78a49e67f7db88680b0bca7cfac54132fc7253cd
[linux-2.6.git] / kernel / trace / trace.c
1 /*
2  * ring buffer based function tracer
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally taken from the RT patch by:
8  *    Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code from the latency_tracer, that is:
11  *  Copyright (C) 2004-2006 Ingo Molnar
12  *  Copyright (C) 2004 William Lee Irwin III
13  */
14 #include <linux/ring_buffer.h>
15 #include <generated/utsrelease.h>
16 #include <linux/stacktrace.h>
17 #include <linux/writeback.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/smp_lock.h>
21 #include <linux/notifier.h>
22 #include <linux/irqflags.h>
23 #include <linux/debugfs.h>
24 #include <linux/pagemap.h>
25 #include <linux/hardirq.h>
26 #include <linux/linkage.h>
27 #include <linux/uaccess.h>
28 #include <linux/kprobes.h>
29 #include <linux/ftrace.h>
30 #include <linux/module.h>
31 #include <linux/percpu.h>
32 #include <linux/splice.h>
33 #include <linux/kdebug.h>
34 #include <linux/string.h>
35 #include <linux/rwsem.h>
36 #include <linux/slab.h>
37 #include <linux/ctype.h>
38 #include <linux/init.h>
39 #include <linux/poll.h>
40 #include <linux/fs.h>
41
42 #include "trace.h"
43 #include "trace_output.h"
44
45 #define TRACE_BUFFER_FLAGS      (RB_FL_OVERWRITE)
46
47 /*
48  * On boot up, the ring buffer is set to the minimum size, so that
49  * we do not waste memory on systems that are not using tracing.
50  */
51 int ring_buffer_expanded;
52
53 /*
54  * We need to change this state when a selftest is running.
55  * A selftest will lurk into the ring-buffer to count the
56  * entries inserted during the selftest although some concurrent
57  * insertions into the ring-buffer such as trace_printk could occurred
58  * at the same time, giving false positive or negative results.
59  */
60 static bool __read_mostly tracing_selftest_running;
61
62 /*
63  * If a tracer is running, we do not want to run SELFTEST.
64  */
65 bool __read_mostly tracing_selftest_disabled;
66
67 /* For tracers that don't implement custom flags */
68 static struct tracer_opt dummy_tracer_opt[] = {
69         { }
70 };
71
72 static struct tracer_flags dummy_tracer_flags = {
73         .val = 0,
74         .opts = dummy_tracer_opt
75 };
76
77 static int dummy_set_flag(u32 old_flags, u32 bit, int set)
78 {
79         return 0;
80 }
81
82 /*
83  * Kill all tracing for good (never come back).
84  * It is initialized to 1 but will turn to zero if the initialization
85  * of the tracer is successful. But that is the only place that sets
86  * this back to zero.
87  */
88 static int tracing_disabled = 1;
89
90 DEFINE_PER_CPU(int, ftrace_cpu_disabled);
91
92 static inline void ftrace_disable_cpu(void)
93 {
94         preempt_disable();
95         __this_cpu_inc(ftrace_cpu_disabled);
96 }
97
98 static inline void ftrace_enable_cpu(void)
99 {
100         __this_cpu_dec(ftrace_cpu_disabled);
101         preempt_enable();
102 }
103
104 static cpumask_var_t __read_mostly      tracing_buffer_mask;
105
106 #define for_each_tracing_cpu(cpu)       \
107         for_each_cpu(cpu, tracing_buffer_mask)
108
109 /*
110  * ftrace_dump_on_oops - variable to dump ftrace buffer on oops
111  *
112  * If there is an oops (or kernel panic) and the ftrace_dump_on_oops
113  * is set, then ftrace_dump is called. This will output the contents
114  * of the ftrace buffers to the console.  This is very useful for
115  * capturing traces that lead to crashes and outputing it to a
116  * serial console.
117  *
118  * It is default off, but you can enable it with either specifying
119  * "ftrace_dump_on_oops" in the kernel command line, or setting
120  * /proc/sys/kernel/ftrace_dump_on_oops
121  * Set 1 if you want to dump buffers of all CPUs
122  * Set 2 if you want to dump the buffer of the CPU that triggered oops
123  */
124
125 enum ftrace_dump_mode ftrace_dump_on_oops;
126
127 static int tracing_set_tracer(const char *buf);
128
129 #define MAX_TRACER_SIZE         100
130 static char bootup_tracer_buf[MAX_TRACER_SIZE] __initdata;
131 static char *default_bootup_tracer;
132
133 static int __init set_cmdline_ftrace(char *str)
134 {
135         strncpy(bootup_tracer_buf, str, MAX_TRACER_SIZE);
136         default_bootup_tracer = bootup_tracer_buf;
137         /* We are using ftrace early, expand it */
138         ring_buffer_expanded = 1;
139         return 1;
140 }
141 __setup("ftrace=", set_cmdline_ftrace);
142
143 static int __init set_ftrace_dump_on_oops(char *str)
144 {
145         if (*str++ != '=' || !*str) {
146                 ftrace_dump_on_oops = DUMP_ALL;
147                 return 1;
148         }
149
150         if (!strcmp("orig_cpu", str)) {
151                 ftrace_dump_on_oops = DUMP_ORIG;
152                 return 1;
153         }
154
155         return 0;
156 }
157 __setup("ftrace_dump_on_oops", set_ftrace_dump_on_oops);
158
159 unsigned long long ns2usecs(cycle_t nsec)
160 {
161         nsec += 500;
162         do_div(nsec, 1000);
163         return nsec;
164 }
165
166 /*
167  * The global_trace is the descriptor that holds the tracing
168  * buffers for the live tracing. For each CPU, it contains
169  * a link list of pages that will store trace entries. The
170  * page descriptor of the pages in the memory is used to hold
171  * the link list by linking the lru item in the page descriptor
172  * to each of the pages in the buffer per CPU.
173  *
174  * For each active CPU there is a data field that holds the
175  * pages for the buffer for that CPU. Each CPU has the same number
176  * of pages allocated for its buffer.
177  */
178 static struct trace_array       global_trace;
179
180 static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
181
182 int filter_current_check_discard(struct ring_buffer *buffer,
183                                  struct ftrace_event_call *call, void *rec,
184                                  struct ring_buffer_event *event)
185 {
186         return filter_check_discard(call, rec, buffer, event);
187 }
188 EXPORT_SYMBOL_GPL(filter_current_check_discard);
189
190 cycle_t ftrace_now(int cpu)
191 {
192         u64 ts;
193
194         /* Early boot up does not have a buffer yet */
195         if (!global_trace.buffer)
196                 return trace_clock_local();
197
198         ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
199         ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
200
201         return ts;
202 }
203
204 /*
205  * The max_tr is used to snapshot the global_trace when a maximum
206  * latency is reached. Some tracers will use this to store a maximum
207  * trace while it continues examining live traces.
208  *
209  * The buffers for the max_tr are set up the same as the global_trace.
210  * When a snapshot is taken, the link list of the max_tr is swapped
211  * with the link list of the global_trace and the buffers are reset for
212  * the global_trace so the tracing can continue.
213  */
214 static struct trace_array       max_tr;
215
216 static DEFINE_PER_CPU(struct trace_array_cpu, max_tr_data);
217
218 /* tracer_enabled is used to toggle activation of a tracer */
219 static int                      tracer_enabled = 1;
220
221 /**
222  * tracing_is_enabled - return tracer_enabled status
223  *
224  * This function is used by other tracers to know the status
225  * of the tracer_enabled flag.  Tracers may use this function
226  * to know if it should enable their features when starting
227  * up. See irqsoff tracer for an example (start_irqsoff_tracer).
228  */
229 int tracing_is_enabled(void)
230 {
231         return tracer_enabled;
232 }
233
234 /*
235  * trace_buf_size is the size in bytes that is allocated
236  * for a buffer. Note, the number of bytes is always rounded
237  * to page size.
238  *
239  * This number is purposely set to a low number of 16384.
240  * If the dump on oops happens, it will be much appreciated
241  * to not have to wait for all that output. Anyway this can be
242  * boot time and run time configurable.
243  */
244 #define TRACE_BUF_SIZE_DEFAULT  1441792UL /* 16384 * 88 (sizeof(entry)) */
245
246 static unsigned long            trace_buf_size = TRACE_BUF_SIZE_DEFAULT;
247
248 /* trace_types holds a link list of available tracers. */
249 static struct tracer            *trace_types __read_mostly;
250
251 /* current_trace points to the tracer that is currently active */
252 static struct tracer            *current_trace __read_mostly;
253
254 /*
255  * trace_types_lock is used to protect the trace_types list.
256  */
257 static DEFINE_MUTEX(trace_types_lock);
258
259 /*
260  * serialize the access of the ring buffer
261  *
262  * ring buffer serializes readers, but it is low level protection.
263  * The validity of the events (which returns by ring_buffer_peek() ..etc)
264  * are not protected by ring buffer.
265  *
266  * The content of events may become garbage if we allow other process consumes
267  * these events concurrently:
268  *   A) the page of the consumed events may become a normal page
269  *      (not reader page) in ring buffer, and this page will be rewrited
270  *      by events producer.
271  *   B) The page of the consumed events may become a page for splice_read,
272  *      and this page will be returned to system.
273  *
274  * These primitives allow multi process access to different cpu ring buffer
275  * concurrently.
276  *
277  * These primitives don't distinguish read-only and read-consume access.
278  * Multi read-only access are also serialized.
279  */
280
281 #ifdef CONFIG_SMP
282 static DECLARE_RWSEM(all_cpu_access_lock);
283 static DEFINE_PER_CPU(struct mutex, cpu_access_lock);
284
285 static inline void trace_access_lock(int cpu)
286 {
287         if (cpu == TRACE_PIPE_ALL_CPU) {
288                 /* gain it for accessing the whole ring buffer. */
289                 down_write(&all_cpu_access_lock);
290         } else {
291                 /* gain it for accessing a cpu ring buffer. */
292
293                 /* Firstly block other trace_access_lock(TRACE_PIPE_ALL_CPU). */
294                 down_read(&all_cpu_access_lock);
295
296                 /* Secondly block other access to this @cpu ring buffer. */
297                 mutex_lock(&per_cpu(cpu_access_lock, cpu));
298         }
299 }
300
301 static inline void trace_access_unlock(int cpu)
302 {
303         if (cpu == TRACE_PIPE_ALL_CPU) {
304                 up_write(&all_cpu_access_lock);
305         } else {
306                 mutex_unlock(&per_cpu(cpu_access_lock, cpu));
307                 up_read(&all_cpu_access_lock);
308         }
309 }
310
311 static inline void trace_access_lock_init(void)
312 {
313         int cpu;
314
315         for_each_possible_cpu(cpu)
316                 mutex_init(&per_cpu(cpu_access_lock, cpu));
317 }
318
319 #else
320
321 static DEFINE_MUTEX(access_lock);
322
323 static inline void trace_access_lock(int cpu)
324 {
325         (void)cpu;
326         mutex_lock(&access_lock);
327 }
328
329 static inline void trace_access_unlock(int cpu)
330 {
331         (void)cpu;
332         mutex_unlock(&access_lock);
333 }
334
335 static inline void trace_access_lock_init(void)
336 {
337 }
338
339 #endif
340
341 /* trace_wait is a waitqueue for tasks blocked on trace_poll */
342 static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
343
344 /* trace_flags holds trace_options default values */
345 unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
346         TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
347         TRACE_ITER_GRAPH_TIME;
348
349 static int trace_stop_count;
350 static DEFINE_SPINLOCK(tracing_start_lock);
351
352 /**
353  * trace_wake_up - wake up tasks waiting for trace input
354  *
355  * Simply wakes up any task that is blocked on the trace_wait
356  * queue. These is used with trace_poll for tasks polling the trace.
357  */
358 void trace_wake_up(void)
359 {
360         int cpu;
361
362         if (trace_flags & TRACE_ITER_BLOCK)
363                 return;
364         /*
365          * The runqueue_is_locked() can fail, but this is the best we
366          * have for now:
367          */
368         cpu = get_cpu();
369         if (!runqueue_is_locked(cpu))
370                 wake_up(&trace_wait);
371         put_cpu();
372 }
373
374 static int __init set_buf_size(char *str)
375 {
376         unsigned long buf_size;
377
378         if (!str)
379                 return 0;
380         buf_size = memparse(str, &str);
381         /* nr_entries can not be zero */
382         if (buf_size == 0)
383                 return 0;
384         trace_buf_size = buf_size;
385         return 1;
386 }
387 __setup("trace_buf_size=", set_buf_size);
388
389 static int __init set_tracing_thresh(char *str)
390 {
391         unsigned long threshhold;
392         int ret;
393
394         if (!str)
395                 return 0;
396         ret = strict_strtoul(str, 0, &threshhold);
397         if (ret < 0)
398                 return 0;
399         tracing_thresh = threshhold * 1000;
400         return 1;
401 }
402 __setup("tracing_thresh=", set_tracing_thresh);
403
404 unsigned long nsecs_to_usecs(unsigned long nsecs)
405 {
406         return nsecs / 1000;
407 }
408
409 /* These must match the bit postions in trace_iterator_flags */
410 static const char *trace_options[] = {
411         "print-parent",
412         "sym-offset",
413         "sym-addr",
414         "verbose",
415         "raw",
416         "hex",
417         "bin",
418         "block",
419         "stacktrace",
420         "trace_printk",
421         "ftrace_preempt",
422         "branch",
423         "annotate",
424         "userstacktrace",
425         "sym-userobj",
426         "printk-msg-only",
427         "context-info",
428         "latency-format",
429         "sleep-time",
430         "graph-time",
431         NULL
432 };
433
434 static struct {
435         u64 (*func)(void);
436         const char *name;
437 } trace_clocks[] = {
438         { trace_clock_local,    "local" },
439         { trace_clock_global,   "global" },
440 };
441
442 int trace_clock_id;
443
444 /*
445  * trace_parser_get_init - gets the buffer for trace parser
446  */
447 int trace_parser_get_init(struct trace_parser *parser, int size)
448 {
449         memset(parser, 0, sizeof(*parser));
450
451         parser->buffer = kmalloc(size, GFP_KERNEL);
452         if (!parser->buffer)
453                 return 1;
454
455         parser->size = size;
456         return 0;
457 }
458
459 /*
460  * trace_parser_put - frees the buffer for trace parser
461  */
462 void trace_parser_put(struct trace_parser *parser)
463 {
464         kfree(parser->buffer);
465 }
466
467 /*
468  * trace_get_user - reads the user input string separated by  space
469  * (matched by isspace(ch))
470  *
471  * For each string found the 'struct trace_parser' is updated,
472  * and the function returns.
473  *
474  * Returns number of bytes read.
475  *
476  * See kernel/trace/trace.h for 'struct trace_parser' details.
477  */
478 int trace_get_user(struct trace_parser *parser, const char __user *ubuf,
479         size_t cnt, loff_t *ppos)
480 {
481         char ch;
482         size_t read = 0;
483         ssize_t ret;
484
485         if (!*ppos)
486                 trace_parser_clear(parser);
487
488         ret = get_user(ch, ubuf++);
489         if (ret)
490                 goto out;
491
492         read++;
493         cnt--;
494
495         /*
496          * The parser is not finished with the last write,
497          * continue reading the user input without skipping spaces.
498          */
499         if (!parser->cont) {
500                 /* skip white space */
501                 while (cnt && isspace(ch)) {
502                         ret = get_user(ch, ubuf++);
503                         if (ret)
504                                 goto out;
505                         read++;
506                         cnt--;
507                 }
508
509                 /* only spaces were written */
510                 if (isspace(ch)) {
511                         *ppos += read;
512                         ret = read;
513                         goto out;
514                 }
515
516                 parser->idx = 0;
517         }
518
519         /* read the non-space input */
520         while (cnt && !isspace(ch)) {
521                 if (parser->idx < parser->size - 1)
522                         parser->buffer[parser->idx++] = ch;
523                 else {
524                         ret = -EINVAL;
525                         goto out;
526                 }
527                 ret = get_user(ch, ubuf++);
528                 if (ret)
529                         goto out;
530                 read++;
531                 cnt--;
532         }
533
534         /* We either got finished input or we have to wait for another call. */
535         if (isspace(ch)) {
536                 parser->buffer[parser->idx] = 0;
537                 parser->cont = false;
538         } else {
539                 parser->cont = true;
540                 parser->buffer[parser->idx++] = ch;
541         }
542
543         *ppos += read;
544         ret = read;
545
546 out:
547         return ret;
548 }
549
550 ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
551 {
552         int len;
553         int ret;
554
555         if (!cnt)
556                 return 0;
557
558         if (s->len <= s->readpos)
559                 return -EBUSY;
560
561         len = s->len - s->readpos;
562         if (cnt > len)
563                 cnt = len;
564         ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
565         if (ret == cnt)
566                 return -EFAULT;
567
568         cnt -= ret;
569
570         s->readpos += cnt;
571         return cnt;
572 }
573
574 static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
575 {
576         int len;
577         void *ret;
578
579         if (s->len <= s->readpos)
580                 return -EBUSY;
581
582         len = s->len - s->readpos;
583         if (cnt > len)
584                 cnt = len;
585         ret = memcpy(buf, s->buffer + s->readpos, cnt);
586         if (!ret)
587                 return -EFAULT;
588
589         s->readpos += cnt;
590         return cnt;
591 }
592
593 /*
594  * ftrace_max_lock is used to protect the swapping of buffers
595  * when taking a max snapshot. The buffers themselves are
596  * protected by per_cpu spinlocks. But the action of the swap
597  * needs its own lock.
598  *
599  * This is defined as a arch_spinlock_t in order to help
600  * with performance when lockdep debugging is enabled.
601  *
602  * It is also used in other places outside the update_max_tr
603  * so it needs to be defined outside of the
604  * CONFIG_TRACER_MAX_TRACE.
605  */
606 static arch_spinlock_t ftrace_max_lock =
607         (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
608
609 unsigned long __read_mostly     tracing_thresh;
610
611 #ifdef CONFIG_TRACER_MAX_TRACE
612 unsigned long __read_mostly     tracing_max_latency;
613
614 /*
615  * Copy the new maximum trace into the separate maximum-trace
616  * structure. (this way the maximum trace is permanently saved,
617  * for later retrieval via /sys/kernel/debug/tracing/latency_trace)
618  */
619 static void
620 __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
621 {
622         struct trace_array_cpu *data = tr->data[cpu];
623         struct trace_array_cpu *max_data;
624
625         max_tr.cpu = cpu;
626         max_tr.time_start = data->preempt_timestamp;
627
628         max_data = max_tr.data[cpu];
629         max_data->saved_latency = tracing_max_latency;
630         max_data->critical_start = data->critical_start;
631         max_data->critical_end = data->critical_end;
632
633         memcpy(max_data->comm, tsk->comm, TASK_COMM_LEN);
634         max_data->pid = tsk->pid;
635         max_data->uid = task_uid(tsk);
636         max_data->nice = tsk->static_prio - 20 - MAX_RT_PRIO;
637         max_data->policy = tsk->policy;
638         max_data->rt_priority = tsk->rt_priority;
639
640         /* record this tasks comm */
641         tracing_record_cmdline(tsk);
642 }
643
644 /**
645  * update_max_tr - snapshot all trace buffers from global_trace to max_tr
646  * @tr: tracer
647  * @tsk: the task with the latency
648  * @cpu: The cpu that initiated the trace.
649  *
650  * Flip the buffers between the @tr and the max_tr and record information
651  * about which task was the cause of this latency.
652  */
653 void
654 update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
655 {
656         struct ring_buffer *buf = tr->buffer;
657
658         if (trace_stop_count)
659                 return;
660
661         WARN_ON_ONCE(!irqs_disabled());
662         arch_spin_lock(&ftrace_max_lock);
663
664         tr->buffer = max_tr.buffer;
665         max_tr.buffer = buf;
666
667         __update_max_tr(tr, tsk, cpu);
668         arch_spin_unlock(&ftrace_max_lock);
669 }
670
671 /**
672  * update_max_tr_single - only copy one trace over, and reset the rest
673  * @tr - tracer
674  * @tsk - task with the latency
675  * @cpu - the cpu of the buffer to copy.
676  *
677  * Flip the trace of a single CPU buffer between the @tr and the max_tr.
678  */
679 void
680 update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
681 {
682         int ret;
683
684         if (trace_stop_count)
685                 return;
686
687         WARN_ON_ONCE(!irqs_disabled());
688         arch_spin_lock(&ftrace_max_lock);
689
690         ftrace_disable_cpu();
691
692         ret = ring_buffer_swap_cpu(max_tr.buffer, tr->buffer, cpu);
693
694         if (ret == -EBUSY) {
695                 /*
696                  * We failed to swap the buffer due to a commit taking
697                  * place on this CPU. We fail to record, but we reset
698                  * the max trace buffer (no one writes directly to it)
699                  * and flag that it failed.
700                  */
701                 trace_array_printk(&max_tr, _THIS_IP_,
702                         "Failed to swap buffers due to commit in progress\n");
703         }
704
705         ftrace_enable_cpu();
706
707         WARN_ON_ONCE(ret && ret != -EAGAIN && ret != -EBUSY);
708
709         __update_max_tr(tr, tsk, cpu);
710         arch_spin_unlock(&ftrace_max_lock);
711 }
712 #endif /* CONFIG_TRACER_MAX_TRACE */
713
714 /**
715  * register_tracer - register a tracer with the ftrace system.
716  * @type - the plugin for the tracer
717  *
718  * Register a new plugin tracer.
719  */
720 int register_tracer(struct tracer *type)
721 __releases(kernel_lock)
722 __acquires(kernel_lock)
723 {
724         struct tracer *t;
725         int ret = 0;
726
727         if (!type->name) {
728                 pr_info("Tracer must have a name\n");
729                 return -1;
730         }
731
732         if (strlen(type->name) > MAX_TRACER_SIZE) {
733                 pr_info("Tracer has a name longer than %d\n", MAX_TRACER_SIZE);
734                 return -1;
735         }
736
737         /*
738          * When this gets called we hold the BKL which means that
739          * preemption is disabled. Various trace selftests however
740          * need to disable and enable preemption for successful tests.
741          * So we drop the BKL here and grab it after the tests again.
742          */
743         unlock_kernel();
744         mutex_lock(&trace_types_lock);
745
746         tracing_selftest_running = true;
747
748         for (t = trace_types; t; t = t->next) {
749                 if (strcmp(type->name, t->name) == 0) {
750                         /* already found */
751                         pr_info("Tracer %s already registered\n",
752                                 type->name);
753                         ret = -1;
754                         goto out;
755                 }
756         }
757
758         if (!type->set_flag)
759                 type->set_flag = &dummy_set_flag;
760         if (!type->flags)
761                 type->flags = &dummy_tracer_flags;
762         else
763                 if (!type->flags->opts)
764                         type->flags->opts = dummy_tracer_opt;
765         if (!type->wait_pipe)
766                 type->wait_pipe = default_wait_pipe;
767
768
769 #ifdef CONFIG_FTRACE_STARTUP_TEST
770         if (type->selftest && !tracing_selftest_disabled) {
771                 struct tracer *saved_tracer = current_trace;
772                 struct trace_array *tr = &global_trace;
773
774                 /*
775                  * Run a selftest on this tracer.
776                  * Here we reset the trace buffer, and set the current
777                  * tracer to be this tracer. The tracer can then run some
778                  * internal tracing to verify that everything is in order.
779                  * If we fail, we do not register this tracer.
780                  */
781                 tracing_reset_online_cpus(tr);
782
783                 current_trace = type;
784                 /* the test is responsible for initializing and enabling */
785                 pr_info("Testing tracer %s: ", type->name);
786                 ret = type->selftest(type, tr);
787                 /* the test is responsible for resetting too */
788                 current_trace = saved_tracer;
789                 if (ret) {
790                         printk(KERN_CONT "FAILED!\n");
791                         goto out;
792                 }
793                 /* Only reset on passing, to avoid touching corrupted buffers */
794                 tracing_reset_online_cpus(tr);
795
796                 printk(KERN_CONT "PASSED\n");
797         }
798 #endif
799
800         type->next = trace_types;
801         trace_types = type;
802
803  out:
804         tracing_selftest_running = false;
805         mutex_unlock(&trace_types_lock);
806
807         if (ret || !default_bootup_tracer)
808                 goto out_unlock;
809
810         if (strncmp(default_bootup_tracer, type->name, MAX_TRACER_SIZE))
811                 goto out_unlock;
812
813         printk(KERN_INFO "Starting tracer '%s'\n", type->name);
814         /* Do we want this tracer to start on bootup? */
815         tracing_set_tracer(type->name);
816         default_bootup_tracer = NULL;
817         /* disable other selftests, since this will break it. */
818         tracing_selftest_disabled = 1;
819 #ifdef CONFIG_FTRACE_STARTUP_TEST
820         printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
821                type->name);
822 #endif
823
824  out_unlock:
825         lock_kernel();
826         return ret;
827 }
828
829 void unregister_tracer(struct tracer *type)
830 {
831         struct tracer **t;
832
833         mutex_lock(&trace_types_lock);
834         for (t = &trace_types; *t; t = &(*t)->next) {
835                 if (*t == type)
836                         goto found;
837         }
838         pr_info("Tracer %s not registered\n", type->name);
839         goto out;
840
841  found:
842         *t = (*t)->next;
843
844         if (type == current_trace && tracer_enabled) {
845                 tracer_enabled = 0;
846                 tracing_stop();
847                 if (current_trace->stop)
848                         current_trace->stop(&global_trace);
849                 current_trace = &nop_trace;
850         }
851 out:
852         mutex_unlock(&trace_types_lock);
853 }
854
855 static void __tracing_reset(struct ring_buffer *buffer, int cpu)
856 {
857         ftrace_disable_cpu();
858         ring_buffer_reset_cpu(buffer, cpu);
859         ftrace_enable_cpu();
860 }
861
862 void tracing_reset(struct trace_array *tr, int cpu)
863 {
864         struct ring_buffer *buffer = tr->buffer;
865
866         ring_buffer_record_disable(buffer);
867
868         /* Make sure all commits have finished */
869         synchronize_sched();
870         __tracing_reset(buffer, cpu);
871
872         ring_buffer_record_enable(buffer);
873 }
874
875 void tracing_reset_online_cpus(struct trace_array *tr)
876 {
877         struct ring_buffer *buffer = tr->buffer;
878         int cpu;
879
880         ring_buffer_record_disable(buffer);
881
882         /* Make sure all commits have finished */
883         synchronize_sched();
884
885         tr->time_start = ftrace_now(tr->cpu);
886
887         for_each_online_cpu(cpu)
888                 __tracing_reset(buffer, cpu);
889
890         ring_buffer_record_enable(buffer);
891 }
892
893 void tracing_reset_current(int cpu)
894 {
895         tracing_reset(&global_trace, cpu);
896 }
897
898 void tracing_reset_current_online_cpus(void)
899 {
900         tracing_reset_online_cpus(&global_trace);
901 }
902
903 #define SAVED_CMDLINES 128
904 #define NO_CMDLINE_MAP UINT_MAX
905 static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
906 static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
907 static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
908 static int cmdline_idx;
909 static arch_spinlock_t trace_cmdline_lock = __ARCH_SPIN_LOCK_UNLOCKED;
910
911 /* temporary disable recording */
912 static atomic_t trace_record_cmdline_disabled __read_mostly;
913
914 static void trace_init_cmdlines(void)
915 {
916         memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
917         memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
918         cmdline_idx = 0;
919 }
920
921 int is_tracing_stopped(void)
922 {
923         return trace_stop_count;
924 }
925
926 /**
927  * ftrace_off_permanent - disable all ftrace code permanently
928  *
929  * This should only be called when a serious anomally has
930  * been detected.  This will turn off the function tracing,
931  * ring buffers, and other tracing utilites. It takes no
932  * locks and can be called from any context.
933  */
934 void ftrace_off_permanent(void)
935 {
936         tracing_disabled = 1;
937         ftrace_stop();
938         tracing_off_permanent();
939 }
940
941 /**
942  * tracing_start - quick start of the tracer
943  *
944  * If tracing is enabled but was stopped by tracing_stop,
945  * this will start the tracer back up.
946  */
947 void tracing_start(void)
948 {
949         struct ring_buffer *buffer;
950         unsigned long flags;
951
952         if (tracing_disabled)
953                 return;
954
955         spin_lock_irqsave(&tracing_start_lock, flags);
956         if (--trace_stop_count) {
957                 if (trace_stop_count < 0) {
958                         /* Someone screwed up their debugging */
959                         WARN_ON_ONCE(1);
960                         trace_stop_count = 0;
961                 }
962                 goto out;
963         }
964
965         /* Prevent the buffers from switching */
966         arch_spin_lock(&ftrace_max_lock);
967
968         buffer = global_trace.buffer;
969         if (buffer)
970                 ring_buffer_record_enable(buffer);
971
972         buffer = max_tr.buffer;
973         if (buffer)
974                 ring_buffer_record_enable(buffer);
975
976         arch_spin_unlock(&ftrace_max_lock);
977
978         ftrace_start();
979  out:
980         spin_unlock_irqrestore(&tracing_start_lock, flags);
981 }
982
983 /**
984  * tracing_stop - quick stop of the tracer
985  *
986  * Light weight way to stop tracing. Use in conjunction with
987  * tracing_start.
988  */
989 void tracing_stop(void)
990 {
991         struct ring_buffer *buffer;
992         unsigned long flags;
993
994         ftrace_stop();
995         spin_lock_irqsave(&tracing_start_lock, flags);
996         if (trace_stop_count++)
997                 goto out;
998
999         /* Prevent the buffers from switching */
1000         arch_spin_lock(&ftrace_max_lock);
1001
1002         buffer = global_trace.buffer;
1003         if (buffer)
1004                 ring_buffer_record_disable(buffer);
1005
1006         buffer = max_tr.buffer;
1007         if (buffer)
1008                 ring_buffer_record_disable(buffer);
1009
1010         arch_spin_unlock(&ftrace_max_lock);
1011
1012  out:
1013         spin_unlock_irqrestore(&tracing_start_lock, flags);
1014 }
1015
1016 void trace_stop_cmdline_recording(void);
1017
1018 static void trace_save_cmdline(struct task_struct *tsk)
1019 {
1020         unsigned pid, idx;
1021
1022         if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
1023                 return;
1024
1025         /*
1026          * It's not the end of the world if we don't get
1027          * the lock, but we also don't want to spin
1028          * nor do we want to disable interrupts,
1029          * so if we miss here, then better luck next time.
1030          */
1031         if (!arch_spin_trylock(&trace_cmdline_lock))
1032                 return;
1033
1034         idx = map_pid_to_cmdline[tsk->pid];
1035         if (idx == NO_CMDLINE_MAP) {
1036                 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
1037
1038                 /*
1039                  * Check whether the cmdline buffer at idx has a pid
1040                  * mapped. We are going to overwrite that entry so we
1041                  * need to clear the map_pid_to_cmdline. Otherwise we
1042                  * would read the new comm for the old pid.
1043                  */
1044                 pid = map_cmdline_to_pid[idx];
1045                 if (pid != NO_CMDLINE_MAP)
1046                         map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
1047
1048                 map_cmdline_to_pid[idx] = tsk->pid;
1049                 map_pid_to_cmdline[tsk->pid] = idx;
1050
1051                 cmdline_idx = idx;
1052         }
1053
1054         memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
1055
1056         arch_spin_unlock(&trace_cmdline_lock);
1057 }
1058
1059 void trace_find_cmdline(int pid, char comm[])
1060 {
1061         unsigned map;
1062
1063         if (!pid) {
1064                 strcpy(comm, "<idle>");
1065                 return;
1066         }
1067
1068         if (WARN_ON_ONCE(pid < 0)) {
1069                 strcpy(comm, "<XXX>");
1070                 return;
1071         }
1072
1073         if (pid > PID_MAX_DEFAULT) {
1074                 strcpy(comm, "<...>");
1075                 return;
1076         }
1077
1078         preempt_disable();
1079         arch_spin_lock(&trace_cmdline_lock);
1080         map = map_pid_to_cmdline[pid];
1081         if (map != NO_CMDLINE_MAP)
1082                 strcpy(comm, saved_cmdlines[map]);
1083         else
1084                 strcpy(comm, "<...>");
1085
1086         arch_spin_unlock(&trace_cmdline_lock);
1087         preempt_enable();
1088 }
1089
1090 void tracing_record_cmdline(struct task_struct *tsk)
1091 {
1092         if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
1093             !tracing_is_on())
1094                 return;
1095
1096         trace_save_cmdline(tsk);
1097 }
1098
1099 void
1100 tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
1101                              int pc)
1102 {
1103         struct task_struct *tsk = current;
1104
1105         entry->preempt_count            = pc & 0xff;
1106         entry->pid                      = (tsk) ? tsk->pid : 0;
1107         entry->lock_depth               = (tsk) ? tsk->lock_depth : 0;
1108         entry->flags =
1109 #ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
1110                 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
1111 #else
1112                 TRACE_FLAG_IRQS_NOSUPPORT |
1113 #endif
1114                 ((pc & HARDIRQ_MASK) ? TRACE_FLAG_HARDIRQ : 0) |
1115                 ((pc & SOFTIRQ_MASK) ? TRACE_FLAG_SOFTIRQ : 0) |
1116                 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
1117 }
1118 EXPORT_SYMBOL_GPL(tracing_generic_entry_update);
1119
1120 struct ring_buffer_event *
1121 trace_buffer_lock_reserve(struct ring_buffer *buffer,
1122                           int type,
1123                           unsigned long len,
1124                           unsigned long flags, int pc)
1125 {
1126         struct ring_buffer_event *event;
1127
1128         event = ring_buffer_lock_reserve(buffer, len);
1129         if (event != NULL) {
1130                 struct trace_entry *ent = ring_buffer_event_data(event);
1131
1132                 tracing_generic_entry_update(ent, flags, pc);
1133                 ent->type = type;
1134         }
1135
1136         return event;
1137 }
1138
1139 static inline void
1140 __trace_buffer_unlock_commit(struct ring_buffer *buffer,
1141                              struct ring_buffer_event *event,
1142                              unsigned long flags, int pc,
1143                              int wake)
1144 {
1145         ring_buffer_unlock_commit(buffer, event);
1146
1147         ftrace_trace_stack(buffer, flags, 6, pc);
1148         ftrace_trace_userstack(buffer, flags, pc);
1149
1150         if (wake)
1151                 trace_wake_up();
1152 }
1153
1154 void trace_buffer_unlock_commit(struct ring_buffer *buffer,
1155                                 struct ring_buffer_event *event,
1156                                 unsigned long flags, int pc)
1157 {
1158         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1159 }
1160
1161 struct ring_buffer_event *
1162 trace_current_buffer_lock_reserve(struct ring_buffer **current_rb,
1163                                   int type, unsigned long len,
1164                                   unsigned long flags, int pc)
1165 {
1166         *current_rb = global_trace.buffer;
1167         return trace_buffer_lock_reserve(*current_rb,
1168                                          type, len, flags, pc);
1169 }
1170 EXPORT_SYMBOL_GPL(trace_current_buffer_lock_reserve);
1171
1172 void trace_current_buffer_unlock_commit(struct ring_buffer *buffer,
1173                                         struct ring_buffer_event *event,
1174                                         unsigned long flags, int pc)
1175 {
1176         __trace_buffer_unlock_commit(buffer, event, flags, pc, 1);
1177 }
1178 EXPORT_SYMBOL_GPL(trace_current_buffer_unlock_commit);
1179
1180 void trace_nowake_buffer_unlock_commit(struct ring_buffer *buffer,
1181                                        struct ring_buffer_event *event,
1182                                        unsigned long flags, int pc)
1183 {
1184         __trace_buffer_unlock_commit(buffer, event, flags, pc, 0);
1185 }
1186 EXPORT_SYMBOL_GPL(trace_nowake_buffer_unlock_commit);
1187
1188 void trace_current_buffer_discard_commit(struct ring_buffer *buffer,
1189                                          struct ring_buffer_event *event)
1190 {
1191         ring_buffer_discard_commit(buffer, event);
1192 }
1193 EXPORT_SYMBOL_GPL(trace_current_buffer_discard_commit);
1194
1195 void
1196 trace_function(struct trace_array *tr,
1197                unsigned long ip, unsigned long parent_ip, unsigned long flags,
1198                int pc)
1199 {
1200         struct ftrace_event_call *call = &event_function;
1201         struct ring_buffer *buffer = tr->buffer;
1202         struct ring_buffer_event *event;
1203         struct ftrace_entry *entry;
1204
1205         /* If we are reading the ring buffer, don't trace */
1206         if (unlikely(__this_cpu_read(ftrace_cpu_disabled)))
1207                 return;
1208
1209         event = trace_buffer_lock_reserve(buffer, TRACE_FN, sizeof(*entry),
1210                                           flags, pc);
1211         if (!event)
1212                 return;
1213         entry   = ring_buffer_event_data(event);
1214         entry->ip                       = ip;
1215         entry->parent_ip                = parent_ip;
1216
1217         if (!filter_check_discard(call, entry, buffer, event))
1218                 ring_buffer_unlock_commit(buffer, event);
1219 }
1220
1221 void
1222 ftrace(struct trace_array *tr, struct trace_array_cpu *data,
1223        unsigned long ip, unsigned long parent_ip, unsigned long flags,
1224        int pc)
1225 {
1226         if (likely(!atomic_read(&data->disabled)))
1227                 trace_function(tr, ip, parent_ip, flags, pc);
1228 }
1229
1230 #ifdef CONFIG_STACKTRACE
1231 static void __ftrace_trace_stack(struct ring_buffer *buffer,
1232                                  unsigned long flags,
1233                                  int skip, int pc)
1234 {
1235         struct ftrace_event_call *call = &event_kernel_stack;
1236         struct ring_buffer_event *event;
1237         struct stack_entry *entry;
1238         struct stack_trace trace;
1239
1240         event = trace_buffer_lock_reserve(buffer, TRACE_STACK,
1241                                           sizeof(*entry), flags, pc);
1242         if (!event)
1243                 return;
1244         entry   = ring_buffer_event_data(event);
1245         memset(&entry->caller, 0, sizeof(entry->caller));
1246
1247         trace.nr_entries        = 0;
1248         trace.max_entries       = FTRACE_STACK_ENTRIES;
1249         trace.skip              = skip;
1250         trace.entries           = entry->caller;
1251
1252         save_stack_trace(&trace);
1253         if (!filter_check_discard(call, entry, buffer, event))
1254                 ring_buffer_unlock_commit(buffer, event);
1255 }
1256
1257 void ftrace_trace_stack(struct ring_buffer *buffer, unsigned long flags,
1258                         int skip, int pc)
1259 {
1260         if (!(trace_flags & TRACE_ITER_STACKTRACE))
1261                 return;
1262
1263         __ftrace_trace_stack(buffer, flags, skip, pc);
1264 }
1265
1266 void __trace_stack(struct trace_array *tr, unsigned long flags, int skip,
1267                    int pc)
1268 {
1269         __ftrace_trace_stack(tr->buffer, flags, skip, pc);
1270 }
1271
1272 /**
1273  * trace_dump_stack - record a stack back trace in the trace buffer
1274  */
1275 void trace_dump_stack(void)
1276 {
1277         unsigned long flags;
1278
1279         if (tracing_disabled || tracing_selftest_running)
1280                 return;
1281
1282         local_save_flags(flags);
1283
1284         /* skipping 3 traces, seems to get us at the caller of this function */
1285         __ftrace_trace_stack(global_trace.buffer, flags, 3, preempt_count());
1286 }
1287
1288 void
1289 ftrace_trace_userstack(struct ring_buffer *buffer, unsigned long flags, int pc)
1290 {
1291         struct ftrace_event_call *call = &event_user_stack;
1292         struct ring_buffer_event *event;
1293         struct userstack_entry *entry;
1294         struct stack_trace trace;
1295
1296         if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1297                 return;
1298
1299         /*
1300          * NMIs can not handle page faults, even with fix ups.
1301          * The save user stack can (and often does) fault.
1302          */
1303         if (unlikely(in_nmi()))
1304                 return;
1305
1306         event = trace_buffer_lock_reserve(buffer, TRACE_USER_STACK,
1307                                           sizeof(*entry), flags, pc);
1308         if (!event)
1309                 return;
1310         entry   = ring_buffer_event_data(event);
1311
1312         entry->tgid             = current->tgid;
1313         memset(&entry->caller, 0, sizeof(entry->caller));
1314
1315         trace.nr_entries        = 0;
1316         trace.max_entries       = FTRACE_STACK_ENTRIES;
1317         trace.skip              = 0;
1318         trace.entries           = entry->caller;
1319
1320         save_stack_trace_user(&trace);
1321         if (!filter_check_discard(call, entry, buffer, event))
1322                 ring_buffer_unlock_commit(buffer, event);
1323 }
1324
1325 #ifdef UNUSED
1326 static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1327 {
1328         ftrace_trace_userstack(tr, flags, preempt_count());
1329 }
1330 #endif /* UNUSED */
1331
1332 #endif /* CONFIG_STACKTRACE */
1333
1334 static void
1335 ftrace_trace_special(void *__tr,
1336                      unsigned long arg1, unsigned long arg2, unsigned long arg3,
1337                      int pc)
1338 {
1339         struct ftrace_event_call *call = &event_special;
1340         struct ring_buffer_event *event;
1341         struct trace_array *tr = __tr;
1342         struct ring_buffer *buffer = tr->buffer;
1343         struct special_entry *entry;
1344
1345         event = trace_buffer_lock_reserve(buffer, TRACE_SPECIAL,
1346                                           sizeof(*entry), 0, pc);
1347         if (!event)
1348                 return;
1349         entry   = ring_buffer_event_data(event);
1350         entry->arg1                     = arg1;
1351         entry->arg2                     = arg2;
1352         entry->arg3                     = arg3;
1353
1354         if (!filter_check_discard(call, entry, buffer, event))
1355                 trace_buffer_unlock_commit(buffer, event, 0, pc);
1356 }
1357
1358 void
1359 __trace_special(void *__tr, void *__data,
1360                 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1361 {
1362         ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1363 }
1364
1365 void
1366 ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1367 {
1368         struct trace_array *tr = &global_trace;
1369         struct trace_array_cpu *data;
1370         unsigned long flags;
1371         int cpu;
1372         int pc;
1373
1374         if (tracing_disabled)
1375                 return;
1376
1377         pc = preempt_count();
1378         local_irq_save(flags);
1379         cpu = raw_smp_processor_id();
1380         data = tr->data[cpu];
1381
1382         if (likely(atomic_inc_return(&data->disabled) == 1))
1383                 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1384
1385         atomic_dec(&data->disabled);
1386         local_irq_restore(flags);
1387 }
1388
1389 /**
1390  * trace_vbprintk - write binary msg to tracing buffer
1391  *
1392  */
1393 int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1394 {
1395         static arch_spinlock_t trace_buf_lock =
1396                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
1397         static u32 trace_buf[TRACE_BUF_SIZE];
1398
1399         struct ftrace_event_call *call = &event_bprint;
1400         struct ring_buffer_event *event;
1401         struct ring_buffer *buffer;
1402         struct trace_array *tr = &global_trace;
1403         struct trace_array_cpu *data;
1404         struct bprint_entry *entry;
1405         unsigned long flags;
1406         int disable;
1407         int cpu, len = 0, size, pc;
1408
1409         if (unlikely(tracing_selftest_running || tracing_disabled))
1410                 return 0;
1411
1412         /* Don't pollute graph traces with trace_vprintk internals */
1413         pause_graph_tracing();
1414
1415         pc = preempt_count();
1416         preempt_disable_notrace();
1417         cpu = raw_smp_processor_id();
1418         data = tr->data[cpu];
1419
1420         disable = atomic_inc_return(&data->disabled);
1421         if (unlikely(disable != 1))
1422                 goto out;
1423
1424         /* Lockdep uses trace_printk for lock tracing */
1425         local_irq_save(flags);
1426         arch_spin_lock(&trace_buf_lock);
1427         len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1428
1429         if (len > TRACE_BUF_SIZE || len < 0)
1430                 goto out_unlock;
1431
1432         size = sizeof(*entry) + sizeof(u32) * len;
1433         buffer = tr->buffer;
1434         event = trace_buffer_lock_reserve(buffer, TRACE_BPRINT, size,
1435                                           flags, pc);
1436         if (!event)
1437                 goto out_unlock;
1438         entry = ring_buffer_event_data(event);
1439         entry->ip                       = ip;
1440         entry->fmt                      = fmt;
1441
1442         memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1443         if (!filter_check_discard(call, entry, buffer, event)) {
1444                 ring_buffer_unlock_commit(buffer, event);
1445                 ftrace_trace_stack(buffer, flags, 6, pc);
1446         }
1447
1448 out_unlock:
1449         arch_spin_unlock(&trace_buf_lock);
1450         local_irq_restore(flags);
1451
1452 out:
1453         atomic_dec_return(&data->disabled);
1454         preempt_enable_notrace();
1455         unpause_graph_tracing();
1456
1457         return len;
1458 }
1459 EXPORT_SYMBOL_GPL(trace_vbprintk);
1460
1461 int trace_array_printk(struct trace_array *tr,
1462                        unsigned long ip, const char *fmt, ...)
1463 {
1464         int ret;
1465         va_list ap;
1466
1467         if (!(trace_flags & TRACE_ITER_PRINTK))
1468                 return 0;
1469
1470         va_start(ap, fmt);
1471         ret = trace_array_vprintk(tr, ip, fmt, ap);
1472         va_end(ap);
1473         return ret;
1474 }
1475
1476 int trace_array_vprintk(struct trace_array *tr,
1477                         unsigned long ip, const char *fmt, va_list args)
1478 {
1479         static arch_spinlock_t trace_buf_lock = __ARCH_SPIN_LOCK_UNLOCKED;
1480         static char trace_buf[TRACE_BUF_SIZE];
1481
1482         struct ftrace_event_call *call = &event_print;
1483         struct ring_buffer_event *event;
1484         struct ring_buffer *buffer;
1485         struct trace_array_cpu *data;
1486         int cpu, len = 0, size, pc;
1487         struct print_entry *entry;
1488         unsigned long irq_flags;
1489         int disable;
1490
1491         if (tracing_disabled || tracing_selftest_running)
1492                 return 0;
1493
1494         pc = preempt_count();
1495         preempt_disable_notrace();
1496         cpu = raw_smp_processor_id();
1497         data = tr->data[cpu];
1498
1499         disable = atomic_inc_return(&data->disabled);
1500         if (unlikely(disable != 1))
1501                 goto out;
1502
1503         pause_graph_tracing();
1504         raw_local_irq_save(irq_flags);
1505         arch_spin_lock(&trace_buf_lock);
1506         len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1507
1508         size = sizeof(*entry) + len + 1;
1509         buffer = tr->buffer;
1510         event = trace_buffer_lock_reserve(buffer, TRACE_PRINT, size,
1511                                           irq_flags, pc);
1512         if (!event)
1513                 goto out_unlock;
1514         entry = ring_buffer_event_data(event);
1515         entry->ip = ip;
1516
1517         memcpy(&entry->buf, trace_buf, len);
1518         entry->buf[len] = '\0';
1519         if (!filter_check_discard(call, entry, buffer, event)) {
1520                 ring_buffer_unlock_commit(buffer, event);
1521                 ftrace_trace_stack(buffer, irq_flags, 6, pc);
1522         }
1523
1524  out_unlock:
1525         arch_spin_unlock(&trace_buf_lock);
1526         raw_local_irq_restore(irq_flags);
1527         unpause_graph_tracing();
1528  out:
1529         atomic_dec_return(&data->disabled);
1530         preempt_enable_notrace();
1531
1532         return len;
1533 }
1534
1535 int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1536 {
1537         return trace_array_vprintk(&global_trace, ip, fmt, args);
1538 }
1539 EXPORT_SYMBOL_GPL(trace_vprintk);
1540
1541 enum trace_file_type {
1542         TRACE_FILE_LAT_FMT      = 1,
1543         TRACE_FILE_ANNOTATE     = 2,
1544 };
1545
1546 static void trace_iterator_increment(struct trace_iterator *iter)
1547 {
1548         /* Don't allow ftrace to trace into the ring buffers */
1549         ftrace_disable_cpu();
1550
1551         iter->idx++;
1552         if (iter->buffer_iter[iter->cpu])
1553                 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1554
1555         ftrace_enable_cpu();
1556 }
1557
1558 static struct trace_entry *
1559 peek_next_entry(struct trace_iterator *iter, int cpu, u64 *ts,
1560                 unsigned long *lost_events)
1561 {
1562         struct ring_buffer_event *event;
1563         struct ring_buffer_iter *buf_iter = iter->buffer_iter[cpu];
1564
1565         /* Don't allow ftrace to trace into the ring buffers */
1566         ftrace_disable_cpu();
1567
1568         if (buf_iter)
1569                 event = ring_buffer_iter_peek(buf_iter, ts);
1570         else
1571                 event = ring_buffer_peek(iter->tr->buffer, cpu, ts,
1572                                          lost_events);
1573
1574         ftrace_enable_cpu();
1575
1576         return event ? ring_buffer_event_data(event) : NULL;
1577 }
1578
1579 static struct trace_entry *
1580 __find_next_entry(struct trace_iterator *iter, int *ent_cpu,
1581                   unsigned long *missing_events, u64 *ent_ts)
1582 {
1583         struct ring_buffer *buffer = iter->tr->buffer;
1584         struct trace_entry *ent, *next = NULL;
1585         unsigned long lost_events = 0, next_lost = 0;
1586         int cpu_file = iter->cpu_file;
1587         u64 next_ts = 0, ts;
1588         int next_cpu = -1;
1589         int cpu;
1590
1591         /*
1592          * If we are in a per_cpu trace file, don't bother by iterating over
1593          * all cpu and peek directly.
1594          */
1595         if (cpu_file > TRACE_PIPE_ALL_CPU) {
1596                 if (ring_buffer_empty_cpu(buffer, cpu_file))
1597                         return NULL;
1598                 ent = peek_next_entry(iter, cpu_file, ent_ts, missing_events);
1599                 if (ent_cpu)
1600                         *ent_cpu = cpu_file;
1601
1602                 return ent;
1603         }
1604
1605         for_each_tracing_cpu(cpu) {
1606
1607                 if (ring_buffer_empty_cpu(buffer, cpu))
1608                         continue;
1609
1610                 ent = peek_next_entry(iter, cpu, &ts, &lost_events);
1611
1612                 /*
1613                  * Pick the entry with the smallest timestamp:
1614                  */
1615                 if (ent && (!next || ts < next_ts)) {
1616                         next = ent;
1617                         next_cpu = cpu;
1618                         next_ts = ts;
1619                         next_lost = lost_events;
1620                 }
1621         }
1622
1623         if (ent_cpu)
1624                 *ent_cpu = next_cpu;
1625
1626         if (ent_ts)
1627                 *ent_ts = next_ts;
1628
1629         if (missing_events)
1630                 *missing_events = next_lost;
1631
1632         return next;
1633 }
1634
1635 /* Find the next real entry, without updating the iterator itself */
1636 struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1637                                           int *ent_cpu, u64 *ent_ts)
1638 {
1639         return __find_next_entry(iter, ent_cpu, NULL, ent_ts);
1640 }
1641
1642 /* Find the next real entry, and increment the iterator to the next entry */
1643 static void *find_next_entry_inc(struct trace_iterator *iter)
1644 {
1645         iter->ent = __find_next_entry(iter, &iter->cpu,
1646                                       &iter->lost_events, &iter->ts);
1647
1648         if (iter->ent)
1649                 trace_iterator_increment(iter);
1650
1651         return iter->ent ? iter : NULL;
1652 }
1653
1654 static void trace_consume(struct trace_iterator *iter)
1655 {
1656         /* Don't allow ftrace to trace into the ring buffers */
1657         ftrace_disable_cpu();
1658         ring_buffer_consume(iter->tr->buffer, iter->cpu, &iter->ts,
1659                             &iter->lost_events);
1660         ftrace_enable_cpu();
1661 }
1662
1663 static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1664 {
1665         struct trace_iterator *iter = m->private;
1666         int i = (int)*pos;
1667         void *ent;
1668
1669         WARN_ON_ONCE(iter->leftover);
1670
1671         (*pos)++;
1672
1673         /* can't go backwards */
1674         if (iter->idx > i)
1675                 return NULL;
1676
1677         if (iter->idx < 0)
1678                 ent = find_next_entry_inc(iter);
1679         else
1680                 ent = iter;
1681
1682         while (ent && iter->idx < i)
1683                 ent = find_next_entry_inc(iter);
1684
1685         iter->pos = *pos;
1686
1687         return ent;
1688 }
1689
1690 static void tracing_iter_reset(struct trace_iterator *iter, int cpu)
1691 {
1692         struct trace_array *tr = iter->tr;
1693         struct ring_buffer_event *event;
1694         struct ring_buffer_iter *buf_iter;
1695         unsigned long entries = 0;
1696         u64 ts;
1697
1698         tr->data[cpu]->skipped_entries = 0;
1699
1700         if (!iter->buffer_iter[cpu])
1701                 return;
1702
1703         buf_iter = iter->buffer_iter[cpu];
1704         ring_buffer_iter_reset(buf_iter);
1705
1706         /*
1707          * We could have the case with the max latency tracers
1708          * that a reset never took place on a cpu. This is evident
1709          * by the timestamp being before the start of the buffer.
1710          */
1711         while ((event = ring_buffer_iter_peek(buf_iter, &ts))) {
1712                 if (ts >= iter->tr->time_start)
1713                         break;
1714                 entries++;
1715                 ring_buffer_read(buf_iter, NULL);
1716         }
1717
1718         tr->data[cpu]->skipped_entries = entries;
1719 }
1720
1721 /*
1722  * The current tracer is copied to avoid a global locking
1723  * all around.
1724  */
1725 static void *s_start(struct seq_file *m, loff_t *pos)
1726 {
1727         struct trace_iterator *iter = m->private;
1728         static struct tracer *old_tracer;
1729         int cpu_file = iter->cpu_file;
1730         void *p = NULL;
1731         loff_t l = 0;
1732         int cpu;
1733
1734         /* copy the tracer to avoid using a global lock all around */
1735         mutex_lock(&trace_types_lock);
1736         if (unlikely(old_tracer != current_trace && current_trace)) {
1737                 old_tracer = current_trace;
1738                 *iter->trace = *current_trace;
1739         }
1740         mutex_unlock(&trace_types_lock);
1741
1742         atomic_inc(&trace_record_cmdline_disabled);
1743
1744         if (*pos != iter->pos) {
1745                 iter->ent = NULL;
1746                 iter->cpu = 0;
1747                 iter->idx = -1;
1748
1749                 ftrace_disable_cpu();
1750
1751                 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1752                         for_each_tracing_cpu(cpu)
1753                                 tracing_iter_reset(iter, cpu);
1754                 } else
1755                         tracing_iter_reset(iter, cpu_file);
1756
1757                 ftrace_enable_cpu();
1758
1759                 iter->leftover = 0;
1760                 for (p = iter; p && l < *pos; p = s_next(m, p, &l))
1761                         ;
1762
1763         } else {
1764                 /*
1765                  * If we overflowed the seq_file before, then we want
1766                  * to just reuse the trace_seq buffer again.
1767                  */
1768                 if (iter->leftover)
1769                         p = iter;
1770                 else {
1771                         l = *pos - 1;
1772                         p = s_next(m, p, &l);
1773                 }
1774         }
1775
1776         trace_event_read_lock();
1777         trace_access_lock(cpu_file);
1778         return p;
1779 }
1780
1781 static void s_stop(struct seq_file *m, void *p)
1782 {
1783         struct trace_iterator *iter = m->private;
1784
1785         atomic_dec(&trace_record_cmdline_disabled);
1786         trace_access_unlock(iter->cpu_file);
1787         trace_event_read_unlock();
1788 }
1789
1790 static void print_lat_help_header(struct seq_file *m)
1791 {
1792         seq_puts(m, "#                  _------=> CPU#            \n");
1793         seq_puts(m, "#                 / _-----=> irqs-off        \n");
1794         seq_puts(m, "#                | / _----=> need-resched    \n");
1795         seq_puts(m, "#                || / _---=> hardirq/softirq \n");
1796         seq_puts(m, "#                ||| / _--=> preempt-depth   \n");
1797         seq_puts(m, "#                |||| /_--=> lock-depth       \n");
1798         seq_puts(m, "#                |||||/     delay             \n");
1799         seq_puts(m, "#  cmd     pid   |||||| time  |   caller      \n");
1800         seq_puts(m, "#     \\   /      ||||||   \\   |   /           \n");
1801 }
1802
1803 static void print_func_help_header(struct seq_file *m)
1804 {
1805         seq_puts(m, "#           TASK-PID    CPU#    TIMESTAMP  FUNCTION\n");
1806         seq_puts(m, "#              | |       |          |         |\n");
1807 }
1808
1809
1810 void
1811 print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1812 {
1813         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1814         struct trace_array *tr = iter->tr;
1815         struct trace_array_cpu *data = tr->data[tr->cpu];
1816         struct tracer *type = current_trace;
1817         unsigned long entries = 0;
1818         unsigned long total = 0;
1819         unsigned long count;
1820         const char *name = "preemption";
1821         int cpu;
1822
1823         if (type)
1824                 name = type->name;
1825
1826
1827         for_each_tracing_cpu(cpu) {
1828                 count = ring_buffer_entries_cpu(tr->buffer, cpu);
1829                 /*
1830                  * If this buffer has skipped entries, then we hold all
1831                  * entries for the trace and we need to ignore the
1832                  * ones before the time stamp.
1833                  */
1834                 if (tr->data[cpu]->skipped_entries) {
1835                         count -= tr->data[cpu]->skipped_entries;
1836                         /* total is the same as the entries */
1837                         total += count;
1838                 } else
1839                         total += count +
1840                                 ring_buffer_overrun_cpu(tr->buffer, cpu);
1841                 entries += count;
1842         }
1843
1844         seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1845                    name, UTS_RELEASE);
1846         seq_puts(m, "# -----------------------------------"
1847                  "---------------------------------\n");
1848         seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1849                    " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1850                    nsecs_to_usecs(data->saved_latency),
1851                    entries,
1852                    total,
1853                    tr->cpu,
1854 #if defined(CONFIG_PREEMPT_NONE)
1855                    "server",
1856 #elif defined(CONFIG_PREEMPT_VOLUNTARY)
1857                    "desktop",
1858 #elif defined(CONFIG_PREEMPT)
1859                    "preempt",
1860 #else
1861                    "unknown",
1862 #endif
1863                    /* These are reserved for later use */
1864                    0, 0, 0, 0);
1865 #ifdef CONFIG_SMP
1866         seq_printf(m, " #P:%d)\n", num_online_cpus());
1867 #else
1868         seq_puts(m, ")\n");
1869 #endif
1870         seq_puts(m, "#    -----------------\n");
1871         seq_printf(m, "#    | task: %.16s-%d "
1872                    "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1873                    data->comm, data->pid, data->uid, data->nice,
1874                    data->policy, data->rt_priority);
1875         seq_puts(m, "#    -----------------\n");
1876
1877         if (data->critical_start) {
1878                 seq_puts(m, "#  => started at: ");
1879                 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1880                 trace_print_seq(m, &iter->seq);
1881                 seq_puts(m, "\n#  => ended at:   ");
1882                 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1883                 trace_print_seq(m, &iter->seq);
1884                 seq_puts(m, "\n#\n");
1885         }
1886
1887         seq_puts(m, "#\n");
1888 }
1889
1890 static void test_cpu_buff_start(struct trace_iterator *iter)
1891 {
1892         struct trace_seq *s = &iter->seq;
1893
1894         if (!(trace_flags & TRACE_ITER_ANNOTATE))
1895                 return;
1896
1897         if (!(iter->iter_flags & TRACE_FILE_ANNOTATE))
1898                 return;
1899
1900         if (cpumask_test_cpu(iter->cpu, iter->started))
1901                 return;
1902
1903         if (iter->tr->data[iter->cpu]->skipped_entries)
1904                 return;
1905
1906         cpumask_set_cpu(iter->cpu, iter->started);
1907
1908         /* Don't print started cpu buffer for the first entry of the trace */
1909         if (iter->idx > 1)
1910                 trace_seq_printf(s, "##### CPU %u buffer started ####\n",
1911                                 iter->cpu);
1912 }
1913
1914 static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1915 {
1916         struct trace_seq *s = &iter->seq;
1917         unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1918         struct trace_entry *entry;
1919         struct trace_event *event;
1920
1921         entry = iter->ent;
1922
1923         test_cpu_buff_start(iter);
1924
1925         event = ftrace_find_event(entry->type);
1926
1927         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1928                 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
1929                         if (!trace_print_lat_context(iter))
1930                                 goto partial;
1931                 } else {
1932                         if (!trace_print_context(iter))
1933                                 goto partial;
1934                 }
1935         }
1936
1937         if (event)
1938                 return event->funcs->trace(iter, sym_flags, event);
1939
1940         if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
1941                 goto partial;
1942
1943         return TRACE_TYPE_HANDLED;
1944 partial:
1945         return TRACE_TYPE_PARTIAL_LINE;
1946 }
1947
1948 static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
1949 {
1950         struct trace_seq *s = &iter->seq;
1951         struct trace_entry *entry;
1952         struct trace_event *event;
1953
1954         entry = iter->ent;
1955
1956         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1957                 if (!trace_seq_printf(s, "%d %d %llu ",
1958                                       entry->pid, iter->cpu, iter->ts))
1959                         goto partial;
1960         }
1961
1962         event = ftrace_find_event(entry->type);
1963         if (event)
1964                 return event->funcs->raw(iter, 0, event);
1965
1966         if (!trace_seq_printf(s, "%d ?\n", entry->type))
1967                 goto partial;
1968
1969         return TRACE_TYPE_HANDLED;
1970 partial:
1971         return TRACE_TYPE_PARTIAL_LINE;
1972 }
1973
1974 static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
1975 {
1976         struct trace_seq *s = &iter->seq;
1977         unsigned char newline = '\n';
1978         struct trace_entry *entry;
1979         struct trace_event *event;
1980
1981         entry = iter->ent;
1982
1983         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
1984                 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
1985                 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
1986                 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
1987         }
1988
1989         event = ftrace_find_event(entry->type);
1990         if (event) {
1991                 enum print_line_t ret = event->funcs->hex(iter, 0, event);
1992                 if (ret != TRACE_TYPE_HANDLED)
1993                         return ret;
1994         }
1995
1996         SEQ_PUT_FIELD_RET(s, newline);
1997
1998         return TRACE_TYPE_HANDLED;
1999 }
2000
2001 static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2002 {
2003         struct trace_seq *s = &iter->seq;
2004         struct trace_entry *entry;
2005         struct trace_event *event;
2006
2007         entry = iter->ent;
2008
2009         if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2010                 SEQ_PUT_FIELD_RET(s, entry->pid);
2011                 SEQ_PUT_FIELD_RET(s, iter->cpu);
2012                 SEQ_PUT_FIELD_RET(s, iter->ts);
2013         }
2014
2015         event = ftrace_find_event(entry->type);
2016         return event ? event->funcs->binary(iter, 0, event) :
2017                 TRACE_TYPE_HANDLED;
2018 }
2019
2020 int trace_empty(struct trace_iterator *iter)
2021 {
2022         int cpu;
2023
2024         /* If we are looking at one CPU buffer, only check that one */
2025         if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
2026                 cpu = iter->cpu_file;
2027                 if (iter->buffer_iter[cpu]) {
2028                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2029                                 return 0;
2030                 } else {
2031                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2032                                 return 0;
2033                 }
2034                 return 1;
2035         }
2036
2037         for_each_tracing_cpu(cpu) {
2038                 if (iter->buffer_iter[cpu]) {
2039                         if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
2040                                 return 0;
2041                 } else {
2042                         if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
2043                                 return 0;
2044                 }
2045         }
2046
2047         return 1;
2048 }
2049
2050 /*  Called with trace_event_read_lock() held. */
2051 static enum print_line_t print_trace_line(struct trace_iterator *iter)
2052 {
2053         enum print_line_t ret;
2054
2055         if (iter->lost_events)
2056                 trace_seq_printf(&iter->seq, "CPU:%d [LOST %lu EVENTS]\n",
2057                                  iter->cpu, iter->lost_events);
2058
2059         if (iter->trace && iter->trace->print_line) {
2060                 ret = iter->trace->print_line(iter);
2061                 if (ret != TRACE_TYPE_UNHANDLED)
2062                         return ret;
2063         }
2064
2065         if (iter->ent->type == TRACE_BPRINT &&
2066                         trace_flags & TRACE_ITER_PRINTK &&
2067                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2068                 return trace_print_bprintk_msg_only(iter);
2069
2070         if (iter->ent->type == TRACE_PRINT &&
2071                         trace_flags & TRACE_ITER_PRINTK &&
2072                         trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2073                 return trace_print_printk_msg_only(iter);
2074
2075         if (trace_flags & TRACE_ITER_BIN)
2076                 return print_bin_fmt(iter);
2077
2078         if (trace_flags & TRACE_ITER_HEX)
2079                 return print_hex_fmt(iter);
2080
2081         if (trace_flags & TRACE_ITER_RAW)
2082                 return print_raw_fmt(iter);
2083
2084         return print_trace_fmt(iter);
2085 }
2086
2087 void trace_default_header(struct seq_file *m)
2088 {
2089         struct trace_iterator *iter = m->private;
2090
2091         if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2092                 /* print nothing if the buffers are empty */
2093                 if (trace_empty(iter))
2094                         return;
2095                 print_trace_header(m, iter);
2096                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2097                         print_lat_help_header(m);
2098         } else {
2099                 if (!(trace_flags & TRACE_ITER_VERBOSE))
2100                         print_func_help_header(m);
2101         }
2102 }
2103
2104 static int s_show(struct seq_file *m, void *v)
2105 {
2106         struct trace_iterator *iter = v;
2107         int ret;
2108
2109         if (iter->ent == NULL) {
2110                 if (iter->tr) {
2111                         seq_printf(m, "# tracer: %s\n", iter->trace->name);
2112                         seq_puts(m, "#\n");
2113                 }
2114                 if (iter->trace && iter->trace->print_header)
2115                         iter->trace->print_header(m);
2116                 else
2117                         trace_default_header(m);
2118
2119         } else if (iter->leftover) {
2120                 /*
2121                  * If we filled the seq_file buffer earlier, we
2122                  * want to just show it now.
2123                  */
2124                 ret = trace_print_seq(m, &iter->seq);
2125
2126                 /* ret should this time be zero, but you never know */
2127                 iter->leftover = ret;
2128
2129         } else {
2130                 print_trace_line(iter);
2131                 ret = trace_print_seq(m, &iter->seq);
2132                 /*
2133                  * If we overflow the seq_file buffer, then it will
2134                  * ask us for this data again at start up.
2135                  * Use that instead.
2136                  *  ret is 0 if seq_file write succeeded.
2137                  *        -1 otherwise.
2138                  */
2139                 iter->leftover = ret;
2140         }
2141
2142         return 0;
2143 }
2144
2145 static const struct seq_operations tracer_seq_ops = {
2146         .start          = s_start,
2147         .next           = s_next,
2148         .stop           = s_stop,
2149         .show           = s_show,
2150 };
2151
2152 static struct trace_iterator *
2153 __tracing_open(struct inode *inode, struct file *file)
2154 {
2155         long cpu_file = (long) inode->i_private;
2156         void *fail_ret = ERR_PTR(-ENOMEM);
2157         struct trace_iterator *iter;
2158         struct seq_file *m;
2159         int cpu, ret;
2160
2161         if (tracing_disabled)
2162                 return ERR_PTR(-ENODEV);
2163
2164         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2165         if (!iter)
2166                 return ERR_PTR(-ENOMEM);
2167
2168         /*
2169          * We make a copy of the current tracer to avoid concurrent
2170          * changes on it while we are reading.
2171          */
2172         mutex_lock(&trace_types_lock);
2173         iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
2174         if (!iter->trace)
2175                 goto fail;
2176
2177         if (current_trace)
2178                 *iter->trace = *current_trace;
2179
2180         if (!zalloc_cpumask_var(&iter->started, GFP_KERNEL))
2181                 goto fail;
2182
2183         if (current_trace && current_trace->print_max)
2184                 iter->tr = &max_tr;
2185         else
2186                 iter->tr = &global_trace;
2187         iter->pos = -1;
2188         mutex_init(&iter->mutex);
2189         iter->cpu_file = cpu_file;
2190
2191         /* Notify the tracer early; before we stop tracing. */
2192         if (iter->trace && iter->trace->open)
2193                 iter->trace->open(iter);
2194
2195         /* Annotate start of buffers if we had overruns */
2196         if (ring_buffer_overruns(iter->tr->buffer))
2197                 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2198
2199         /* stop the trace while dumping */
2200         tracing_stop();
2201
2202         if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
2203                 for_each_tracing_cpu(cpu) {
2204                         iter->buffer_iter[cpu] =
2205                                 ring_buffer_read_prepare(iter->tr->buffer, cpu);
2206                 }
2207                 ring_buffer_read_prepare_sync();
2208                 for_each_tracing_cpu(cpu) {
2209                         ring_buffer_read_start(iter->buffer_iter[cpu]);
2210                         tracing_iter_reset(iter, cpu);
2211                 }
2212         } else {
2213                 cpu = iter->cpu_file;
2214                 iter->buffer_iter[cpu] =
2215                         ring_buffer_read_prepare(iter->tr->buffer, cpu);
2216                 ring_buffer_read_prepare_sync();
2217                 ring_buffer_read_start(iter->buffer_iter[cpu]);
2218                 tracing_iter_reset(iter, cpu);
2219         }
2220
2221         ret = seq_open(file, &tracer_seq_ops);
2222         if (ret < 0) {
2223                 fail_ret = ERR_PTR(ret);
2224                 goto fail_buffer;
2225         }
2226
2227         m = file->private_data;
2228         m->private = iter;
2229
2230         mutex_unlock(&trace_types_lock);
2231
2232         return iter;
2233
2234  fail_buffer:
2235         for_each_tracing_cpu(cpu) {
2236                 if (iter->buffer_iter[cpu])
2237                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2238         }
2239         free_cpumask_var(iter->started);
2240         tracing_start();
2241  fail:
2242         mutex_unlock(&trace_types_lock);
2243         kfree(iter->trace);
2244         kfree(iter);
2245
2246         return fail_ret;
2247 }
2248
2249 int tracing_open_generic(struct inode *inode, struct file *filp)
2250 {
2251         if (tracing_disabled)
2252                 return -ENODEV;
2253
2254         filp->private_data = inode->i_private;
2255         return 0;
2256 }
2257
2258 static int tracing_release(struct inode *inode, struct file *file)
2259 {
2260         struct seq_file *m = (struct seq_file *)file->private_data;
2261         struct trace_iterator *iter;
2262         int cpu;
2263
2264         if (!(file->f_mode & FMODE_READ))
2265                 return 0;
2266
2267         iter = m->private;
2268
2269         mutex_lock(&trace_types_lock);
2270         for_each_tracing_cpu(cpu) {
2271                 if (iter->buffer_iter[cpu])
2272                         ring_buffer_read_finish(iter->buffer_iter[cpu]);
2273         }
2274
2275         if (iter->trace && iter->trace->close)
2276                 iter->trace->close(iter);
2277
2278         /* reenable tracing if it was previously enabled */
2279         tracing_start();
2280         mutex_unlock(&trace_types_lock);
2281
2282         seq_release(inode, file);
2283         mutex_destroy(&iter->mutex);
2284         free_cpumask_var(iter->started);
2285         kfree(iter->trace);
2286         kfree(iter);
2287         return 0;
2288 }
2289
2290 static int tracing_open(struct inode *inode, struct file *file)
2291 {
2292         struct trace_iterator *iter;
2293         int ret = 0;
2294
2295         /* If this file was open for write, then erase contents */
2296         if ((file->f_mode & FMODE_WRITE) &&
2297             (file->f_flags & O_TRUNC)) {
2298                 long cpu = (long) inode->i_private;
2299
2300                 if (cpu == TRACE_PIPE_ALL_CPU)
2301                         tracing_reset_online_cpus(&global_trace);
2302                 else
2303                         tracing_reset(&global_trace, cpu);
2304         }
2305
2306         if (file->f_mode & FMODE_READ) {
2307                 iter = __tracing_open(inode, file);
2308                 if (IS_ERR(iter))
2309                         ret = PTR_ERR(iter);
2310                 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
2311                         iter->iter_flags |= TRACE_FILE_LAT_FMT;
2312         }
2313         return ret;
2314 }
2315
2316 static void *
2317 t_next(struct seq_file *m, void *v, loff_t *pos)
2318 {
2319         struct tracer *t = v;
2320
2321         (*pos)++;
2322
2323         if (t)
2324                 t = t->next;
2325
2326         return t;
2327 }
2328
2329 static void *t_start(struct seq_file *m, loff_t *pos)
2330 {
2331         struct tracer *t;
2332         loff_t l = 0;
2333
2334         mutex_lock(&trace_types_lock);
2335         for (t = trace_types; t && l < *pos; t = t_next(m, t, &l))
2336                 ;
2337
2338         return t;
2339 }
2340
2341 static void t_stop(struct seq_file *m, void *p)
2342 {
2343         mutex_unlock(&trace_types_lock);
2344 }
2345
2346 static int t_show(struct seq_file *m, void *v)
2347 {
2348         struct tracer *t = v;
2349
2350         if (!t)
2351                 return 0;
2352
2353         seq_printf(m, "%s", t->name);
2354         if (t->next)
2355                 seq_putc(m, ' ');
2356         else
2357                 seq_putc(m, '\n');
2358
2359         return 0;
2360 }
2361
2362 static const struct seq_operations show_traces_seq_ops = {
2363         .start          = t_start,
2364         .next           = t_next,
2365         .stop           = t_stop,
2366         .show           = t_show,
2367 };
2368
2369 static int show_traces_open(struct inode *inode, struct file *file)
2370 {
2371         if (tracing_disabled)
2372                 return -ENODEV;
2373
2374         return seq_open(file, &show_traces_seq_ops);
2375 }
2376
2377 static ssize_t
2378 tracing_write_stub(struct file *filp, const char __user *ubuf,
2379                    size_t count, loff_t *ppos)
2380 {
2381         return count;
2382 }
2383
2384 static const struct file_operations tracing_fops = {
2385         .open           = tracing_open,
2386         .read           = seq_read,
2387         .write          = tracing_write_stub,
2388         .llseek         = seq_lseek,
2389         .release        = tracing_release,
2390 };
2391
2392 static const struct file_operations show_traces_fops = {
2393         .open           = show_traces_open,
2394         .read           = seq_read,
2395         .release        = seq_release,
2396 };
2397
2398 /*
2399  * Only trace on a CPU if the bitmask is set:
2400  */
2401 static cpumask_var_t tracing_cpumask;
2402
2403 /*
2404  * The tracer itself will not take this lock, but still we want
2405  * to provide a consistent cpumask to user-space:
2406  */
2407 static DEFINE_MUTEX(tracing_cpumask_update_lock);
2408
2409 /*
2410  * Temporary storage for the character representation of the
2411  * CPU bitmask (and one more byte for the newline):
2412  */
2413 static char mask_str[NR_CPUS + 1];
2414
2415 static ssize_t
2416 tracing_cpumask_read(struct file *filp, char __user *ubuf,
2417                      size_t count, loff_t *ppos)
2418 {
2419         int len;
2420
2421         mutex_lock(&tracing_cpumask_update_lock);
2422
2423         len = cpumask_scnprintf(mask_str, count, tracing_cpumask);
2424         if (count - len < 2) {
2425                 count = -EINVAL;
2426                 goto out_err;
2427         }
2428         len += sprintf(mask_str + len, "\n");
2429         count = simple_read_from_buffer(ubuf, count, ppos, mask_str, NR_CPUS+1);
2430
2431 out_err:
2432         mutex_unlock(&tracing_cpumask_update_lock);
2433
2434         return count;
2435 }
2436
2437 static ssize_t
2438 tracing_cpumask_write(struct file *filp, const char __user *ubuf,
2439                       size_t count, loff_t *ppos)
2440 {
2441         int err, cpu;
2442         cpumask_var_t tracing_cpumask_new;
2443
2444         if (!alloc_cpumask_var(&tracing_cpumask_new, GFP_KERNEL))
2445                 return -ENOMEM;
2446
2447         err = cpumask_parse_user(ubuf, count, tracing_cpumask_new);
2448         if (err)
2449                 goto err_unlock;
2450
2451         mutex_lock(&tracing_cpumask_update_lock);
2452
2453         local_irq_disable();
2454         arch_spin_lock(&ftrace_max_lock);
2455         for_each_tracing_cpu(cpu) {
2456                 /*
2457                  * Increase/decrease the disabled counter if we are
2458                  * about to flip a bit in the cpumask:
2459                  */
2460                 if (cpumask_test_cpu(cpu, tracing_cpumask) &&
2461                                 !cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2462                         atomic_inc(&global_trace.data[cpu]->disabled);
2463                 }
2464                 if (!cpumask_test_cpu(cpu, tracing_cpumask) &&
2465                                 cpumask_test_cpu(cpu, tracing_cpumask_new)) {
2466                         atomic_dec(&global_trace.data[cpu]->disabled);
2467                 }
2468         }
2469         arch_spin_unlock(&ftrace_max_lock);
2470         local_irq_enable();
2471
2472         cpumask_copy(tracing_cpumask, tracing_cpumask_new);
2473
2474         mutex_unlock(&tracing_cpumask_update_lock);
2475         free_cpumask_var(tracing_cpumask_new);
2476
2477         return count;
2478
2479 err_unlock:
2480         free_cpumask_var(tracing_cpumask_new);
2481
2482         return err;
2483 }
2484
2485 static const struct file_operations tracing_cpumask_fops = {
2486         .open           = tracing_open_generic,
2487         .read           = tracing_cpumask_read,
2488         .write          = tracing_cpumask_write,
2489 };
2490
2491 static int tracing_trace_options_show(struct seq_file *m, void *v)
2492 {
2493         struct tracer_opt *trace_opts;
2494         u32 tracer_flags;
2495         int i;
2496
2497         mutex_lock(&trace_types_lock);
2498         tracer_flags = current_trace->flags->val;
2499         trace_opts = current_trace->flags->opts;
2500
2501         for (i = 0; trace_options[i]; i++) {
2502                 if (trace_flags & (1 << i))
2503                         seq_printf(m, "%s\n", trace_options[i]);
2504                 else
2505                         seq_printf(m, "no%s\n", trace_options[i]);
2506         }
2507
2508         for (i = 0; trace_opts[i].name; i++) {
2509                 if (tracer_flags & trace_opts[i].bit)
2510                         seq_printf(m, "%s\n", trace_opts[i].name);
2511                 else
2512                         seq_printf(m, "no%s\n", trace_opts[i].name);
2513         }
2514         mutex_unlock(&trace_types_lock);
2515
2516         return 0;
2517 }
2518
2519 static int __set_tracer_option(struct tracer *trace,
2520                                struct tracer_flags *tracer_flags,
2521                                struct tracer_opt *opts, int neg)
2522 {
2523         int ret;
2524
2525         ret = trace->set_flag(tracer_flags->val, opts->bit, !neg);
2526         if (ret)
2527                 return ret;
2528
2529         if (neg)
2530                 tracer_flags->val &= ~opts->bit;
2531         else
2532                 tracer_flags->val |= opts->bit;
2533         return 0;
2534 }
2535
2536 /* Try to assign a tracer specific option */
2537 static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2538 {
2539         struct tracer_flags *tracer_flags = trace->flags;
2540         struct tracer_opt *opts = NULL;
2541         int i;
2542
2543         for (i = 0; tracer_flags->opts[i].name; i++) {
2544                 opts = &tracer_flags->opts[i];
2545
2546                 if (strcmp(cmp, opts->name) == 0)
2547                         return __set_tracer_option(trace, trace->flags,
2548                                                    opts, neg);
2549         }
2550
2551         return -EINVAL;
2552 }
2553
2554 static void set_tracer_flags(unsigned int mask, int enabled)
2555 {
2556         /* do nothing if flag is already set */
2557         if (!!(trace_flags & mask) == !!enabled)
2558                 return;
2559
2560         if (enabled)
2561                 trace_flags |= mask;
2562         else
2563                 trace_flags &= ~mask;
2564 }
2565
2566 static ssize_t
2567 tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2568                         size_t cnt, loff_t *ppos)
2569 {
2570         char buf[64];
2571         char *cmp;
2572         int neg = 0;
2573         int ret;
2574         int i;
2575
2576         if (cnt >= sizeof(buf))
2577                 return -EINVAL;
2578
2579         if (copy_from_user(&buf, ubuf, cnt))
2580                 return -EFAULT;
2581
2582         buf[cnt] = 0;
2583         cmp = strstrip(buf);
2584
2585         if (strncmp(cmp, "no", 2) == 0) {
2586                 neg = 1;
2587                 cmp += 2;
2588         }
2589
2590         for (i = 0; trace_options[i]; i++) {
2591                 if (strcmp(cmp, trace_options[i]) == 0) {
2592                         set_tracer_flags(1 << i, !neg);
2593                         break;
2594                 }
2595         }
2596
2597         /* If no option could be set, test the specific tracer options */
2598         if (!trace_options[i]) {
2599                 mutex_lock(&trace_types_lock);
2600                 ret = set_tracer_option(current_trace, cmp, neg);
2601                 mutex_unlock(&trace_types_lock);
2602                 if (ret)
2603                         return ret;
2604         }
2605
2606         *ppos += cnt;
2607
2608         return cnt;
2609 }
2610
2611 static int tracing_trace_options_open(struct inode *inode, struct file *file)
2612 {
2613         if (tracing_disabled)
2614                 return -ENODEV;
2615         return single_open(file, tracing_trace_options_show, NULL);
2616 }
2617
2618 static const struct file_operations tracing_iter_fops = {
2619         .open           = tracing_trace_options_open,
2620         .read           = seq_read,
2621         .llseek         = seq_lseek,
2622         .release        = single_release,
2623         .write          = tracing_trace_options_write,
2624 };
2625
2626 static const char readme_msg[] =
2627         "tracing mini-HOWTO:\n\n"
2628         "# mount -t debugfs nodev /sys/kernel/debug\n\n"
2629         "# cat /sys/kernel/debug/tracing/available_tracers\n"
2630         "wakeup preemptirqsoff preemptoff irqsoff function sched_switch nop\n\n"
2631         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2632         "nop\n"
2633         "# echo sched_switch > /sys/kernel/debug/tracing/current_tracer\n"
2634         "# cat /sys/kernel/debug/tracing/current_tracer\n"
2635         "sched_switch\n"
2636         "# cat /sys/kernel/debug/tracing/trace_options\n"
2637         "noprint-parent nosym-offset nosym-addr noverbose\n"
2638         "# echo print-parent > /sys/kernel/debug/tracing/trace_options\n"
2639         "# echo 1 > /sys/kernel/debug/tracing/tracing_enabled\n"
2640         "# cat /sys/kernel/debug/tracing/trace > /tmp/trace.txt\n"
2641         "# echo 0 > /sys/kernel/debug/tracing/tracing_enabled\n"
2642 ;
2643
2644 static ssize_t
2645 tracing_readme_read(struct file *filp, char __user *ubuf,
2646                        size_t cnt, loff_t *ppos)
2647 {
2648         return simple_read_from_buffer(ubuf, cnt, ppos,
2649                                         readme_msg, strlen(readme_msg));
2650 }
2651
2652 static const struct file_operations tracing_readme_fops = {
2653         .open           = tracing_open_generic,
2654         .read           = tracing_readme_read,
2655 };
2656
2657 static ssize_t
2658 tracing_saved_cmdlines_read(struct file *file, char __user *ubuf,
2659                                 size_t cnt, loff_t *ppos)
2660 {
2661         char *buf_comm;
2662         char *file_buf;
2663         char *buf;
2664         int len = 0;
2665         int pid;
2666         int i;
2667
2668         file_buf = kmalloc(SAVED_CMDLINES*(16+TASK_COMM_LEN), GFP_KERNEL);
2669         if (!file_buf)
2670                 return -ENOMEM;
2671
2672         buf_comm = kmalloc(TASK_COMM_LEN, GFP_KERNEL);
2673         if (!buf_comm) {
2674                 kfree(file_buf);
2675                 return -ENOMEM;
2676         }
2677
2678         buf = file_buf;
2679
2680         for (i = 0; i < SAVED_CMDLINES; i++) {
2681                 int r;
2682
2683                 pid = map_cmdline_to_pid[i];
2684                 if (pid == -1 || pid == NO_CMDLINE_MAP)
2685                         continue;
2686
2687                 trace_find_cmdline(pid, buf_comm);
2688                 r = sprintf(buf, "%d %s\n", pid, buf_comm);
2689                 buf += r;
2690                 len += r;
2691         }
2692
2693         len = simple_read_from_buffer(ubuf, cnt, ppos,
2694                                       file_buf, len);
2695
2696         kfree(file_buf);
2697         kfree(buf_comm);
2698
2699         return len;
2700 }
2701
2702 static const struct file_operations tracing_saved_cmdlines_fops = {
2703     .open       = tracing_open_generic,
2704     .read       = tracing_saved_cmdlines_read,
2705 };
2706
2707 static ssize_t
2708 tracing_ctrl_read(struct file *filp, char __user *ubuf,
2709                   size_t cnt, loff_t *ppos)
2710 {
2711         char buf[64];
2712         int r;
2713
2714         r = sprintf(buf, "%u\n", tracer_enabled);
2715         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2716 }
2717
2718 static ssize_t
2719 tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2720                    size_t cnt, loff_t *ppos)
2721 {
2722         struct trace_array *tr = filp->private_data;
2723         char buf[64];
2724         unsigned long val;
2725         int ret;
2726
2727         if (cnt >= sizeof(buf))
2728                 return -EINVAL;
2729
2730         if (copy_from_user(&buf, ubuf, cnt))
2731                 return -EFAULT;
2732
2733         buf[cnt] = 0;
2734
2735         ret = strict_strtoul(buf, 10, &val);
2736         if (ret < 0)
2737                 return ret;
2738
2739         val = !!val;
2740
2741         mutex_lock(&trace_types_lock);
2742         if (tracer_enabled ^ val) {
2743                 if (val) {
2744                         tracer_enabled = 1;
2745                         if (current_trace->start)
2746                                 current_trace->start(tr);
2747                         tracing_start();
2748                 } else {
2749                         tracer_enabled = 0;
2750                         tracing_stop();
2751                         if (current_trace->stop)
2752                                 current_trace->stop(tr);
2753                 }
2754         }
2755         mutex_unlock(&trace_types_lock);
2756
2757         *ppos += cnt;
2758
2759         return cnt;
2760 }
2761
2762 static ssize_t
2763 tracing_set_trace_read(struct file *filp, char __user *ubuf,
2764                        size_t cnt, loff_t *ppos)
2765 {
2766         char buf[MAX_TRACER_SIZE+2];
2767         int r;
2768
2769         mutex_lock(&trace_types_lock);
2770         if (current_trace)
2771                 r = sprintf(buf, "%s\n", current_trace->name);
2772         else
2773                 r = sprintf(buf, "\n");
2774         mutex_unlock(&trace_types_lock);
2775
2776         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2777 }
2778
2779 int tracer_init(struct tracer *t, struct trace_array *tr)
2780 {
2781         tracing_reset_online_cpus(tr);
2782         return t->init(tr);
2783 }
2784
2785 static int tracing_resize_ring_buffer(unsigned long size)
2786 {
2787         int ret;
2788
2789         /*
2790          * If kernel or user changes the size of the ring buffer
2791          * we use the size that was given, and we can forget about
2792          * expanding it later.
2793          */
2794         ring_buffer_expanded = 1;
2795
2796         ret = ring_buffer_resize(global_trace.buffer, size);
2797         if (ret < 0)
2798                 return ret;
2799
2800         ret = ring_buffer_resize(max_tr.buffer, size);
2801         if (ret < 0) {
2802                 int r;
2803
2804                 r = ring_buffer_resize(global_trace.buffer,
2805                                        global_trace.entries);
2806                 if (r < 0) {
2807                         /*
2808                          * AARGH! We are left with different
2809                          * size max buffer!!!!
2810                          * The max buffer is our "snapshot" buffer.
2811                          * When a tracer needs a snapshot (one of the
2812                          * latency tracers), it swaps the max buffer
2813                          * with the saved snap shot. We succeeded to
2814                          * update the size of the main buffer, but failed to
2815                          * update the size of the max buffer. But when we tried
2816                          * to reset the main buffer to the original size, we
2817                          * failed there too. This is very unlikely to
2818                          * happen, but if it does, warn and kill all
2819                          * tracing.
2820                          */
2821                         WARN_ON(1);
2822                         tracing_disabled = 1;
2823                 }
2824                 return ret;
2825         }
2826
2827         global_trace.entries = size;
2828
2829         return ret;
2830 }
2831
2832 /**
2833  * tracing_update_buffers - used by tracing facility to expand ring buffers
2834  *
2835  * To save on memory when the tracing is never used on a system with it
2836  * configured in. The ring buffers are set to a minimum size. But once
2837  * a user starts to use the tracing facility, then they need to grow
2838  * to their default size.
2839  *
2840  * This function is to be called when a tracer is about to be used.
2841  */
2842 int tracing_update_buffers(void)
2843 {
2844         int ret = 0;
2845
2846         mutex_lock(&trace_types_lock);
2847         if (!ring_buffer_expanded)
2848                 ret = tracing_resize_ring_buffer(trace_buf_size);
2849         mutex_unlock(&trace_types_lock);
2850
2851         return ret;
2852 }
2853
2854 struct trace_option_dentry;
2855
2856 static struct trace_option_dentry *
2857 create_trace_option_files(struct tracer *tracer);
2858
2859 static void
2860 destroy_trace_option_files(struct trace_option_dentry *topts);
2861
2862 static int tracing_set_tracer(const char *buf)
2863 {
2864         static struct trace_option_dentry *topts;
2865         struct trace_array *tr = &global_trace;
2866         struct tracer *t;
2867         int ret = 0;
2868
2869         mutex_lock(&trace_types_lock);
2870
2871         if (!ring_buffer_expanded) {
2872                 ret = tracing_resize_ring_buffer(trace_buf_size);
2873                 if (ret < 0)
2874                         goto out;
2875                 ret = 0;
2876         }
2877
2878         for (t = trace_types; t; t = t->next) {
2879                 if (strcmp(t->name, buf) == 0)
2880                         break;
2881         }
2882         if (!t) {
2883                 ret = -EINVAL;
2884                 goto out;
2885         }
2886         if (t == current_trace)
2887                 goto out;
2888
2889         trace_branch_disable();
2890         if (current_trace && current_trace->reset)
2891                 current_trace->reset(tr);
2892
2893         destroy_trace_option_files(topts);
2894
2895         current_trace = t;
2896
2897         topts = create_trace_option_files(current_trace);
2898
2899         if (t->init) {
2900                 ret = tracer_init(t, tr);
2901                 if (ret)
2902                         goto out;
2903         }
2904
2905         trace_branch_enable(tr);
2906  out:
2907         mutex_unlock(&trace_types_lock);
2908
2909         return ret;
2910 }
2911
2912 static ssize_t
2913 tracing_set_trace_write(struct file *filp, const char __user *ubuf,
2914                         size_t cnt, loff_t *ppos)
2915 {
2916         char buf[MAX_TRACER_SIZE+1];
2917         int i;
2918         size_t ret;
2919         int err;
2920
2921         ret = cnt;
2922
2923         if (cnt > MAX_TRACER_SIZE)
2924                 cnt = MAX_TRACER_SIZE;
2925
2926         if (copy_from_user(&buf, ubuf, cnt))
2927                 return -EFAULT;
2928
2929         buf[cnt] = 0;
2930
2931         /* strip ending whitespace. */
2932         for (i = cnt - 1; i > 0 && isspace(buf[i]); i--)
2933                 buf[i] = 0;
2934
2935         err = tracing_set_tracer(buf);
2936         if (err)
2937                 return err;
2938
2939         *ppos += ret;
2940
2941         return ret;
2942 }
2943
2944 static ssize_t
2945 tracing_max_lat_read(struct file *filp, char __user *ubuf,
2946                      size_t cnt, loff_t *ppos)
2947 {
2948         unsigned long *ptr = filp->private_data;
2949         char buf[64];
2950         int r;
2951
2952         r = snprintf(buf, sizeof(buf), "%ld\n",
2953                      *ptr == (unsigned long)-1 ? -1 : nsecs_to_usecs(*ptr));
2954         if (r > sizeof(buf))
2955                 r = sizeof(buf);
2956         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2957 }
2958
2959 static ssize_t
2960 tracing_max_lat_write(struct file *filp, const char __user *ubuf,
2961                       size_t cnt, loff_t *ppos)
2962 {
2963         unsigned long *ptr = filp->private_data;
2964         char buf[64];
2965         unsigned long val;
2966         int ret;
2967
2968         if (cnt >= sizeof(buf))
2969                 return -EINVAL;
2970
2971         if (copy_from_user(&buf, ubuf, cnt))
2972                 return -EFAULT;
2973
2974         buf[cnt] = 0;
2975
2976         ret = strict_strtoul(buf, 10, &val);
2977         if (ret < 0)
2978                 return ret;
2979
2980         *ptr = val * 1000;
2981
2982         return cnt;
2983 }
2984
2985 static int tracing_open_pipe(struct inode *inode, struct file *filp)
2986 {
2987         long cpu_file = (long) inode->i_private;
2988         struct trace_iterator *iter;
2989         int ret = 0;
2990
2991         if (tracing_disabled)
2992                 return -ENODEV;
2993
2994         mutex_lock(&trace_types_lock);
2995
2996         /* create a buffer to store the information to pass to userspace */
2997         iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2998         if (!iter) {
2999                 ret = -ENOMEM;
3000                 goto out;
3001         }
3002
3003         /*
3004          * We make a copy of the current tracer to avoid concurrent
3005          * changes on it while we are reading.
3006          */
3007         iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
3008         if (!iter->trace) {
3009                 ret = -ENOMEM;
3010                 goto fail;
3011         }
3012         if (current_trace)
3013                 *iter->trace = *current_trace;
3014
3015         if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
3016                 ret = -ENOMEM;
3017                 goto fail;
3018         }
3019
3020         /* trace pipe does not show start of buffer */
3021         cpumask_setall(iter->started);
3022
3023         if (trace_flags & TRACE_ITER_LATENCY_FMT)
3024                 iter->iter_flags |= TRACE_FILE_LAT_FMT;
3025
3026         iter->cpu_file = cpu_file;
3027         iter->tr = &global_trace;
3028         mutex_init(&iter->mutex);
3029         filp->private_data = iter;
3030
3031         if (iter->trace->pipe_open)
3032                 iter->trace->pipe_open(iter);
3033
3034 out:
3035         mutex_unlock(&trace_types_lock);
3036         return ret;
3037
3038 fail:
3039         kfree(iter->trace);
3040         kfree(iter);
3041         mutex_unlock(&trace_types_lock);
3042         return ret;
3043 }
3044
3045 static int tracing_release_pipe(struct inode *inode, struct file *file)
3046 {
3047         struct trace_iterator *iter = file->private_data;
3048
3049         mutex_lock(&trace_types_lock);
3050
3051         if (iter->trace->pipe_close)
3052                 iter->trace->pipe_close(iter);
3053
3054         mutex_unlock(&trace_types_lock);
3055
3056         free_cpumask_var(iter->started);
3057         mutex_destroy(&iter->mutex);
3058         kfree(iter->trace);
3059         kfree(iter);
3060
3061         return 0;
3062 }
3063
3064 static unsigned int
3065 tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3066 {
3067         struct trace_iterator *iter = filp->private_data;
3068
3069         if (trace_flags & TRACE_ITER_BLOCK) {
3070                 /*
3071                  * Always select as readable when in blocking mode
3072                  */
3073                 return POLLIN | POLLRDNORM;
3074         } else {
3075                 if (!trace_empty(iter))
3076                         return POLLIN | POLLRDNORM;
3077                 poll_wait(filp, &trace_wait, poll_table);
3078                 if (!trace_empty(iter))
3079                         return POLLIN | POLLRDNORM;
3080
3081                 return 0;
3082         }
3083 }
3084
3085
3086 void default_wait_pipe(struct trace_iterator *iter)
3087 {
3088         DEFINE_WAIT(wait);
3089
3090         prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3091
3092         if (trace_empty(iter))
3093                 schedule();
3094
3095         finish_wait(&trace_wait, &wait);
3096 }
3097
3098 /*
3099  * This is a make-shift waitqueue.
3100  * A tracer might use this callback on some rare cases:
3101  *
3102  *  1) the current tracer might hold the runqueue lock when it wakes up
3103  *     a reader, hence a deadlock (sched, function, and function graph tracers)
3104  *  2) the function tracers, trace all functions, we don't want
3105  *     the overhead of calling wake_up and friends
3106  *     (and tracing them too)
3107  *
3108  *     Anyway, this is really very primitive wakeup.
3109  */
3110 void poll_wait_pipe(struct trace_iterator *iter)
3111 {
3112         set_current_state(TASK_INTERRUPTIBLE);
3113         /* sleep for 100 msecs, and try again. */
3114         schedule_timeout(HZ / 10);
3115 }
3116
3117 /* Must be called with trace_types_lock mutex held. */
3118 static int tracing_wait_pipe(struct file *filp)
3119 {
3120         struct trace_iterator *iter = filp->private_data;
3121
3122         while (trace_empty(iter)) {
3123
3124                 if ((filp->f_flags & O_NONBLOCK)) {
3125                         return -EAGAIN;
3126                 }
3127
3128                 mutex_unlock(&iter->mutex);
3129
3130                 iter->trace->wait_pipe(iter);
3131
3132                 mutex_lock(&iter->mutex);
3133
3134                 if (signal_pending(current))
3135                         return -EINTR;
3136
3137                 /*
3138                  * We block until we read something and tracing is disabled.
3139                  * We still block if tracing is disabled, but we have never
3140                  * read anything. This allows a user to cat this file, and
3141                  * then enable tracing. But after we have read something,
3142                  * we give an EOF when tracing is again disabled.
3143                  *
3144                  * iter->pos will be 0 if we haven't read anything.
3145                  */
3146                 if (!tracer_enabled && iter->pos)
3147                         break;
3148         }
3149
3150         return 1;
3151 }
3152
3153 /*
3154  * Consumer reader.
3155  */
3156 static ssize_t
3157 tracing_read_pipe(struct file *filp, char __user *ubuf,
3158                   size_t cnt, loff_t *ppos)
3159 {
3160         struct trace_iterator *iter = filp->private_data;
3161         static struct tracer *old_tracer;
3162         ssize_t sret;
3163
3164         /* return any leftover data */
3165         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3166         if (sret != -EBUSY)
3167                 return sret;
3168
3169         trace_seq_init(&iter->seq);
3170
3171         /* copy the tracer to avoid using a global lock all around */
3172         mutex_lock(&trace_types_lock);
3173         if (unlikely(old_tracer != current_trace && current_trace)) {
3174                 old_tracer = current_trace;
3175                 *iter->trace = *current_trace;
3176         }
3177         mutex_unlock(&trace_types_lock);
3178
3179         /*
3180          * Avoid more than one consumer on a single file descriptor
3181          * This is just a matter of traces coherency, the ring buffer itself
3182          * is protected.
3183          */
3184         mutex_lock(&iter->mutex);
3185         if (iter->trace->read) {
3186                 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
3187                 if (sret)
3188                         goto out;
3189         }
3190
3191 waitagain:
3192         sret = tracing_wait_pipe(filp);
3193         if (sret <= 0)
3194                 goto out;
3195
3196         /* stop when tracing is finished */
3197         if (trace_empty(iter)) {
3198                 sret = 0;
3199                 goto out;
3200         }
3201
3202         if (cnt >= PAGE_SIZE)
3203                 cnt = PAGE_SIZE - 1;
3204
3205         /* reset all but tr, trace, and overruns */
3206         memset(&iter->seq, 0,
3207                sizeof(struct trace_iterator) -
3208                offsetof(struct trace_iterator, seq));
3209         iter->pos = -1;
3210
3211         trace_event_read_lock();
3212         trace_access_lock(iter->cpu_file);
3213         while (find_next_entry_inc(iter) != NULL) {
3214                 enum print_line_t ret;
3215                 int len = iter->seq.len;
3216
3217                 ret = print_trace_line(iter);
3218                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3219                         /* don't print partial lines */
3220                         iter->seq.len = len;
3221                         break;
3222                 }
3223                 if (ret != TRACE_TYPE_NO_CONSUME)
3224                         trace_consume(iter);
3225
3226                 if (iter->seq.len >= cnt)
3227                         break;
3228         }
3229         trace_access_unlock(iter->cpu_file);
3230         trace_event_read_unlock();
3231
3232         /* Now copy what we have to the user */
3233         sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3234         if (iter->seq.readpos >= iter->seq.len)
3235                 trace_seq_init(&iter->seq);
3236
3237         /*
3238          * If there was nothing to send to user, inspite of consuming trace
3239          * entries, go back to wait for more entries.
3240          */
3241         if (sret == -EBUSY)
3242                 goto waitagain;
3243
3244 out:
3245         mutex_unlock(&iter->mutex);
3246
3247         return sret;
3248 }
3249
3250 static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
3251                                      struct pipe_buffer *buf)
3252 {
3253         __free_page(buf->page);
3254 }
3255
3256 static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
3257                                      unsigned int idx)
3258 {
3259         __free_page(spd->pages[idx]);
3260 }
3261
3262 static const struct pipe_buf_operations tracing_pipe_buf_ops = {
3263         .can_merge              = 0,
3264         .map                    = generic_pipe_buf_map,
3265         .unmap                  = generic_pipe_buf_unmap,
3266         .confirm                = generic_pipe_buf_confirm,
3267         .release                = tracing_pipe_buf_release,
3268         .steal                  = generic_pipe_buf_steal,
3269         .get                    = generic_pipe_buf_get,
3270 };
3271
3272 static size_t
3273 tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
3274 {
3275         size_t count;
3276         int ret;
3277
3278         /* Seq buffer is page-sized, exactly what we need. */
3279         for (;;) {
3280                 count = iter->seq.len;
3281                 ret = print_trace_line(iter);
3282                 count = iter->seq.len - count;
3283                 if (rem < count) {
3284                         rem = 0;
3285                         iter->seq.len -= count;
3286                         break;
3287                 }
3288                 if (ret == TRACE_TYPE_PARTIAL_LINE) {
3289                         iter->seq.len -= count;
3290                         break;
3291                 }
3292
3293                 if (ret != TRACE_TYPE_NO_CONSUME)
3294                         trace_consume(iter);
3295                 rem -= count;
3296                 if (!find_next_entry_inc(iter)) {
3297                         rem = 0;
3298                         iter->ent = NULL;
3299                         break;
3300                 }
3301         }
3302
3303         return rem;
3304 }
3305
3306 static ssize_t tracing_splice_read_pipe(struct file *filp,
3307                                         loff_t *ppos,
3308                                         struct pipe_inode_info *pipe,
3309                                         size_t len,
3310                                         unsigned int flags)
3311 {
3312         struct page *pages_def[PIPE_DEF_BUFFERS];
3313         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3314         struct trace_iterator *iter = filp->private_data;
3315         struct splice_pipe_desc spd = {
3316                 .pages          = pages_def,
3317                 .partial        = partial_def,
3318                 .nr_pages       = 0, /* This gets updated below. */
3319                 .flags          = flags,
3320                 .ops            = &tracing_pipe_buf_ops,
3321                 .spd_release    = tracing_spd_release_pipe,
3322         };
3323         static struct tracer *old_tracer;
3324         ssize_t ret;
3325         size_t rem;
3326         unsigned int i;
3327
3328         if (splice_grow_spd(pipe, &spd))
3329                 return -ENOMEM;
3330
3331         /* copy the tracer to avoid using a global lock all around */
3332         mutex_lock(&trace_types_lock);
3333         if (unlikely(old_tracer != current_trace && current_trace)) {
3334                 old_tracer = current_trace;
3335                 *iter->trace = *current_trace;
3336         }
3337         mutex_unlock(&trace_types_lock);
3338
3339         mutex_lock(&iter->mutex);
3340
3341         if (iter->trace->splice_read) {
3342                 ret = iter->trace->splice_read(iter, filp,
3343                                                ppos, pipe, len, flags);
3344                 if (ret)
3345                         goto out_err;
3346         }
3347
3348         ret = tracing_wait_pipe(filp);
3349         if (ret <= 0)
3350                 goto out_err;
3351
3352         if (!iter->ent && !find_next_entry_inc(iter)) {
3353                 ret = -EFAULT;
3354                 goto out_err;
3355         }
3356
3357         trace_event_read_lock();
3358         trace_access_lock(iter->cpu_file);
3359
3360         /* Fill as many pages as possible. */
3361         for (i = 0, rem = len; i < pipe->buffers && rem; i++) {
3362                 spd.pages[i] = alloc_page(GFP_KERNEL);
3363                 if (!spd.pages[i])
3364                         break;
3365
3366                 rem = tracing_fill_pipe_page(rem, iter);
3367
3368                 /* Copy the data into the page, so we can start over. */
3369                 ret = trace_seq_to_buffer(&iter->seq,
3370                                           page_address(spd.pages[i]),
3371                                           iter->seq.len);
3372                 if (ret < 0) {
3373                         __free_page(spd.pages[i]);
3374                         break;
3375                 }
3376                 spd.partial[i].offset = 0;
3377                 spd.partial[i].len = iter->seq.len;
3378
3379                 trace_seq_init(&iter->seq);
3380         }
3381
3382         trace_access_unlock(iter->cpu_file);
3383         trace_event_read_unlock();
3384         mutex_unlock(&iter->mutex);
3385
3386         spd.nr_pages = i;
3387
3388         ret = splice_to_pipe(pipe, &spd);
3389 out:
3390         splice_shrink_spd(pipe, &spd);
3391         return ret;
3392
3393 out_err:
3394         mutex_unlock(&iter->mutex);
3395         goto out;
3396 }
3397
3398 static ssize_t
3399 tracing_entries_read(struct file *filp, char __user *ubuf,
3400                      size_t cnt, loff_t *ppos)
3401 {
3402         struct trace_array *tr = filp->private_data;
3403         char buf[96];
3404         int r;
3405
3406         mutex_lock(&trace_types_lock);
3407         if (!ring_buffer_expanded)
3408                 r = sprintf(buf, "%lu (expanded: %lu)\n",
3409                             tr->entries >> 10,
3410                             trace_buf_size >> 10);
3411         else
3412                 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3413         mutex_unlock(&trace_types_lock);
3414
3415         return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3416 }
3417
3418 static ssize_t
3419 tracing_entries_write(struct file *filp, const char __user *ubuf,
3420                       size_t cnt, loff_t *ppos)
3421 {
3422         unsigned long val;
3423         char buf[64];
3424         int ret, cpu;
3425
3426         if (cnt >= sizeof(buf))
3427                 return -EINVAL;
3428
3429         if (copy_from_user(&buf, ubuf, cnt))
3430                 return -EFAULT;
3431
3432         buf[cnt] = 0;
3433
3434         ret = strict_strtoul(buf, 10, &val);
3435         if (ret < 0)
3436                 return ret;
3437
3438         /* must have at least 1 entry */
3439         if (!val)
3440                 return -EINVAL;
3441
3442         mutex_lock(&trace_types_lock);
3443
3444         tracing_stop();
3445
3446         /* disable all cpu buffers */
3447         for_each_tracing_cpu(cpu) {
3448                 if (global_trace.data[cpu])
3449                         atomic_inc(&global_trace.data[cpu]->disabled);
3450                 if (max_tr.data[cpu])
3451                         atomic_inc(&max_tr.data[cpu]->disabled);
3452         }
3453
3454         /* value is in KB */
3455         val <<= 10;
3456
3457         if (val != global_trace.entries) {
3458                 ret = tracing_resize_ring_buffer(val);
3459                 if (ret < 0) {
3460                         cnt = ret;
3461                         goto out;
3462                 }
3463         }
3464
3465         *ppos += cnt;
3466
3467         /* If check pages failed, return ENOMEM */
3468         if (tracing_disabled)
3469                 cnt = -ENOMEM;
3470  out:
3471         for_each_tracing_cpu(cpu) {
3472                 if (global_trace.data[cpu])
3473                         atomic_dec(&global_trace.data[cpu]->disabled);
3474                 if (max_tr.data[cpu])
3475                         atomic_dec(&max_tr.data[cpu]->disabled);
3476         }
3477
3478         tracing_start();
3479         max_tr.entries = global_trace.entries;
3480         mutex_unlock(&trace_types_lock);
3481
3482         return cnt;
3483 }
3484
3485 static int mark_printk(const char *fmt, ...)
3486 {
3487         int ret;
3488         va_list args;
3489         va_start(args, fmt);
3490         ret = trace_vprintk(0, fmt, args);
3491         va_end(args);
3492         return ret;
3493 }
3494
3495 static ssize_t
3496 tracing_mark_write(struct file *filp, const char __user *ubuf,
3497                                         size_t cnt, loff_t *fpos)
3498 {
3499         char *buf;
3500
3501         if (tracing_disabled)
3502                 return -EINVAL;
3503
3504         if (cnt > TRACE_BUF_SIZE)
3505                 cnt = TRACE_BUF_SIZE;
3506
3507         buf = kmalloc(cnt + 2, GFP_KERNEL);
3508         if (buf == NULL)
3509                 return -ENOMEM;
3510
3511         if (copy_from_user(buf, ubuf, cnt)) {
3512                 kfree(buf);
3513                 return -EFAULT;
3514         }
3515         if (buf[cnt-1] != '\n') {
3516                 buf[cnt] = '\n';
3517                 buf[cnt+1] = '\0';
3518         } else
3519                 buf[cnt] = '\0';
3520
3521         cnt = mark_printk("%s", buf);
3522         kfree(buf);
3523         *fpos += cnt;
3524
3525         return cnt;
3526 }
3527
3528 static int tracing_clock_show(struct seq_file *m, void *v)
3529 {
3530         int i;
3531
3532         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++)
3533                 seq_printf(m,
3534                         "%s%s%s%s", i ? " " : "",
3535                         i == trace_clock_id ? "[" : "", trace_clocks[i].name,
3536                         i == trace_clock_id ? "]" : "");
3537         seq_putc(m, '\n');
3538
3539         return 0;
3540 }
3541
3542 static ssize_t tracing_clock_write(struct file *filp, const char __user *ubuf,
3543                                    size_t cnt, loff_t *fpos)
3544 {
3545         char buf[64];
3546         const char *clockstr;
3547         int i;
3548
3549         if (cnt >= sizeof(buf))
3550                 return -EINVAL;
3551
3552         if (copy_from_user(&buf, ubuf, cnt))
3553                 return -EFAULT;
3554
3555         buf[cnt] = 0;
3556
3557         clockstr = strstrip(buf);
3558
3559         for (i = 0; i < ARRAY_SIZE(trace_clocks); i++) {
3560                 if (strcmp(trace_clocks[i].name, clockstr) == 0)
3561                         break;
3562         }
3563         if (i == ARRAY_SIZE(trace_clocks))
3564                 return -EINVAL;
3565
3566         trace_clock_id = i;
3567
3568         mutex_lock(&trace_types_lock);
3569
3570         ring_buffer_set_clock(global_trace.buffer, trace_clocks[i].func);
3571         if (max_tr.buffer)
3572                 ring_buffer_set_clock(max_tr.buffer, trace_clocks[i].func);
3573
3574         mutex_unlock(&trace_types_lock);
3575
3576         *fpos += cnt;
3577
3578         return cnt;
3579 }
3580
3581 static int tracing_clock_open(struct inode *inode, struct file *file)
3582 {
3583         if (tracing_disabled)
3584                 return -ENODEV;
3585         return single_open(file, tracing_clock_show, NULL);
3586 }
3587
3588 static const struct file_operations tracing_max_lat_fops = {
3589         .open           = tracing_open_generic,
3590         .read           = tracing_max_lat_read,
3591         .write          = tracing_max_lat_write,
3592 };
3593
3594 static const struct file_operations tracing_ctrl_fops = {
3595         .open           = tracing_open_generic,
3596         .read           = tracing_ctrl_read,
3597         .write          = tracing_ctrl_write,
3598 };
3599
3600 static const struct file_operations set_tracer_fops = {
3601         .open           = tracing_open_generic,
3602         .read           = tracing_set_trace_read,
3603         .write          = tracing_set_trace_write,
3604 };
3605
3606 static const struct file_operations tracing_pipe_fops = {
3607         .open           = tracing_open_pipe,
3608         .poll           = tracing_poll_pipe,
3609         .read           = tracing_read_pipe,
3610         .splice_read    = tracing_splice_read_pipe,
3611         .release        = tracing_release_pipe,
3612 };
3613
3614 static const struct file_operations tracing_entries_fops = {
3615         .open           = tracing_open_generic,
3616         .read           = tracing_entries_read,
3617         .write          = tracing_entries_write,
3618 };
3619
3620 static const struct file_operations tracing_mark_fops = {
3621         .open           = tracing_open_generic,
3622         .write          = tracing_mark_write,
3623 };
3624
3625 static const struct file_operations trace_clock_fops = {
3626         .open           = tracing_clock_open,
3627         .read           = seq_read,
3628         .llseek         = seq_lseek,
3629         .release        = single_release,
3630         .write          = tracing_clock_write,
3631 };
3632
3633 struct ftrace_buffer_info {
3634         struct trace_array      *tr;
3635         void                    *spare;
3636         int                     cpu;
3637         unsigned int            read;
3638 };
3639
3640 static int tracing_buffers_open(struct inode *inode, struct file *filp)
3641 {
3642         int cpu = (int)(long)inode->i_private;
3643         struct ftrace_buffer_info *info;
3644
3645         if (tracing_disabled)
3646                 return -ENODEV;
3647
3648         info = kzalloc(sizeof(*info), GFP_KERNEL);
3649         if (!info)
3650                 return -ENOMEM;
3651
3652         info->tr        = &global_trace;
3653         info->cpu       = cpu;
3654         info->spare     = NULL;
3655         /* Force reading ring buffer for first read */
3656         info->read      = (unsigned int)-1;
3657
3658         filp->private_data = info;
3659
3660         return nonseekable_open(inode, filp);
3661 }
3662
3663 static ssize_t
3664 tracing_buffers_read(struct file *filp, char __user *ubuf,
3665                      size_t count, loff_t *ppos)
3666 {
3667         struct ftrace_buffer_info *info = filp->private_data;
3668         ssize_t ret;
3669         size_t size;
3670
3671         if (!count)
3672                 return 0;
3673
3674         if (!info->spare)
3675                 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3676         if (!info->spare)
3677                 return -ENOMEM;
3678
3679         /* Do we have previous read data to read? */
3680         if (info->read < PAGE_SIZE)
3681                 goto read;
3682
3683         info->read = 0;
3684
3685         trace_access_lock(info->cpu);
3686         ret = ring_buffer_read_page(info->tr->buffer,
3687                                     &info->spare,
3688                                     count,
3689                                     info->cpu, 0);
3690         trace_access_unlock(info->cpu);
3691         if (ret < 0)
3692                 return 0;
3693
3694 read:
3695         size = PAGE_SIZE - info->read;
3696         if (size > count)
3697                 size = count;
3698
3699         ret = copy_to_user(ubuf, info->spare + info->read, size);
3700         if (ret == size)
3701                 return -EFAULT;
3702         size -= ret;
3703
3704         *ppos += size;
3705         info->read += size;
3706
3707         return size;
3708 }
3709
3710 static int tracing_buffers_release(struct inode *inode, struct file *file)
3711 {
3712         struct ftrace_buffer_info *info = file->private_data;
3713
3714         if (info->spare)
3715                 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3716         kfree(info);
3717
3718         return 0;
3719 }
3720
3721 struct buffer_ref {
3722         struct ring_buffer      *buffer;
3723         void                    *page;
3724         int                     ref;
3725 };
3726
3727 static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3728                                     struct pipe_buffer *buf)
3729 {
3730         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3731
3732         if (--ref->ref)
3733                 return;
3734
3735         ring_buffer_free_read_page(ref->buffer, ref->page);
3736         kfree(ref);
3737         buf->private = 0;
3738 }
3739
3740 static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3741                                  struct pipe_buffer *buf)
3742 {
3743         return 1;
3744 }
3745
3746 static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3747                                 struct pipe_buffer *buf)
3748 {
3749         struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3750
3751         ref->ref++;
3752 }
3753
3754 /* Pipe buffer operations for a buffer. */
3755 static const struct pipe_buf_operations buffer_pipe_buf_ops = {
3756         .can_merge              = 0,
3757         .map                    = generic_pipe_buf_map,
3758         .unmap                  = generic_pipe_buf_unmap,
3759         .confirm                = generic_pipe_buf_confirm,
3760         .release                = buffer_pipe_buf_release,
3761         .steal                  = buffer_pipe_buf_steal,
3762         .get                    = buffer_pipe_buf_get,
3763 };
3764
3765 /*
3766  * Callback from splice_to_pipe(), if we need to release some pages
3767  * at the end of the spd in case we error'ed out in filling the pipe.
3768  */
3769 static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3770 {
3771         struct buffer_ref *ref =
3772                 (struct buffer_ref *)spd->partial[i].private;
3773
3774         if (--ref->ref)
3775                 return;
3776
3777         ring_buffer_free_read_page(ref->buffer, ref->page);
3778         kfree(ref);
3779         spd->partial[i].private = 0;
3780 }
3781
3782 static ssize_t
3783 tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3784                             struct pipe_inode_info *pipe, size_t len,
3785                             unsigned int flags)
3786 {
3787         struct ftrace_buffer_info *info = file->private_data;
3788         struct partial_page partial_def[PIPE_DEF_BUFFERS];
3789         struct page *pages_def[PIPE_DEF_BUFFERS];
3790         struct splice_pipe_desc spd = {
3791                 .pages          = pages_def,
3792                 .partial        = partial_def,
3793                 .flags          = flags,
3794                 .ops            = &buffer_pipe_buf_ops,
3795                 .spd_release    = buffer_spd_release,
3796         };
3797         struct buffer_ref *ref;
3798         int entries, size, i;
3799         size_t ret;
3800
3801         if (splice_grow_spd(pipe, &spd))
3802                 return -ENOMEM;
3803
3804         if (*ppos & (PAGE_SIZE - 1)) {
3805                 WARN_ONCE(1, "Ftrace: previous read must page-align\n");
3806                 ret = -EINVAL;
3807                 goto out;
3808         }
3809
3810         if (len & (PAGE_SIZE - 1)) {
3811                 WARN_ONCE(1, "Ftrace: splice_read should page-align\n");
3812                 if (len < PAGE_SIZE) {
3813                         ret = -EINVAL;
3814                         goto out;
3815                 }
3816                 len &= PAGE_MASK;
3817         }
3818
3819         trace_access_lock(info->cpu);
3820         entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3821
3822         for (i = 0; i < pipe->buffers && len && entries; i++, len -= PAGE_SIZE) {
3823                 struct page *page;
3824                 int r;
3825
3826                 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3827                 if (!ref)
3828                         break;
3829
3830                 ref->ref = 1;
3831                 ref->buffer = info->tr->buffer;
3832                 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3833                 if (!ref->page) {
3834                         kfree(ref);
3835                         break;
3836                 }
3837
3838                 r = ring_buffer_read_page(ref->buffer, &ref->page,
3839                                           len, info->cpu, 1);
3840                 if (r < 0) {
3841                         ring_buffer_free_read_page(ref->buffer,
3842                                                    ref->page);
3843                         kfree(ref);
3844                         break;
3845                 }
3846
3847                 /*
3848                  * zero out any left over data, this is going to
3849                  * user land.
3850                  */
3851                 size = ring_buffer_page_len(ref->page);
3852                 if (size < PAGE_SIZE)
3853                         memset(ref->page + size, 0, PAGE_SIZE - size);
3854
3855                 page = virt_to_page(ref->page);
3856
3857                 spd.pages[i] = page;
3858                 spd.partial[i].len = PAGE_SIZE;
3859                 spd.partial[i].offset = 0;
3860                 spd.partial[i].private = (unsigned long)ref;
3861                 spd.nr_pages++;
3862                 *ppos += PAGE_SIZE;
3863
3864                 entries = ring_buffer_entries_cpu(info->tr->buffer, info->cpu);
3865         }
3866
3867         trace_access_unlock(info->cpu);
3868         spd.nr_pages = i;
3869
3870         /* did we read anything? */
3871         if (!spd.nr_pages) {
3872                 if (flags & SPLICE_F_NONBLOCK)
3873                         ret = -EAGAIN;
3874                 else
3875                         ret = 0;
3876                 /* TODO: block */
3877                 goto out;
3878         }
3879
3880         ret = splice_to_pipe(pipe, &spd);
3881         splice_shrink_spd(pipe, &spd);
3882 out:
3883         return ret;
3884 }
3885
3886 static const struct file_operations tracing_buffers_fops = {
3887         .open           = tracing_buffers_open,
3888         .read           = tracing_buffers_read,
3889         .release        = tracing_buffers_release,
3890         .splice_read    = tracing_buffers_splice_read,
3891         .llseek         = no_llseek,
3892 };
3893
3894 static ssize_t
3895 tracing_stats_read(struct file *filp, char __user *ubuf,
3896                    size_t count, loff_t *ppos)
3897 {
3898         unsigned long cpu = (unsigned long)filp->private_data;
3899         struct trace_array *tr = &global_trace;
3900         struct trace_seq *s;
3901         unsigned long cnt;
3902
3903         s = kmalloc(sizeof(*s), GFP_KERNEL);
3904         if (!s)
3905                 return -ENOMEM;
3906
3907         trace_seq_init(s);
3908
3909         cnt = ring_buffer_entries_cpu(tr->buffer, cpu);
3910         trace_seq_printf(s, "entries: %ld\n", cnt);
3911
3912         cnt = ring_buffer_overrun_cpu(tr->buffer, cpu);
3913         trace_seq_printf(s, "overrun: %ld\n", cnt);
3914
3915         cnt = ring_buffer_commit_overrun_cpu(tr->buffer, cpu);
3916         trace_seq_printf(s, "commit overrun: %ld\n", cnt);
3917
3918         count = simple_read_from_buffer(ubuf, count, ppos, s->buffer, s->len);
3919
3920         kfree(s);
3921
3922         return count;
3923 }
3924
3925 static const struct file_operations tracing_stats_fops = {
3926         .open           = tracing_open_generic,
3927         .read           = tracing_stats_read,
3928 };
3929
3930 #ifdef CONFIG_DYNAMIC_FTRACE
3931
3932 int __weak ftrace_arch_read_dyn_info(char *buf, int size)
3933 {
3934         return 0;
3935 }
3936
3937 static ssize_t
3938 tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3939                   size_t cnt, loff_t *ppos)
3940 {
3941         static char ftrace_dyn_info_buffer[1024];
3942         static DEFINE_MUTEX(dyn_info_mutex);
3943         unsigned long *p = filp->private_data;
3944         char *buf = ftrace_dyn_info_buffer;
3945         int size = ARRAY_SIZE(ftrace_dyn_info_buffer);
3946         int r;
3947
3948         mutex_lock(&dyn_info_mutex);
3949         r = sprintf(buf, "%ld ", *p);
3950
3951         r += ftrace_arch_read_dyn_info(buf+r, (size-1)-r);
3952         buf[r++] = '\n';
3953
3954         r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3955
3956         mutex_unlock(&dyn_info_mutex);
3957
3958         return r;
3959 }
3960
3961 static const struct file_operations tracing_dyn_info_fops = {
3962         .open           = tracing_open_generic,
3963         .read           = tracing_read_dyn_info,
3964 };
3965 #endif
3966
3967 static struct dentry *d_tracer;
3968
3969 struct dentry *tracing_init_dentry(void)
3970 {
3971         static int once;
3972
3973         if (d_tracer)
3974                 return d_tracer;
3975
3976         if (!debugfs_initialized())
3977                 return NULL;
3978
3979         d_tracer = debugfs_create_dir("tracing", NULL);
3980
3981         if (!d_tracer && !once) {
3982                 once = 1;
3983                 pr_warning("Could not create debugfs directory 'tracing'\n");
3984                 return NULL;
3985         }
3986
3987         return d_tracer;
3988 }
3989
3990 static struct dentry *d_percpu;
3991
3992 struct dentry *tracing_dentry_percpu(void)
3993 {
3994         static int once;
3995         struct dentry *d_tracer;
3996
3997         if (d_percpu)
3998                 return d_percpu;
3999
4000         d_tracer = tracing_init_dentry();
4001
4002         if (!d_tracer)
4003                 return NULL;
4004
4005         d_percpu = debugfs_create_dir("per_cpu", d_tracer);
4006
4007         if (!d_percpu && !once) {
4008                 once = 1;
4009                 pr_warning("Could not create debugfs directory 'per_cpu'\n");
4010                 return NULL;
4011         }
4012
4013         return d_percpu;
4014 }
4015
4016 static void tracing_init_debugfs_percpu(long cpu)
4017 {
4018         struct dentry *d_percpu = tracing_dentry_percpu();
4019         struct dentry *d_cpu;
4020         /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
4021         char cpu_dir[7];
4022
4023         if (cpu > 999 || cpu < 0)
4024                 return;
4025
4026         sprintf(cpu_dir, "cpu%ld", cpu);
4027         d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
4028         if (!d_cpu) {
4029                 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
4030                 return;
4031         }
4032
4033         /* per cpu trace_pipe */
4034         trace_create_file("trace_pipe", 0444, d_cpu,
4035                         (void *) cpu, &tracing_pipe_fops);
4036
4037         /* per cpu trace */
4038         trace_create_file("trace", 0644, d_cpu,
4039                         (void *) cpu, &tracing_fops);
4040
4041         trace_create_file("trace_pipe_raw", 0444, d_cpu,
4042                         (void *) cpu, &tracing_buffers_fops);
4043
4044         trace_create_file("stats", 0444, d_cpu,
4045                         (void *) cpu, &tracing_stats_fops);
4046 }
4047
4048 #ifdef CONFIG_FTRACE_SELFTEST
4049 /* Let selftest have access to static functions in this file */
4050 #include "trace_selftest.c"
4051 #endif
4052
4053 struct trace_option_dentry {
4054         struct tracer_opt               *opt;
4055         struct tracer_flags             *flags;
4056         struct dentry                   *entry;
4057 };
4058
4059 static ssize_t
4060 trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
4061                         loff_t *ppos)
4062 {
4063         struct trace_option_dentry *topt = filp->private_data;
4064         char *buf;
4065
4066         if (topt->flags->val & topt->opt->bit)
4067                 buf = "1\n";
4068         else
4069                 buf = "0\n";
4070
4071         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4072 }
4073
4074 static ssize_t
4075 trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
4076                          loff_t *ppos)
4077 {
4078         struct trace_option_dentry *topt = filp->private_data;
4079         unsigned long val;
4080         char buf[64];
4081         int ret;
4082
4083         if (cnt >= sizeof(buf))
4084                 return -EINVAL;
4085
4086         if (copy_from_user(&buf, ubuf, cnt))
4087                 return -EFAULT;
4088
4089         buf[cnt] = 0;
4090
4091         ret = strict_strtoul(buf, 10, &val);
4092         if (ret < 0)
4093                 return ret;
4094
4095         if (val != 0 && val != 1)
4096                 return -EINVAL;
4097
4098         if (!!(topt->flags->val & topt->opt->bit) != val) {
4099                 mutex_lock(&trace_types_lock);
4100                 ret = __set_tracer_option(current_trace, topt->flags,
4101                                           topt->opt, !val);
4102                 mutex_unlock(&trace_types_lock);
4103                 if (ret)
4104                         return ret;
4105         }
4106
4107         *ppos += cnt;
4108
4109         return cnt;
4110 }
4111
4112
4113 static const struct file_operations trace_options_fops = {
4114         .open = tracing_open_generic,
4115         .read = trace_options_read,
4116         .write = trace_options_write,
4117 };
4118
4119 static ssize_t
4120 trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
4121                         loff_t *ppos)
4122 {
4123         long index = (long)filp->private_data;
4124         char *buf;
4125
4126         if (trace_flags & (1 << index))
4127                 buf = "1\n";
4128         else
4129                 buf = "0\n";
4130
4131         return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
4132 }
4133
4134 static ssize_t
4135 trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
4136                          loff_t *ppos)
4137 {
4138         long index = (long)filp->private_data;
4139         char buf[64];
4140         unsigned long val;
4141         int ret;
4142
4143         if (cnt >= sizeof(buf))
4144                 return -EINVAL;
4145
4146         if (copy_from_user(&buf, ubuf, cnt))
4147                 return -EFAULT;
4148
4149         buf[cnt] = 0;
4150
4151         ret = strict_strtoul(buf, 10, &val);
4152         if (ret < 0)
4153                 return ret;
4154
4155         if (val != 0 && val != 1)
4156                 return -EINVAL;
4157         set_tracer_flags(1 << index, val);
4158
4159         *ppos += cnt;
4160
4161         return cnt;
4162 }
4163
4164 static const struct file_operations trace_options_core_fops = {
4165         .open = tracing_open_generic,
4166         .read = trace_options_core_read,
4167         .write = trace_options_core_write,
4168 };
4169
4170 struct dentry *trace_create_file(const char *name,
4171                                  mode_t mode,
4172                                  struct dentry *parent,
4173                                  void *data,
4174                                  const struct file_operations *fops)
4175 {
4176         struct dentry *ret;
4177
4178         ret = debugfs_create_file(name, mode, parent, data, fops);
4179         if (!ret)
4180                 pr_warning("Could not create debugfs '%s' entry\n", name);
4181
4182         return ret;
4183 }
4184
4185
4186 static struct dentry *trace_options_init_dentry(void)
4187 {
4188         struct dentry *d_tracer;
4189         static struct dentry *t_options;
4190
4191         if (t_options)
4192                 return t_options;
4193
4194         d_tracer = tracing_init_dentry();
4195         if (!d_tracer)
4196                 return NULL;
4197
4198         t_options = debugfs_create_dir("options", d_tracer);
4199         if (!t_options) {
4200                 pr_warning("Could not create debugfs directory 'options'\n");
4201                 return NULL;
4202         }
4203
4204         return t_options;
4205 }
4206
4207 static void
4208 create_trace_option_file(struct trace_option_dentry *topt,
4209                          struct tracer_flags *flags,
4210                          struct tracer_opt *opt)
4211 {
4212         struct dentry *t_options;
4213
4214         t_options = trace_options_init_dentry();
4215         if (!t_options)
4216                 return;
4217
4218         topt->flags = flags;
4219         topt->opt = opt;
4220
4221         topt->entry = trace_create_file(opt->name, 0644, t_options, topt,
4222                                     &trace_options_fops);
4223
4224 }
4225
4226 static struct trace_option_dentry *
4227 create_trace_option_files(struct tracer *tracer)
4228 {
4229         struct trace_option_dentry *topts;
4230         struct tracer_flags *flags;
4231         struct tracer_opt *opts;
4232         int cnt;
4233
4234         if (!tracer)
4235                 return NULL;
4236
4237         flags = tracer->flags;
4238
4239         if (!flags || !flags->opts)
4240                 return NULL;
4241
4242         opts = flags->opts;
4243
4244         for (cnt = 0; opts[cnt].name; cnt++)
4245                 ;
4246
4247         topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
4248         if (!topts)
4249                 return NULL;
4250
4251         for (cnt = 0; opts[cnt].name; cnt++)
4252                 create_trace_option_file(&topts[cnt], flags,
4253                                          &opts[cnt]);
4254
4255         return topts;
4256 }
4257
4258 static void
4259 destroy_trace_option_files(struct trace_option_dentry *topts)
4260 {
4261         int cnt;
4262
4263         if (!topts)
4264                 return;
4265
4266         for (cnt = 0; topts[cnt].opt; cnt++) {
4267                 if (topts[cnt].entry)
4268                         debugfs_remove(topts[cnt].entry);
4269         }
4270
4271         kfree(topts);
4272 }
4273
4274 static struct dentry *
4275 create_trace_option_core_file(const char *option, long index)
4276 {
4277         struct dentry *t_options;
4278
4279         t_options = trace_options_init_dentry();
4280         if (!t_options)
4281                 return NULL;
4282
4283         return trace_create_file(option, 0644, t_options, (void *)index,
4284                                     &trace_options_core_fops);
4285 }
4286
4287 static __init void create_trace_options_dir(void)
4288 {
4289         struct dentry *t_options;
4290         int i;
4291
4292         t_options = trace_options_init_dentry();
4293         if (!t_options)
4294                 return;
4295
4296         for (i = 0; trace_options[i]; i++)
4297                 create_trace_option_core_file(trace_options[i], i);
4298 }
4299
4300 static __init int tracer_init_debugfs(void)
4301 {
4302         struct dentry *d_tracer;
4303         int cpu;
4304
4305         trace_access_lock_init();
4306
4307         d_tracer = tracing_init_dentry();
4308
4309         trace_create_file("tracing_enabled", 0644, d_tracer,
4310                         &global_trace, &tracing_ctrl_fops);
4311
4312         trace_create_file("trace_options", 0644, d_tracer,
4313                         NULL, &tracing_iter_fops);
4314
4315         trace_create_file("tracing_cpumask", 0644, d_tracer,
4316                         NULL, &tracing_cpumask_fops);
4317
4318         trace_create_file("trace", 0644, d_tracer,
4319                         (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
4320
4321         trace_create_file("available_tracers", 0444, d_tracer,
4322                         &global_trace, &show_traces_fops);
4323
4324         trace_create_file("current_tracer", 0644, d_tracer,
4325                         &global_trace, &set_tracer_fops);
4326
4327 #ifdef CONFIG_TRACER_MAX_TRACE
4328         trace_create_file("tracing_max_latency", 0644, d_tracer,
4329                         &tracing_max_latency, &tracing_max_lat_fops);
4330 #endif
4331
4332         trace_create_file("tracing_thresh", 0644, d_tracer,
4333                         &tracing_thresh, &tracing_max_lat_fops);
4334
4335         trace_create_file("README", 0444, d_tracer,
4336                         NULL, &tracing_readme_fops);
4337
4338         trace_create_file("trace_pipe", 0444, d_tracer,
4339                         (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
4340
4341         trace_create_file("buffer_size_kb", 0644, d_tracer,
4342                         &global_trace, &tracing_entries_fops);
4343
4344         trace_create_file("trace_marker", 0220, d_tracer,
4345                         NULL, &tracing_mark_fops);
4346
4347         trace_create_file("saved_cmdlines", 0444, d_tracer,
4348                         NULL, &tracing_saved_cmdlines_fops);
4349
4350         trace_create_file("trace_clock", 0644, d_tracer, NULL,
4351                           &trace_clock_fops);
4352
4353 #ifdef CONFIG_DYNAMIC_FTRACE
4354         trace_create_file("dyn_ftrace_total_info", 0444, d_tracer,
4355                         &ftrace_update_tot_cnt, &tracing_dyn_info_fops);
4356 #endif
4357
4358         create_trace_options_dir();
4359
4360         for_each_tracing_cpu(cpu)
4361                 tracing_init_debugfs_percpu(cpu);
4362
4363         return 0;
4364 }
4365
4366 static int trace_panic_handler(struct notifier_block *this,
4367                                unsigned long event, void *unused)
4368 {
4369         if (ftrace_dump_on_oops)
4370                 ftrace_dump(ftrace_dump_on_oops);
4371         return NOTIFY_OK;
4372 }
4373
4374 static struct notifier_block trace_panic_notifier = {
4375         .notifier_call  = trace_panic_handler,
4376         .next           = NULL,
4377         .priority       = 150   /* priority: INT_MAX >= x >= 0 */
4378 };
4379
4380 static int trace_die_handler(struct notifier_block *self,
4381                              unsigned long val,
4382                              void *data)
4383 {
4384         switch (val) {
4385         case DIE_OOPS:
4386                 if (ftrace_dump_on_oops)
4387                         ftrace_dump(ftrace_dump_on_oops);
4388                 break;
4389         default:
4390                 break;
4391         }
4392         return NOTIFY_OK;
4393 }
4394
4395 static struct notifier_block trace_die_notifier = {
4396         .notifier_call = trace_die_handler,
4397         .priority = 200
4398 };
4399
4400 /*
4401  * printk is set to max of 1024, we really don't need it that big.
4402  * Nothing should be printing 1000 characters anyway.
4403  */
4404 #define TRACE_MAX_PRINT         1000
4405
4406 /*
4407  * Define here KERN_TRACE so that we have one place to modify
4408  * it if we decide to change what log level the ftrace dump
4409  * should be at.
4410  */
4411 #define KERN_TRACE              KERN_EMERG
4412
4413 static void
4414 trace_printk_seq(struct trace_seq *s)
4415 {
4416         /* Probably should print a warning here. */
4417         if (s->len >= 1000)
4418                 s->len = 1000;
4419
4420         /* should be zero ended, but we are paranoid. */
4421         s->buffer[s->len] = 0;
4422
4423         printk(KERN_TRACE "%s", s->buffer);
4424
4425         trace_seq_init(s);
4426 }
4427
4428 static void
4429 __ftrace_dump(bool disable_tracing, enum ftrace_dump_mode oops_dump_mode)
4430 {
4431         static arch_spinlock_t ftrace_dump_lock =
4432                 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
4433         /* use static because iter can be a bit big for the stack */
4434         static struct trace_iterator iter;
4435         unsigned int old_userobj;
4436         static int dump_ran;
4437         unsigned long flags;
4438         int cnt = 0, cpu;
4439
4440         /* only one dump */
4441         local_irq_save(flags);
4442         arch_spin_lock(&ftrace_dump_lock);
4443         if (dump_ran)
4444                 goto out;
4445
4446         dump_ran = 1;
4447
4448         tracing_off();
4449
4450         if (disable_tracing)
4451                 ftrace_kill();
4452
4453         for_each_tracing_cpu(cpu) {
4454                 atomic_inc(&global_trace.data[cpu]->disabled);
4455         }
4456
4457         old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4458
4459         /* don't look at user memory in panic mode */
4460         trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
4461
4462         /* Simulate the iterator */
4463         iter.tr = &global_trace;
4464         iter.trace = current_trace;
4465
4466         switch (oops_dump_mode) {
4467         case DUMP_ALL:
4468                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4469                 break;
4470         case DUMP_ORIG:
4471                 iter.cpu_file = raw_smp_processor_id();
4472                 break;
4473         case DUMP_NONE:
4474                 goto out_enable;
4475         default:
4476                 printk(KERN_TRACE "Bad dumping mode, switching to all CPUs dump\n");
4477                 iter.cpu_file = TRACE_PIPE_ALL_CPU;
4478         }
4479
4480         printk(KERN_TRACE "Dumping ftrace buffer:\n");
4481
4482         /*
4483          * We need to stop all tracing on all CPUS to read the
4484          * the next buffer. This is a bit expensive, but is
4485          * not done often. We fill all what we can read,
4486          * and then release the locks again.
4487          */
4488
4489         while (!trace_empty(&iter)) {
4490
4491                 if (!cnt)
4492                         printk(KERN_TRACE "---------------------------------\n");
4493
4494                 cnt++;
4495
4496                 /* reset all but tr, trace, and overruns */
4497                 memset(&iter.seq, 0,
4498                        sizeof(struct trace_iterator) -
4499                        offsetof(struct trace_iterator, seq));
4500                 iter.iter_flags |= TRACE_FILE_LAT_FMT;
4501                 iter.pos = -1;
4502
4503                 if (find_next_entry_inc(&iter) != NULL) {
4504                         int ret;
4505
4506                         ret = print_trace_line(&iter);
4507                         if (ret != TRACE_TYPE_NO_CONSUME)
4508                                 trace_consume(&iter);
4509                 }
4510
4511                 trace_printk_seq(&iter.seq);
4512         }
4513
4514         if (!cnt)
4515                 printk(KERN_TRACE "   (ftrace buffer empty)\n");
4516         else
4517                 printk(KERN_TRACE "---------------------------------\n");
4518
4519  out_enable:
4520         /* Re-enable tracing if requested */
4521         if (!disable_tracing) {
4522                 trace_flags |= old_userobj;
4523
4524                 for_each_tracing_cpu(cpu) {
4525                         atomic_dec(&global_trace.data[cpu]->disabled);
4526                 }
4527                 tracing_on();
4528         }
4529
4530  out:
4531         arch_spin_unlock(&ftrace_dump_lock);
4532         local_irq_restore(flags);
4533 }
4534
4535 /* By default: disable tracing after the dump */
4536 void ftrace_dump(enum ftrace_dump_mode oops_dump_mode)
4537 {
4538         __ftrace_dump(true, oops_dump_mode);
4539 }
4540
4541 __init static int tracer_alloc_buffers(void)
4542 {
4543         int ring_buf_size;
4544         int i;
4545         int ret = -ENOMEM;
4546
4547         if (!alloc_cpumask_var(&tracing_buffer_mask, GFP_KERNEL))
4548                 goto out;
4549
4550         if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
4551                 goto out_free_buffer_mask;
4552
4553         /* To save memory, keep the ring buffer size to its minimum */
4554         if (ring_buffer_expanded)
4555                 ring_buf_size = trace_buf_size;
4556         else
4557                 ring_buf_size = 1;
4558
4559         cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
4560         cpumask_copy(tracing_cpumask, cpu_all_mask);
4561
4562         /* TODO: make the number of buffers hot pluggable with CPUS */
4563         global_trace.buffer = ring_buffer_alloc(ring_buf_size,
4564                                                    TRACE_BUFFER_FLAGS);
4565         if (!global_trace.buffer) {
4566                 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
4567                 WARN_ON(1);
4568                 goto out_free_cpumask;
4569         }
4570         global_trace.entries = ring_buffer_size(global_trace.buffer);
4571
4572
4573 #ifdef CONFIG_TRACER_MAX_TRACE
4574         max_tr.buffer = ring_buffer_alloc(ring_buf_size,
4575                                              TRACE_BUFFER_FLAGS);
4576         if (!max_tr.buffer) {
4577                 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
4578                 WARN_ON(1);
4579                 ring_buffer_free(global_trace.buffer);
4580                 goto out_free_cpumask;
4581         }
4582         max_tr.entries = ring_buffer_size(max_tr.buffer);
4583         WARN_ON(max_tr.entries != global_trace.entries);
4584 #endif
4585
4586         /* Allocate the first page for all buffers */
4587         for_each_tracing_cpu(i) {
4588                 global_trace.data[i] = &per_cpu(global_trace_cpu, i);
4589                 max_tr.data[i] = &per_cpu(max_tr_data, i);
4590         }
4591
4592         trace_init_cmdlines();
4593
4594         register_tracer(&nop_trace);
4595         current_trace = &nop_trace;
4596         /* All seems OK, enable tracing */
4597         tracing_disabled = 0;
4598
4599         atomic_notifier_chain_register(&panic_notifier_list,
4600                                        &trace_panic_notifier);
4601
4602         register_die_notifier(&trace_die_notifier);
4603
4604         return 0;
4605
4606 out_free_cpumask:
4607         free_cpumask_var(tracing_cpumask);
4608 out_free_buffer_mask:
4609         free_cpumask_var(tracing_buffer_mask);
4610 out:
4611         return ret;
4612 }
4613
4614 __init static int clear_boot_tracer(void)
4615 {
4616         /*
4617          * The default tracer at boot buffer is an init section.
4618          * This function is called in lateinit. If we did not
4619          * find the boot tracer, then clear it out, to prevent
4620          * later registration from accessing the buffer that is
4621          * about to be freed.
4622          */
4623         if (!default_bootup_tracer)
4624                 return 0;
4625
4626         printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4627                default_bootup_tracer);
4628         default_bootup_tracer = NULL;
4629
4630         return 0;
4631 }
4632
4633 early_initcall(tracer_alloc_buffers);
4634 fs_initcall(tracer_init_debugfs);
4635 late_initcall(clear_boot_tracer);