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