_pipe,
3472         .read           = tracing_read_pipe,
3473         .splice_read    = tracing_splice_read_pipe,
3474         .release        = tracing_release_pipe,
3475 };
3476
3477 static const struct file_operations tracing_entries_fops = {
3478         .open           = tracing_open_generic,
3479         .read           = tracing_entries_read,
3480         .write          = tracing_entries_write,
3481 };
3482
3483 static const struct file_operations tracing_mark_fops = {
3484         .open           = tracing_open_generic,
3485         .write          = tracing_mark_write,
3486 };
3487
3488 static const struct file_operations trace_clock_fops = {
3489         .open           = tracing_open_generic,
3490         .read           = tracing_clock_read,
3491         .write          = tracing_clock_write,
3492 };
3493
3494 struct ftrace_buffer_info {
3495         struct trace_array      *tr;
3496         void                    *spare;
3497         int                     cpu;
3498         unsigned int            read;
3499 };
3500
3501 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3502 {
3503         int cpu = (int)(long)inode->i_private;
3504         struct ftrace_buffer_info *info;
3505
3506         if (tracing_disabled)
3507                 return -ENODEV;
3508
3509         info = kzalloc(sizeof(*info), GFP_KERNEL);
3510         if (!info)
3511                 return -ENOMEM;
3512
3513         info->tr        = &global_trace;
3514         info->cpu       = cpu;
3515         info->spare     = NULL;
3516         /* Force reading ring buffer for first read */
3517         info->read      = (unsigned int)-1;
3518
3519         filp->private_data = info;
3520
3521         return nonseekable_open(inode, filp);
3522 }
3523
3524 static ssize_t
3525 tracing_buffers_read(struct file *filp, char __user *ubuf,
3526                      size_t count, loff_t *ppos)
3527 {
3528         struct ftrace_buffer_info *info = filp->private_data;
3529         unsigned int pos;
3530         ssize_t ret;
3531         size_t size;
3532
3533         if (!count)
3534                 return 0;
3535
3536         if (!info->spare)
3537                 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3538         if (!info->spare)
3539                 return -ENOMEM;
3540
3541         /* Do we have previous read data to read? */
3542         if (info->read < PAGE_SIZE)
3543                 goto read;
3544
3545         info->read = 0;
3546
3547         ret = ring_buffer_read_page(info->tr->buffer,
3548                                     &info->spare,
3549                                     count,
3550                                     info->cpu, 0);
3551         if (ret < 0)
3552                 return 0;
3553
3554         pos = ring_buffer_page_len(info->spare);
3555
3556         if (pos < PAGE_SIZE)
3557                 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3558
3559 read:
3560         size = PAGE_SIZE - info->read;
3561         if (size > count)
3562                 size = count;
3563
3564         ret = copy_to_user(ubuf, info->spare + info->read, size);
3565         if (ret == size)
3566                 return -EFAULT;
3567         size -= ret;
3568
3569         *ppos += size;
3570         info->read += size;
3571
3572         return size;
3573 }
3574
3575 static int tracing_buffers_release(struct inode *inode, struct file *file)
3576 {
3577         struct ftrace_buffer_info *info = file->private_data;
3578
3579         if (info->spare)
3580                 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3581         kfree(info);
3582
3583         return 0;
3584 }
3585
3586 struct buffer_ref {
3587         struct ring_buffer      *buffer;
3588         void                    *page;
3589         int                     ref;
3590 };
3591
3592 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3593                                     struct pipe_buffer *buf)
3594 {
3595         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3596
3597         if (--ref->ref)
3598                 return;
3599
3600         ring_buffer_free_read_page(ref->buffer, ref->page);
3601         kfree(ref);
3602         buf->private = 0;
3603 }
3604
3605 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3606                                  struct pipe_buffer *buf)
3607 {
3608         return 1;
3609 }
3610
3611 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3612                                 struct pipe_buffer *buf)
3613 {
3614         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3615
3616         ref->ref++;
3617 }
3618
3619 /* Pipe buffer operations for a buffer. */
3620 static struct pipe_buf_operations buffer_pipe_buf_ops = {
3621         .can_merge              = 0,
3622         .map                    = generic_pipe_buf_map,
3623         .unmap                  = generic_pipe_buf_unmap,
3624         .confirm                = generic_pipe_buf_confirm,
3625         .release                = buffer_pipe_buf_release,
3626         .steal                  = buffer_pipe_buf_steal,
3627         .get                    = buffer_pipe_buf_get,
3628 };
3629
3630 /*
3631  * Callback from splice_to_pipe(), if we need to release some pages
3632  * at the end of the spd in case we error'ed out in filling the pipe.
3633  */
3634 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3635 {
3636         struct buffer_ref *ref =
3637                 (struct buffer_ref *)spd->partial[i].private;
3638
3639         if (--ref->ref)
3640                 return;
3641
3642         ring_buffer_free_read_page(ref->buffer, ref->page);
3643         kfree(ref);
3644         spd->partial[i].private = 0;
3645 }
3646
3647 static ssize_t
3648 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3649                             struct pipe_inode_info *pipe, size_t len,
3650                             unsigned int flags)
3651 {
3652         struct ftrace_buffer_info *info = file->private_data;
3653         struct partial_page partial[PIPE_BUFFERS];
3654         struct page *pages[PIPE_BUFFERS];
3655         struct splice_pipe_desc spd = {
3656                 .pages          = pages,
3657                 .partial        = partial,
3658                 .flags          = flags,
3659                 .ops            = &buffer_pipe_buf_ops,
3660                 .spd_release    = buffer_spd_release,
3661         };
3662         struct buffer_ref *ref;
3663         int entries, size, i;
3664         size_t ret;
3665
3666         if (*ppos & (PAGE_SIZE - 1)) {
3667                 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3668                 return -EINVAL;
3669         }
3670
3671         if (len & (PAGE_SIZE - 1)) {
3672                 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3673                 if (len < PAGE_SIZE)
3674                         return -EINVAL;
3675                 len &= PAGE_MASK;
3676         }
3677
3678         entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3679
3680         for (i = 0; i < PIPE_BUFFERS && len && entries; i++, len -= PAGE_SIZE) {
3681                 struct page *page;
3682                 int r;
3683
3684                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3685                 if (!ref)
3686                         break;
3687
3688                 ref->ref = 1;
3689                 ref->buffer = info->tr->buffer;
3690                 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3691                 if (!ref->page) {
3692                         kfree(ref);
3693                         break;
3694                 }
3695
3696                 r = ring_buffer_read_page(ref->buffer, &ref->page,
3697                                           len, info->cpu, 1);
3698                 if (r < 0) {
3699                         ring_buffer_free_read_page(ref->buffer,
3700                                                    ref->page);
3701                         kfree(ref);
3702                         break;
3703                 }
3704
3705                 /*
3706                  * zero out any left over data, this is going to
3707                  * user land.
3708                  */
3709                 size = ring_buffer_page_len(ref->page);
3710                 if (size < PAGE_SIZE)
3711                         memset(ref->page + size, 0, PAGE_SIZE - size);
3712
3713                 page = virt_to_page(ref->page);
3714
3715                 spd.pages[i] = page;
3716                 spd.partial[i].len = PAGE_SIZE;
3717                 spd.partial[i].offset = 0;
3718                 spd.partial[i].private = (unsigned long)ref;
3719                 spd.nr_pages++;
3720                 *ppos += PAGE_SIZE;
3721
3722                 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3723         }
3724
3725         spd.nr_pages = i;
3726
3727         /* did we read anything? */
3728         if (!spd.nr_pages) {
3729                 if (flags & SPLICE_F_NONBLOCK)
3730                         ret = -EAGAIN;
3731                 else
3732                         ret = 0;
3733                 /* TODO: block */
3734                 return ret;
3735         }
3736
3737         ret = splice_to_pipe(pipe, &spd);
3738
3739         return ret;
3740 }
3741
3742 static const struct file_operations tracing_buffers_fops = {
3743         .open           = tracing_buffers_open,
3744         .read           = tracing_buffers_read,
3745         .release        = tracing_buffers_release,
3746         .splice_read    = tracing_buffers_splice_read,
3747         .llseek         = no_llseek,
3748 };
3749
3750 static ssize_t
3751 tracing_stats_read(struct file *filp, char __user *ubuf,
3752                    size_t count, loff_t *ppos)
3753 {
3754         unsigned long cpu = (unsigned long)filp->private_data;
3755         struct trace_array *tr = &global_trace;
3756         struct trace_seq *s;
3757         unsigned long cnt;
3758
3759         s = kmalloc(sizeof(*s), GFP_KERNEL);
3760         if (!s)
3761                 return -ENOMEM;
3762
3763         trace_seq_init(s);
3764
3765         cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3766         trace_seq_printf(s, "entries: %ld\n", cnt);
3767
3768         cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3769         trace_seq_printf(s, "overrun: %ld\n", cnt);
3770
3771         cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3772         trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3773
3774         count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3775
3776         kfree(s);
3777
3778         return count;
3779 }
3780
3781 static const struct file_operations tracing_stats_fops = {
3782         .open           = tracing_open_generic,
3783         .read           = tracing_stats_read,
3784 };
3785
3786 #ifdef CONFIG_DYNAMIC_FTRACE
3787
3788 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3789 {
3790         return 0;
3791 }
3792
3793 static ssize_t
3794 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3795                   size_t cnt, loff_t *ppos)
3796 {
3797         static char ftrace_dyn_info_buffer[1024];
3798         static DEFINE_MUTEX(dyn_info_mutex);
3799         unsigned long *p = filp->private_data;
3800         char *buf = ftrace_dyn_info_buffer;
3801         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3802         int r;
3803
3804         mutex_lock(&dyn_info_mutex);
3805         r = sprintf(buf, "%ld ", *p);
3806
3807         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3808         buf[r++] = '\n';
3809
3810         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3811
3812         mutex_unlock(&dyn_info_mutex);
3813
3814         return r;
3815 }
3816
3817 static const struct file_operations tracing_dyn_info_fops = {
3818         .open           = tracing_open_generic,
3819         .read           = tracing_read_dyn_info,
3820 };
3821 #endif
3822
3823 static struct dentry *d_tracer;
3824
3825 struct dentry *tracing_init_dentry(void)
3826 {
3827         static int once;
3828
3829         if (d_tracer)
3830                 return d_tracer;
3831
3832         if (!debugfs_initialized())
3833                 return NULL;
3834
3835         d_tracer = debugfs_create_dir("tracing", NULL);
3836
3837         if (!d_tracer && !once) {
3838                 once = 1;
3839                 pr_warning("Could not create debugfs directory 'tracing'\n");
3840                 return NULL;
3841         }
3842
3843         return d_tracer;
3844 }
3845
3846 static struct dentry *d_percpu;
3847
3848 struct dentry *tracing_dentry_percpu(void)
3849 {
3850         static int once;
3851         struct dentry *d_tracer;
3852
3853         if (d_percpu)
3854                 return d_percpu;
3855
3856         d_tracer = tracing_init_dentry();
3857
3858         if (!d_tracer)
3859                 return NULL;
3860
3861         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3862
3863         if (!d_percpu && !once) {
3864                 once = 1;
3865                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3866                 return NULL;
3867         }
3868
3869         return d_percpu;
3870 }
3871
3872 static void tracing_init_debugfs_percpu(long cpu)
3873 {
3874         struct dentry *d_percpu = tracing_dentry_percpu();
3875         struct dentry *d_cpu;
3876         /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3877         char cpu_dir[7];
3878
3879         if (cpu > 999 || cpu < 0)
3880                 return;
3881
3882         sprintf(cpu_dir, "cpu%ld", cpu);
3883         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3884         if (!d_cpu) {
3885                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3886                 return;
3887         }
3888
3889         /* per cpu trace_pipe */
3890         trace_create_file("trace_pipe", 0444, d_cpu,
3891                         (void *) cpu, &tracing_pipe_fops);
3892
3893         /* per cpu trace */
3894         trace_create_file("trace", 0644, d_cpu,
3895                         (void *) cpu, &tracing_fops);
3896
3897         trace_create_file("trace_pipe_raw", 0444, d_cpu,
3898                         (void *) cpu, &tracing_buffers_fops);
3899
3900         trace_create_file("stats", 0444, d_cpu,
3901                         (void *) cpu, &tracing_stats_fops);
3902 }
3903
3904 #ifdef CONFIG_FTRACE_SELFTEST
3905 /* Let selftest have access to static functions in this file */
3906 #include "trace_selftest.c"
3907 #endif
3908
3909 struct trace_option_dentry {
3910         struct tracer_opt               *opt;
3911         struct tracer_flags             *flags;
3912         struct dentry                   *entry;
3913 };
3914
3915 static ssize_t
3916 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3917                         loff_t *ppos)
3918 {
3919         struct trace_option_dentry *topt = filp->private_data;
3920         char *buf;
3921
3922         if (topt->flags->val & topt->opt->bit)
3923                 buf = "1\n";
3924         else
3925                 buf = "0\n";
3926
3927         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3928 }
3929
3930 static ssize_t
3931 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3932                          loff_t *ppos)
3933 {
3934         struct trace_option_dentry *topt = filp->private_data;
3935         unsigned long val;
3936         char buf[64];
3937         int ret;
3938
3939         if (cnt >= sizeof(buf))
3940                 return -EINVAL;
3941
3942         if (copy_from_user(&buf, ubuf, cnt))
3943                 return -EFAULT;
3944
3945         buf[cnt] = 0;
3946
3947         ret = strict_strtoul(buf, 10, &val);
3948         if (ret < 0)
3949                 return ret;
3950
3951         ret = 0;
3952         switch (val) {
3953         case 0:
3954                 /* do nothing if already cleared */
3955                 if (!(topt->flags->val & topt->opt->bit))
3956                         break;
3957
3958                 mutex_lock(&trace_types_lock);
3959                 if (current_trace->set_flag)
3960                         ret = current_trace->set_flag(topt->flags->val,
3961                                                       topt->opt->bit, 0);
3962                 mutex_unlock(&trace_types_lock);
3963                 if (ret)
3964                         return ret;
3965                 topt->flags->val &= ~topt->opt->bit;
3966                 break;
3967         case 1:
3968                 /* do nothing if already set */
3969                 if (topt->flags->val & topt->opt->bit)
3970                         break;
3971
3972                 mutex_lock(&trace_types_lock);
3973                 if (current_trace->set_flag)
3974                         ret = current_trace->set_flag(topt->flags->val,
3975                                                       topt->opt->bit, 1);
3976                 mutex_unlock(&trace_types_lock);
3977                 if (ret)
3978                         return ret;
3979                 topt->flags->val |= topt->opt->bit;
3980                 break;
3981
3982         default:
3983                 return -EINVAL;
3984         }
3985
3986         *ppos += cnt;
3987
3988         return cnt;
3989 }
3990
3991
3992 static const struct file_operations trace_options_fops = {
3993         .open = tracing_open_generic,
3994         .read = trace_options_read,
3995         .write = trace_options_write,
3996 };
3997
3998 static ssize_t
3999 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4000                         loff_t *ppos)
4001 {
4002         long index = (long)filp->private_data;
4003         char *buf;
4004
4005         if (trace_flags & (1 << index))
4006                 buf = "1\n";
4007         else
4008                 buf = "0\n";
4009
4010         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4011 }
4012
4013 static ssize_t
4014 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4015                          loff_t *ppos)
4016 {
4017         long index = (long)filp->private_data;
4018         char buf[64];
4019         unsigned long val;
4020         int ret;
4021
4022         if (cnt >= sizeof(buf))
4023                 return -EINVAL;
4024
4025         if (copy_from_user(&buf, ubuf, cnt))
4026                 return -EFAULT;
4027
4028         buf[cnt] = 0;
4029
4030         ret = strict_strtoul(buf, 10, &val);
4031         if (ret < 0)
4032                 return ret;
4033
4034         if (val != 0 && val != 1)
4035                 return -EINVAL;
4036         set_tracer_flags(1 << index, val);
4037
4038         *ppos += cnt;
4039
4040         return cnt;
4041 }
4042
4043 static const struct file_operations trace_options_core_fops = {
4044         .open = tracing_open_generic,
4045         .read = trace_options_core_read,
4046         .write = trace_options_core_write,
4047 };
4048
4049 struct dentry *trace_create_file(const char *name,
4050                                  mode_t mode,
4051                                  struct dentry *parent,
4052                                  void *data,
4053                                  const struct file_operations *fops)
4054 {
4055         struct dentry *ret;
4056
4057         ret = debugfs_create_file(name, mode, parent, data, fops);
4058         if (!ret)
4059                 pr_warning("Could not create debugfs '%s' entry\n", name);
4060
4061         return ret;
4062 }
4063
4064
4065 static struct dentry *trace_options_init_dentry(void)
4066 {
4067         struct dentry *d_tracer;
4068         static struct dentry *t_options;
4069
4070         if (t_options)
4071                 return t_options;
4072
4073         d_tracer = tracing_init_dentry();
4074         if (!d_tracer)
4075                 return NULL;
4076
4077         t_options = debugfs_create_dir("options", d_tracer);
4078         if (!t_options) {
4079                 pr_warning("Could not create debugfs directory 'options'\n");
4080                 return NULL;
4081         }
4082
4083         return t_options;
4084 }
4085
4086 static void
4087 create_trace_option_file(struct trace_option_dentry *topt,
4088                          struct tracer_flags *flags,
4089                          struct tracer_opt *opt)
4090 {
4091         struct dentry *t_options;
4092
4093         t_options = trace_options_init_dentry();
4094         if (!t_options)
4095                 return;
4096
4097         topt->flags = flags;
4098         topt->opt = opt;
4099
4100         topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4101                                     &trace_options_fops);
4102
4103 }
4104
4105 static struct trace_option_dentry *
4106 create_trace_option_files(struct tracer *tracer)
4107 {
4108         struct trace_option_dentry *topts;
4109         struct tracer_flags *flags;
4110         struct tracer_opt *opts;
4111         int cnt;
4112
4113         if (!tracer)
4114                 return NULL;
4115
4116         flags = tracer->flags;
4117
4118         if (!flags || !flags->opts)
4119                 return NULL;
4120
4121         opts = flags->opts;
4122
4123         for (cnt = 0; opts[cnt].name; cnt++)
4124                 ;
4125
4126         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4127         if (!topts)
4128                 return NULL;
4129
4130         for (cnt = 0; opts[cnt].name; cnt++)
4131                 create_trace_option_file(&topts[cnt], flags,
4132                                          &opts[cnt]);
4133
4134         return topts;
4135 }
4136
4137 static void
4138 destroy_trace_option_files(struct trace_option_dentry *topts)
4139 {
4140         int cnt;
4141
4142         if (!topts)
4143                 return;
4144
4145         for (cnt = 0; topts[cnt].opt; cnt++) {
4146                 if (topts[cnt].entry)
4147                         debugfs_remove(topts[cnt].entry);
4148         }
4149
4150         kfree(topts);
4151 }
4152
4153 static struct dentry *
4154 create_trace_option_core_file(const char *option, long index)
4155 {
4156         struct dentry *t_options;
4157
4158         t_options = trace_options_init_dentry();
4159         if (!t_options)
4160                 return NULL;
4161
4162         return trace_create_file(option, 0644, t_options, (void *)index,
4163                                     &trace_options_core_fops);
4164 }
4165
4166 static __init void create_trace_options_dir(void)
4167 {
4168         struct dentry *t_options;
4169         int i;
4170
4171         t_options = trace_options_init_dentry();
4172         if (!t_options)
4173                 return;
4174
4175         for (i = 0; trace_options[i]; i++)
4176                 create_trace_option_core_file(trace_options[i], i);
4177 }
4178
4179 static __init int tracer_init_debugfs(void)
4180 {
4181         struct dentry *d_tracer;
4182         int cpu;
4183
4184         d_tracer = tracing_init_dentry();
4185
4186         trace_create_file("tracing_enabled", 0644, d_tracer,
4187                         &global_trace, &tracing_ctrl_fops);
4188
4189         trace_create_file("trace_options", 0644, d_tracer,
4190                         NULL, &tracing_iter_fops);
4191
4192         trace_create_file("tracing_cpumask", 0644, d_tracer,
4193                         NULL, &tracing_cpumask_fops);
4194
4195         trace_create_file("trace", 0644, d_tracer,
4196                         (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4197
4198         trace_create_file("available_tracers", 0444, d_tracer,
4199                         &global_trace, &show_traces_fops);
4200
4201         trace_create_file("current_tracer", 0644, d_tracer,
4202                         &global_trace, &set_tracer_fops);
4203
4204 #ifdef CONFIG_TRACER_MAX_TRACE
4205         trace_create_file("tracing_max_latency", 0644, d_tracer,
4206                         &tracing_max_latency, &tracing_max_lat_fops);
4207
4208         trace_create_file("tracing_thresh", 0644, d_tracer,
4209                         &tracing_thresh, &tracing_max_lat_fops);
4210 #endif
4211
4212         trace_create_file("README", 0444, d_tracer,
4213                         NULL, &tracing_readme_fops);
4214
4215         trace_create_file("trace_pipe", 0444, d_tracer,
4216                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4217
4218         trace_create_file("buffer_size_kb", 0644, d_tracer,
4219                         &global_trace, &tracing_entries_fops);
4220
4221         trace_create_file("trace_marker", 0220, d_tracer,
4222                         NULL, &tracing_mark_fops);
4223
4224         trace_create_file("saved_cmdlines", 0444, d_tracer,
4225                         NULL, &tracing_saved_cmdlines_fops);
4226
4227         trace_create_file("trace_clock", 0644, d_tracer, NULL,
4228                           &trace_clock_fops);
4229
4230 #ifdef CONFIG_DYNAMIC_FTRACE
4231         trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4232                         &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4233 #endif
4234 #ifdef CONFIG_SYSPROF_TRACER
4235         init_tracer_sysprof_debugfs(d_tracer);
4236 #endif
4237
4238         create_trace_options_dir();
4239
4240         for_each_tracing_cpu(cpu)
4241                 tracing_init_debugfs_percpu(cpu);
4242
4243         return 0;
4244 }
4245
4246 static int trace_panic_handler(struct notifier_block *this,
4247                                unsigned long event, void *unused)
4248 {
4249         if (ftrace_dump_on_oops)
4250                 ftrace_dump();
4251         return NOTIFY_OK;
4252 }
4253
4254 static struct notifier_block trace_panic_notifier = {
4255         .notifier_call  = trace_panic_handler,
4256         .next           = NULL,
4257         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
4258 };
4259
4260 static int trace_die_handler(struct notifier_block *self,
4261                              unsigned long val,
4262                              void *data)
4263 {
4264         switch (val) {
4265         case DIE_OOPS:
4266                 if (ftrace_dump_on_oops)
4267                         ftrace_dump();
4268                 break;
4269         default:
4270                 break;
4271         }
4272         return NOTIFY_OK;
4273 }
4274
4275 static struct notifier_block trace_die_notifier = {
4276         .notifier_call = trace_die_handler,
4277         .priority = 200
4278 };
4279
4280 /*
4281  * printk is set to max of 1024, we really don't need it that big.
4282  * Nothing should be printing 1000 characters anyway.
4283  */
4284 #define TRACE_MAX_PRINT         1000
4285
4286 /*
4287  * Define here KERN_TRACE so that we have one place to modify
4288  * it if we decide to change what log level the ftrace dump
4289  * should be at.
4290  */
4291 #define KERN_TRACE              KERN_EMERG
4292
4293 static void
4294 trace_printk_seq(struct trace_seq *s)
4295 {
4296         /* Probably should print a warning here. */
4297         if (s->len >= 1000)
4298                 s->len = 1000;
4299
4300         /* should be zero ended, but we are paranoid. */
4301         s->buffer[s->len] = 0;
4302
4303         printk(KERN_TRACE "%s", s->buffer);
4304
4305         trace_seq_init(s);
4306 }
4307
4308 static void __ftrace_dump(bool disable_tracing)
4309 {
4310         static arch_spinlock_t ftrace_dump_lock =
4311                 (arch_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
4312         /* use static because iter can be a bit big for the stack */
4313         static struct trace_iterator iter;
4314         unsigned int old_userobj;
4315         static int dump_ran;
4316         unsigned long flags;
4317         int cnt = 0, cpu;
4318
4319         /* only one dump */
4320         local_irq_save(flags);
4321         __raw_spin_lock(&ftrace_dump_lock);
4322         if (dump_ran)
4323                 goto out;
4324
4325         dump_ran = 1;
4326
4327         tracing_off();
4328
4329         if (disable_tracing)
4330                 ftrace_kill();
4331
4332         for_each_tracing_cpu(cpu) {
4333                 atomic_inc(&global_trace.data[cpu]->disabled);
4334         }
4335
4336         old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4337
4338         /* don't look at user memory in panic mode */
4339         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4340
4341         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4342
4343         /* Simulate the iterator */
4344         iter.tr = &global_trace;
4345         iter.trace = current_trace;
4346         iter.cpu_file = TRACE_PIPE_ALL_CPU;
4347
4348         /*
4349          * We need to stop all tracing on all CPUS to read the
4350          * the next buffer. This is a bit expensive, but is
4351          * not done often. We fill all what we can read,
4352          * and then release the locks again.
4353          */
4354
4355         while (!trace_empty(&iter)) {
4356
4357                 if (!cnt)
4358                         printk(KERN_TRACE "---------------------------------\n");
4359
4360                 cnt++;
4361
4362                 /* reset all but tr, trace, and overruns */
4363                 memset(&iter.seq, 0,
4364                        sizeof(struct trace_iterator) -
4365                        offsetof(struct trace_iterator, seq));
4366                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4367                 iter.pos = -1;
4368
4369                 if (find_next_entry_inc(&iter) != NULL) {
4370                         int ret;
4371
4372                         ret = print_trace_line(&iter);
4373                         if (ret != TRACE_TYPE_NO_CONSUME)
4374                                 trace_consume(&iter);
4375                 }
4376
4377                 trace_printk_seq(&iter.seq);
4378         }
4379
4380         if (!cnt)
4381                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4382         else
4383                 printk(KERN_TRACE "---------------------------------\n");
4384
4385         /* Re-enable tracing if requested */
4386         if (!disable_tracing) {
4387                 trace_flags |= old_userobj;
4388
4389                 for_each_tracing_cpu(cpu) {
4390                         atomic_dec(&global_trace.data[cpu]->disabled);
4391                 }
4392                 tracing_on();
4393         }
4394
4395  out:
4396         __raw_spin_unlock(&ftrace_dump_lock);
4397         local_irq_restore(flags);
4398 }
4399
4400 /* By default: disable tracing after the dump */
4401 void ftrace_dump(void)
4402 {
4403         __ftrace_dump(true);
4404 }
4405
4406 __init static int tracer_alloc_buffers(void)
4407 {
4408         int ring_buf_size;
4409         int i;
4410         int ret = -ENOMEM;
4411
4412         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4413                 goto out;
4414
4415         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4416                 goto out_free_buffer_mask;
4417
4418         if (!zalloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4419                 goto out_free_tracing_cpumask;
4420
4421         /* To save memory, keep the ring buffer size to its minimum */
4422         if (ring_buffer_expanded)
4423                 ring_buf_size = trace_buf_size;
4424         else
4425                 ring_buf_size = 1;
4426
4427         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4428         cpumask_copy(tracing_cpumask, cpu_all_mask);
4429
4430         /* TODO: make the number of buffers hot pluggable with CPUS */
4431         global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4432                                                    TRACE_BUFFER_FLAGS);
4433         if (!global_trace.buffer) {
4434                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4435                 WARN_ON(1);
4436                 goto out_free_cpumask;
4437         }
4438         global_trace.entries = ring_buffer_size(global_trace.buffer);
4439
4440
4441 #ifdef CONFIG_TRACER_MAX_TRACE
4442         max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4443                                              TRACE_BUFFER_FLAGS);
4444         if (!max_tr.buffer) {
4445                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4446                 WARN_ON(1);
4447                 ring_buffer_free(global_trace.buffer);
4448                 goto out_free_cpumask;
4449         }
4450         max_tr.entries = ring_buffer_size(max_tr.buffer);
4451         WARN_ON(max_tr.entries != global_trace.entries);
4452 #endif
4453
4454         /* Allocate the first page for all buffers */
4455         for_each_tracing_cpu(i) {
4456                 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4457                 max_tr.data[i] = &per_cpu(max_tr_data, i);
4458         }
4459
4460         trace_init_cmdlines();
4461
4462         register_tracer(&nop_trace);
4463         current_trace = &nop_trace;
4464 #ifdef CONFIG_BOOT_TRACER
4465         register_tracer(&boot_tracer);
4466 #endif
4467         /* All seems OK, enable tracing */
4468         tracing_disabled = 0;
4469
4470         atomic_notifier_chain_register(&panic_notifier_list,
4471                                        &trace_panic_notifier);
4472
4473         register_die_notifier(&trace_die_notifier);
4474
4475         return 0;
4476
4477 out_free_cpumask:
4478         free_cpumask_var(tracing_reader_cpumask);
4479 out_free_tracing_cpumask:
4480         free_cpumask_var(tracing_cpumask);
4481 out_free_buffer_mask:
4482         free_cpumask_var(tracing_buffer_mask);
4483 out:
4484         return ret;
4485 }
4486
4487 __init static int clear_boot_tracer(void)
4488 {
4489         /*
4490          * The default tracer at boot buffer is an init section.
4491          * This function is called in lateinit. If we did not
4492          * find the boot tracer, then clear it out, to prevent
4493          * later registration from accessing the buffer that is
4494          * about to be freed.
4495          */
4496         if (!default_bootup_tracer)
4497                 return 0;
4498
4499         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4500                default_bootup_tracer);
4501         default_bootup_tracer = NULL;
4502
4503         return 0;
4504 }
4505
4506 early_initcall(tracer_alloc_buffers);
4507 fs_initcall(tracer_init_debugfs);
4508 late_initcall(clear_boot_tracer);