c5e39cd7310db2a371a6ff1c872b4bf473750b12
[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/utsrelease.h>
15 #include <linux/kallsyms.h>
16 #include <linux/seq_file.h>
17 #include <linux/notifier.h>
18 #include <linux/debugfs.h>
19 #include <linux/pagemap.h>
20 #include <linux/hardirq.h>
21 #include <linux/linkage.h>
22 #include <linux/uaccess.h>
23 #include <linux/ftrace.h>
24 #include <linux/module.h>
25 #include <linux/percpu.h>
26 #include <linux/kdebug.h>
27 #include <linux/ctype.h>
28 #include <linux/init.h>
29 #include <linux/poll.h>
30 #include <linux/gfp.h>
31 #include <linux/fs.h>
32 #include <linux/kprobes.h>
33 #include <linux/writeback.h>
34 #include <linux/splice.h>
35
36 #include <linux/stacktrace.h>
37 #include <linux/ring_buffer.h>
38 #include <linux/irqflags.h>
39
40 #include "trace.h"
41 #include "trace_output.h"
42
43 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
44
45 unsigned long __read_mostly     tracing_max_latency;
46 unsigned long __read_mostly     tracing_thresh;
47
48 /*
49  * We need to change this state when a selftest is running.
50  * A selftest will lurk into the ring-buffer to count the
51  * entries inserted during the selftest although some concurrent
52  * insertions into the ring-buffer such as ftrace_printk could occurred
53  * at the same time, giving false positive or negative results.
54  */
55 static bool __read_mostly tracing_selftest_running;
56
57 /*
58  * If a tracer is running, we do not want to run SELFTEST.
59  */
60 static bool __read_mostly tracing_selftest_disabled;
61
62 /* For tracers that don't implement custom flags */
63 static struct tracer_opt dummy_tracer_opt[] = {
64         { }
65 };
66
67 static struct tracer_flags dummy_tracer_flags = {
68         .val = 0,
69         .opts = dummy_tracer_opt
70 };
71
72 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
73 {
74         return 0;
75 }
76
77 /*
78  * Kill all tracing for good (never come back).
79  * It is initialized to 1 but will turn to zero if the initialization
80  * of the tracer is successful. But that is the only place that sets
81  * this back to zero.
82  */
83 static int tracing_disabled = 1;
84
85 static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
86
87 static inline void ftrace_disable_cpu(void)
88 {
89         preempt_disable();
90         local_inc(&__get_cpu_var(ftrace_cpu_disabled));
91 }
92
93 static inline void ftrace_enable_cpu(void)
94 {
95         local_dec(&__get_cpu_var(ftrace_cpu_disabled));
96         preempt_enable();
97 }
98
99 static cpumask_var_t __read_mostly      tracing_buffer_mask;
100
101 /* Define which cpu buffers are currently read in trace_pipe */
102 static cpumask_var_t                    tracing_reader_cpumask;
103
104 #define for_each_tracing_cpu(cpu)       \
105         for_each_cpu(cpu, tracing_buffer_mask)
106
107 /*
108  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
109  *
110  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
111  * is set, then ftrace_dump is called. This will output the contents
112  * of the ftrace buffers to the console.  This is very useful for
113  * capturing traces that lead to crashes and outputing it to a
114  * serial console.
115  *
116  * It is default off, but you can enable it with either specifying
117  * "ftrace_dump_on_oops" in the kernel command line, or setting
118  * /proc/sys/kernel/ftrace_dump_on_oops to true.
119  */
120 int ftrace_dump_on_oops;
121
122 static int tracing_set_tracer(const char *buf);
123
124 #define BOOTUP_TRACER_SIZE              100
125 static char bootup_tracer_buf[BOOTUP_TRACER_SIZE] __initdata;
126 static char *default_bootup_tracer;
127
128 static int __init set_ftrace(char *str)
129 {
130         strncpy(bootup_tracer_buf, str, BOOTUP_TRACER_SIZE);
131         default_bootup_tracer = bootup_tracer_buf;
132         return 1;
133 }
134 __setup("ftrace=", set_ftrace);
135
136 static int __init set_ftrace_dump_on_oops(char *str)
137 {
138         ftrace_dump_on_oops = 1;
139         return 1;
140 }
141 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
142
143 long
144 ns2usecs(cycle_t nsec)
145 {
146         nsec += 500;
147         do_div(nsec, 1000);
148         return nsec;
149 }
150
151 cycle_t ftrace_now(int cpu)
152 {
153         u64 ts = ring_buffer_time_stamp(cpu);
154         ring_buffer_normalize_time_stamp(cpu, &ts);
155         return ts;
156 }
157
158 /*
159  * The global_trace is the descriptor that holds the tracing
160  * buffers for the live tracing. For each CPU, it contains
161  * a link list of pages that will store trace entries. The
162  * page descriptor of the pages in the memory is used to hold
163  * the link list by linking the lru item in the page descriptor
164  * to each of the pages in the buffer per CPU.
165  *
166  * For each active CPU there is a data field that holds the
167  * pages for the buffer for that CPU. Each CPU has the same number
168  * of pages allocated for its buffer.
169  */
170 static struct trace_array       global_trace;
171
172 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
173
174 /*
175  * The max_tr is used to snapshot the global_trace when a maximum
176  * latency is reached. Some tracers will use this to store a maximum
177  * trace while it continues examining live traces.
178  *
179  * The buffers for the max_tr are set up the same as the global_trace.
180  * When a snapshot is taken, the link list of the max_tr is swapped
181  * with the link list of the global_trace and the buffers are reset for
182  * the global_trace so the tracing can continue.
183  */
184 static struct trace_array       max_tr;
185
186 static DEFINE_PER_CPU(struct trace_array_cpu, max_data);
187
188 /* tracer_enabled is used to toggle activation of a tracer */
189 static int                      tracer_enabled = 1;
190
191 /**
192  * tracing_is_enabled - return tracer_enabled status
193  *
194  * This function is used by other tracers to know the status
195  * of the tracer_enabled flag.  Tracers may use this function
196  * to know if it should enable their features when starting
197  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
198  */
199 int tracing_is_enabled(void)
200 {
201         return tracer_enabled;
202 }
203
204 /*
205  * trace_buf_size is the size in bytes that is allocated
206  * for a buffer. Note, the number of bytes is always rounded
207  * to page size.
208  *
209  * This number is purposely set to a low number of 16384.
210  * If the dump on oops happens, it will be much appreciated
211  * to not have to wait for all that output. Anyway this can be
212  * boot time and run time configurable.
213  */
214 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
215
216 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
217
218 /* trace_types holds a link list of available tracers. */
219 static struct tracer            *trace_types __read_mostly;
220
221 /* current_trace points to the tracer that is currently active */
222 static struct tracer            *current_trace __read_mostly;
223
224 /*
225  * max_tracer_type_len is used to simplify the allocating of
226  * buffers to read userspace tracer names. We keep track of
227  * the longest tracer name registered.
228  */
229 static int                      max_tracer_type_len;
230
231 /*
232  * trace_types_lock is used to protect the trace_types list.
233  * This lock is also used to keep user access serialized.
234  * Accesses from userspace will grab this lock while userspace
235  * activities happen inside the kernel.
236  */
237 static DEFINE_MUTEX(trace_types_lock);
238
239 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
240 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
241
242 /* trace_flags holds trace_options default values */
243 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
244         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO;
245
246 /**
247  * trace_wake_up - wake up tasks waiting for trace input
248  *
249  * Simply wakes up any task that is blocked on the trace_wait
250  * queue. These is used with trace_poll for tasks polling the trace.
251  */
252 void trace_wake_up(void)
253 {
254         /*
255          * The runqueue_is_locked() can fail, but this is the best we
256          * have for now:
257          */
258         if (!(trace_flags & TRACE_ITER_BLOCK) && !runqueue_is_locked())
259                 wake_up(&trace_wait);
260 }
261
262 static int __init set_buf_size(char *str)
263 {
264         unsigned long buf_size;
265         int ret;
266
267         if (!str)
268                 return 0;
269         ret = strict_strtoul(str, 0, &buf_size);
270         /* nr_entries can not be zero */
271         if (ret < 0 || buf_size == 0)
272                 return 0;
273         trace_buf_size = buf_size;
274         return 1;
275 }
276 __setup("trace_buf_size=", set_buf_size);
277
278 unsigned long nsecs_to_usecs(unsigned long nsecs)
279 {
280         return nsecs / 1000;
281 }
282
283 /* These must match the bit postions in trace_iterator_flags */
284 static const char *trace_options[] = {
285         "print-parent",
286         "sym-offset",
287         "sym-addr",
288         "verbose",
289         "raw",
290         "hex",
291         "bin",
292         "block",
293         "stacktrace",
294         "sched-tree",
295         "ftrace_printk",
296         "ftrace_preempt",
297         "branch",
298         "annotate",
299         "userstacktrace",
300         "sym-userobj",
301         "printk-msg-only",
302         "context-info",
303         NULL
304 };
305
306 /*
307  * ftrace_max_lock is used to protect the swapping of buffers
308  * when taking a max snapshot. The buffers themselves are
309  * protected by per_cpu spinlocks. But the action of the swap
310  * needs its own lock.
311  *
312  * This is defined as a raw_spinlock_t in order to help
313  * with performance when lockdep debugging is enabled.
314  */
315 static raw_spinlock_t ftrace_max_lock =
316         (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
317
318 /*
319  * Copy the new maximum trace into the separate maximum-trace
320  * structure. (this way the maximum trace is permanently saved,
321  * for later retrieval via /debugfs/tracing/latency_trace)
322  */
323 static void
324 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
325 {
326         struct trace_array_cpu *data = tr->data[cpu];
327
328         max_tr.cpu = cpu;
329         max_tr.time_start = data->preempt_timestamp;
330
331         data = max_tr.data[cpu];
332         data->saved_latency = tracing_max_latency;
333
334         memcpy(data->comm, tsk->comm, TASK_COMM_LEN);
335         data->pid = tsk->pid;
336         data->uid = task_uid(tsk);
337         data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
338         data->policy = tsk->policy;
339         data->rt_priority = tsk->rt_priority;
340
341         /* record this tasks comm */
342         tracing_record_cmdline(tsk);
343 }
344
345 static void
346 trace_seq_reset(struct trace_seq *s)
347 {
348         s->len = 0;
349         s->readpos = 0;
350 }
351
352 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
353 {
354         int len;
355         int ret;
356
357         if (s->len <= s->readpos)
358                 return -EBUSY;
359
360         len = s->len - s->readpos;
361         if (cnt > len)
362                 cnt = len;
363         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
364         if (ret)
365                 return -EFAULT;
366
367         s->readpos += len;
368         return cnt;
369 }
370
371 ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
372 {
373         int len;
374         void *ret;
375
376         if (s->len <= s->readpos)
377                 return -EBUSY;
378
379         len = s->len - s->readpos;
380         if (cnt > len)
381                 cnt = len;
382         ret = memcpy(buf, s->buffer + s->readpos, cnt);
383         if (!ret)
384                 return -EFAULT;
385
386         s->readpos += len;
387         return cnt;
388 }
389
390 static void
391 trace_print_seq(struct seq_file *m, struct trace_seq *s)
392 {
393         int len = s->len >= PAGE_SIZE ? PAGE_SIZE - 1 : s->len;
394
395         s->buffer[len] = 0;
396         seq_puts(m, s->buffer);
397
398         trace_seq_reset(s);
399 }
400
401 /**
402  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
403  * @tr: tracer
404  * @tsk: the task with the latency
405  * @cpu: The cpu that initiated the trace.
406  *
407  * Flip the buffers between the @tr and the max_tr and record information
408  * about which task was the cause of this latency.
409  */
410 void
411 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
412 {
413         struct ring_buffer *buf = tr->buffer;
414
415         WARN_ON_ONCE(!irqs_disabled());
416         __raw_spin_lock(&ftrace_max_lock);
417
418         tr->buffer = max_tr.buffer;
419         max_tr.buffer = buf;
420
421         ftrace_disable_cpu();
422         ring_buffer_reset(tr->buffer);
423         ftrace_enable_cpu();
424
425         __update_max_tr(tr, tsk, cpu);
426         __raw_spin_unlock(&ftrace_max_lock);
427 }
428
429 /**
430  * update_max_tr_single - only copy one trace over, and reset the rest
431  * @tr - tracer
432  * @tsk - task with the latency
433  * @cpu - the cpu of the buffer to copy.
434  *
435  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
436  */
437 void
438 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
439 {
440         int ret;
441
442         WARN_ON_ONCE(!irqs_disabled());
443         __raw_spin_lock(&ftrace_max_lock);
444
445         ftrace_disable_cpu();
446
447         ring_buffer_reset(max_tr.buffer);
448         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
449
450         ftrace_enable_cpu();
451
452         WARN_ON_ONCE(ret && ret != -EAGAIN);
453
454         __update_max_tr(tr, tsk, cpu);
455         __raw_spin_unlock(&ftrace_max_lock);
456 }
457
458 /**
459  * register_tracer - register a tracer with the ftrace system.
460  * @type - the plugin for the tracer
461  *
462  * Register a new plugin tracer.
463  */
464 int register_tracer(struct tracer *type)
465 __releases(kernel_lock)
466 __acquires(kernel_lock)
467 {
468         struct tracer *t;
469         int len;
470         int ret = 0;
471
472         if (!type->name) {
473                 pr_info("Tracer must have a name\n");
474                 return -1;
475         }
476
477         /*
478          * When this gets called we hold the BKL which means that
479          * preemption is disabled. Various trace selftests however
480          * need to disable and enable preemption for successful tests.
481          * So we drop the BKL here and grab it after the tests again.
482          */
483         unlock_kernel();
484         mutex_lock(&trace_types_lock);
485
486         tracing_selftest_running = true;
487
488         for (t = trace_types; t; t = t->next) {
489                 if (strcmp(type->name, t->name) == 0) {
490                         /* already found */
491                         pr_info("Trace %s already registered\n",
492                                 type->name);
493                         ret = -1;
494                         goto out;
495                 }
496         }
497
498         if (!type->set_flag)
499                 type->set_flag = &dummy_set_flag;
500         if (!type->flags)
501                 type->flags = &dummy_tracer_flags;
502         else
503                 if (!type->flags->opts)
504                         type->flags->opts = dummy_tracer_opt;
505         if (!type->wait_pipe)
506                 type->wait_pipe = default_wait_pipe;
507
508
509 #ifdef CONFIG_FTRACE_STARTUP_TEST
510         if (type->selftest && !tracing_selftest_disabled) {
511                 struct tracer *saved_tracer = current_trace;
512                 struct trace_array *tr = &global_trace;
513                 int i;
514
515                 /*
516                  * Run a selftest on this tracer.
517                  * Here we reset the trace buffer, and set the current
518                  * tracer to be this tracer. The tracer can then run some
519                  * internal tracing to verify that everything is in order.
520                  * If we fail, we do not register this tracer.
521                  */
522                 for_each_tracing_cpu(i)
523                         tracing_reset(tr, i);
524
525                 current_trace = type;
526                 /* the test is responsible for initializing and enabling */
527                 pr_info("Testing tracer %s: ", type->name);
528                 ret = type->selftest(type, tr);
529                 /* the test is responsible for resetting too */
530                 current_trace = saved_tracer;
531                 if (ret) {
532                         printk(KERN_CONT "FAILED!\n");
533                         goto out;
534                 }
535                 /* Only reset on passing, to avoid touching corrupted buffers */
536                 for_each_tracing_cpu(i)
537                         tracing_reset(tr, i);
538
539                 printk(KERN_CONT "PASSED\n");
540         }
541 #endif
542
543         type->next = trace_types;
544         trace_types = type;
545         len = strlen(type->name);
546         if (len > max_tracer_type_len)
547                 max_tracer_type_len = len;
548
549  out:
550         tracing_selftest_running = false;
551         mutex_unlock(&trace_types_lock);
552
553         if (ret || !default_bootup_tracer)
554                 goto out_unlock;
555
556         if (strncmp(default_bootup_tracer, type->name, BOOTUP_TRACER_SIZE))
557                 goto out_unlock;
558
559         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
560         /* Do we want this tracer to start on bootup? */
561         tracing_set_tracer(type->name);
562         default_bootup_tracer = NULL;
563         /* disable other selftests, since this will break it. */
564         tracing_selftest_disabled = 1;
565 #ifdef CONFIG_FTRACE_STARTUP_TEST
566         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
567                type->name);
568 #endif
569
570  out_unlock:
571         lock_kernel();
572         return ret;
573 }
574
575 void unregister_tracer(struct tracer *type)
576 {
577         struct tracer **t;
578         int len;
579
580         mutex_lock(&trace_types_lock);
581         for (t = &trace_types; *t; t = &(*t)->next) {
582                 if (*t == type)
583                         goto found;
584         }
585         pr_info("Trace %s not registered\n", type->name);
586         goto out;
587
588  found:
589         *t = (*t)->next;
590
591         if (type == current_trace && tracer_enabled) {
592                 tracer_enabled = 0;
593                 tracing_stop();
594                 if (current_trace->stop)
595                         current_trace->stop(&global_trace);
596                 current_trace = &nop_trace;
597         }
598
599         if (strlen(type->name) != max_tracer_type_len)
600                 goto out;
601
602         max_tracer_type_len = 0;
603         for (t = &trace_types; *t; t = &(*t)->next) {
604                 len = strlen((*t)->name);
605                 if (len > max_tracer_type_len)
606                         max_tracer_type_len = len;
607         }
608  out:
609         mutex_unlock(&trace_types_lock);
610 }
611
612 void tracing_reset(struct trace_array *tr, int cpu)
613 {
614         ftrace_disable_cpu();
615         ring_buffer_reset_cpu(tr->buffer, cpu);
616         ftrace_enable_cpu();
617 }
618
619 void tracing_reset_online_cpus(struct trace_array *tr)
620 {
621         int cpu;
622
623         tr->time_start = ftrace_now(tr->cpu);
624
625         for_each_online_cpu(cpu)
626                 tracing_reset(tr, cpu);
627 }
628
629 #define SAVED_CMDLINES 128
630 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
631 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
632 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
633 static int cmdline_idx;
634 static DEFINE_SPINLOCK(trace_cmdline_lock);
635
636 /* temporary disable recording */
637 static atomic_t trace_record_cmdline_disabled __read_mostly;
638
639 static void trace_init_cmdlines(void)
640 {
641         memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline));
642         memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid));
643         cmdline_idx = 0;
644 }
645
646 static int trace_stop_count;
647 static DEFINE_SPINLOCK(tracing_start_lock);
648
649 /**
650  * ftrace_off_permanent - disable all ftrace code permanently
651  *
652  * This should only be called when a serious anomally has
653  * been detected.  This will turn off the function tracing,
654  * ring buffers, and other tracing utilites. It takes no
655  * locks and can be called from any context.
656  */
657 void ftrace_off_permanent(void)
658 {
659         tracing_disabled = 1;
660         ftrace_stop();
661         tracing_off_permanent();
662 }
663
664 /**
665  * tracing_start - quick start of the tracer
666  *
667  * If tracing is enabled but was stopped by tracing_stop,
668  * this will start the tracer back up.
669  */
670 void tracing_start(void)
671 {
672         struct ring_buffer *buffer;
673         unsigned long flags;
674
675         if (tracing_disabled)
676                 return;
677
678         spin_lock_irqsave(&tracing_start_lock, flags);
679         if (--trace_stop_count) {
680                 if (trace_stop_count < 0) {
681                         /* Someone screwed up their debugging */
682                         WARN_ON_ONCE(1);
683                         trace_stop_count = 0;
684                 }
685                 goto out;
686         }
687
688
689         buffer = global_trace.buffer;
690         if (buffer)
691                 ring_buffer_record_enable(buffer);
692
693         buffer = max_tr.buffer;
694         if (buffer)
695                 ring_buffer_record_enable(buffer);
696
697         ftrace_start();
698  out:
699         spin_unlock_irqrestore(&tracing_start_lock, flags);
700 }
701
702 /**
703  * tracing_stop - quick stop of the tracer
704  *
705  * Light weight way to stop tracing. Use in conjunction with
706  * tracing_start.
707  */
708 void tracing_stop(void)
709 {
710         struct ring_buffer *buffer;
711         unsigned long flags;
712
713         ftrace_stop();
714         spin_lock_irqsave(&tracing_start_lock, flags);
715         if (trace_stop_count++)
716                 goto out;
717
718         buffer = global_trace.buffer;
719         if (buffer)
720                 ring_buffer_record_disable(buffer);
721
722         buffer = max_tr.buffer;
723         if (buffer)
724                 ring_buffer_record_disable(buffer);
725
726  out:
727         spin_unlock_irqrestore(&tracing_start_lock, flags);
728 }
729
730 void trace_stop_cmdline_recording(void);
731
732 static void trace_save_cmdline(struct task_struct *tsk)
733 {
734         unsigned map;
735         unsigned idx;
736
737         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
738                 return;
739
740         /*
741          * It's not the end of the world if we don't get
742          * the lock, but we also don't want to spin
743          * nor do we want to disable interrupts,
744          * so if we miss here, then better luck next time.
745          */
746         if (!spin_trylock(&trace_cmdline_lock))
747                 return;
748
749         idx = map_pid_to_cmdline[tsk->pid];
750         if (idx >= SAVED_CMDLINES) {
751                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
752
753                 map = map_cmdline_to_pid[idx];
754                 if (map <= PID_MAX_DEFAULT)
755                         map_pid_to_cmdline[map] = (unsigned)-1;
756
757                 map_pid_to_cmdline[tsk->pid] = idx;
758
759                 cmdline_idx = idx;
760         }
761
762         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
763
764         spin_unlock(&trace_cmdline_lock);
765 }
766
767 char *trace_find_cmdline(int pid)
768 {
769         char *cmdline = "<...>";
770         unsigned map;
771
772         if (!pid)
773                 return "<idle>";
774
775         if (pid > PID_MAX_DEFAULT)
776                 goto out;
777
778         map = map_pid_to_cmdline[pid];
779         if (map >= SAVED_CMDLINES)
780                 goto out;
781
782         cmdline = saved_cmdlines[map];
783
784  out:
785         return cmdline;
786 }
787
788 void tracing_record_cmdline(struct task_struct *tsk)
789 {
790         if (atomic_read(&trace_record_cmdline_disabled))
791                 return;
792
793         trace_save_cmdline(tsk);
794 }
795
796 void
797 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
798                              int pc)
799 {
800         struct task_struct *tsk = current;
801
802         entry->preempt_count            = pc & 0xff;
803         entry->pid                      = (tsk) ? tsk->pid : 0;
804         entry->tgid                     = (tsk) ? tsk->tgid : 0;
805         entry->flags =
806 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
807                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
808 #else
809                 TRACE_FLAG_IRQS_NOSUPPORT |
810 #endif
811                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
812                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
813                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
814 }
815
816 struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
817                                                     unsigned char type,
818                                                     unsigned long len,
819                                                     unsigned long flags, int pc)
820 {
821         struct ring_buffer_event *event;
822
823         event = ring_buffer_lock_reserve(tr->buffer, len);
824         if (event != NULL) {
825                 struct trace_entry *ent = ring_buffer_event_data(event);
826
827                 tracing_generic_entry_update(ent, flags, pc);
828                 ent->type = type;
829         }
830
831         return event;
832 }
833 static void ftrace_trace_stack(struct trace_array *tr,
834                                unsigned long flags, int skip, int pc);
835 static void ftrace_trace_userstack(struct trace_array *tr,
836                                    unsigned long flags, int pc);
837
838 void trace_buffer_unlock_commit(struct trace_array *tr,
839                                 struct ring_buffer_event *event,
840                                 unsigned long flags, int pc)
841 {
842         ring_buffer_unlock_commit(tr->buffer, event);
843
844         ftrace_trace_stack(tr, flags, 6, pc);
845         ftrace_trace_userstack(tr, flags, pc);
846         trace_wake_up();
847 }
848
849 struct ring_buffer_event *
850 trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
851                                   unsigned long flags, int pc)
852 {
853         return trace_buffer_lock_reserve(&global_trace,
854                                          type, len, flags, pc);
855 }
856
857 void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
858                                         unsigned long flags, int pc)
859 {
860         return trace_buffer_unlock_commit(&global_trace, event, flags, pc);
861 }
862
863 void
864 trace_function(struct trace_array *tr,
865                unsigned long ip, unsigned long parent_ip, unsigned long flags,
866                int pc)
867 {
868         struct ring_buffer_event *event;
869         struct ftrace_entry *entry;
870
871         /* If we are reading the ring buffer, don't trace */
872         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
873                 return;
874
875         event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
876                                           flags, pc);
877         if (!event)
878                 return;
879         entry   = ring_buffer_event_data(event);
880         entry->ip                       = ip;
881         entry->parent_ip                = parent_ip;
882         ring_buffer_unlock_commit(tr->buffer, event);
883 }
884
885 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
886 static void __trace_graph_entry(struct trace_array *tr,
887                                 struct ftrace_graph_ent *trace,
888                                 unsigned long flags,
889                                 int pc)
890 {
891         struct ring_buffer_event *event;
892         struct ftrace_graph_ent_entry *entry;
893
894         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
895                 return;
896
897         event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
898                                           sizeof(*entry), flags, pc);
899         if (!event)
900                 return;
901         entry   = ring_buffer_event_data(event);
902         entry->graph_ent                        = *trace;
903         ring_buffer_unlock_commit(global_trace.buffer, event);
904 }
905
906 static void __trace_graph_return(struct trace_array *tr,
907                                 struct ftrace_graph_ret *trace,
908                                 unsigned long flags,
909                                 int pc)
910 {
911         struct ring_buffer_event *event;
912         struct ftrace_graph_ret_entry *entry;
913
914         if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
915                 return;
916
917         event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
918                                           sizeof(*entry), flags, pc);
919         if (!event)
920                 return;
921         entry   = ring_buffer_event_data(event);
922         entry->ret                              = *trace;
923         ring_buffer_unlock_commit(global_trace.buffer, event);
924 }
925 #endif
926
927 void
928 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
929        unsigned long ip, unsigned long parent_ip, unsigned long flags,
930        int pc)
931 {
932         if (likely(!atomic_read(&data->disabled)))
933                 trace_function(tr, ip, parent_ip, flags, pc);
934 }
935
936 static void __ftrace_trace_stack(struct trace_array *tr,
937                                  unsigned long flags,
938                                  int skip, int pc)
939 {
940 #ifdef CONFIG_STACKTRACE
941         struct ring_buffer_event *event;
942         struct stack_entry *entry;
943         struct stack_trace trace;
944
945         event = trace_buffer_lock_reserve(tr, TRACE_STACK,
946                                           sizeof(*entry), flags, pc);
947         if (!event)
948                 return;
949         entry   = ring_buffer_event_data(event);
950         memset(&entry->caller, 0, sizeof(entry->caller));
951
952         trace.nr_entries        = 0;
953         trace.max_entries       = FTRACE_STACK_ENTRIES;
954         trace.skip              = skip;
955         trace.entries           = entry->caller;
956
957         save_stack_trace(&trace);
958         ring_buffer_unlock_commit(tr->buffer, event);
959 #endif
960 }
961
962 static void ftrace_trace_stack(struct trace_array *tr,
963                                unsigned long flags,
964                                int skip, int pc)
965 {
966         if (!(trace_flags & TRACE_ITER_STACKTRACE))
967                 return;
968
969         __ftrace_trace_stack(tr, flags, skip, pc);
970 }
971
972 void __trace_stack(struct trace_array *tr,
973                    unsigned long flags,
974                    int skip, int pc)
975 {
976         __ftrace_trace_stack(tr, flags, skip, pc);
977 }
978
979 static void ftrace_trace_userstack(struct trace_array *tr,
980                                    unsigned long flags, int pc)
981 {
982 #ifdef CONFIG_STACKTRACE
983         struct ring_buffer_event *event;
984         struct userstack_entry *entry;
985         struct stack_trace trace;
986
987         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
988                 return;
989
990         event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
991                                           sizeof(*entry), flags, pc);
992         if (!event)
993                 return;
994         entry   = ring_buffer_event_data(event);
995
996         memset(&entry->caller, 0, sizeof(entry->caller));
997
998         trace.nr_entries        = 0;
999         trace.max_entries       = FTRACE_STACK_ENTRIES;
1000         trace.skip              = 0;
1001         trace.entries           = entry->caller;
1002
1003         save_stack_trace_user(&trace);
1004         ring_buffer_unlock_commit(tr->buffer, event);
1005 #endif
1006 }
1007
1008 #ifdef UNUSED
1009 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1010 {
1011         ftrace_trace_userstack(tr, flags, preempt_count());
1012 }
1013 #endif /* UNUSED */
1014
1015 static void
1016 ftrace_trace_special(void *__tr,
1017                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1018                      int pc)
1019 {
1020         struct ring_buffer_event *event;
1021         struct trace_array *tr = __tr;
1022         struct special_entry *entry;
1023
1024         event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1025                                           sizeof(*entry), 0, pc);
1026         if (!event)
1027                 return;
1028         entry   = ring_buffer_event_data(event);
1029         entry->arg1                     = arg1;
1030         entry->arg2                     = arg2;
1031         entry->arg3                     = arg3;
1032         trace_buffer_unlock_commit(tr, event, 0, pc);
1033 }
1034
1035 void
1036 __trace_special(void *__tr, void *__data,
1037                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1038 {
1039         ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1040 }
1041
1042 void
1043 tracing_sched_switch_trace(struct trace_array *tr,
1044                            struct task_struct *prev,
1045                            struct task_struct *next,
1046                            unsigned long flags, int pc)
1047 {
1048         struct ring_buffer_event *event;
1049         struct ctx_switch_entry *entry;
1050
1051         event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1052                                           sizeof(*entry), flags, pc);
1053         if (!event)
1054                 return;
1055         entry   = ring_buffer_event_data(event);
1056         entry->prev_pid                 = prev->pid;
1057         entry->prev_prio                = prev->prio;
1058         entry->prev_state               = prev->state;
1059         entry->next_pid                 = next->pid;
1060         entry->next_prio                = next->prio;
1061         entry->next_state               = next->state;
1062         entry->next_cpu = task_cpu(next);
1063         trace_buffer_unlock_commit(tr, event, flags, pc);
1064 }
1065
1066 void
1067 tracing_sched_wakeup_trace(struct trace_array *tr,
1068                            struct task_struct *wakee,
1069                            struct task_struct *curr,
1070                            unsigned long flags, int pc)
1071 {
1072         struct ring_buffer_event *event;
1073         struct ctx_switch_entry *entry;
1074
1075         event = trace_buffer_lock_reserve(tr, TRACE_WAKE,
1076                                           sizeof(*entry), flags, pc);
1077         if (!event)
1078                 return;
1079         entry   = ring_buffer_event_data(event);
1080         entry->prev_pid                 = curr->pid;
1081         entry->prev_prio                = curr->prio;
1082         entry->prev_state               = curr->state;
1083         entry->next_pid                 = wakee->pid;
1084         entry->next_prio                = wakee->prio;
1085         entry->next_state               = wakee->state;
1086         entry->next_cpu                 = task_cpu(wakee);
1087
1088         ring_buffer_unlock_commit(tr->buffer, event);
1089         ftrace_trace_stack(tr, flags, 6, pc);
1090         ftrace_trace_userstack(tr, flags, pc);
1091 }
1092
1093 void
1094 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1095 {
1096         struct trace_array *tr = &global_trace;
1097         struct trace_array_cpu *data;
1098         unsigned long flags;
1099         int cpu;
1100         int pc;
1101
1102         if (tracing_disabled)
1103                 return;
1104
1105         pc = preempt_count();
1106         local_irq_save(flags);
1107         cpu = raw_smp_processor_id();
1108         data = tr->data[cpu];
1109
1110         if (likely(atomic_inc_return(&data->disabled) == 1))
1111                 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1112
1113         atomic_dec(&data->disabled);
1114         local_irq_restore(flags);
1115 }
1116
1117 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1118 int trace_graph_entry(struct ftrace_graph_ent *trace)
1119 {
1120         struct trace_array *tr = &global_trace;
1121         struct trace_array_cpu *data;
1122         unsigned long flags;
1123         long disabled;
1124         int cpu;
1125         int pc;
1126
1127         if (!ftrace_trace_task(current))
1128                 return 0;
1129
1130         if (!ftrace_graph_addr(trace->func))
1131                 return 0;
1132
1133         local_irq_save(flags);
1134         cpu = raw_smp_processor_id();
1135         data = tr->data[cpu];
1136         disabled = atomic_inc_return(&data->disabled);
1137         if (likely(disabled == 1)) {
1138                 pc = preempt_count();
1139                 __trace_graph_entry(tr, trace, flags, pc);
1140         }
1141         /* Only do the atomic if it is not already set */
1142         if (!test_tsk_trace_graph(current))
1143                 set_tsk_trace_graph(current);
1144         atomic_dec(&data->disabled);
1145         local_irq_restore(flags);
1146
1147         return 1;
1148 }
1149
1150 void trace_graph_return(struct ftrace_graph_ret *trace)
1151 {
1152         struct trace_array *tr = &global_trace;
1153         struct trace_array_cpu *data;
1154         unsigned long flags;
1155         long disabled;
1156         int cpu;
1157         int pc;
1158
1159         local_irq_save(flags);
1160         cpu = raw_smp_processor_id();
1161         data = tr->data[cpu];
1162         disabled = atomic_inc_return(&data->disabled);
1163         if (likely(disabled == 1)) {
1164                 pc = preempt_count();
1165                 __trace_graph_return(tr, trace, flags, pc);
1166         }
1167         if (!trace->depth)
1168                 clear_tsk_trace_graph(current);
1169         atomic_dec(&data->disabled);
1170         local_irq_restore(flags);
1171 }
1172 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1173
1174 enum trace_file_type {
1175         TRACE_FILE_LAT_FMT      = 1,
1176         TRACE_FILE_ANNOTATE     = 2,
1177 };
1178
1179 static void trace_iterator_increment(struct trace_iterator *iter)
1180 {
1181         /* Don't allow ftrace to trace into the ring buffers */
1182         ftrace_disable_cpu();
1183
1184         iter->idx++;
1185         if (iter->buffer_iter[iter->cpu])
1186                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1187
1188         ftrace_enable_cpu();
1189 }
1190
1191 static struct trace_entry *
1192 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts)
1193 {
1194         struct ring_buffer_event *event;
1195         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1196
1197         /* Don't allow ftrace to trace into the ring buffers */
1198         ftrace_disable_cpu();
1199
1200         if (buf_iter)
1201                 event = ring_buffer_iter_peek(buf_iter, ts);
1202         else
1203                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts);
1204
1205         ftrace_enable_cpu();
1206
1207         return event ? ring_buffer_event_data(event) : NULL;
1208 }
1209
1210 static struct trace_entry *
1211 __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1212 {
1213         struct ring_buffer *buffer = iter->tr->buffer;
1214         struct trace_entry *ent, *next = NULL;
1215         int cpu_file = iter->cpu_file;
1216         u64 next_ts = 0, ts;
1217         int next_cpu = -1;
1218         int cpu;
1219
1220         /*
1221          * If we are in a per_cpu trace file, don't bother by iterating over
1222          * all cpu and peek directly.
1223          */
1224         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1225                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1226                         return NULL;
1227                 ent = peek_next_entry(iter, cpu_file, ent_ts);
1228                 if (ent_cpu)
1229                         *ent_cpu = cpu_file;
1230
1231                 return ent;
1232         }
1233
1234         for_each_tracing_cpu(cpu) {
1235
1236                 if (ring_buffer_empty_cpu(buffer, cpu))
1237                         continue;
1238
1239                 ent = peek_next_entry(iter, cpu, &ts);
1240
1241                 /*
1242                  * Pick the entry with the smallest timestamp:
1243                  */
1244                 if (ent && (!next || ts < next_ts)) {
1245                         next = ent;
1246                         next_cpu = cpu;
1247                         next_ts = ts;
1248                 }
1249         }
1250
1251         if (ent_cpu)
1252                 *ent_cpu = next_cpu;
1253
1254         if (ent_ts)
1255                 *ent_ts = next_ts;
1256
1257         return next;
1258 }
1259
1260 /* Find the next real entry, without updating the iterator itself */
1261 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1262                                           int *ent_cpu, u64 *ent_ts)
1263 {
1264         return __find_next_entry(iter, ent_cpu, ent_ts);
1265 }
1266
1267 /* Find the next real entry, and increment the iterator to the next entry */
1268 static void *find_next_entry_inc(struct trace_iterator *iter)
1269 {
1270         iter->ent = __find_next_entry(iter, &iter->cpu, &iter->ts);
1271
1272         if (iter->ent)
1273                 trace_iterator_increment(iter);
1274
1275         return iter->ent ? iter : NULL;
1276 }
1277
1278 static void trace_consume(struct trace_iterator *iter)
1279 {
1280         /* Don't allow ftrace to trace into the ring buffers */
1281         ftrace_disable_cpu();
1282         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts);
1283         ftrace_enable_cpu();
1284 }
1285
1286 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1287 {
1288         struct trace_iterator *iter = m->private;
1289         int i = (int)*pos;
1290         void *ent;
1291
1292         (*pos)++;
1293
1294         /* can't go backwards */
1295         if (iter->idx > i)
1296                 return NULL;
1297
1298         if (iter->idx < 0)
1299                 ent = find_next_entry_inc(iter);
1300         else
1301                 ent = iter;
1302
1303         while (ent && iter->idx < i)
1304                 ent = find_next_entry_inc(iter);
1305
1306         iter->pos = *pos;
1307
1308         return ent;
1309 }
1310
1311 /*
1312  * No necessary locking here. The worst thing which can
1313  * happen is loosing events consumed at the same time
1314  * by a trace_pipe reader.
1315  * Other than that, we don't risk to crash the ring buffer
1316  * because it serializes the readers.
1317  *
1318  * The current tracer is copied to avoid a global locking
1319  * all around.
1320  */
1321 static void *s_start(struct seq_file *m, loff_t *pos)
1322 {
1323         struct trace_iterator *iter = m->private;
1324         static struct tracer *old_tracer;
1325         int cpu_file = iter->cpu_file;
1326         void *p = NULL;
1327         loff_t l = 0;
1328         int cpu;
1329
1330         /* copy the tracer to avoid using a global lock all around */
1331         mutex_lock(&trace_types_lock);
1332         if (unlikely(old_tracer != current_trace && current_trace)) {
1333                 old_tracer = current_trace;
1334                 *iter->trace = *current_trace;
1335         }
1336         mutex_unlock(&trace_types_lock);
1337
1338         atomic_inc(&trace_record_cmdline_disabled);
1339
1340         if (*pos != iter->pos) {
1341                 iter->ent = NULL;
1342                 iter->cpu = 0;
1343                 iter->idx = -1;
1344
1345                 ftrace_disable_cpu();
1346
1347                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1348                         for_each_tracing_cpu(cpu)
1349                                 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1350                 } else
1351                         ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1352
1353
1354                 ftrace_enable_cpu();
1355
1356                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1357                         ;
1358
1359         } else {
1360                 l = *pos - 1;
1361                 p = s_next(m, p, &l);
1362         }
1363
1364         return p;
1365 }
1366
1367 static void s_stop(struct seq_file *m, void *p)
1368 {
1369         atomic_dec(&trace_record_cmdline_disabled);
1370 }
1371
1372 static void print_lat_help_header(struct seq_file *m)
1373 {
1374         seq_puts(m, "#                  _------=> CPU#            \n");
1375         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1376         seq_puts(m, "#                | / _----=> need-resched    \n");
1377         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1378         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1379         seq_puts(m, "#                |||| /                      \n");
1380         seq_puts(m, "#                |||||     delay             \n");
1381         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1382         seq_puts(m, "#     \\   /      |||||   \\   |   /           \n");
1383 }
1384
1385 static void print_func_help_header(struct seq_file *m)
1386 {
1387         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1388         seq_puts(m, "#              | |       |          |         |\n");
1389 }
1390
1391
1392 static void
1393 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1394 {
1395         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1396         struct trace_array *tr = iter->tr;
1397         struct trace_array_cpu *data = tr->data[tr->cpu];
1398         struct tracer *type = current_trace;
1399         unsigned long total;
1400         unsigned long entries;
1401         const char *name = "preemption";
1402
1403         if (type)
1404                 name = type->name;
1405
1406         entries = ring_buffer_entries(iter->tr->buffer);
1407         total = entries +
1408                 ring_buffer_overruns(iter->tr->buffer);
1409
1410         seq_printf(m, "%s latency trace v1.1.5 on %s\n",
1411                    name, UTS_RELEASE);
1412         seq_puts(m, "-----------------------------------"
1413                  "---------------------------------\n");
1414         seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |"
1415                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1416                    nsecs_to_usecs(data->saved_latency),
1417                    entries,
1418                    total,
1419                    tr->cpu,
1420 #if defined(CONFIG_PREEMPT_NONE)
1421                    "server",
1422 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1423                    "desktop",
1424 #elif defined(CONFIG_PREEMPT)
1425                    "preempt",
1426 #else
1427                    "unknown",
1428 #endif
1429                    /* These are reserved for later use */
1430                    0, 0, 0, 0);
1431 #ifdef CONFIG_SMP
1432         seq_printf(m, " #P:%d)\n", num_online_cpus());
1433 #else
1434         seq_puts(m, ")\n");
1435 #endif
1436         seq_puts(m, "    -----------------\n");
1437         seq_printf(m, "    | task: %.16s-%d "
1438                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1439                    data->comm, data->pid, data->uid, data->nice,
1440                    data->policy, data->rt_priority);
1441         seq_puts(m, "    -----------------\n");
1442
1443         if (data->critical_start) {
1444                 seq_puts(m, " => started at: ");
1445                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1446                 trace_print_seq(m, &iter->seq);
1447                 seq_puts(m, "\n => ended at:   ");
1448                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1449                 trace_print_seq(m, &iter->seq);
1450                 seq_puts(m, "\n");
1451         }
1452
1453         seq_puts(m, "\n");
1454 }
1455
1456 static void test_cpu_buff_start(struct trace_iterator *iter)
1457 {
1458         struct trace_seq *s = &iter->seq;
1459
1460         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1461                 return;
1462
1463         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1464                 return;
1465
1466         if (cpumask_test_cpu(iter->cpu, iter->started))
1467                 return;
1468
1469         cpumask_set_cpu(iter->cpu, iter->started);
1470         trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1471 }
1472
1473 static enum print_line_t print_lat_fmt(struct trace_iterator *iter)
1474 {
1475         struct trace_seq *s = &iter->seq;
1476         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1477         struct trace_event *event;
1478         struct trace_entry *entry = iter->ent;
1479
1480         test_cpu_buff_start(iter);
1481
1482         event = ftrace_find_event(entry->type);
1483
1484         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1485                 if (!trace_print_lat_context(iter))
1486                         goto partial;
1487         }
1488
1489         if (event)
1490                 return event->latency_trace(iter, sym_flags);
1491
1492         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1493                 goto partial;
1494
1495         return TRACE_TYPE_HANDLED;
1496 partial:
1497         return TRACE_TYPE_PARTIAL_LINE;
1498 }
1499
1500 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1501 {
1502         struct trace_seq *s = &iter->seq;
1503         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1504         struct trace_entry *entry;
1505         struct trace_event *event;
1506
1507         entry = iter->ent;
1508
1509         test_cpu_buff_start(iter);
1510
1511         event = ftrace_find_event(entry->type);
1512
1513         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1514                 if (!trace_print_context(iter))
1515                         goto partial;
1516         }
1517
1518         if (event)
1519                 return event->trace(iter, sym_flags);
1520
1521         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1522                 goto partial;
1523
1524         return TRACE_TYPE_HANDLED;
1525 partial:
1526         return TRACE_TYPE_PARTIAL_LINE;
1527 }
1528
1529 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1530 {
1531         struct trace_seq *s = &iter->seq;
1532         struct trace_entry *entry;
1533         struct trace_event *event;
1534
1535         entry = iter->ent;
1536
1537         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1538                 if (!trace_seq_printf(s, "%d %d %llu ",
1539                                       entry->pid, iter->cpu, iter->ts))
1540                         goto partial;
1541         }
1542
1543         event = ftrace_find_event(entry->type);
1544         if (event)
1545                 return event->raw(iter, 0);
1546
1547         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1548                 goto partial;
1549
1550         return TRACE_TYPE_HANDLED;
1551 partial:
1552         return TRACE_TYPE_PARTIAL_LINE;
1553 }
1554
1555 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1556 {
1557         struct trace_seq *s = &iter->seq;
1558         unsigned char newline = '\n';
1559         struct trace_entry *entry;
1560         struct trace_event *event;
1561
1562         entry = iter->ent;
1563
1564         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1565                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1566                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1567                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1568         }
1569
1570         event = ftrace_find_event(entry->type);
1571         if (event) {
1572                 enum print_line_t ret = event->hex(iter, 0);
1573                 if (ret != TRACE_TYPE_HANDLED)
1574                         return ret;
1575         }
1576
1577         SEQ_PUT_FIELD_RET(s, newline);
1578
1579         return TRACE_TYPE_HANDLED;
1580 }
1581
1582 static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
1583 {
1584         struct trace_seq *s = &iter->seq;
1585         struct trace_entry *entry = iter->ent;
1586         struct print_entry *field;
1587         int ret;
1588
1589         trace_assign_type(field, entry);
1590
1591         ret = trace_seq_printf(s, "%s", field->buf);
1592         if (!ret)
1593                 return TRACE_TYPE_PARTIAL_LINE;
1594
1595         return TRACE_TYPE_HANDLED;
1596 }
1597
1598 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1599 {
1600         struct trace_seq *s = &iter->seq;
1601         struct trace_entry *entry;
1602         struct trace_event *event;
1603
1604         entry = iter->ent;
1605
1606         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1607                 SEQ_PUT_FIELD_RET(s, entry->pid);
1608                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1609                 SEQ_PUT_FIELD_RET(s, iter->ts);
1610         }
1611
1612         event = ftrace_find_event(entry->type);
1613         return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
1614 }
1615
1616 static int trace_empty(struct trace_iterator *iter)
1617 {
1618         int cpu;
1619
1620         for_each_tracing_cpu(cpu) {
1621                 if (iter->buffer_iter[cpu]) {
1622                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1623                                 return 0;
1624                 } else {
1625                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1626                                 return 0;
1627                 }
1628         }
1629
1630         return 1;
1631 }
1632
1633 static enum print_line_t print_trace_line(struct trace_iterator *iter)
1634 {
1635         enum print_line_t ret;
1636
1637         if (iter->trace && iter->trace->print_line) {
1638                 ret = iter->trace->print_line(iter);
1639                 if (ret != TRACE_TYPE_UNHANDLED)
1640                         return ret;
1641         }
1642
1643         if (iter->ent->type == TRACE_PRINT &&
1644                         trace_flags & TRACE_ITER_PRINTK &&
1645                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1646                 return print_printk_msg_only(iter);
1647
1648         if (trace_flags & TRACE_ITER_BIN)
1649                 return print_bin_fmt(iter);
1650
1651         if (trace_flags & TRACE_ITER_HEX)
1652                 return print_hex_fmt(iter);
1653
1654         if (trace_flags & TRACE_ITER_RAW)
1655                 return print_raw_fmt(iter);
1656
1657         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
1658                 return print_lat_fmt(iter);
1659
1660         return print_trace_fmt(iter);
1661 }
1662
1663 static int s_show(struct seq_file *m, void *v)
1664 {
1665         struct trace_iterator *iter = v;
1666
1667         if (iter->ent == NULL) {
1668                 if (iter->tr) {
1669                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
1670                         seq_puts(m, "#\n");
1671                 }
1672                 if (iter->trace && iter->trace->print_header)
1673                         iter->trace->print_header(m);
1674                 else if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1675                         /* print nothing if the buffers are empty */
1676                         if (trace_empty(iter))
1677                                 return 0;
1678                         print_trace_header(m, iter);
1679                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1680                                 print_lat_help_header(m);
1681                 } else {
1682                         if (!(trace_flags & TRACE_ITER_VERBOSE))
1683                                 print_func_help_header(m);
1684                 }
1685         } else {
1686                 print_trace_line(iter);
1687                 trace_print_seq(m, &iter->seq);
1688         }
1689
1690         return 0;
1691 }
1692
1693 static struct seq_operations tracer_seq_ops = {
1694         .start          = s_start,
1695         .next           = s_next,
1696         .stop           = s_stop,
1697         .show           = s_show,
1698 };
1699
1700 static struct trace_iterator *
1701 __tracing_open(struct inode *inode, struct file *file)
1702 {
1703         long cpu_file = (long) inode->i_private;
1704         void *fail_ret = ERR_PTR(-ENOMEM);
1705         struct trace_iterator *iter;
1706         struct seq_file *m;
1707         int cpu, ret;
1708
1709         if (tracing_disabled)
1710                 return ERR_PTR(-ENODEV);
1711
1712         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
1713         if (!iter)
1714                 return ERR_PTR(-ENOMEM);
1715
1716         /*
1717          * We make a copy of the current tracer to avoid concurrent
1718          * changes on it while we are reading.
1719          */
1720         mutex_lock(&trace_types_lock);
1721         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1722         if (!iter->trace)
1723                 goto fail;
1724
1725         if (current_trace)
1726                 *iter->trace = *current_trace;
1727
1728         if (current_trace && current_trace->print_max)
1729                 iter->tr = &max_tr;
1730         else
1731                 iter->tr = &global_trace;
1732         iter->pos = -1;
1733         mutex_init(&iter->mutex);
1734         iter->cpu_file = cpu_file;
1735
1736         /* Notify the tracer early; before we stop tracing. */
1737         if (iter->trace && iter->trace->open)
1738                 iter->trace->open(iter);
1739
1740         /* Annotate start of buffers if we had overruns */
1741         if (ring_buffer_overruns(iter->tr->buffer))
1742                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
1743
1744         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1745                 for_each_tracing_cpu(cpu) {
1746
1747                         iter->buffer_iter[cpu] =
1748                                 ring_buffer_read_start(iter->tr->buffer, cpu);
1749
1750                         if (!iter->buffer_iter[cpu])
1751                                 goto fail_buffer;
1752                 }
1753         } else {
1754                 cpu = iter->cpu_file;
1755                 iter->buffer_iter[cpu] =
1756                                 ring_buffer_read_start(iter->tr->buffer, cpu);
1757
1758                 if (!iter->buffer_iter[cpu])
1759                         goto fail;
1760         }
1761
1762         /* TODO stop tracer */
1763         ret = seq_open(file, &tracer_seq_ops);
1764         if (ret < 0) {
1765                 fail_ret = ERR_PTR(ret);
1766                 goto fail_buffer;
1767         }
1768
1769         m = file->private_data;
1770         m->private = iter;
1771
1772         /* stop the trace while dumping */
1773         tracing_stop();
1774
1775         mutex_unlock(&trace_types_lock);
1776
1777         return iter;
1778
1779  fail_buffer:
1780         for_each_tracing_cpu(cpu) {
1781                 if (iter->buffer_iter[cpu])
1782                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
1783         }
1784  fail:
1785         mutex_unlock(&trace_types_lock);
1786         kfree(iter->trace);
1787         kfree(iter);
1788
1789         return fail_ret;
1790 }
1791
1792 int tracing_open_generic(struct inode *inode, struct file *filp)
1793 {
1794         if (tracing_disabled)
1795                 return -ENODEV;
1796
1797         filp->private_data = inode->i_private;
1798         return 0;
1799 }
1800
1801 static int tracing_release(struct inode *inode, struct file *file)
1802 {
1803         struct seq_file *m = (struct seq_file *)file->private_data;
1804         struct trace_iterator *iter = m->private;
1805         int cpu;
1806
1807         mutex_lock(&trace_types_lock);
1808         for_each_tracing_cpu(cpu) {
1809                 if (iter->buffer_iter[cpu])
1810                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
1811         }
1812
1813         if (iter->trace && iter->trace->close)
1814                 iter->trace->close(iter);
1815
1816         /* reenable tracing if it was previously enabled */
1817         tracing_start();
1818         mutex_unlock(&trace_types_lock);
1819
1820         seq_release(inode, file);
1821         mutex_destroy(&iter->mutex);
1822         kfree(iter->trace);
1823         kfree(iter);
1824         return 0;
1825 }
1826
1827 static int tracing_open(struct inode *inode, struct file *file)
1828 {
1829         struct trace_iterator *iter;
1830         int ret = 0;
1831
1832         iter = __tracing_open(inode, file);
1833         if (IS_ERR(iter))
1834                 ret = PTR_ERR(iter);
1835
1836         return ret;
1837 }
1838
1839 static int tracing_lt_open(struct inode *inode, struct file *file)
1840 {
1841         struct trace_iterator *iter;
1842         int ret = 0;
1843
1844         iter = __tracing_open(inode, file);
1845
1846         if (IS_ERR(iter))
1847                 ret = PTR_ERR(iter);
1848         else
1849                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1850
1851         return ret;
1852 }
1853
1854
1855 static void *
1856 t_next(struct seq_file *m, void *v, loff_t *pos)
1857 {
1858         struct tracer *t = m->private;
1859
1860         (*pos)++;
1861
1862         if (t)
1863                 t = t->next;
1864
1865         m->private = t;
1866
1867         return t;
1868 }
1869
1870 static void *t_start(struct seq_file *m, loff_t *pos)
1871 {
1872         struct tracer *t = m->private;
1873         loff_t l = 0;
1874
1875         mutex_lock(&trace_types_lock);
1876         for (; t && l < *pos; t = t_next(m, t, &l))
1877                 ;
1878
1879         return t;
1880 }
1881
1882 static void t_stop(struct seq_file *m, void *p)
1883 {
1884         mutex_unlock(&trace_types_lock);
1885 }
1886
1887 static int t_show(struct seq_file *m, void *v)
1888 {
1889         struct tracer *t = v;
1890
1891         if (!t)
1892                 return 0;
1893
1894         seq_printf(m, "%s", t->name);
1895         if (t->next)
1896                 seq_putc(m, ' ');
1897         else
1898                 seq_putc(m, '\n');
1899
1900         return 0;
1901 }
1902
1903 static struct seq_operations show_traces_seq_ops = {
1904         .start          = t_start,
1905         .next           = t_next,
1906         .stop           = t_stop,
1907         .show           = t_show,
1908 };
1909
1910 static int show_traces_open(struct inode *inode, struct file *file)
1911 {
1912         int ret;
1913
1914         if (tracing_disabled)
1915                 return -ENODEV;
1916
1917         ret = seq_open(file, &show_traces_seq_ops);
1918         if (!ret) {
1919                 struct seq_file *m = file->private_data;
1920                 m->private = trace_types;
1921         }
1922
1923         return ret;
1924 }
1925
1926 static struct file_operations tracing_fops = {
1927         .open           = tracing_open,
1928         .read           = seq_read,
1929         .llseek         = seq_lseek,
1930         .release        = tracing_release,
1931 };
1932
1933 static struct file_operations tracing_lt_fops = {
1934         .open           = tracing_lt_open,
1935         .read           = seq_read,
1936         .llseek         = seq_lseek,
1937         .release        = tracing_release,
1938 };
1939
1940 static struct file_operations show_traces_fops = {
1941         .open           = show_traces_open,
1942         .read           = seq_read,
1943         .release        = seq_release,
1944 };
1945
1946 /*
1947  * Only trace on a CPU if the bitmask is set:
1948  */
1949 static cpumask_var_t tracing_cpumask;
1950
1951 /*
1952  * The tracer itself will not take this lock, but still we want
1953  * to provide a consistent cpumask to user-space:
1954  */
1955 static DEFINE_MUTEX(tracing_cpumask_update_lock);
1956
1957 /*
1958  * Temporary storage for the character representation of the
1959  * CPU bitmask (and one more byte for the newline):
1960  */
1961 static char mask_str[NR_CPUS + 1];
1962
1963 static ssize_t
1964 tracing_cpumask_read(struct file *filp, char __user *ubuf,
1965                      size_t count, loff_t *ppos)
1966 {
1967         int len;
1968
1969         mutex_lock(&tracing_cpumask_update_lock);
1970
1971         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
1972         if (count - len < 2) {
1973                 count = -EINVAL;
1974                 goto out_err;
1975         }
1976         len += sprintf(mask_str + len, "\n");
1977         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
1978
1979 out_err:
1980         mutex_unlock(&tracing_cpumask_update_lock);
1981
1982         return count;
1983 }
1984
1985 static ssize_t
1986 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
1987                       size_t count, loff_t *ppos)
1988 {
1989         int err, cpu;
1990         cpumask_var_t tracing_cpumask_new;
1991
1992         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
1993                 return -ENOMEM;
1994
1995         mutex_lock(&tracing_cpumask_update_lock);
1996         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
1997         if (err)
1998                 goto err_unlock;
1999
2000         local_irq_disable();
2001         __raw_spin_lock(&ftrace_max_lock);
2002         for_each_tracing_cpu(cpu) {
2003                 /*
2004                  * Increase/decrease the disabled counter if we are
2005                  * about to flip a bit in the cpumask:
2006                  */
2007                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2008                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2009                         atomic_inc(&global_trace.data[cpu]->disabled);
2010                 }
2011                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2012                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2013                         atomic_dec(&global_trace.data[cpu]->disabled);
2014                 }
2015         }
2016         __raw_spin_unlock(&ftrace_max_lock);
2017         local_irq_enable();
2018
2019         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2020
2021         mutex_unlock(&tracing_cpumask_update_lock);
2022         free_cpumask_var(tracing_cpumask_new);
2023
2024         return count;
2025
2026 err_unlock:
2027         mutex_unlock(&tracing_cpumask_update_lock);
2028         free_cpumask_var(tracing_cpumask);
2029
2030         return err;
2031 }
2032
2033 static struct file_operations tracing_cpumask_fops = {
2034         .open           = tracing_open_generic,
2035         .read           = tracing_cpumask_read,
2036         .write          = tracing_cpumask_write,
2037 };
2038
2039 static ssize_t
2040 tracing_trace_options_read(struct file *filp, char __user *ubuf,
2041                        size_t cnt, loff_t *ppos)
2042 {
2043         struct tracer_opt *trace_opts;
2044         u32 tracer_flags;
2045         int len = 0;
2046         char *buf;
2047         int r = 0;
2048         int i;
2049
2050
2051         /* calculate max size */
2052         for (i = 0; trace_options[i]; i++) {
2053                 len += strlen(trace_options[i]);
2054                 len += 3; /* "no" and newline */
2055         }
2056
2057         mutex_lock(&trace_types_lock);
2058         tracer_flags = current_trace->flags->val;
2059         trace_opts = current_trace->flags->opts;
2060
2061         /*
2062          * Increase the size with names of options specific
2063          * of the current tracer.
2064          */
2065         for (i = 0; trace_opts[i].name; i++) {
2066                 len += strlen(trace_opts[i].name);
2067                 len += 3; /* "no" and newline */
2068         }
2069
2070         /* +2 for \n and \0 */
2071         buf = kmalloc(len + 2, GFP_KERNEL);
2072         if (!buf) {
2073                 mutex_unlock(&trace_types_lock);
2074                 return -ENOMEM;
2075         }
2076
2077         for (i = 0; trace_options[i]; i++) {
2078                 if (trace_flags & (1 << i))
2079                         r += sprintf(buf + r, "%s\n", trace_options[i]);
2080                 else
2081                         r += sprintf(buf + r, "no%s\n", trace_options[i]);
2082         }
2083
2084         for (i = 0; trace_opts[i].name; i++) {
2085                 if (tracer_flags & trace_opts[i].bit)
2086                         r += sprintf(buf + r, "%s\n",
2087                                 trace_opts[i].name);
2088                 else
2089                         r += sprintf(buf + r, "no%s\n",
2090                                 trace_opts[i].name);
2091         }
2092         mutex_unlock(&trace_types_lock);
2093
2094         WARN_ON(r >= len + 2);
2095
2096         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2097
2098         kfree(buf);
2099         return r;
2100 }
2101
2102 /* Try to assign a tracer specific option */
2103 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2104 {
2105         struct tracer_flags *trace_flags = trace->flags;
2106         struct tracer_opt *opts = NULL;
2107         int ret = 0, i = 0;
2108         int len;
2109
2110         for (i = 0; trace_flags->opts[i].name; i++) {
2111                 opts = &trace_flags->opts[i];
2112                 len = strlen(opts->name);
2113
2114                 if (strncmp(cmp, opts->name, len) == 0) {
2115                         ret = trace->set_flag(trace_flags->val,
2116                                 opts->bit, !neg);
2117                         break;
2118                 }
2119         }
2120         /* Not found */
2121         if (!trace_flags->opts[i].name)
2122                 return -EINVAL;
2123
2124         /* Refused to handle */
2125         if (ret)
2126                 return ret;
2127
2128         if (neg)
2129                 trace_flags->val &= ~opts->bit;
2130         else
2131                 trace_flags->val |= opts->bit;
2132
2133         return 0;
2134 }
2135
2136 static ssize_t
2137 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2138                         size_t cnt, loff_t *ppos)
2139 {
2140         char buf[64];
2141         char *cmp = buf;
2142         int neg = 0;
2143         int ret;
2144         int i;
2145
2146         if (cnt >= sizeof(buf))
2147                 return -EINVAL;
2148
2149         if (copy_from_user(&buf, ubuf, cnt))
2150                 return -EFAULT;
2151
2152         buf[cnt] = 0;
2153
2154         if (strncmp(buf, "no", 2) == 0) {
2155                 neg = 1;
2156                 cmp += 2;
2157         }
2158
2159         for (i = 0; trace_options[i]; i++) {
2160                 int len = strlen(trace_options[i]);
2161
2162                 if (strncmp(cmp, trace_options[i], len) == 0) {
2163                         if (neg)
2164                                 trace_flags &= ~(1 << i);
2165                         else
2166                                 trace_flags |= (1 << i);
2167                         break;
2168                 }
2169         }
2170
2171         /* If no option could be set, test the specific tracer options */
2172         if (!trace_options[i]) {
2173                 mutex_lock(&trace_types_lock);
2174                 ret = set_tracer_option(current_trace, cmp, neg);
2175                 mutex_unlock(&trace_types_lock);
2176                 if (ret)
2177                         return ret;
2178         }
2179
2180         filp->f_pos += cnt;
2181
2182         return cnt;
2183 }
2184
2185 static struct file_operations tracing_iter_fops = {
2186         .open           = tracing_open_generic,
2187         .read           = tracing_trace_options_read,
2188         .write          = tracing_trace_options_write,
2189 };
2190
2191 static const char readme_msg[] =
2192         "tracing mini-HOWTO:\n\n"
2193         "# mkdir /debug\n"
2194         "# mount -t debugfs nodev /debug\n\n"
2195         "# cat /debug/tracing/available_tracers\n"
2196         "wakeup preemptirqsoff preemptoff irqsoff ftrace sched_switch none\n\n"
2197         "# cat /debug/tracing/current_tracer\n"
2198         "none\n"
2199         "# echo sched_switch > /debug/tracing/current_tracer\n"
2200         "# cat /debug/tracing/current_tracer\n"
2201         "sched_switch\n"
2202         "# cat /debug/tracing/trace_options\n"
2203         "noprint-parent nosym-offset nosym-addr noverbose\n"
2204         "# echo print-parent > /debug/tracing/trace_options\n"
2205         "# echo 1 > /debug/tracing/tracing_enabled\n"
2206         "# cat /debug/tracing/trace > /tmp/trace.txt\n"
2207         "echo 0 > /debug/tracing/tracing_enabled\n"
2208 ;
2209
2210 static ssize_t
2211 tracing_readme_read(struct file *filp, char __user *ubuf,
2212                        size_t cnt, loff_t *ppos)
2213 {
2214         return simple_read_from_buffer(ubuf, cnt, ppos,
2215                                         readme_msg, strlen(readme_msg));
2216 }
2217
2218 static struct file_operations tracing_readme_fops = {
2219         .open           = tracing_open_generic,
2220         .read           = tracing_readme_read,
2221 };
2222
2223 static ssize_t
2224 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2225                   size_t cnt, loff_t *ppos)
2226 {
2227         char buf[64];
2228         int r;
2229
2230         r = sprintf(buf, "%u\n", tracer_enabled);
2231         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2232 }
2233
2234 static ssize_t
2235 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2236                    size_t cnt, loff_t *ppos)
2237 {
2238         struct trace_array *tr = filp->private_data;
2239         char buf[64];
2240         unsigned long val;
2241         int ret;
2242
2243         if (cnt >= sizeof(buf))
2244                 return -EINVAL;
2245
2246         if (copy_from_user(&buf, ubuf, cnt))
2247                 return -EFAULT;
2248
2249         buf[cnt] = 0;
2250
2251         ret = strict_strtoul(buf, 10, &val);
2252         if (ret < 0)
2253                 return ret;
2254
2255         val = !!val;
2256
2257         mutex_lock(&trace_types_lock);
2258         if (tracer_enabled ^ val) {
2259                 if (val) {
2260                         tracer_enabled = 1;
2261                         if (current_trace->start)
2262                                 current_trace->start(tr);
2263                         tracing_start();
2264                 } else {
2265                         tracer_enabled = 0;
2266                         tracing_stop();
2267                         if (current_trace->stop)
2268                                 current_trace->stop(tr);
2269                 }
2270         }
2271         mutex_unlock(&trace_types_lock);
2272
2273         filp->f_pos += cnt;
2274
2275         return cnt;
2276 }
2277
2278 static ssize_t
2279 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2280                        size_t cnt, loff_t *ppos)
2281 {
2282         char buf[max_tracer_type_len+2];
2283         int r;
2284
2285         mutex_lock(&trace_types_lock);
2286         if (current_trace)
2287                 r = sprintf(buf, "%s\n", current_trace->name);
2288         else
2289                 r = sprintf(buf, "\n");
2290         mutex_unlock(&trace_types_lock);
2291
2292         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2293 }
2294
2295 int tracer_init(struct tracer *t, struct trace_array *tr)
2296 {
2297         tracing_reset_online_cpus(tr);
2298         return t->init(tr);
2299 }
2300
2301 struct trace_option_dentry;
2302
2303 static struct trace_option_dentry *
2304 create_trace_option_files(struct tracer *tracer);
2305
2306 static void
2307 destroy_trace_option_files(struct trace_option_dentry *topts);
2308
2309 static int tracing_set_tracer(const char *buf)
2310 {
2311         static struct trace_option_dentry *topts;
2312         struct trace_array *tr = &global_trace;
2313         struct tracer *t;
2314         int ret = 0;
2315
2316         mutex_lock(&trace_types_lock);
2317         for (t = trace_types; t; t = t->next) {
2318                 if (strcmp(t->name, buf) == 0)
2319                         break;
2320         }
2321         if (!t) {
2322                 ret = -EINVAL;
2323                 goto out;
2324         }
2325         if (t == current_trace)
2326                 goto out;
2327
2328         trace_branch_disable();
2329         if (current_trace && current_trace->reset)
2330                 current_trace->reset(tr);
2331
2332         destroy_trace_option_files(topts);
2333
2334         current_trace = t;
2335
2336         topts = create_trace_option_files(current_trace);
2337
2338         if (t->init) {
2339                 ret = tracer_init(t, tr);
2340                 if (ret)
2341                         goto out;
2342         }
2343
2344         trace_branch_enable(tr);
2345  out:
2346         mutex_unlock(&trace_types_lock);
2347
2348         return ret;
2349 }
2350
2351 static ssize_t
2352 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2353                         size_t cnt, loff_t *ppos)
2354 {
2355         char buf[max_tracer_type_len+1];
2356         int i;
2357         size_t ret;
2358         int err;
2359
2360         ret = cnt;
2361
2362         if (cnt > max_tracer_type_len)
2363                 cnt = max_tracer_type_len;
2364
2365         if (copy_from_user(&buf, ubuf, cnt))
2366                 return -EFAULT;
2367
2368         buf[cnt] = 0;
2369
2370         /* strip ending whitespace. */
2371         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2372                 buf[i] = 0;
2373
2374         err = tracing_set_tracer(buf);
2375         if (err)
2376                 return err;
2377
2378         filp->f_pos += ret;
2379
2380         return ret;
2381 }
2382
2383 static ssize_t
2384 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2385                      size_t cnt, loff_t *ppos)
2386 {
2387         unsigned long *ptr = filp->private_data;
2388         char buf[64];
2389         int r;
2390
2391         r = snprintf(buf, sizeof(buf), "%ld\n",
2392                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2393         if (r > sizeof(buf))
2394                 r = sizeof(buf);
2395         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2396 }
2397
2398 static ssize_t
2399 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2400                       size_t cnt, loff_t *ppos)
2401 {
2402         unsigned long *ptr = filp->private_data;
2403         char buf[64];
2404         unsigned long val;
2405         int ret;
2406
2407         if (cnt >= sizeof(buf))
2408                 return -EINVAL;
2409
2410         if (copy_from_user(&buf, ubuf, cnt))
2411                 return -EFAULT;
2412
2413         buf[cnt] = 0;
2414
2415         ret = strict_strtoul(buf, 10, &val);
2416         if (ret < 0)
2417                 return ret;
2418
2419         *ptr = val * 1000;
2420
2421         return cnt;
2422 }
2423
2424 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2425 {
2426         long cpu_file = (long) inode->i_private;
2427         struct trace_iterator *iter;
2428         int ret = 0;
2429
2430         if (tracing_disabled)
2431                 return -ENODEV;
2432
2433         mutex_lock(&trace_types_lock);
2434
2435         /* We only allow one reader per cpu */
2436         if (cpu_file == TRACE_PIPE_ALL_CPU) {
2437                 if (!cpumask_empty(tracing_reader_cpumask)) {
2438                         ret = -EBUSY;
2439                         goto out;
2440                 }
2441                 cpumask_setall(tracing_reader_cpumask);
2442         } else {
2443                 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2444                         cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2445                 else {
2446                         ret = -EBUSY;
2447                         goto out;
2448                 }
2449         }
2450
2451         /* create a buffer to store the information to pass to userspace */
2452         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2453         if (!iter) {
2454                 ret = -ENOMEM;
2455                 goto out;
2456         }
2457
2458         /*
2459          * We make a copy of the current tracer to avoid concurrent
2460          * changes on it while we are reading.
2461          */
2462         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2463         if (!iter->trace) {
2464                 ret = -ENOMEM;
2465                 goto fail;
2466         }
2467         if (current_trace)
2468                 *iter->trace = *current_trace;
2469
2470         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2471                 ret = -ENOMEM;
2472                 goto fail;
2473         }
2474
2475         /* trace pipe does not show start of buffer */
2476         cpumask_setall(iter->started);
2477
2478         iter->cpu_file = cpu_file;
2479         iter->tr = &global_trace;
2480         mutex_init(&iter->mutex);
2481         filp->private_data = iter;
2482
2483         if (iter->trace->pipe_open)
2484                 iter->trace->pipe_open(iter);
2485
2486 out:
2487         mutex_unlock(&trace_types_lock);
2488         return ret;
2489
2490 fail:
2491         kfree(iter->trace);
2492         kfree(iter);
2493         mutex_unlock(&trace_types_lock);
2494         return ret;
2495 }
2496
2497 static int tracing_release_pipe(struct inode *inode, struct file *file)
2498 {
2499         struct trace_iterator *iter = file->private_data;
2500
2501         mutex_lock(&trace_types_lock);
2502
2503         if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2504                 cpumask_clear(tracing_reader_cpumask);
2505         else
2506                 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2507
2508         mutex_unlock(&trace_types_lock);
2509
2510         free_cpumask_var(iter->started);
2511         mutex_destroy(&iter->mutex);
2512         kfree(iter->trace);
2513         kfree(iter);
2514
2515         return 0;
2516 }
2517
2518 static unsigned int
2519 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
2520 {
2521         struct trace_iterator *iter = filp->private_data;
2522
2523         if (trace_flags & TRACE_ITER_BLOCK) {
2524                 /*
2525                  * Always select as readable when in blocking mode
2526                  */
2527                 return POLLIN | POLLRDNORM;
2528         } else {
2529                 if (!trace_empty(iter))
2530                         return POLLIN | POLLRDNORM;
2531                 poll_wait(filp, &trace_wait, poll_table);
2532                 if (!trace_empty(iter))
2533                         return POLLIN | POLLRDNORM;
2534
2535                 return 0;
2536         }
2537 }
2538
2539
2540 void default_wait_pipe(struct trace_iterator *iter)
2541 {
2542         DEFINE_WAIT(wait);
2543
2544         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
2545
2546         if (trace_empty(iter))
2547                 schedule();
2548
2549         finish_wait(&trace_wait, &wait);
2550 }
2551
2552 /*
2553  * This is a make-shift waitqueue.
2554  * A tracer might use this callback on some rare cases:
2555  *
2556  *  1) the current tracer might hold the runqueue lock when it wakes up
2557  *     a reader, hence a deadlock (sched, function, and function graph tracers)
2558  *  2) the function tracers, trace all functions, we don't want
2559  *     the overhead of calling wake_up and friends
2560  *     (and tracing them too)
2561  *
2562  *     Anyway, this is really very primitive wakeup.
2563  */
2564 void poll_wait_pipe(struct trace_iterator *iter)
2565 {
2566         set_current_state(TASK_INTERRUPTIBLE);
2567         /* sleep for 100 msecs, and try again. */
2568         schedule_timeout(HZ / 10);
2569 }
2570
2571 /* Must be called with trace_types_lock mutex held. */
2572 static int tracing_wait_pipe(struct file *filp)
2573 {
2574         struct trace_iterator *iter = filp->private_data;
2575
2576         while (trace_empty(iter)) {
2577
2578                 if ((filp->f_flags & O_NONBLOCK)) {
2579                         return -EAGAIN;
2580                 }
2581
2582                 mutex_unlock(&iter->mutex);
2583
2584                 iter->trace->wait_pipe(iter);
2585
2586                 mutex_lock(&iter->mutex);
2587
2588                 if (signal_pending(current))
2589                         return -EINTR;
2590
2591                 /*
2592                  * We block until we read something and tracing is disabled.
2593                  * We still block if tracing is disabled, but we have never
2594                  * read anything. This allows a user to cat this file, and
2595                  * then enable tracing. But after we have read something,
2596                  * we give an EOF when tracing is again disabled.
2597                  *
2598                  * iter->pos will be 0 if we haven't read anything.
2599                  */
2600                 if (!tracer_enabled && iter->pos)
2601                         break;
2602         }
2603
2604         return 1;
2605 }
2606
2607 /*
2608  * Consumer reader.
2609  */
2610 static ssize_t
2611 tracing_read_pipe(struct file *filp, char __user *ubuf,
2612                   size_t cnt, loff_t *ppos)
2613 {
2614         struct trace_iterator *iter = filp->private_data;
2615         static struct tracer *old_tracer;
2616         ssize_t sret;
2617
2618         /* return any leftover data */
2619         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2620         if (sret != -EBUSY)
2621                 return sret;
2622
2623         trace_seq_reset(&iter->seq);
2624
2625         /* copy the tracer to avoid using a global lock all around */
2626         mutex_lock(&trace_types_lock);
2627         if (unlikely(old_tracer != current_trace && current_trace)) {
2628                 old_tracer = current_trace;
2629                 *iter->trace = *current_trace;
2630         }
2631         mutex_unlock(&trace_types_lock);
2632
2633         /*
2634          * Avoid more than one consumer on a single file descriptor
2635          * This is just a matter of traces coherency, the ring buffer itself
2636          * is protected.
2637          */
2638         mutex_lock(&iter->mutex);
2639         if (iter->trace->read) {
2640                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2641                 if (sret)
2642                         goto out;
2643         }
2644
2645 waitagain:
2646         sret = tracing_wait_pipe(filp);
2647         if (sret <= 0)
2648                 goto out;
2649
2650         /* stop when tracing is finished */
2651         if (trace_empty(iter)) {
2652                 sret = 0;
2653                 goto out;
2654         }
2655
2656         if (cnt >= PAGE_SIZE)
2657                 cnt = PAGE_SIZE - 1;
2658
2659         /* reset all but tr, trace, and overruns */
2660         memset(&iter->seq, 0,
2661                sizeof(struct trace_iterator) -
2662                offsetof(struct trace_iterator, seq));
2663         iter->pos = -1;
2664
2665         while (find_next_entry_inc(iter) != NULL) {
2666                 enum print_line_t ret;
2667                 int len = iter->seq.len;
2668
2669                 ret = print_trace_line(iter);
2670                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2671                         /* don't print partial lines */
2672                         iter->seq.len = len;
2673                         break;
2674                 }
2675                 if (ret != TRACE_TYPE_NO_CONSUME)
2676                         trace_consume(iter);
2677
2678                 if (iter->seq.len >= cnt)
2679                         break;
2680         }
2681
2682         /* Now copy what we have to the user */
2683         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2684         if (iter->seq.readpos >= iter->seq.len)
2685                 trace_seq_reset(&iter->seq);
2686
2687         /*
2688          * If there was nothing to send to user, inspite of consuming trace
2689          * entries, go back to wait for more entries.
2690          */
2691         if (sret == -EBUSY)
2692                 goto waitagain;
2693
2694 out:
2695         mutex_unlock(&iter->mutex);
2696
2697         return sret;
2698 }
2699
2700 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
2701                                      struct pipe_buffer *buf)
2702 {
2703         __free_page(buf->page);
2704 }
2705
2706 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
2707                                      unsigned int idx)
2708 {
2709         __free_page(spd->pages[idx]);
2710 }
2711
2712 static struct pipe_buf_operations tracing_pipe_buf_ops = {
2713         .can_merge              = 0,
2714         .map                    = generic_pipe_buf_map,
2715         .unmap                  = generic_pipe_buf_unmap,
2716         .confirm                = generic_pipe_buf_confirm,
2717         .release                = tracing_pipe_buf_release,
2718         .steal                  = generic_pipe_buf_steal,
2719         .get                    = generic_pipe_buf_get,
2720 };
2721
2722 static size_t
2723 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
2724 {
2725         size_t count;
2726         int ret;
2727
2728         /* Seq buffer is page-sized, exactly what we need. */
2729         for (;;) {
2730                 count = iter->seq.len;
2731                 ret = print_trace_line(iter);
2732                 count = iter->seq.len - count;
2733                 if (rem < count) {
2734                         rem = 0;
2735                         iter->seq.len -= count;
2736                         break;
2737                 }
2738                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2739                         iter->seq.len -= count;
2740                         break;
2741                 }
2742
2743                 trace_consume(iter);
2744                 rem -= count;
2745                 if (!find_next_entry_inc(iter)) {
2746                         rem = 0;
2747                         iter->ent = NULL;
2748                         break;
2749                 }
2750         }
2751
2752         return rem;
2753 }
2754
2755 static ssize_t tracing_splice_read_pipe(struct file *filp,
2756                                         loff_t *ppos,
2757                                         struct pipe_inode_info *pipe,
2758                                         size_t len,
2759                                         unsigned int flags)
2760 {
2761         struct page *pages[PIPE_BUFFERS];
2762         struct partial_page partial[PIPE_BUFFERS];
2763         struct trace_iterator *iter = filp->private_data;
2764         struct splice_pipe_desc spd = {
2765                 .pages          = pages,
2766                 .partial        = partial,
2767                 .nr_pages       = 0, /* This gets updated below. */
2768                 .flags          = flags,
2769                 .ops            = &tracing_pipe_buf_ops,
2770                 .spd_release    = tracing_spd_release_pipe,
2771         };
2772         static struct tracer *old_tracer;
2773         ssize_t ret;
2774         size_t rem;
2775         unsigned int i;
2776
2777         /* copy the tracer to avoid using a global lock all around */
2778         mutex_lock(&trace_types_lock);
2779         if (unlikely(old_tracer != current_trace && current_trace)) {
2780                 old_tracer = current_trace;
2781                 *iter->trace = *current_trace;
2782         }
2783         mutex_unlock(&trace_types_lock);
2784
2785         mutex_lock(&iter->mutex);
2786
2787         if (iter->trace->splice_read) {
2788                 ret = iter->trace->splice_read(iter, filp,
2789                                                ppos, pipe, len, flags);
2790                 if (ret)
2791                         goto out_err;
2792         }
2793
2794         ret = tracing_wait_pipe(filp);
2795         if (ret <= 0)
2796                 goto out_err;
2797
2798         if (!iter->ent && !find_next_entry_inc(iter)) {
2799                 ret = -EFAULT;
2800                 goto out_err;
2801         }
2802
2803         /* Fill as many pages as possible. */
2804         for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
2805                 pages[i] = alloc_page(GFP_KERNEL);
2806                 if (!pages[i])
2807                         break;
2808
2809                 rem = tracing_fill_pipe_page(rem, iter);
2810
2811                 /* Copy the data into the page, so we can start over. */
2812                 ret = trace_seq_to_buffer(&iter->seq,
2813                                           page_address(pages[i]),
2814                                           iter->seq.len);
2815                 if (ret < 0) {
2816                         __free_page(pages[i]);
2817                         break;
2818                 }
2819                 partial[i].offset = 0;
2820                 partial[i].len = iter->seq.len;
2821
2822                 trace_seq_reset(&iter->seq);
2823         }
2824
2825         mutex_unlock(&iter->mutex);
2826
2827         spd.nr_pages = i;
2828
2829         return splice_to_pipe(pipe, &spd);
2830
2831 out_err:
2832         mutex_unlock(&iter->mutex);
2833
2834         return ret;
2835 }
2836
2837 static ssize_t
2838 tracing_entries_read(struct file *filp, char __user *ubuf,
2839                      size_t cnt, loff_t *ppos)
2840 {
2841         struct trace_array *tr = filp->private_data;
2842         char buf[64];
2843         int r;
2844
2845         r = sprintf(buf, "%lu\n", tr->entries >> 10);
2846         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2847 }
2848
2849 static ssize_t
2850 tracing_entries_write(struct file *filp, const char __user *ubuf,
2851                       size_t cnt, loff_t *ppos)
2852 {
2853         unsigned long val;
2854         char buf[64];
2855         int ret, cpu;
2856
2857         if (cnt >= sizeof(buf))
2858                 return -EINVAL;
2859
2860         if (copy_from_user(&buf, ubuf, cnt))
2861                 return -EFAULT;
2862
2863         buf[cnt] = 0;
2864
2865         ret = strict_strtoul(buf, 10, &val);
2866         if (ret < 0)
2867                 return ret;
2868
2869         /* must have at least 1 entry */
2870         if (!val)
2871                 return -EINVAL;
2872
2873         mutex_lock(&trace_types_lock);
2874
2875         tracing_stop();
2876
2877         /* disable all cpu buffers */
2878         for_each_tracing_cpu(cpu) {
2879                 if (global_trace.data[cpu])
2880                         atomic_inc(&global_trace.data[cpu]->disabled);
2881                 if (max_tr.data[cpu])
2882                         atomic_inc(&max_tr.data[cpu]->disabled);
2883         }
2884
2885         /* value is in KB */
2886         val <<= 10;
2887
2888         if (val != global_trace.entries) {
2889                 ret = ring_buffer_resize(global_trace.buffer, val);
2890                 if (ret < 0) {
2891                         cnt = ret;
2892                         goto out;
2893                 }
2894
2895                 ret = ring_buffer_resize(max_tr.buffer, val);
2896                 if (ret < 0) {
2897                         int r;
2898                         cnt = ret;
2899                         r = ring_buffer_resize(global_trace.buffer,
2900                                                global_trace.entries);
2901                         if (r < 0) {
2902                                 /* AARGH! We are left with different
2903                                  * size max buffer!!!! */
2904                                 WARN_ON(1);
2905                                 tracing_disabled = 1;
2906                         }
2907                         goto out;
2908                 }
2909
2910                 global_trace.entries = val;
2911         }
2912
2913         filp->f_pos += cnt;
2914
2915         /* If check pages failed, return ENOMEM */
2916         if (tracing_disabled)
2917                 cnt = -ENOMEM;
2918  out:
2919         for_each_tracing_cpu(cpu) {
2920                 if (global_trace.data[cpu])
2921                         atomic_dec(&global_trace.data[cpu]->disabled);
2922                 if (max_tr.data[cpu])
2923                         atomic_dec(&max_tr.data[cpu]->disabled);
2924         }
2925
2926         tracing_start();
2927         max_tr.entries = global_trace.entries;
2928         mutex_unlock(&trace_types_lock);
2929
2930         return cnt;
2931 }
2932
2933 static int mark_printk(const char *fmt, ...)
2934 {
2935         int ret;
2936         va_list args;
2937         va_start(args, fmt);
2938         ret = trace_vprintk(0, -1, fmt, args);
2939         va_end(args);
2940         return ret;
2941 }
2942
2943 static ssize_t
2944 tracing_mark_write(struct file *filp, const char __user *ubuf,
2945                                         size_t cnt, loff_t *fpos)
2946 {
2947         char *buf;
2948         char *end;
2949
2950         if (tracing_disabled)
2951                 return -EINVAL;
2952
2953         if (cnt > TRACE_BUF_SIZE)
2954                 cnt = TRACE_BUF_SIZE;
2955
2956         buf = kmalloc(cnt + 1, GFP_KERNEL);
2957         if (buf == NULL)
2958                 return -ENOMEM;
2959
2960         if (copy_from_user(buf, ubuf, cnt)) {
2961                 kfree(buf);
2962                 return -EFAULT;
2963         }
2964
2965         /* Cut from the first nil or newline. */
2966         buf[cnt] = '\0';
2967         end = strchr(buf, '\n');
2968         if (end)
2969                 *end = '\0';
2970
2971         cnt = mark_printk("%s\n", buf);
2972         kfree(buf);
2973         *fpos += cnt;
2974
2975         return cnt;
2976 }
2977
2978 static struct file_operations tracing_max_lat_fops = {
2979         .open           = tracing_open_generic,
2980         .read           = tracing_max_lat_read,
2981         .write          = tracing_max_lat_write,
2982 };
2983
2984 static struct file_operations tracing_ctrl_fops = {
2985         .open           = tracing_open_generic,
2986         .read           = tracing_ctrl_read,
2987         .write          = tracing_ctrl_write,
2988 };
2989
2990 static struct file_operations set_tracer_fops = {
2991         .open           = tracing_open_generic,
2992         .read           = tracing_set_trace_read,
2993         .write          = tracing_set_trace_write,
2994 };
2995
2996 static struct file_operations tracing_pipe_fops = {
2997         .open           = tracing_open_pipe,
2998         .poll           = tracing_poll_pipe,
2999         .read           = tracing_read_pipe,
3000         .splice_read    = tracing_splice_read_pipe,
3001         .release        = tracing_release_pipe,
3002 };
3003
3004 static struct file_operations tracing_entries_fops = {
3005         .open           = tracing_open_generic,
3006         .read           = tracing_entries_read,
3007         .write          = tracing_entries_write,
3008 };
3009
3010 static struct file_operations tracing_mark_fops = {
3011         .open           = tracing_open_generic,
3012         .write          = tracing_mark_write,
3013 };
3014
3015 #ifdef CONFIG_DYNAMIC_FTRACE
3016
3017 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3018 {
3019         return 0;
3020 }
3021
3022 static ssize_t
3023 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3024                   size_t cnt, loff_t *ppos)
3025 {
3026         static char ftrace_dyn_info_buffer[1024];
3027         static DEFINE_MUTEX(dyn_info_mutex);
3028         unsigned long *p = filp->private_data;
3029         char *buf = ftrace_dyn_info_buffer;
3030         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3031         int r;
3032
3033         mutex_lock(&dyn_info_mutex);
3034         r = sprintf(buf, "%ld ", *p);
3035
3036         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3037         buf[r++] = '\n';
3038
3039         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3040
3041         mutex_unlock(&dyn_info_mutex);
3042
3043         return r;
3044 }
3045
3046 static struct file_operations tracing_dyn_info_fops = {
3047         .open           = tracing_open_generic,
3048         .read           = tracing_read_dyn_info,
3049 };
3050 #endif
3051
3052 static struct dentry *d_tracer;
3053
3054 struct dentry *tracing_init_dentry(void)
3055 {
3056         static int once;
3057
3058         if (d_tracer)
3059                 return d_tracer;
3060
3061         d_tracer = debugfs_create_dir("tracing", NULL);
3062
3063         if (!d_tracer && !once) {
3064                 once = 1;
3065                 pr_warning("Could not create debugfs directory 'tracing'\n");
3066                 return NULL;
3067         }
3068
3069         return d_tracer;
3070 }
3071
3072 static struct dentry *d_percpu;
3073
3074 struct dentry *tracing_dentry_percpu(void)
3075 {
3076         static int once;
3077         struct dentry *d_tracer;
3078
3079         if (d_percpu)
3080                 return d_percpu;
3081
3082         d_tracer = tracing_init_dentry();
3083
3084         if (!d_tracer)
3085                 return NULL;
3086
3087         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3088
3089         if (!d_percpu && !once) {
3090                 once = 1;
3091                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3092                 return NULL;
3093         }
3094
3095         return d_percpu;
3096 }
3097
3098 static void tracing_init_debugfs_percpu(long cpu)
3099 {
3100         struct dentry *d_percpu = tracing_dentry_percpu();
3101         struct dentry *entry, *d_cpu;
3102         /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3103         char cpu_dir[7];
3104
3105         if (cpu > 999 || cpu < 0)
3106                 return;
3107
3108         sprintf(cpu_dir, "cpu%ld", cpu);
3109         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3110         if (!d_cpu) {
3111                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3112                 return;
3113         }
3114
3115         /* per cpu trace_pipe */
3116         entry = debugfs_create_file("trace_pipe", 0444, d_cpu,
3117                                 (void *) cpu, &tracing_pipe_fops);
3118         if (!entry)
3119                 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3120
3121         /* per cpu trace */
3122         entry = debugfs_create_file("trace", 0444, d_cpu,
3123                                 (void *) cpu, &tracing_fops);
3124         if (!entry)
3125                 pr_warning("Could not create debugfs 'trace' entry\n");
3126 }
3127
3128 #ifdef CONFIG_FTRACE_SELFTEST
3129 /* Let selftest have access to static functions in this file */
3130 #include "trace_selftest.c"
3131 #endif
3132
3133 struct trace_option_dentry {
3134         struct tracer_opt               *opt;
3135         struct tracer_flags             *flags;
3136         struct dentry                   *entry;
3137 };
3138
3139 static ssize_t
3140 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3141                         loff_t *ppos)
3142 {
3143         struct trace_option_dentry *topt = filp->private_data;
3144         char *buf;
3145
3146         if (topt->flags->val & topt->opt->bit)
3147                 buf = "1\n";
3148         else
3149                 buf = "0\n";
3150
3151         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3152 }
3153
3154 static ssize_t
3155 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3156                          loff_t *ppos)
3157 {
3158         struct trace_option_dentry *topt = filp->private_data;
3159         unsigned long val;
3160         char buf[64];
3161         int ret;
3162
3163         if (cnt >= sizeof(buf))
3164                 return -EINVAL;
3165
3166         if (copy_from_user(&buf, ubuf, cnt))
3167                 return -EFAULT;
3168
3169         buf[cnt] = 0;
3170
3171         ret = strict_strtoul(buf, 10, &val);
3172         if (ret < 0)
3173                 return ret;
3174
3175         ret = 0;
3176         switch (val) {
3177         case 0:
3178                 /* do nothing if already cleared */
3179                 if (!(topt->flags->val & topt->opt->bit))
3180                         break;
3181
3182                 mutex_lock(&trace_types_lock);
3183                 if (current_trace->set_flag)
3184                         ret = current_trace->set_flag(topt->flags->val,
3185                                                       topt->opt->bit, 0);
3186                 mutex_unlock(&trace_types_lock);
3187                 if (ret)
3188                         return ret;
3189                 topt->flags->val &= ~topt->opt->bit;
3190                 break;
3191         case 1:
3192                 /* do nothing if already set */
3193                 if (topt->flags->val & topt->opt->bit)
3194                         break;
3195
3196                 mutex_lock(&trace_types_lock);
3197                 if (current_trace->set_flag)
3198                         ret = current_trace->set_flag(topt->flags->val,
3199                                                       topt->opt->bit, 1);
3200                 mutex_unlock(&trace_types_lock);
3201                 if (ret)
3202                         return ret;
3203                 topt->flags->val |= topt->opt->bit;
3204                 break;
3205
3206         default:
3207                 return -EINVAL;
3208         }
3209
3210         *ppos += cnt;
3211
3212         return cnt;
3213 }
3214
3215
3216 static const struct file_operations trace_options_fops = {
3217         .open = tracing_open_generic,
3218         .read = trace_options_read,
3219         .write = trace_options_write,
3220 };
3221
3222 static ssize_t
3223 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3224                         loff_t *ppos)
3225 {
3226         long index = (long)filp->private_data;
3227         char *buf;
3228
3229         if (trace_flags & (1 << index))
3230                 buf = "1\n";
3231         else
3232                 buf = "0\n";
3233
3234         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3235 }
3236
3237 static ssize_t
3238 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3239                          loff_t *ppos)
3240 {
3241         long index = (long)filp->private_data;
3242         char buf[64];
3243         unsigned long val;
3244         int ret;
3245
3246         if (cnt >= sizeof(buf))
3247                 return -EINVAL;
3248
3249         if (copy_from_user(&buf, ubuf, cnt))
3250                 return -EFAULT;
3251
3252         buf[cnt] = 0;
3253
3254         ret = strict_strtoul(buf, 10, &val);
3255         if (ret < 0)
3256                 return ret;
3257
3258         switch (val) {
3259         case 0:
3260                 trace_flags &= ~(1 << index);
3261                 break;
3262         case 1:
3263                 trace_flags |= 1 << index;
3264                 break;
3265
3266         default:
3267                 return -EINVAL;
3268         }
3269
3270         *ppos += cnt;
3271
3272         return cnt;
3273 }
3274
3275 static const struct file_operations trace_options_core_fops = {
3276         .open = tracing_open_generic,
3277         .read = trace_options_core_read,
3278         .write = trace_options_core_write,
3279 };
3280
3281 static struct dentry *trace_options_init_dentry(void)
3282 {
3283         struct dentry *d_tracer;
3284         static struct dentry *t_options;
3285
3286         if (t_options)
3287                 return t_options;
3288
3289         d_tracer = tracing_init_dentry();
3290         if (!d_tracer)
3291                 return NULL;
3292
3293         t_options = debugfs_create_dir("options", d_tracer);
3294         if (!t_options) {
3295                 pr_warning("Could not create debugfs directory 'options'\n");
3296                 return NULL;
3297         }
3298
3299         return t_options;
3300 }
3301
3302 static void
3303 create_trace_option_file(struct trace_option_dentry *topt,
3304                          struct tracer_flags *flags,
3305                          struct tracer_opt *opt)
3306 {
3307         struct dentry *t_options;
3308         struct dentry *entry;
3309
3310         t_options = trace_options_init_dentry();
3311         if (!t_options)
3312                 return;
3313
3314         topt->flags = flags;
3315         topt->opt = opt;
3316
3317         entry = debugfs_create_file(opt->name, 0644, t_options, topt,
3318                                     &trace_options_fops);
3319
3320         topt->entry = entry;
3321
3322 }
3323
3324 static struct trace_option_dentry *
3325 create_trace_option_files(struct tracer *tracer)
3326 {
3327         struct trace_option_dentry *topts;
3328         struct tracer_flags *flags;
3329         struct tracer_opt *opts;
3330         int cnt;
3331
3332         if (!tracer)
3333                 return NULL;
3334
3335         flags = tracer->flags;
3336
3337         if (!flags || !flags->opts)
3338                 return NULL;
3339
3340         opts = flags->opts;
3341
3342         for (cnt = 0; opts[cnt].name; cnt++)
3343                 ;
3344
3345         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3346         if (!topts)
3347                 return NULL;
3348
3349         for (cnt = 0; opts[cnt].name; cnt++)
3350                 create_trace_option_file(&topts[cnt], flags,
3351                                          &opts[cnt]);
3352
3353         return topts;
3354 }
3355
3356 static void
3357 destroy_trace_option_files(struct trace_option_dentry *topts)
3358 {
3359         int cnt;
3360
3361         if (!topts)
3362                 return;
3363
3364         for (cnt = 0; topts[cnt].opt; cnt++) {
3365                 if (topts[cnt].entry)
3366                         debugfs_remove(topts[cnt].entry);
3367         }
3368
3369         kfree(topts);
3370 }
3371
3372 static struct dentry *
3373 create_trace_option_core_file(const char *option, long index)
3374 {
3375         struct dentry *t_options;
3376         struct dentry *entry;
3377
3378         t_options = trace_options_init_dentry();
3379         if (!t_options)
3380                 return NULL;
3381
3382         entry = debugfs_create_file(option, 0644, t_options, (void *)index,
3383                                     &trace_options_core_fops);
3384
3385         return entry;
3386 }
3387
3388 static __init void create_trace_options_dir(void)
3389 {
3390         struct dentry *t_options;
3391         struct dentry *entry;
3392         int i;
3393
3394         t_options = trace_options_init_dentry();
3395         if (!t_options)
3396                 return;
3397
3398         for (i = 0; trace_options[i]; i++) {
3399                 entry = create_trace_option_core_file(trace_options[i], i);
3400                 if (!entry)
3401                         pr_warning("Could not create debugfs %s entry\n",
3402                                    trace_options[i]);
3403         }
3404 }
3405
3406 static __init int tracer_init_debugfs(void)
3407 {
3408         struct dentry *d_tracer;
3409         struct dentry *entry;
3410         int cpu;
3411
3412         d_tracer = tracing_init_dentry();
3413
3414         entry = debugfs_create_file("tracing_enabled", 0644, d_tracer,
3415                                     &global_trace, &tracing_ctrl_fops);
3416         if (!entry)
3417                 pr_warning("Could not create debugfs 'tracing_enabled' entry\n");
3418
3419         entry = debugfs_create_file("trace_options", 0644, d_tracer,
3420                                     NULL, &tracing_iter_fops);
3421         if (!entry)
3422                 pr_warning("Could not create debugfs 'trace_options' entry\n");
3423
3424         create_trace_options_dir();
3425
3426         entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3427                                     NULL, &tracing_cpumask_fops);
3428         if (!entry)
3429                 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3430
3431         entry = debugfs_create_file("trace", 0444, d_tracer,
3432                                  (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
3433         if (!entry)
3434                 pr_warning("Could not create debugfs 'trace' entry\n");
3435
3436         entry = debugfs_create_file("available_tracers", 0444, d_tracer,
3437                                     &global_trace, &show_traces_fops);
3438         if (!entry)
3439                 pr_warning("Could not create debugfs 'available_tracers' entry\n");
3440
3441         entry = debugfs_create_file("current_tracer", 0444, d_tracer,
3442                                     &global_trace, &set_tracer_fops);
3443         if (!entry)
3444                 pr_warning("Could not create debugfs 'current_tracer' entry\n");
3445
3446         entry = debugfs_create_file("tracing_max_latency", 0644, d_tracer,
3447                                     &tracing_max_latency,
3448                                     &tracing_max_lat_fops);
3449         if (!entry)
3450                 pr_warning("Could not create debugfs "
3451                            "'tracing_max_latency' entry\n");
3452
3453         entry = debugfs_create_file("tracing_thresh", 0644, d_tracer,
3454                                     &tracing_thresh, &tracing_max_lat_fops);
3455         if (!entry)
3456                 pr_warning("Could not create debugfs "
3457                            "'tracing_thresh' entry\n");
3458         entry = debugfs_create_file("README", 0644, d_tracer,
3459                                     NULL, &tracing_readme_fops);
3460         if (!entry)
3461                 pr_warning("Could not create debugfs 'README' entry\n");
3462
3463         entry = debugfs_create_file("trace_pipe", 0444, d_tracer,
3464                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
3465         if (!entry)
3466                 pr_warning("Could not create debugfs "
3467                            "'trace_pipe' entry\n");
3468
3469         entry = debugfs_create_file("buffer_size_kb", 0644, d_tracer,
3470                                     &global_trace, &tracing_entries_fops);
3471         if (!entry)
3472                 pr_warning("Could not create debugfs "
3473                            "'buffer_size_kb' entry\n");
3474
3475         entry = debugfs_create_file("trace_marker", 0220, d_tracer,
3476                                     NULL, &tracing_mark_fops);
3477         if (!entry)
3478                 pr_warning("Could not create debugfs "
3479                            "'trace_marker' entry\n");
3480
3481 #ifdef CONFIG_DYNAMIC_FTRACE
3482         entry = debugfs_create_file("dyn_ftrace_total_info", 0444, d_tracer,
3483                                     &ftrace_update_tot_cnt,
3484                                     &tracing_dyn_info_fops);
3485         if (!entry)
3486                 pr_warning("Could not create debugfs "
3487                            "'dyn_ftrace_total_info' entry\n");
3488 #endif
3489 #ifdef CONFIG_SYSPROF_TRACER
3490         init_tracer_sysprof_debugfs(d_tracer);
3491 #endif
3492
3493         for_each_tracing_cpu(cpu)
3494                 tracing_init_debugfs_percpu(cpu);
3495
3496         return 0;
3497 }
3498
3499 int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
3500 {
3501         static DEFINE_SPINLOCK(trace_buf_lock);
3502         static char trace_buf[TRACE_BUF_SIZE];
3503
3504         struct ring_buffer_event *event;
3505         struct trace_array *tr = &global_trace;
3506         struct trace_array_cpu *data;
3507         int cpu, len = 0, size, pc;
3508         struct print_entry *entry;
3509         unsigned long irq_flags;
3510
3511         if (tracing_disabled || tracing_selftest_running)
3512                 return 0;
3513
3514         pc = preempt_count();
3515         preempt_disable_notrace();
3516         cpu = raw_smp_processor_id();
3517         data = tr->data[cpu];
3518
3519         if (unlikely(atomic_read(&data->disabled)))
3520                 goto out;
3521
3522         pause_graph_tracing();
3523         spin_lock_irqsave(&trace_buf_lock, irq_flags);
3524         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3525
3526         len = min(len, TRACE_BUF_SIZE-1);
3527         trace_buf[len] = 0;
3528
3529         size = sizeof(*entry) + len + 1;
3530         event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
3531         if (!event)
3532                 goto out_unlock;
3533         entry = ring_buffer_event_data(event);
3534         entry->ip                       = ip;
3535         entry->depth                    = depth;
3536
3537         memcpy(&entry->buf, trace_buf, len);
3538         entry->buf[len] = 0;
3539         ring_buffer_unlock_commit(tr->buffer, event);
3540
3541  out_unlock:
3542         spin_unlock_irqrestore(&trace_buf_lock, irq_flags);
3543         unpause_graph_tracing();
3544  out:
3545         preempt_enable_notrace();
3546
3547         return len;
3548 }
3549 EXPORT_SYMBOL_GPL(trace_vprintk);
3550
3551 int __ftrace_printk(unsigned long ip, const char *fmt, ...)
3552 {
3553         int ret;
3554         va_list ap;
3555
3556         if (!(trace_flags & TRACE_ITER_PRINTK))
3557                 return 0;
3558
3559         va_start(ap, fmt);
3560         ret = trace_vprintk(ip, task_curr_ret_stack(current), fmt, ap);
3561         va_end(ap);
3562         return ret;
3563 }
3564 EXPORT_SYMBOL_GPL(__ftrace_printk);
3565
3566 int __ftrace_vprintk(unsigned long ip, const char *fmt, va_list ap)
3567 {
3568         if (!(trace_flags & TRACE_ITER_PRINTK))
3569                 return 0;
3570
3571         return trace_vprintk(ip, task_curr_ret_stack(current), fmt, ap);
3572 }
3573 EXPORT_SYMBOL_GPL(__ftrace_vprintk);
3574
3575 static int trace_panic_handler(struct notifier_block *this,
3576                                unsigned long event, void *unused)
3577 {
3578         if (ftrace_dump_on_oops)
3579                 ftrace_dump();
3580         return NOTIFY_OK;
3581 }
3582
3583 static struct notifier_block trace_panic_notifier = {
3584         .notifier_call  = trace_panic_handler,
3585         .next           = NULL,
3586         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
3587 };
3588
3589 static int trace_die_handler(struct notifier_block *self,
3590                              unsigned long val,
3591                              void *data)
3592 {
3593         switch (val) {
3594         case DIE_OOPS:
3595                 if (ftrace_dump_on_oops)
3596                         ftrace_dump();
3597                 break;
3598         default:
3599                 break;
3600         }
3601         return NOTIFY_OK;
3602 }
3603
3604 static struct notifier_block trace_die_notifier = {
3605         .notifier_call = trace_die_handler,
3606         .priority = 200
3607 };
3608
3609 /*
3610  * printk is set to max of 1024, we really don't need it that big.
3611  * Nothing should be printing 1000 characters anyway.
3612  */
3613 #define TRACE_MAX_PRINT         1000
3614
3615 /*
3616  * Define here KERN_TRACE so that we have one place to modify
3617  * it if we decide to change what log level the ftrace dump
3618  * should be at.
3619  */
3620 #define KERN_TRACE              KERN_EMERG
3621
3622 static void
3623 trace_printk_seq(struct trace_seq *s)
3624 {
3625         /* Probably should print a warning here. */
3626         if (s->len >= 1000)
3627                 s->len = 1000;
3628
3629         /* should be zero ended, but we are paranoid. */
3630         s->buffer[s->len] = 0;
3631
3632         printk(KERN_TRACE "%s", s->buffer);
3633
3634         trace_seq_reset(s);
3635 }
3636
3637 void ftrace_dump(void)
3638 {
3639         static DEFINE_SPINLOCK(ftrace_dump_lock);
3640         /* use static because iter can be a bit big for the stack */
3641         static struct trace_iterator iter;
3642         static int dump_ran;
3643         unsigned long flags;
3644         int cnt = 0, cpu;
3645
3646         /* only one dump */
3647         spin_lock_irqsave(&ftrace_dump_lock, flags);
3648         if (dump_ran)
3649                 goto out;
3650
3651         dump_ran = 1;
3652
3653         /* No turning back! */
3654         tracing_off();
3655         ftrace_kill();
3656
3657         for_each_tracing_cpu(cpu) {
3658                 atomic_inc(&global_trace.data[cpu]->disabled);
3659         }
3660
3661         /* don't look at user memory in panic mode */
3662         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3663
3664         printk(KERN_TRACE "Dumping ftrace buffer:\n");
3665
3666         iter.tr = &global_trace;
3667         iter.trace = current_trace;
3668
3669         /*
3670          * We need to stop all tracing on all CPUS to read the
3671          * the next buffer. This is a bit expensive, but is
3672          * not done often. We fill all what we can read,
3673          * and then release the locks again.
3674          */
3675
3676         while (!trace_empty(&iter)) {
3677
3678                 if (!cnt)
3679                         printk(KERN_TRACE "---------------------------------\n");
3680
3681                 cnt++;
3682
3683                 /* reset all but tr, trace, and overruns */
3684                 memset(&iter.seq, 0,
3685                        sizeof(struct trace_iterator) -
3686                        offsetof(struct trace_iterator, seq));
3687                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
3688                 iter.pos = -1;
3689
3690                 if (find_next_entry_inc(&iter) != NULL) {
3691                         print_trace_line(&iter);
3692                         trace_consume(&iter);
3693                 }
3694
3695                 trace_printk_seq(&iter.seq);
3696         }
3697
3698         if (!cnt)
3699                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
3700         else
3701                 printk(KERN_TRACE "---------------------------------\n");
3702
3703  out:
3704         spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3705 }
3706
3707 __init static int tracer_alloc_buffers(void)
3708 {
3709         struct trace_array_cpu *data;
3710         int i;
3711         int ret = -ENOMEM;
3712
3713         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
3714                 goto out;
3715
3716         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
3717                 goto out_free_buffer_mask;
3718
3719         if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
3720                 goto out_free_tracing_cpumask;
3721
3722         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
3723         cpumask_copy(tracing_cpumask, cpu_all_mask);
3724         cpumask_clear(tracing_reader_cpumask);
3725
3726         /* TODO: make the number of buffers hot pluggable with CPUS */
3727         global_trace.buffer = ring_buffer_alloc(trace_buf_size,
3728                                                    TRACE_BUFFER_FLAGS);
3729         if (!global_trace.buffer) {
3730                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
3731                 WARN_ON(1);
3732                 goto out_free_cpumask;
3733         }
3734         global_trace.entries = ring_buffer_size(global_trace.buffer);
3735
3736
3737 #ifdef CONFIG_TRACER_MAX_TRACE
3738         max_tr.buffer = ring_buffer_alloc(trace_buf_size,
3739                                              TRACE_BUFFER_FLAGS);
3740         if (!max_tr.buffer) {
3741                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
3742                 WARN_ON(1);
3743                 ring_buffer_free(global_trace.buffer);
3744                 goto out_free_cpumask;
3745         }
3746         max_tr.entries = ring_buffer_size(max_tr.buffer);
3747         WARN_ON(max_tr.entries != global_trace.entries);
3748 #endif
3749
3750         /* Allocate the first page for all buffers */
3751         for_each_tracing_cpu(i) {
3752                 data = global_trace.data[i] = &per_cpu(global_trace_cpu, i);
3753                 max_tr.data[i] = &per_cpu(max_data, i);
3754         }
3755
3756         trace_init_cmdlines();
3757
3758         register_tracer(&nop_trace);
3759         current_trace = &nop_trace;
3760 #ifdef CONFIG_BOOT_TRACER
3761         register_tracer(&boot_tracer);
3762 #endif
3763         /* All seems OK, enable tracing */
3764         tracing_disabled = 0;
3765
3766         atomic_notifier_chain_register(&panic_notifier_list,
3767                                        &trace_panic_notifier);
3768
3769         register_die_notifier(&trace_die_notifier);
3770         ret = 0;
3771
3772 out_free_cpumask:
3773         free_cpumask_var(tracing_reader_cpumask);
3774 out_free_tracing_cpumask:
3775         free_cpumask_var(tracing_cpumask);
3776 out_free_buffer_mask:
3777         free_cpumask_var(tracing_buffer_mask);
3778 out:
3779         return ret;
3780 }
3781
3782 __init static int clear_boot_tracer(void)
3783 {
3784         /*
3785          * The default tracer at boot buffer is an init section.
3786          * This function is called in lateinit. If we did not
3787          * find the boot tracer, then clear it out, to prevent
3788          * later registration from accessing the buffer that is
3789          * about to be freed.
3790          */
3791         if (!default_bootup_tracer)
3792                 return 0;
3793
3794         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
3795                default_bootup_tracer);
3796         default_bootup_tracer = NULL;
3797
3798         return 0;
3799 }
3800
3801 early_initcall(tracer_alloc_buffers);
3802 fs_initcall(tracer_init_debugfs);
3803 late_initcall(clear_boot_tracer);