Merge commit 'v3.4.4' into android-tegra-nv-3.4
[linux-2.6.git] / kernel / trace / trace.c
1 /*
2  * ring buffer based function tracer
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally taken from the RT patch by:
8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code from the latency_tracer, that is:
11  *  Copyright (C) 2004-2006 Ingo Molnar
12  *  Copyright (C) 2004 William Lee Irwin III
13  */
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/notifier.h>
21 #include <linux/irqflags.h>
22 #include <linux/debugfs.h>
23 #include <linux/pagemap.h>
24 #include <linux/hardirq.h>
25 #include <linux/linkage.h>
26 #include <linux/uaccess.h>
27 #include <linux/kprobes.h>
28 #include <linux/ftrace.h>
29 #include <linux/module.h>
30 #include <linux/percpu.h>
31 #include <linux/splice.h>
32 #include <linux/kdebug.h>
33 #include <linux/string.h>
34 #include <linux/rwsem.h>
35 #include <linux/slab.h>
36 #include <linux/ctype.h>
37 #include <linux/init.h>
38 #include <linux/poll.h>
39 #include <linux/nmi.h>
40 #include <linux/fs.h>
41
42 #include "trace.h"
43 #include "trace_output.h"
44
45 /*
46  * On boot up, the ring buffer is set to the minimum size, so that
47  * we do not waste memory on systems that are not using tracing.
48  */
49 int ring_buffer_expanded;
50
51 /*
52  * We need to change this state when a selftest is running.
53  * A selftest will lurk into the ring-buffer to count the
54  * entries inserted during the selftest although some concurrent
55  * insertions into the ring-buffer such as trace_printk could occurred
56  * at the same time, giving false positive or negative results.
57  */
58 static bool __read_mostly tracing_selftest_running;
59
60 /*
61  * If a tracer is running, we do not want to run SELFTEST.
62  */
63 bool __read_mostly tracing_selftest_disabled;
64
65 /* For tracers that don't implement custom flags */
66 static struct tracer_opt dummy_tracer_opt[] = {
67         { }
68 };
69
70 static struct tracer_flags dummy_tracer_flags = {
71         .val = 0,
72         .opts = dummy_tracer_opt
73 };
74
75 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
76 {
77         return 0;
78 }
79
80 /*
81  * Kill all tracing for good (never come back).
82  * It is initialized to 1 but will turn to zero if the initialization
83  * of the tracer is successful. But that is the only place that sets
84  * this back to zero.
85  */
86 static int tracing_disabled = 1;
87
88 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
89
90 static inline void ftrace_disable_cpu(void)
91 {
92         preempt_disable();
93         __this_cpu_inc(ftrace_cpu_disabled);
94 }
95
96 static inline void ftrace_enable_cpu(void)
97 {
98         __this_cpu_dec(ftrace_cpu_disabled);
99         preempt_enable();
100 }
101
102 cpumask_var_t __read_mostly     tracing_buffer_mask;
103
104 /*
105  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
106  *
107  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
108  * is set, then ftrace_dump is called. This will output the contents
109  * of the ftrace buffers to the console.  This is very useful for
110  * capturing traces that lead to crashes and outputing it to a
111  * serial console.
112  *
113  * It is default off, but you can enable it with either specifying
114  * "ftrace_dump_on_oops" in the kernel command line, or setting
115  * /proc/sys/kernel/ftrace_dump_on_oops
116  * Set 1 if you want to dump buffers of all CPUs
117  * Set 2 if you want to dump the buffer of the CPU that triggered oops
118  */
119
120 enum ftrace_dump_mode ftrace_dump_on_oops;
121
122 static int tracing_set_tracer(const char *buf);
123
124 #define MAX_TRACER_SIZE         100
125 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
126 static char *default_bootup_tracer;
127
128 static int __init set_cmdline_ftrace(char *str)
129 {
130         strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
131         default_bootup_tracer = bootup_tracer_buf;
132         /* We are using ftrace early, expand it */
133         ring_buffer_expanded = 1;
134         return 1;
135 }
136 __setup("ftrace=", set_cmdline_ftrace);
137
138 static int __init set_ftrace_dump_on_oops(char *str)
139 {
140         if (*str++ != '=' || !*str) {
141                 ftrace_dump_on_oops = DUMP_ALL;
142                 return 1;
143         }
144
145         if (!strcmp("orig_cpu", str)) {
146                 ftrace_dump_on_oops = DUMP_ORIG;
147                 return 1;
148         }
149
150         return 0;
151 }
152 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
153
154 unsigned long long ns2usecs(cycle_t nsec)
155 {
156         nsec += 500;
157         do_div(nsec, 1000);
158         return nsec;
159 }
160
161 /*
162  * The global_trace is the descriptor that holds the tracing
163  * buffers for the live tracing. For each CPU, it contains
164  * a link list of pages that will store trace entries. The
165  * page descriptor of the pages in the memory is used to hold
166  * the link list by linking the lru item in the page descriptor
167  * to each of the pages in the buffer per CPU.
168  *
169  * For each active CPU there is a data field that holds the
170  * pages for the buffer for that CPU. Each CPU has the same number
171  * of pages allocated for its buffer.
172  */
173 static struct trace_array       global_trace;
174
175 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
176
177 int filter_current_check_discard(struct ring_buffer *buffer,
178                                  struct ftrace_event_call *call, void *rec,
179                                  struct ring_buffer_event *event)
180 {
181         return filter_check_discard(call, rec, buffer, event);
182 }
183 EXPORT_SYMBOL_GPL(filter_current_check_discard);
184
185 cycle_t ftrace_now(int cpu)
186 {
187         u64 ts;
188
189         /* Early boot up does not have a buffer yet */
190         if (!global_trace.buffer)
191                 return trace_clock_local();
192
193         ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
194         ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
195
196         return ts;
197 }
198
199 /*
200  * The max_tr is used to snapshot the global_trace when a maximum
201  * latency is reached. Some tracers will use this to store a maximum
202  * trace while it continues examining live traces.
203  *
204  * The buffers for the max_tr are set up the same as the global_trace.
205  * When a snapshot is taken, the link list of the max_tr is swapped
206  * with the link list of the global_trace and the buffers are reset for
207  * the global_trace so the tracing can continue.
208  */
209 static struct trace_array       max_tr;
210
211 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
212
213 /* tracer_enabled is used to toggle activation of a tracer */
214 static int                      tracer_enabled = 1;
215
216 /**
217  * tracing_is_enabled - return tracer_enabled status
218  *
219  * This function is used by other tracers to know the status
220  * of the tracer_enabled flag.  Tracers may use this function
221  * to know if it should enable their features when starting
222  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
223  */
224 int tracing_is_enabled(void)
225 {
226         return tracer_enabled;
227 }
228
229 /*
230  * trace_buf_size is the size in bytes that is allocated
231  * for a buffer. Note, the number of bytes is always rounded
232  * to page size.
233  *
234  * This number is purposely set to a low number of 16384.
235  * If the dump on oops happens, it will be much appreciated
236  * to not have to wait for all that output. Anyway this can be
237  * boot time and run time configurable.
238  */
239 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
240
241 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
242
243 /* trace_types holds a link list of available tracers. */
244 static struct tracer            *trace_types __read_mostly;
245
246 /* current_trace points to the tracer that is currently active */
247 static struct tracer            *current_trace __read_mostly;
248
249 /*
250  * trace_types_lock is used to protect the trace_types list.
251  */
252 static DEFINE_MUTEX(trace_types_lock);
253
254 /*
255  * serialize the access of the ring buffer
256  *
257  * ring buffer serializes readers, but it is low level protection.
258  * The validity of the events (which returns by ring_buffer_peek() ..etc)
259  * are not protected by ring buffer.
260  *
261  * The content of events may become garbage if we allow other process consumes
262  * these events concurrently:
263  *   A) the page of the consumed events may become a normal page
264  *      (not reader page) in ring buffer, and this page will be rewrited
265  *      by events producer.
266  *   B) The page of the consumed events may become a page for splice_read,
267  *      and this page will be returned to system.
268  *
269  * These primitives allow multi process access to different cpu ring buffer
270  * concurrently.
271  *
272  * These primitives don't distinguish read-only and read-consume access.
273  * Multi read-only access are also serialized.
274  */
275
276 #ifdef CONFIG_SMP
277 static DECLARE_RWSEM(all_cpu_access_lock);
278 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
279
280 static inline void trace_access_lock(int cpu)
281 {
282         if (cpu == TRACE_PIPE_ALL_CPU) {
283                 /* gain it for accessing the whole ring buffer. */
284                 down_write(&all_cpu_access_lock);
285         } else {
286                 /* gain it for accessing a cpu ring buffer. */
287
288                 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
289                 down_read(&all_cpu_access_lock);
290
291                 /* Secondly block other access to this @cpu ring buffer. */
292                 mutex_lock(&per_cpu(cpu_access_lock, cpu));
293         }
294 }
295
296 static inline void trace_access_unlock(int cpu)
297 {
298         if (cpu == TRACE_PIPE_ALL_CPU) {
299                 up_write(&all_cpu_access_lock);
300         } else {
301                 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
302                 up_read(&all_cpu_access_lock);
303         }
304 }
305
306 static inline void trace_access_lock_init(void)
307 {
308         int cpu;
309
310         for_each_possible_cpu(cpu)
311                 mutex_init(&per_cpu(cpu_access_lock, cpu));
312 }
313
314 #else
315
316 static DEFINE_MUTEX(access_lock);
317
318 static inline void trace_access_lock(int cpu)
319 {
320         (void)cpu;
321         mutex_lock(&access_lock);
322 }
323
324 static inline void trace_access_unlock(int cpu)
325 {
326         (void)cpu;
327         mutex_unlock(&access_lock);
328 }
329
330 static inline void trace_access_lock_init(void)
331 {
332 }
333
334 #endif
335
336 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
337 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
338
339 /* trace_flags holds trace_options default values */
340 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
341         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
342         TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
343         TRACE_ITER_IRQ_INFO;
344
345 static int trace_stop_count;
346 static DEFINE_RAW_SPINLOCK(tracing_start_lock);
347
348 static void wakeup_work_handler(struct work_struct *work)
349 {
350         wake_up(&trace_wait);
351 }
352
353 static DECLARE_DELAYED_WORK(wakeup_work, wakeup_work_handler);
354
355 /**
356  * tracing_on - enable tracing buffers
357  *
358  * This function enables tracing buffers that may have been
359  * disabled with tracing_off.
360  */
361 void tracing_on(void)
362 {
363         if (global_trace.buffer)
364                 ring_buffer_record_on(global_trace.buffer);
365         /*
366          * This flag is only looked at when buffers haven't been
367          * allocated yet. We don't really care about the race
368          * between setting this flag and actually turning
369          * on the buffer.
370          */
371         global_trace.buffer_disabled = 0;
372 }
373 EXPORT_SYMBOL_GPL(tracing_on);
374
375 /**
376  * tracing_off - turn off tracing buffers
377  *
378  * This function stops the tracing buffers from recording data.
379  * It does not disable any overhead the tracers themselves may
380  * be causing. This function simply causes all recording to
381  * the ring buffers to fail.
382  */
383 void tracing_off(void)
384 {
385         if (global_trace.buffer)
386                 ring_buffer_record_off(global_trace.buffer);
387         /*
388          * This flag is only looked at when buffers haven't been
389          * allocated yet. We don't really care about the race
390          * between setting this flag and actually turning
391          * on the buffer.
392          */
393         global_trace.buffer_disabled = 1;
394 }
395 EXPORT_SYMBOL_GPL(tracing_off);
396
397 /**
398  * tracing_is_on - show state of ring buffers enabled
399  */
400 int tracing_is_on(void)
401 {
402         if (global_trace.buffer)
403                 return ring_buffer_record_is_on(global_trace.buffer);
404         return !global_trace.buffer_disabled;
405 }
406 EXPORT_SYMBOL_GPL(tracing_is_on);
407
408 /**
409  * trace_wake_up - wake up tasks waiting for trace input
410  *
411  * Schedules a delayed work to wake up any task that is blocked on the
412  * trace_wait queue. These is used with trace_poll for tasks polling the
413  * trace.
414  */
415 void trace_wake_up(void)
416 {
417         const unsigned long delay = msecs_to_jiffies(2);
418
419         if (trace_flags & TRACE_ITER_BLOCK)
420                 return;
421         schedule_delayed_work(&wakeup_work, delay);
422 }
423
424 static int __init set_buf_size(char *str)
425 {
426         unsigned long buf_size;
427
428         if (!str)
429                 return 0;
430         buf_size = memparse(str, &str);
431         /* nr_entries can not be zero */
432         if (buf_size == 0)
433                 return 0;
434         trace_buf_size = buf_size;
435         return 1;
436 }
437 __setup("trace_buf_size=", set_buf_size);
438
439 static int __init set_tracing_thresh(char *str)
440 {
441         unsigned long threshhold;
442         int ret;
443
444         if (!str)
445                 return 0;
446         ret = strict_strtoul(str, 0, &threshhold);
447         if (ret < 0)
448                 return 0;
449         tracing_thresh = threshhold * 1000;
450         return 1;
451 }
452 __setup("tracing_thresh=", set_tracing_thresh);
453
454 unsigned long nsecs_to_usecs(unsigned long nsecs)
455 {
456         return nsecs / 1000;
457 }
458
459 /* These must match the bit postions in trace_iterator_flags */
460 static const char *trace_options[] = {
461         "print-parent",
462         "sym-offset",
463         "sym-addr",
464         "verbose",
465         "raw",
466         "hex",
467         "bin",
468         "block",
469         "stacktrace",
470         "trace_printk",
471         "ftrace_preempt",
472         "branch",
473         "annotate",
474         "userstacktrace",
475         "sym-userobj",
476         "printk-msg-only",
477         "context-info",
478         "latency-format",
479         "sleep-time",
480         "graph-time",
481         "record-cmd",
482         "overwrite",
483         "disable_on_free",
484         "irq-info",
485         NULL
486 };
487
488 static struct {
489         u64 (*func)(void);
490         const char *name;
491 } trace_clocks[] = {
492         { trace_clock_local,    "local" },
493         { trace_clock_global,   "global" },
494         { trace_clock_counter,  "counter" },
495 };
496
497 int trace_clock_id;
498
499 /*
500  * trace_parser_get_init - gets the buffer for trace parser
501  */
502 int trace_parser_get_init(struct trace_parser *parser, int size)
503 {
504         memset(parser, 0, sizeof(*parser));
505
506         parser->buffer = kmalloc(size, GFP_KERNEL);
507         if (!parser->buffer)
508                 return 1;
509
510         parser->size = size;
511         return 0;
512 }
513
514 /*
515  * trace_parser_put - frees the buffer for trace parser
516  */
517 void trace_parser_put(struct trace_parser *parser)
518 {
519         kfree(parser->buffer);
520 }
521
522 /*
523  * trace_get_user - reads the user input string separated by  space
524  * (matched by isspace(ch))
525  *
526  * For each string found the 'struct trace_parser' is updated,
527  * and the function returns.
528  *
529  * Returns number of bytes read.
530  *
531  * See kernel/trace/trace.h for 'struct trace_parser' details.
532  */
533 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
534         size_t cnt, loff_t *ppos)
535 {
536         char ch;
537         size_t read = 0;
538         ssize_t ret;
539
540         if (!*ppos)
541                 trace_parser_clear(parser);
542
543         ret = get_user(ch, ubuf++);
544         if (ret)
545                 goto out;
546
547         read++;
548         cnt--;
549
550         /*
551          * The parser is not finished with the last write,
552          * continue reading the user input without skipping spaces.
553          */
554         if (!parser->cont) {
555                 /* skip white space */
556                 while (cnt && isspace(ch)) {
557                         ret = get_user(ch, ubuf++);
558                         if (ret)
559                                 goto out;
560                         read++;
561                         cnt--;
562                 }
563
564                 /* only spaces were written */
565                 if (isspace(ch)) {
566                         *ppos += read;
567                         ret = read;
568                         goto out;
569                 }
570
571                 parser->idx = 0;
572         }
573
574         /* read the non-space input */
575         while (cnt && !isspace(ch)) {
576                 if (parser->idx < parser->size - 1)
577                         parser->buffer[parser->idx++] = ch;
578                 else {
579                         ret = -EINVAL;
580                         goto out;
581                 }
582                 ret = get_user(ch, ubuf++);
583                 if (ret)
584                         goto out;
585                 read++;
586                 cnt--;
587         }
588
589         /* We either got finished input or we have to wait for another call. */
590         if (isspace(ch)) {
591                 parser->buffer[parser->idx] = 0;
592                 parser->cont = false;
593         } else {
594                 parser->cont = true;
595                 parser->buffer[parser->idx++] = ch;
596         }
597
598         *ppos += read;
599         ret = read;
600
601 out:
602         return ret;
603 }
604
605 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
606 {
607         int len;
608         int ret;
609
610         if (!cnt)
611                 return 0;
612
613         if (s->len <= s->readpos)
614                 return -EBUSY;
615
616         len = s->len - s->readpos;
617         if (cnt > len)
618                 cnt = len;
619         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
620         if (ret == cnt)
621                 return -EFAULT;
622
623         cnt -= ret;
624
625         s->readpos += cnt;
626         return cnt;
627 }
628
629 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
630 {
631         int len;
632         void *ret;
633
634         if (s->len <= s->readpos)
635                 return -EBUSY;
636
637         len = s->len - s->readpos;
638         if (cnt > len)
639                 cnt = len;
640         ret = memcpy(buf, s->buffer + s->readpos, cnt);
641         if (!ret)
642                 return -EFAULT;
643
644         s->readpos += cnt;
645         return cnt;
646 }
647
648 /*
649  * ftrace_max_lock is used to protect the swapping of buffers
650  * when taking a max snapshot. The buffers themselves are
651  * protected by per_cpu spinlocks. But the action of the swap
652  * needs its own lock.
653  *
654  * This is defined as a arch_spinlock_t in order to help
655  * with performance when lockdep debugging is enabled.
656  *
657  * It is also used in other places outside the update_max_tr
658  * so it needs to be defined outside of the
659  * CONFIG_TRACER_MAX_TRACE.
660  */
661 static arch_spinlock_t ftrace_max_lock =
662         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
663
664 unsigned long __read_mostly     tracing_thresh;
665
666 #ifdef CONFIG_TRACER_MAX_TRACE
667 unsigned long __read_mostly     tracing_max_latency;
668
669 /*
670  * Copy the new maximum trace into the separate maximum-trace
671  * structure. (this way the maximum trace is permanently saved,
672  * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
673  */
674 static void
675 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
676 {
677         struct trace_array_cpu *data = tr->data[cpu];
678         struct trace_array_cpu *max_data;
679
680         max_tr.cpu = cpu;
681         max_tr.time_start = data->preempt_timestamp;
682
683         max_data = max_tr.data[cpu];
684         max_data->saved_latency = tracing_max_latency;
685         max_data->critical_start = data->critical_start;
686         max_data->critical_end = data->critical_end;
687
688         memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
689         max_data->pid = tsk->pid;
690         max_data->uid = task_uid(tsk);
691         max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
692         max_data->policy = tsk->policy;
693         max_data->rt_priority = tsk->rt_priority;
694
695         /* record this tasks comm */
696         tracing_record_cmdline(tsk);
697 }
698
699 /**
700  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
701  * @tr: tracer
702  * @tsk: the task with the latency
703  * @cpu: The cpu that initiated the trace.
704  *
705  * Flip the buffers between the @tr and the max_tr and record information
706  * about which task was the cause of this latency.
707  */
708 void
709 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
710 {
711         struct ring_buffer *buf = tr->buffer;
712
713         if (trace_stop_count)
714                 return;
715
716         WARN_ON_ONCE(!irqs_disabled());
717         if (!current_trace->use_max_tr) {
718                 WARN_ON_ONCE(1);
719                 return;
720         }
721         arch_spin_lock(&ftrace_max_lock);
722
723         tr->buffer = max_tr.buffer;
724         max_tr.buffer = buf;
725
726         __update_max_tr(tr, tsk, cpu);
727         arch_spin_unlock(&ftrace_max_lock);
728 }
729
730 /**
731  * update_max_tr_single - only copy one trace over, and reset the rest
732  * @tr - tracer
733  * @tsk - task with the latency
734  * @cpu - the cpu of the buffer to copy.
735  *
736  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
737  */
738 void
739 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
740 {
741         int ret;
742
743         if (trace_stop_count)
744                 return;
745
746         WARN_ON_ONCE(!irqs_disabled());
747         if (!current_trace->use_max_tr) {
748                 WARN_ON_ONCE(1);
749                 return;
750         }
751
752         arch_spin_lock(&ftrace_max_lock);
753
754         ftrace_disable_cpu();
755
756         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
757
758         if (ret == -EBUSY) {
759                 /*
760                  * We failed to swap the buffer due to a commit taking
761                  * place on this CPU. We fail to record, but we reset
762                  * the max trace buffer (no one writes directly to it)
763                  * and flag that it failed.
764                  */
765                 trace_array_printk(&max_tr, _THIS_IP_,
766                         "Failed to swap buffers due to commit in progress\n");
767         }
768
769         ftrace_enable_cpu();
770
771         WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
772
773         __update_max_tr(tr, tsk, cpu);
774         arch_spin_unlock(&ftrace_max_lock);
775 }
776 #endif /* CONFIG_TRACER_MAX_TRACE */
777
778 /**
779  * register_tracer - register a tracer with the ftrace system.
780  * @type - the plugin for the tracer
781  *
782  * Register a new plugin tracer.
783  */
784 int register_tracer(struct tracer *type)
785 __releases(kernel_lock)
786 __acquires(kernel_lock)
787 {
788         struct tracer *t;
789         int ret = 0;
790
791         if (!type->name) {
792                 pr_info("Tracer must have a name\n");
793                 return -1;
794         }
795
796         if (strlen(type->name) >= MAX_TRACER_SIZE) {
797                 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
798                 return -1;
799         }
800
801         mutex_lock(&trace_types_lock);
802
803         tracing_selftest_running = true;
804
805         for (t = trace_types; t; t = t->next) {
806                 if (strcmp(type->name, t->name) == 0) {
807                         /* already found */
808                         pr_info("Tracer %s already registered\n",
809                                 type->name);
810                         ret = -1;
811                         goto out;
812                 }
813         }
814
815         if (!type->set_flag)
816                 type->set_flag = &dummy_set_flag;
817         if (!type->flags)
818                 type->flags = &dummy_tracer_flags;
819         else
820                 if (!type->flags->opts)
821                         type->flags->opts = dummy_tracer_opt;
822         if (!type->wait_pipe)
823                 type->wait_pipe = default_wait_pipe;
824
825
826 #ifdef CONFIG_FTRACE_STARTUP_TEST
827         if (type->selftest && !tracing_selftest_disabled) {
828                 struct tracer *saved_tracer = current_trace;
829                 struct trace_array *tr = &global_trace;
830
831                 /*
832                  * Run a selftest on this tracer.
833                  * Here we reset the trace buffer, and set the current
834                  * tracer to be this tracer. The tracer can then run some
835                  * internal tracing to verify that everything is in order.
836                  * If we fail, we do not register this tracer.
837                  */
838                 tracing_reset_online_cpus(tr);
839
840                 current_trace = type;
841
842                 /* If we expanded the buffers, make sure the max is expanded too */
843                 if (ring_buffer_expanded && type->use_max_tr)
844                         ring_buffer_resize(max_tr.buffer, trace_buf_size);
845
846                 /* the test is responsible for initializing and enabling */
847                 pr_info("Testing tracer %s: ", type->name);
848                 ret = type->selftest(type, tr);
849                 /* the test is responsible for resetting too */
850                 current_trace = saved_tracer;
851                 if (ret) {
852                         printk(KERN_CONT "FAILED!\n");
853                         goto out;
854                 }
855                 /* Only reset on passing, to avoid touching corrupted buffers */
856                 tracing_reset_online_cpus(tr);
857
858                 /* Shrink the max buffer again */
859                 if (ring_buffer_expanded && type->use_max_tr)
860                         ring_buffer_resize(max_tr.buffer, 1);
861
862                 printk(KERN_CONT "PASSED\n");
863         }
864 #endif
865
866         type->next = trace_types;
867         trace_types = type;
868
869  out:
870         tracing_selftest_running = false;
871         mutex_unlock(&trace_types_lock);
872
873         if (ret || !default_bootup_tracer)
874                 goto out_unlock;
875
876         if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
877                 goto out_unlock;
878
879         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
880         /* Do we want this tracer to start on bootup? */
881         tracing_set_tracer(type->name);
882         default_bootup_tracer = NULL;
883         /* disable other selftests, since this will break it. */
884         tracing_selftest_disabled = 1;
885 #ifdef CONFIG_FTRACE_STARTUP_TEST
886         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
887                type->name);
888 #endif
889
890  out_unlock:
891         return ret;
892 }
893
894 void unregister_tracer(struct tracer *type)
895 {
896         struct tracer **t;
897
898         mutex_lock(&trace_types_lock);
899         for (t = &trace_types; *t; t = &(*t)->next) {
900                 if (*t == type)
901                         goto found;
902         }
903         pr_info("Tracer %s not registered\n", type->name);
904         goto out;
905
906  found:
907         *t = (*t)->next;
908
909         if (type == current_trace && tracer_enabled) {
910                 tracer_enabled = 0;
911                 tracing_stop();
912                 if (current_trace->stop)
913                         current_trace->stop(&global_trace);
914                 current_trace = &nop_trace;
915         }
916 out:
917         mutex_unlock(&trace_types_lock);
918 }
919
920 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
921 {
922         ftrace_disable_cpu();
923         ring_buffer_reset_cpu(buffer, cpu);
924         ftrace_enable_cpu();
925 }
926
927 void tracing_reset(struct trace_array *tr, int cpu)
928 {
929         struct ring_buffer *buffer = tr->buffer;
930
931         ring_buffer_record_disable(buffer);
932
933         /* Make sure all commits have finished */
934         synchronize_sched();
935         __tracing_reset(buffer, cpu);
936
937         ring_buffer_record_enable(buffer);
938 }
939
940 void tracing_reset_online_cpus(struct trace_array *tr)
941 {
942         struct ring_buffer *buffer = tr->buffer;
943         int cpu;
944
945         ring_buffer_record_disable(buffer);
946
947         /* Make sure all commits have finished */
948         synchronize_sched();
949
950         tr->time_start = ftrace_now(tr->cpu);
951
952         for_each_online_cpu(cpu)
953                 __tracing_reset(buffer, cpu);
954
955         ring_buffer_record_enable(buffer);
956 }
957
958 void tracing_reset_current(int cpu)
959 {
960         tracing_reset(&global_trace, cpu);
961 }
962
963 void tracing_reset_current_online_cpus(void)
964 {
965         tracing_reset_online_cpus(&global_trace);
966 }
967
968 #define SAVED_CMDLINES 128
969 #define NO_CMDLINE_MAP UINT_MAX
970 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
971 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
972 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
973 static int cmdline_idx;
974 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
975
976 /* temporary disable recording */
977 static atomic_t trace_record_cmdline_disabled __read_mostly;
978
979 static void trace_init_cmdlines(void)
980 {
981         memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
982         memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
983         cmdline_idx = 0;
984 }
985
986 int is_tracing_stopped(void)
987 {
988         return trace_stop_count;
989 }
990
991 /**
992  * ftrace_off_permanent - disable all ftrace code permanently
993  *
994  * This should only be called when a serious anomally has
995  * been detected.  This will turn off the function tracing,
996  * ring buffers, and other tracing utilites. It takes no
997  * locks and can be called from any context.
998  */
999 void ftrace_off_permanent(void)
1000 {
1001         tracing_disabled = 1;
1002         ftrace_stop();
1003         tracing_off_permanent();
1004 }
1005
1006 /**
1007  * tracing_start - quick start of the tracer
1008  *
1009  * If tracing is enabled but was stopped by tracing_stop,
1010  * this will start the tracer back up.
1011  */
1012 void tracing_start(void)
1013 {
1014         struct ring_buffer *buffer;
1015         unsigned long flags;
1016
1017         if (tracing_disabled)
1018                 return;
1019
1020         raw_spin_lock_irqsave(&tracing_start_lock, flags);
1021         if (--trace_stop_count) {
1022                 if (trace_stop_count < 0) {
1023                         /* Someone screwed up their debugging */
1024                         WARN_ON_ONCE(1);
1025                         trace_stop_count = 0;
1026                 }
1027                 goto out;
1028         }
1029
1030         /* Prevent the buffers from switching */
1031         arch_spin_lock(&ftrace_max_lock);
1032
1033         buffer = global_trace.buffer;
1034         if (buffer)
1035                 ring_buffer_record_enable(buffer);
1036
1037         buffer = max_tr.buffer;
1038         if (buffer)
1039                 ring_buffer_record_enable(buffer);
1040
1041         arch_spin_unlock(&ftrace_max_lock);
1042
1043         ftrace_start();
1044  out:
1045         raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1046 }
1047
1048 /**
1049  * tracing_stop - quick stop of the tracer
1050  *
1051  * Light weight way to stop tracing. Use in conjunction with
1052  * tracing_start.
1053  */
1054 void tracing_stop(void)
1055 {
1056         struct ring_buffer *buffer;
1057         unsigned long flags;
1058
1059         ftrace_stop();
1060         raw_spin_lock_irqsave(&tracing_start_lock, flags);
1061         if (trace_stop_count++)
1062                 goto out;
1063
1064         /* Prevent the buffers from switching */
1065         arch_spin_lock(&ftrace_max_lock);
1066
1067         buffer = global_trace.buffer;
1068         if (buffer)
1069                 ring_buffer_record_disable(buffer);
1070
1071         buffer = max_tr.buffer;
1072         if (buffer)
1073                 ring_buffer_record_disable(buffer);
1074
1075         arch_spin_unlock(&ftrace_max_lock);
1076
1077  out:
1078         raw_spin_unlock_irqrestore(&tracing_start_lock, flags);
1079 }
1080
1081 void trace_stop_cmdline_recording(void);
1082
1083 static void trace_save_cmdline(struct task_struct *tsk)
1084 {
1085         unsigned pid, idx;
1086
1087         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1088                 return;
1089
1090         /*
1091          * It's not the end of the world if we don't get
1092          * the lock, but we also don't want to spin
1093          * nor do we want to disable interrupts,
1094          * so if we miss here, then better luck next time.
1095          */
1096         if (!arch_spin_trylock(&trace_cmdline_lock))
1097                 return;
1098
1099         idx = map_pid_to_cmdline[tsk->pid];
1100         if (idx == NO_CMDLINE_MAP) {
1101                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1102
1103                 /*
1104                  * Check whether the cmdline buffer at idx has a pid
1105                  * mapped. We are going to overwrite that entry so we
1106                  * need to clear the map_pid_to_cmdline. Otherwise we
1107                  * would read the new comm for the old pid.
1108                  */
1109                 pid = map_cmdline_to_pid[idx];
1110                 if (pid != NO_CMDLINE_MAP)
1111                         map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1112
1113                 map_cmdline_to_pid[idx] = tsk->pid;
1114                 map_pid_to_cmdline[tsk->pid] = idx;
1115
1116                 cmdline_idx = idx;
1117         }
1118
1119         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1120
1121         arch_spin_unlock(&trace_cmdline_lock);
1122 }
1123
1124 void trace_find_cmdline(int pid, char comm[])
1125 {
1126         unsigned map;
1127
1128         if (!pid) {
1129                 strcpy(comm, "<idle>");
1130                 return;
1131         }
1132
1133         if (WARN_ON_ONCE(pid < 0)) {
1134                 strcpy(comm, "<XXX>");
1135                 return;
1136         }
1137
1138         if (pid > PID_MAX_DEFAULT) {
1139                 strcpy(comm, "<...>");
1140                 return;
1141         }
1142
1143         preempt_disable();
1144         arch_spin_lock(&trace_cmdline_lock);
1145         map = map_pid_to_cmdline[pid];
1146         if (map != NO_CMDLINE_MAP)
1147                 strcpy(comm, saved_cmdlines[map]);
1148         else
1149                 strcpy(comm, "<...>");
1150
1151         arch_spin_unlock(&trace_cmdline_lock);
1152         preempt_enable();
1153 }
1154
1155 void tracing_record_cmdline(struct task_struct *tsk)
1156 {
1157         if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1158             !tracing_is_on())
1159                 return;
1160
1161         trace_save_cmdline(tsk);
1162 }
1163
1164 void
1165 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1166                              int pc)
1167 {
1168         struct task_struct *tsk = current;
1169
1170         entry->preempt_count            = pc & 0xff;
1171         entry->pid                      = (tsk) ? tsk->pid : 0;
1172         entry->padding                  = 0;
1173         entry->flags =
1174 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1175                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1176 #else
1177                 TRACE_FLAG_IRQS_NOSUPPORT |
1178 #endif
1179                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1180                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1181                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1182 }
1183 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1184
1185 struct ring_buffer_event *
1186 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1187                           int type,
1188                           unsigned long len,
1189                           unsigned long flags, int pc)
1190 {
1191         struct ring_buffer_event *event;
1192
1193         event = ring_buffer_lock_reserve(buffer, len);
1194         if (event != NULL) {
1195                 struct trace_entry *ent = ring_buffer_event_data(event);
1196
1197                 tracing_generic_entry_update(ent, flags, pc);
1198                 ent->type = type;
1199         }
1200
1201         return event;
1202 }
1203
1204 static inline void
1205 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1206                              struct ring_buffer_event *event,
1207                              unsigned long flags, int pc,
1208                              int wake)
1209 {
1210         ring_buffer_unlock_commit(buffer, event);
1211
1212         ftrace_trace_stack(buffer, flags, 6, pc);
1213         ftrace_trace_userstack(buffer, flags, pc);
1214
1215         if (wake)
1216                 trace_wake_up();
1217 }
1218
1219 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1220                                 struct ring_buffer_event *event,
1221                                 unsigned long flags, int pc)
1222 {
1223         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1224 }
1225
1226 struct ring_buffer_event *
1227 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1228                                   int type, unsigned long len,
1229                                   unsigned long flags, int pc)
1230 {
1231         *current_rb = global_trace.buffer;
1232         return trace_buffer_lock_reserve(*current_rb,
1233                                          type, len, flags, pc);
1234 }
1235 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1236
1237 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1238                                         struct ring_buffer_event *event,
1239                                         unsigned long flags, int pc)
1240 {
1241         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1242 }
1243 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1244
1245 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1246                                        struct ring_buffer_event *event,
1247                                        unsigned long flags, int pc)
1248 {
1249         __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1250 }
1251 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1252
1253 void trace_nowake_buffer_unlock_commit_regs(struct ring_buffer *buffer,
1254                                             struct ring_buffer_event *event,
1255                                             unsigned long flags, int pc,
1256                                             struct pt_regs *regs)
1257 {
1258         ring_buffer_unlock_commit(buffer, event);
1259
1260         ftrace_trace_stack_regs(buffer, flags, 0, pc, regs);
1261         ftrace_trace_userstack(buffer, flags, pc);
1262 }
1263 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit_regs);
1264
1265 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1266                                          struct ring_buffer_event *event)
1267 {
1268         ring_buffer_discard_commit(buffer, event);
1269 }
1270 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1271
1272 void
1273 trace_function(struct trace_array *tr,
1274                unsigned long ip, unsigned long parent_ip, unsigned long flags,
1275                int pc)
1276 {
1277         struct ftrace_event_call *call = &event_function;
1278         struct ring_buffer *buffer = tr->buffer;
1279         struct ring_buffer_event *event;
1280         struct ftrace_entry *entry;
1281
1282         /* If we are reading the ring buffer, don't trace */
1283         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1284                 return;
1285
1286         event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1287                                           flags, pc);
1288         if (!event)
1289                 return;
1290         entry   = ring_buffer_event_data(event);
1291         entry->ip                       = ip;
1292         entry->parent_ip                = parent_ip;
1293
1294         if (!filter_check_discard(call, entry, buffer, event))
1295                 ring_buffer_unlock_commit(buffer, event);
1296 }
1297
1298 void
1299 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1300        unsigned long ip, unsigned long parent_ip, unsigned long flags,
1301        int pc)
1302 {
1303         if (likely(!atomic_read(&data->disabled)))
1304                 trace_function(tr, ip, parent_ip, flags, pc);
1305 }
1306
1307 #ifdef CONFIG_STACKTRACE
1308
1309 #define FTRACE_STACK_MAX_ENTRIES (PAGE_SIZE / sizeof(unsigned long))
1310 struct ftrace_stack {
1311         unsigned long           calls[FTRACE_STACK_MAX_ENTRIES];
1312 };
1313
1314 static DEFINE_PER_CPU(struct ftrace_stack, ftrace_stack);
1315 static DEFINE_PER_CPU(int, ftrace_stack_reserve);
1316
1317 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1318                                  unsigned long flags,
1319                                  int skip, int pc, struct pt_regs *regs)
1320 {
1321         struct ftrace_event_call *call = &event_kernel_stack;
1322         struct ring_buffer_event *event;
1323         struct stack_entry *entry;
1324         struct stack_trace trace;
1325         int use_stack;
1326         int size = FTRACE_STACK_ENTRIES;
1327
1328         trace.nr_entries        = 0;
1329         trace.skip              = skip;
1330
1331         /*
1332          * Since events can happen in NMIs there's no safe way to
1333          * use the per cpu ftrace_stacks. We reserve it and if an interrupt
1334          * or NMI comes in, it will just have to use the default
1335          * FTRACE_STACK_SIZE.
1336          */
1337         preempt_disable_notrace();
1338
1339         use_stack = ++__get_cpu_var(ftrace_stack_reserve);
1340         /*
1341          * We don't need any atomic variables, just a barrier.
1342          * If an interrupt comes in, we don't care, because it would
1343          * have exited and put the counter back to what we want.
1344          * We just need a barrier to keep gcc from moving things
1345          * around.
1346          */
1347         barrier();
1348         if (use_stack == 1) {
1349                 trace.entries           = &__get_cpu_var(ftrace_stack).calls[0];
1350                 trace.max_entries       = FTRACE_STACK_MAX_ENTRIES;
1351
1352                 if (regs)
1353                         save_stack_trace_regs(regs, &trace);
1354                 else
1355                         save_stack_trace(&trace);
1356
1357                 if (trace.nr_entries > size)
1358                         size = trace.nr_entries;
1359         } else
1360                 /* From now on, use_stack is a boolean */
1361                 use_stack = 0;
1362
1363         size *= sizeof(unsigned long);
1364
1365         event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1366                                           sizeof(*entry) + size, flags, pc);
1367         if (!event)
1368                 goto out;
1369         entry = ring_buffer_event_data(event);
1370
1371         memset(&entry->caller, 0, size);
1372
1373         if (use_stack)
1374                 memcpy(&entry->caller, trace.entries,
1375                        trace.nr_entries * sizeof(unsigned long));
1376         else {
1377                 trace.max_entries       = FTRACE_STACK_ENTRIES;
1378                 trace.entries           = entry->caller;
1379                 if (regs)
1380                         save_stack_trace_regs(regs, &trace);
1381                 else
1382                         save_stack_trace(&trace);
1383         }
1384
1385         entry->size = trace.nr_entries;
1386
1387         if (!filter_check_discard(call, entry, buffer, event))
1388                 ring_buffer_unlock_commit(buffer, event);
1389
1390  out:
1391         /* Again, don't let gcc optimize things here */
1392         barrier();
1393         __get_cpu_var(ftrace_stack_reserve)--;
1394         preempt_enable_notrace();
1395
1396 }
1397
1398 void ftrace_trace_stack_regs(struct ring_buffer *buffer, unsigned long flags,
1399                              int skip, int pc, struct pt_regs *regs)
1400 {
1401         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1402                 return;
1403
1404         __ftrace_trace_stack(buffer, flags, skip, pc, regs);
1405 }
1406
1407 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1408                         int skip, int pc)
1409 {
1410         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1411                 return;
1412
1413         __ftrace_trace_stack(buffer, flags, skip, pc, NULL);
1414 }
1415
1416 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1417                    int pc)
1418 {
1419         __ftrace_trace_stack(tr->buffer, flags, skip, pc, NULL);
1420 }
1421
1422 /**
1423  * trace_dump_stack - record a stack back trace in the trace buffer
1424  */
1425 void trace_dump_stack(void)
1426 {
1427         unsigned long flags;
1428
1429         if (tracing_disabled || tracing_selftest_running)
1430                 return;
1431
1432         local_save_flags(flags);
1433
1434         /* skipping 3 traces, seems to get us at the caller of this function */
1435         __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count(), NULL);
1436 }
1437
1438 static DEFINE_PER_CPU(int, user_stack_count);
1439
1440 void
1441 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1442 {
1443         struct ftrace_event_call *call = &event_user_stack;
1444         struct ring_buffer_event *event;
1445         struct userstack_entry *entry;
1446         struct stack_trace trace;
1447
1448         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1449                 return;
1450
1451         /*
1452          * NMIs can not handle page faults, even with fix ups.
1453          * The save user stack can (and often does) fault.
1454          */
1455         if (unlikely(in_nmi()))
1456                 return;
1457
1458         /*
1459          * prevent recursion, since the user stack tracing may
1460          * trigger other kernel events.
1461          */
1462         preempt_disable();
1463         if (__this_cpu_read(user_stack_count))
1464                 goto out;
1465
1466         __this_cpu_inc(user_stack_count);
1467
1468         event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1469                                           sizeof(*entry), flags, pc);
1470         if (!event)
1471                 goto out_drop_count;
1472         entry   = ring_buffer_event_data(event);
1473
1474         entry->tgid             = current->tgid;
1475         memset(&entry->caller, 0, sizeof(entry->caller));
1476
1477         trace.nr_entries        = 0;
1478         trace.max_entries       = FTRACE_STACK_ENTRIES;
1479         trace.skip              = 0;
1480         trace.entries           = entry->caller;
1481
1482         save_stack_trace_user(&trace);
1483         if (!filter_check_discard(call, entry, buffer, event))
1484                 ring_buffer_unlock_commit(buffer, event);
1485
1486  out_drop_count:
1487         __this_cpu_dec(user_stack_count);
1488  out:
1489         preempt_enable();
1490 }
1491
1492 #ifdef UNUSED
1493 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1494 {
1495         ftrace_trace_userstack(tr, flags, preempt_count());
1496 }
1497 #endif /* UNUSED */
1498
1499 #endif /* CONFIG_STACKTRACE */
1500
1501 /**
1502  * trace_vbprintk - write binary msg to tracing buffer
1503  *
1504  */
1505 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1506 {
1507         static arch_spinlock_t trace_buf_lock =
1508                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1509         static u32 trace_buf[TRACE_BUF_SIZE];
1510
1511         struct ftrace_event_call *call = &event_bprint;
1512         struct ring_buffer_event *event;
1513         struct ring_buffer *buffer;
1514         struct trace_array *tr = &global_trace;
1515         struct trace_array_cpu *data;
1516         struct bprint_entry *entry;
1517         unsigned long flags;
1518         int disable;
1519         int cpu, len = 0, size, pc;
1520
1521         if (unlikely(tracing_selftest_running || tracing_disabled))
1522                 return 0;
1523
1524         /* Don't pollute graph traces with trace_vprintk internals */
1525         pause_graph_tracing();
1526
1527         pc = preempt_count();
1528         preempt_disable_notrace();
1529         cpu = raw_smp_processor_id();
1530         data = tr->data[cpu];
1531
1532         disable = atomic_inc_return(&data->disabled);
1533         if (unlikely(disable != 1))
1534                 goto out;
1535
1536         /* Lockdep uses trace_printk for lock tracing */
1537         local_irq_save(flags);
1538         arch_spin_lock(&trace_buf_lock);
1539         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1540
1541         if (len > TRACE_BUF_SIZE || len < 0)
1542                 goto out_unlock;
1543
1544         size = sizeof(*entry) + sizeof(u32) * len;
1545         buffer = tr->buffer;
1546         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1547                                           flags, pc);
1548         if (!event)
1549                 goto out_unlock;
1550         entry = ring_buffer_event_data(event);
1551         entry->ip                       = ip;
1552         entry->fmt                      = fmt;
1553
1554         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1555         if (!filter_check_discard(call, entry, buffer, event)) {
1556                 ring_buffer_unlock_commit(buffer, event);
1557                 ftrace_trace_stack(buffer, flags, 6, pc);
1558         }
1559
1560 out_unlock:
1561         arch_spin_unlock(&trace_buf_lock);
1562         local_irq_restore(flags);
1563
1564 out:
1565         atomic_dec_return(&data->disabled);
1566         preempt_enable_notrace();
1567         unpause_graph_tracing();
1568
1569         return len;
1570 }
1571 EXPORT_SYMBOL_GPL(trace_vbprintk);
1572
1573 int trace_array_printk(struct trace_array *tr,
1574                        unsigned long ip, const char *fmt, ...)
1575 {
1576         int ret;
1577         va_list ap;
1578
1579         if (!(trace_flags & TRACE_ITER_PRINTK))
1580                 return 0;
1581
1582         va_start(ap, fmt);
1583         ret = trace_array_vprintk(tr, ip, fmt, ap);
1584         va_end(ap);
1585         return ret;
1586 }
1587
1588 int trace_array_vprintk(struct trace_array *tr,
1589                         unsigned long ip, const char *fmt, va_list args)
1590 {
1591         static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1592         static char trace_buf[TRACE_BUF_SIZE];
1593
1594         struct ftrace_event_call *call = &event_print;
1595         struct ring_buffer_event *event;
1596         struct ring_buffer *buffer;
1597         struct trace_array_cpu *data;
1598         int cpu, len = 0, size, pc;
1599         struct print_entry *entry;
1600         unsigned long irq_flags;
1601         int disable;
1602
1603         if (tracing_disabled || tracing_selftest_running)
1604                 return 0;
1605
1606         pc = preempt_count();
1607         preempt_disable_notrace();
1608         cpu = raw_smp_processor_id();
1609         data = tr->data[cpu];
1610
1611         disable = atomic_inc_return(&data->disabled);
1612         if (unlikely(disable != 1))
1613                 goto out;
1614
1615         pause_graph_tracing();
1616         raw_local_irq_save(irq_flags);
1617         arch_spin_lock(&trace_buf_lock);
1618         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1619
1620         size = sizeof(*entry) + len + 1;
1621         buffer = tr->buffer;
1622         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1623                                           irq_flags, pc);
1624         if (!event)
1625                 goto out_unlock;
1626         entry = ring_buffer_event_data(event);
1627         entry->ip = ip;
1628
1629         memcpy(&entry->buf, trace_buf, len);
1630         entry->buf[len] = '\0';
1631         if (!filter_check_discard(call, entry, buffer, event)) {
1632                 ring_buffer_unlock_commit(buffer, event);
1633                 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1634         }
1635
1636  out_unlock:
1637         arch_spin_unlock(&trace_buf_lock);
1638         raw_local_irq_restore(irq_flags);
1639         unpause_graph_tracing();
1640  out:
1641         atomic_dec_return(&data->disabled);
1642         preempt_enable_notrace();
1643
1644         return len;
1645 }
1646
1647 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1648 {
1649         return trace_array_vprintk(&global_trace, ip, fmt, args);
1650 }
1651 EXPORT_SYMBOL_GPL(trace_vprintk);
1652
1653 static void trace_iterator_increment(struct trace_iterator *iter)
1654 {
1655         /* Don't allow ftrace to trace into the ring buffers */
1656         ftrace_disable_cpu();
1657
1658         iter->idx++;
1659         if (iter->buffer_iter[iter->cpu])
1660                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1661
1662         ftrace_enable_cpu();
1663 }
1664
1665 static struct trace_entry *
1666 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1667                 unsigned long *lost_events)
1668 {
1669         struct ring_buffer_event *event;
1670         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1671
1672         /* Don't allow ftrace to trace into the ring buffers */
1673         ftrace_disable_cpu();
1674
1675         if (buf_iter)
1676                 event = ring_buffer_iter_peek(buf_iter, ts);
1677         else
1678                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1679                                          lost_events);
1680
1681         ftrace_enable_cpu();
1682
1683         if (event) {
1684                 iter->ent_size = ring_buffer_event_length(event);
1685                 return ring_buffer_event_data(event);
1686         }
1687         iter->ent_size = 0;
1688         return NULL;
1689 }
1690
1691 static struct trace_entry *
1692 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1693                   unsigned long *missing_events, u64 *ent_ts)
1694 {
1695         struct ring_buffer *buffer = iter->tr->buffer;
1696         struct trace_entry *ent, *next = NULL;
1697         unsigned long lost_events = 0, next_lost = 0;
1698         int cpu_file = iter->cpu_file;
1699         u64 next_ts = 0, ts;
1700         int next_cpu = -1;
1701         int next_size = 0;
1702         int cpu;
1703
1704         /*
1705          * If we are in a per_cpu trace file, don't bother by iterating over
1706          * all cpu and peek directly.
1707          */
1708         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1709                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1710                         return NULL;
1711                 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1712                 if (ent_cpu)
1713                         *ent_cpu = cpu_file;
1714
1715                 return ent;
1716         }
1717
1718         for_each_tracing_cpu(cpu) {
1719
1720                 if (ring_buffer_empty_cpu(buffer, cpu))
1721                         continue;
1722
1723                 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1724
1725                 /*
1726                  * Pick the entry with the smallest timestamp:
1727                  */
1728                 if (ent && (!next || ts < next_ts)) {
1729                         next = ent;
1730                         next_cpu = cpu;
1731                         next_ts = ts;
1732                         next_lost = lost_events;
1733                         next_size = iter->ent_size;
1734                 }
1735         }
1736
1737         iter->ent_size = next_size;
1738
1739         if (ent_cpu)
1740                 *ent_cpu = next_cpu;
1741
1742         if (ent_ts)
1743                 *ent_ts = next_ts;
1744
1745         if (missing_events)
1746                 *missing_events = next_lost;
1747
1748         return next;
1749 }
1750
1751 /* Find the next real entry, without updating the iterator itself */
1752 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1753                                           int *ent_cpu, u64 *ent_ts)
1754 {
1755         return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1756 }
1757
1758 /* Find the next real entry, and increment the iterator to the next entry */
1759 void *trace_find_next_entry_inc(struct trace_iterator *iter)
1760 {
1761         iter->ent = __find_next_entry(iter, &iter->cpu,
1762                                       &iter->lost_events, &iter->ts);
1763
1764         if (iter->ent)
1765                 trace_iterator_increment(iter);
1766
1767         return iter->ent ? iter : NULL;
1768 }
1769
1770 static void trace_consume(struct trace_iterator *iter)
1771 {
1772         /* Don't allow ftrace to trace into the ring buffers */
1773         ftrace_disable_cpu();
1774         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1775                             &iter->lost_events);
1776         ftrace_enable_cpu();
1777 }
1778
1779 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1780 {
1781         struct trace_iterator *iter = m->private;
1782         int i = (int)*pos;
1783         void *ent;
1784
1785         WARN_ON_ONCE(iter->leftover);
1786
1787         (*pos)++;
1788
1789         /* can't go backwards */
1790         if (iter->idx > i)
1791                 return NULL;
1792
1793         if (iter->idx < 0)
1794                 ent = trace_find_next_entry_inc(iter);
1795         else
1796                 ent = iter;
1797
1798         while (ent && iter->idx < i)
1799                 ent = trace_find_next_entry_inc(iter);
1800
1801         iter->pos = *pos;
1802
1803         return ent;
1804 }
1805
1806 void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1807 {
1808         struct trace_array *tr = iter->tr;
1809         struct ring_buffer_event *event;
1810         struct ring_buffer_iter *buf_iter;
1811         unsigned long entries = 0;
1812         u64 ts;
1813
1814         tr->data[cpu]->skipped_entries = 0;
1815
1816         if (!iter->buffer_iter[cpu])
1817                 return;
1818
1819         buf_iter = iter->buffer_iter[cpu];
1820         ring_buffer_iter_reset(buf_iter);
1821
1822         /*
1823          * We could have the case with the max latency tracers
1824          * that a reset never took place on a cpu. This is evident
1825          * by the timestamp being before the start of the buffer.
1826          */
1827         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1828                 if (ts >= iter->tr->time_start)
1829                         break;
1830                 entries++;
1831                 ring_buffer_read(buf_iter, NULL);
1832         }
1833
1834         tr->data[cpu]->skipped_entries = entries;
1835 }
1836
1837 /*
1838  * The current tracer is copied to avoid a global locking
1839  * all around.
1840  */
1841 static void *s_start(struct seq_file *m, loff_t *pos)
1842 {
1843         struct trace_iterator *iter = m->private;
1844         static struct tracer *old_tracer;
1845         int cpu_file = iter->cpu_file;
1846         void *p = NULL;
1847         loff_t l = 0;
1848         int cpu;
1849
1850         /* copy the tracer to avoid using a global lock all around */
1851         mutex_lock(&trace_types_lock);
1852         if (unlikely(old_tracer != current_trace && current_trace)) {
1853                 old_tracer = current_trace;
1854                 *iter->trace = *current_trace;
1855         }
1856         mutex_unlock(&trace_types_lock);
1857
1858         atomic_inc(&trace_record_cmdline_disabled);
1859
1860         if (*pos != iter->pos) {
1861                 iter->ent = NULL;
1862                 iter->cpu = 0;
1863                 iter->idx = -1;
1864
1865                 ftrace_disable_cpu();
1866
1867                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1868                         for_each_tracing_cpu(cpu)
1869                                 tracing_iter_reset(iter, cpu);
1870                 } else
1871                         tracing_iter_reset(iter, cpu_file);
1872
1873                 ftrace_enable_cpu();
1874
1875                 iter->leftover = 0;
1876                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1877                         ;
1878
1879         } else {
1880                 /*
1881                  * If we overflowed the seq_file before, then we want
1882                  * to just reuse the trace_seq buffer again.
1883                  */
1884                 if (iter->leftover)
1885                         p = iter;
1886                 else {
1887                         l = *pos - 1;
1888                         p = s_next(m, p, &l);
1889                 }
1890         }
1891
1892         trace_event_read_lock();
1893         trace_access_lock(cpu_file);
1894         return p;
1895 }
1896
1897 static void s_stop(struct seq_file *m, void *p)
1898 {
1899         struct trace_iterator *iter = m->private;
1900
1901         atomic_dec(&trace_record_cmdline_disabled);
1902         trace_access_unlock(iter->cpu_file);
1903         trace_event_read_unlock();
1904 }
1905
1906 static void
1907 get_total_entries(struct trace_array *tr, unsigned long *total, unsigned long *entries)
1908 {
1909         unsigned long count;
1910         int cpu;
1911
1912         *total = 0;
1913         *entries = 0;
1914
1915         for_each_tracing_cpu(cpu) {
1916                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1917                 /*
1918                  * If this buffer has skipped entries, then we hold all
1919                  * entries for the trace and we need to ignore the
1920                  * ones before the time stamp.
1921                  */
1922                 if (tr->data[cpu]->skipped_entries) {
1923                         count -= tr->data[cpu]->skipped_entries;
1924                         /* total is the same as the entries */
1925                         *total += count;
1926                 } else
1927                         *total += count +
1928                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1929                 *entries += count;
1930         }
1931 }
1932
1933 static void print_lat_help_header(struct seq_file *m)
1934 {
1935         seq_puts(m, "#                  _------=> CPU#            \n");
1936         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1937         seq_puts(m, "#                | / _----=> need-resched    \n");
1938         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1939         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1940         seq_puts(m, "#                |||| /     delay             \n");
1941         seq_puts(m, "#  cmd     pid   ||||| time  |   caller      \n");
1942         seq_puts(m, "#     \\   /      |||||  \\    |   /           \n");
1943 }
1944
1945 static void print_event_info(struct trace_array *tr, struct seq_file *m)
1946 {
1947         unsigned long total;
1948         unsigned long entries;
1949
1950         get_total_entries(tr, &total, &entries);
1951         seq_printf(m, "# entries-in-buffer/entries-written: %lu/%lu   #P:%d\n",
1952                    entries, total, num_online_cpus());
1953         seq_puts(m, "#\n");
1954 }
1955
1956 static void print_func_help_header(struct trace_array *tr, struct seq_file *m)
1957 {
1958         print_event_info(tr, m);
1959         seq_puts(m, "#           TASK-PID   CPU#      TIMESTAMP  FUNCTION\n");
1960         seq_puts(m, "#              | |       |          |         |\n");
1961 }
1962
1963 static void print_func_help_header_irq(struct trace_array *tr, struct seq_file *m)
1964 {
1965         print_event_info(tr, m);
1966         seq_puts(m, "#                              _-----=> irqs-off\n");
1967         seq_puts(m, "#                             / _----=> need-resched\n");
1968         seq_puts(m, "#                            | / _---=> hardirq/softirq\n");
1969         seq_puts(m, "#                            || / _--=> preempt-depth\n");
1970         seq_puts(m, "#                            ||| /     delay\n");
1971         seq_puts(m, "#           TASK-PID   CPU#  ||||    TIMESTAMP  FUNCTION\n");
1972         seq_puts(m, "#              | |       |   ||||       |         |\n");
1973 }
1974
1975 void
1976 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1977 {
1978         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1979         struct trace_array *tr = iter->tr;
1980         struct trace_array_cpu *data = tr->data[tr->cpu];
1981         struct tracer *type = current_trace;
1982         unsigned long entries;
1983         unsigned long total;
1984         const char *name = "preemption";
1985
1986         if (type)
1987                 name = type->name;
1988
1989         get_total_entries(tr, &total, &entries);
1990
1991         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1992                    name, UTS_RELEASE);
1993         seq_puts(m, "# -----------------------------------"
1994                  "---------------------------------\n");
1995         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1996                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1997                    nsecs_to_usecs(data->saved_latency),
1998                    entries,
1999                    total,
2000                    tr->cpu,
2001 #if defined(CONFIG_PREEMPT_NONE)
2002                    "server",
2003 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
2004                    "desktop",
2005 #elif defined(CONFIG_PREEMPT)
2006                    "preempt",
2007 #else
2008                    "unknown",
2009 #endif
2010                    /* These are reserved for later use */
2011                    0, 0, 0, 0);
2012 #ifdef CONFIG_SMP
2013         seq_printf(m, " #P:%d)\n", num_online_cpus());
2014 #else
2015         seq_puts(m, ")\n");
2016 #endif
2017         seq_puts(m, "#    -----------------\n");
2018         seq_printf(m, "#    | task: %.16s-%d "
2019                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2020                    data->comm, data->pid, data->uid, data->nice,
2021                    data->policy, data->rt_priority);
2022         seq_puts(m, "#    -----------------\n");
2023
2024         if (data->critical_start) {
2025                 seq_puts(m, "#  => started at: ");
2026                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
2027                 trace_print_seq(m, &iter->seq);
2028                 seq_puts(m, "\n#  => ended at:   ");
2029                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
2030                 trace_print_seq(m, &iter->seq);
2031                 seq_puts(m, "\n#\n");
2032         }
2033
2034         seq_puts(m, "#\n");
2035 }
2036
2037 static void test_cpu_buff_start(struct trace_iterator *iter)
2038 {
2039         struct trace_seq *s = &iter->seq;
2040
2041         if (!(trace_flags & TRACE_ITER_ANNOTATE))
2042                 return;
2043
2044         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
2045                 return;
2046
2047         if (cpumask_test_cpu(iter->cpu, iter->started))
2048                 return;
2049
2050         if (iter->tr->data[iter->cpu]->skipped_entries)
2051                 return;
2052
2053         cpumask_set_cpu(iter->cpu, iter->started);
2054
2055         /* Don't print started cpu buffer for the first entry of the trace */
2056         if (iter->idx > 1)
2057                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
2058                                 iter->cpu);
2059 }
2060
2061 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
2062 {
2063         struct trace_seq *s = &iter->seq;
2064         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
2065         struct trace_entry *entry;
2066         struct trace_event *event;
2067
2068         entry = iter->ent;
2069
2070         test_cpu_buff_start(iter);
2071
2072         event = ftrace_find_event(entry->type);
2073
2074         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2075                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2076                         if (!trace_print_lat_context(iter))
2077                                 goto partial;
2078                 } else {
2079                         if (!trace_print_context(iter))
2080                                 goto partial;
2081                 }
2082         }
2083
2084         if (event)
2085                 return event->funcs->trace(iter, sym_flags, event);
2086
2087         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2088                 goto partial;
2089
2090         return TRACE_TYPE_HANDLED;
2091 partial:
2092         return TRACE_TYPE_PARTIAL_LINE;
2093 }
2094
2095 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2096 {
2097         struct trace_seq *s = &iter->seq;
2098         struct trace_entry *entry;
2099         struct trace_event *event;
2100
2101         entry = iter->ent;
2102
2103         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2104                 if (!trace_seq_printf(s, "%d %d %llu ",
2105                                       entry->pid, iter->cpu, iter->ts))
2106                         goto partial;
2107         }
2108
2109         event = ftrace_find_event(entry->type);
2110         if (event)
2111                 return event->funcs->raw(iter, 0, event);
2112
2113         if (!trace_seq_printf(s, "%d ?\n", entry->type))
2114                 goto partial;
2115
2116         return TRACE_TYPE_HANDLED;
2117 partial:
2118         return TRACE_TYPE_PARTIAL_LINE;
2119 }
2120
2121 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2122 {
2123         struct trace_seq *s = &iter->seq;
2124         unsigned char newline = '\n';
2125         struct trace_entry *entry;
2126         struct trace_event *event;
2127
2128         entry = iter->ent;
2129
2130         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2131                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2132                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2133                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2134         }
2135
2136         event = ftrace_find_event(entry->type);
2137         if (event) {
2138                 enum print_line_t ret = event->funcs->hex(iter, 0, event);
2139                 if (ret != TRACE_TYPE_HANDLED)
2140                         return ret;
2141         }
2142
2143         SEQ_PUT_FIELD_RET(s, newline);
2144
2145         return TRACE_TYPE_HANDLED;
2146 }
2147
2148 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2149 {
2150         struct trace_seq *s = &iter->seq;
2151         struct trace_entry *entry;
2152         struct trace_event *event;
2153
2154         entry = iter->ent;
2155
2156         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2157                 SEQ_PUT_FIELD_RET(s, entry->pid);
2158                 SEQ_PUT_FIELD_RET(s, iter->cpu);
2159                 SEQ_PUT_FIELD_RET(s, iter->ts);
2160         }
2161
2162         event = ftrace_find_event(entry->type);
2163         return event ? event->funcs->binary(iter, 0, event) :
2164                 TRACE_TYPE_HANDLED;
2165 }
2166
2167 int trace_empty(struct trace_iterator *iter)
2168 {
2169         int cpu;
2170
2171         /* If we are looking at one CPU buffer, only check that one */
2172         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2173                 cpu = iter->cpu_file;
2174                 if (iter->buffer_iter[cpu]) {
2175                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2176                                 return 0;
2177                 } else {
2178                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2179                                 return 0;
2180                 }
2181                 return 1;
2182         }
2183
2184         for_each_tracing_cpu(cpu) {
2185                 if (iter->buffer_iter[cpu]) {
2186                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2187                                 return 0;
2188                 } else {
2189                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2190                                 return 0;
2191                 }
2192         }
2193
2194         return 1;
2195 }
2196
2197 /*  Called with trace_event_read_lock() held. */
2198 enum print_line_t print_trace_line(struct trace_iterator *iter)
2199 {
2200         enum print_line_t ret;
2201
2202         if (iter->lost_events &&
2203             !trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2204                                  iter->cpu, iter->lost_events))
2205                 return TRACE_TYPE_PARTIAL_LINE;
2206
2207         if (iter->trace && iter->trace->print_line) {
2208                 ret = iter->trace->print_line(iter);
2209                 if (ret != TRACE_TYPE_UNHANDLED)
2210                         return ret;
2211         }
2212
2213         if (iter->ent->type == TRACE_BPRINT &&
2214                         trace_flags & TRACE_ITER_PRINTK &&
2215                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2216                 return trace_print_bprintk_msg_only(iter);
2217
2218         if (iter->ent->type == TRACE_PRINT &&
2219                         trace_flags & TRACE_ITER_PRINTK &&
2220                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2221                 return trace_print_printk_msg_only(iter);
2222
2223         if (trace_flags & TRACE_ITER_BIN)
2224                 return print_bin_fmt(iter);
2225
2226         if (trace_flags & TRACE_ITER_HEX)
2227                 return print_hex_fmt(iter);
2228
2229         if (trace_flags & TRACE_ITER_RAW)
2230                 return print_raw_fmt(iter);
2231
2232         return print_trace_fmt(iter);
2233 }
2234
2235 void trace_latency_header(struct seq_file *m)
2236 {
2237         struct trace_iterator *iter = m->private;
2238
2239         /* print nothing if the buffers are empty */
2240         if (trace_empty(iter))
2241                 return;
2242
2243         if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2244                 print_trace_header(m, iter);
2245
2246         if (!(trace_flags & TRACE_ITER_VERBOSE))
2247                 print_lat_help_header(m);
2248 }
2249
2250 void trace_default_header(struct seq_file *m)
2251 {
2252         struct trace_iterator *iter = m->private;
2253
2254         if (!(trace_flags & TRACE_ITER_CONTEXT_INFO))
2255                 return;
2256
2257         if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2258                 /* print nothing if the buffers are empty */
2259                 if (trace_empty(iter))
2260                         return;
2261                 print_trace_header(m, iter);
2262                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2263                         print_lat_help_header(m);
2264         } else {
2265                 if (!(trace_flags & TRACE_ITER_VERBOSE)) {
2266                         if (trace_flags & TRACE_ITER_IRQ_INFO)
2267                                 print_func_help_header_irq(iter->tr, m);
2268                         else
2269                                 print_func_help_header(iter->tr, m);
2270                 }
2271         }
2272 }
2273
2274 static void test_ftrace_alive(struct seq_file *m)
2275 {
2276         if (!ftrace_is_dead())
2277                 return;
2278         seq_printf(m, "# WARNING: FUNCTION TRACING IS CORRUPTED\n");
2279         seq_printf(m, "#          MAY BE MISSING FUNCTION EVENTS\n");
2280 }
2281
2282 static int s_show(struct seq_file *m, void *v)
2283 {
2284         struct trace_iterator *iter = v;
2285         int ret;
2286
2287         if (iter->ent == NULL) {
2288                 if (iter->tr) {
2289                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2290                         seq_puts(m, "#\n");
2291                         test_ftrace_alive(m);
2292                 }
2293                 if (iter->trace && iter->trace->print_header)
2294                         iter->trace->print_header(m);
2295                 else
2296                         trace_default_header(m);
2297
2298         } else if (iter->leftover) {
2299                 /*
2300                  * If we filled the seq_file buffer earlier, we
2301                  * want to just show it now.
2302                  */
2303                 ret = trace_print_seq(m, &iter->seq);
2304
2305                 /* ret should this time be zero, but you never know */
2306                 iter->leftover = ret;
2307
2308         } else {
2309                 print_trace_line(iter);
2310                 ret = trace_print_seq(m, &iter->seq);
2311                 /*
2312                  * If we overflow the seq_file buffer, then it will
2313                  * ask us for this data again at start up.
2314                  * Use that instead.
2315                  *  ret is 0 if seq_file write succeeded.
2316                  *        -1 otherwise.
2317                  */
2318                 iter->leftover = ret;
2319         }
2320
2321         return 0;
2322 }
2323
2324 static const struct seq_operations tracer_seq_ops = {
2325         .start          = s_start,
2326         .next           = s_next,
2327         .stop           = s_stop,
2328         .show           = s_show,
2329 };
2330
2331 static struct trace_iterator *
2332 __tracing_open(struct inode *inode, struct file *file)
2333 {
2334         long cpu_file = (long) inode->i_private;
2335         void *fail_ret = ERR_PTR(-ENOMEM);
2336         struct trace_iterator *iter;
2337         struct seq_file *m;
2338         int cpu, ret;
2339
2340         if (tracing_disabled)
2341                 return ERR_PTR(-ENODEV);
2342
2343         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2344         if (!iter)
2345                 return ERR_PTR(-ENOMEM);
2346
2347         /*
2348          * We make a copy of the current tracer to avoid concurrent
2349          * changes on it while we are reading.
2350          */
2351         mutex_lock(&trace_types_lock);
2352         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2353         if (!iter->trace)
2354                 goto fail;
2355
2356         if (current_trace)
2357                 *iter->trace = *current_trace;
2358
2359         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2360                 goto fail;
2361
2362         if (current_trace && current_trace->print_max)
2363                 iter->tr = &max_tr;
2364         else
2365                 iter->tr = &global_trace;
2366         iter->pos = -1;
2367         mutex_init(&iter->mutex);
2368         iter->cpu_file = cpu_file;
2369
2370         /* Notify the tracer early; before we stop tracing. */
2371         if (iter->trace && iter->trace->open)
2372                 iter->trace->open(iter);
2373
2374         /* Annotate start of buffers if we had overruns */
2375         if (ring_buffer_overruns(iter->tr->buffer))
2376                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2377
2378         /* stop the trace while dumping */
2379         tracing_stop();
2380
2381         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2382                 for_each_tracing_cpu(cpu) {
2383                         iter->buffer_iter[cpu] =
2384                                 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2385                 }
2386                 ring_buffer_read_prepare_sync();
2387                 for_each_tracing_cpu(cpu) {
2388                         ring_buffer_read_start(iter->buffer_iter[cpu]);
2389                         tracing_iter_reset(iter, cpu);
2390                 }
2391         } else {
2392                 cpu = iter->cpu_file;
2393                 iter->buffer_iter[cpu] =
2394                         ring_buffer_read_prepare(iter->tr->buffer, cpu);
2395                 ring_buffer_read_prepare_sync();
2396                 ring_buffer_read_start(iter->buffer_iter[cpu]);
2397                 tracing_iter_reset(iter, cpu);
2398         }
2399
2400         ret = seq_open(file, &tracer_seq_ops);
2401         if (ret < 0) {
2402                 fail_ret = ERR_PTR(ret);
2403                 goto fail_buffer;
2404         }
2405
2406         m = file->private_data;
2407         m->private = iter;
2408
2409         mutex_unlock(&trace_types_lock);
2410
2411         return iter;
2412
2413  fail_buffer:
2414         for_each_tracing_cpu(cpu) {
2415                 if (iter->buffer_iter[cpu])
2416                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2417         }
2418         free_cpumask_var(iter->started);
2419         tracing_start();
2420  fail:
2421         mutex_unlock(&trace_types_lock);
2422         kfree(iter->trace);
2423         kfree(iter);
2424
2425         return fail_ret;
2426 }
2427
2428 int tracing_open_generic(struct inode *inode, struct file *filp)
2429 {
2430         if (tracing_disabled)
2431                 return -ENODEV;
2432
2433         filp->private_data = inode->i_private;
2434         return 0;
2435 }
2436
2437 static int tracing_release(struct inode *inode, struct file *file)
2438 {
2439         struct seq_file *m = file->private_data;
2440         struct trace_iterator *iter;
2441         int cpu;
2442
2443         if (!(file->f_mode & FMODE_READ))
2444                 return 0;
2445
2446         iter = m->private;
2447
2448         mutex_lock(&trace_types_lock);
2449         for_each_tracing_cpu(cpu) {
2450                 if (iter->buffer_iter[cpu])
2451                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2452         }
2453
2454         if (iter->trace && iter->trace->close)
2455                 iter->trace->close(iter);
2456
2457         /* reenable tracing if it was previously enabled */
2458         tracing_start();
2459         mutex_unlock(&trace_types_lock);
2460
2461         seq_release(inode, file);
2462         mutex_destroy(&iter->mutex);
2463         free_cpumask_var(iter->started);
2464         kfree(iter->trace);
2465         kfree(iter);
2466         return 0;
2467 }
2468
2469 static int tracing_open(struct inode *inode, struct file *file)
2470 {
2471         struct trace_iterator *iter;
2472         int ret = 0;
2473
2474         /* If this file was open for write, then erase contents */
2475         if ((file->f_mode & FMODE_WRITE) &&
2476             (file->f_flags & O_TRUNC)) {
2477                 long cpu = (long) inode->i_private;
2478
2479                 if (cpu == TRACE_PIPE_ALL_CPU)
2480                         tracing_reset_online_cpus(&global_trace);
2481                 else
2482                         tracing_reset(&global_trace, cpu);
2483         }
2484
2485         if (file->f_mode & FMODE_READ) {
2486                 iter = __tracing_open(inode, file);
2487                 if (IS_ERR(iter))
2488                         ret = PTR_ERR(iter);
2489                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2490                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2491         }
2492         return ret;
2493 }
2494
2495 static void *
2496 t_next(struct seq_file *m, void *v, loff_t *pos)
2497 {
2498         struct tracer *t = v;
2499
2500         (*pos)++;
2501
2502         if (t)
2503                 t = t->next;
2504
2505         return t;
2506 }
2507
2508 static void *t_start(struct seq_file *m, loff_t *pos)
2509 {
2510         struct tracer *t;
2511         loff_t l = 0;
2512
2513         mutex_lock(&trace_types_lock);
2514         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2515                 ;
2516
2517         return t;
2518 }
2519
2520 static void t_stop(struct seq_file *m, void *p)
2521 {
2522         mutex_unlock(&trace_types_lock);
2523 }
2524
2525 static int t_show(struct seq_file *m, void *v)
2526 {
2527         struct tracer *t = v;
2528
2529         if (!t)
2530                 return 0;
2531
2532         seq_printf(m, "%s", t->name);
2533         if (t->next)
2534                 seq_putc(m, ' ');
2535         else
2536                 seq_putc(m, '\n');
2537
2538         return 0;
2539 }
2540
2541 static const struct seq_operations show_traces_seq_ops = {
2542         .start          = t_start,
2543         .next           = t_next,
2544         .stop           = t_stop,
2545         .show           = t_show,
2546 };
2547
2548 static int show_traces_open(struct inode *inode, struct file *file)
2549 {
2550         if (tracing_disabled)
2551                 return -ENODEV;
2552
2553         return seq_open(file, &show_traces_seq_ops);
2554 }
2555
2556 static ssize_t
2557 tracing_write_stub(struct file *filp, const char __user *ubuf,
2558                    size_t count, loff_t *ppos)
2559 {
2560         return count;
2561 }
2562
2563 static loff_t tracing_seek(struct file *file, loff_t offset, int origin)
2564 {
2565         if (file->f_mode & FMODE_READ)
2566                 return seq_lseek(file, offset, origin);
2567         else
2568                 return 0;
2569 }
2570
2571 static const struct file_operations tracing_fops = {
2572         .open           = tracing_open,
2573         .read           = seq_read,
2574         .write          = tracing_write_stub,
2575         .llseek         = tracing_seek,
2576         .release        = tracing_release,
2577 };
2578
2579 static const struct file_operations show_traces_fops = {
2580         .open           = show_traces_open,
2581         .read           = seq_read,
2582         .release        = seq_release,
2583         .llseek         = seq_lseek,
2584 };
2585
2586 /*
2587  * Only trace on a CPU if the bitmask is set:
2588  */
2589 static cpumask_var_t tracing_cpumask;
2590
2591 /*
2592  * The tracer itself will not take this lock, but still we want
2593  * to provide a consistent cpumask to user-space:
2594  */
2595 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2596
2597 /*
2598  * Temporary storage for the character representation of the
2599  * CPU bitmask (and one more byte for the newline):
2600  */
2601 static char mask_str[NR_CPUS + 1];
2602
2603 static ssize_t
2604 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2605                      size_t count, loff_t *ppos)
2606 {
2607         int len;
2608
2609         mutex_lock(&tracing_cpumask_update_lock);
2610
2611         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2612         if (count - len < 2) {
2613                 count = -EINVAL;
2614                 goto out_err;
2615         }
2616         len += sprintf(mask_str + len, "\n");
2617         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2618
2619 out_err:
2620         mutex_unlock(&tracing_cpumask_update_lock);
2621
2622         return count;
2623 }
2624
2625 static ssize_t
2626 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2627                       size_t count, loff_t *ppos)
2628 {
2629         int err, cpu;
2630         cpumask_var_t tracing_cpumask_new;
2631
2632         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2633                 return -ENOMEM;
2634
2635         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2636         if (err)
2637                 goto err_unlock;
2638
2639         mutex_lock(&tracing_cpumask_update_lock);
2640
2641         local_irq_disable();
2642         arch_spin_lock(&ftrace_max_lock);
2643         for_each_tracing_cpu(cpu) {
2644                 /*
2645                  * Increase/decrease the disabled counter if we are
2646                  * about to flip a bit in the cpumask:
2647                  */
2648                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2649                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2650                         atomic_inc(&global_trace.data[cpu]->disabled);
2651                 }
2652                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2653                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2654                         atomic_dec(&global_trace.data[cpu]->disabled);
2655                 }
2656         }
2657         arch_spin_unlock(&ftrace_max_lock);
2658         local_irq_enable();
2659
2660         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2661
2662         mutex_unlock(&tracing_cpumask_update_lock);
2663         free_cpumask_var(tracing_cpumask_new);
2664
2665         return count;
2666
2667 err_unlock:
2668         free_cpumask_var(tracing_cpumask_new);
2669
2670         return err;
2671 }
2672
2673 static const struct file_operations tracing_cpumask_fops = {
2674         .open           = tracing_open_generic,
2675         .read           = tracing_cpumask_read,
2676         .write          = tracing_cpumask_write,
2677         .llseek         = generic_file_llseek,
2678 };
2679
2680 static int tracing_trace_options_show(struct seq_file *m, void *v)
2681 {
2682         struct tracer_opt *trace_opts;
2683         u32 tracer_flags;
2684         int i;
2685
2686         mutex_lock(&trace_types_lock);
2687         tracer_flags = current_trace->flags->val;
2688         trace_opts = current_trace->flags->opts;
2689
2690         for (i = 0; trace_options[i]; i++) {
2691                 if (trace_flags & (1 << i))
2692                         seq_printf(m, "%s\n", trace_options[i]);
2693                 else
2694                         seq_printf(m, "no%s\n", trace_options[i]);
2695         }
2696
2697         for (i = 0; trace_opts[i].name; i++) {
2698                 if (tracer_flags & trace_opts[i].bit)
2699                         seq_printf(m, "%s\n", trace_opts[i].name);
2700                 else
2701                         seq_printf(m, "no%s\n", trace_opts[i].name);
2702         }
2703         mutex_unlock(&trace_types_lock);
2704
2705         return 0;
2706 }
2707
2708 static int __set_tracer_option(struct tracer *trace,
2709                                struct tracer_flags *tracer_flags,
2710                                struct tracer_opt *opts, int neg)
2711 {
2712         int ret;
2713
2714         ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2715         if (ret)
2716                 return ret;
2717
2718         if (neg)
2719                 tracer_flags->val &= ~opts->bit;
2720         else
2721                 tracer_flags->val |= opts->bit;
2722         return 0;
2723 }
2724
2725 /* Try to assign a tracer specific option */
2726 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2727 {
2728         struct tracer_flags *tracer_flags = trace->flags;
2729         struct tracer_opt *opts = NULL;
2730         int i;
2731
2732         for (i = 0; tracer_flags->opts[i].name; i++) {
2733                 opts = &tracer_flags->opts[i];
2734
2735                 if (strcmp(cmp, opts->name) == 0)
2736                         return __set_tracer_option(trace, trace->flags,
2737                                                    opts, neg);
2738         }
2739
2740         return -EINVAL;
2741 }
2742
2743 static void set_tracer_flags(unsigned int mask, int enabled)
2744 {
2745         /* do nothing if flag is already set */
2746         if (!!(trace_flags & mask) == !!enabled)
2747                 return;
2748
2749         if (enabled)
2750                 trace_flags |= mask;
2751         else
2752                 trace_flags &= ~mask;
2753
2754         if (mask == TRACE_ITER_RECORD_CMD)
2755                 trace_event_enable_cmd_record(enabled);
2756
2757         if (mask == TRACE_ITER_OVERWRITE)
2758                 ring_buffer_change_overwrite(global_trace.buffer, enabled);
2759 }
2760
2761 static ssize_t
2762 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2763                         size_t cnt, loff_t *ppos)
2764 {
2765         char buf[64];
2766         char *cmp;
2767         int neg = 0;
2768         int ret;
2769         int i;
2770
2771         if (cnt >= sizeof(buf))
2772                 return -EINVAL;
2773
2774         if (copy_from_user(&buf, ubuf, cnt))
2775                 return -EFAULT;
2776
2777         buf[cnt] = 0;
2778         cmp = strstrip(buf);
2779
2780         if (strncmp(cmp, "no", 2) == 0) {
2781                 neg = 1;
2782                 cmp += 2;
2783         }
2784
2785         for (i = 0; trace_options[i]; i++) {
2786                 if (strcmp(cmp, trace_options[i]) == 0) {
2787                         set_tracer_flags(1 << i, !neg);
2788                         break;
2789                 }
2790         }
2791
2792         /* If no option could be set, test the specific tracer options */
2793         if (!trace_options[i]) {
2794                 mutex_lock(&trace_types_lock);
2795                 ret = set_tracer_option(current_trace, cmp, neg);
2796                 mutex_unlock(&trace_types_lock);
2797                 if (ret)
2798                         return ret;
2799         }
2800
2801         *ppos += cnt;
2802
2803         return cnt;
2804 }
2805
2806 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2807 {
2808         if (tracing_disabled)
2809                 return -ENODEV;
2810         return single_open(file, tracing_trace_options_show, NULL);
2811 }
2812
2813 static const struct file_operations tracing_iter_fops = {
2814         .open           = tracing_trace_options_open,
2815         .read           = seq_read,
2816         .llseek         = seq_lseek,
2817         .release        = single_release,
2818         .write          = tracing_trace_options_write,
2819 };
2820
2821 static const char readme_msg[] =
2822         "tracing mini-HOWTO:\n\n"
2823         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2824         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2825         "wakeup wakeup_rt preemptirqsoff preemptoff irqsoff function nop\n\n"
2826         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2827         "nop\n"
2828         "# echo wakeup > /sys/kernel/debug/tracing/current_tracer\n"
2829         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2830         "wakeup\n"
2831         "# cat /sys/kernel/debug/tracing/trace_options\n"
2832         "noprint-parent nosym-offset nosym-addr noverbose\n"
2833         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2834         "# echo 1 > /sys/kernel/debug/tracing/tracing_on\n"
2835         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2836         "# echo 0 > /sys/kernel/debug/tracing/tracing_on\n"
2837 ;
2838
2839 static ssize_t
2840 tracing_readme_read(struct file *filp, char __user *ubuf,
2841                        size_t cnt, loff_t *ppos)
2842 {
2843         return simple_read_from_buffer(ubuf, cnt, ppos,
2844                                         readme_msg, strlen(readme_msg));
2845 }
2846
2847 static const struct file_operations tracing_readme_fops = {
2848         .open           = tracing_open_generic,
2849         .read           = tracing_readme_read,
2850         .llseek         = generic_file_llseek,
2851 };
2852
2853 static ssize_t
2854 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2855                                 size_t cnt, loff_t *ppos)
2856 {
2857         char *buf_comm;
2858         char *file_buf;
2859         char *buf;
2860         int len = 0;
2861         int pid;
2862         int i;
2863
2864         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2865         if (!file_buf)
2866                 return -ENOMEM;
2867
2868         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2869         if (!buf_comm) {
2870                 kfree(file_buf);
2871                 return -ENOMEM;
2872         }
2873
2874         buf = file_buf;
2875
2876         for (i = 0; i < SAVED_CMDLINES; i++) {
2877                 int r;
2878
2879                 pid = map_cmdline_to_pid[i];
2880                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2881                         continue;
2882
2883                 trace_find_cmdline(pid, buf_comm);
2884                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2885                 buf += r;
2886                 len += r;
2887         }
2888
2889         len = simple_read_from_buffer(ubuf, cnt, ppos,
2890                                       file_buf, len);
2891
2892         kfree(file_buf);
2893         kfree(buf_comm);
2894
2895         return len;
2896 }
2897
2898 static const struct file_operations tracing_saved_cmdlines_fops = {
2899     .open       = tracing_open_generic,
2900     .read       = tracing_saved_cmdlines_read,
2901     .llseek     = generic_file_llseek,
2902 };
2903
2904 static ssize_t
2905 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2906                   size_t cnt, loff_t *ppos)
2907 {
2908         char buf[64];
2909         int r;
2910
2911         r = sprintf(buf, "%u\n", tracer_enabled);
2912         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2913 }
2914
2915 static ssize_t
2916 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2917                    size_t cnt, loff_t *ppos)
2918 {
2919         struct trace_array *tr = filp->private_data;
2920         unsigned long val;
2921         int ret;
2922
2923         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
2924         if (ret)
2925                 return ret;
2926
2927         val = !!val;
2928
2929         mutex_lock(&trace_types_lock);
2930         if (tracer_enabled ^ val) {
2931
2932                 /* Only need to warn if this is used to change the state */
2933                 WARN_ONCE(1, "tracing_enabled is deprecated. Use tracing_on");
2934
2935                 if (val) {
2936                         tracer_enabled = 1;
2937                         if (current_trace->start)
2938                                 current_trace->start(tr);
2939                         tracing_start();
2940                 } else {
2941                         tracer_enabled = 0;
2942                         tracing_stop();
2943                         if (current_trace->stop)
2944                                 current_trace->stop(tr);
2945                 }
2946         }
2947         mutex_unlock(&trace_types_lock);
2948
2949         *ppos += cnt;
2950
2951         return cnt;
2952 }
2953
2954 static ssize_t
2955 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2956                        size_t cnt, loff_t *ppos)
2957 {
2958         char buf[MAX_TRACER_SIZE+2];
2959         int r;
2960
2961         mutex_lock(&trace_types_lock);
2962         if (current_trace)
2963                 r = sprintf(buf, "%s\n", current_trace->name);
2964         else
2965                 r = sprintf(buf, "\n");
2966         mutex_unlock(&trace_types_lock);
2967
2968         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2969 }
2970
2971 int tracer_init(struct tracer *t, struct trace_array *tr)
2972 {
2973         tracing_reset_online_cpus(tr);
2974         return t->init(tr);
2975 }
2976
2977 static int __tracing_resize_ring_buffer(unsigned long size)
2978 {
2979         int ret;
2980
2981         /*
2982          * If kernel or user changes the size of the ring buffer
2983          * we use the size that was given, and we can forget about
2984          * expanding it later.
2985          */
2986         ring_buffer_expanded = 1;
2987
2988         ret = ring_buffer_resize(global_trace.buffer, size);
2989         if (ret < 0)
2990                 return ret;
2991
2992         if (!current_trace->use_max_tr)
2993                 goto out;
2994
2995         ret = ring_buffer_resize(max_tr.buffer, size);
2996         if (ret < 0) {
2997                 int r;
2998
2999                 r = ring_buffer_resize(global_trace.buffer,
3000                                        global_trace.entries);
3001                 if (r < 0) {
3002                         /*
3003                          * AARGH! We are left with different
3004                          * size max buffer!!!!
3005                          * The max buffer is our "snapshot" buffer.
3006                          * When a tracer needs a snapshot (one of the
3007                          * latency tracers), it swaps the max buffer
3008                          * with the saved snap shot. We succeeded to
3009                          * update the size of the main buffer, but failed to
3010                          * update the size of the max buffer. But when we tried
3011                          * to reset the main buffer to the original size, we
3012                          * failed there too. This is very unlikely to
3013                          * happen, but if it does, warn and kill all
3014                          * tracing.
3015                          */
3016                         WARN_ON(1);
3017                         tracing_disabled = 1;
3018                 }
3019                 return ret;
3020         }
3021
3022         max_tr.entries = size;
3023  out:
3024         global_trace.entries = size;
3025
3026         return ret;
3027 }
3028
3029 static ssize_t tracing_resize_ring_buffer(unsigned long size)
3030 {
3031         int cpu, ret = size;
3032
3033         mutex_lock(&trace_types_lock);
3034
3035         tracing_stop();
3036
3037         /* disable all cpu buffers */
3038         for_each_tracing_cpu(cpu) {
3039                 if (global_trace.data[cpu])
3040                         atomic_inc(&global_trace.data[cpu]->disabled);
3041                 if (max_tr.data[cpu])
3042                         atomic_inc(&max_tr.data[cpu]->disabled);
3043         }
3044
3045         if (size != global_trace.entries)
3046                 ret = __tracing_resize_ring_buffer(size);
3047
3048         if (ret < 0)
3049                 ret = -ENOMEM;
3050
3051         for_each_tracing_cpu(cpu) {
3052                 if (global_trace.data[cpu])
3053                         atomic_dec(&global_trace.data[cpu]->disabled);
3054                 if (max_tr.data[cpu])
3055                         atomic_dec(&max_tr.data[cpu]->disabled);
3056         }
3057
3058         tracing_start();
3059         mutex_unlock(&trace_types_lock);
3060
3061         return ret;
3062 }
3063
3064
3065 /**
3066  * tracing_update_buffers - used by tracing facility to expand ring buffers
3067  *
3068  * To save on memory when the tracing is never used on a system with it
3069  * configured in. The ring buffers are set to a minimum size. But once
3070  * a user starts to use the tracing facility, then they need to grow
3071  * to their default size.
3072  *
3073  * This function is to be called when a tracer is about to be used.
3074  */
3075 int tracing_update_buffers(void)
3076 {
3077         int ret = 0;
3078
3079         mutex_lock(&trace_types_lock);
3080         if (!ring_buffer_expanded)
3081                 ret = __tracing_resize_ring_buffer(trace_buf_size);
3082         mutex_unlock(&trace_types_lock);
3083
3084         return ret;
3085 }
3086
3087 struct trace_option_dentry;
3088
3089 static struct trace_option_dentry *
3090 create_trace_option_files(struct tracer *tracer);
3091
3092 static void
3093 destroy_trace_option_files(struct trace_option_dentry *topts);
3094
3095 static int tracing_set_tracer(const char *buf)
3096 {
3097         static struct trace_option_dentry *topts;
3098         struct trace_array *tr = &global_trace;
3099         struct tracer *t;
3100         int ret = 0;
3101
3102         mutex_lock(&trace_types_lock);
3103
3104         if (!ring_buffer_expanded) {
3105                 ret = __tracing_resize_ring_buffer(trace_buf_size);
3106                 if (ret < 0)
3107                         goto out;
3108                 ret = 0;
3109         }
3110
3111         for (t = trace_types; t; t = t->next) {
3112                 if (strcmp(t->name, buf) == 0)
3113                         break;
3114         }
3115         if (!t) {
3116                 ret = -EINVAL;
3117                 goto out;
3118         }
3119         if (t == current_trace)
3120                 goto out;
3121
3122         trace_branch_disable();
3123         if (current_trace && current_trace->reset)
3124                 current_trace->reset(tr);
3125         if (current_trace && current_trace->use_max_tr) {
3126                 /*
3127                  * We don't free the ring buffer. instead, resize it because
3128                  * The max_tr ring buffer has some state (e.g. ring->clock) and
3129                  * we want preserve it.
3130                  */
3131                 ring_buffer_resize(max_tr.buffer, 1);
3132                 max_tr.entries = 1;
3133         }
3134         destroy_trace_option_files(topts);
3135
3136         current_trace = t;
3137
3138         topts = create_trace_option_files(current_trace);
3139         if (current_trace->use_max_tr) {
3140                 ret = ring_buffer_resize(max_tr.buffer, global_trace.entries);
3141                 if (ret < 0)
3142                         goto out;
3143                 max_tr.entries = global_trace.entries;
3144         }
3145
3146         if (t->init) {
3147                 ret = tracer_init(t, tr);
3148                 if (ret)
3149                         goto out;
3150         }
3151
3152         trace_branch_enable(tr);
3153  out:
3154         mutex_unlock(&trace_types_lock);
3155
3156         return ret;
3157 }
3158
3159 static ssize_t
3160 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
3161                         size_t cnt, loff_t *ppos)
3162 {
3163         char buf[MAX_TRACER_SIZE+1];
3164         int i;
3165         size_t ret;
3166         int err;
3167
3168         ret = cnt;
3169
3170         if (cnt > MAX_TRACER_SIZE)
3171                 cnt = MAX_TRACER_SIZE;
3172
3173         if (copy_from_user(&buf, ubuf, cnt))
3174                 return -EFAULT;
3175
3176         buf[cnt] = 0;
3177
3178         /* strip ending whitespace. */
3179         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
3180                 buf[i] = 0;
3181
3182         err = tracing_set_tracer(buf);
3183         if (err)
3184                 return err;
3185
3186         *ppos += ret;
3187
3188         return ret;
3189 }
3190
3191 static ssize_t
3192 tracing_max_lat_read(struct file *filp, char __user *ubuf,
3193                      size_t cnt, loff_t *ppos)
3194 {
3195         unsigned long *ptr = filp->private_data;
3196         char buf[64];
3197         int r;
3198
3199         r = snprintf(buf, sizeof(buf), "%ld\n",
3200                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
3201         if (r > sizeof(buf))
3202                 r = sizeof(buf);
3203         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3204 }
3205
3206 static ssize_t
3207 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3208                       size_t cnt, loff_t *ppos)
3209 {
3210         unsigned long *ptr = filp->private_data;
3211         unsigned long val;
3212         int ret;
3213
3214         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3215         if (ret)
3216                 return ret;
3217
3218         *ptr = val * 1000;
3219
3220         return cnt;
3221 }
3222
3223 static int tracing_open_pipe(struct inode *inode, struct file *filp)
3224 {
3225         long cpu_file = (long) inode->i_private;
3226         struct trace_iterator *iter;
3227         int ret = 0;
3228
3229         if (tracing_disabled)
3230                 return -ENODEV;
3231
3232         mutex_lock(&trace_types_lock);
3233
3234         /* create a buffer to store the information to pass to userspace */
3235         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3236         if (!iter) {
3237                 ret = -ENOMEM;
3238                 goto out;
3239         }
3240
3241         /*
3242          * We make a copy of the current tracer to avoid concurrent
3243          * changes on it while we are reading.
3244          */
3245         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3246         if (!iter->trace) {
3247                 ret = -ENOMEM;
3248                 goto fail;
3249         }
3250         if (current_trace)
3251                 *iter->trace = *current_trace;
3252
3253         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3254                 ret = -ENOMEM;
3255                 goto fail;
3256         }
3257
3258         /* trace pipe does not show start of buffer */
3259         cpumask_setall(iter->started);
3260
3261         if (trace_flags & TRACE_ITER_LATENCY_FMT)
3262                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3263
3264         iter->cpu_file = cpu_file;
3265         iter->tr = &global_trace;
3266         mutex_init(&iter->mutex);
3267         filp->private_data = iter;
3268
3269         if (iter->trace->pipe_open)
3270                 iter->trace->pipe_open(iter);
3271
3272         nonseekable_open(inode, filp);
3273 out:
3274         mutex_unlock(&trace_types_lock);
3275         return ret;
3276
3277 fail:
3278         kfree(iter->trace);
3279         kfree(iter);
3280         mutex_unlock(&trace_types_lock);
3281         return ret;
3282 }
3283
3284 static int tracing_release_pipe(struct inode *inode, struct file *file)
3285 {
3286         struct trace_iterator *iter = file->private_data;
3287
3288         mutex_lock(&trace_types_lock);
3289
3290         if (iter->trace->pipe_close)
3291                 iter->trace->pipe_close(iter);
3292
3293         mutex_unlock(&trace_types_lock);
3294
3295         free_cpumask_var(iter->started);
3296         mutex_destroy(&iter->mutex);
3297         kfree(iter->trace);
3298         kfree(iter);
3299
3300         return 0;
3301 }
3302
3303 static unsigned int
3304 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3305 {
3306         struct trace_iterator *iter = filp->private_data;
3307
3308         if (trace_flags & TRACE_ITER_BLOCK) {
3309                 /*
3310                  * Always select as readable when in blocking mode
3311                  */
3312                 return POLLIN | POLLRDNORM;
3313         } else {
3314                 if (!trace_empty(iter))
3315                         return POLLIN | POLLRDNORM;
3316                 poll_wait(filp, &trace_wait, poll_table);
3317                 if (!trace_empty(iter))
3318                         return POLLIN | POLLRDNORM;
3319
3320                 return 0;
3321         }
3322 }
3323
3324
3325 void default_wait_pipe(struct trace_iterator *iter)
3326 {
3327         DEFINE_WAIT(wait);
3328
3329         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3330
3331         if (trace_empty(iter))
3332                 schedule();
3333
3334         finish_wait(&trace_wait, &wait);
3335 }
3336
3337 /*
3338  * This is a make-shift waitqueue.
3339  * A tracer might use this callback on some rare cases:
3340  *
3341  *  1) the current tracer might hold the runqueue lock when it wakes up
3342  *     a reader, hence a deadlock (sched, function, and function graph tracers)
3343  *  2) the function tracers, trace all functions, we don't want
3344  *     the overhead of calling wake_up and friends
3345  *     (and tracing them too)
3346  *
3347  *     Anyway, this is really very primitive wakeup.
3348  */
3349 void poll_wait_pipe(struct trace_iterator *iter)
3350 {
3351         set_current_state(TASK_INTERRUPTIBLE);
3352         /* sleep for 100 msecs, and try again. */
3353         schedule_timeout(HZ / 10);
3354 }
3355
3356 /* Must be called with trace_types_lock mutex held. */
3357 static int tracing_wait_pipe(struct file *filp)
3358 {
3359         struct trace_iterator *iter = filp->private_data;
3360
3361         while (trace_empty(iter)) {
3362
3363                 if ((filp->f_flags & O_NONBLOCK)) {
3364                         return -EAGAIN;
3365                 }
3366
3367                 mutex_unlock(&iter->mutex);
3368
3369                 iter->trace->wait_pipe(iter);
3370
3371                 mutex_lock(&iter->mutex);
3372
3373                 if (signal_pending(current))
3374                         return -EINTR;
3375
3376                 /*
3377                  * We block until we read something and tracing is disabled.
3378                  * We still block if tracing is disabled, but we have never
3379                  * read anything. This allows a user to cat this file, and
3380                  * then enable tracing. But after we have read something,
3381                  * we give an EOF when tracing is again disabled.
3382                  *
3383                  * iter->pos will be 0 if we haven't read anything.
3384                  */
3385                 if (!tracer_enabled && iter->pos)
3386                         break;
3387         }
3388
3389         return 1;
3390 }
3391
3392 /*
3393  * Consumer reader.
3394  */
3395 static ssize_t
3396 tracing_read_pipe(struct file *filp, char __user *ubuf,
3397                   size_t cnt, loff_t *ppos)
3398 {
3399         struct trace_iterator *iter = filp->private_data;
3400         static struct tracer *old_tracer;
3401         ssize_t sret;
3402
3403         /* return any leftover data */
3404         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3405         if (sret != -EBUSY)
3406                 return sret;
3407
3408         trace_seq_init(&iter->seq);
3409
3410         /* copy the tracer to avoid using a global lock all around */
3411         mutex_lock(&trace_types_lock);
3412         if (unlikely(old_tracer != current_trace && current_trace)) {
3413                 old_tracer = current_trace;
3414                 *iter->trace = *current_trace;
3415         }
3416         mutex_unlock(&trace_types_lock);
3417
3418         /*
3419          * Avoid more than one consumer on a single file descriptor
3420          * This is just a matter of traces coherency, the ring buffer itself
3421          * is protected.
3422          */
3423         mutex_lock(&iter->mutex);
3424         if (iter->trace->read) {
3425                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3426                 if (sret)
3427                         goto out;
3428         }
3429
3430 waitagain:
3431         sret = tracing_wait_pipe(filp);
3432         if (sret <= 0)
3433                 goto out;
3434
3435         /* stop when tracing is finished */
3436         if (trace_empty(iter)) {
3437                 sret = 0;
3438                 goto out;
3439         }
3440
3441         if (cnt >= PAGE_SIZE)
3442                 cnt = PAGE_SIZE - 1;
3443
3444         /* reset all but tr, trace, and overruns */
3445         memset(&iter->seq, 0,
3446                sizeof(struct trace_iterator) -
3447                offsetof(struct trace_iterator, seq));
3448         iter->pos = -1;
3449
3450         trace_event_read_lock();
3451         trace_access_lock(iter->cpu_file);
3452         while (trace_find_next_entry_inc(iter) != NULL) {
3453                 enum print_line_t ret;
3454                 int len = iter->seq.len;
3455
3456                 ret = print_trace_line(iter);
3457                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3458                         /* don't print partial lines */
3459                         iter->seq.len = len;
3460                         break;
3461                 }
3462                 if (ret != TRACE_TYPE_NO_CONSUME)
3463                         trace_consume(iter);
3464
3465                 if (iter->seq.len >= cnt)
3466                         break;
3467
3468                 /*
3469                  * Setting the full flag means we reached the trace_seq buffer
3470                  * size and we should leave by partial output condition above.
3471                  * One of the trace_seq_* functions is not used properly.
3472                  */
3473                 WARN_ONCE(iter->seq.full, "full flag set for trace type %d",
3474                           iter->ent->type);
3475         }
3476         trace_access_unlock(iter->cpu_file);
3477         trace_event_read_unlock();
3478
3479         /* Now copy what we have to the user */
3480         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3481         if (iter->seq.readpos >= iter->seq.len)
3482                 trace_seq_init(&iter->seq);
3483
3484         /*
3485          * If there was nothing to send to user, in spite of consuming trace
3486          * entries, go back to wait for more entries.
3487          */
3488         if (sret == -EBUSY)
3489                 goto waitagain;
3490
3491 out:
3492         mutex_unlock(&iter->mutex);
3493
3494         return sret;
3495 }
3496
3497 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3498                                      struct pipe_buffer *buf)
3499 {
3500         __free_page(buf->page);
3501 }
3502
3503 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3504                                      unsigned int idx)
3505 {
3506         __free_page(spd->pages[idx]);
3507 }
3508
3509 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3510         .can_merge              = 0,
3511         .map                    = generic_pipe_buf_map,
3512         .unmap                  = generic_pipe_buf_unmap,
3513         .confirm                = generic_pipe_buf_confirm,
3514         .release                = tracing_pipe_buf_release,
3515         .steal                  = generic_pipe_buf_steal,
3516         .get                    = generic_pipe_buf_get,
3517 };
3518
3519 static size_t
3520 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3521 {
3522         size_t count;
3523         int ret;
3524
3525         /* Seq buffer is page-sized, exactly what we need. */
3526         for (;;) {
3527                 count = iter->seq.len;
3528                 ret = print_trace_line(iter);
3529                 count = iter->seq.len - count;
3530                 if (rem < count) {
3531                         rem = 0;
3532                         iter->seq.len -= count;
3533                         break;
3534                 }
3535                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3536                         iter->seq.len -= count;
3537                         break;
3538                 }
3539
3540                 if (ret != TRACE_TYPE_NO_CONSUME)
3541                         trace_consume(iter);
3542                 rem -= count;
3543                 if (!trace_find_next_entry_inc(iter))   {
3544                         rem = 0;
3545                         iter->ent = NULL;
3546                         break;
3547                 }
3548         }
3549
3550         return rem;
3551 }
3552
3553 static ssize_t tracing_splice_read_pipe(struct file *filp,
3554                                         loff_t *ppos,
3555                                         struct pipe_inode_info *pipe,
3556                                         size_t len,
3557                                         unsigned int flags)
3558 {
3559         struct page *pages_def[PIPE_DEF_BUFFERS];
3560         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3561         struct trace_iterator *iter = filp->private_data;
3562         struct splice_pipe_desc spd = {
3563                 .pages          = pages_def,
3564                 .partial        = partial_def,
3565                 .nr_pages       = 0, /* This gets updated below. */
3566                 .flags          = flags,
3567                 .ops            = &tracing_pipe_buf_ops,
3568                 .spd_release    = tracing_spd_release_pipe,
3569         };
3570         static struct tracer *old_tracer;
3571         ssize_t ret;
3572         size_t rem;
3573         unsigned int i;
3574
3575         if (splice_grow_spd(pipe, &spd))
3576                 return -ENOMEM;
3577
3578         /* copy the tracer to avoid using a global lock all around */
3579         mutex_lock(&trace_types_lock);
3580         if (unlikely(old_tracer != current_trace && current_trace)) {
3581                 old_tracer = current_trace;
3582                 *iter->trace = *current_trace;
3583         }
3584         mutex_unlock(&trace_types_lock);
3585
3586         mutex_lock(&iter->mutex);
3587
3588         if (iter->trace->splice_read) {
3589                 ret = iter->trace->splice_read(iter, filp,
3590                                                ppos, pipe, len, flags);
3591                 if (ret)
3592                         goto out_err;
3593         }
3594
3595         ret = tracing_wait_pipe(filp);
3596         if (ret <= 0)
3597                 goto out_err;
3598
3599         if (!iter->ent && !trace_find_next_entry_inc(iter)) {
3600                 ret = -EFAULT;
3601                 goto out_err;
3602         }
3603
3604         trace_event_read_lock();
3605         trace_access_lock(iter->cpu_file);
3606
3607         /* Fill as many pages as possible. */
3608         for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3609                 spd.pages[i] = alloc_page(GFP_KERNEL);
3610                 if (!spd.pages[i])
3611                         break;
3612
3613                 rem = tracing_fill_pipe_page(rem, iter);
3614
3615                 /* Copy the data into the page, so we can start over. */
3616                 ret = trace_seq_to_buffer(&iter->seq,
3617                                           page_address(spd.pages[i]),
3618                                           iter->seq.len);
3619                 if (ret < 0) {
3620                         __free_page(spd.pages[i]);
3621                         break;
3622                 }
3623                 spd.partial[i].offset = 0;
3624                 spd.partial[i].len = iter->seq.len;
3625
3626                 trace_seq_init(&iter->seq);
3627         }
3628
3629         trace_access_unlock(iter->cpu_file);
3630         trace_event_read_unlock();
3631         mutex_unlock(&iter->mutex);
3632
3633         spd.nr_pages = i;
3634
3635         ret = splice_to_pipe(pipe, &spd);
3636 out:
3637         splice_shrink_spd(pipe, &spd);
3638         return ret;
3639
3640 out_err:
3641         mutex_unlock(&iter->mutex);
3642         goto out;
3643 }
3644
3645 static ssize_t
3646 tracing_entries_read(struct file *filp, char __user *ubuf,
3647                      size_t cnt, loff_t *ppos)
3648 {
3649         struct trace_array *tr = filp->private_data;
3650         char buf[96];
3651         int r;
3652
3653         mutex_lock(&trace_types_lock);
3654         if (!ring_buffer_expanded)
3655                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3656                             tr->entries >> 10,
3657                             trace_buf_size >> 10);
3658         else
3659                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3660         mutex_unlock(&trace_types_lock);
3661
3662         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3663 }
3664
3665 static ssize_t
3666 tracing_entries_write(struct file *filp, const char __user *ubuf,
3667                       size_t cnt, loff_t *ppos)
3668 {
3669         unsigned long val;
3670         int ret;
3671
3672         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
3673         if (ret)
3674                 return ret;
3675
3676         /* must have at least 1 entry */
3677         if (!val)
3678                 return -EINVAL;
3679
3680         /* value is in KB */
3681         val <<= 10;
3682
3683         ret = tracing_resize_ring_buffer(val);
3684         if (ret < 0)
3685                 return ret;
3686
3687         *ppos += cnt;
3688
3689         return cnt;
3690 }
3691
3692 static ssize_t
3693 tracing_total_entries_read(struct file *filp, char __user *ubuf,
3694                                 size_t cnt, loff_t *ppos)
3695 {
3696         struct trace_array *tr = filp->private_data;
3697         char buf[64];
3698         int r, cpu;
3699         unsigned long size = 0, expanded_size = 0;
3700
3701         mutex_lock(&trace_types_lock);
3702         for_each_tracing_cpu(cpu) {
3703                 size += tr->entries >> 10;
3704                 if (!ring_buffer_expanded)
3705                         expanded_size += trace_buf_size >> 10;
3706         }
3707         if (ring_buffer_expanded)
3708                 r = sprintf(buf, "%lu\n", size);
3709         else
3710                 r = sprintf(buf, "%lu (expanded: %lu)\n", size, expanded_size);
3711         mutex_unlock(&trace_types_lock);
3712
3713         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3714 }
3715
3716 static ssize_t
3717 tracing_free_buffer_write(struct file *filp, const char __user *ubuf,
3718                           size_t cnt, loff_t *ppos)
3719 {
3720         /*
3721          * There is no need to read what the user has written, this function
3722          * is just to make sure that there is no error when "echo" is used
3723          */
3724
3725         *ppos += cnt;
3726
3727         return cnt;
3728 }
3729
3730 static int
3731 tracing_free_buffer_release(struct inode *inode, struct file *filp)
3732 {
3733         /* disable tracing ? */
3734         if (trace_flags & TRACE_ITER_STOP_ON_FREE)
3735                 tracing_off();
3736         /* resize the ring buffer to 0 */
3737         tracing_resize_ring_buffer(0);
3738
3739         return 0;
3740 }
3741
3742 static ssize_t
3743 tracing_mark_write(struct file *filp, const char __user *ubuf,
3744                                         size_t cnt, loff_t *fpos)
3745 {
3746         unsigned long addr = (unsigned long)ubuf;
3747         struct ring_buffer_event *event;
3748         struct ring_buffer *buffer;
3749         struct print_entry *entry;
3750         unsigned long irq_flags;
3751         struct page *pages[2];
3752         int nr_pages = 1;
3753         ssize_t written;
3754         void *page1;
3755         void *page2;
3756         int offset;
3757         int size;
3758         int len;
3759         int ret;
3760
3761         if (tracing_disabled)
3762                 return -EINVAL;
3763
3764         if (cnt > TRACE_BUF_SIZE)
3765                 cnt = TRACE_BUF_SIZE;
3766
3767         /*
3768          * Userspace is injecting traces into the kernel trace buffer.
3769          * We want to be as non intrusive as possible.
3770          * To do so, we do not want to allocate any special buffers
3771          * or take any locks, but instead write the userspace data
3772          * straight into the ring buffer.
3773          *
3774          * First we need to pin the userspace buffer into memory,
3775          * which, most likely it is, because it just referenced it.
3776          * But there's no guarantee that it is. By using get_user_pages_fast()
3777          * and kmap_atomic/kunmap_atomic() we can get access to the
3778          * pages directly. We then write the data directly into the
3779          * ring buffer.
3780          */
3781         BUILD_BUG_ON(TRACE_BUF_SIZE >= PAGE_SIZE);
3782
3783         /* check if we cross pages */
3784         if ((addr & PAGE_MASK) != ((addr + cnt) & PAGE_MASK))
3785                 nr_pages = 2;
3786
3787         offset = addr & (PAGE_SIZE - 1);
3788         addr &= PAGE_MASK;
3789
3790         ret = get_user_pages_fast(addr, nr_pages, 0, pages);
3791         if (ret < nr_pages) {
3792                 while (--ret >= 0)
3793                         put_page(pages[ret]);
3794                 written = -EFAULT;
3795                 goto out;
3796         }
3797
3798         page1 = kmap_atomic(pages[0]);
3799         if (nr_pages == 2)
3800                 page2 = kmap_atomic(pages[1]);
3801
3802         local_save_flags(irq_flags);
3803         size = sizeof(*entry) + cnt + 2; /* possible \n added */
3804         buffer = global_trace.buffer;
3805         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
3806                                           irq_flags, preempt_count());
3807         if (!event) {
3808                 /* Ring buffer disabled, return as if not open for write */
3809                 written = -EBADF;
3810                 goto out_unlock;
3811         }
3812
3813         entry = ring_buffer_event_data(event);
3814         entry->ip = _THIS_IP_;
3815
3816         if (nr_pages == 2) {
3817                 len = PAGE_SIZE - offset;
3818                 memcpy(&entry->buf, page1 + offset, len);
3819                 memcpy(&entry->buf[len], page2, cnt - len);
3820         } else
3821                 memcpy(&entry->buf, page1 + offset, cnt);
3822
3823         if (entry->buf[cnt - 1] != '\n') {
3824                 entry->buf[cnt] = '\n';
3825                 entry->buf[cnt + 1] = '\0';
3826         } else
3827                 entry->buf[cnt] = '\0';
3828
3829         ring_buffer_unlock_commit(buffer, event);
3830
3831         written = cnt;
3832
3833         *fpos += written;
3834
3835  out_unlock:
3836         if (nr_pages == 2)
3837                 kunmap_atomic(page2);
3838         kunmap_atomic(page1);
3839         while (nr_pages > 0)
3840                 put_page(pages[--nr_pages]);
3841  out:
3842         return written;
3843 }
3844
3845 static int tracing_clock_show(struct seq_file *m, void *v)
3846 {
3847         int i;
3848
3849         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3850                 seq_printf(m,
3851                         "%s%s%s%s", i ? " " : "",
3852                         i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3853                         i == trace_clock_id ? "]" : "");
3854         seq_putc(m, '\n');
3855
3856         return 0;
3857 }
3858
3859 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3860                                    size_t cnt, loff_t *fpos)
3861 {
3862         char buf[64];
3863         const char *clockstr;
3864         int i;
3865
3866         if (cnt >= sizeof(buf))
3867                 return -EINVAL;
3868
3869         if (copy_from_user(&buf, ubuf, cnt))
3870                 return -EFAULT;
3871
3872         buf[cnt] = 0;
3873
3874         clockstr = strstrip(buf);
3875
3876         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3877                 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3878                         break;
3879         }
3880         if (i == ARRAY_SIZE(trace_clocks))
3881                 return -EINVAL;
3882
3883         trace_clock_id = i;
3884
3885         mutex_lock(&trace_types_lock);
3886
3887         ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3888         if (max_tr.buffer)
3889                 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3890
3891         mutex_unlock(&trace_types_lock);
3892
3893         *fpos += cnt;
3894
3895         return cnt;
3896 }
3897
3898 static int tracing_clock_open(struct inode *inode, struct file *file)
3899 {
3900         if (tracing_disabled)
3901                 return -ENODEV;
3902         return single_open(file, tracing_clock_show, NULL);
3903 }
3904
3905 static const struct file_operations tracing_max_lat_fops = {
3906         .open           = tracing_open_generic,
3907         .read           = tracing_max_lat_read,
3908         .write          = tracing_max_lat_write,
3909         .llseek         = generic_file_llseek,
3910 };
3911
3912 static const struct file_operations tracing_ctrl_fops = {
3913         .open           = tracing_open_generic,
3914         .read           = tracing_ctrl_read,
3915         .write          = tracing_ctrl_write,
3916         .llseek         = generic_file_llseek,
3917 };
3918
3919 static const struct file_operations set_tracer_fops = {
3920         .open           = tracing_open_generic,
3921         .read           = tracing_set_trace_read,
3922         .write          = tracing_set_trace_write,
3923         .llseek         = generic_file_llseek,
3924 };
3925
3926 static const struct file_operations tracing_pipe_fops = {
3927         .open           = tracing_open_pipe,
3928         .poll           = tracing_poll_pipe,
3929         .read           = tracing_read_pipe,
3930         .splice_read    = tracing_splice_read_pipe,
3931         .release        = tracing_release_pipe,
3932         .llseek         = no_llseek,
3933 };
3934
3935 static const struct file_operations tracing_entries_fops = {
3936         .open           = tracing_open_generic,
3937         .read           = tracing_entries_read,
3938         .write          = tracing_entries_write,
3939         .llseek         = generic_file_llseek,
3940 };
3941
3942 static const struct file_operations tracing_total_entries_fops = {
3943         .open           = tracing_open_generic,
3944         .read           = tracing_total_entries_read,
3945         .llseek         = generic_file_llseek,
3946 };
3947
3948 static const struct file_operations tracing_free_buffer_fops = {
3949         .write          = tracing_free_buffer_write,
3950         .release        = tracing_free_buffer_release,
3951 };
3952
3953 static const struct file_operations tracing_mark_fops = {
3954         .open           = tracing_open_generic,
3955         .write          = tracing_mark_write,
3956         .llseek         = generic_file_llseek,
3957 };
3958
3959 static const struct file_operations trace_clock_fops = {
3960         .open           = tracing_clock_open,
3961         .read           = seq_read,
3962         .llseek         = seq_lseek,
3963         .release        = single_release,
3964         .write          = tracing_clock_write,
3965 };
3966
3967 struct ftrace_buffer_info {
3968         struct trace_array      *tr;
3969         void                    *spare;
3970         int                     cpu;
3971         unsigned int            read;
3972 };
3973
3974 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3975 {
3976         int cpu = (int)(long)inode->i_private;
3977         struct ftrace_buffer_info *info;
3978
3979         if (tracing_disabled)
3980                 return -ENODEV;
3981
3982         info = kzalloc(sizeof(*info), GFP_KERNEL);
3983         if (!info)
3984                 return -ENOMEM;
3985
3986         info->tr        = &global_trace;
3987         info->cpu       = cpu;
3988         info->spare     = NULL;
3989         /* Force reading ring buffer for first read */
3990         info->read      = (unsigned int)-1;
3991
3992         filp->private_data = info;
3993
3994         return nonseekable_open(inode, filp);
3995 }
3996
3997 static ssize_t
3998 tracing_buffers_read(struct file *filp, char __user *ubuf,
3999                      size_t count, loff_t *ppos)
4000 {
4001         struct ftrace_buffer_info *info = filp->private_data;
4002         ssize_t ret;
4003         size_t size;
4004
4005         if (!count)
4006                 return 0;
4007
4008         if (!info->spare)
4009                 info->spare = ring_buffer_alloc_read_page(info->tr->buffer, info->cpu);
4010         if (!info->spare)
4011                 return -ENOMEM;
4012
4013         /* Do we have previous read data to read? */
4014         if (info->read < PAGE_SIZE)
4015                 goto read;
4016
4017         trace_access_lock(info->cpu);
4018         ret = ring_buffer_read_page(info->tr->buffer,
4019                                     &info->spare,
4020                                     count,
4021                                     info->cpu, 0);
4022         trace_access_unlock(info->cpu);
4023         if (ret < 0)
4024                 return 0;
4025
4026         info->read = 0;
4027
4028 read:
4029         size = PAGE_SIZE - info->read;
4030         if (size > count)
4031                 size = count;
4032
4033         ret = copy_to_user(ubuf, info->spare + info->read, size);
4034         if (ret == size)
4035                 return -EFAULT;
4036         size -= ret;
4037
4038         *ppos += size;
4039         info->read += size;
4040
4041         return size;
4042 }
4043
4044 static int tracing_buffers_release(struct inode *inode, struct file *file)
4045 {
4046         struct ftrace_buffer_info *info = file->private_data;
4047
4048         if (info->spare)
4049                 ring_buffer_free_read_page(info->tr->buffer, info->spare);
4050         kfree(info);
4051
4052         return 0;
4053 }
4054
4055 struct buffer_ref {
4056         struct ring_buffer      *buffer;
4057         void                    *page;
4058         int                     ref;
4059 };
4060
4061 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4062                                     struct pipe_buffer *buf)
4063 {
4064         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4065
4066         if (--ref->ref)
4067                 return;
4068
4069         ring_buffer_free_read_page(ref->buffer, ref->page);
4070         kfree(ref);
4071         buf->private = 0;
4072 }
4073
4074 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
4075                                  struct pipe_buffer *buf)
4076 {
4077         return 1;
4078 }
4079
4080 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4081                                 struct pipe_buffer *buf)
4082 {
4083         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
4084
4085         ref->ref++;
4086 }
4087
4088 /* Pipe buffer operations for a buffer. */
4089 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4090         .can_merge              = 0,
4091         .map                    = generic_pipe_buf_map,
4092         .unmap                  = generic_pipe_buf_unmap,
4093         .confirm                = generic_pipe_buf_confirm,
4094         .release                = buffer_pipe_buf_release,
4095         .steal                  = buffer_pipe_buf_steal,
4096         .get                    = buffer_pipe_buf_get,
4097 };
4098
4099 /*
4100  * Callback from splice_to_pipe(), if we need to release some pages
4101  * at the end of the spd in case we error'ed out in filling the pipe.
4102  */
4103 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
4104 {
4105         struct buffer_ref *ref =
4106                 (struct buffer_ref *)spd->partial[i].private;
4107
4108         if (--ref->ref)
4109                 return;
4110
4111         ring_buffer_free_read_page(ref->buffer, ref->page);
4112         kfree(ref);
4113         spd->partial[i].private = 0;
4114 }
4115
4116 static ssize_t
4117 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
4118                             struct pipe_inode_info *pipe, size_t len,
4119                             unsigned int flags)
4120 {
4121         struct ftrace_buffer_info *info = file->private_data;
4122         struct partial_page partial_def[PIPE_DEF_BUFFERS];
4123         struct page *pages_def[PIPE_DEF_BUFFERS];
4124         struct splice_pipe_desc spd = {
4125                 .pages          = pages_def,
4126                 .partial        = partial_def,
4127                 .flags          = flags,
4128                 .ops            = &buffer_pipe_buf_ops,
4129                 .spd_release    = buffer_spd_release,
4130         };
4131         struct buffer_ref *ref;
4132         int entries, size, i;
4133         size_t ret;
4134
4135         if (splice_grow_spd(pipe, &spd))
4136                 return -ENOMEM;
4137
4138         if (*ppos & (PAGE_SIZE - 1)) {
4139                 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
4140                 ret = -EINVAL;
4141                 goto out;
4142         }
4143
4144         if (len & (PAGE_SIZE - 1)) {
4145                 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
4146                 if (len < PAGE_SIZE) {
4147                         ret = -EINVAL;
4148                         goto out;
4149                 }
4150                 len &= PAGE_MASK;
4151         }
4152
4153         trace_access_lock(info->cpu);
4154         entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4155
4156         for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
4157                 struct page *page;
4158                 int r;
4159
4160                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
4161                 if (!ref)
4162                         break;
4163
4164                 ref->ref = 1;
4165                 ref->buffer = info->tr->buffer;
4166                 ref->page = ring_buffer_alloc_read_page(ref->buffer, info->cpu);
4167                 if (!ref->page) {
4168                         kfree(ref);
4169                         break;
4170                 }
4171
4172                 r = ring_buffer_read_page(ref->buffer, &ref->page,
4173                                           len, info->cpu, 1);
4174                 if (r < 0) {
4175                         ring_buffer_free_read_page(ref->buffer, ref->page);
4176                         kfree(ref);
4177                         break;
4178                 }
4179
4180                 /*
4181                  * zero out any left over data, this is going to
4182                  * user land.
4183                  */
4184                 size = ring_buffer_page_len(ref->page);
4185                 if (size < PAGE_SIZE)
4186                         memset(ref->page + size, 0, PAGE_SIZE - size);
4187
4188                 page = virt_to_page(ref->page);
4189
4190                 spd.pages[i] = page;
4191                 spd.partial[i].len = PAGE_SIZE;
4192                 spd.partial[i].offset = 0;
4193                 spd.partial[i].private = (unsigned long)ref;
4194                 spd.nr_pages++;
4195                 *ppos += PAGE_SIZE;
4196
4197                 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
4198         }
4199
4200         trace_access_unlock(info->cpu);
4201         spd.nr_pages = i;
4202
4203         /* did we read anything? */
4204         if (!spd.nr_pages) {
4205                 if (flags & SPLICE_F_NONBLOCK)
4206                         ret = -EAGAIN;
4207                 else
4208                         ret = 0;
4209                 /* TODO: block */
4210                 goto out;
4211         }
4212
4213         ret = splice_to_pipe(pipe, &spd);
4214         splice_shrink_spd(pipe, &spd);
4215 out:
4216         return ret;
4217 }
4218
4219 static const struct file_operations tracing_buffers_fops = {
4220         .open           = tracing_buffers_open,
4221         .read           = tracing_buffers_read,
4222         .release        = tracing_buffers_release,
4223         .splice_read    = tracing_buffers_splice_read,
4224         .llseek         = no_llseek,
4225 };
4226
4227 static ssize_t
4228 tracing_stats_read(struct file *filp, char __user *ubuf,
4229                    size_t count, loff_t *ppos)
4230 {
4231         unsigned long cpu = (unsigned long)filp->private_data;
4232         struct trace_array *tr = &global_trace;
4233         struct trace_seq *s;
4234         unsigned long cnt;
4235         unsigned long long t;
4236         unsigned long usec_rem;
4237
4238         s = kmalloc(sizeof(*s), GFP_KERNEL);
4239         if (!s)
4240                 return -ENOMEM;
4241
4242         trace_seq_init(s);
4243
4244         cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
4245         trace_seq_printf(s, "entries: %ld\n", cnt);
4246
4247         cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
4248         trace_seq_printf(s, "overrun: %ld\n", cnt);
4249
4250         cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
4251         trace_seq_printf(s, "commit overrun: %ld\n", cnt);
4252
4253         cnt = ring_buffer_bytes_cpu(tr->buffer, cpu);
4254         trace_seq_printf(s, "bytes: %ld\n", cnt);
4255
4256         t = ns2usecs(ring_buffer_oldest_event_ts(tr->buffer, cpu));
4257         usec_rem = do_div(t, USEC_PER_SEC);
4258         trace_seq_printf(s, "oldest event ts: %5llu.%06lu\n", t, usec_rem);
4259
4260         t = ns2usecs(ring_buffer_time_stamp(tr->buffer, cpu));
4261         usec_rem = do_div(t, USEC_PER_SEC);
4262         trace_seq_printf(s, "now ts: %5llu.%06lu\n", t, usec_rem);
4263
4264         count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
4265
4266         kfree(s);
4267
4268         return count;
4269 }
4270
4271 static const struct file_operations tracing_stats_fops = {
4272         .open           = tracing_open_generic,
4273         .read           = tracing_stats_read,
4274         .llseek         = generic_file_llseek,
4275 };
4276
4277 #ifdef CONFIG_DYNAMIC_FTRACE
4278
4279 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
4280 {
4281         return 0;
4282 }
4283
4284 static ssize_t
4285 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
4286                   size_t cnt, loff_t *ppos)
4287 {
4288         static char ftrace_dyn_info_buffer[1024];
4289         static DEFINE_MUTEX(dyn_info_mutex);
4290         unsigned long *p = filp->private_data;
4291         char *buf = ftrace_dyn_info_buffer;
4292         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
4293         int r;
4294
4295         mutex_lock(&dyn_info_mutex);
4296         r = sprintf(buf, "%ld ", *p);
4297
4298         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
4299         buf[r++] = '\n';
4300
4301         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4302
4303         mutex_unlock(&dyn_info_mutex);
4304
4305         return r;
4306 }
4307
4308 static const struct file_operations tracing_dyn_info_fops = {
4309         .open           = tracing_open_generic,
4310         .read           = tracing_read_dyn_info,
4311         .llseek         = generic_file_llseek,
4312 };
4313 #endif
4314
4315 static struct dentry *d_tracer;
4316
4317 struct dentry *tracing_init_dentry(void)
4318 {
4319         static int once;
4320
4321         if (d_tracer)
4322                 return d_tracer;
4323
4324         if (!debugfs_initialized())
4325                 return NULL;
4326
4327         d_tracer = debugfs_create_dir("tracing", NULL);
4328
4329         if (!d_tracer && !once) {
4330                 once = 1;
4331                 pr_warning("Could not create debugfs directory 'tracing'\n");
4332                 return NULL;
4333         }
4334
4335         return d_tracer;
4336 }
4337
4338 static struct dentry *d_percpu;
4339
4340 struct dentry *tracing_dentry_percpu(void)
4341 {
4342         static int once;
4343         struct dentry *d_tracer;
4344
4345         if (d_percpu)
4346                 return d_percpu;
4347
4348         d_tracer = tracing_init_dentry();
4349
4350         if (!d_tracer)
4351                 return NULL;
4352
4353         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4354
4355         if (!d_percpu && !once) {
4356                 once = 1;
4357                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4358                 return NULL;
4359         }
4360
4361         return d_percpu;
4362 }
4363
4364 static void tracing_init_debugfs_percpu(long cpu)
4365 {
4366         struct dentry *d_percpu = tracing_dentry_percpu();
4367         struct dentry *d_cpu;
4368         char cpu_dir[30]; /* 30 characters should be more than enough */
4369
4370         snprintf(cpu_dir, 30, "cpu%ld", cpu);
4371         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4372         if (!d_cpu) {
4373                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4374                 return;
4375         }
4376
4377         /* per cpu trace_pipe */
4378         trace_create_file("trace_pipe", 0444, d_cpu,
4379                         (void *) cpu, &tracing_pipe_fops);
4380
4381         /* per cpu trace */
4382         trace_create_file("trace", 0644, d_cpu,
4383                         (void *) cpu, &tracing_fops);
4384
4385         trace_create_file("trace_pipe_raw", 0444, d_cpu,
4386                         (void *) cpu, &tracing_buffers_fops);
4387
4388         trace_create_file("stats", 0444, d_cpu,
4389                         (void *) cpu, &tracing_stats_fops);
4390 }
4391
4392 #ifdef CONFIG_FTRACE_SELFTEST
4393 /* Let selftest have access to static functions in this file */
4394 #include "trace_selftest.c"
4395 #endif
4396
4397 struct trace_option_dentry {
4398         struct tracer_opt               *opt;
4399         struct tracer_flags             *flags;
4400         struct dentry                   *entry;
4401 };
4402
4403 static ssize_t
4404 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4405                         loff_t *ppos)
4406 {
4407         struct trace_option_dentry *topt = filp->private_data;
4408         char *buf;
4409
4410         if (topt->flags->val & topt->opt->bit)
4411                 buf = "1\n";
4412         else
4413                 buf = "0\n";
4414
4415         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4416 }
4417
4418 static ssize_t
4419 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4420                          loff_t *ppos)
4421 {
4422         struct trace_option_dentry *topt = filp->private_data;
4423         unsigned long val;
4424         int ret;
4425
4426         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4427         if (ret)
4428                 return ret;
4429
4430         if (val != 0 && val != 1)
4431                 return -EINVAL;
4432
4433         if (!!(topt->flags->val & topt->opt->bit) != val) {
4434                 mutex_lock(&trace_types_lock);
4435                 ret = __set_tracer_option(current_trace, topt->flags,
4436                                           topt->opt, !val);
4437                 mutex_unlock(&trace_types_lock);
4438                 if (ret)
4439                         return ret;
4440         }
4441
4442         *ppos += cnt;
4443
4444         return cnt;
4445 }
4446
4447
4448 static const struct file_operations trace_options_fops = {
4449         .open = tracing_open_generic,
4450         .read = trace_options_read,
4451         .write = trace_options_write,
4452         .llseek = generic_file_llseek,
4453 };
4454
4455 static ssize_t
4456 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4457                         loff_t *ppos)
4458 {
4459         long index = (long)filp->private_data;
4460         char *buf;
4461
4462         if (trace_flags & (1 << index))
4463                 buf = "1\n";
4464         else
4465                 buf = "0\n";
4466
4467         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4468 }
4469
4470 static ssize_t
4471 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4472                          loff_t *ppos)
4473 {
4474         long index = (long)filp->private_data;
4475         unsigned long val;
4476         int ret;
4477
4478         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4479         if (ret)
4480                 return ret;
4481
4482         if (val != 0 && val != 1)
4483                 return -EINVAL;
4484         set_tracer_flags(1 << index, val);
4485
4486         *ppos += cnt;
4487
4488         return cnt;
4489 }
4490
4491 static const struct file_operations trace_options_core_fops = {
4492         .open = tracing_open_generic,
4493         .read = trace_options_core_read,
4494         .write = trace_options_core_write,
4495         .llseek = generic_file_llseek,
4496 };
4497
4498 struct dentry *trace_create_file(const char *name,
4499                                  umode_t mode,
4500                                  struct dentry *parent,
4501                                  void *data,
4502                                  const struct file_operations *fops)
4503 {
4504         struct dentry *ret;
4505
4506         ret = debugfs_create_file(name, mode, parent, data, fops);
4507         if (!ret)
4508                 pr_warning("Could not create debugfs '%s' entry\n", name);
4509
4510         return ret;
4511 }
4512
4513
4514 static struct dentry *trace_options_init_dentry(void)
4515 {
4516         struct dentry *d_tracer;
4517         static struct dentry *t_options;
4518
4519         if (t_options)
4520                 return t_options;
4521
4522         d_tracer = tracing_init_dentry();
4523         if (!d_tracer)
4524                 return NULL;
4525
4526         t_options = debugfs_create_dir("options", d_tracer);
4527         if (!t_options) {
4528                 pr_warning("Could not create debugfs directory 'options'\n");
4529                 return NULL;
4530         }
4531
4532         return t_options;
4533 }
4534
4535 static void
4536 create_trace_option_file(struct trace_option_dentry *topt,
4537                          struct tracer_flags *flags,
4538                          struct tracer_opt *opt)
4539 {
4540         struct dentry *t_options;
4541
4542         t_options = trace_options_init_dentry();
4543         if (!t_options)
4544                 return;
4545
4546         topt->flags = flags;
4547         topt->opt = opt;
4548
4549         topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4550                                     &trace_options_fops);
4551
4552 }
4553
4554 static struct trace_option_dentry *
4555 create_trace_option_files(struct tracer *tracer)
4556 {
4557         struct trace_option_dentry *topts;
4558         struct tracer_flags *flags;
4559         struct tracer_opt *opts;
4560         int cnt;
4561
4562         if (!tracer)
4563                 return NULL;
4564
4565         flags = tracer->flags;
4566
4567         if (!flags || !flags->opts)
4568                 return NULL;
4569
4570         opts = flags->opts;
4571
4572         for (cnt = 0; opts[cnt].name; cnt++)
4573                 ;
4574
4575         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4576         if (!topts)
4577                 return NULL;
4578
4579         for (cnt = 0; opts[cnt].name; cnt++)
4580                 create_trace_option_file(&topts[cnt], flags,
4581                                          &opts[cnt]);
4582
4583         return topts;
4584 }
4585
4586 static void
4587 destroy_trace_option_files(struct trace_option_dentry *topts)
4588 {
4589         int cnt;
4590
4591         if (!topts)
4592                 return;
4593
4594         for (cnt = 0; topts[cnt].opt; cnt++) {
4595                 if (topts[cnt].entry)
4596                         debugfs_remove(topts[cnt].entry);
4597         }
4598
4599         kfree(topts);
4600 }
4601
4602 static struct dentry *
4603 create_trace_option_core_file(const char *option, long index)
4604 {
4605         struct dentry *t_options;
4606
4607         t_options = trace_options_init_dentry();
4608         if (!t_options)
4609                 return NULL;
4610
4611         return trace_create_file(option, 0644, t_options, (void *)index,
4612                                     &trace_options_core_fops);
4613 }
4614
4615 static __init void create_trace_options_dir(void)
4616 {
4617         struct dentry *t_options;
4618         int i;
4619
4620         t_options = trace_options_init_dentry();
4621         if (!t_options)
4622                 return;
4623
4624         for (i = 0; trace_options[i]; i++)
4625                 create_trace_option_core_file(trace_options[i], i);
4626 }
4627
4628 static ssize_t
4629 rb_simple_read(struct file *filp, char __user *ubuf,
4630                size_t cnt, loff_t *ppos)
4631 {
4632         struct trace_array *tr = filp->private_data;
4633         struct ring_buffer *buffer = tr->buffer;
4634         char buf[64];
4635         int r;
4636
4637         if (buffer)
4638                 r = ring_buffer_record_is_on(buffer);
4639         else
4640                 r = 0;
4641
4642         r = sprintf(buf, "%d\n", r);
4643
4644         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
4645 }
4646
4647 static ssize_t
4648 rb_simple_write(struct file *filp, const char __user *ubuf,
4649                 size_t cnt, loff_t *ppos)
4650 {
4651         struct trace_array *tr = filp->private_data;
4652         struct ring_buffer *buffer = tr->buffer;
4653         unsigned long val;
4654         int ret;
4655
4656         ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
4657         if (ret)
4658                 return ret;
4659
4660         if (buffer) {
4661                 if (val)
4662                         ring_buffer_record_on(buffer);
4663                 else
4664                         ring_buffer_record_off(buffer);
4665         }
4666
4667         (*ppos)++;
4668
4669         return cnt;
4670 }
4671
4672 static const struct file_operations rb_simple_fops = {
4673         .open           = tracing_open_generic,
4674         .read           = rb_simple_read,
4675         .write          = rb_simple_write,
4676         .llseek         = default_llseek,
4677 };
4678
4679 static __init int tracer_init_debugfs(void)
4680 {
4681         struct dentry *d_tracer;
4682         int cpu;
4683
4684         trace_access_lock_init();
4685
4686         d_tracer = tracing_init_dentry();
4687
4688         trace_create_file("tracing_enabled", 0644, d_tracer,
4689                         &global_trace, &tracing_ctrl_fops);
4690
4691         trace_create_file("trace_options", 0644, d_tracer,
4692                         NULL, &tracing_iter_fops);
4693
4694         trace_create_file("tracing_cpumask", 0644, d_tracer,
4695                         NULL, &tracing_cpumask_fops);
4696
4697         trace_create_file("trace", 0644, d_tracer,
4698                         (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4699
4700         trace_create_file("available_tracers", 0444, d_tracer,
4701                         &global_trace, &show_traces_fops);
4702
4703         trace_create_file("current_tracer", 0644, d_tracer,
4704                         &global_trace, &set_tracer_fops);
4705
4706 #ifdef CONFIG_TRACER_MAX_TRACE
4707         trace_create_file("tracing_max_latency", 0644, d_tracer,
4708                         &tracing_max_latency, &tracing_max_lat_fops);
4709 #endif
4710
4711         trace_create_file("tracing_thresh", 0644, d_tracer,
4712                         &tracing_thresh, &tracing_max_lat_fops);
4713
4714         trace_create_file("README", 0444, d_tracer,
4715                         NULL, &tracing_readme_fops);
4716
4717         trace_create_file("trace_pipe", 0444, d_tracer,
4718                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4719
4720         trace_create_file("buffer_size_kb", 0644, d_tracer,
4721                         &global_trace, &tracing_entries_fops);
4722
4723         trace_create_file("buffer_total_size_kb", 0444, d_tracer,
4724                         &global_trace, &tracing_total_entries_fops);
4725
4726         trace_create_file("free_buffer", 0644, d_tracer,
4727                         &global_trace, &tracing_free_buffer_fops);
4728
4729         trace_create_file("trace_marker", 0220, d_tracer,
4730                         NULL, &tracing_mark_fops);
4731
4732         trace_create_file("saved_cmdlines", 0444, d_tracer,
4733                         NULL, &tracing_saved_cmdlines_fops);
4734
4735         trace_create_file("trace_clock", 0644, d_tracer, NULL,
4736                           &trace_clock_fops);
4737
4738         trace_create_file("tracing_on", 0644, d_tracer,
4739                             &global_trace, &rb_simple_fops);
4740
4741 #ifdef CONFIG_DYNAMIC_FTRACE
4742         trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4743                         &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4744 #endif
4745
4746         create_trace_options_dir();
4747
4748         for_each_tracing_cpu(cpu)
4749                 tracing_init_debugfs_percpu(cpu);
4750
4751         return 0;
4752 }
4753
4754 static int trace_panic_handler(struct notifier_block *this,
4755                                unsigned long event, void *unused)
4756 {
4757         if (ftrace_dump_on_oops)
4758                 ftrace_dump(ftrace_dump_on_oops);
4759         return NOTIFY_OK;
4760 }
4761
4762 static struct notifier_block trace_panic_notifier = {
4763         .notifier_call  = trace_panic_handler,
4764         .next           = NULL,
4765         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
4766 };
4767
4768 static int trace_die_handler(struct notifier_block *self,
4769                              unsigned long val,
4770                              void *data)
4771 {
4772         switch (val) {
4773         case DIE_OOPS:
4774                 if (ftrace_dump_on_oops)
4775                         ftrace_dump(ftrace_dump_on_oops);
4776                 break;
4777         default:
4778                 break;
4779         }
4780         return NOTIFY_OK;
4781 }
4782
4783 static struct notifier_block trace_die_notifier = {
4784         .notifier_call = trace_die_handler,
4785         .priority = 200
4786 };
4787
4788 /*
4789  * printk is set to max of 1024, we really don't need it that big.
4790  * Nothing should be printing 1000 characters anyway.
4791  */
4792 #define TRACE_MAX_PRINT         1000
4793
4794 /*
4795  * Define here KERN_TRACE so that we have one place to modify
4796  * it if we decide to change what log level the ftrace dump
4797  * should be at.
4798  */
4799 #define KERN_TRACE              KERN_EMERG
4800
4801 void
4802 trace_printk_seq(struct trace_seq *s)
4803 {
4804         /* Probably should print a warning here. */
4805         if (s->len >= 1000)
4806                 s->len = 1000;
4807
4808         /* should be zero ended, but we are paranoid. */
4809         s->buffer[s->len] = 0;
4810
4811         printk(KERN_TRACE "%s", s->buffer);
4812
4813         trace_seq_init(s);
4814 }
4815
4816 void trace_init_global_iter(struct trace_iterator *iter)
4817 {
4818         iter->tr = &global_trace;
4819         iter->trace = current_trace;
4820         iter->cpu_file = TRACE_PIPE_ALL_CPU;
4821 }
4822
4823 static void
4824 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4825 {
4826         static arch_spinlock_t ftrace_dump_lock =
4827                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4828         /* use static because iter can be a bit big for the stack */
4829         static struct trace_iterator iter;
4830         unsigned int old_userobj;
4831         static int dump_ran;
4832         unsigned long flags;
4833         int cnt = 0, cpu;
4834
4835         /* only one dump */
4836         local_irq_save(flags);
4837         arch_spin_lock(&ftrace_dump_lock);
4838         if (dump_ran)
4839                 goto out;
4840
4841         dump_ran = 1;
4842
4843         tracing_off();
4844
4845         /* Did function tracer already get disabled? */
4846         if (ftrace_is_dead()) {
4847                 printk("# WARNING: FUNCTION TRACING IS CORRUPTED\n");
4848                 printk("#          MAY BE MISSING FUNCTION EVENTS\n");
4849         }
4850
4851         if (disable_tracing)
4852                 ftrace_kill();
4853
4854         trace_init_global_iter(&iter);
4855
4856         for_each_tracing_cpu(cpu) {
4857                 atomic_inc(&iter.tr->data[cpu]->disabled);
4858         }
4859
4860         old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4861
4862         /* don't look at user memory in panic mode */
4863         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4864
4865         /* Simulate the iterator */
4866         iter.tr = &global_trace;
4867         iter.trace = current_trace;
4868
4869         switch (oops_dump_mode) {
4870         case DUMP_ALL:
4871                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4872                 break;
4873         case DUMP_ORIG:
4874                 iter.cpu_file = raw_smp_processor_id();
4875                 break;
4876         case DUMP_NONE:
4877                 goto out_enable;
4878         default:
4879                 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4880                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4881         }
4882
4883         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4884
4885         /*
4886          * We need to stop all tracing on all CPUS to read the
4887          * the next buffer. This is a bit expensive, but is
4888          * not done often. We fill all what we can read,
4889          * and then release the locks again.
4890          */
4891
4892         while (!trace_empty(&iter)) {
4893
4894                 if (!cnt)
4895                         printk(KERN_TRACE "---------------------------------\n");
4896
4897                 cnt++;
4898
4899                 /* reset all but tr, trace, and overruns */
4900                 memset(&iter.seq, 0,
4901                        sizeof(struct trace_iterator) -
4902                        offsetof(struct trace_iterator, seq));
4903                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4904                 iter.pos = -1;
4905
4906                 if (trace_find_next_entry_inc(&iter) != NULL) {
4907                         int ret;
4908
4909                         ret = print_trace_line(&iter);
4910                         if (ret != TRACE_TYPE_NO_CONSUME)
4911                                 trace_consume(&iter);
4912                 }
4913                 touch_nmi_watchdog();
4914
4915                 trace_printk_seq(&iter.seq);
4916         }
4917
4918         if (!cnt)
4919                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4920         else
4921                 printk(KERN_TRACE "---------------------------------\n");
4922
4923  out_enable:
4924         /* Re-enable tracing if requested */
4925         if (!disable_tracing) {
4926                 trace_flags |= old_userobj;
4927
4928                 for_each_tracing_cpu(cpu) {
4929                         atomic_dec(&iter.tr->data[cpu]->disabled);
4930                 }
4931                 tracing_on();
4932         }
4933
4934  out:
4935         arch_spin_unlock(&ftrace_dump_lock);
4936         local_irq_restore(flags);
4937 }
4938
4939 /* By default: disable tracing after the dump */
4940 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4941 {
4942         __ftrace_dump(true, oops_dump_mode);
4943 }
4944 EXPORT_SYMBOL_GPL(ftrace_dump);
4945
4946 __init static int tracer_alloc_buffers(void)
4947 {
4948         int ring_buf_size;
4949         enum ring_buffer_flags rb_flags;
4950         int i;
4951         int ret = -ENOMEM;
4952
4953
4954         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4955                 goto out;
4956
4957         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4958                 goto out_free_buffer_mask;
4959
4960         /* To save memory, keep the ring buffer size to its minimum */
4961         if (ring_buffer_expanded)
4962                 ring_buf_size = trace_buf_size;
4963         else
4964                 ring_buf_size = 1;
4965
4966         rb_flags = trace_flags & TRACE_ITER_OVERWRITE ? RB_FL_OVERWRITE : 0;
4967
4968         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4969         cpumask_copy(tracing_cpumask, cpu_all_mask);
4970
4971         /* TODO: make the number of buffers hot pluggable with CPUS */
4972         global_trace.buffer = ring_buffer_alloc(ring_buf_size, rb_flags);
4973         if (!global_trace.buffer) {
4974                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4975                 WARN_ON(1);
4976                 goto out_free_cpumask;
4977         }
4978         global_trace.entries = ring_buffer_size(global_trace.buffer);
4979         if (global_trace.buffer_disabled)
4980                 tracing_off();
4981
4982
4983 #ifdef CONFIG_TRACER_MAX_TRACE
4984         max_tr.buffer = ring_buffer_alloc(1, rb_flags);
4985         if (!max_tr.buffer) {
4986                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4987                 WARN_ON(1);
4988                 ring_buffer_free(global_trace.buffer);
4989                 goto out_free_cpumask;
4990         }
4991         max_tr.entries = 1;
4992 #endif
4993
4994         /* Allocate the first page for all buffers */
4995         for_each_tracing_cpu(i) {
4996                 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4997                 max_tr.data[i] = &per_cpu(max_tr_data, i);
4998         }
4999
5000         trace_init_cmdlines();
5001
5002         register_tracer(&nop_trace);
5003         current_trace = &nop_trace;
5004         /* All seems OK, enable tracing */
5005         tracing_disabled = 0;
5006
5007         atomic_notifier_chain_register(&panic_notifier_list,
5008                                        &trace_panic_notifier);
5009
5010         register_die_notifier(&trace_die_notifier);
5011
5012         return 0;
5013
5014 out_free_cpumask:
5015         free_cpumask_var(tracing_cpumask);
5016 out_free_buffer_mask:
5017         free_cpumask_var(tracing_buffer_mask);
5018 out:
5019         return ret;
5020 }
5021
5022 __init static int clear_boot_tracer(void)
5023 {
5024         /*
5025          * The default tracer at boot buffer is an init section.
5026          * This function is called in lateinit. If we did not
5027          * find the boot tracer, then clear it out, to prevent
5028          * later registration from accessing the buffer that is
5029          * about to be freed.
5030          */
5031         if (!default_bootup_tracer)
5032                 return 0;
5033
5034         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
5035                default_bootup_tracer);
5036         default_bootup_tracer = NULL;
5037
5038         return 0;
5039 }
5040
5041 early_initcall(tracer_alloc_buffers);
5042 fs_initcall(tracer_init_debugfs);
5043 late_initcall(clear_boot_tracer);