Merge branch 'tip/perf/core' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
[linux-2.6.git] / kernel / trace / trace.c
1 /*
2  * ring buffer based function tracer
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally taken from the RT patch by:
8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code from the latency_tracer, that is:
11  *  Copyright (C) 2004-2006 Ingo Molnar
12  *  Copyright (C) 2004 William Lee Irwin III
13  */
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/rwsem.h>
36 #include <linux/slab.h>
37 #include <linux/ctype.h>
38 #include <linux/init.h>
39 #include <linux/poll.h>
40 #include <linux/fs.h>
41
42 #include "trace.h"
43 #include "trace_output.h"
44
45 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
46
47 /*
48  * On boot up, the ring buffer is set to the minimum size, so that
49  * we do not waste memory on systems that are not using tracing.
50  */
51 int ring_buffer_expanded;
52
53 /*
54  * We need to change this state when a selftest is running.
55  * A selftest will lurk into the ring-buffer to count the
56  * entries inserted during the selftest although some concurrent
57  * insertions into the ring-buffer such as trace_printk could occurred
58  * at the same time, giving false positive or negative results.
59  */
60 static bool __read_mostly tracing_selftest_running;
61
62 /*
63  * If a tracer is running, we do not want to run SELFTEST.
64  */
65 bool __read_mostly tracing_selftest_disabled;
66
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt[] = {
69         { }
70 };
71
72 static struct tracer_flags dummy_tracer_flags = {
73         .val = 0,
74         .opts = dummy_tracer_opt
75 };
76
77 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
78 {
79         return 0;
80 }
81
82 /*
83  * Kill all tracing for good (never come back).
84  * It is initialized to 1 but will turn to zero if the initialization
85  * of the tracer is successful. But that is the only place that sets
86  * this back to zero.
87  */
88 static int tracing_disabled = 1;
89
90 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
91
92 static inline void ftrace_disable_cpu(void)
93 {
94         preempt_disable();
95         __this_cpu_inc(ftrace_cpu_disabled);
96 }
97
98 static inline void ftrace_enable_cpu(void)
99 {
100         __this_cpu_dec(ftrace_cpu_disabled);
101         preempt_enable();
102 }
103
104 static cpumask_var_t __read_mostly      tracing_buffer_mask;
105
106 #define for_each_tracing_cpu(cpu)       \
107         for_each_cpu(cpu, tracing_buffer_mask)
108
109 /*
110  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
111  *
112  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
113  * is set, then ftrace_dump is called. This will output the contents
114  * of the ftrace buffers to the console.  This is very useful for
115  * capturing traces that lead to crashes and outputing it to a
116  * serial console.
117  *
118  * It is default off, but you can enable it with either specifying
119  * "ftrace_dump_on_oops" in the kernel command line, or setting
120  * /proc/sys/kernel/ftrace_dump_on_oops
121  * Set 1 if you want to dump buffers of all CPUs
122  * Set 2 if you want to dump the buffer of the CPU that triggered oops
123  */
124
125 enum ftrace_dump_mode ftrace_dump_on_oops;
126
127 static int tracing_set_tracer(const char *buf);
128
129 #define MAX_TRACER_SIZE         100
130 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
131 static char *default_bootup_tracer;
132
133 static int __init set_cmdline_ftrace(char *str)
134 {
135         strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
136         default_bootup_tracer = bootup_tracer_buf;
137         /* We are using ftrace early, expand it */
138         ring_buffer_expanded = 1;
139         return 1;
140 }
141 __setup("ftrace=", set_cmdline_ftrace);
142
143 static int __init set_ftrace_dump_on_oops(char *str)
144 {
145         if (*str++ != '=' || !*str) {
146                 ftrace_dump_on_oops = DUMP_ALL;
147                 return 1;
148         }
149
150         if (!strcmp("orig_cpu", str)) {
151                 ftrace_dump_on_oops = DUMP_ORIG;
152                 return 1;
153         }
154
155         return 0;
156 }
157 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
158
159 unsigned long long ns2usecs(cycle_t nsec)
160 {
161         nsec += 500;
162         do_div(nsec, 1000);
163         return nsec;
164 }
165
166 /*
167  * The global_trace is the descriptor that holds the tracing
168  * buffers for the live tracing. For each CPU, it contains
169  * a link list of pages that will store trace entries. The
170  * page descriptor of the pages in the memory is used to hold
171  * the link list by linking the lru item in the page descriptor
172  * to each of the pages in the buffer per CPU.
173  *
174  * For each active CPU there is a data field that holds the
175  * pages for the buffer for that CPU. Each CPU has the same number
176  * of pages allocated for its buffer.
177  */
178 static struct trace_array       global_trace;
179
180 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
181
182 int filter_current_check_discard(struct ring_buffer *buffer,
183                                  struct ftrace_event_call *call, void *rec,
184                                  struct ring_buffer_event *event)
185 {
186         return filter_check_discard(call, rec, buffer, event);
187 }
188 EXPORT_SYMBOL_GPL(filter_current_check_discard);
189
190 cycle_t ftrace_now(int cpu)
191 {
192         u64 ts;
193
194         /* Early boot up does not have a buffer yet */
195         if (!global_trace.buffer)
196                 return trace_clock_local();
197
198         ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
199         ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
200
201         return ts;
202 }
203
204 /*
205  * The max_tr is used to snapshot the global_trace when a maximum
206  * latency is reached. Some tracers will use this to store a maximum
207  * trace while it continues examining live traces.
208  *
209  * The buffers for the max_tr are set up the same as the global_trace.
210  * When a snapshot is taken, the link list of the max_tr is swapped
211  * with the link list of the global_trace and the buffers are reset for
212  * the global_trace so the tracing can continue.
213  */
214 static struct trace_array       max_tr;
215
216 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
217
218 /* tracer_enabled is used to toggle activation of a tracer */
219 static int                      tracer_enabled = 1;
220
221 /**
222  * tracing_is_enabled - return tracer_enabled status
223  *
224  * This function is used by other tracers to know the status
225  * of the tracer_enabled flag.  Tracers may use this function
226  * to know if it should enable their features when starting
227  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
228  */
229 int tracing_is_enabled(void)
230 {
231         return tracer_enabled;
232 }
233
234 /*
235  * trace_buf_size is the size in bytes that is allocated
236  * for a buffer. Note, the number of bytes is always rounded
237  * to page size.
238  *
239  * This number is purposely set to a low number of 16384.
240  * If the dump on oops happens, it will be much appreciated
241  * to not have to wait for all that output. Anyway this can be
242  * boot time and run time configurable.
243  */
244 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
245
246 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
247
248 /* trace_types holds a link list of available tracers. */
249 static struct tracer            *trace_types __read_mostly;
250
251 /* current_trace points to the tracer that is currently active */
252 static struct tracer            *current_trace __read_mostly;
253
254 /*
255  * trace_types_lock is used to protect the trace_types list.
256  */
257 static DEFINE_MUTEX(trace_types_lock);
258
259 /*
260  * serialize the access of the ring buffer
261  *
262  * ring buffer serializes readers, but it is low level protection.
263  * The validity of the events (which returns by ring_buffer_peek() ..etc)
264  * are not protected by ring buffer.
265  *
266  * The content of events may become garbage if we allow other process consumes
267  * these events concurrently:
268  *   A) the page of the consumed events may become a normal page
269  *      (not reader page) in ring buffer, and this page will be rewrited
270  *      by events producer.
271  *   B) The page of the consumed events may become a page for splice_read,
272  *      and this page will be returned to system.
273  *
274  * These primitives allow multi process access to different cpu ring buffer
275  * concurrently.
276  *
277  * These primitives don't distinguish read-only and read-consume access.
278  * Multi read-only access are also serialized.
279  */
280
281 #ifdef CONFIG_SMP
282 static DECLARE_RWSEM(all_cpu_access_lock);
283 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
284
285 static inline void trace_access_lock(int cpu)
286 {
287         if (cpu == TRACE_PIPE_ALL_CPU) {
288                 /* gain it for accessing the whole ring buffer. */
289                 down_write(&all_cpu_access_lock);
290         } else {
291                 /* gain it for accessing a cpu ring buffer. */
292
293                 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
294                 down_read(&all_cpu_access_lock);
295
296                 /* Secondly block other access to this @cpu ring buffer. */
297                 mutex_lock(&per_cpu(cpu_access_lock, cpu));
298         }
299 }
300
301 static inline void trace_access_unlock(int cpu)
302 {
303         if (cpu == TRACE_PIPE_ALL_CPU) {
304                 up_write(&all_cpu_access_lock);
305         } else {
306                 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
307                 up_read(&all_cpu_access_lock);
308         }
309 }
310
311 static inline void trace_access_lock_init(void)
312 {
313         int cpu;
314
315         for_each_possible_cpu(cpu)
316                 mutex_init(&per_cpu(cpu_access_lock, cpu));
317 }
318
319 #else
320
321 static DEFINE_MUTEX(access_lock);
322
323 static inline void trace_access_lock(int cpu)
324 {
325         (void)cpu;
326         mutex_lock(&access_lock);
327 }
328
329 static inline void trace_access_unlock(int cpu)
330 {
331         (void)cpu;
332         mutex_unlock(&access_lock);
333 }
334
335 static inline void trace_access_lock_init(void)
336 {
337 }
338
339 #endif
340
341 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
342 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
343
344 /* trace_flags holds trace_options default values */
345 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
346         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
347         TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD;
348
349 static int trace_stop_count;
350 static DEFINE_SPINLOCK(tracing_start_lock);
351
352 /**
353  * trace_wake_up - wake up tasks waiting for trace input
354  *
355  * Simply wakes up any task that is blocked on the trace_wait
356  * queue. These is used with trace_poll for tasks polling the trace.
357  */
358 void trace_wake_up(void)
359 {
360         int cpu;
361
362         if (trace_flags & TRACE_ITER_BLOCK)
363                 return;
364         /*
365          * The runqueue_is_locked() can fail, but this is the best we
366          * have for now:
367          */
368         cpu = get_cpu();
369         if (!runqueue_is_locked(cpu))
370                 wake_up(&trace_wait);
371         put_cpu();
372 }
373
374 static int __init set_buf_size(char *str)
375 {
376         unsigned long buf_size;
377
378         if (!str)
379                 return 0;
380         buf_size = memparse(str, &str);
381         /* nr_entries can not be zero */
382         if (buf_size == 0)
383                 return 0;
384         trace_buf_size = buf_size;
385         return 1;
386 }
387 __setup("trace_buf_size=", set_buf_size);
388
389 static int __init set_tracing_thresh(char *str)
390 {
391         unsigned long threshhold;
392         int ret;
393
394         if (!str)
395                 return 0;
396         ret = strict_strtoul(str, 0, &threshhold);
397         if (ret < 0)
398                 return 0;
399         tracing_thresh = threshhold * 1000;
400         return 1;
401 }
402 __setup("tracing_thresh=", set_tracing_thresh);
403
404 unsigned long nsecs_to_usecs(unsigned long nsecs)
405 {
406         return nsecs / 1000;
407 }
408
409 /* These must match the bit postions in trace_iterator_flags */
410 static const char *trace_options[] = {
411         "print-parent",
412         "sym-offset",
413         "sym-addr",
414         "verbose",
415         "raw",
416         "hex",
417         "bin",
418         "block",
419         "stacktrace",
420         "trace_printk",
421         "ftrace_preempt",
422         "branch",
423         "annotate",
424         "userstacktrace",
425         "sym-userobj",
426         "printk-msg-only",
427         "context-info",
428         "latency-format",
429         "sleep-time",
430         "graph-time",
431         "record-cmd",
432         NULL
433 };
434
435 static struct {
436         u64 (*func)(void);
437         const char *name;
438 } trace_clocks[] = {
439         { trace_clock_local,    "local" },
440         { trace_clock_global,   "global" },
441 };
442
443 int trace_clock_id;
444
445 /*
446  * trace_parser_get_init - gets the buffer for trace parser
447  */
448 int trace_parser_get_init(struct trace_parser *parser, int size)
449 {
450         memset(parser, 0, sizeof(*parser));
451
452         parser->buffer = kmalloc(size, GFP_KERNEL);
453         if (!parser->buffer)
454                 return 1;
455
456         parser->size = size;
457         return 0;
458 }
459
460 /*
461  * trace_parser_put - frees the buffer for trace parser
462  */
463 void trace_parser_put(struct trace_parser *parser)
464 {
465         kfree(parser->buffer);
466 }
467
468 /*
469  * trace_get_user - reads the user input string separated by  space
470  * (matched by isspace(ch))
471  *
472  * For each string found the 'struct trace_parser' is updated,
473  * and the function returns.
474  *
475  * Returns number of bytes read.
476  *
477  * See kernel/trace/trace.h for 'struct trace_parser' details.
478  */
479 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
480         size_t cnt, loff_t *ppos)
481 {
482         char ch;
483         size_t read = 0;
484         ssize_t ret;
485
486         if (!*ppos)
487                 trace_parser_clear(parser);
488
489         ret = get_user(ch, ubuf++);
490         if (ret)
491                 goto out;
492
493         read++;
494         cnt--;
495
496         /*
497          * The parser is not finished with the last write,
498          * continue reading the user input without skipping spaces.
499          */
500         if (!parser->cont) {
501                 /* skip white space */
502                 while (cnt && isspace(ch)) {
503                         ret = get_user(ch, ubuf++);
504                         if (ret)
505                                 goto out;
506                         read++;
507                         cnt--;
508                 }
509
510                 /* only spaces were written */
511                 if (isspace(ch)) {
512                         *ppos += read;
513                         ret = read;
514                         goto out;
515                 }
516
517                 parser->idx = 0;
518         }
519
520         /* read the non-space input */
521         while (cnt && !isspace(ch)) {
522                 if (parser->idx < parser->size - 1)
523                         parser->buffer[parser->idx++] = ch;
524                 else {
525                         ret = -EINVAL;
526                         goto out;
527                 }
528                 ret = get_user(ch, ubuf++);
529                 if (ret)
530                         goto out;
531                 read++;
532                 cnt--;
533         }
534
535         /* We either got finished input or we have to wait for another call. */
536         if (isspace(ch)) {
537                 parser->buffer[parser->idx] = 0;
538                 parser->cont = false;
539         } else {
540                 parser->cont = true;
541                 parser->buffer[parser->idx++] = ch;
542         }
543
544         *ppos += read;
545         ret = read;
546
547 out:
548         return ret;
549 }
550
551 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
552 {
553         int len;
554         int ret;
555
556         if (!cnt)
557                 return 0;
558
559         if (s->len <= s->readpos)
560                 return -EBUSY;
561
562         len = s->len - s->readpos;
563         if (cnt > len)
564                 cnt = len;
565         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
566         if (ret == cnt)
567                 return -EFAULT;
568
569         cnt -= ret;
570
571         s->readpos += cnt;
572         return cnt;
573 }
574
575 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
576 {
577         int len;
578         void *ret;
579
580         if (s->len <= s->readpos)
581                 return -EBUSY;
582
583         len = s->len - s->readpos;
584         if (cnt > len)
585                 cnt = len;
586         ret = memcpy(buf, s->buffer + s->readpos, cnt);
587         if (!ret)
588                 return -EFAULT;
589
590         s->readpos += cnt;
591         return cnt;
592 }
593
594 /*
595  * ftrace_max_lock is used to protect the swapping of buffers
596  * when taking a max snapshot. The buffers themselves are
597  * protected by per_cpu spinlocks. But the action of the swap
598  * needs its own lock.
599  *
600  * This is defined as a arch_spinlock_t in order to help
601  * with performance when lockdep debugging is enabled.
602  *
603  * It is also used in other places outside the update_max_tr
604  * so it needs to be defined outside of the
605  * CONFIG_TRACER_MAX_TRACE.
606  */
607 static arch_spinlock_t ftrace_max_lock =
608         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
609
610 unsigned long __read_mostly     tracing_thresh;
611
612 #ifdef CONFIG_TRACER_MAX_TRACE
613 unsigned long __read_mostly     tracing_max_latency;
614
615 /*
616  * Copy the new maximum trace into the separate maximum-trace
617  * structure. (this way the maximum trace is permanently saved,
618  * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
619  */
620 static void
621 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
622 {
623         struct trace_array_cpu *data = tr->data[cpu];
624         struct trace_array_cpu *max_data;
625
626         max_tr.cpu = cpu;
627         max_tr.time_start = data->preempt_timestamp;
628
629         max_data = max_tr.data[cpu];
630         max_data->saved_latency = tracing_max_latency;
631         max_data->critical_start = data->critical_start;
632         max_data->critical_end = data->critical_end;
633
634         memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
635         max_data->pid = tsk->pid;
636         max_data->uid = task_uid(tsk);
637         max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
638         max_data->policy = tsk->policy;
639         max_data->rt_priority = tsk->rt_priority;
640
641         /* record this tasks comm */
642         tracing_record_cmdline(tsk);
643 }
644
645 /**
646  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
647  * @tr: tracer
648  * @tsk: the task with the latency
649  * @cpu: The cpu that initiated the trace.
650  *
651  * Flip the buffers between the @tr and the max_tr and record information
652  * about which task was the cause of this latency.
653  */
654 void
655 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
656 {
657         struct ring_buffer *buf = tr->buffer;
658
659         if (trace_stop_count)
660                 return;
661
662         WARN_ON_ONCE(!irqs_disabled());
663         if (!current_trace->use_max_tr) {
664                 WARN_ON_ONCE(1);
665                 return;
666         }
667         arch_spin_lock(&ftrace_max_lock);
668
669         tr->buffer = max_tr.buffer;
670         max_tr.buffer = buf;
671
672         __update_max_tr(tr, tsk, cpu);
673         arch_spin_unlock(&ftrace_max_lock);
674 }
675
676 /**
677  * update_max_tr_single - only copy one trace over, and reset the rest
678  * @tr - tracer
679  * @tsk - task with the latency
680  * @cpu - the cpu of the buffer to copy.
681  *
682  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
683  */
684 void
685 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
686 {
687         int ret;
688
689         if (trace_stop_count)
690                 return;
691
692         WARN_ON_ONCE(!irqs_disabled());
693         if (!current_trace->use_max_tr) {
694                 WARN_ON_ONCE(1);
695                 return;
696         }
697
698         arch_spin_lock(&ftrace_max_lock);
699
700         ftrace_disable_cpu();
701
702         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
703
704         if (ret == -EBUSY) {
705                 /*
706                  * We failed to swap the buffer due to a commit taking
707                  * place on this CPU. We fail to record, but we reset
708                  * the max trace buffer (no one writes directly to it)
709                  * and flag that it failed.
710                  */
711                 trace_array_printk(&max_tr, _THIS_IP_,
712                         "Failed to swap buffers due to commit in progress\n");
713         }
714
715         ftrace_enable_cpu();
716
717         WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
718
719         __update_max_tr(tr, tsk, cpu);
720         arch_spin_unlock(&ftrace_max_lock);
721 }
722 #endif /* CONFIG_TRACER_MAX_TRACE */
723
724 /**
725  * register_tracer - register a tracer with the ftrace system.
726  * @type - the plugin for the tracer
727  *
728  * Register a new plugin tracer.
729  */
730 int register_tracer(struct tracer *type)
731 __releases(kernel_lock)
732 __acquires(kernel_lock)
733 {
734         struct tracer *t;
735         int ret = 0;
736
737         if (!type->name) {
738                 pr_info("Tracer must have a name\n");
739                 return -1;
740         }
741
742         if (strlen(type->name) >= MAX_TRACER_SIZE) {
743                 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
744                 return -1;
745         }
746
747         /*
748          * When this gets called we hold the BKL which means that
749          * preemption is disabled. Various trace selftests however
750          * need to disable and enable preemption for successful tests.
751          * So we drop the BKL here and grab it after the tests again.
752          */
753         unlock_kernel();
754         mutex_lock(&trace_types_lock);
755
756         tracing_selftest_running = true;
757
758         for (t = trace_types; t; t = t->next) {
759                 if (strcmp(type->name, t->name) == 0) {
760                         /* already found */
761                         pr_info("Tracer %s already registered\n",
762                                 type->name);
763                         ret = -1;
764                         goto out;
765                 }
766         }
767
768         if (!type->set_flag)
769                 type->set_flag = &dummy_set_flag;
770         if (!type->flags)
771                 type->flags = &dummy_tracer_flags;
772         else
773                 if (!type->flags->opts)
774                         type->flags->opts = dummy_tracer_opt;
775         if (!type->wait_pipe)
776                 type->wait_pipe = default_wait_pipe;
777
778
779 #ifdef CONFIG_FTRACE_STARTUP_TEST
780         if (type->selftest && !tracing_selftest_disabled) {
781                 struct tracer *saved_tracer = current_trace;
782                 struct trace_array *tr = &global_trace;
783
784                 /*
785                  * Run a selftest on this tracer.
786                  * Here we reset the trace buffer, and set the current
787                  * tracer to be this tracer. The tracer can then run some
788                  * internal tracing to verify that everything is in order.
789                  * If we fail, we do not register this tracer.
790                  */
791                 tracing_reset_online_cpus(tr);
792
793                 current_trace = type;
794                 /* the test is responsible for initializing and enabling */
795                 pr_info("Testing tracer %s: ", type->name);
796                 ret = type->selftest(type, tr);
797                 /* the test is responsible for resetting too */
798                 current_trace = saved_tracer;
799                 if (ret) {
800                         printk(KERN_CONT "FAILED!\n");
801                         goto out;
802                 }
803                 /* Only reset on passing, to avoid touching corrupted buffers */
804                 tracing_reset_online_cpus(tr);
805
806                 printk(KERN_CONT "PASSED\n");
807         }
808 #endif
809
810         type->next = trace_types;
811         trace_types = type;
812
813  out:
814         tracing_selftest_running = false;
815         mutex_unlock(&trace_types_lock);
816
817         if (ret || !default_bootup_tracer)
818                 goto out_unlock;
819
820         if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
821                 goto out_unlock;
822
823         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
824         /* Do we want this tracer to start on bootup? */
825         tracing_set_tracer(type->name);
826         default_bootup_tracer = NULL;
827         /* disable other selftests, since this will break it. */
828         tracing_selftest_disabled = 1;
829 #ifdef CONFIG_FTRACE_STARTUP_TEST
830         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
831                type->name);
832 #endif
833
834  out_unlock:
835         lock_kernel();
836         return ret;
837 }
838
839 void unregister_tracer(struct tracer *type)
840 {
841         struct tracer **t;
842
843         mutex_lock(&trace_types_lock);
844         for (t = &trace_types; *t; t = &(*t)->next) {
845                 if (*t == type)
846                         goto found;
847         }
848         pr_info("Tracer %s not registered\n", type->name);
849         goto out;
850
851  found:
852         *t = (*t)->next;
853
854         if (type == current_trace && tracer_enabled) {
855                 tracer_enabled = 0;
856                 tracing_stop();
857                 if (current_trace->stop)
858                         current_trace->stop(&global_trace);
859                 current_trace = &nop_trace;
860         }
861 out:
862         mutex_unlock(&trace_types_lock);
863 }
864
865 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
866 {
867         ftrace_disable_cpu();
868         ring_buffer_reset_cpu(buffer, cpu);
869         ftrace_enable_cpu();
870 }
871
872 void tracing_reset(struct trace_array *tr, int cpu)
873 {
874         struct ring_buffer *buffer = tr->buffer;
875
876         ring_buffer_record_disable(buffer);
877
878         /* Make sure all commits have finished */
879         synchronize_sched();
880         __tracing_reset(buffer, cpu);
881
882         ring_buffer_record_enable(buffer);
883 }
884
885 void tracing_reset_online_cpus(struct trace_array *tr)
886 {
887         struct ring_buffer *buffer = tr->buffer;
888         int cpu;
889
890         ring_buffer_record_disable(buffer);
891
892         /* Make sure all commits have finished */
893         synchronize_sched();
894
895         tr->time_start = ftrace_now(tr->cpu);
896
897         for_each_online_cpu(cpu)
898                 __tracing_reset(buffer, cpu);
899
900         ring_buffer_record_enable(buffer);
901 }
902
903 void tracing_reset_current(int cpu)
904 {
905         tracing_reset(&global_trace, cpu);
906 }
907
908 void tracing_reset_current_online_cpus(void)
909 {
910         tracing_reset_online_cpus(&global_trace);
911 }
912
913 #define SAVED_CMDLINES 128
914 #define NO_CMDLINE_MAP UINT_MAX
915 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
916 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
917 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
918 static int cmdline_idx;
919 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
920
921 /* temporary disable recording */
922 static atomic_t trace_record_cmdline_disabled __read_mostly;
923
924 static void trace_init_cmdlines(void)
925 {
926         memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
927         memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
928         cmdline_idx = 0;
929 }
930
931 int is_tracing_stopped(void)
932 {
933         return trace_stop_count;
934 }
935
936 /**
937  * ftrace_off_permanent - disable all ftrace code permanently
938  *
939  * This should only be called when a serious anomally has
940  * been detected.  This will turn off the function tracing,
941  * ring buffers, and other tracing utilites. It takes no
942  * locks and can be called from any context.
943  */
944 void ftrace_off_permanent(void)
945 {
946         tracing_disabled = 1;
947         ftrace_stop();
948         tracing_off_permanent();
949 }
950
951 /**
952  * tracing_start - quick start of the tracer
953  *
954  * If tracing is enabled but was stopped by tracing_stop,
955  * this will start the tracer back up.
956  */
957 void tracing_start(void)
958 {
959         struct ring_buffer *buffer;
960         unsigned long flags;
961
962         if (tracing_disabled)
963                 return;
964
965         spin_lock_irqsave(&tracing_start_lock, flags);
966         if (--trace_stop_count) {
967                 if (trace_stop_count < 0) {
968                         /* Someone screwed up their debugging */
969                         WARN_ON_ONCE(1);
970                         trace_stop_count = 0;
971                 }
972                 goto out;
973         }
974
975         /* Prevent the buffers from switching */
976         arch_spin_lock(&ftrace_max_lock);
977
978         buffer = global_trace.buffer;
979         if (buffer)
980                 ring_buffer_record_enable(buffer);
981
982         buffer = max_tr.buffer;
983         if (buffer)
984                 ring_buffer_record_enable(buffer);
985
986         arch_spin_unlock(&ftrace_max_lock);
987
988         ftrace_start();
989  out:
990         spin_unlock_irqrestore(&tracing_start_lock, flags);
991 }
992
993 /**
994  * tracing_stop - quick stop of the tracer
995  *
996  * Light weight way to stop tracing. Use in conjunction with
997  * tracing_start.
998  */
999 void tracing_stop(void)
1000 {
1001         struct ring_buffer *buffer;
1002         unsigned long flags;
1003
1004         ftrace_stop();
1005         spin_lock_irqsave(&tracing_start_lock, flags);
1006         if (trace_stop_count++)
1007                 goto out;
1008
1009         /* Prevent the buffers from switching */
1010         arch_spin_lock(&ftrace_max_lock);
1011
1012         buffer = global_trace.buffer;
1013         if (buffer)
1014                 ring_buffer_record_disable(buffer);
1015
1016         buffer = max_tr.buffer;
1017         if (buffer)
1018                 ring_buffer_record_disable(buffer);
1019
1020         arch_spin_unlock(&ftrace_max_lock);
1021
1022  out:
1023         spin_unlock_irqrestore(&tracing_start_lock, flags);
1024 }
1025
1026 void trace_stop_cmdline_recording(void);
1027
1028 static void trace_save_cmdline(struct task_struct *tsk)
1029 {
1030         unsigned pid, idx;
1031
1032         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1033                 return;
1034
1035         /*
1036          * It's not the end of the world if we don't get
1037          * the lock, but we also don't want to spin
1038          * nor do we want to disable interrupts,
1039          * so if we miss here, then better luck next time.
1040          */
1041         if (!arch_spin_trylock(&trace_cmdline_lock))
1042                 return;
1043
1044         idx = map_pid_to_cmdline[tsk->pid];
1045         if (idx == NO_CMDLINE_MAP) {
1046                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1047
1048                 /*
1049                  * Check whether the cmdline buffer at idx has a pid
1050                  * mapped. We are going to overwrite that entry so we
1051                  * need to clear the map_pid_to_cmdline. Otherwise we
1052                  * would read the new comm for the old pid.
1053                  */
1054                 pid = map_cmdline_to_pid[idx];
1055                 if (pid != NO_CMDLINE_MAP)
1056                         map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1057
1058                 map_cmdline_to_pid[idx] = tsk->pid;
1059                 map_pid_to_cmdline[tsk->pid] = idx;
1060
1061                 cmdline_idx = idx;
1062         }
1063
1064         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1065
1066         arch_spin_unlock(&trace_cmdline_lock);
1067 }
1068
1069 void trace_find_cmdline(int pid, char comm[])
1070 {
1071         unsigned map;
1072
1073         if (!pid) {
1074                 strcpy(comm, "<idle>");
1075                 return;
1076         }
1077
1078         if (WARN_ON_ONCE(pid < 0)) {
1079                 strcpy(comm, "<XXX>");
1080                 return;
1081         }
1082
1083         if (pid > PID_MAX_DEFAULT) {
1084                 strcpy(comm, "<...>");
1085                 return;
1086         }
1087
1088         preempt_disable();
1089         arch_spin_lock(&trace_cmdline_lock);
1090         map = map_pid_to_cmdline[pid];
1091         if (map != NO_CMDLINE_MAP)
1092                 strcpy(comm, saved_cmdlines[map]);
1093         else
1094                 strcpy(comm, "<...>");
1095
1096         arch_spin_unlock(&trace_cmdline_lock);
1097         preempt_enable();
1098 }
1099
1100 void tracing_record_cmdline(struct task_struct *tsk)
1101 {
1102         if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1103             !tracing_is_on())
1104                 return;
1105
1106         trace_save_cmdline(tsk);
1107 }
1108
1109 void
1110 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1111                              int pc)
1112 {
1113         struct task_struct *tsk = current;
1114
1115         entry->preempt_count            = pc & 0xff;
1116         entry->pid                      = (tsk) ? tsk->pid : 0;
1117         entry->lock_depth               = (tsk) ? tsk->lock_depth : 0;
1118         entry->flags =
1119 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1120                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1121 #else
1122                 TRACE_FLAG_IRQS_NOSUPPORT |
1123 #endif
1124                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1125                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1126                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1127 }
1128 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1129
1130 struct ring_buffer_event *
1131 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1132                           int type,
1133                           unsigned long len,
1134                           unsigned long flags, int pc)
1135 {
1136         struct ring_buffer_event *event;
1137
1138         event = ring_buffer_lock_reserve(buffer, len);
1139         if (event != NULL) {
1140                 struct trace_entry *ent = ring_buffer_event_data(event);
1141
1142                 tracing_generic_entry_update(ent, flags, pc);
1143                 ent->type = type;
1144         }
1145
1146         return event;
1147 }
1148
1149 static inline void
1150 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1151                              struct ring_buffer_event *event,
1152                              unsigned long flags, int pc,
1153                              int wake)
1154 {
1155         ring_buffer_unlock_commit(buffer, event);
1156
1157         ftrace_trace_stack(buffer, flags, 6, pc);
1158         ftrace_trace_userstack(buffer, flags, pc);
1159
1160         if (wake)
1161                 trace_wake_up();
1162 }
1163
1164 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1165                                 struct ring_buffer_event *event,
1166                                 unsigned long flags, int pc)
1167 {
1168         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1169 }
1170
1171 struct ring_buffer_event *
1172 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1173                                   int type, unsigned long len,
1174                                   unsigned long flags, int pc)
1175 {
1176         *current_rb = global_trace.buffer;
1177         return trace_buffer_lock_reserve(*current_rb,
1178                                          type, len, flags, pc);
1179 }
1180 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1181
1182 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1183                                         struct ring_buffer_event *event,
1184                                         unsigned long flags, int pc)
1185 {
1186         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1187 }
1188 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1189
1190 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1191                                        struct ring_buffer_event *event,
1192                                        unsigned long flags, int pc)
1193 {
1194         __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1195 }
1196 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1197
1198 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1199                                          struct ring_buffer_event *event)
1200 {
1201         ring_buffer_discard_commit(buffer, event);
1202 }
1203 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1204
1205 void
1206 trace_function(struct trace_array *tr,
1207                unsigned long ip, unsigned long parent_ip, unsigned long flags,
1208                int pc)
1209 {
1210         struct ftrace_event_call *call = &event_function;
1211         struct ring_buffer *buffer = tr->buffer;
1212         struct ring_buffer_event *event;
1213         struct ftrace_entry *entry;
1214
1215         /* If we are reading the ring buffer, don't trace */
1216         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1217                 return;
1218
1219         event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1220                                           flags, pc);
1221         if (!event)
1222                 return;
1223         entry   = ring_buffer_event_data(event);
1224         entry->ip                       = ip;
1225         entry->parent_ip                = parent_ip;
1226
1227         if (!filter_check_discard(call, entry, buffer, event))
1228                 ring_buffer_unlock_commit(buffer, event);
1229 }
1230
1231 void
1232 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1233        unsigned long ip, unsigned long parent_ip, unsigned long flags,
1234        int pc)
1235 {
1236         if (likely(!atomic_read(&data->disabled)))
1237                 trace_function(tr, ip, parent_ip, flags, pc);
1238 }
1239
1240 #ifdef CONFIG_STACKTRACE
1241 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1242                                  unsigned long flags,
1243                                  int skip, int pc)
1244 {
1245         struct ftrace_event_call *call = &event_kernel_stack;
1246         struct ring_buffer_event *event;
1247         struct stack_entry *entry;
1248         struct stack_trace trace;
1249
1250         event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1251                                           sizeof(*entry), flags, pc);
1252         if (!event)
1253                 return;
1254         entry   = ring_buffer_event_data(event);
1255         memset(&entry->caller, 0, sizeof(entry->caller));
1256
1257         trace.nr_entries        = 0;
1258         trace.max_entries       = FTRACE_STACK_ENTRIES;
1259         trace.skip              = skip;
1260         trace.entries           = entry->caller;
1261
1262         save_stack_trace(&trace);
1263         if (!filter_check_discard(call, entry, buffer, event))
1264                 ring_buffer_unlock_commit(buffer, event);
1265 }
1266
1267 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1268                         int skip, int pc)
1269 {
1270         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1271                 return;
1272
1273         __ftrace_trace_stack(buffer, flags, skip, pc);
1274 }
1275
1276 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1277                    int pc)
1278 {
1279         __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1280 }
1281
1282 /**
1283  * trace_dump_stack - record a stack back trace in the trace buffer
1284  */
1285 void trace_dump_stack(void)
1286 {
1287         unsigned long flags;
1288
1289         if (tracing_disabled || tracing_selftest_running)
1290                 return;
1291
1292         local_save_flags(flags);
1293
1294         /* skipping 3 traces, seems to get us at the caller of this function */
1295         __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1296 }
1297
1298 void
1299 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1300 {
1301         struct ftrace_event_call *call = &event_user_stack;
1302         struct ring_buffer_event *event;
1303         struct userstack_entry *entry;
1304         struct stack_trace trace;
1305
1306         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1307                 return;
1308
1309         /*
1310          * NMIs can not handle page faults, even with fix ups.
1311          * The save user stack can (and often does) fault.
1312          */
1313         if (unlikely(in_nmi()))
1314                 return;
1315
1316         event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1317                                           sizeof(*entry), flags, pc);
1318         if (!event)
1319                 return;
1320         entry   = ring_buffer_event_data(event);
1321
1322         entry->tgid             = current->tgid;
1323         memset(&entry->caller, 0, sizeof(entry->caller));
1324
1325         trace.nr_entries        = 0;
1326         trace.max_entries       = FTRACE_STACK_ENTRIES;
1327         trace.skip              = 0;
1328         trace.entries           = entry->caller;
1329
1330         save_stack_trace_user(&trace);
1331         if (!filter_check_discard(call, entry, buffer, event))
1332                 ring_buffer_unlock_commit(buffer, event);
1333 }
1334
1335 #ifdef UNUSED
1336 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1337 {
1338         ftrace_trace_userstack(tr, flags, preempt_count());
1339 }
1340 #endif /* UNUSED */
1341
1342 #endif /* CONFIG_STACKTRACE */
1343
1344 /**
1345  * trace_vbprintk - write binary msg to tracing buffer
1346  *
1347  */
1348 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1349 {
1350         static arch_spinlock_t trace_buf_lock =
1351                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1352         static u32 trace_buf[TRACE_BUF_SIZE];
1353
1354         struct ftrace_event_call *call = &event_bprint;
1355         struct ring_buffer_event *event;
1356         struct ring_buffer *buffer;
1357         struct trace_array *tr = &global_trace;
1358         struct trace_array_cpu *data;
1359         struct bprint_entry *entry;
1360         unsigned long flags;
1361         int disable;
1362         int cpu, len = 0, size, pc;
1363
1364         if (unlikely(tracing_selftest_running || tracing_disabled))
1365                 return 0;
1366
1367         /* Don't pollute graph traces with trace_vprintk internals */
1368         pause_graph_tracing();
1369
1370         pc = preempt_count();
1371         preempt_disable_notrace();
1372         cpu = raw_smp_processor_id();
1373         data = tr->data[cpu];
1374
1375         disable = atomic_inc_return(&data->disabled);
1376         if (unlikely(disable != 1))
1377                 goto out;
1378
1379         /* Lockdep uses trace_printk for lock tracing */
1380         local_irq_save(flags);
1381         arch_spin_lock(&trace_buf_lock);
1382         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1383
1384         if (len > TRACE_BUF_SIZE || len < 0)
1385                 goto out_unlock;
1386
1387         size = sizeof(*entry) + sizeof(u32) * len;
1388         buffer = tr->buffer;
1389         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1390                                           flags, pc);
1391         if (!event)
1392                 goto out_unlock;
1393         entry = ring_buffer_event_data(event);
1394         entry->ip                       = ip;
1395         entry->fmt                      = fmt;
1396
1397         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1398         if (!filter_check_discard(call, entry, buffer, event)) {
1399                 ring_buffer_unlock_commit(buffer, event);
1400                 ftrace_trace_stack(buffer, flags, 6, pc);
1401         }
1402
1403 out_unlock:
1404         arch_spin_unlock(&trace_buf_lock);
1405         local_irq_restore(flags);
1406
1407 out:
1408         atomic_dec_return(&data->disabled);
1409         preempt_enable_notrace();
1410         unpause_graph_tracing();
1411
1412         return len;
1413 }
1414 EXPORT_SYMBOL_GPL(trace_vbprintk);
1415
1416 int trace_array_printk(struct trace_array *tr,
1417                        unsigned long ip, const char *fmt, ...)
1418 {
1419         int ret;
1420         va_list ap;
1421
1422         if (!(trace_flags & TRACE_ITER_PRINTK))
1423                 return 0;
1424
1425         va_start(ap, fmt);
1426         ret = trace_array_vprintk(tr, ip, fmt, ap);
1427         va_end(ap);
1428         return ret;
1429 }
1430
1431 int trace_array_vprintk(struct trace_array *tr,
1432                         unsigned long ip, const char *fmt, va_list args)
1433 {
1434         static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1435         static char trace_buf[TRACE_BUF_SIZE];
1436
1437         struct ftrace_event_call *call = &event_print;
1438         struct ring_buffer_event *event;
1439         struct ring_buffer *buffer;
1440         struct trace_array_cpu *data;
1441         int cpu, len = 0, size, pc;
1442         struct print_entry *entry;
1443         unsigned long irq_flags;
1444         int disable;
1445
1446         if (tracing_disabled || tracing_selftest_running)
1447                 return 0;
1448
1449         pc = preempt_count();
1450         preempt_disable_notrace();
1451         cpu = raw_smp_processor_id();
1452         data = tr->data[cpu];
1453
1454         disable = atomic_inc_return(&data->disabled);
1455         if (unlikely(disable != 1))
1456                 goto out;
1457
1458         pause_graph_tracing();
1459         raw_local_irq_save(irq_flags);
1460         arch_spin_lock(&trace_buf_lock);
1461         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1462
1463         size = sizeof(*entry) + len + 1;
1464         buffer = tr->buffer;
1465         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1466                                           irq_flags, pc);
1467         if (!event)
1468                 goto out_unlock;
1469         entry = ring_buffer_event_data(event);
1470         entry->ip = ip;
1471
1472         memcpy(&entry->buf, trace_buf, len);
1473         entry->buf[len] = '\0';
1474         if (!filter_check_discard(call, entry, buffer, event)) {
1475                 ring_buffer_unlock_commit(buffer, event);
1476                 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1477         }
1478
1479  out_unlock:
1480         arch_spin_unlock(&trace_buf_lock);
1481         raw_local_irq_restore(irq_flags);
1482         unpause_graph_tracing();
1483  out:
1484         atomic_dec_return(&data->disabled);
1485         preempt_enable_notrace();
1486
1487         return len;
1488 }
1489
1490 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1491 {
1492         return trace_array_vprintk(&global_trace, ip, fmt, args);
1493 }
1494 EXPORT_SYMBOL_GPL(trace_vprintk);
1495
1496 enum trace_file_type {
1497         TRACE_FILE_LAT_FMT      = 1,
1498         TRACE_FILE_ANNOTATE     = 2,
1499 };
1500
1501 static void trace_iterator_increment(struct trace_iterator *iter)
1502 {
1503         /* Don't allow ftrace to trace into the ring buffers */
1504         ftrace_disable_cpu();
1505
1506         iter->idx++;
1507         if (iter->buffer_iter[iter->cpu])
1508                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1509
1510         ftrace_enable_cpu();
1511 }
1512
1513 static struct trace_entry *
1514 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1515                 unsigned long *lost_events)
1516 {
1517         struct ring_buffer_event *event;
1518         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1519
1520         /* Don't allow ftrace to trace into the ring buffers */
1521         ftrace_disable_cpu();
1522
1523         if (buf_iter)
1524                 event = ring_buffer_iter_peek(buf_iter, ts);
1525         else
1526                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1527                                          lost_events);
1528
1529         ftrace_enable_cpu();
1530
1531         return event ? ring_buffer_event_data(event) : NULL;
1532 }
1533
1534 static struct trace_entry *
1535 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1536                   unsigned long *missing_events, u64 *ent_ts)
1537 {
1538         struct ring_buffer *buffer = iter->tr->buffer;
1539         struct trace_entry *ent, *next = NULL;
1540         unsigned long lost_events = 0, next_lost = 0;
1541         int cpu_file = iter->cpu_file;
1542         u64 next_ts = 0, ts;
1543         int next_cpu = -1;
1544         int cpu;
1545
1546         /*
1547          * If we are in a per_cpu trace file, don't bother by iterating over
1548          * all cpu and peek directly.
1549          */
1550         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1551                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1552                         return NULL;
1553                 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1554                 if (ent_cpu)
1555                         *ent_cpu = cpu_file;
1556
1557                 return ent;
1558         }
1559
1560         for_each_tracing_cpu(cpu) {
1561
1562                 if (ring_buffer_empty_cpu(buffer, cpu))
1563                         continue;
1564
1565                 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1566
1567                 /*
1568                  * Pick the entry with the smallest timestamp:
1569                  */
1570                 if (ent && (!next || ts < next_ts)) {
1571                         next = ent;
1572                         next_cpu = cpu;
1573                         next_ts = ts;
1574                         next_lost = lost_events;
1575                 }
1576         }
1577
1578         if (ent_cpu)
1579                 *ent_cpu = next_cpu;
1580
1581         if (ent_ts)
1582                 *ent_ts = next_ts;
1583
1584         if (missing_events)
1585                 *missing_events = next_lost;
1586
1587         return next;
1588 }
1589
1590 /* Find the next real entry, without updating the iterator itself */
1591 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1592                                           int *ent_cpu, u64 *ent_ts)
1593 {
1594         return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1595 }
1596
1597 /* Find the next real entry, and increment the iterator to the next entry */
1598 static void *find_next_entry_inc(struct trace_iterator *iter)
1599 {
1600         iter->ent = __find_next_entry(iter, &iter->cpu,
1601                                       &iter->lost_events, &iter->ts);
1602
1603         if (iter->ent)
1604                 trace_iterator_increment(iter);
1605
1606         return iter->ent ? iter : NULL;
1607 }
1608
1609 static void trace_consume(struct trace_iterator *iter)
1610 {
1611         /* Don't allow ftrace to trace into the ring buffers */
1612         ftrace_disable_cpu();
1613         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1614                             &iter->lost_events);
1615         ftrace_enable_cpu();
1616 }
1617
1618 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1619 {
1620         struct trace_iterator *iter = m->private;
1621         int i = (int)*pos;
1622         void *ent;
1623
1624         WARN_ON_ONCE(iter->leftover);
1625
1626         (*pos)++;
1627
1628         /* can't go backwards */
1629         if (iter->idx > i)
1630                 return NULL;
1631
1632         if (iter->idx < 0)
1633                 ent = find_next_entry_inc(iter);
1634         else
1635                 ent = iter;
1636
1637         while (ent && iter->idx < i)
1638                 ent = find_next_entry_inc(iter);
1639
1640         iter->pos = *pos;
1641
1642         return ent;
1643 }
1644
1645 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1646 {
1647         struct trace_array *tr = iter->tr;
1648         struct ring_buffer_event *event;
1649         struct ring_buffer_iter *buf_iter;
1650         unsigned long entries = 0;
1651         u64 ts;
1652
1653         tr->data[cpu]->skipped_entries = 0;
1654
1655         if (!iter->buffer_iter[cpu])
1656                 return;
1657
1658         buf_iter = iter->buffer_iter[cpu];
1659         ring_buffer_iter_reset(buf_iter);
1660
1661         /*
1662          * We could have the case with the max latency tracers
1663          * that a reset never took place on a cpu. This is evident
1664          * by the timestamp being before the start of the buffer.
1665          */
1666         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1667                 if (ts >= iter->tr->time_start)
1668                         break;
1669                 entries++;
1670                 ring_buffer_read(buf_iter, NULL);
1671         }
1672
1673         tr->data[cpu]->skipped_entries = entries;
1674 }
1675
1676 /*
1677  * The current tracer is copied to avoid a global locking
1678  * all around.
1679  */
1680 static void *s_start(struct seq_file *m, loff_t *pos)
1681 {
1682         struct trace_iterator *iter = m->private;
1683         static struct tracer *old_tracer;
1684         int cpu_file = iter->cpu_file;
1685         void *p = NULL;
1686         loff_t l = 0;
1687         int cpu;
1688
1689         /* copy the tracer to avoid using a global lock all around */
1690         mutex_lock(&trace_types_lock);
1691         if (unlikely(old_tracer != current_trace && current_trace)) {
1692                 old_tracer = current_trace;
1693                 *iter->trace = *current_trace;
1694         }
1695         mutex_unlock(&trace_types_lock);
1696
1697         atomic_inc(&trace_record_cmdline_disabled);
1698
1699         if (*pos != iter->pos) {
1700                 iter->ent = NULL;
1701                 iter->cpu = 0;
1702                 iter->idx = -1;
1703
1704                 ftrace_disable_cpu();
1705
1706                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1707                         for_each_tracing_cpu(cpu)
1708                                 tracing_iter_reset(iter, cpu);
1709                 } else
1710                         tracing_iter_reset(iter, cpu_file);
1711
1712                 ftrace_enable_cpu();
1713
1714                 iter->leftover = 0;
1715                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1716                         ;
1717
1718         } else {
1719                 /*
1720                  * If we overflowed the seq_file before, then we want
1721                  * to just reuse the trace_seq buffer again.
1722                  */
1723                 if (iter->leftover)
1724                         p = iter;
1725                 else {
1726                         l = *pos - 1;
1727                         p = s_next(m, p, &l);
1728                 }
1729         }
1730
1731         trace_event_read_lock();
1732         trace_access_lock(cpu_file);
1733         return p;
1734 }
1735
1736 static void s_stop(struct seq_file *m, void *p)
1737 {
1738         struct trace_iterator *iter = m->private;
1739
1740         atomic_dec(&trace_record_cmdline_disabled);
1741         trace_access_unlock(iter->cpu_file);
1742         trace_event_read_unlock();
1743 }
1744
1745 static void print_lat_help_header(struct seq_file *m)
1746 {
1747         seq_puts(m, "#                  _------=> CPU#            \n");
1748         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1749         seq_puts(m, "#                | / _----=> need-resched    \n");
1750         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1751         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1752         seq_puts(m, "#                |||| /_--=> lock-depth       \n");
1753         seq_puts(m, "#                |||||/     delay             \n");
1754         seq_puts(m, "#  cmd     pid   |||||| time  |   caller      \n");
1755         seq_puts(m, "#     \\   /      ||||||   \\   |   /           \n");
1756 }
1757
1758 static void print_func_help_header(struct seq_file *m)
1759 {
1760         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1761         seq_puts(m, "#              | |       |          |         |\n");
1762 }
1763
1764
1765 void
1766 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1767 {
1768         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1769         struct trace_array *tr = iter->tr;
1770         struct trace_array_cpu *data = tr->data[tr->cpu];
1771         struct tracer *type = current_trace;
1772         unsigned long entries = 0;
1773         unsigned long total = 0;
1774         unsigned long count;
1775         const char *name = "preemption";
1776         int cpu;
1777
1778         if (type)
1779                 name = type->name;
1780
1781
1782         for_each_tracing_cpu(cpu) {
1783                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1784                 /*
1785                  * If this buffer has skipped entries, then we hold all
1786                  * entries for the trace and we need to ignore the
1787                  * ones before the time stamp.
1788                  */
1789                 if (tr->data[cpu]->skipped_entries) {
1790                         count -= tr->data[cpu]->skipped_entries;
1791                         /* total is the same as the entries */
1792                         total += count;
1793                 } else
1794                         total += count +
1795                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1796                 entries += count;
1797         }
1798
1799         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1800                    name, UTS_RELEASE);
1801         seq_puts(m, "# -----------------------------------"
1802                  "---------------------------------\n");
1803         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1804                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1805                    nsecs_to_usecs(data->saved_latency),
1806                    entries,
1807                    total,
1808                    tr->cpu,
1809 #if defined(CONFIG_PREEMPT_NONE)
1810                    "server",
1811 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1812                    "desktop",
1813 #elif defined(CONFIG_PREEMPT)
1814                    "preempt",
1815 #else
1816                    "unknown",
1817 #endif
1818                    /* These are reserved for later use */
1819                    0, 0, 0, 0);
1820 #ifdef CONFIG_SMP
1821         seq_printf(m, " #P:%d)\n", num_online_cpus());
1822 #else
1823         seq_puts(m, ")\n");
1824 #endif
1825         seq_puts(m, "#    -----------------\n");
1826         seq_printf(m, "#    | task: %.16s-%d "
1827                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1828                    data->comm, data->pid, data->uid, data->nice,
1829                    data->policy, data->rt_priority);
1830         seq_puts(m, "#    -----------------\n");
1831
1832         if (data->critical_start) {
1833                 seq_puts(m, "#  => started at: ");
1834                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1835                 trace_print_seq(m, &iter->seq);
1836                 seq_puts(m, "\n#  => ended at:   ");
1837                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1838                 trace_print_seq(m, &iter->seq);
1839                 seq_puts(m, "\n#\n");
1840         }
1841
1842         seq_puts(m, "#\n");
1843 }
1844
1845 static void test_cpu_buff_start(struct trace_iterator *iter)
1846 {
1847         struct trace_seq *s = &iter->seq;
1848
1849         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1850                 return;
1851
1852         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1853                 return;
1854
1855         if (cpumask_test_cpu(iter->cpu, iter->started))
1856                 return;
1857
1858         if (iter->tr->data[iter->cpu]->skipped_entries)
1859                 return;
1860
1861         cpumask_set_cpu(iter->cpu, iter->started);
1862
1863         /* Don't print started cpu buffer for the first entry of the trace */
1864         if (iter->idx > 1)
1865                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1866                                 iter->cpu);
1867 }
1868
1869 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1870 {
1871         struct trace_seq *s = &iter->seq;
1872         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1873         struct trace_entry *entry;
1874         struct trace_event *event;
1875
1876         entry = iter->ent;
1877
1878         test_cpu_buff_start(iter);
1879
1880         event = ftrace_find_event(entry->type);
1881
1882         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1883                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1884                         if (!trace_print_lat_context(iter))
1885                                 goto partial;
1886                 } else {
1887                         if (!trace_print_context(iter))
1888                                 goto partial;
1889                 }
1890         }
1891
1892         if (event)
1893                 return event->funcs->trace(iter, sym_flags, event);
1894
1895         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1896                 goto partial;
1897
1898         return TRACE_TYPE_HANDLED;
1899 partial:
1900         return TRACE_TYPE_PARTIAL_LINE;
1901 }
1902
1903 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1904 {
1905         struct trace_seq *s = &iter->seq;
1906         struct trace_entry *entry;
1907         struct trace_event *event;
1908
1909         entry = iter->ent;
1910
1911         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1912                 if (!trace_seq_printf(s, "%d %d %llu ",
1913                                       entry->pid, iter->cpu, iter->ts))
1914                         goto partial;
1915         }
1916
1917         event = ftrace_find_event(entry->type);
1918         if (event)
1919                 return event->funcs->raw(iter, 0, event);
1920
1921         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1922                 goto partial;
1923
1924         return TRACE_TYPE_HANDLED;
1925 partial:
1926         return TRACE_TYPE_PARTIAL_LINE;
1927 }
1928
1929 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1930 {
1931         struct trace_seq *s = &iter->seq;
1932         unsigned char newline = '\n';
1933         struct trace_entry *entry;
1934         struct trace_event *event;
1935
1936         entry = iter->ent;
1937
1938         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1939                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1940                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1941                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1942         }
1943
1944         event = ftrace_find_event(entry->type);
1945         if (event) {
1946                 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1947                 if (ret != TRACE_TYPE_HANDLED)
1948                         return ret;
1949         }
1950
1951         SEQ_PUT_FIELD_RET(s, newline);
1952
1953         return TRACE_TYPE_HANDLED;
1954 }
1955
1956 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
1957 {
1958         struct trace_seq *s = &iter->seq;
1959         struct trace_entry *entry;
1960         struct trace_event *event;
1961
1962         entry = iter->ent;
1963
1964         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1965                 SEQ_PUT_FIELD_RET(s, entry->pid);
1966                 SEQ_PUT_FIELD_RET(s, iter->cpu);
1967                 SEQ_PUT_FIELD_RET(s, iter->ts);
1968         }
1969
1970         event = ftrace_find_event(entry->type);
1971         return event ? event->funcs->binary(iter, 0, event) :
1972                 TRACE_TYPE_HANDLED;
1973 }
1974
1975 int trace_empty(struct trace_iterator *iter)
1976 {
1977         int cpu;
1978
1979         /* If we are looking at one CPU buffer, only check that one */
1980         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1981                 cpu = iter->cpu_file;
1982                 if (iter->buffer_iter[cpu]) {
1983                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1984                                 return 0;
1985                 } else {
1986                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1987                                 return 0;
1988                 }
1989                 return 1;
1990         }
1991
1992         for_each_tracing_cpu(cpu) {
1993                 if (iter->buffer_iter[cpu]) {
1994                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1995                                 return 0;
1996                 } else {
1997                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1998                                 return 0;
1999                 }
2000         }
2001
2002         return 1;
2003 }
2004
2005 /*  Called with trace_event_read_lock() held. */
2006 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2007 {
2008         enum print_line_t ret;
2009
2010         if (iter->lost_events)
2011                 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2012                                  iter->cpu, iter->lost_events);
2013
2014         if (iter->trace && iter->trace->print_line) {
2015                 ret = iter->trace->print_line(iter);
2016                 if (ret != TRACE_TYPE_UNHANDLED)
2017                         return ret;
2018         }
2019
2020         if (iter->ent->type == TRACE_BPRINT &&
2021                         trace_flags & TRACE_ITER_PRINTK &&
2022                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2023                 return trace_print_bprintk_msg_only(iter);
2024
2025         if (iter->ent->type == TRACE_PRINT &&
2026                         trace_flags & TRACE_ITER_PRINTK &&
2027                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2028                 return trace_print_printk_msg_only(iter);
2029
2030         if (trace_flags & TRACE_ITER_BIN)
2031                 return print_bin_fmt(iter);
2032
2033         if (trace_flags & TRACE_ITER_HEX)
2034                 return print_hex_fmt(iter);
2035
2036         if (trace_flags & TRACE_ITER_RAW)
2037                 return print_raw_fmt(iter);
2038
2039         return print_trace_fmt(iter);
2040 }
2041
2042 void trace_default_header(struct seq_file *m)
2043 {
2044         struct trace_iterator *iter = m->private;
2045
2046         if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2047                 /* print nothing if the buffers are empty */
2048                 if (trace_empty(iter))
2049                         return;
2050                 print_trace_header(m, iter);
2051                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2052                         print_lat_help_header(m);
2053         } else {
2054                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2055                         print_func_help_header(m);
2056         }
2057 }
2058
2059 static int s_show(struct seq_file *m, void *v)
2060 {
2061         struct trace_iterator *iter = v;
2062         int ret;
2063
2064         if (iter->ent == NULL) {
2065                 if (iter->tr) {
2066                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2067                         seq_puts(m, "#\n");
2068                 }
2069                 if (iter->trace && iter->trace->print_header)
2070                         iter->trace->print_header(m);
2071                 else
2072                         trace_default_header(m);
2073
2074         } else if (iter->leftover) {
2075                 /*
2076                  * If we filled the seq_file buffer earlier, we
2077                  * want to just show it now.
2078                  */
2079                 ret = trace_print_seq(m, &iter->seq);
2080
2081                 /* ret should this time be zero, but you never know */
2082                 iter->leftover = ret;
2083
2084         } else {
2085                 print_trace_line(iter);
2086                 ret = trace_print_seq(m, &iter->seq);
2087                 /*
2088                  * If we overflow the seq_file buffer, then it will
2089                  * ask us for this data again at start up.
2090                  * Use that instead.
2091                  *  ret is 0 if seq_file write succeeded.
2092                  *        -1 otherwise.
2093                  */
2094                 iter->leftover = ret;
2095         }
2096
2097         return 0;
2098 }
2099
2100 static const struct seq_operations tracer_seq_ops = {
2101         .start          = s_start,
2102         .next           = s_next,
2103         .stop           = s_stop,
2104         .show           = s_show,
2105 };
2106
2107 static struct trace_iterator *
2108 __tracing_open(struct inode *inode, struct file *file)
2109 {
2110         long cpu_file = (long) inode->i_private;
2111         void *fail_ret = ERR_PTR(-ENOMEM);
2112         struct trace_iterator *iter;
2113         struct seq_file *m;
2114         int cpu, ret;
2115
2116         if (tracing_disabled)
2117                 return ERR_PTR(-ENODEV);
2118
2119         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2120         if (!iter)
2121                 return ERR_PTR(-ENOMEM);
2122
2123         /*
2124          * We make a copy of the current tracer to avoid concurrent
2125          * changes on it while we are reading.
2126          */
2127         mutex_lock(&trace_types_lock);
2128         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2129         if (!iter->trace)
2130                 goto fail;
2131
2132         if (current_trace)
2133                 *iter->trace = *current_trace;
2134
2135         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2136                 goto fail;
2137
2138         if (current_trace && current_trace->print_max)
2139                 iter->tr = &max_tr;
2140         else
2141                 iter->tr = &global_trace;
2142         iter->pos = -1;
2143         mutex_init(&iter->mutex);
2144         iter->cpu_file = cpu_file;
2145
2146         /* Notify the tracer early; before we stop tracing. */
2147         if (iter->trace && iter->trace->open)
2148                 iter->trace->open(iter);
2149
2150         /* Annotate start of buffers if we had overruns */
2151         if (ring_buffer_overruns(iter->tr->buffer))
2152                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2153
2154         /* stop the trace while dumping */
2155         tracing_stop();
2156
2157         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2158                 for_each_tracing_cpu(cpu) {
2159                         iter->buffer_iter[cpu] =
2160                                 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2161                 }
2162                 ring_buffer_read_prepare_sync();
2163                 for_each_tracing_cpu(cpu) {
2164                         ring_buffer_read_start(iter->buffer_iter[cpu]);
2165                         tracing_iter_reset(iter, cpu);
2166                 }
2167         } else {
2168                 cpu = iter->cpu_file;
2169                 iter->buffer_iter[cpu] =
2170                         ring_buffer_read_prepare(iter->tr->buffer, cpu);
2171                 ring_buffer_read_prepare_sync();
2172                 ring_buffer_read_start(iter->buffer_iter[cpu]);
2173                 tracing_iter_reset(iter, cpu);
2174         }
2175
2176         ret = seq_open(file, &tracer_seq_ops);
2177         if (ret < 0) {
2178                 fail_ret = ERR_PTR(ret);
2179                 goto fail_buffer;
2180         }
2181
2182         m = file->private_data;
2183         m->private = iter;
2184
2185         mutex_unlock(&trace_types_lock);
2186
2187         return iter;
2188
2189  fail_buffer:
2190         for_each_tracing_cpu(cpu) {
2191                 if (iter->buffer_iter[cpu])
2192                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2193         }
2194         free_cpumask_var(iter->started);
2195         tracing_start();
2196  fail:
2197         mutex_unlock(&trace_types_lock);
2198         kfree(iter->trace);
2199         kfree(iter);
2200
2201         return fail_ret;
2202 }
2203
2204 int tracing_open_generic(struct inode *inode, struct file *filp)
2205 {
2206         if (tracing_disabled)
2207                 return -ENODEV;
2208
2209         filp->private_data = inode->i_private;
2210         return 0;
2211 }
2212
2213 static int tracing_release(struct inode *inode, struct file *file)
2214 {
2215         struct seq_file *m = (struct seq_file *)file->private_data;
2216         struct trace_iterator *iter;
2217         int cpu;
2218
2219         if (!(file->f_mode & FMODE_READ))
2220                 return 0;
2221
2222         iter = m->private;
2223
2224         mutex_lock(&trace_types_lock);
2225         for_each_tracing_cpu(cpu) {
2226                 if (iter->buffer_iter[cpu])
2227                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2228         }
2229
2230         if (iter->trace && iter->trace->close)
2231                 iter->trace->close(iter);
2232
2233         /* reenable tracing if it was previously enabled */
2234         tracing_start();
2235         mutex_unlock(&trace_types_lock);
2236
2237         seq_release(inode, file);
2238         mutex_destroy(&iter->mutex);
2239         free_cpumask_var(iter->started);
2240         kfree(iter->trace);
2241         kfree(iter);
2242         return 0;
2243 }
2244
2245 static int tracing_open(struct inode *inode, struct file *file)
2246 {
2247         struct trace_iterator *iter;
2248         int ret = 0;
2249
2250         /* If this file was open for write, then erase contents */
2251         if ((file->f_mode & FMODE_WRITE) &&
2252             (file->f_flags & O_TRUNC)) {
2253                 long cpu = (long) inode->i_private;
2254
2255                 if (cpu == TRACE_PIPE_ALL_CPU)
2256                         tracing_reset_online_cpus(&global_trace);
2257                 else
2258                         tracing_reset(&global_trace, cpu);
2259         }
2260
2261         if (file->f_mode & FMODE_READ) {
2262                 iter = __tracing_open(inode, file);
2263                 if (IS_ERR(iter))
2264                         ret = PTR_ERR(iter);
2265                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2266                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2267         }
2268         return ret;
2269 }
2270
2271 static void *
2272 t_next(struct seq_file *m, void *v, loff_t *pos)
2273 {
2274         struct tracer *t = v;
2275
2276         (*pos)++;
2277
2278         if (t)
2279                 t = t->next;
2280
2281         return t;
2282 }
2283
2284 static void *t_start(struct seq_file *m, loff_t *pos)
2285 {
2286         struct tracer *t;
2287         loff_t l = 0;
2288
2289         mutex_lock(&trace_types_lock);
2290         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2291                 ;
2292
2293         return t;
2294 }
2295
2296 static void t_stop(struct seq_file *m, void *p)
2297 {
2298         mutex_unlock(&trace_types_lock);
2299 }
2300
2301 static int t_show(struct seq_file *m, void *v)
2302 {
2303         struct tracer *t = v;
2304
2305         if (!t)
2306                 return 0;
2307
2308         seq_printf(m, "%s", t->name);
2309         if (t->next)
2310                 seq_putc(m, ' ');
2311         else
2312                 seq_putc(m, '\n');
2313
2314         return 0;
2315 }
2316
2317 static const struct seq_operations show_traces_seq_ops = {
2318         .start          = t_start,
2319         .next           = t_next,
2320         .stop           = t_stop,
2321         .show           = t_show,
2322 };
2323
2324 static int show_traces_open(struct inode *inode, struct file *file)
2325 {
2326         if (tracing_disabled)
2327                 return -ENODEV;
2328
2329         return seq_open(file, &show_traces_seq_ops);
2330 }
2331
2332 static ssize_t
2333 tracing_write_stub(struct file *filp, const char __user *ubuf,
2334                    size_t count, loff_t *ppos)
2335 {
2336         return count;
2337 }
2338
2339 static const struct file_operations tracing_fops = {
2340         .open           = tracing_open,
2341         .read           = seq_read,
2342         .write          = tracing_write_stub,
2343         .llseek         = seq_lseek,
2344         .release        = tracing_release,
2345 };
2346
2347 static const struct file_operations show_traces_fops = {
2348         .open           = show_traces_open,
2349         .read           = seq_read,
2350         .release        = seq_release,
2351         .llseek         = seq_lseek,
2352 };
2353
2354 /*
2355  * Only trace on a CPU if the bitmask is set:
2356  */
2357 static cpumask_var_t tracing_cpumask;
2358
2359 /*
2360  * The tracer itself will not take this lock, but still we want
2361  * to provide a consistent cpumask to user-space:
2362  */
2363 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2364
2365 /*
2366  * Temporary storage for the character representation of the
2367  * CPU bitmask (and one more byte for the newline):
2368  */
2369 static char mask_str[NR_CPUS + 1];
2370
2371 static ssize_t
2372 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2373                      size_t count, loff_t *ppos)
2374 {
2375         int len;
2376
2377         mutex_lock(&tracing_cpumask_update_lock);
2378
2379         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2380         if (count - len < 2) {
2381                 count = -EINVAL;
2382                 goto out_err;
2383         }
2384         len += sprintf(mask_str + len, "\n");
2385         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2386
2387 out_err:
2388         mutex_unlock(&tracing_cpumask_update_lock);
2389
2390         return count;
2391 }
2392
2393 static ssize_t
2394 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2395                       size_t count, loff_t *ppos)
2396 {
2397         int err, cpu;
2398         cpumask_var_t tracing_cpumask_new;
2399
2400         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2401                 return -ENOMEM;
2402
2403         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2404         if (err)
2405                 goto err_unlock;
2406
2407         mutex_lock(&tracing_cpumask_update_lock);
2408
2409         local_irq_disable();
2410         arch_spin_lock(&ftrace_max_lock);
2411         for_each_tracing_cpu(cpu) {
2412                 /*
2413                  * Increase/decrease the disabled counter if we are
2414                  * about to flip a bit in the cpumask:
2415                  */
2416                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2417                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2418                         atomic_inc(&global_trace.data[cpu]->disabled);
2419                 }
2420                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2421                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2422                         atomic_dec(&global_trace.data[cpu]->disabled);
2423                 }
2424         }
2425         arch_spin_unlock(&ftrace_max_lock);
2426         local_irq_enable();
2427
2428         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2429
2430         mutex_unlock(&tracing_cpumask_update_lock);
2431         free_cpumask_var(tracing_cpumask_new);
2432
2433         return count;
2434
2435 err_unlock:
2436         free_cpumask_var(tracing_cpumask_new);
2437
2438         return err;
2439 }
2440
2441 static const struct file_operations tracing_cpumask_fops = {
2442         .open           = tracing_open_generic,
2443         .read           = tracing_cpumask_read,
2444         .write          = tracing_cpumask_write,
2445         .llseek         = generic_file_llseek,
2446 };
2447
2448 static int tracing_trace_options_show(struct seq_file *m, void *v)
2449 {
2450         struct tracer_opt *trace_opts;
2451         u32 tracer_flags;
2452         int i;
2453
2454         mutex_lock(&trace_types_lock);
2455         tracer_flags = current_trace->flags->val;
2456         trace_opts = current_trace->flags->opts;
2457
2458         for (i = 0; trace_options[i]; i++) {
2459                 if (trace_flags & (1 << i))
2460                         seq_printf(m, "%s\n", trace_options[i]);
2461                 else
2462                         seq_printf(m, "no%s\n", trace_options[i]);
2463         }
2464
2465         for (i = 0; trace_opts[i].name; i++) {
2466                 if (tracer_flags & trace_opts[i].bit)
2467                         seq_printf(m, "%s\n", trace_opts[i].name);
2468                 else
2469                         seq_printf(m, "no%s\n", trace_opts[i].name);
2470         }
2471         mutex_unlock(&trace_types_lock);
2472
2473         return 0;
2474 }
2475
2476 static int __set_tracer_option(struct tracer *trace,
2477                                struct tracer_flags *tracer_flags,
2478                                struct tracer_opt *opts, int neg)
2479 {
2480         int ret;
2481
2482         ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2483         if (ret)
2484                 return ret;
2485
2486         if (neg)
2487                 tracer_flags->val &= ~opts->bit;
2488         else
2489                 tracer_flags->val |= opts->bit;
2490         return 0;
2491 }
2492
2493 /* Try to assign a tracer specific option */
2494 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2495 {
2496         struct tracer_flags *tracer_flags = trace->flags;
2497         struct tracer_opt *opts = NULL;
2498         int i;
2499
2500         for (i = 0; tracer_flags->opts[i].name; i++) {
2501                 opts = &tracer_flags->opts[i];
2502
2503                 if (strcmp(cmp, opts->name) == 0)
2504                         return __set_tracer_option(trace, trace->flags,
2505                                                    opts, neg);
2506         }
2507
2508         return -EINVAL;
2509 }
2510
2511 static void set_tracer_flags(unsigned int mask, int enabled)
2512 {
2513         /* do nothing if flag is already set */
2514         if (!!(trace_flags & mask) == !!enabled)
2515                 return;
2516
2517         if (enabled)
2518                 trace_flags |= mask;
2519         else
2520                 trace_flags &= ~mask;
2521
2522         if (mask == TRACE_ITER_RECORD_CMD)
2523                 trace_event_enable_cmd_record(enabled);
2524 }
2525
2526 static ssize_t
2527 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2528                         size_t cnt, loff_t *ppos)
2529 {
2530         char buf[64];
2531         char *cmp;
2532         int neg = 0;
2533         int ret;
2534         int i;
2535
2536         if (cnt >= sizeof(buf))
2537                 return -EINVAL;
2538
2539         if (copy_from_user(&buf, ubuf, cnt))
2540                 return -EFAULT;
2541
2542         buf[cnt] = 0;
2543         cmp = strstrip(buf);
2544
2545         if (strncmp(cmp, "no", 2) == 0) {
2546                 neg = 1;
2547                 cmp += 2;
2548         }
2549
2550         for (i = 0; trace_options[i]; i++) {
2551                 if (strcmp(cmp, trace_options[i]) == 0) {
2552                         set_tracer_flags(1 << i, !neg);
2553                         break;
2554                 }
2555         }
2556
2557         /* If no option could be set, test the specific tracer options */
2558         if (!trace_options[i]) {
2559                 mutex_lock(&trace_types_lock);
2560                 ret = set_tracer_option(current_trace, cmp, neg);
2561                 mutex_unlock(&trace_types_lock);
2562                 if (ret)
2563                         return ret;
2564         }
2565
2566         *ppos += cnt;
2567
2568         return cnt;
2569 }
2570
2571 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2572 {
2573         if (tracing_disabled)
2574                 return -ENODEV;
2575         return single_open(file, tracing_trace_options_show, NULL);
2576 }
2577
2578 static const struct file_operations tracing_iter_fops = {
2579         .open           = tracing_trace_options_open,
2580         .read           = seq_read,
2581         .llseek         = seq_lseek,
2582         .release        = single_release,
2583         .write          = tracing_trace_options_write,
2584 };
2585
2586 static const char readme_msg[] =
2587         "tracing mini-HOWTO:\n\n"
2588         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2589         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2590         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2591         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2592         "nop\n"
2593         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2594         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2595         "sched_switch\n"
2596         "# cat /sys/kernel/debug/tracing/trace_options\n"
2597         "noprint-parent nosym-offset nosym-addr noverbose\n"
2598         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2599         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2600         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2601         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2602 ;
2603
2604 static ssize_t
2605 tracing_readme_read(struct file *filp, char __user *ubuf,
2606                        size_t cnt, loff_t *ppos)
2607 {
2608         return simple_read_from_buffer(ubuf, cnt, ppos,
2609                                         readme_msg, strlen(readme_msg));
2610 }
2611
2612 static const struct file_operations tracing_readme_fops = {
2613         .open           = tracing_open_generic,
2614         .read           = tracing_readme_read,
2615         .llseek         = generic_file_llseek,
2616 };
2617
2618 static ssize_t
2619 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2620                                 size_t cnt, loff_t *ppos)
2621 {
2622         char *buf_comm;
2623         char *file_buf;
2624         char *buf;
2625         int len = 0;
2626         int pid;
2627         int i;
2628
2629         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2630         if (!file_buf)
2631                 return -ENOMEM;
2632
2633         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2634         if (!buf_comm) {
2635                 kfree(file_buf);
2636                 return -ENOMEM;
2637         }
2638
2639         buf = file_buf;
2640
2641         for (i = 0; i < SAVED_CMDLINES; i++) {
2642                 int r;
2643
2644                 pid = map_cmdline_to_pid[i];
2645                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2646                         continue;
2647
2648                 trace_find_cmdline(pid, buf_comm);
2649                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2650                 buf += r;
2651                 len += r;
2652         }
2653
2654         len = simple_read_from_buffer(ubuf, cnt, ppos,
2655                                       file_buf, len);
2656
2657         kfree(file_buf);
2658         kfree(buf_comm);
2659
2660         return len;
2661 }
2662
2663 static const struct file_operations tracing_saved_cmdlines_fops = {
2664     .open       = tracing_open_generic,
2665     .read       = tracing_saved_cmdlines_read,
2666     .llseek     = generic_file_llseek,
2667 };
2668
2669 static ssize_t
2670 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2671                   size_t cnt, loff_t *ppos)
2672 {
2673         char buf[64];
2674         int r;
2675
2676         r = sprintf(buf, "%u\n", tracer_enabled);
2677         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2678 }
2679
2680 static ssize_t
2681 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2682                    size_t cnt, loff_t *ppos)
2683 {
2684         struct trace_array *tr = filp->private_data;
2685         char buf[64];
2686         unsigned long val;
2687         int ret;
2688
2689         if (cnt >= sizeof(buf))
2690                 return -EINVAL;
2691
2692         if (copy_from_user(&buf, ubuf, cnt))
2693                 return -EFAULT;
2694
2695         buf[cnt] = 0;
2696
2697         ret = strict_strtoul(buf, 10, &val);
2698         if (ret < 0)
2699                 return ret;
2700
2701         val = !!val;
2702
2703         mutex_lock(&trace_types_lock);
2704         if (tracer_enabled ^ val) {
2705                 if (val) {
2706                         tracer_enabled = 1;
2707                         if (current_trace->start)
2708                                 current_trace->start(tr);
2709                         tracing_start();
2710                 } else {
2711                         tracer_enabled = 0;
2712                         tracing_stop();
2713                         if (current_trace->stop)
2714                                 current_trace->stop(tr);
2715                 }
2716         }
2717         mutex_unlock(&trace_types_lock);
2718
2719         *ppos += cnt;
2720
2721         return cnt;
2722 }
2723
2724 static ssize_t
2725 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2726                        size_t cnt, loff_t *ppos)
2727 {
2728         char buf[MAX_TRACER_SIZE+2];
2729         int r;
2730
2731         mutex_lock(&trace_types_lock);
2732         if (current_trace)
2733                 r = sprintf(buf, "%s\n", current_trace->name);
2734         else
2735                 r = sprintf(buf, "\n");
2736         mutex_unlock(&trace_types_lock);
2737
2738         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2739 }
2740
2741 int tracer_init(struct tracer *t, struct trace_array *tr)
2742 {
2743         tracing_reset_online_cpus(tr);
2744         return t->init(tr);
2745 }
2746
2747 static int tracing_resize_ring_buffer(unsigned long size)
2748 {
2749         int ret;
2750
2751         /*
2752          * If kernel or user changes the size of the ring buffer
2753          * we use the size that was given, and we can forget about
2754          * expanding it later.
2755          */
2756         ring_buffer_expanded = 1;
2757
2758         ret = ring_buffer_resize(global_trace.buffer, size);
2759         if (ret < 0)
2760                 return ret;
2761
2762         if (!current_trace->use_max_tr)
2763                 goto out;
2764
2765         ret = ring_buffer_resize(max_tr.buffer, size);
2766         if (ret < 0) {
2767                 int r;
2768
2769                 r = ring_buffer_resize(global_trace.buffer,
2770                                        global_trace.entries);
2771                 if (r < 0) {
2772                         /*
2773                          * AARGH! We are left with different
2774                          * size max buffer!!!!
2775                          * The max buffer is our "snapshot" buffer.
2776                          * When a tracer needs a snapshot (one of the
2777                          * latency tracers), it swaps the max buffer
2778                          * with the saved snap shot. We succeeded to
2779                          * update the size of the main buffer, but failed to
2780                          * update the size of the max buffer. But when we tried
2781                          * to reset the main buffer to the original size, we
2782                          * failed there too. This is very unlikely to
2783                          * happen, but if it does, warn and kill all
2784                          * tracing.
2785                          */
2786                         WARN_ON(1);
2787                         tracing_disabled = 1;
2788                 }
2789                 return ret;
2790         }
2791
2792         max_tr.entries = size;
2793  out:
2794         global_trace.entries = size;
2795
2796         return ret;
2797 }
2798
2799
2800 /**
2801  * tracing_update_buffers - used by tracing facility to expand ring buffers
2802  *
2803  * To save on memory when the tracing is never used on a system with it
2804  * configured in. The ring buffers are set to a minimum size. But once
2805  * a user starts to use the tracing facility, then they need to grow
2806  * to their default size.
2807  *
2808  * This function is to be called when a tracer is about to be used.
2809  */
2810 int tracing_update_buffers(void)
2811 {
2812         int ret = 0;
2813
2814         mutex_lock(&trace_types_lock);
2815         if (!ring_buffer_expanded)
2816                 ret = tracing_resize_ring_buffer(trace_buf_size);
2817         mutex_unlock(&trace_types_lock);
2818
2819         return ret;
2820 }
2821
2822 struct trace_option_dentry;
2823
2824 static struct trace_option_dentry *
2825 create_trace_option_files(struct tracer *tracer);
2826
2827 static void
2828 destroy_trace_option_files(struct trace_option_dentry *topts);
2829
2830 static int tracing_set_tracer(const char *buf)
2831 {
2832         static struct trace_option_dentry *topts;
2833         struct trace_array *tr = &global_trace;
2834         struct tracer *t;
2835         int ret = 0;
2836
2837         mutex_lock(&trace_types_lock);
2838
2839         if (!ring_buffer_expanded) {
2840                 ret = tracing_resize_ring_buffer(trace_buf_size);
2841                 if (ret < 0)
2842                         goto out;
2843                 ret = 0;
2844         }
2845
2846         for (t = trace_types; t; t = t->next) {
2847                 if (strcmp(t->name, buf) == 0)
2848                         break;
2849         }
2850         if (!t) {
2851                 ret = -EINVAL;
2852                 goto out;
2853         }
2854         if (t == current_trace)
2855                 goto out;
2856
2857         trace_branch_disable();
2858         if (current_trace && current_trace->reset)
2859                 current_trace->reset(tr);
2860         if (current_trace && current_trace->use_max_tr) {
2861                 /*
2862                  * We don't free the ring buffer. instead, resize it because
2863                  * The max_tr ring buffer has some state (e.g. ring->clock) and
2864                  * we want preserve it.
2865                  */
2866                 ring_buffer_resize(max_tr.buffer, 1);
2867                 max_tr.entries = 1;
2868         }
2869         destroy_trace_option_files(topts);
2870
2871         current_trace = t;
2872
2873         topts = create_trace_option_files(current_trace);
2874         if (current_trace->use_max_tr) {
2875                 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
2876                 if (ret < 0)
2877                         goto out;
2878                 max_tr.entries = global_trace.entries;
2879         }
2880
2881         if (t->init) {
2882                 ret = tracer_init(t, tr);
2883                 if (ret)
2884                         goto out;
2885         }
2886
2887         trace_branch_enable(tr);
2888  out:
2889         mutex_unlock(&trace_types_lock);
2890
2891         return ret;
2892 }
2893
2894 static ssize_t
2895 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2896                         size_t cnt, loff_t *ppos)
2897 {
2898         char buf[MAX_TRACER_SIZE+1];
2899         int i;
2900         size_t ret;
2901         int err;
2902
2903         ret = cnt;
2904
2905         if (cnt > MAX_TRACER_SIZE)
2906                 cnt = MAX_TRACER_SIZE;
2907
2908         if (copy_from_user(&buf, ubuf, cnt))
2909                 return -EFAULT;
2910
2911         buf[cnt] = 0;
2912
2913         /* strip ending whitespace. */
2914         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2915                 buf[i] = 0;
2916
2917         err = tracing_set_tracer(buf);
2918         if (err)
2919                 return err;
2920
2921         *ppos += ret;
2922
2923         return ret;
2924 }
2925
2926 static ssize_t
2927 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2928                      size_t cnt, loff_t *ppos)
2929 {
2930         unsigned long *ptr = filp->private_data;
2931         char buf[64];
2932         int r;
2933
2934         r = snprintf(buf, sizeof(buf), "%ld\n",
2935                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2936         if (r > sizeof(buf))
2937                 r = sizeof(buf);
2938         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2939 }
2940
2941 static ssize_t
2942 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2943                       size_t cnt, loff_t *ppos)
2944 {
2945         unsigned long *ptr = filp->private_data;
2946         char buf[64];
2947         unsigned long val;
2948         int ret;
2949
2950         if (cnt >= sizeof(buf))
2951                 return -EINVAL;
2952
2953         if (copy_from_user(&buf, ubuf, cnt))
2954                 return -EFAULT;
2955
2956         buf[cnt] = 0;
2957
2958         ret = strict_strtoul(buf, 10, &val);
2959         if (ret < 0)
2960                 return ret;
2961
2962         *ptr = val * 1000;
2963
2964         return cnt;
2965 }
2966
2967 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2968 {
2969         long cpu_file = (long) inode->i_private;
2970         struct trace_iterator *iter;
2971         int ret = 0;
2972
2973         if (tracing_disabled)
2974                 return -ENODEV;
2975
2976         mutex_lock(&trace_types_lock);
2977
2978         /* create a buffer to store the information to pass to userspace */
2979         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2980         if (!iter) {
2981                 ret = -ENOMEM;
2982                 goto out;
2983         }
2984
2985         /*
2986          * We make a copy of the current tracer to avoid concurrent
2987          * changes on it while we are reading.
2988          */
2989         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2990         if (!iter->trace) {
2991                 ret = -ENOMEM;
2992                 goto fail;
2993         }
2994         if (current_trace)
2995                 *iter->trace = *current_trace;
2996
2997         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2998                 ret = -ENOMEM;
2999                 goto fail;
3000         }
3001
3002         /* trace pipe does not show start of buffer */
3003         cpumask_setall(iter->started);
3004
3005         if (trace_flags & TRACE_ITER_LATENCY_FMT)
3006                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3007
3008         iter->cpu_file = cpu_file;
3009         iter->tr = &global_trace;
3010         mutex_init(&iter->mutex);
3011         filp->private_data = iter;
3012
3013         if (iter->trace->pipe_open)
3014                 iter->trace->pipe_open(iter);
3015
3016         nonseekable_open(inode, filp);
3017 out:
3018         mutex_unlock(&trace_types_lock);
3019         return ret;
3020
3021 fail:
3022         kfree(iter->trace);
3023         kfree(iter);
3024         mutex_unlock(&trace_types_lock);
3025         return ret;
3026 }
3027
3028 static int tracing_release_pipe(struct inode *inode, struct file *file)
3029 {
3030         struct trace_iterator *iter = file->private_data;
3031
3032         mutex_lock(&trace_types_lock);
3033
3034         if (iter->trace->pipe_close)
3035                 iter->trace->pipe_close(iter);
3036
3037         mutex_unlock(&trace_types_lock);
3038
3039         free_cpumask_var(iter->started);
3040         mutex_destroy(&iter->mutex);
3041         kfree(iter->trace);
3042         kfree(iter);
3043
3044         return 0;
3045 }
3046
3047 static unsigned int
3048 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3049 {
3050         struct trace_iterator *iter = filp->private_data;
3051
3052         if (trace_flags & TRACE_ITER_BLOCK) {
3053                 /*
3054                  * Always select as readable when in blocking mode
3055                  */
3056                 return POLLIN | POLLRDNORM;
3057         } else {
3058                 if (!trace_empty(iter))
3059                         return POLLIN | POLLRDNORM;
3060                 poll_wait(filp, &trace_wait, poll_table);
3061                 if (!trace_empty(iter))
3062                         return POLLIN | POLLRDNORM;
3063
3064                 return 0;
3065         }
3066 }
3067
3068
3069 void default_wait_pipe(struct trace_iterator *iter)
3070 {
3071         DEFINE_WAIT(wait);
3072
3073         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3074
3075         if (trace_empty(iter))
3076                 schedule();
3077
3078         finish_wait(&trace_wait, &wait);
3079 }
3080
3081 /*
3082  * This is a make-shift waitqueue.
3083  * A tracer might use this callback on some rare cases:
3084  *
3085  *  1) the current tracer might hold the runqueue lock when it wakes up
3086  *     a reader, hence a deadlock (sched, function, and function graph tracers)
3087  *  2) the function tracers, trace all functions, we don't want
3088  *     the overhead of calling wake_up and friends
3089  *     (and tracing them too)
3090  *
3091  *     Anyway, this is really very primitive wakeup.
3092  */
3093 void poll_wait_pipe(struct trace_iterator *iter)
3094 {
3095         set_current_state(TASK_INTERRUPTIBLE);
3096         /* sleep for 100 msecs, and try again. */
3097         schedule_timeout(HZ / 10);
3098 }
3099
3100 /* Must be called with trace_types_lock mutex held. */
3101 static int tracing_wait_pipe(struct file *filp)
3102 {
3103         struct trace_iterator *iter = filp->private_data;
3104
3105         while (trace_empty(iter)) {
3106
3107                 if ((filp->f_flags & O_NONBLOCK)) {
3108                         return -EAGAIN;
3109                 }
3110
3111                 mutex_unlock(&iter->mutex);
3112
3113                 iter->trace->wait_pipe(iter);
3114
3115                 mutex_lock(&iter->mutex);
3116
3117                 if (signal_pending(current))
3118                         return -EINTR;
3119
3120                 /*
3121                  * We block until we read something and tracing is disabled.
3122                  * We still block if tracing is disabled, but we have never
3123                  * read anything. This allows a user to cat this file, and
3124                  * then enable tracing. But after we have read something,
3125                  * we give an EOF when tracing is again disabled.
3126                  *
3127                  * iter->pos will be 0 if we haven't read anything.
3128                  */
3129                 if (!tracer_enabled && iter->pos)
3130                         break;
3131         }
3132
3133         return 1;
3134 }
3135
3136 /*
3137  * Consumer reader.
3138  */
3139 static ssize_t
3140 tracing_read_pipe(struct file *filp, char __user *ubuf,
3141                   size_t cnt, loff_t *ppos)
3142 {
3143         struct trace_iterator *iter = filp->private_data;
3144         static struct tracer *old_tracer;
3145         ssize_t sret;
3146
3147         /* return any leftover data */
3148         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3149         if (sret != -EBUSY)
3150                 return sret;
3151
3152         trace_seq_init(&iter->seq);
3153
3154         /* copy the tracer to avoid using a global lock all around */
3155         mutex_lock(&trace_types_lock);
3156         if (unlikely(old_tracer != current_trace && current_trace)) {
3157                 old_tracer = current_trace;
3158                 *iter->trace = *current_trace;
3159         }
3160         mutex_unlock(&trace_types_lock);
3161
3162         /*
3163          * Avoid more than one consumer on a single file descriptor
3164          * This is just a matter of traces coherency, the ring buffer itself
3165          * is protected.
3166          */
3167         mutex_lock(&iter->mutex);
3168         if (iter->trace->read) {
3169                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3170                 if (sret)
3171                         goto out;
3172         }
3173
3174 waitagain:
3175         sret = tracing_wait_pipe(filp);
3176         if (sret <= 0)
3177                 goto out;
3178
3179         /* stop when tracing is finished */
3180         if (trace_empty(iter)) {
3181                 sret = 0;
3182                 goto out;
3183         }
3184
3185         if (cnt >= PAGE_SIZE)
3186                 cnt = PAGE_SIZE - 1;
3187
3188         /* reset all but tr, trace, and overruns */
3189         memset(&iter->seq, 0,
3190                sizeof(struct trace_iterator) -
3191                offsetof(struct trace_iterator, seq));
3192         iter->pos = -1;
3193
3194         trace_event_read_lock();
3195         trace_access_lock(iter->cpu_file);
3196         while (find_next_entry_inc(iter) != NULL) {
3197                 enum print_line_t ret;
3198                 int len = iter->seq.len;
3199
3200                 ret = print_trace_line(iter);
3201                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3202                         /* don't print partial lines */
3203                         iter->seq.len = len;
3204                         break;
3205                 }
3206                 if (ret != TRACE_TYPE_NO_CONSUME)
3207                         trace_consume(iter);
3208
3209                 if (iter->seq.len >= cnt)
3210                         break;
3211         }
3212         trace_access_unlock(iter->cpu_file);
3213         trace_event_read_unlock();
3214
3215         /* Now copy what we have to the user */
3216         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3217         if (iter->seq.readpos >= iter->seq.len)
3218                 trace_seq_init(&iter->seq);
3219
3220         /*
3221          * If there was nothing to send to user, inspite of consuming trace
3222          * entries, go back to wait for more entries.
3223          */
3224         if (sret == -EBUSY)
3225                 goto waitagain;
3226
3227 out:
3228         mutex_unlock(&iter->mutex);
3229
3230         return sret;
3231 }
3232
3233 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3234                                      struct pipe_buffer *buf)
3235 {
3236         __free_page(buf->page);
3237 }
3238
3239 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3240                                      unsigned int idx)
3241 {
3242         __free_page(spd->pages[idx]);
3243 }
3244
3245 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3246         .can_merge              = 0,
3247         .map                    = generic_pipe_buf_map,
3248         .unmap                  = generic_pipe_buf_unmap,
3249         .confirm                = generic_pipe_buf_confirm,
3250         .release                = tracing_pipe_buf_release,
3251         .steal                  = generic_pipe_buf_steal,
3252         .get                    = generic_pipe_buf_get,
3253 };
3254
3255 static size_t
3256 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3257 {
3258         size_t count;
3259         int ret;
3260
3261         /* Seq buffer is page-sized, exactly what we need. */
3262         for (;;) {
3263                 count = iter->seq.len;
3264                 ret = print_trace_line(iter);
3265                 count = iter->seq.len - count;
3266                 if (rem < count) {
3267                         rem = 0;
3268                         iter->seq.len -= count;
3269                         break;
3270                 }
3271                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3272                         iter->seq.len -= count;
3273                         break;
3274                 }
3275
3276                 if (ret != TRACE_TYPE_NO_CONSUME)
3277                         trace_consume(iter);
3278                 rem -= count;
3279                 if (!find_next_entry_inc(iter)) {
3280                         rem = 0;
3281                         iter->ent = NULL;
3282                         break;
3283                 }
3284         }
3285
3286         return rem;
3287 }
3288
3289 static ssize_t tracing_splice_read_pipe(struct file *filp,
3290                                         loff_t *ppos,
3291                                         struct pipe_inode_info *pipe,
3292                                         size_t len,
3293                                         unsigned int flags)
3294 {
3295         struct page *pages_def[PIPE_DEF_BUFFERS];
3296         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3297         struct trace_iterator *iter = filp->private_data;
3298         struct splice_pipe_desc spd = {
3299                 .pages          = pages_def,
3300                 .partial        = partial_def,
3301                 .nr_pages       = 0, /* This gets updated below. */
3302                 .flags          = flags,
3303                 .ops            = &tracing_pipe_buf_ops,
3304                 .spd_release    = tracing_spd_release_pipe,
3305         };
3306         static struct tracer *old_tracer;
3307         ssize_t ret;
3308         size_t rem;
3309         unsigned int i;
3310
3311         if (splice_grow_spd(pipe, &spd))
3312                 return -ENOMEM;
3313
3314         /* copy the tracer to avoid using a global lock all around */
3315         mutex_lock(&trace_types_lock);
3316         if (unlikely(old_tracer != current_trace && current_trace)) {
3317                 old_tracer = current_trace;
3318                 *iter->trace = *current_trace;
3319         }
3320         mutex_unlock(&trace_types_lock);
3321
3322         mutex_lock(&iter->mutex);
3323
3324         if (iter->trace->splice_read) {
3325                 ret = iter->trace->splice_read(iter, filp,
3326                                                ppos, pipe, len, flags);
3327                 if (ret)
3328                         goto out_err;
3329         }
3330
3331         ret = tracing_wait_pipe(filp);
3332         if (ret <= 0)
3333                 goto out_err;
3334
3335         if (!iter->ent && !find_next_entry_inc(iter)) {
3336                 ret = -EFAULT;
3337                 goto out_err;
3338         }
3339
3340         trace_event_read_lock();
3341         trace_access_lock(iter->cpu_file);
3342
3343         /* Fill as many pages as possible. */
3344         for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3345                 spd.pages[i] = alloc_page(GFP_KERNEL);
3346                 if (!spd.pages[i])
3347                         break;
3348
3349                 rem = tracing_fill_pipe_page(rem, iter);
3350
3351                 /* Copy the data into the page, so we can start over. */
3352                 ret = trace_seq_to_buffer(&iter->seq,
3353                                           page_address(spd.pages[i]),
3354                                           iter->seq.len);
3355                 if (ret < 0) {
3356                         __free_page(spd.pages[i]);
3357                         break;
3358                 }
3359                 spd.partial[i].offset = 0;
3360                 spd.partial[i].len = iter->seq.len;
3361
3362                 trace_seq_init(&iter->seq);
3363         }
3364
3365         trace_access_unlock(iter->cpu_file);
3366         trace_event_read_unlock();
3367         mutex_unlock(&iter->mutex);
3368
3369         spd.nr_pages = i;
3370
3371         ret = splice_to_pipe(pipe, &spd);
3372 out:
3373         splice_shrink_spd(pipe, &spd);
3374         return ret;
3375
3376 out_err:
3377         mutex_unlock(&iter->mutex);
3378         goto out;
3379 }
3380
3381 static ssize_t
3382 tracing_entries_read(struct file *filp, char __user *ubuf,
3383                      size_t cnt, loff_t *ppos)
3384 {
3385         struct trace_array *tr = filp->private_data;
3386         char buf[96];
3387         int r;
3388
3389         mutex_lock(&trace_types_lock);
3390         if (!ring_buffer_expanded)
3391                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3392                             tr->entries >> 10,
3393                             trace_buf_size >> 10);
3394         else
3395                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3396         mutex_unlock(&trace_types_lock);
3397
3398         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3399 }
3400
3401 static ssize_t
3402 tracing_entries_write(struct file *filp, const char __user *ubuf,
3403                       size_t cnt, loff_t *ppos)
3404 {
3405         unsigned long val;
3406         char buf[64];
3407         int ret, cpu;
3408
3409         if (cnt >= sizeof(buf))
3410                 return -EINVAL;
3411
3412         if (copy_from_user(&buf, ubuf, cnt))
3413                 return -EFAULT;
3414
3415         buf[cnt] = 0;
3416
3417         ret = strict_strtoul(buf, 10, &val);
3418         if (ret < 0)
3419                 return ret;
3420
3421         /* must have at least 1 entry */
3422         if (!val)
3423                 return -EINVAL;
3424
3425         mutex_lock(&trace_types_lock);
3426
3427         tracing_stop();
3428
3429         /* disable all cpu buffers */
3430         for_each_tracing_cpu(cpu) {
3431                 if (global_trace.data[cpu])
3432                         atomic_inc(&global_trace.data[cpu]->disabled);
3433                 if (max_tr.data[cpu])
3434                         atomic_inc(&max_tr.data[cpu]->disabled);
3435         }
3436
3437         /* value is in KB */
3438         val <<= 10;
3439
3440         if (val != global_trace.entries) {
3441                 ret = tracing_resize_ring_buffer(val);
3442                 if (ret < 0) {
3443                         cnt = ret;
3444                         goto out;
3445                 }
3446         }
3447
3448         *ppos += cnt;
3449
3450         /* If check pages failed, return ENOMEM */
3451         if (tracing_disabled)
3452                 cnt = -ENOMEM;
3453  out:
3454         for_each_tracing_cpu(cpu) {
3455                 if (global_trace.data[cpu])
3456                         atomic_dec(&global_trace.data[cpu]->disabled);
3457                 if (max_tr.data[cpu])
3458                         atomic_dec(&max_tr.data[cpu]->disabled);
3459         }
3460
3461         tracing_start();
3462         mutex_unlock(&trace_types_lock);
3463
3464         return cnt;
3465 }
3466
3467 static int mark_printk(const char *fmt, ...)
3468 {
3469         int ret;
3470         va_list args;
3471         va_start(args, fmt);
3472         ret = trace_vprintk(0, fmt, args);
3473         va_end(args);
3474         return ret;
3475 }
3476
3477 static ssize_t
3478 tracing_mark_write(struct file *filp, const char __user *ubuf,
3479                                         size_t cnt, loff_t *fpos)
3480 {
3481         char *buf;
3482
3483         if (tracing_disabled)
3484                 return -EINVAL;
3485
3486         if (cnt > TRACE_BUF_SIZE)
3487                 cnt = TRACE_BUF_SIZE;
3488
3489         buf = kmalloc(cnt + 2, GFP_KERNEL);
3490         if (buf == NULL)
3491                 return -ENOMEM;
3492
3493         if (copy_from_user(buf, ubuf, cnt)) {
3494                 kfree(buf);
3495                 return -EFAULT;
3496         }
3497         if (buf[cnt-1] != '\n') {
3498                 buf[cnt] = '\n';
3499                 buf[cnt+1] = '\0';
3500         } else
3501                 buf[cnt] = '\0';
3502
3503         cnt = mark_printk("%s", buf);
3504         kfree(buf);
3505         *fpos += cnt;
3506
3507         return cnt;
3508 }
3509
3510 static int tracing_clock_show(struct seq_file *m, void *v)
3511 {
3512         int i;
3513
3514         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3515                 seq_printf(m,
3516                         "%s%s%s%s", i ? " " : "",
3517                         i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3518                         i == trace_clock_id ? "]" : "");
3519         seq_putc(m, '\n');
3520
3521         return 0;
3522 }
3523
3524 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3525                                    size_t cnt, loff_t *fpos)
3526 {
3527         char buf[64];
3528         const char *clockstr;
3529         int i;
3530
3531         if (cnt >= sizeof(buf))
3532                 return -EINVAL;
3533
3534         if (copy_from_user(&buf, ubuf, cnt))
3535                 return -EFAULT;
3536
3537         buf[cnt] = 0;
3538
3539         clockstr = strstrip(buf);
3540
3541         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3542                 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3543                         break;
3544         }
3545         if (i == ARRAY_SIZE(trace_clocks))
3546                 return -EINVAL;
3547
3548         trace_clock_id = i;
3549
3550         mutex_lock(&trace_types_lock);
3551
3552         ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3553         if (max_tr.buffer)
3554                 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3555
3556         mutex_unlock(&trace_types_lock);
3557
3558         *fpos += cnt;
3559
3560         return cnt;
3561 }
3562
3563 static int tracing_clock_open(struct inode *inode, struct file *file)
3564 {
3565         if (tracing_disabled)
3566                 return -ENODEV;
3567         return single_open(file, tracing_clock_show, NULL);
3568 }
3569
3570 static const struct file_operations tracing_max_lat_fops = {
3571         .open           = tracing_open_generic,
3572         .read           = tracing_max_lat_read,
3573         .write          = tracing_max_lat_write,
3574         .llseek         = generic_file_llseek,
3575 };
3576
3577 static const struct file_operations tracing_ctrl_fops = {
3578         .open           = tracing_open_generic,
3579         .read           = tracing_ctrl_read,
3580         .write          = tracing_ctrl_write,
3581         .llseek         = generic_file_llseek,
3582 };
3583
3584 static const struct file_operations set_tracer_fops = {
3585         .open           = tracing_open_generic,
3586         .read           = tracing_set_trace_read,
3587         .write          = tracing_set_trace_write,
3588         .llseek         = generic_file_llseek,
3589 };
3590
3591 static const struct file_operations tracing_pipe_fops = {
3592         .open           = tracing_open_pipe,
3593         .poll           = tracing_poll_pipe,
3594         .read           = tracing_read_pipe,
3595         .splice_read    = tracing_splice_read_pipe,
3596         .release        = tracing_release_pipe,
3597         .llseek         = no_llseek,
3598 };
3599
3600 static const struct file_operations tracing_entries_fops = {
3601         .open           = tracing_open_generic,
3602         .read           = tracing_entries_read,
3603         .write          = tracing_entries_write,
3604         .llseek         = generic_file_llseek,
3605 };
3606
3607 static const struct file_operations tracing_mark_fops = {
3608         .open           = tracing_open_generic,
3609         .write          = tracing_mark_write,
3610         .llseek         = generic_file_llseek,
3611 };
3612
3613 static const struct file_operations trace_clock_fops = {
3614         .open           = tracing_clock_open,
3615         .read           = seq_read,
3616         .llseek         = seq_lseek,
3617         .release        = single_release,
3618         .write          = tracing_clock_write,
3619 };
3620
3621 struct ftrace_buffer_info {
3622         struct trace_array      *tr;
3623         void                    *spare;
3624         int                     cpu;
3625         unsigned int            read;
3626 };
3627
3628 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3629 {
3630         int cpu = (int)(long)inode->i_private;
3631         struct ftrace_buffer_info *info;
3632
3633         if (tracing_disabled)
3634                 return -ENODEV;
3635
3636         info = kzalloc(sizeof(*info), GFP_KERNEL);
3637         if (!info)
3638                 return -ENOMEM;
3639
3640         info->tr        = &global_trace;
3641         info->cpu       = cpu;
3642         info->spare     = NULL;
3643         /* Force reading ring buffer for first read */
3644         info->read      = (unsigned int)-1;
3645
3646         filp->private_data = info;
3647
3648         return nonseekable_open(inode, filp);
3649 }
3650
3651 static ssize_t
3652 tracing_buffers_read(struct file *filp, char __user *ubuf,
3653                      size_t count, loff_t *ppos)
3654 {
3655         struct ftrace_buffer_info *info = filp->private_data;
3656         ssize_t ret;
3657         size_t size;
3658
3659         if (!count)
3660                 return 0;
3661
3662         if (!info->spare)
3663                 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3664         if (!info->spare)
3665                 return -ENOMEM;
3666
3667         /* Do we have previous read data to read? */
3668         if (info->read < PAGE_SIZE)
3669                 goto read;
3670
3671         info->read = 0;
3672
3673         trace_access_lock(info->cpu);
3674         ret = ring_buffer_read_page(info->tr->buffer,
3675                                     &info->spare,
3676                                     count,
3677                                     info->cpu, 0);
3678         trace_access_unlock(info->cpu);
3679         if (ret < 0)
3680                 return 0;
3681
3682 read:
3683         size = PAGE_SIZE - info->read;
3684         if (size > count)
3685                 size = count;
3686
3687         ret = copy_to_user(ubuf, info->spare + info->read, size);
3688         if (ret == size)
3689                 return -EFAULT;
3690         size -= ret;
3691
3692         *ppos += size;
3693         info->read += size;
3694
3695         return size;
3696 }
3697
3698 static int tracing_buffers_release(struct inode *inode, struct file *file)
3699 {
3700         struct ftrace_buffer_info *info = file->private_data;
3701
3702         if (info->spare)
3703                 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3704         kfree(info);
3705
3706         return 0;
3707 }
3708
3709 struct buffer_ref {
3710         struct ring_buffer      *buffer;
3711         void                    *page;
3712         int                     ref;
3713 };
3714
3715 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3716                                     struct pipe_buffer *buf)
3717 {
3718         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3719
3720         if (--ref->ref)
3721                 return;
3722
3723         ring_buffer_free_read_page(ref->buffer, ref->page);
3724         kfree(ref);
3725         buf->private = 0;
3726 }
3727
3728 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3729                                  struct pipe_buffer *buf)
3730 {
3731         return 1;
3732 }
3733
3734 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3735                                 struct pipe_buffer *buf)
3736 {
3737         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3738
3739         ref->ref++;
3740 }
3741
3742 /* Pipe buffer operations for a buffer. */
3743 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3744         .can_merge              = 0,
3745         .map                    = generic_pipe_buf_map,
3746         .unmap                  = generic_pipe_buf_unmap,
3747         .confirm                = generic_pipe_buf_confirm,
3748         .release                = buffer_pipe_buf_release,
3749         .steal                  = buffer_pipe_buf_steal,
3750         .get                    = buffer_pipe_buf_get,
3751 };
3752
3753 /*
3754  * Callback from splice_to_pipe(), if we need to release some pages
3755  * at the end of the spd in case we error'ed out in filling the pipe.
3756  */
3757 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3758 {
3759         struct buffer_ref *ref =
3760                 (struct buffer_ref *)spd->partial[i].private;
3761
3762         if (--ref->ref)
3763                 return;
3764
3765         ring_buffer_free_read_page(ref->buffer, ref->page);
3766         kfree(ref);
3767         spd->partial[i].private = 0;
3768 }
3769
3770 static ssize_t
3771 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3772                             struct pipe_inode_info *pipe, size_t len,
3773                             unsigned int flags)
3774 {
3775         struct ftrace_buffer_info *info = file->private_data;
3776         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3777         struct page *pages_def[PIPE_DEF_BUFFERS];
3778         struct splice_pipe_desc spd = {
3779                 .pages          = pages_def,
3780                 .partial        = partial_def,
3781                 .flags          = flags,
3782                 .ops            = &buffer_pipe_buf_ops,
3783                 .spd_release    = buffer_spd_release,
3784         };
3785         struct buffer_ref *ref;
3786         int entries, size, i;
3787         size_t ret;
3788
3789         if (splice_grow_spd(pipe, &spd))
3790                 return -ENOMEM;
3791
3792         if (*ppos & (PAGE_SIZE - 1)) {
3793                 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3794                 ret = -EINVAL;
3795                 goto out;
3796         }
3797
3798         if (len & (PAGE_SIZE - 1)) {
3799                 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3800                 if (len < PAGE_SIZE) {
3801                         ret = -EINVAL;
3802                         goto out;
3803                 }
3804                 len &= PAGE_MASK;
3805         }
3806
3807         trace_access_lock(info->cpu);
3808         entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3809
3810         for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3811                 struct page *page;
3812                 int r;
3813
3814                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3815                 if (!ref)
3816                         break;
3817
3818                 ref->ref = 1;
3819                 ref->buffer = info->tr->buffer;
3820                 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3821                 if (!ref->page) {
3822                         kfree(ref);
3823                         break;
3824                 }
3825
3826                 r = ring_buffer_read_page(ref->buffer, &ref->page,
3827                                           len, info->cpu, 1);
3828                 if (r < 0) {
3829                         ring_buffer_free_read_page(ref->buffer,
3830                                                    ref->page);
3831                         kfree(ref);
3832                         break;
3833                 }
3834
3835                 /*
3836                  * zero out any left over data, this is going to
3837                  * user land.
3838                  */
3839                 size = ring_buffer_page_len(ref->page);
3840                 if (size < PAGE_SIZE)
3841                         memset(ref->page + size, 0, PAGE_SIZE - size);
3842
3843                 page = virt_to_page(ref->page);
3844
3845                 spd.pages[i] = page;
3846                 spd.partial[i].len = PAGE_SIZE;
3847                 spd.partial[i].offset = 0;
3848                 spd.partial[i].private = (unsigned long)ref;
3849                 spd.nr_pages++;
3850                 *ppos += PAGE_SIZE;
3851
3852                 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3853         }
3854
3855         trace_access_unlock(info->cpu);
3856         spd.nr_pages = i;
3857
3858         /* did we read anything? */
3859         if (!spd.nr_pages) {
3860                 if (flags & SPLICE_F_NONBLOCK)
3861                         ret = -EAGAIN;
3862                 else
3863                         ret = 0;
3864                 /* TODO: block */
3865                 goto out;
3866         }
3867
3868         ret = splice_to_pipe(pipe, &spd);
3869         splice_shrink_spd(pipe, &spd);
3870 out:
3871         return ret;
3872 }
3873
3874 static const struct file_operations tracing_buffers_fops = {
3875         .open           = tracing_buffers_open,
3876         .read           = tracing_buffers_read,
3877         .release        = tracing_buffers_release,
3878         .splice_read    = tracing_buffers_splice_read,
3879         .llseek         = no_llseek,
3880 };
3881
3882 static ssize_t
3883 tracing_stats_read(struct file *filp, char __user *ubuf,
3884                    size_t count, loff_t *ppos)
3885 {
3886         unsigned long cpu = (unsigned long)filp->private_data;
3887         struct trace_array *tr = &global_trace;
3888         struct trace_seq *s;
3889         unsigned long cnt;
3890
3891         s = kmalloc(sizeof(*s), GFP_KERNEL);
3892         if (!s)
3893                 return -ENOMEM;
3894
3895         trace_seq_init(s);
3896
3897         cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3898         trace_seq_printf(s, "entries: %ld\n", cnt);
3899
3900         cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3901         trace_seq_printf(s, "overrun: %ld\n", cnt);
3902
3903         cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3904         trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3905
3906         count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3907
3908         kfree(s);
3909
3910         return count;
3911 }
3912
3913 static const struct file_operations tracing_stats_fops = {
3914         .open           = tracing_open_generic,
3915         .read           = tracing_stats_read,
3916         .llseek         = generic_file_llseek,
3917 };
3918
3919 #ifdef CONFIG_DYNAMIC_FTRACE
3920
3921 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3922 {
3923         return 0;
3924 }
3925
3926 static ssize_t
3927 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3928                   size_t cnt, loff_t *ppos)
3929 {
3930         static char ftrace_dyn_info_buffer[1024];
3931         static DEFINE_MUTEX(dyn_info_mutex);
3932         unsigned long *p = filp->private_data;
3933         char *buf = ftrace_dyn_info_buffer;
3934         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3935         int r;
3936
3937         mutex_lock(&dyn_info_mutex);
3938         r = sprintf(buf, "%ld ", *p);
3939
3940         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3941         buf[r++] = '\n';
3942
3943         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3944
3945         mutex_unlock(&dyn_info_mutex);
3946
3947         return r;
3948 }
3949
3950 static const struct file_operations tracing_dyn_info_fops = {
3951         .open           = tracing_open_generic,
3952         .read           = tracing_read_dyn_info,
3953         .llseek         = generic_file_llseek,
3954 };
3955 #endif
3956
3957 static struct dentry *d_tracer;
3958
3959 struct dentry *tracing_init_dentry(void)
3960 {
3961         static int once;
3962
3963         if (d_tracer)
3964                 return d_tracer;
3965
3966         if (!debugfs_initialized())
3967                 return NULL;
3968
3969         d_tracer = debugfs_create_dir("tracing", NULL);
3970
3971         if (!d_tracer && !once) {
3972                 once = 1;
3973                 pr_warning("Could not create debugfs directory 'tracing'\n");
3974                 return NULL;
3975         }
3976
3977         return d_tracer;
3978 }
3979
3980 static struct dentry *d_percpu;
3981
3982 struct dentry *tracing_dentry_percpu(void)
3983 {
3984         static int once;
3985         struct dentry *d_tracer;
3986
3987         if (d_percpu)
3988                 return d_percpu;
3989
3990         d_tracer = tracing_init_dentry();
3991
3992         if (!d_tracer)
3993                 return NULL;
3994
3995         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3996
3997         if (!d_percpu && !once) {
3998                 once = 1;
3999                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4000                 return NULL;
4001         }
4002
4003         return d_percpu;
4004 }
4005
4006 static void tracing_init_debugfs_percpu(long cpu)
4007 {
4008         struct dentry *d_percpu = tracing_dentry_percpu();
4009         struct dentry *d_cpu;
4010         /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
4011         char cpu_dir[7];
4012
4013         if (cpu > 999 || cpu < 0)
4014                 return;
4015
4016         sprintf(cpu_dir, "cpu%ld", cpu);
4017         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4018         if (!d_cpu) {
4019                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4020                 return;
4021         }
4022
4023         /* per cpu trace_pipe */
4024         trace_create_file("trace_pipe", 0444, d_cpu,
4025                         (void *) cpu, &tracing_pipe_fops);
4026
4027         /* per cpu trace */
4028         trace_create_file("trace", 0644, d_cpu,
4029                         (void *) cpu, &tracing_fops);
4030
4031         trace_create_file("trace_pipe_raw", 0444, d_cpu,
4032                         (void *) cpu, &tracing_buffers_fops);
4033
4034         trace_create_file("stats", 0444, d_cpu,
4035                         (void *) cpu, &tracing_stats_fops);
4036 }
4037
4038 #ifdef CONFIG_FTRACE_SELFTEST
4039 /* Let selftest have access to static functions in this file */
4040 #include "trace_selftest.c"
4041 #endif
4042
4043 struct trace_option_dentry {
4044         struct tracer_opt               *opt;
4045         struct tracer_flags             *flags;
4046         struct dentry                   *entry;
4047 };
4048
4049 static ssize_t
4050 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4051                         loff_t *ppos)
4052 {
4053         struct trace_option_dentry *topt = filp->private_data;
4054         char *buf;
4055
4056         if (topt->flags->val & topt->opt->bit)
4057                 buf = "1\n";
4058         else
4059                 buf = "0\n";
4060
4061         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4062 }
4063
4064 static ssize_t
4065 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4066                          loff_t *ppos)
4067 {
4068         struct trace_option_dentry *topt = filp->private_data;
4069         unsigned long val;
4070         char buf[64];
4071         int ret;
4072
4073         if (cnt >= sizeof(buf))
4074                 return -EINVAL;
4075
4076         if (copy_from_user(&buf, ubuf, cnt))
4077                 return -EFAULT;
4078
4079         buf[cnt] = 0;
4080
4081         ret = strict_strtoul(buf, 10, &val);
4082         if (ret < 0)
4083                 return ret;
4084
4085         if (val != 0 && val != 1)
4086                 return -EINVAL;
4087
4088         if (!!(topt->flags->val & topt->opt->bit) != val) {
4089                 mutex_lock(&trace_types_lock);
4090                 ret = __set_tracer_option(current_trace, topt->flags,
4091                                           topt->opt, !val);
4092                 mutex_unlock(&trace_types_lock);
4093                 if (ret)
4094                         return ret;
4095         }
4096
4097         *ppos += cnt;
4098
4099         return cnt;
4100 }
4101
4102
4103 static const struct file_operations trace_options_fops = {
4104         .open = tracing_open_generic,
4105         .read = trace_options_read,
4106         .write = trace_options_write,
4107         .llseek = generic_file_llseek,
4108 };
4109
4110 static ssize_t
4111 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4112                         loff_t *ppos)
4113 {
4114         long index = (long)filp->private_data;
4115         char *buf;
4116
4117         if (trace_flags & (1 << index))
4118                 buf = "1\n";
4119         else
4120                 buf = "0\n";
4121
4122         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4123 }
4124
4125 static ssize_t
4126 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4127                          loff_t *ppos)
4128 {
4129         long index = (long)filp->private_data;
4130         char buf[64];
4131         unsigned long val;
4132         int ret;
4133
4134         if (cnt >= sizeof(buf))
4135                 return -EINVAL;
4136
4137         if (copy_from_user(&buf, ubuf, cnt))
4138                 return -EFAULT;
4139
4140         buf[cnt] = 0;
4141
4142         ret = strict_strtoul(buf, 10, &val);
4143         if (ret < 0)
4144                 return ret;
4145
4146         if (val != 0 && val != 1)
4147                 return -EINVAL;
4148         set_tracer_flags(1 << index, val);
4149
4150         *ppos += cnt;
4151
4152         return cnt;
4153 }
4154
4155 static const struct file_operations trace_options_core_fops = {
4156         .open = tracing_open_generic,
4157         .read = trace_options_core_read,
4158         .write = trace_options_core_write,
4159         .llseek = generic_file_llseek,
4160 };
4161
4162 struct dentry *trace_create_file(const char *name,
4163                                  mode_t mode,
4164                                  struct dentry *parent,
4165                                  void *data,
4166                                  const struct file_operations *fops)
4167 {
4168         struct dentry *ret;
4169
4170         ret = debugfs_create_file(name, mode, parent, data, fops);
4171         if (!ret)
4172                 pr_warning("Could not create debugfs '%s' entry\n", name);
4173
4174         return ret;
4175 }
4176
4177
4178 static struct dentry *trace_options_init_dentry(void)
4179 {
4180         struct dentry *d_tracer;
4181         static struct dentry *t_options;
4182
4183         if (t_options)
4184                 return t_options;
4185
4186         d_tracer = tracing_init_dentry();
4187         if (!d_tracer)
4188                 return NULL;
4189
4190         t_options = debugfs_create_dir("options", d_tracer);
4191         if (!t_options) {
4192                 pr_warning("Could not create debugfs directory 'options'\n");
4193                 return NULL;
4194         }
4195
4196         return t_options;
4197 }
4198
4199 static void
4200 create_trace_option_file(struct trace_option_dentry *topt,
4201                          struct tracer_flags *flags,
4202                          struct tracer_opt *opt)
4203 {
4204         struct dentry *t_options;
4205
4206         t_options = trace_options_init_dentry();
4207         if (!t_options)
4208                 return;
4209
4210         topt->flags = flags;
4211         topt->opt = opt;
4212
4213         topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4214                                     &trace_options_fops);
4215
4216 }
4217
4218 static struct trace_option_dentry *
4219 create_trace_option_files(struct tracer *tracer)
4220 {
4221         struct trace_option_dentry *topts;
4222         struct tracer_flags *flags;
4223         struct tracer_opt *opts;
4224         int cnt;
4225
4226         if (!tracer)
4227                 return NULL;
4228
4229         flags = tracer->flags;
4230
4231         if (!flags || !flags->opts)
4232                 return NULL;
4233
4234         opts = flags->opts;
4235
4236         for (cnt = 0; opts[cnt].name; cnt++)
4237                 ;
4238
4239         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4240         if (!topts)
4241                 return NULL;
4242
4243         for (cnt = 0; opts[cnt].name; cnt++)
4244                 create_trace_option_file(&topts[cnt], flags,
4245                                          &opts[cnt]);
4246
4247         return topts;
4248 }
4249
4250 static void
4251 destroy_trace_option_files(struct trace_option_dentry *topts)
4252 {
4253         int cnt;
4254
4255         if (!topts)
4256                 return;
4257
4258         for (cnt = 0; topts[cnt].opt; cnt++) {
4259                 if (topts[cnt].entry)
4260                         debugfs_remove(topts[cnt].entry);
4261         }
4262
4263         kfree(topts);
4264 }
4265
4266 static struct dentry *
4267 create_trace_option_core_file(const char *option, long index)
4268 {
4269         struct dentry *t_options;
4270
4271         t_options = trace_options_init_dentry();
4272         if (!t_options)
4273                 return NULL;
4274
4275         return trace_create_file(option, 0644, t_options, (void *)index,
4276                                     &trace_options_core_fops);
4277 }
4278
4279 static __init void create_trace_options_dir(void)
4280 {
4281         struct dentry *t_options;
4282         int i;
4283
4284         t_options = trace_options_init_dentry();
4285         if (!t_options)
4286                 return;
4287
4288         for (i = 0; trace_options[i]; i++)
4289                 create_trace_option_core_file(trace_options[i], i);
4290 }
4291
4292 static __init int tracer_init_debugfs(void)
4293 {
4294         struct dentry *d_tracer;
4295         int cpu;
4296
4297         trace_access_lock_init();
4298
4299         d_tracer = tracing_init_dentry();
4300
4301         trace_create_file("tracing_enabled", 0644, d_tracer,
4302                         &global_trace, &tracing_ctrl_fops);
4303
4304         trace_create_file("trace_options", 0644, d_tracer,
4305                         NULL, &tracing_iter_fops);
4306
4307         trace_create_file("tracing_cpumask", 0644, d_tracer,
4308                         NULL, &tracing_cpumask_fops);
4309
4310         trace_create_file("trace", 0644, d_tracer,
4311                         (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4312
4313         trace_create_file("available_tracers", 0444, d_tracer,
4314                         &global_trace, &show_traces_fops);
4315
4316         trace_create_file("current_tracer", 0644, d_tracer,
4317                         &global_trace, &set_tracer_fops);
4318
4319 #ifdef CONFIG_TRACER_MAX_TRACE
4320         trace_create_file("tracing_max_latency", 0644, d_tracer,
4321                         &tracing_max_latency, &tracing_max_lat_fops);
4322 #endif
4323
4324         trace_create_file("tracing_thresh", 0644, d_tracer,
4325                         &tracing_thresh, &tracing_max_lat_fops);
4326
4327         trace_create_file("README", 0444, d_tracer,
4328                         NULL, &tracing_readme_fops);
4329
4330         trace_create_file("trace_pipe", 0444, d_tracer,
4331                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4332
4333         trace_create_file("buffer_size_kb", 0644, d_tracer,
4334                         &global_trace, &tracing_entries_fops);
4335
4336         trace_create_file("trace_marker", 0220, d_tracer,
4337                         NULL, &tracing_mark_fops);
4338
4339         trace_create_file("saved_cmdlines", 0444, d_tracer,
4340                         NULL, &tracing_saved_cmdlines_fops);
4341
4342         trace_create_file("trace_clock", 0644, d_tracer, NULL,
4343                           &trace_clock_fops);
4344
4345 #ifdef CONFIG_DYNAMIC_FTRACE
4346         trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4347                         &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4348 #endif
4349
4350         create_trace_options_dir();
4351
4352         for_each_tracing_cpu(cpu)
4353                 tracing_init_debugfs_percpu(cpu);
4354
4355         return 0;
4356 }
4357
4358 static int trace_panic_handler(struct notifier_block *this,
4359                                unsigned long event, void *unused)
4360 {
4361         if (ftrace_dump_on_oops)
4362                 ftrace_dump(ftrace_dump_on_oops);
4363         return NOTIFY_OK;
4364 }
4365
4366 static struct notifier_block trace_panic_notifier = {
4367         .notifier_call  = trace_panic_handler,
4368         .next           = NULL,
4369         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
4370 };
4371
4372 static int trace_die_handler(struct notifier_block *self,
4373                              unsigned long val,
4374                              void *data)
4375 {
4376         switch (val) {
4377         case DIE_OOPS:
4378                 if (ftrace_dump_on_oops)
4379                         ftrace_dump(ftrace_dump_on_oops);
4380                 break;
4381         default:
4382                 break;
4383         }
4384         return NOTIFY_OK;
4385 }
4386
4387 static struct notifier_block trace_die_notifier = {
4388         .notifier_call = trace_die_handler,
4389         .priority = 200
4390 };
4391
4392 /*
4393  * printk is set to max of 1024, we really don't need it that big.
4394  * Nothing should be printing 1000 characters anyway.
4395  */
4396 #define TRACE_MAX_PRINT         1000
4397
4398 /*
4399  * Define here KERN_TRACE so that we have one place to modify
4400  * it if we decide to change what log level the ftrace dump
4401  * should be at.
4402  */
4403 #define KERN_TRACE              KERN_EMERG
4404
4405 static void
4406 trace_printk_seq(struct trace_seq *s)
4407 {
4408         /* Probably should print a warning here. */
4409         if (s->len >= 1000)
4410                 s->len = 1000;
4411
4412         /* should be zero ended, but we are paranoid. */
4413         s->buffer[s->len] = 0;
4414
4415         printk(KERN_TRACE "%s", s->buffer);
4416
4417         trace_seq_init(s);
4418 }
4419
4420 static void
4421 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4422 {
4423         static arch_spinlock_t ftrace_dump_lock =
4424                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4425         /* use static because iter can be a bit big for the stack */
4426         static struct trace_iterator iter;
4427         unsigned int old_userobj;
4428         static int dump_ran;
4429         unsigned long flags;
4430         int cnt = 0, cpu;
4431
4432         /* only one dump */
4433         local_irq_save(flags);
4434         arch_spin_lock(&ftrace_dump_lock);
4435         if (dump_ran)
4436                 goto out;
4437
4438         dump_ran = 1;
4439
4440         tracing_off();
4441
4442         if (disable_tracing)
4443                 ftrace_kill();
4444
4445         for_each_tracing_cpu(cpu) {
4446                 atomic_inc(&global_trace.data[cpu]->disabled);
4447         }
4448
4449         old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4450
4451         /* don't look at user memory in panic mode */
4452         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4453
4454         /* Simulate the iterator */
4455         iter.tr = &global_trace;
4456         iter.trace = current_trace;
4457
4458         switch (oops_dump_mode) {
4459         case DUMP_ALL:
4460                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4461                 break;
4462         case DUMP_ORIG:
4463                 iter.cpu_file = raw_smp_processor_id();
4464                 break;
4465         case DUMP_NONE:
4466                 goto out_enable;
4467         default:
4468                 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4469                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4470         }
4471
4472         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4473
4474         /*
4475          * We need to stop all tracing on all CPUS to read the
4476          * the next buffer. This is a bit expensive, but is
4477          * not done often. We fill all what we can read,
4478          * and then release the locks again.
4479          */
4480
4481         while (!trace_empty(&iter)) {
4482
4483                 if (!cnt)
4484                         printk(KERN_TRACE "---------------------------------\n");
4485
4486                 cnt++;
4487
4488                 /* reset all but tr, trace, and overruns */
4489                 memset(&iter.seq, 0,
4490                        sizeof(struct trace_iterator) -
4491                        offsetof(struct trace_iterator, seq));
4492                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4493                 iter.pos = -1;
4494
4495                 if (find_next_entry_inc(&iter) != NULL) {
4496                         int ret;
4497
4498                         ret = print_trace_line(&iter);
4499                         if (ret != TRACE_TYPE_NO_CONSUME)
4500                                 trace_consume(&iter);
4501                 }
4502
4503                 trace_printk_seq(&iter.seq);
4504         }
4505
4506         if (!cnt)
4507                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4508         else
4509                 printk(KERN_TRACE "---------------------------------\n");
4510
4511  out_enable:
4512         /* Re-enable tracing if requested */
4513         if (!disable_tracing) {
4514                 trace_flags |= old_userobj;
4515
4516                 for_each_tracing_cpu(cpu) {
4517                         atomic_dec(&global_trace.data[cpu]->disabled);
4518                 }
4519                 tracing_on();
4520         }
4521
4522  out:
4523         arch_spin_unlock(&ftrace_dump_lock);
4524         local_irq_restore(flags);
4525 }
4526
4527 /* By default: disable tracing after the dump */
4528 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4529 {
4530         __ftrace_dump(true, oops_dump_mode);
4531 }
4532
4533 __init static int tracer_alloc_buffers(void)
4534 {
4535         int ring_buf_size;
4536         int i;
4537         int ret = -ENOMEM;
4538
4539         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4540                 goto out;
4541
4542         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4543                 goto out_free_buffer_mask;
4544
4545         /* To save memory, keep the ring buffer size to its minimum */
4546         if (ring_buffer_expanded)
4547                 ring_buf_size = trace_buf_size;
4548         else
4549                 ring_buf_size = 1;
4550
4551         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4552         cpumask_copy(tracing_cpumask, cpu_all_mask);
4553
4554         /* TODO: make the number of buffers hot pluggable with CPUS */
4555         global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4556                                                    TRACE_BUFFER_FLAGS);
4557         if (!global_trace.buffer) {
4558                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4559                 WARN_ON(1);
4560                 goto out_free_cpumask;
4561         }
4562         global_trace.entries = ring_buffer_size(global_trace.buffer);
4563
4564
4565 #ifdef CONFIG_TRACER_MAX_TRACE
4566         max_tr.buffer = ring_buffer_alloc(1, TRACE_BUFFER_FLAGS);
4567         if (!max_tr.buffer) {
4568                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4569                 WARN_ON(1);
4570                 ring_buffer_free(global_trace.buffer);
4571                 goto out_free_cpumask;
4572         }
4573         max_tr.entries = 1;
4574 #endif
4575
4576         /* Allocate the first page for all buffers */
4577         for_each_tracing_cpu(i) {
4578                 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4579                 max_tr.data[i] = &per_cpu(max_tr_data, i);
4580         }
4581
4582         trace_init_cmdlines();
4583
4584         register_tracer(&nop_trace);
4585         current_trace = &nop_trace;
4586         /* All seems OK, enable tracing */
4587         tracing_disabled = 0;
4588
4589         atomic_notifier_chain_register(&panic_notifier_list,
4590                                        &trace_panic_notifier);
4591
4592         register_die_notifier(&trace_die_notifier);
4593
4594         return 0;
4595
4596 out_free_cpumask:
4597         free_cpumask_var(tracing_cpumask);
4598 out_free_buffer_mask:
4599         free_cpumask_var(tracing_buffer_mask);
4600 out:
4601         return ret;
4602 }
4603
4604 __init static int clear_boot_tracer(void)
4605 {
4606         /*
4607          * The default tracer at boot buffer is an init section.
4608          * This function is called in lateinit. If we did not
4609          * find the boot tracer, then clear it out, to prevent
4610          * later registration from accessing the buffer that is
4611          * about to be freed.
4612          */
4613         if (!default_bootup_tracer)
4614                 return 0;
4615
4616         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4617                default_bootup_tracer);
4618         default_bootup_tracer = NULL;
4619
4620         return 0;
4621 }
4622
4623 early_initcall(tracer_alloc_buffers);
4624 fs_initcall(tracer_init_debugfs);
4625 late_initcall(clear_boot_tracer);